Volume 2 - International Environmental Modelling and Software ...
Volume 2 - International Environmental Modelling and Software ...
Volume 2 - International Environmental Modelling and Software ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Complexity <strong>and</strong> Integrated<br />
Resources Management<br />
Transactions<br />
of the 2nd Biennial Meeting of the<br />
<strong>International</strong> <strong>Environmental</strong> <strong>Modelling</strong><br />
<strong>and</strong> <strong>Software</strong> Society<br />
Editors<br />
Claudia Pahl-Wostl<br />
Sonja Schmidt<br />
Andrea E. Rizzoli<br />
Anthony J. Jakeman<br />
IEMSs 2004 – 14-17 June 2004,<br />
University of Osnabrück, Germany
Complexity <strong>and</strong> Integrated Resources Management - Transactions of the 2nd Biennial<br />
Meeting of the <strong>International</strong> <strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong> <strong>Software</strong> Society<br />
<strong>Volume</strong> Editors<br />
Claudia Pahl-Wostl<br />
Sonja Schmidt<br />
Institut für Umweltsystemforschung<br />
Universität Osnabrück<br />
Artilleriestr. 34<br />
D 49076 Osnabrück, Germany<br />
Andrea E. Rizzoli<br />
IDSIA Istituto Dalle Molle di studi sull'intelligenza artificiale<br />
Galleria 2<br />
CH 6928 Manno, Switzerl<strong>and</strong><br />
Anthony J. Jakeman<br />
The Centre for Resource <strong>and</strong> <strong>Environmental</strong> Studies (Bldg 43)<br />
The Australian National University<br />
ACT 0200, Canberra, Australia<br />
Each paper in this volume was refereed by an Editor, a member of the Editorial Board <strong>and</strong><br />
two anonymous referees.<br />
The copyright of all papers is an exclusive right of the authors. No work can be reproduced<br />
without written permission of the authors.<br />
Responsibility for the contents of these papers rests upon the authors <strong>and</strong> not on the<br />
<strong>International</strong> <strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong> <strong>Software</strong> Society.<br />
ISBN 88-900787-1-5<br />
iEMSs 2004<br />
Published by the <strong>International</strong> <strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong> <strong>Software</strong> Society (iEMSs)<br />
President: Anthony J. Jakeman.<br />
Address: iEMSs, c/- IDSIA, Galleria 2, 6928 Manno, Switzerl<strong>and</strong><br />
Email: secretary@iemss.org<br />
Website: http://www.iemss.org<br />
II<br />
Typeset in Como (Italy) by SEA, Servizi Editoriali Associati
IEMSs 2004 – 14-17 June 2004,<br />
University of Osnabrück, Germany<br />
Complexity <strong>and</strong> Integrated Resources<br />
Management Transactions of the 2nd Biennial<br />
Meeting of the <strong>International</strong> <strong>Environmental</strong><br />
<strong>Modelling</strong> <strong>and</strong> <strong>Software</strong> Society<br />
Claudia Pahl-Wostl, Sonja Schmidt, Andrea<br />
E. Rizzoli, Anthony J. Jakeman ( E d i t o r s )<br />
Co-editors<br />
David Batten Keith Jeffery Dale S. Rothman<br />
Michel Blind Kostas Karatzas Miquel Sànchez-Marrè<br />
Felix Chan Markus Knoflacher Dragan Savic<br />
Barry Croke Peter Krause Huub Scholten<br />
Wolfgang-Albert Flügel Christine Lim Boris Schröder<br />
Carlo Giupponi Ian Littlewood Jan Sendzimir<br />
Romy Greiner Michael Matthies Ralf Seppelt<br />
Carlo Gualtieri Michael McAleer Achim Sydow<br />
Nigel Hall Dragutin T. Mihailovic Hilde Passier<br />
Matt Hare Les Oxley David Post<br />
Reinout Heijungs Jens C. Refsgaard Peter Vanrolleghem<br />
Stefanie Hellweg<br />
Otto Richter<br />
Suhejla Hoti<br />
Michela Robba<br />
Organizers<br />
The conference has been organized by iEMSs<br />
(the <strong>International</strong> <strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong> <strong>Software</strong> Society) in cooperation with:<br />
Harmoni-CA (Concerted Action on Harmonizing <strong>Modelling</strong> Tools for River Basin Management)<br />
TIAS (Integrated Assessment Society)<br />
IAHS (<strong>International</strong> Association of Hydrological Sciences)<br />
BESAI (Binding <strong>Environmental</strong> Sciences <strong>and</strong> Artificial Intelligence)<br />
MODSS <strong>International</strong> Conference on Multi-objective Decision Support Systems for L<strong>and</strong>,<br />
Water <strong>and</strong> <strong>Environmental</strong> Management<br />
ISESS <strong>International</strong> Symposium on <strong>Environmental</strong> <strong>Software</strong> Systems<br />
ERCIM the European Research Consortium for Informatics <strong>and</strong> Mathematics<br />
Local Organizers<br />
The local organization of the conference has been managed by:<br />
DBU (German <strong>Environmental</strong> Foundation)<br />
USF (Institute of <strong>Environmental</strong> Systems Research, University of Osnabrück)<br />
III
Editorial<br />
Dear Reader,<br />
The 2nd Biennial Meeting of the <strong>International</strong> <strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong> <strong>Software</strong> Society (iEMSS 2004) was dedicated<br />
to the theme: Complexity <strong>and</strong> Integrated Resources Management”, a very topical theme given the increasing complexity<br />
of contemporary resource management problems <strong>and</strong> the increasing uncertainties from global change. The meeting<br />
assembled nearly 300 researchers from all over the globe <strong>and</strong> from a wide range of disciplines. Presentations<br />
discussed latest developments in modelling methodologies <strong>and</strong> software tools applied to different areas of resources<br />
management. Contributions provided evidence of the important role of models to improve our underst<strong>and</strong>ing of the complexity<br />
of socio-ecological systems <strong>and</strong> to develop appropriate management strategies. Increasing attention was paid<br />
to the role of stakeholders in model development <strong>and</strong> application <strong>and</strong> to a new role for models in processes of social<br />
learning in participatory resources management.<br />
The conference took place in the facilities of the German <strong>Environmental</strong> Foundation in Osnabrück. The ambience of<br />
these low-energy buildings, designed to minimise their impact on the environment, was well suited to the conference<br />
theme <strong>and</strong> their open <strong>and</strong> flexible structure facilitated intense discussions <strong>and</strong> exchange not only during but also<br />
between sessions.<br />
I hope that readers will share the excitement of conference participants when browsing through the conference proceedings<br />
<strong>and</strong> reading some of the papers in more detail. Interested readers are advised to consult the journals<br />
<strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong> <strong>Software</strong> <strong>and</strong> Ecological <strong>Modelling</strong> <strong>and</strong> Advances in Geosciences where special issues<br />
emanating from this conference will be published. We also look forward to the third biennial meeting, iEMSs 2006, which<br />
will be held in Burlington, Vermont, USA (see http://www.iemss.org/iemss2006).<br />
October 2004<br />
Claudia Pahl-Wostl<br />
The <strong>International</strong> <strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong> <strong>Software</strong> Society acknowledges gratefully the assistance of the<br />
following people in realizing the iEMSs 2004 conference:<br />
• Claudia Pahl-Wostl for convening the conference<br />
• Sonja Schmidt for organising the conference<br />
• Andrea Rizzoli for the web-based conference management tool, creating <strong>and</strong> updating the conference website<br />
<strong>and</strong> for expert <strong>and</strong> technical advice<br />
• all session organizers <strong>and</strong> reviewers<br />
• Antje Braeuer <strong>and</strong> Georg Johann for supporting the organisation whenever <strong>and</strong> wherever necessary<br />
• all members of the Institute of <strong>Environmental</strong> Systems Research <strong>and</strong> the department of resource flow management<br />
for supporting iEMSs 2004, especially Ilke Borowski, Frank Hilker, Maja Schlüter <strong>and</strong> Dominik Reusser<br />
• all members of ZUK “Zentrum für Umweltkommunikation“ of the German <strong>Environmental</strong> Foundation<br />
IV
TABLE OF CONTENTS<br />
iEMSs 2004 sessions (part two)<br />
<strong>Environmental</strong> Informatics Towards Citizen-centred Electronic<br />
Information Services: the Urban Environment Example<br />
Using FLOSS towards Building <strong>Environmental</strong>e Information<br />
K. Karatzas, A. Masouras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 525<br />
Applying agent technologyin <strong>Environmental</strong> Management Systems underreal-time constraints<br />
I.N. Athanasiadis, P.A. Mitkas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 531<br />
Supporting the Strategic Objectives of Participative Water Resources Management;<br />
an Evaluation of the Performance of Four ICT Tools<br />
A. Swinford, B. McIntosh, P. Jeffrey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 537<br />
Web Services for <strong>Environmental</strong> Informatics<br />
E. Arauco, L. Sommaruga . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 543<br />
<strong>Environmental</strong> Decision Support Systems<br />
Concepts of Decision Support for River Rehabilitation P. Reichert, M. Borsuk, M. Hostmann,<br />
S. Schweizer, C. Spörri, K. Tockner, B. Truffer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 550<br />
Decision Making under Uncertainty in a Decision Support System for the Red River Inge<br />
A.T. de Kort, M.J. Booij . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 556<br />
Development of a GIS-based Decision Support Tool for Integrated Water Resources Management in<br />
Southern Africa<br />
M. Märker, K.Bongartz, W.A. Flügel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 562<br />
Possible Courses: Multi-Objective <strong>Modelling</strong> <strong>and</strong> Decision Support Using a Bayesian Network<br />
Approximation to a Nonpoint Source Pollution Model<br />
D. Swayne, J. Shi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 568<br />
V
A Spatial DSS for South Australia's Prawn Fisheries. Using Historic Knowledge Towards <strong>Environmental</strong><br />
<strong>and</strong> Economical Sustainability<br />
B. Ostendorf, N. Carrick . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 574<br />
Optimum Sustainable Water Management in an Urbanizing River Basin in Japan, Based on Integrated<br />
<strong>Modelling</strong> Techniques<br />
E. Kudo, M. Ostrowski . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 580<br />
Application of a GIS-based Simulation Tool to Analyze <strong>and</strong> Communicate Uncertainties in Future Water<br />
Availability in the Amudarya River Delta<br />
M. Schlüter, N. Rüger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 586<br />
Integration of MONERIS <strong>and</strong> GREAT-ER in the Decision Support System for the German Elbe River Basin<br />
J. Berlekamp, N. Graf, O. Hess, S. Lautenbach, S. Reimer, M. Matthies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p . 5 9 3<br />
An integrated tool for water policy in agriculture<br />
G.M. Bazzana . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 599<br />
Towards a Decision Support System for Real Time Risk Assessment of Hazardous Material Transport on<br />
Road<br />
D. Giglio, R. Minciardi, D. Pizzorni, R. Rudari, R. Sacile, A. Tomasoni, E. Trasforini . . . . . . . . . . . . . .p. 605<br />
Appropriate <strong>Modelling</strong> in DSSs for River Basin Management<br />
Y. Xu, M.J. Booij . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 611<br />
Water Management, Public Participation <strong>and</strong> Decision Support Systems: the MULINO Approach<br />
J. Feás, C. Giupponi, P. Rosato . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 617<br />
A Dual-scale <strong>Modelling</strong> approach to Integrated Resource Management in East <strong>and</strong> South-east Asia:<br />
Challenges <strong>and</strong> Potential solutions<br />
R. Rötter, M. van den Berg, H. Hengsdijk, J. Wolf, M. van Ittersum, H. van Keulen, E.O. Agustin, T. T. Son,<br />
N.X. Lai, W. Guanghuo, A.G. Laborte . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p . 6 2 3<br />
The role of Multi-Criteria Decision Analysis in a DEcision Support sYstem for REhabilitation<br />
of contaminated sites (the DESYRE software)<br />
C. Carlon, S. Giove, P. Agostini, A. Critto, A. Marcomini . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 629<br />
ICT Requirements for an 'evolutionary' development of WFD compliant River Basin Management Plans<br />
M. Blind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 635<br />
DAWN:A platform for evaluating water-pricing policies using a software agent society<br />
I.N. Athanasiadis, P. Vartalas, P.A. Mitkasa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 643<br />
Empirical Evaluation of Decision Support Systems: Concepts <strong>and</strong> an Example for Trumpeter Swan<br />
Management<br />
R.S. Sojda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 649<br />
An integrated modelling approach to conduct multi-factorial analyses on the impacts<br />
of climate change on whole-farm systems<br />
M. Rivington, G. Bellocchi, K.B. Matthews, K. Buchan, M. Donatelli . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 656<br />
VI
Some Methodological Concepts to Analyse the Role of IC-tools in Social Learning Processes<br />
P. Maurel, F. Cernesson, N, Ferr<strong>and</strong>, M. Craps, P. Valkering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 662<br />
Tools to Think With? Towards Underst<strong>and</strong>ing the Use <strong>and</strong> Impact of Model-Based Support Tools<br />
B.S. McIntosh, R.A.F. Seaton, P. Jeffrey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 668<br />
Uncertainty in the Water Framework Directive: Implications for Economic Analysis<br />
J. Mysiak, K. Sigel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 674<br />
An Interactive Spatial Optimisation Tool for Systematic L<strong>and</strong>scape Restoration B.A. Bryana, L.M. Perryb,<br />
D. Gerner, B. Ostendorf, N.D. Crossman . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 680<br />
Assessing the Feasibility of Using Radar Satellite Data to Detect Flood Extent <strong>and</strong> Floodplain Structures<br />
Edith Stabel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 686<br />
Optimal Groundwater Exploitation <strong>and</strong> Pollution Control<br />
A. Bagnera, M. Massabò, R. Minciardi, L. Molini, M. Robba, R. Sacile . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 693<br />
Towards an <strong>Environmental</strong> DSS basedon Spatio-Temporal Markov Chain Approximation<br />
G. Balent, M. Deconchat, S. Ladet, R. Martin-Clouaire, R. Sabbadin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 699<br />
Regional Dynamic <strong>Modelling</strong><br />
L<strong>and</strong> Use <strong>and</strong> Hydrological Management: ICHAM, an Integrated Model at a Regional Scale in<br />
Northeastern Thail<strong>and</strong><br />
N. Hall, R. Lertsirivorakul, R. Gre i n e r, S. Yongvanit, A. Yuvaniyama, R. Lastf, W. Milne-Homef . . . . . . . .p. 705<br />
Forecasting Municipal Solid Waste Generation in MajorEuropean Cities<br />
P. Beigl, G. Wassermann, F. Schneider, S. Salhofer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 711<br />
Real Time Optimal Resource Allocation in Natural Hazard Management<br />
P. Fiorucci, F. Gaetani, R. Minciardi, R. Sacile, E. Trasforini . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 717<br />
Combining Dynamic Economic Analysis <strong>and</strong> Environ-mental Impact <strong>Modelling</strong>: Addressing Uncertainty<br />
<strong>and</strong> Complexity of Agricultural Development<br />
H. Lehtonen, I. Bärlund, S. Tattari, M. Hilden . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 723<br />
Simulation of Water <strong>and</strong> Carbon Fluxes in Agro- <strong>and</strong> forest Ecosystems at the Regional Scale<br />
J. Post, V. Krysanova, F. Suckow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 730<br />
An Integrated Geomorphological <strong>and</strong> Hydrogeological MMS <strong>Modelling</strong> Framework for a Semi-Arid<br />
Mountain Basin in the High Atlas, Southern Morocco<br />
C. de Jong, R. Machauer, B. Reichert, S. Cappy, R. Viger, G. Leavesley . . . . . . . . . . . . . . . . . . . . . . . .p. 736<br />
Anticipated Effects of Re-Allocation of Intensive Livestock in S<strong>and</strong>y Areas in the Netherl<strong>and</strong>s<br />
A. van Wezel, J.D. van Dam, P. Cleij . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 742<br />
An Integrated System for the Forest Fires Dynamic Hazard Assessment Over a Wide Area<br />
P. Fiorucci, F. Gaetani, R. Minciardi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 748<br />
VII
Scenario Development <strong>and</strong> Integrated Scenario <strong>Modelling</strong><br />
Linking Narrative Storylines <strong>and</strong> Quantitative Models to Combat Desertification in the Guadalentín, Spain<br />
K. Kok, H. Van Delden . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 754<br />
Integrated Assessment of Water Stress in Ceara, Brazil, under Climate Change Forcing<br />
M.S. Krol. P. van Oel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 760<br />
From Narrative to Number: A Role for Quantitative Models in Scenario Analysis<br />
E. Kemp-Benedict . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 765<br />
Scenario Reoptimization under Data Uncertainty<br />
P. Zuddas, G.M. Sechi, A. Manca . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 771<br />
Reliable <strong>and</strong> Valid Identification of a Small Number of Global Emission Scenarios<br />
O. Tietje . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 777<br />
Simulating Global Feedbacks Between Sea Level Rise, Water for Agriculture <strong>and</strong> the Complex<br />
Socio-Economic Development of the IPCC Scenarios<br />
S. Werners, R. Boumans, L. Bouwer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 783<br />
Biocomplexity <strong>and</strong> Adaptive Ecosystem Management<br />
Principles of Human-Enviroment Systems (HES) Research<br />
R. Scholz, C. Binder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 791<br />
Addressing Sustainability, HIV-AIDS, <strong>and</strong> Water Resource Questions in Botswana<br />
M. Hellmuth, J. Sendzimir, D. Yates, K. Strzepek, W. S<strong>and</strong>erson . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 797<br />
<strong>Modelling</strong> Biocomplexity in the Tisza River Basin within a Participatory Adaptive Framework<br />
J. Sendzimir, P. Balogh, A. Vári . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 803<br />
Linking Hydrologic Modeling <strong>and</strong> Ecologic Modeling: An Application of Adaptive Ecosystem Management<br />
in the Everglades Mangrove Zone of Florida Bay<br />
J.C. Cline, J. Lorenz, E. Swain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 810<br />
On the Local Coexistence of Species in Plant Communities<br />
J. Yoshimura, K. Tainaka, T. Suzuki, M. Shiyomi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 816<br />
Ecosystems as Evolutionary Complex Systems: A Synthesis of Two System-Theoretic Approaches Based<br />
on Boolean Network<br />
B. Fath, W. Grant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 822<br />
Benthic Macroinvertebrates <strong>Modelling</strong> Using Artificial Neural Networks (ANN):<br />
Case Study of a Subtropical Brazilian River<br />
D. Pereira, M. de A. Vitola, O.C. Pedrollo, I.C. Junqueira, S.J. de Luca . . . . . . . . . . . . . . . . . . . . . . . . .p. 828<br />
Interspecific Segregation <strong>and</strong> Phase Transition in a Lattice Ecosystem with Intraspecific Competition<br />
K. Tainaka, M. Kushida, Y. Itoh, J. Yoshimura . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 834<br />
VIII
A Model of the Biocomplexity of Deforestation in Tropical Forest: Caparo Case Study<br />
R. Quintero, R. Barros, J. Dàvila, N. Moreno, G. Tonella, M. Ablan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 840<br />
Developing Tools for Adaptive Integrated Water Resource Management in a Semi-Arid Region:<br />
Possibilities, Probabilities <strong>and</strong> Uncertainties<br />
D. Eisenhuth, J.B. Abad, A. Bonnet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 846<br />
Ecological <strong>Modelling</strong><br />
Stability Analyses of the 50/50 Sex Ratio Using Lattice Simulation<br />
Y. Itoh, J. Yoshimura, K. Tainaka . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 852<br />
Reproductive Strategies of Marine Green Algae: the Evolution of Slight Anisogamy <strong>and</strong> <strong>Environmental</strong><br />
Conditions of Habitat<br />
T. Togashi, T. Miyazaki, J. Yoshimura, J.L. Bartelt, P.A. Cox . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 858<br />
Predicting Predation Efficiency of Biocontrol Agents: Linking Behavior of Individuals <strong>and</strong> Population<br />
Dynamics<br />
B. Tenhumberg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 864<br />
The Coexistence of Plankton Species with Various Nutrient Conditions: nutrient conditions: A Lattice<br />
Simulation Model<br />
T. Miyazaki, T. Togashi, T. Suzuki, T. Hashimoto, K. Tainaka, J. Yo s h i m u r a . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 870<br />
Mathematical <strong>Modelling</strong> of Harmful Algal Blooms<br />
R.R. Sarkar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 876<br />
<strong>and</strong> Calibrating Models<br />
L<strong>and</strong>scape Patterns: Simulating Changes, Identifying Driving Forces<br />
Implications of Processing Spatial Data from a Forested Catchment for a Hillslope Hydrological Model<br />
T. Kokkonen, H. Koivusalo, A. Laurén, S. Penttinen, S. Piirainen, M. Starr, L. Finér . . . . . . . . . . . . . .p. 783<br />
Generic Process-Based Plant Models for the Analysis of L<strong>and</strong>scape Change<br />
B. Reineking, A. Huth, C. Wissel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 889<br />
The Role of Local Spatial Heterogeneity in the Maintenance of Parapatric Boundaries: Agent Based<br />
Models of Competition Between two Parasitic Ticks<br />
A. Tyre, B. Tenhumberg, C.M. Bull . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 895<br />
How to Compare Different Conceptual Approaches to Metapopulation <strong>Modelling</strong><br />
F.M. Hilker, M. Hinsch, H.J. Poethke . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 902<br />
Simulation of Dynamic Tree Species Patterns in the Alpine region of Valais (Switzerl<strong>and</strong>) during the Holocene<br />
H. Lischke . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 908<br />
Aphid Population Dynamics in Agricultural L<strong>and</strong>scapes: An Agent-based Simulation Model<br />
H. Parry, A.J. Evans, D. Morgan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 914<br />
IX
Integrating Wetl<strong>and</strong>s <strong>and</strong> Riparian Zones in Regional Hydrological Modeling<br />
F.F. Hattermann, V. Krysanova, A. Habeck . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 920<br />
Ecoregion Classification Using a Bayesian Approach <strong>and</strong> Centre-Focused Clusters<br />
D. Pullar, S. Low Choy, W. Rochester . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 927<br />
Assessing Management Systems for the Conservation of Open L<strong>and</strong>scapes Using an Integrated<br />
L<strong>and</strong>scape Model Approach<br />
M. Rudner, R. Biedermann, B. Schröder, M. Kleyer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 933<br />
<strong>Environmental</strong> Interfaces<br />
Physics <strong>and</strong> <strong>Modelling</strong> of Transport <strong>and</strong> Transformation Processes at<br />
Forecasting UV Index by NEOPLANTA Model: Methodology <strong>and</strong> Validation<br />
S. Malinovic, D. Mihailovic, D. Kapor, Z. Mijatovic, I. Arsenic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 939<br />
Mathematical Models for Gene Flow from GM Crops in the Environment<br />
O. Richter, K. Foit, R. Seppelt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 945<br />
Simulation of Herbicide Transport in an Alluvial Plain<br />
K. Meiwirth, A. Mermoud . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 951<br />
The Influence of the Averaging Period on Calculation of Air Pollution Using a Puff Model<br />
B. Rajkovic, G. Zoran, P. Zlatica, D. Vladimir . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 956<br />
Interaction Between Hydrodynamics <strong>and</strong> Mass-Transfer at the Sediment-Water Interface<br />
C. Gualtieri . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 962<br />
A Spatially-Distributed Conceptual Model for Reactive Transport of Phosphorus from Diffuse Sources:<br />
an Object-Oriented Approach<br />
B. Koo, S. Dunn, R. Ferrier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 970<br />
A Probabilistic <strong>Modelling</strong> Concept for the Quantification of Flood Risks <strong>and</strong> Associated Uncertainties<br />
H. Apel, A. Thieken, B. Merz, G. Blöschl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 977<br />
Parameters Estimation Using Some Analytical Solutions of the Anisotropic Advection-Dispersion Model<br />
F. Catania, M. Massabo, O. Paladino . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 984<br />
Soil Hydraulics Properties Estimation by using Pedotransfer Functions in a Northeastern Semiarid Zone<br />
Catchment, Brazil<br />
L. Moreira, A. Marozzi Righetto, V. Medeiros . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 990<br />
An Approach for Calculating the Turbulent Transfer Coefficient Inside the Sparse Tall Vegetation<br />
D. Mihailovic, M. Budincevic, B. Lalic, D. Kapor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p. 996<br />
X
River Basin Management<br />
The Utility of GIS Delivered <strong>Environmental</strong> Models in the Characterisation of Surface Water Bodies under<br />
the Water Framework Directive: Low Flows 2000 - a Case Study<br />
T. Goodwin, M. Fry, M. Holmes, A. Young . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p.1002<br />
A Tool for Evaluating Risk to Surface Water Quality Status<br />
N. McIntyre . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p.1008<br />
Spatially Distributed Investment Prioritization for Sediment Control over the Murray Darling Basin, Australia<br />
H. Lu, C. Moran, I. Prosser, R. DeRose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p.1014<br />
Appropriate Accuracy of Models for Decision-Support Systems: Case Example for the Elbe River Basin<br />
J.L. de Kok, K.U. van der Wal, M.J. Booij . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p.1021<br />
River Basin Management Plans <strong>and</strong> Decision Support<br />
C. Giupponi, R. Camera, V. Cogan, F. Anita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p.1027<br />
Introducing River <strong>Modelling</strong> in the Implementation of the DPSIR Scheme in the Water Framework Directive<br />
S. Marsili-Libelli, S. Cavalieri, F. Betti . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p.1033<br />
Sensitivity analysis of a network-based, catchment scale water quality model<br />
L. Newham, F.T. Andrews, J. Norton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p.1039<br />
Dealing with Unidentifiable Sources of Uncertainty within <strong>Environmental</strong> Models<br />
A. van Griensven, T. Meixner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p.1045<br />
Assessing SWAT Model Performance in the Evaluation of Management Actions for the Implementation of<br />
the Water Framework Directive in a Finnish Catchment<br />
I. Bärlund, T. Kirkkala, O. Malve, J. Kämäri . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p.1051<br />
Assessing the Effects of Agricultural Change on Nitrogen Fluxes Using the Integrated<br />
Nitrogen CAtchment (INCA) Model<br />
K. Rankinen, H. Lehtonen, K. Granlund, I. Bärlund . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p.1057<br />
Implications of Complexity <strong>and</strong> Uncertainty for Integrated <strong>Modelling</strong> <strong>and</strong> Impact Assessment in River Basins<br />
V. Krysanova, F.F. Hattermann, F. Wechsung . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p.1064<br />
Coupling Surface <strong>and</strong> Ground Water Processes For Water Resources Simulation in Irrigated Alluvial Basins<br />
C. G<strong>and</strong>olfi, A. Facchi, D. Maggi, B. Ortuani . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p.1069<br />
Investigating Spatial Pattern Comparison Methods for Distributed Hydrological Model Assessment<br />
S. Weal<strong>and</strong>s, R. Grayson, J. Walker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p.1075<br />
Reduced Models of the Retention of Nitrogen in Catchments<br />
K. Wahlin, D. Shahsavani, A. Grimvall, A. Wade, D. Butterfield, H.P. Jarvie . . . . . . . . . . . . . . . . . . . . .p.1081<br />
The Evaluation of Uncertainty Propagation into River Water Quality Predictions to Guide Future<br />
Momintoring Campaigns<br />
V. V<strong>and</strong>enberghe, W. Bauwens, P.A. Vanrolleghem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .p.1087<br />
XI
Using FLOSS towards Βuilding <strong>Environmental</strong><br />
Information Systems<br />
Dr. Eng. Kostas Karatzas <strong>and</strong> Asteris Masouras<br />
Aristotle University, Department of Mechanical Engineering, 54124 Thessaloniki, Greece<br />
Abstract: Public access to environmental information is the basis for a higher degree of involvement of<br />
citizens <strong>and</strong> stakeholders in environmental decision-making [Haklay 2003][EU ISPO 1999]. <strong>Environmental</strong><br />
Information Systems play a key role in contemporary urban environmental management strategies, <strong>and</strong> are a<br />
prerequisite for the proper, timely information of the public [Kampinnen 2001]; yet the fuzzy nature of<br />
environmental information [Denzer 2002] requires for systems that can make optimum use of informatics<br />
<strong>and</strong> telecommunications infrastructures to address environmental management needs, while remaining openended,<br />
easy to use <strong>and</strong> inexpensive to implement <strong>and</strong> operate. In this paper, we will attempt to present the<br />
characteristics of FLOSS software that render it appropriate for use in developing <strong>Environmental</strong><br />
Information Systems, accompanied by real world project examples.<br />
Keywords: <strong>Environmental</strong> Informatics, <strong>Environmental</strong> Information Systems, FLOSS, Free <strong>Software</strong>, Libre<br />
<strong>Software</strong>, Open Source<br />
1. INTRODUCTION<br />
Free - Libre - Open Source <strong>Software</strong> (FLOSS,<br />
Infonomics, 2002]) is a new software development<br />
paradigm that emerged in the last decade <strong>and</strong> relies<br />
directly on the volunteer efforts of geographically<br />
dispersed developers of varying professional<br />
affiliations <strong>and</strong> proficiencies.<br />
In direct contrast with previously established<br />
business practices [Raymond, 2000], this software<br />
development paradigm is fuelled by full disclosure<br />
of the source code, volunteer effort <strong>and</strong> a number<br />
of “freedoms” granted to the software user<br />
regarding his ability to interact with the software<br />
<strong>and</strong> propagate its use.<br />
By promoting code reuse <strong>and</strong> the adaptation of<br />
freely available best practices, FLOSS<br />
development practices minimize redundancy <strong>and</strong><br />
concentrate investment on innovation [Von Hippel<br />
2003]. The support FLOSS projects receive from<br />
the user-developer community serves to provide<br />
guidance, reduce maintenance costs <strong>and</strong> enhance<br />
software sustainability, while the service-oriented<br />
model of FLOSS allows for a broad range of<br />
contractors to provide support, <strong>and</strong> helps in<br />
minimizing the Total Cost of Ownership.<br />
It is these characteristics FLOSS, as we will<br />
demonstrate in this paper, that render it flexible,<br />
economical <strong>and</strong> reusable, <strong>and</strong> thus appropriate for<br />
use in building publicly funded ICT projects<br />
[Infonomics, 2002], especially those aiming at the<br />
dissemination of information to citizens, such as<br />
online environmental portals.<br />
2. USING FLOSS SOFTWARE<br />
RESOURCES<br />
Historically, although the software model itself<br />
could be said to derive from UNIX, the FLOSS<br />
development community <strong>and</strong> underlying<br />
ideological movement is a little more than a<br />
decade old: it was officially set in motion with the<br />
first version of the GNU General Purpose License<br />
(1989) <strong>and</strong> Linus Torvalds decision to release the<br />
Linux kernel to the public (1991). FLOSS<br />
represents a software development paradigm, <strong>and</strong><br />
as such, it is fairly new, compared to its precursors<br />
whose roots go back to the '50s <strong>and</strong> '70s.<br />
The FLOSS development community consists of<br />
individuals or groups of individuals who<br />
contribute to a particular FLOSS product or<br />
technology: as a consequence of the previous<br />
statement, this also includes the users of the<br />
software. Although referencing various forms of<br />
voluntary affiliation around FLOSS projects, the<br />
community is the driving force of FLOSS software<br />
development. It constitutes a Community of<br />
Practice (CoP) [Kimble 2001], <strong>and</strong> its motivations<br />
<strong>and</strong> processes have been recorded elsewhere in<br />
525
detail [Ghosh], [Shah], [Lerner 2001]. CoP’s are<br />
described as “intrinsic conditions for the existence<br />
of knowledge”, [Lave 1991] attested to by the fact<br />
that the FLOSS community provides fertile ground<br />
for user-consumer involvement in online joint<br />
innovation [Hemetsberger 2003]. The FLOSS<br />
process refers to the approach for developing <strong>and</strong><br />
maintaining FLOSS products <strong>and</strong> technologies,<br />
including software, computers, devices, technical<br />
formats, <strong>and</strong> computer languages.<br />
The definition of Free <strong>Software</strong> recognizes some<br />
fundamental freedoms as imparted by the author<br />
(http://www.gnu.org/philosophy/free-sw.html) to<br />
the user, inside a license agreement:<br />
- The freedom to study how the program works,<br />
<strong>and</strong> the freedom to adapt the code according<br />
specific needs<br />
- The freedom to improve the program (enlarge,<br />
add functions);<br />
- The freedom to run the program, for any<br />
purpose <strong>and</strong> on any number of machines;<br />
- The freedom to redistribute copies to other<br />
users.<br />
The Open Source definition<br />
(http://www.opensource.org/docs/definition.php)<br />
further extended these principles <strong>and</strong> focused on<br />
the development process rather than the political<br />
ideology underlying the Free <strong>Software</strong> movement.<br />
The terms Open Source <strong>and</strong> Free <strong>Software</strong> refer to<br />
software developed <strong>and</strong> distributed on the above<br />
principles, with terms such as Libre software<br />
[EWGLS, 2001], or the FLOSS aggregate used to<br />
describe them together [Infonomics, 2002].<br />
Although these terms are not fully<br />
interchangeable, this paper focuses on the software<br />
development process common to both movements.<br />
Unrestricted access to the software source code is<br />
a precondition for most of these freedoms, <strong>and</strong> it is<br />
implied that the usefulness <strong>and</strong> potential for reuse<br />
of such software is dependent on the continual<br />
revision <strong>and</strong> adaptation of its source code. In<br />
proprietary <strong>and</strong> closed development environments,<br />
the frequency of revisions is dominated by the<br />
sales cycle but can also be stilled by managerial<br />
decree. In FLOSS, the “life expectancy” of<br />
software developed in is a direct outcome of its<br />
popularity with developers, who will choose to<br />
devote time to improve functionality, <strong>and</strong> users,<br />
who will provide constant feedback to developers<br />
on needed improvements <strong>and</strong> fixes.<br />
The use of FLOSS software towards building<br />
environmental information systems hinges on<br />
three points [IDA, 2002] providing benefits to<br />
users, developers <strong>and</strong> operators of the software:<br />
economy, quality <strong>and</strong> philosophy.<br />
Economy<br />
Reusing <strong>and</strong> adapting freely available best practice<br />
software, instead of resorting to monolithic<br />
proprietary solutions or developing everything<br />
from scratch leads to minimizing redundancy in<br />
development efforts <strong>and</strong> by extension, in<br />
concentrating investment on innovation. Relying<br />
on the community to spark developer interest in<br />
the software <strong>and</strong> provide user feedback reduces<br />
maintenance costs <strong>and</strong> prolongs its’ useful life<br />
cycle. A corollary of this is that the functionality<br />
<strong>and</strong> maintainability of the software is not impaired<br />
by artificial limitations (i.e. not intrinsic to the<br />
software itself), such as expiring licenses <strong>and</strong><br />
financial plights affecting a single developing<br />
entity.<br />
The Total Cost of Ownership i (TCO) of solutions<br />
based on FLOSS from a contractor point of view is<br />
alleviated [EWGLS, 2001, The Mitre Corp.,<br />
2001], since consulting fees are fully useful<br />
expenditures, in contrast with licensing fees which<br />
mostly serve as instruments of amortization for<br />
developing companies. Since <strong>Environmental</strong><br />
Information Systems development is largely<br />
supported by public funding, such amortization<br />
should not burden beneficiaries of their services.<br />
For the service-oriented model of FLOSS, it<br />
should be noted that costs of support <strong>and</strong><br />
maintenance can be contracted out to a range of<br />
suppliers, as per the competitive nature of the<br />
market ensured by source code disclosure [Lerner<br />
<strong>and</strong> Tirole, 2001].<br />
Quality<br />
The main objective in software engineering is not<br />
necessarily to spend less but rather to obtain a<br />
higher quality for the same amount of money, <strong>and</strong><br />
aim to enforce the best possible safeguards for<br />
quality <strong>and</strong> safety in the product. Avoiding to<br />
“reinvent the wheel” by using funds to develop<br />
new applications rather than re-inventing already<br />
developed parts, speeds up technological<br />
innovation -as is also the case with the increased<br />
cooperation <strong>and</strong> full source code disclosure <strong>and</strong><br />
availability required by FLOSS tenets. Finally, as<br />
has been repeatedly demonstrated in recent years<br />
[Perens, 2001, Schneier 2001)], software security<br />
concerns are better addressed through a continuous<br />
process of issue disclosure <strong>and</strong> user-developer<br />
cooperation in order to overcome them.<br />
Philosophy<br />
FLOSS presents the potential for a Social Return<br />
of Investment on public funding, by virtue of<br />
constituting a Global Public Good ii [UNDP, 2002],<br />
<strong>and</strong> by its’ potential to produce non-monetizable<br />
benefits for society, in the form a body of code<br />
526
that can be utilized in building sustainable<br />
informatics infrastructures for the public.<br />
Reliance on proprietary software for science<br />
results in vendor “lock-in” as regards to data<br />
formats, making it difficult to pursue common<br />
protocols for data interchange <strong>and</strong> storage, for<br />
instance, as it is required by modern systems<br />
dealing with the problems of environmental data<br />
heterogeneity [Visser et. al, 2001]. In contrast,<br />
FLOSS developers <strong>and</strong> proponents promote the<br />
use of open scientific st<strong>and</strong>ards, through their use<br />
in applications, as a means of consolidating<br />
researcher efforts, minimizing the cost <strong>and</strong><br />
dependencies of technical innovation. In addition,<br />
the FLOSS software movement serves the further<br />
collaboration between public bodies, professional<br />
communities <strong>and</strong> the private sector in the interests<br />
of creating a flexible <strong>and</strong> lasting service<br />
environment for the public iii . The free<br />
dissemination of technological advances (both in<br />
terms of cost <strong>and</strong> material availability) relating to<br />
informatics services, although not a panacea, can<br />
be seen to eventually help eclipse the digital divide<br />
[Schauer, 2003], by allowing poorer countries to<br />
“catch up”.<br />
3. ENVIRONMENTAL INFORMATION<br />
SYSTEMS ASPECTS<br />
<strong>Environmental</strong> Information Systems are<br />
informatics systems concerned with the<br />
management of data about the status of the<br />
environment <strong>and</strong> related scientific, regulatory,<br />
legal, managerial or other information, <strong>and</strong> are<br />
used by authorities, policy <strong>and</strong> decision makers<br />
<strong>and</strong> or experts for environmental monitoring,<br />
management planning <strong>and</strong> coordination,<br />
environmental impact assessment, urban planning<br />
<strong>and</strong> decision support. etc. Due to the complexity of<br />
the decision processes involved, disparate data<br />
from a variety of sources must be combined. This<br />
“holistic nature” of environmental information<br />
systems leads to heterogeneity problems regarding<br />
the syntax, structure, <strong>and</strong> semantics of<br />
environmental data [Visser 2001]. Overcoming<br />
them requires, among others, the adoption of<br />
common protocols for data exchange <strong>and</strong> storage,<br />
<strong>and</strong> making use of metadata to facilitate<br />
interoperability between subsystems.<br />
FLOSS promotes the use of open data st<strong>and</strong>ards as<br />
a means of consolidating researcher efforts <strong>and</strong><br />
increasing technical interoperability. Thus it can<br />
be demonstrated that the right of public access to<br />
environmental information, as has been defined in<br />
contemporary legislation [EU/EC 2003a][EU/EC<br />
2003a], is better served by utilizing open, flexible<br />
<strong>and</strong> low cost dissemination platforms that make<br />
use of software developed by the community. In<br />
the following chapter, examples of FLOSS<br />
applications are presented, all related to air quality<br />
management systems <strong>and</strong> all addressing problems<br />
that converge into the need of openness,<br />
flexibility, adaptability, resource optimum,<br />
environmental management solutions.<br />
4. PROJECT EXAMPLES<br />
In the following, we demonstrate the utilization of<br />
FLOSS resources towards building public<br />
<strong>Environmental</strong> Information Systems [Haklay<br />
2003], on the basis of EU supported projects. In<br />
the core of our approach, we realise<br />
<strong>Environmental</strong> Information being a public good<br />
<strong>and</strong> in parallel the raw material for the compilation<br />
of electronic information services. FLOSS may<br />
then be considered as a “public good” in the sense<br />
of publicly available software infrastructure <strong>and</strong><br />
functionality potential that may be used for the<br />
development of electronic environmental<br />
information services to support personal well<br />
being in accordance with environmental awareness<br />
raising, thus framing a “healthy” sustainable<br />
development paradigm. To this end, the humancentred<br />
approach of <strong>Environmental</strong> Informatics<br />
applications may be served.<br />
4.1 The APNEE/APNEE-TU projects:<br />
The APNEE project (http://www.apnee.org)<br />
contributed to the European research on public<br />
information systems <strong>and</strong> services, by developing<br />
citizen-centered dynamic information services<br />
aimed at providing intelligence about the ambient<br />
environment. These services advise the citizen<br />
about the air quality in terms of air quality indexes<br />
<strong>and</strong> offer guidelines for behavioural change.<br />
Awareness services are based upon an array of<br />
information channels to reach the citizen. APNEE<br />
further utilises various intuitive presentation<br />
formats to convey information. The configuration<br />
of such ambient technologies <strong>and</strong> the selection<br />
specific information channels has been evaluated<br />
in field trials in different European regions.<br />
APNEE-TU further investigated with success the<br />
feasibility <strong>and</strong> adaptability of the APNEE<br />
approach in relation to new technologies, extended<br />
<strong>and</strong> updated content, <strong>and</strong> new application sites<br />
It was apparent from the beginning of the APNEE<br />
project that a flexible, modular <strong>and</strong> cost-effective<br />
architecture was needed, to support the<br />
environmental information needs of urban<br />
agglomerations through easy-to-use <strong>and</strong> easy-toaccess<br />
interfaces that would allow a measure of<br />
personalization / customization in order to prove<br />
attractive to citizens. For this reason, development<br />
527
of the APNEE regional server was based on<br />
FLOSS technologies.<br />
APNEE / APNEE-TU is composed of a set of<br />
reference core modules, including the database,<br />
the service triggers, the regional server application<br />
<strong>and</strong> basic functionality modules (licensed as Open<br />
Source), as well as proprietary extension modules<br />
developed by telecommunication partners to<br />
provide services based on local ICT infrastructure<br />
conditions. Although the core modules are<br />
considered to be the heart of the system, yet, they<br />
may be “by-passed” or not implemented, in cases<br />
where only the electronic services are of interest<br />
for installation <strong>and</strong> operational usage, provided<br />
that there is a database <strong>and</strong> a pull <strong>and</strong> push scheme<br />
provided via alternative software infrastructures.<br />
The modules <strong>and</strong> the interfaces that have been<br />
developed by the project consortium, to build the<br />
regional server, are the following:<br />
• APNEE <strong>Environmental</strong> Database: the database<br />
forms the back-end of all APNEE-TU services<br />
<strong>and</strong> consists of a schema for environmental data<br />
series, as well as warnings, medical advises,<br />
pollutant information; spatial data for the<br />
WebGIS component, <strong>and</strong> user information <strong>and</strong><br />
personalization data for subscribers. APNEE-TU<br />
provides an object-relational persistence layer to<br />
allow cooperation with a variety of FLOSS <strong>and</strong><br />
proprietary RDBMS systems.<br />
• APNEE Regional Server: the centerpiece of the<br />
APNEE platform, the regional server provides a<br />
web-based anchoring point for APNEE services,<br />
configured <strong>and</strong> localized as per the needs of each<br />
installation site; also provides administrative<br />
interfaces for a variety of functionalities, such as<br />
subscription to the newsletter, <strong>and</strong> email<br />
services. The materialisation of the regional<br />
Server for Thessaloniki-Greece is presented in<br />
Figure 1.<br />
• Push services: these services consist of modules<br />
that are executed on when changes in the<br />
database occur. What kind of database change<br />
that will launch a module, is configured in the<br />
trigger. Push services consist of sending SMS<br />
<strong>and</strong> email messages to the citizen periodically, or<br />
upon user specified conditions, <strong>and</strong> are mostly<br />
used to send out alerts <strong>and</strong> warnings.<br />
• Pull services: these services are used whenever<br />
another application requests information from<br />
the database. This includes requests made from<br />
users via WWW, PDA, or WAP, <strong>and</strong> requests<br />
from automatic processes using the XML-RPC<br />
or SOAP interface.<br />
In addition, a variety of technologies was used for<br />
the development of electronic environmental<br />
information service applications, including:<br />
• J2ME Applications: Based on Java 2 Platform,<br />
Micro Edition (J2ME) for mobile devices, these<br />
applications serve the need for on-time<br />
information of remote users. They require to be<br />
downloaded to a J2ME-enabled mobile phone<br />
<strong>and</strong> provide static <strong>and</strong> dynamic information<br />
which is updated by using GPRS <strong>and</strong> http<br />
protocols.<br />
• PDA web application: This is a “lighter”<br />
version of the regional server, with trimmed<br />
layout, navigation, images <strong>and</strong> content to allow<br />
for the smaller display size of the average PDA.<br />
It allows the browsing of web application via a<br />
PDA, by automatic adaptation to the device<br />
characteristics to display the same content <strong>and</strong><br />
give access to the same bundle of information<br />
services.<br />
Database<br />
Data<br />
provider<br />
Application<br />
framework<br />
Auxilliary technologies<br />
(Perl, WML, C)<br />
Servlet<br />
container<br />
Business<br />
logic<br />
Template<br />
engine<br />
Development<br />
environment<br />
Figure 1. The APNEE-TU services<br />
architecture materialisation for Thessaloniki,<br />
Greece.<br />
Web server<br />
Services<br />
interface<br />
Implementation of the APNEE regional server<br />
(ARS) was based on Java servlets, server-based<br />
web applications, <strong>and</strong> a variety of technologies<br />
made available by the Apache Foundation (Figure<br />
3). At the inception of APNEE, in early 1999, Java<br />
2 Enterprise Edition (J2EE) had not been finalized,<br />
while the first FLOSS J2EE-compatible container<br />
was certified in 2002. The ARS, while nominally a<br />
J2EE application, does not take full advantage of<br />
the J2EE framework, as it was based on Jakarta<br />
Turbine, a modular service oriented web<br />
application framework that provides the Torque<br />
object-relational mapping layer, the Velocity<br />
dynamic HTML-based template language, a<br />
comprehensive RBAC security system that<br />
incorporates groups of users, a templating<br />
framework <strong>and</strong> an intra-application service<br />
management layer for pluggable services. The<br />
templating framework of Turbine worked very<br />
well in our case, allowing separate teams to focus<br />
on presentation <strong>and</strong> logic, in conjunction with the<br />
SMS<br />
WAP<br />
GSM<br />
Email<br />
528
automatic build <strong>and</strong> deployment scripting system<br />
based on Apache Ant. We found out that the<br />
Turbine security model is rich enough for our<br />
need, but that it does not map too well to the J2EE<br />
security model. The ARS programmatically<br />
provided service front-ends to the environmental<br />
database through Torque, as well as Web Services<br />
interfaces, via XML-RPC <strong>and</strong> Apache Axis.<br />
Finally, Jakarta Tomcat was chosen as the default<br />
servlet container for both development <strong>and</strong><br />
production use.<br />
In recognition of the value of the FLOSS software<br />
paradigm towards building informatics systems for<br />
the public, <strong>and</strong> in order to preserve <strong>and</strong><br />
disseminate development efforts, the APNEE-TU<br />
project produced a “reference implementation”,<br />
composed of the environmental database, regional<br />
server <strong>and</strong> core modules, which is licensed as<br />
Open Source, thus making it’s embracement <strong>and</strong><br />
support of use cases by anyone interested much<br />
easier.<br />
4.2 APNEE Use Cases:<br />
Citizens might daily traverse several parts of the<br />
city, in commuting to <strong>and</strong> from their places of<br />
work, <strong>and</strong> in attending to personal <strong>and</strong> family<br />
needs (access to local markets <strong>and</strong> public services<br />
etc.).<br />
- Certain population groups being sensitive to air<br />
quality levels, as they can affect their health,<br />
especially in regard with the respiratory system,<br />
leads to specific needs for environmental<br />
information.<br />
- The parents of children affected by respiratory<br />
problems, allergies <strong>and</strong> other related health<br />
problems would benefit from advance warning,<br />
based on scientifically authoritative predictions,<br />
on possible increases in air pollution levels, the<br />
discomfort index <strong>and</strong> any other parameter that<br />
affects the quality of the urban environment in<br />
residential areas, school districts etc.<br />
- People with respiratory problems would be<br />
interested in being informed early on sudden<br />
increases of air pollution levels, before<br />
commuting to their places of work, or visiting<br />
friends <strong>and</strong> relatives in remote parts of the city<br />
- Elderly people <strong>and</strong> performers of strenuous<br />
physical exercise or continuous labor in open<br />
spaces would also be interested in timely <strong>and</strong><br />
authoritative information on the state of the<br />
ambient air, <strong>and</strong> the possible negative health<br />
effects of air pollution.<br />
Citizens concerned about these <strong>and</strong> other issues<br />
connected with the urban environment can make<br />
use of the environmental information services<br />
offered by the APNEE portal, through the<br />
multitude of complementary communications<br />
channels supported <strong>and</strong> made possible by the<br />
technological advances of the Information Society.<br />
5. CONCLUSIONS<br />
The need for collaborative environmental<br />
information management <strong>and</strong> dissemination<br />
modules that would allow for implementing a<br />
homogenized, service-based, user perspective of<br />
heterogeneous data <strong>and</strong> computational resources<br />
was the main drive towards modular <strong>and</strong> open<br />
software architectures in projects such as the ones<br />
previously mentioned. It was also made apparent<br />
that project design <strong>and</strong> development could benefit<br />
from the use of Free/Libre/Open Source <strong>Software</strong>.<br />
This was mainly initiated within the last decade<br />
via the usage of Internet based communication<br />
infrastructures, leading to the flourishing of<br />
platform-independent (eg. Java based) software<br />
module implementations <strong>and</strong> the need for costeffective<br />
<strong>and</strong> reliable informatics solutions.<br />
APNEE/APNEE-TU followed the trends outlined<br />
above <strong>and</strong> provided a flexible, cost-effective<br />
working solution for implementing a public<br />
environmental information ICT infrastructure,<br />
making use of resources developed by the<br />
aggregate FLOSS community.<br />
6. ACKNOWLEDGEMENTS<br />
The authors greatly acknowledge the European<br />
Commission for supporting research projects<br />
APNEE (IST 1999-11517) <strong>and</strong> APNEE-TU (IST<br />
2001-34154), <strong>and</strong> their partners herein.<br />
7. REFERENCES<br />
Bøhler, T., K. Karatzas, G. Peinel, Th. Rose <strong>and</strong><br />
R. San Jose, Providing multi-modal access to<br />
environmental data–customizable information<br />
services for disseminating urban air quality<br />
information in APNEE, Computers,<br />
Environment <strong>and</strong> Urban Systems, 26(1), 39-61,<br />
2002<br />
Chan, T., J. Lee, A Comparative Study of Online<br />
User Communities Involvement, Product<br />
Innovation <strong>and</strong> Development, 2004,<br />
http://opensource.mit.edu/papers/chanlee.pdf<br />
Denzer, R., Generic Integration in <strong>Environmental</strong><br />
Information <strong>and</strong> Decision Support Systems,<br />
ieMSs Proceedings, 2002,<br />
http://www.iemss.org/iemss2002/proceedings/<br />
pdf/volume%20tre/denzer.pdf<br />
European Parliament, European Council, Directive<br />
2003/4/EC on public access to environmental<br />
information <strong>and</strong> repealing Council Directive<br />
90/313/EEC, 2003<br />
529
European Parliament, European Council, Dir.<br />
2003/35/EC, 2003a<br />
European Parliament, European Council, eEurope<br />
2002: An Information Society For All Action<br />
Plan, 2002.<br />
EU ISPO, Public Sector Information: A key<br />
resource for Europe, COM(98)585final,<br />
adopted on 20 January 1999.<br />
EWGLS-European Working Group on Libre<br />
<strong>Software</strong>, Free <strong>Software</strong> / Open Source:<br />
Information Society Opportunities for Europe,<br />
2001, http://eu.conecta.it/paper.pdf<br />
Ghosh, R. A. Underst<strong>and</strong>ing Free <strong>Software</strong><br />
Developers: Findings from the FLOSS Study<br />
http://opensource.mit.edu/papers/ghosh.pdf<br />
Haklay, M., Public Access to <strong>Environmental</strong><br />
Information: Past, Present <strong>and</strong> Future,<br />
Computers, Environment <strong>and</strong> Urban Systems,<br />
27, 163-180, 2003.<br />
Hemetsberger, A., When Consumers Produce on<br />
the Internet: The Relationship between<br />
Cognitive-affective, Socially-based, <strong>and</strong><br />
Behavioral Involvement of Prosumers, 2004,<br />
http://opensource.mit.edu/papers/hemetsberger<br />
1.pdf<br />
IDA-European Commission, DG Enterprise,<br />
Pooling Open Source <strong>Software</strong> An IDA<br />
Feasibility Study, 2002.<br />
Int. Institute of Infonomics, Berlecon Research,<br />
ProActive Int., Free/Libre <strong>and</strong> Open Source<br />
<strong>Software</strong> (FLOSS) Survey <strong>and</strong> Study, 2002,<br />
http://www.infonomics.nl/FLOSS/index.htm<br />
Kamppinen, M., P. Malaska,, M. Wilenius,<br />
Citizenship <strong>and</strong> ecological modernization in<br />
the information society.Futures 33,219–223,<br />
2001.<br />
Kimble, C, P. Hildreth, P. Wright, Knowledge<br />
Management <strong>and</strong> Business Model Innovation,<br />
Chapter 13, 220 – 234, 2001,<br />
http://www.getcited.org/pub/103396967<br />
Lave J. <strong>and</strong> E. Wenger, Situated learning.<br />
Legitimate peripheral participation Cambridge<br />
University Press, 1991,<br />
http://citeseer.ist.psu.edu/context/98865/0<br />
Lerner, J. <strong>and</strong> J. Tirole, The Simple Economics of<br />
Open Source, 2002,<br />
http://ideas.repec.org/p/nbr/nberwo/7600.html<br />
Lerner, J., <strong>and</strong> J. Tirole, The open source<br />
movement: Key research questions, European<br />
Economic Review 45, 819-826, 2001.<br />
Mitre Corporation. A Business Case Study of<br />
Open Source <strong>Software</strong>, 2002,<br />
http://www.mitre.org/<br />
work/tech_papers/tech_papers_01/kenwood_so<br />
ftware/kenwood_software.pdf<br />
Neus, A., Managing Information Quality in Virtual<br />
Communities of Practice, The 6th <strong>International</strong><br />
Conference on Information Quality at MIT,<br />
2001, http://citeseer.ist.psu.edu/567684.html<br />
Nichols, D.M., K. Thomson, S.A. Yeates,<br />
Usability <strong>and</strong> open-source software<br />
development, Proceedings of the Symposium<br />
on Computer Human Interaction, (eds.) Kemp,<br />
E., Phillips, C., Kinshuk & Haynes, J., 6 July<br />
2001, Palmerston North, New Zeal<strong>and</strong>. ACM<br />
SIGCHI New Zeal<strong>and</strong>. 49-54. ISBN: 0-473-<br />
07559-8.<br />
Perens, B., “Why Security through Obscurity<br />
Won't Work”, 2001, http://slashdot.org/<br />
features/980720/0819202.shtml<br />
Raymond, E. S., The Cathedral <strong>and</strong> the Bazaar,<br />
2000,http://citeseer.ist.psu.edu/context/66413/0<br />
Schauer, Th., The Sustainable Information<br />
Society, Visions <strong>and</strong> Risks, 2003,<br />
http://www.global-society dialogue.org/saskia.pdf<br />
Shah, S., Underst<strong>and</strong>ing the Nature of<br />
Participation & Coordination in Open <strong>and</strong><br />
Gated Source <strong>Software</strong> Development<br />
Communities<br />
Schneier, B., “Full Disclosure”, Crypto-Gram<br />
Newsleter 111, 2001, http://www.counterpane.<br />
com/crypto-gram-0111.html<br />
The Apache Foundation, Apache Web Services<br />
Project, http://ws.apache.org<br />
The Jakarta Project – Open Source Java-based<br />
tools. http://jakarta.apache.org/<br />
United Nations Development Programme, Global<br />
Public Goods Economic Briefing Paper no.3<br />
on Sustaining Our Global Public Goods, Earth<br />
Summit 2002,<br />
http://www.earthsummit2002.org/es/issues/GP<br />
G/gpg.htm<br />
Visser U., H. Stuckenschmidt, H. Wache, Th.<br />
Vögele, Using <strong>Environmental</strong> Information<br />
Efficiently: Sharing Data <strong>and</strong> Knowledge from<br />
Heterogeneous Sources, 2001,<br />
http://citeseer.nj.nec.com/309536.html<br />
Von Hippel, E. <strong>and</strong> G. Von Krogh, Open source<br />
software <strong>and</strong> the private-collective innovation<br />
model: issues for organization science.<br />
Organization Science 14 (2), 209–233, 2003.<br />
i “The effective, combined cost of acquisition <strong>and</strong><br />
deployment of an information technology<br />
throughout all its perceived useful life” [EWGLS<br />
2001]<br />
ii<br />
Earth Summit 2002, Economic Briefing Paper<br />
no.3 on Sustaining Our Global Public Goods<br />
denotes “free access to basic <strong>and</strong> essential<br />
software” as a “key policy option”.<br />
iii<br />
The 2002 <strong>and</strong> 2005 Action Plans of the<br />
European Commission’s eEurope initiative<br />
promoted FLOSS <strong>and</strong> open st<strong>and</strong>ards for use in<br />
the public sector <strong>and</strong> e-government.<br />
530
Applying agent technology in <strong>Environmental</strong><br />
Management Systems under real-time constraints<br />
I. N. Athanasiadis <strong>and</strong> P. A. Mitkas ab<br />
a Informatics <strong>and</strong> Telematics Institute, Centre for Research <strong>and</strong> Technology Hellas, GR-57001 Thermi, Greece<br />
b Department of Electrical <strong>and</strong> Computer Engineering, Aristotle University of Thessaloniki,<br />
GR-54124 Thessaloniki, Greece<br />
ionathan@iti.gr<br />
Abstract: Changes in the natural environment affect our quality of life. Thus, government, industry, <strong>and</strong> the<br />
public call for integrated environmental management systems capable of supplying all parties with validated,<br />
accurate <strong>and</strong> timely information. The ‘near real-time’ constraint reveals two critical problems in delivering<br />
such tasks: the low quality or absence of data, <strong>and</strong> the changing conditions over a long period. These problems<br />
are common in environmental monitoring networks <strong>and</strong> although harmless for off-line studies, they may be<br />
serious for near real-time systems.<br />
In this work, we discuss the problem space of near real-time reporting <strong>Environmental</strong> Management Systems <strong>and</strong><br />
present a methodology for applying agent technology this area. The proposed methodology applies powerful<br />
tools from the IT sector, such as software agents <strong>and</strong> machine learning, <strong>and</strong> identifies the potential use for<br />
solving real-world problems. An experimental agent-based prototype developed for monitoring <strong>and</strong> assessing<br />
air-quality in near real time is presented. A community of software agents is assigned to monitor <strong>and</strong> validate<br />
measurements coming from several sensors, to assess air-quality, <strong>and</strong>, finally, to deliver air quality indicators<br />
<strong>and</strong> alarms to appropriate recipients, when needed, over the web. The architecture of the developed system is<br />
presented <strong>and</strong> the deployment of a real-world test case is demonstrated.<br />
Keywords: Agent-based systems; <strong>Environmental</strong> monitoring systems; Decision support systems<br />
1 INTRODUCTION<br />
<strong>Environmental</strong> monitoring networks have been established<br />
worldwide, primarily in areas with potential<br />
pollution problems, in order to observe <strong>and</strong><br />
record the conditions of the natural environment.<br />
Through these networks, vast volumes of raw data<br />
are captured, while information systems, called <strong>Environmental</strong><br />
Management Systems (EMS), are in<br />
charge of integrating all recorded data-streams. A<br />
typical EMS installation involves the fusion into a<br />
central database of all data sensed at distributed locations.<br />
Until lately, all recorded data were meant<br />
for environmental scientists occupied with off-line<br />
studies <strong>and</strong> post-processing activities in their effort<br />
to underst<strong>and</strong> the natural phenomena involved.<br />
However, during the last few years there has been<br />
a transition in environmental monitoring systems.<br />
The aftermath of the growing societal interest for<br />
the environment <strong>and</strong> sustainable development was<br />
the emerging need for providing environmental information<br />
to the public. The challenge for EMS is<br />
to embrace the new users in the administration, industry,<br />
<strong>and</strong> the society. Unfortunately, stakeholders<br />
still hold varying interpretations of the environmental<br />
values, thus different types of information are requested<br />
by each one. In spite of their diverse needs,<br />
all users agree on the necessity to access trustworthy<br />
information on time. Near real-time identification<br />
of environmental incidents affects the response of<br />
all stakeholders <strong>and</strong> the effectiveness of prevention<br />
measures.<br />
In this paper near real-time reporting <strong>Environmental</strong><br />
Management Systems are considered, focusing<br />
on recent developments that used software agents.<br />
The “near real time” term emphasizes that such systems<br />
are capable to deliver timely information, with<br />
respect to user- or application- imposed deadlines.<br />
In the following sections, a short review of various<br />
agent-based EMS is presented <strong>and</strong> a generic<br />
methodology for applying software agent technology<br />
to this kind of applications is detailed. Finally,<br />
an experimental agent-based prototype developed<br />
for monitoring <strong>and</strong> assessing air-quality in near real<br />
time is presented.<br />
531
2 BACKGROUND<br />
2.1 <strong>Environmental</strong> Management Systems<br />
<strong>Environmental</strong> Management Systems (EMS) is a<br />
generic term used for describing structures that allow<br />
an organization to assess <strong>and</strong> control the environmental<br />
impact of its activities, products or services<br />
1 . EMS vary from systems that provide a<br />
framework for monitoring <strong>and</strong> reporting on an organization’s<br />
environmental performance (as the auditing<br />
schemes of ISO 14001 <strong>and</strong> EMAS), to systematic<br />
processes for assessing, managing <strong>and</strong> reducing<br />
environmental risk. Considering the quest for<br />
environmental information involving public, industry<br />
<strong>and</strong> administration, the challenge for EMS is to<br />
provide advanced, citizen-centered information services.<br />
To address such a challenge, <strong>Environmental</strong><br />
Informatics 2 , the research initiative examining<br />
the application of Information Technology in environmental<br />
research, monitoring, assessment, management<br />
<strong>and</strong> policy has emerged. Advances in the<br />
IT sector provide capable infrastructure for fusing<br />
knowledge into the every-day life of citizens, which<br />
is expected to lead to a new paradigm for the quality<br />
of life within the urban web, with citizen centered,<br />
environmental information services that will<br />
support societal sustainability while promoting personal<br />
well being [Karatzas et al., 2003].<br />
In the aforementioned context, EMS goals are no<br />
more fettered to integrating raw data-measurements,<br />
rather is to fuse information <strong>and</strong> diffuse knowledge,<br />
in a form comprehensible by everyone, not only the<br />
environmentalists. One could say that EMS have<br />
extended their services from simple Integration, to<br />
Assessment <strong>and</strong> Warning Services, incorporating capabilities<br />
for decision support. Following a different<br />
pathway, EMS can be categorized based on their<br />
overall goals. EMS are called to fulfill dissimilar<br />
needs, thus system goals can be classified to the following<br />
three categories:<br />
a. Off-line analysis systems. Such systems are<br />
geared towards gathering historical data in a systematic<br />
way <strong>and</strong> making them available for indepth<br />
analysis of the phenomena involved.<br />
b. Real-time reporting systems. These are systems<br />
responsible for identifying <strong>and</strong> reporting<br />
the current environmental conditions. They satisfy<br />
the public need for environmental awareness<br />
<strong>and</strong> the administrative <strong>and</strong> industrial needs<br />
for precaution measures.<br />
c. Forecasting Systems. In this case, the goal is to<br />
prognosticate the future conditions of the envi-<br />
1 Definition given by the St<strong>and</strong>ards Council of Canada.<br />
2 <strong>International</strong> Federation for Information Processing, WG 5.11:<br />
Computers <strong>and</strong> the environment, www.environmatics.org<br />
ronment, either in the long or in the short term.<br />
The need to foresee <strong>and</strong> forewarn about potential<br />
environmental problems is the key for preserving<br />
nature <strong>and</strong> taking preventive actions.<br />
Several EMS systems have been developed worldwide<br />
to realize the abovementioned goals. The evolution<br />
of EMS systems started with the off-line analysis<br />
systems, gathering information used for experimental<br />
evaluations of ecological theories. Next<br />
came the long-term forecasting systems, starting<br />
with the Climate Change Models developed in the<br />
70’s, as a response to depreciation of the environmental<br />
conditions. The last few years, public growing<br />
concern has led governments in Europe <strong>and</strong> the<br />
US to ask for real-time reporting of environmental<br />
quality. These actions are on the direction imposed<br />
by legislation (i.e. the US Clean Air Act 1990<br />
<strong>and</strong> the European Directive on Ambient Air Quality,<br />
1996). The European Directive 92/72/EEC aims<br />
to provide the public with information when warning<br />
<strong>and</strong> information threshold levels are exceeded.<br />
Thus, the “real-time” reporting EMS have emerged.<br />
2.2 <strong>Software</strong> Agent Technology<br />
This paper focuses on the design <strong>and</strong> development<br />
of EMS systems using software agent technology.<br />
Agent-Oriented <strong>Software</strong> Engineering has emerged<br />
as a novel paradigm for building software applications.<br />
The key abstraction used is that of an agent,<br />
as a software entity characterized by autonomy, reactivity,<br />
pro-activity, <strong>and</strong> social ability [Jennings,<br />
2000]. Certain types of software agents have abilities<br />
to infer rationally <strong>and</strong> support the decision making<br />
process [Jennings et al., 1998].<br />
Agent-based systems may rely on a single agent,<br />
but the advantages of this initiative are revealed in<br />
the case of Multi-Agent Systems, which consist of a<br />
community of co-operating agents. Several agents,<br />
structured in groups, can share perceptions <strong>and</strong> operate<br />
synergistically to achieve overall goals.<br />
2.3 Related Work<br />
The characteristics of agents <strong>and</strong> multi-agent systems<br />
enable them to process information <strong>and</strong> solve<br />
problems in distributed environments, as those of<br />
EMS. Thus, several agent-based EMS have been developed,<br />
in an approach to improve the performance<br />
of EMS. All these projects used agents or agentrelated<br />
techniques to achieve EMS goals <strong>and</strong> supply<br />
services, such as Integration, Assessment <strong>and</strong><br />
Warnings. In a schematic representation, (Figure<br />
1), EMS services <strong>and</strong> goals are considered as the<br />
two axes for a unified classification of agent-based<br />
developed systems.<br />
532
Agent-based EMS development is concentrated in<br />
two directions. One is the transparent integration<br />
of environmental information. Such systems are InfoSleuth<br />
[Pitts <strong>and</strong> Fowler, 2001], EDEN-IW [Felluga<br />
et al., 2003], NZDIS [Purvis et al., 2000] <strong>and</strong><br />
Kaleidoscope [Micucci, 2002]. A common practice<br />
adopted by these projects is to use software agents<br />
for distributed information processing <strong>and</strong> propagation.<br />
The second direction drives towards forecasting systems,<br />
which take advantage of the agent abilities<br />
for distributed problem solving, in an effort to provide<br />
warning services. These systems include FSEP<br />
[Dance et al., 2003] <strong>and</strong> DNEMO [Kalapanidas<br />
<strong>and</strong> Avouris, 2002], which realize intelligent agent<br />
features for the identification of environmental incidents<br />
in advance. Agent intelligence is implemented<br />
using case-based reasoning engines, regression<br />
trees, <strong>and</strong> neural networks.<br />
Agent technology was adopted by all the aforementioned<br />
projects successfully, indicating that it is a<br />
promising approach to both unify distributed information<br />
<strong>and</strong> implement warning services. Obviously,<br />
there is a gap between the two clusters of applications,<br />
which comprises the real-time reporting<br />
EMS, supporting assessment services. These systems<br />
will be discussed in the following section.<br />
3 NEAR REAL-TIME EMS<br />
3.1 Problem formulation<br />
Integrated EMS need to supply administration, industry<br />
<strong>and</strong> the public with validated, accurate information<br />
related to the environmental conditions. Human<br />
experts <strong>and</strong> scientists must have adequate data<br />
support in their efforts to assess environmental quality<br />
<strong>and</strong> issue alarms on time. The ‘near real-time’<br />
constraint unfolds the most critical problems in delivering<br />
such tasks: (a) the low quality or absence of<br />
data, <strong>and</strong> (b) the changing conditions over a long period.<br />
These problems are common in environmental<br />
monitoring networks <strong>and</strong> although harmless for offline<br />
studies, they may prove to be critical for near<br />
real-time systems.<br />
The main objective of a near real-time EMS is to<br />
provide citizen-centred Electronic Information Services,<br />
including the following:<br />
a. Acquisition of information from distributed locations,<br />
b. Information fusion <strong>and</strong> preprocessing,<br />
c. Data storage <strong>and</strong> organization,<br />
d. <strong>Environmental</strong> assessment, <strong>and</strong><br />
e. Qualitative indicators circulation over the internet<br />
Services<br />
Integration Assessment Warnings<br />
Kaleidoscope<br />
NZDIS<br />
EDEN-IW<br />
Off-line studies<br />
<strong>and</strong> analysis<br />
InfoSleuth<br />
Real-time<br />
reporting<br />
FSEP<br />
DNEMO<br />
Forecasting<br />
System Goals<br />
Figure 1: Classification of agent-based EMS<br />
The overall problem that a near real-time EMS is<br />
called to solve can be summarized as follows: Let a<br />
sensor network monitoring the environmental conditions<br />
at distributed locations. An EMS is installed<br />
over this network, capturing the sensed measurements,<br />
assessing environmental quality <strong>and</strong> delivering<br />
preprocessed information to the final users of<br />
the system, over the internet.<br />
These core activities impose the requirements for<br />
both distributed information fusion <strong>and</strong> distributed<br />
problem solving abilities. Agent success stories in<br />
both information integration <strong>and</strong> warning services<br />
need to be coupled in a common methodology.<br />
3.2 Methodology<br />
Advancing on the way earlier research work has<br />
dealt with EMS using Agent Technology, we propose<br />
a methodology for the development of a near<br />
real-time EMS as a multi-agent system (MAS). Our<br />
goal is to assign all tasks involved in the near realtime<br />
EMS operation to a software agent society.<br />
Through this approach, agents are considered as<br />
both information carriers <strong>and</strong> decision-makers.<br />
Agents as information carriers, act as a distributed<br />
community of data processing units, able to capture,<br />
manipulate <strong>and</strong> propagate information efficiently.<br />
Agents as decision-makers, behave as a network of<br />
problem-solvers that work together to reach solutions.<br />
Our integrated methodology provides a pathway,<br />
which defines both modes of agent operation<br />
in a MAS. An abstract view of our methodology is<br />
depicted in Figure 2.<br />
The starting point is to identify all the appropriate<br />
resources hidden in the application domain.<br />
In-depth underst<strong>and</strong>ing of the related domain affects<br />
the specifications of the information flows<br />
<strong>and</strong> domain knowledge. Information flows impose<br />
533
Domain<br />
Knowledge<br />
Application<br />
Domain<br />
Decision<br />
Making<br />
Distributed Problem Solving<br />
Distributed Information Fusion<br />
Information<br />
Flows<br />
Inference<br />
Models<br />
Agent<br />
<strong>Modelling</strong><br />
Synthesis<br />
Near real-time<br />
reporting EMS<br />
Integration<br />
Figure 2: An abstract view of the method<br />
how agents should manipulate data, while domain<br />
knowledge implies the decision making process incorporated<br />
into the agents. Information flows are<br />
specified through agent communication modeling,<br />
while the decision-making processes have to be<br />
transformed into agent reasoning models.<br />
In the following section, these two procedures are<br />
further explained.<br />
3.3 Agents as information carriers<br />
Modeling agents as information carriers involves<br />
four steps:<br />
Step 1. Identify system inputs <strong>and</strong> outputs.<br />
Consider the interfaces between the system <strong>and</strong> both the<br />
sensors <strong>and</strong> the end-user electronic services. Assign<br />
agents to realize these interfaces acting either as data<br />
fountains or data sinks.<br />
Step 2. Formulate information channels.<br />
Detail how information flows through the system. Specify<br />
possible data transformations needed. Assign those<br />
tasks to data management agents.<br />
Step 3. Conceptualize agent messaging.<br />
Based on the two previous steps, realize inter-agent<br />
communications for smooth information propagation.<br />
Specify the semantics of the communications using ontologies.<br />
Step 4. Specify delivery deadlines.<br />
Concrete deadlines are enjoined to agent communication,<br />
in order to ensure ‘on-time’ delivery of information.<br />
Exit on failure strategies need to be considered<br />
too.<br />
The outcome of the above procedure is materialized<br />
to the specifications of a multi-agent system architecture,<br />
in the form of:<br />
MAS = 〈A, O, I, D〉 (1)<br />
where:<br />
- A = {A 1 , . . . , A n }, is a countable set of software<br />
agents, each one of which is assigned either<br />
to introduce, manipulate or export data.<br />
- O is the domain ontology, which specifies the<br />
common vocabulary in order to represent the<br />
system environment.<br />
- I = {I k = (A i , A j )/A i , A j ∈ A}, is a set<br />
of interactions between agents. These interactions<br />
show the relations in the system organization<br />
<strong>and</strong> they allow the definition of a social<br />
framework determining the information flows in<br />
the system.<br />
- D = {D k , ∀ I k ∈ I}, is a set of the delivery<br />
deadlines assigned to each agent communication.<br />
The modeling procedure <strong>and</strong> the resulted specifications,<br />
formulated in Eq.1, define in detail the architecture<br />
<strong>and</strong> operation of a multi-agent system acting<br />
as a near real-time EMS, from an information fusion<br />
perspective. State-of-the-art methodologies for software<br />
agent modeling, as GAIA [Wooldridge et al.,<br />
2000], AUML [Odell et al., 2000], AORML [Wagner,<br />
2003] or iSTAR [Yu, 1997] are used for defining<br />
agent roles, types, protocols, <strong>and</strong> interactions.<br />
EMS critical services, such as information acquisition,<br />
preprocessing, storage <strong>and</strong> organization are<br />
organized methodically to ensure efficient, on time<br />
electronic services to the public.<br />
3.4 Agents as decision-makers<br />
Agents as decision makers are employed to deliver<br />
the reasoning abilities of the MAS. Indicatively,<br />
decision-making in a real-time EMS involves either<br />
assessment services or activities to overcome data<br />
uncertainty problems. Based on the domain knowledge,<br />
agent decision-making strategies are identified<br />
through the following procedure.<br />
Step 1. Problem formulation <strong>and</strong> decomposition.<br />
Consider the overall problem at h<strong>and</strong> <strong>and</strong> try to break it<br />
down into sub-problems.<br />
Step 2. Construction of decision points.<br />
Assign specific agents to solve each sub-problem, taking<br />
under account their resources, specified by the system’s<br />
architecture.<br />
Step 3. Decision strategy specification.<br />
For each sub-problem provide a strategy to solve it using<br />
the available resources. Consider that in a near realtime<br />
system the goal is to find the best solution possible<br />
in the available timeframe.<br />
Step 4. Realization of Inference models.<br />
Implement the decision strategies designed in the previous<br />
step as inference models of the respective agents.<br />
Inference agents will be embedded into agents as reasoning<br />
engines.<br />
This procedure is highly correlated with the application<br />
under consideration. Finding an optimal decision<br />
strategy is a rather difficult task, especially<br />
when execution time is a parameter of success.<br />
However, three distinct cases of decision-making<br />
strategies, suitable for the case discussed, can be<br />
identified:<br />
534
Case 1 Deterministic Strategies<br />
These are applied, when domain-specific, certain, explainable<br />
rules for decision-making exist. Such rules<br />
may encompass natural laws, logical rules or physical<br />
constrains. In such cases, these rules are incorporated<br />
as a static, confident, explainable expert system into the<br />
agents.<br />
Case 2 Data-driven Strategies.<br />
When historical datasets are available, the application<br />
of machine learning algorithms for knowledge discovery<br />
can yield interesting knowledge models. These<br />
models can be used for agent reasoning in a dynamic,<br />
inductive way. In EMS, there are large volumes of data<br />
continually recorded. When natural laws describing the<br />
monitored phenomena do not exist, or they are too complex,<br />
data-driven models, such as decision trees, casebased<br />
reasoning, or neural networks provide an option<br />
to the application developer. In this case, the procedure<br />
involves the creation of an inference model from historical<br />
data. This model is later incorporated into the<br />
agents.<br />
Case 3 Heuristic strategies.<br />
When neither of the above cases is applicable, heuristic<br />
models or ‘rules of thumb’ may be incorporated into<br />
agents.<br />
Following the aforementioned procedure, <strong>and</strong> having<br />
identified the appropriate decision strategies<br />
among the three cases, decision-making required by<br />
a near-real-time EMS can be realized.<br />
The combined methodology, presented in this section,<br />
provides an integrated pathway for developing<br />
a near real-time EMS using software agents.<br />
4 AN EMS FOR AIR QUALITY<br />
The methodology described in the previous section<br />
has been evaluated for the development<br />
of O 3 RTAA, a near real-time reporting EMS.<br />
O 3 RTAA is a multi-agent system for monitoring<br />
<strong>and</strong> assessing air-quality attributes, which uses data<br />
coming from a meteorological station. A community<br />
of software agents is assigned to monitor <strong>and</strong><br />
validate measurements coming from several sensors,<br />
to assess air-quality, <strong>and</strong>, finally, to fire alarms<br />
to appropriate recipients, when needed, via the Internet.<br />
Agents as information carriers undertake the following<br />
main functions of the system:<br />
A. Data collection from field sensors.<br />
B. Data management, including activities as data<br />
preprocessing, normalization <strong>and</strong> transformation.<br />
C. Information propagation, which involves posting<br />
information over the internet.<br />
Thus, system agents are organized into three groups<br />
(or layers): Contribution, Management <strong>and</strong> Distribution.<br />
Contribution Agents (CA) act as the data<br />
Sensor<br />
Network<br />
O 3 Sensor<br />
…<br />
NO Sensor<br />
X Sensor<br />
Contribution<br />
O 3 CA<br />
NO CA<br />
X CA<br />
Management<br />
Database<br />
DMA Agent<br />
Ozone Alarm<br />
DMA Agent<br />
Distribution<br />
Database<br />
DA Agent<br />
Web<br />
DA Agent<br />
O 3 RTAA System<br />
Figure 3: O 3 RTAA System Architecture<br />
End User<br />
Applications<br />
Measurements<br />
Database<br />
fountain for the system, realizing the appropriate<br />
interfaces with the sensors. Each CA is associated<br />
with a single sensor. Data Management Agents<br />
(DMA) are responsible to fuse data coming from<br />
CAs. Each DMA produces a joint view on the<br />
sensed data in the appropriate format required by<br />
the end-user applications. In O 3 RTAA two DMA<br />
agents are employed. The first is responsible for<br />
posting sensed data into a Measurements Database,<br />
for future use. The second is assigned to calculate<br />
Ozone Alarms, to be distributed over the Internet.<br />
Finally, two Distribution Agents (DA) are instantiated<br />
occupied to realize the interfaces to the enduser<br />
applications. One is in charge of the Measurements<br />
Database, while the second posts Air Quality<br />
Alarms over the Internet.<br />
The overall O 3 RTAA system architecture <strong>and</strong> the<br />
agent communication are shown in Figure 3. Intralayer<br />
communication amplifies individual agent<br />
ability to make trustworthy decisions, in a distributed<br />
AI manner. Inter-layer communication ensures<br />
the successful transfer of data to the final destination.<br />
O 3 RTAA agent messages follow a generic<br />
ontology developed using the Protégé-2000, ontology<br />
editor 3 . Agent delivery deadlines have been<br />
specified to less than a minute, while FIPA 4 compliance<br />
in agent communication ensures the proper<br />
h<strong>and</strong>ling of missing or erroneous message transmission.<br />
O 3 RTAA agent architecture is described in<br />
detail in Athanasiadis <strong>and</strong> Mitkas [2004].<br />
O 3 RTAA agents as decision-makers are responsible<br />
for the following activities:<br />
A. Incoming measurement validation, inspecting<br />
the quality of the data sensed.<br />
B. Estimation of erroneous measurements, substituting<br />
the missing values.<br />
C. Estimation of qualitative indicators, assessing<br />
the overall environmental quality.<br />
The first two activities are left to CAs, while Alarm<br />
DMA is in charge of the third. Each CA is respon-<br />
3 http://protege.stanford.edu<br />
4 Foundation of Physical Intelligent Agents, http://www.fipa.org<br />
@<br />
Web<br />
535
sible for suppling the O 3 RTAA with data coming<br />
from a particular sensor. Thus, they are assigned to<br />
validate incoming measurements <strong>and</strong> estimate the<br />
missing ones. These two activities are both realized<br />
using data-driven strategies. This comes as a<br />
consequence of the nature of these tasks, which are<br />
subject to the local conditions. Data validation <strong>and</strong><br />
missing measurement estimation activities are both<br />
related to the changing conditions over a long period.<br />
Deterministic strategies are unsuitable, while<br />
vast volumes of historical data are available. Thus,<br />
two types of data-driven reasoning engines are incorporated<br />
in each CA. One is the Validation Reasoning<br />
Engine <strong>and</strong> the second is the Estimation Reasoning<br />
Engine. Details on the specification on these<br />
Engines are given in Athanasiadis et al. [2003a, b].<br />
Estimating air quality indicators involves the application<br />
of specific thresholds, imposed by legislation.<br />
These thresholds represent a deterministic<br />
decision-making strategy. Thus, the Alarm DMA<br />
implemented an ozone alert system, by the use of a<br />
Deterministic Inference Reasoning Engine.<br />
The O 3 RTAA system has been successfully installed<br />
as a pilot case at the Mediterranean Centre<br />
for <strong>Environmental</strong> Studies Foundation (CEAM),<br />
Valencia, Spain, in collaboration with IDI-EIKON,<br />
Valencia, Spain.<br />
5 CONCLUSIONS<br />
In this paper, we presented a methodology for developing<br />
near real-time reporting EMS using software<br />
agents, <strong>and</strong> evaluated it for the development<br />
of the O 3 RTAA prototype. The benefits of this<br />
methodology are twofold: First, it can h<strong>and</strong>le data<br />
uncertainty problems through the employment of<br />
a distributed problem solving approach, employing<br />
agents that collaborate <strong>and</strong> synergistically make decisions.<br />
Secondly, it employs a distributed information<br />
processing approach, using software agents for<br />
data fusion, in order to provide at near real time,<br />
trustworthy information over the web.<br />
REFERENCES<br />
Athanasiadis, I. N. <strong>and</strong> P. A. Mitkas. An agent-based intelligent<br />
environmental monitoring system. Management<br />
of <strong>Environmental</strong> Quality, 15(3):238–249, 2004.<br />
Athanasiadis, I. N., V. G. Kaburlasos, P. A. Mitkas, <strong>and</strong><br />
V. Petridis. Applying machine learning techniques on<br />
air quality data for real-time decision support. In Information<br />
Technologies in <strong>Environmental</strong> Engineering,<br />
Gdansk, Pol<strong>and</strong>, June 2003a. ICSC-NAISO Publishers.<br />
Athanasiadis, I. N., P. A. Mitkas, G. B. Laleci, <strong>and</strong><br />
Y. Kabak. Embedding data-driven decision strategies<br />
on software agents: The case of a multi-agent system<br />
for monitoring air-quality indexes. In Concurrent<br />
Engineering: The Vision for the Future Generation<br />
in Research <strong>and</strong> Applications, Madeira, Portugal,<br />
July 2003b. Balkema Publishers.<br />
Dance, S., M. Gorman, L. Padgham, <strong>and</strong> M. Winikoff. An<br />
evolving multi agent system for meteorological alerts.<br />
In Proc. of the second international joint conference<br />
on Autonomous agents <strong>and</strong> multiagent systems, pages<br />
966–967. ACM Press, 2003.<br />
Felluga, B. et al. <strong>Environmental</strong> data exchange for inl<strong>and</strong><br />
waters using independed software agents. Report<br />
20549 EN, Institute for Environment <strong>and</strong> Sustainability,<br />
European Joint Research Centre, Ispra, Italy, April<br />
2003.<br />
Jennings, N. R., K. Sycara, <strong>and</strong> M. J. Wooldridge. A<br />
roadmap of agent research <strong>and</strong> development. In Autonomous<br />
Agents <strong>and</strong> Multi-Agent Systems 1, pages 7–<br />
38, Boston, 1998. Kluwer Academic Publishers.<br />
Jennings, N. R. On agent-based software engineering.<br />
Artificial Intelligence, 117:277–296, 2000.<br />
Kalapanidas, E. <strong>and</strong> N. Avouris. Air quality management<br />
using a multi-agent system. <strong>International</strong> Journal of<br />
Computer Aided Civil <strong>and</strong> Infrastructure Engineering,<br />
17:119–130, 2002.<br />
Karatzas, K., A. Bassoukos, A. Kaprara, A. Masouras,<br />
T. Slini, J. Papaioannou, <strong>and</strong> N. Moussiopoulos. <strong>Environmental</strong><br />
informatics <strong>and</strong> urban air quality management<br />
systems. In 5th <strong>International</strong> Symposium on <strong>Environmental</strong><br />
<strong>Software</strong> Systems (ISESS 2003), volume 5<br />
of <strong>Environmental</strong> <strong>Software</strong> Systems, pages 134–146,<br />
Semmering, Austria, 2003. IFIP.<br />
Micucci, D. Exploiting the kaleidoscope architecture in<br />
an industrial environmental monitoring system with<br />
heterogeneous devices <strong>and</strong> a knowledge-based supervisor.<br />
In Proc. of the 14th international conference<br />
on <strong>Software</strong> engineering <strong>and</strong> knowledge engineering,<br />
pages 685–688. ACM Press, 2002.<br />
Odell, J., H. van Dyke Parunak, <strong>and</strong> B. Bauer. Extending<br />
UML for agents. In Proc. of the 2nd Int. Workshop<br />
on Agent-Oriented Information Systems, Berlin, Germany,<br />
2000. iCue Publishing.<br />
Pitts, G. <strong>and</strong> J. Fowler. InfoSleuth: An emerging technology<br />
for sharing distributed environmental information.<br />
In Information Systems <strong>and</strong> the Environment, pages<br />
159–172. National Academy Press, 2001.<br />
Purvis, M., S. Cranefield, <strong>and</strong> M. Nowostawski. A<br />
distributed architecture for environmental information<br />
systems. In <strong>Environmental</strong> <strong>Software</strong> Systems-<br />
<strong>Environmental</strong> Information <strong>and</strong> Decision Support,<br />
pages 49–56. Kluwer Academic Publishers, 2000.<br />
Wagner, G. The Agent–Object–Relationship metamodel:<br />
Towards a unified conceptual view of state <strong>and</strong> behavior.<br />
Information Systems, 28(5):475–504, 2003.<br />
Wooldridge, M., N. R. Jennings, <strong>and</strong> D. Kinny. The<br />
Gaia methodology for agent-oriented analysis <strong>and</strong> design.<br />
Autonomous Agents <strong>and</strong> Multi-Agent Systems, 3<br />
(3):285–312, 2000.<br />
Yu, E. Towards modelling <strong>and</strong> reasoning support for<br />
early-phase requirements engineering. In Proc. of the<br />
3rd IEEE Int. Symp. on Requirements Engineering,<br />
Washington, USA, 1997. IEEE.<br />
536
Supporting the Strategic Objectives of Participative<br />
Water Resources Management; an Evaluation of the<br />
Performance of Four ICT Tools<br />
Swinford, A., McIntosh, B., Jeffrey, P.<br />
a.swinford@cranfield.ac.uk<br />
School of Water Sciences, Cranfield University. Cranfield. Beds. MK43 0AL.<br />
Article 14 of the Water Framework Directive promotes a social learning model of participative planning <strong>and</strong><br />
creates a broader stakeholder <strong>and</strong> public constituency for water management. Such natural resource<br />
management processes are key testing grounds for the development of new Information <strong>and</strong><br />
Communication Technology (ICT) tools designed to support wider citizen participation in local <strong>and</strong><br />
regional governance. Several types of purpose designed ICT tool are available, but there is a distinct lack of<br />
empirical research into their design <strong>and</strong> effectiveness. Strategic objectives performance take the central role<br />
in the work reported here. Six strategic objectives of the use of ICT tools were identified; learning, trust in<br />
the institution (the developers of the tool), trust in the computer tool (<strong>and</strong> the information contained within),<br />
trust in the decisions made (during a post interaction scenario), motivation <strong>and</strong> inclusion. A number of preexisting<br />
software platforms, each specifically designed to either educate or support decision making in the<br />
area of water management, were selected <strong>and</strong> formally evaluated under controlled conditions with small<br />
groups of evaluators. Results from the evaluation sessions were analysed using statistical analysis<br />
techniques. The discussion focus is primarily on the performance of each evaluated tool with respect to<br />
achieving the strategic objectives.<br />
Keywords: Public participation; ICT tools; Design; Evaluation<br />
1. INTRODUCTION<br />
1.1 The need for citizen inclusion<br />
Implementation of the Water Framework<br />
Directive, specifically Article 14 saw the first<br />
steps towards initiating a two way flow of<br />
information <strong>and</strong> decision making with regards to<br />
water management. According to the European<br />
Commission, this pan-European piece of<br />
legislation was proposed due to pressure from<br />
environmental organisations <strong>and</strong> citizens for<br />
cleaner water resources. Therefore the EC took<br />
upon itself to make the remediation of polluted<br />
water bodies <strong>and</strong> the safeguarding of such areas a<br />
priority (European Union, 2000). The<br />
involvement of organisations <strong>and</strong> citizen groups<br />
was considered to be essential to ensure that the<br />
EC would achieve these objectives.<br />
Following initial ideas on public participation as<br />
presented in Agenda 21 (UNCED, 1992), the<br />
Aarhus convention (CEC, 2003) <strong>and</strong> the Water<br />
Framework Directive (European Union, 2000),<br />
the European Commission (EC) proposed via a<br />
White Paper the ‘opening up’ of the policy<br />
making process, whereby the involvement of<br />
members of the public in ‘shaping <strong>and</strong> delivering’<br />
EU policy would take place (CEC, 2001a).<br />
Reforming European governance requires the<br />
commitment of European member states, regional<br />
<strong>and</strong> local authorities <strong>and</strong> citizens. To determine<br />
good governance, five political principles were<br />
devised which included Openness (in terms of<br />
communicating information to citizens),<br />
Participation (involving citizens would increase<br />
confidence in any final decisions reached by the<br />
EC), Accountability (Member states take<br />
responsibility for their actions), effectiveness (of<br />
polices) <strong>and</strong> coherence (of policies <strong>and</strong> actions).<br />
537
1.2 Bridging the knowledge gap<br />
It is outlined in the legislation (European Union,<br />
2000) that in future, organisations will have to<br />
involve members of the public in decision making<br />
regarding the environment. Perhaps<br />
underst<strong>and</strong>ably organisations may be a little<br />
reluctant to involve members of the public in<br />
decision making regarding issues for which they<br />
have no expert training or prior knowledge of.<br />
Therefore it is imperative that the correct level of<br />
decision making power <strong>and</strong> the most suitable<br />
participative fora are selected based on the<br />
environmental issue to be discussed. A number of<br />
papers (House, 1999; Konisky <strong>and</strong> Beierle, 2001;<br />
Aldred <strong>and</strong> Jacobs, 2000,) have dealt with the<br />
level of decision making power afforded to the<br />
public in a decision a making situation, but<br />
Arnstein was the first to develop a ladder of<br />
citizen participation (Figure 1, Arnstein, 1969).<br />
8. Citizen Control<br />
7. Delegated Power<br />
6. Partnership<br />
5. Placation<br />
4. Consultation<br />
3. Informing<br />
2. Therapy<br />
1. Manipulation<br />
Degrees of<br />
citizen<br />
power<br />
Degrees of<br />
tokenism<br />
Non<br />
participation<br />
Figure 1 – Arnstein’s ladder of citizen<br />
participation<br />
When considering Arnstein’s ladder with respect<br />
to the Water Framework Directive, the degree of<br />
citizen participation stated is unclear. However,<br />
the wording contained within Article 14 of the<br />
Water Framework Directive (European Union,<br />
2000) implies that the level of public participation<br />
(according to Arnstein’s ladder) will manifest<br />
itself either in the form of a partnership or<br />
delegated power. It is unlikely that the level of<br />
citizen power implied in the article is meant to<br />
exist in the realms of tokenism, although there is a<br />
possibility that a subsequent decline to this level<br />
may occur. It is equally unlikely that the article is<br />
actually stating that the public involved should<br />
have complete control, as this could lead to<br />
citizens making ill-informed decisions with<br />
regard to water resource management. In their<br />
2002 paper, Moorhouse <strong>and</strong> Ellif addressed the<br />
benefits of involving members of the public in<br />
decision making, reasoning that inclusion would<br />
reduce uneasiness between experts <strong>and</strong> non<br />
experts.<br />
1.3 The Need for ICT Tools to support<br />
participative processes<br />
So that members of the public are able to interact<br />
successfully <strong>and</strong> make meaningful decisions<br />
regarding water environment issues, certain tools<br />
have been identified which facilitate the decision<br />
making process. Other than the availability of<br />
obvious reference aids such as books or<br />
television, these include ICT (Information <strong>and</strong><br />
communication technology) tools which can be<br />
designed for use to allow citizens to fill a<br />
knowledge gap, or in the form of decision support<br />
tools, which present the user with options<br />
concerning a specific problem or environmental<br />
issue (e.g. Water Ware, Jamieson <strong>and</strong> Fedra,<br />
1996). Other suggested tools include the use of<br />
metaphors (Cronje, 2001) <strong>and</strong> scenarios (Van<br />
Nottes et al, 2003).<br />
In order to enable natural resource management<br />
processes to take place it is widely advocated that<br />
there is a need for the development of new<br />
Information <strong>and</strong> Communication Technology<br />
tools (ICT) specifically to support participative<br />
management tools (Guimãres Pereira, et al. 2003).<br />
Such tools exist, but there is a distinct lack of<br />
research into the design performance,<br />
effectiveness <strong>and</strong> intended use of such tools. As<br />
limited work has been carried out on the design<br />
<strong>and</strong> evaluation of tools specifically designed to<br />
facilitate decision making processes it is<br />
important to first define possible areas that can be<br />
evaluated within an ICT tool. Within this<br />
research, three main areas have been identified to<br />
focus on in terms of evaluation, which are<br />
elements of the Human Computer Interface<br />
(HCI), the deployment context <strong>and</strong> finally the<br />
presence of certain strategic objectives.<br />
2 EVALUATION RESEARCH<br />
Whilst the HCI <strong>and</strong> deployment context are<br />
clearly of influence <strong>and</strong> are important focuses for<br />
study, it is the strategic objectives that we are<br />
concerned with here. Our motive for focusing so<br />
clearly on the strategic objectives of ICT tools is<br />
that strategic objectives constitute the avowed<br />
utility or benefit of engaging stakeholders in the<br />
first place. Without demonstrable<br />
accomplishment of the strategic objectives of ICT<br />
tool deployment, the whole process becomes<br />
somewhat notional <strong>and</strong> speculative. Our challenge<br />
is therefore to identify a set of strategic outcomes<br />
which ICT tools designed to support participative<br />
processes should be achieving. What is the nature<br />
of the change or transition which users of the tool<br />
will undergo? How will their opinions,<br />
538
perspectives, underst<strong>and</strong>ings, or knowledges be<br />
modified / enhanced?<br />
These strategic functions are, in fact, described<br />
reasonably well in the literature. However, a first<br />
principles approach should start with a set of<br />
reasons why wider participation in natural<br />
resource planning <strong>and</strong> management is desirous.<br />
Table 1 provides a suggested set of such reasons.<br />
From the strategic objectives listed in the second<br />
column of Table 1 we can list a preliminary set of<br />
aspirations for participative planning processes;<br />
Learning, Trust, Motivation, Inclusion,<br />
Consensus, Justice, <strong>and</strong> Openness. These strategic<br />
functions or objectives of participation constitute<br />
a set of objectives for not only the participation<br />
process, but also for tools <strong>and</strong> techniques<br />
designed to support such processes.<br />
Strategic objectives of Keywords<br />
participative processes<br />
Better solutions <strong>and</strong><br />
Efficiency<br />
deployment strategies can<br />
be identified.<br />
All interested parties are Fairness<br />
provided with<br />
opportunities to contribute<br />
<strong>and</strong> engage in debate.<br />
Collaboration supports Knowledge<br />
elicitation of both expert pooling<br />
<strong>and</strong> local knowledge.<br />
The bringing together of Trust<br />
members of the public <strong>and</strong><br />
experts can help dispel the<br />
general mistrust of science<br />
that non-experts might<br />
possess.<br />
All parties are aware of the<br />
issues <strong>and</strong> the process by<br />
Transparency of<br />
process<br />
which decisions are made.<br />
Confidence in decisions is Trust / Fairness<br />
likely to be enhanced<br />
under conditions where<br />
inclusiveness <strong>and</strong> openness<br />
are promoted.<br />
Wider participation Consensus<br />
provides opportunities for<br />
broader agreement on<br />
diagnosis, prognosis, <strong>and</strong><br />
solution selection.<br />
Wider participation meets Democracy<br />
the ambitions of<br />
governance principles<br />
based on extending<br />
democracy<br />
Broadening the Inclusion<br />
constituency being<br />
consulted creates wider<br />
ownership of the issue.<br />
Participative processes<br />
result in outcomes which<br />
are considered fairer or<br />
less discriminatory.<br />
Wider participation<br />
provides opportunities for<br />
information <strong>and</strong><br />
knowledge acquisition <strong>and</strong><br />
for social learning.<br />
Justice<br />
Learning<br />
Table 1: Derivation of strategic functions<br />
3. EVALUATION METHODOLOGY<br />
3.1 Platform <strong>and</strong> Respondent Selection<br />
As stated previously existing tools designed<br />
specifically to promote citizen participation or<br />
empowerment are extremely sparse <strong>and</strong> therefore<br />
limited work has been carried out to analyse the<br />
interactions between the user <strong>and</strong> interface <strong>and</strong><br />
more importantly whether tools developed<br />
achieve certain strategic objectives. Early on in<br />
the investigation the decision was made to only<br />
include tools developed within the UK, as it was<br />
felt that it would be unfair to ask residents of the<br />
UK questions regarding unfamiliar locations.<br />
Therefore it was decided that the platforms to be<br />
used in the evaluation would be those that focus<br />
on locations in the UK, so therefore would be<br />
developed by organisations situated in the UK. It<br />
was also decided that the tools should specifically<br />
focus on water related environmental issues. The<br />
aforementioned factors warranted consideration<br />
as they would particularly effect the trust<br />
questions to be asked during the evaluations.<br />
Asking an individual whether they trust an<br />
organisation, or the content within a tool<br />
developed by an organisation requires that the<br />
respondents must at least have had the chance to<br />
find out about or hear of the company in question,<br />
for example the Environment Agency.<br />
The platforms selected for the evaluation included<br />
the Riverside Explorer (Environment Agency),<br />
Ecopod (Environment Agency), The Water Aid<br />
Game (Water Aid) <strong>and</strong> the Personal Barometer<br />
(Cranfield University).<br />
As most of the pre-existing tools mentioned<br />
above were designed for students aged between<br />
10-16 years it was decided that this target<br />
audience should be involved in the testing of the<br />
platforms. To be able to carry out the evaluation<br />
with student respondents it was decided that<br />
contact should be made with different secondary<br />
539
schools in the Bedfordshire area. As well as<br />
including both the developers <strong>and</strong> target audience<br />
in the evaluations, a further respondent group was<br />
involved. Postgraduates from Cranfield<br />
University were also asked to volunteer to take<br />
part in the evaluation work. Both groups<br />
(postgraduates <strong>and</strong> target audience) took part in<br />
the evaluation because designing future tools to<br />
aid participatory process would require that<br />
individuals of all ages <strong>and</strong> ability would need to<br />
be able to use the tools.<br />
Both time <strong>and</strong> monetary constraints limited the<br />
number of evaluation sessions that took place <strong>and</strong><br />
therefore the number of participants that took part<br />
in the investigation. The length of time it took to<br />
plan the sessions reduced time to actually carry<br />
out the sessions. Also as a financial reward was<br />
offered to any postgraduate volunteers willing to<br />
take part in the session. This was a predetermined<br />
amount, therefore limiting the number of<br />
respondents who could take part.<br />
3.2 Evaluation techniques<br />
After much deliberation as to the best way to<br />
discover strategic objective presence in each tool,<br />
it was decided that questions related to each of the<br />
strategic objectives would be asked both before<br />
<strong>and</strong> after interaction with the specific platform.<br />
Both the pre <strong>and</strong> post interaction questions were<br />
exactly the same so that the respondent’s prior<br />
knowledge <strong>and</strong> opinions could be gauged both<br />
before <strong>and</strong> following platform use. This would<br />
also enable a direct comparison between answers<br />
to the two sets of questions pre <strong>and</strong> post platform<br />
use.<br />
A self complete questionnaire was designed to<br />
ask questions relating to the strategic objectives<br />
both before <strong>and</strong> after interaction. A platform<br />
specific scenario was also proposed to the group<br />
before <strong>and</strong> after interaction with the tool. The<br />
format of the evaluation sessions was as follows:<br />
1. The administering of a pre interaction self<br />
complete questionnaire.<br />
2. The discussion of a platform specific pre<br />
interaction scenario (to be taped).<br />
3. Interaction with the tool<br />
4. The administering of a post interaction self<br />
complete questionnaire (Same wording as pre<br />
interaction questionnaire).<br />
5. The discussion of a platform specific post<br />
interaction scenario (to be taped).<br />
4 RESULTS<br />
Data collected from a total of 21 sessions<br />
(involving a total of 69 respondents) are from the<br />
evaluations involving the target audience (10-16<br />
year olds) <strong>and</strong> postgraduate volunteers. This small<br />
sample therefore means that significance testing<br />
would not be very robust or meaningful. This<br />
combined data is represented in Figures 2-6.<br />
Figure 2 shows the total percentage strategic<br />
objectives achieved for each platform tested.<br />
Strategic objecitves achieved (%)<br />
100<br />
80<br />
60<br />
40<br />
20<br />
0<br />
Total strategic objectives achieved per platform (%)<br />
Ecopod Riverside Explorer Water Aid Game Personal Barometer<br />
Platform<br />
Figure 2 – Total percentage strategic objectives<br />
tested.<br />
Through analysis of the individual strategic<br />
objectives achieved by each platform, the degree<br />
to which each strategic objective was achieved<br />
could be observed. Figures 3-6 present the<br />
strategic objectives achieved for each platform.<br />
Key<br />
TII – Trust in the institution<br />
TID – Trust in the decisions made<br />
TICT – Trust in the computer tool<br />
Strategic objectives achieved (%)<br />
Strategic objectives achieved (%)<br />
100<br />
80<br />
60<br />
40<br />
20<br />
0<br />
Strategic objectives - Ecopod<br />
Learning TII Motivation Inclusion TID TICT<br />
Strategic objecitves<br />
Figure 3 – Strategic objectives achieved by<br />
Ecopod (Environment Agency, 2002).<br />
100<br />
80<br />
60<br />
40<br />
20<br />
0<br />
Strategic objecitves - Riverside Explorer<br />
Learning TII Motivation Inclusion TID TICT<br />
Strategic objectives<br />
Figure 4 – Strategic Objectives achieved by The<br />
Riverside Explorer (Environment Agency, 2002)<br />
540
Strategic objectives achieved (%)<br />
Strategic objectives achieved (%)<br />
100<br />
80<br />
60<br />
40<br />
20<br />
0<br />
Strategic objecitves achieved - Water Aid Game<br />
Learning TII Motivation Inclusion TID TICT<br />
Strategic objectives<br />
Figure 5 – Strategic Objectives achieved by the<br />
Water Aid Game (Water Aid, 1999)<br />
100<br />
80<br />
60<br />
40<br />
20<br />
0<br />
Strategic Objectives achieved - Personal Barometer<br />
Learning TII Motivation Inclusion TID TICT<br />
Strategic objectives<br />
Figure 6 – Strategic Objectives achieved by the<br />
Personal Barometer (Cranfield University, 2003)<br />
5 DISCUSSION<br />
By looking at Figure 2 it can be seen that the<br />
Ecopod application achieved the highest overall<br />
percentage of strategic objectives, with a mean<br />
score of 73%. This was followed by the Water<br />
Aid game achieving a score of 71%, the Personal<br />
Barometer achieving 66% <strong>and</strong> finally the<br />
Riverside Explorer achieving a score of 63%.<br />
Ecopod, designed by the Environment Agency<br />
gained the highest overall score during the<br />
evaluation, meaning that it achieved the total<br />
strategic objectives to the highest degree.<br />
However, from the results it can be seen that there<br />
is only a 10% difference between the highest <strong>and</strong><br />
lowest scoring tools, so therefore it is necessary to<br />
consider the degree to which individual tools<br />
achieved strategic objectives.<br />
With all tools, learning was the strategic objective<br />
that was achieved to the lowest degree, which<br />
suggests that particular attention needs to be<br />
focussed on this area when designing a generic<br />
evaluation methodology <strong>and</strong> in future tool design.<br />
The low score for learning for all tools implies<br />
that although they are designed for learning <strong>and</strong><br />
even if they possess all of the relevant<br />
information regarding the subject area, the<br />
respondents have failed to answer questions<br />
regarding a major learning goal within the tool.<br />
This could be because of the way in which the<br />
information is presented within each tool, perhaps<br />
it was difficult for the user to navigate the tool.<br />
This finding has implications for the future design<br />
of computer tools for learning, especially those<br />
used in a decision support context.<br />
The results for the strategic objectives vary<br />
according to each tool evaluated. Beginning with<br />
Ecopod, the joint highest strategic objectives<br />
achieved were ‘inclusion’ <strong>and</strong> ‘trust in the<br />
computer tool’. The second highest jointly, were<br />
‘motivation’ <strong>and</strong> ‘trust in the decisions made’.<br />
The strategic objective ‘trust in the institution’<br />
scored poorly. When looking at the results from<br />
the Riverside Explorer, the strategic objective<br />
achieved to the highest degree was ‘trust in the<br />
institution’, followed by ‘inclusion’ <strong>and</strong> then<br />
jointly by ‘motivation’ <strong>and</strong> ‘trust in the computer<br />
tool’. Results suggest that the tool did not help the<br />
respondents gain confidence in the decision<br />
making scenario, therefore the strategic objective<br />
‘trust in the decisions made’ received a low score.<br />
Adversely, ‘trust in the decision’ was the strategic<br />
objective to be achieved to the highest degree<br />
during the evaluation of the Water Aid Game<br />
tool, followed by ‘inclusion’. This tool achieved<br />
the strategic objectives ‘trust in the institution’,<br />
‘motivation’ <strong>and</strong> ‘trust in the computer tool’ to<br />
the same degree. Finally, ‘inclusion’ was<br />
achieved to the highest degree when the Personal<br />
Barometer was evaluated, followed by<br />
‘motivation’ <strong>and</strong> then jointly by ‘trust in the<br />
computer tool’ <strong>and</strong> ‘trust in the institution’. ‘Trust<br />
in the decisions made’ was the poorest scoring<br />
objective.<br />
From the evaluation sessions it was found that<br />
‘inclusion’ was the easiest objective to achieve<br />
overall. During the questionnaire the respondents<br />
were asked whether a certain environmental issue<br />
(for example, world drought) was a problem that<br />
they thought that they should be concerned with.<br />
It was found that following tool interaction most<br />
respondent’s opinions had changed <strong>and</strong> the tool<br />
demonstrated that it was important for them to<br />
consider the issue. When asked whether they<br />
would get involved in helping solve an<br />
environmental issue affecting their local area,<br />
most respondents answered in a positive way<br />
following tool use. However, the strategic<br />
objectives related to trust varied greatly, the<br />
respondents trusted the computer tool, but when<br />
asked if they trusted the institution that developed<br />
the tool, the responses depended on whether the<br />
respondents had heard of the institutions in the<br />
first place. This would therefore be affected by<br />
age (children would be less likely to be concerned<br />
with environmental matters) or duration of<br />
residency in the UK. Finally the most varied<br />
strategic objective score was trust in decisions<br />
made during the scenario section of the<br />
evaluation. It would seem that after using the tool<br />
541
the respondents did not feel confident about the<br />
decision that they had made aided by the tool.<br />
6 CONCLUSIONS<br />
• Learning was the poorest strategic objective<br />
achieved <strong>and</strong> special attention needs to be<br />
focussed on this in future tool development.<br />
• Inclusion was the highest strategic objective<br />
achieved.<br />
• Respondents tended to be more motivated<br />
following tool use.<br />
• The elements of trust were varied.<br />
7 ACKNOWLEDGEMENTS<br />
The work on which this paper is based was<br />
supported by the European Commission under the<br />
VIRTU@LIS project (IST-2000-28121). The<br />
authors would also like to thank the respondents<br />
who took part in the evaluation study.<br />
8 REFERENCES<br />
Aldred, J. <strong>and</strong> Jacobs, M. (2000) Citizens <strong>and</strong><br />
Wetl<strong>and</strong>s: evaluating the Ely citizens' jury.<br />
Ecological Economics 34 217-232.<br />
Arnstein, S.J. (1969) A ladder of citizen<br />
participation. Journal of the environmental<br />
institute of planners 35 216-224.<br />
CEC (Commission of the European<br />
Communities).(2003)Directive 2003/35/EC<br />
of the European Parliament <strong>and</strong> of the<br />
Council providing for public participation<br />
in respect of the drawing up of certain<br />
plans <strong>and</strong> programmes relating to the<br />
environment <strong>and</strong> amending with regard to<br />
public participation <strong>and</strong> access to justice<br />
Council Directives 85/337/EEC <strong>and</strong><br />
96/61/EC.<br />
CEC (Commission of the European Communities<br />
(2001a)) European Governance: A White<br />
Paper; Com (2001) 428, Brussels,<br />
27.7.2001, available at:<br />
http://europa.eu.int/comm/governance/area<br />
s/index_en,htm.<br />
Cranfield University, 2003. Personal Barometer<br />
[CD-ROM] Copyright [2004, Feb. 25]<br />
Cronje, J.C., (2001) Metaphors <strong>and</strong> model based<br />
learning. Computers <strong>and</strong> Education, 3. pp<br />
241-256.<br />
Environment Agency, 2002. ECOPOD. [Online],<br />
Available:http://www.environmentagency.<br />
gov.uk/yourenv/education/111357/?version<br />
=1&lang=_e [2002] Crown copyright<br />
[2002, Sept. 23].<br />
Environment Agency, 2000. Riverside Explorer.<br />
[CD-ROM] Crown copyright, [2002, Sept.<br />
23].<br />
European Union. (2000) Water European Union<br />
(2002) Guidance on the public<br />
participation in relation to the Water<br />
Framework Directive: Active involvement,<br />
Consultation <strong>and</strong> Public access to<br />
information. EC WFD Common<br />
Implementation Strategy. Working Group<br />
on Public Participation. Framework<br />
Directive. 2000/60/EC.<br />
Guimãres Pereira, Â. Rinaudo, J.D., Jeffrey, P.<br />
Blasques, J. Corral Quintana, S. Courtois,<br />
N. Funtowicz, S. & Petit, V. (2003) ICT<br />
tools to support public participation in<br />
water resources governance <strong>and</strong> planning:<br />
Experiences from the design <strong>and</strong> testing of<br />
a multi-media platform. Journal of<br />
<strong>Environmental</strong> Policy, Assessment <strong>and</strong><br />
Management. 5 (3), 395-420.<br />
House, M. (1999) Citizen Participation in Water<br />
Management. Water, Science <strong>and</strong><br />
Technology 40 (10), 125-130.<br />
Konisky, D.M. <strong>and</strong> Beierle, T.C. (2001)<br />
Innovations in Public Participation <strong>and</strong><br />
<strong>Environmental</strong> Decision-making:<br />
Examples from the Great Lakes Region.<br />
Society <strong>and</strong> Natural Resources 14 815-826.<br />
Jamieson, D.G <strong>and</strong> Fedra, K (1996) The<br />
‘WaterWare’ decision-support systems for<br />
river basin planning. 1. Conceptual design.<br />
Journal of Hydrology. 177, pp163-175.<br />
Moorhouse, M. <strong>and</strong> Ellif, S. (2002) Planning<br />
process for public participation in regional<br />
water resources planning. Journal of the<br />
American Water Resources Association.<br />
38(2), 531.<br />
UNCED (1992) Agenda 21, United Nations<br />
conference on Environment <strong>and</strong><br />
Development (UNCED), Rio de Janeiro,<br />
Brazil.<br />
Van Nottes, P.W.F., Rotmans, J., Van Asselt,<br />
M.B.A., Rothman, D.S., (2003) An<br />
environmental issues with interactive<br />
information <strong>and</strong> communication<br />
technologies.’ Contract # IST-2000-28121<br />
Water Aid. 1999. Water Aid Game [Online],<br />
Available:http://www.wateraid.org/learn_z<br />
one/1114/default.asp [1999] Crown<br />
copyright [2004, Feb. 25].<br />
542
Web Services for <strong>Environmental</strong> Informatics<br />
Erick Arauco a<br />
<strong>and</strong> Lorenzo Sommaruga b<br />
a<br />
University of Piura - Engineering Department ,Piura, Perú- earauco@udep.edu.pe<br />
b<br />
University of Applied Sciences of Southern Switzerl<strong>and</strong> - Innovative Technologies Department (DTI),<br />
Switzerl<strong>and</strong> – lorenzo.sommaruga@supsi.ch<br />
Abstract: This paper presents the description of an open architecture for the management of environmental<br />
content using Web Services. The Web Services technology can be effectively exploited for integrating on one<br />
h<strong>and</strong> the needs for dissemination of analytical data about environment, such as air, noise, traffic, etc., <strong>and</strong> on<br />
the other h<strong>and</strong> the needs of different users concerning the accessibility requirements of their devices,<br />
distributed <strong>and</strong> heterogeneous systems, remote <strong>and</strong> mobile control access. The case study of this paper is<br />
based on the OASI (<strong>Environmental</strong> Observatory of Southern Switzerl<strong>and</strong>) project that permits access to the<br />
air, noise <strong>and</strong> traffic measures for Southern Switzerl<strong>and</strong>. Other than through traditional web pages, this access<br />
is also made possible thanks to the deployment of software applications based on Web Services. To this aim,<br />
a number of Web Services are defined using the UML analisys <strong>and</strong> developed. They identify a) a group<br />
h<strong>and</strong>ling the air information, b) a group implementing the access to the noise information <strong>and</strong> c) a group<br />
implementing the access to the traffic information using the OASI’s database. In addition, another group<br />
allows users to manage his/her own information, e.g. e-mail, OASI’s news, user’s group, etc. In conclusion,<br />
the Web Services Technology could be a good solution for the management of environmental content because<br />
it provides open <strong>and</strong> mobile access to data, interoperability among different client-server nodes, easy<br />
extensibility for integrating any kind of device into the system.<br />
Keywords: Web Services; mobile; J2ME; WAP; SOAP<br />
1. Introduction<br />
The Web Services technology<br />
[W3CWebServices, ApacheWebServices] can be<br />
effectively exploited for integrating on one h<strong>and</strong><br />
the needs for dissemination of analytical data about<br />
environment such as air, noise, traffic, etc., <strong>and</strong> on<br />
the other h<strong>and</strong> the needs of different users<br />
concerning the accessibility requirements of their<br />
devices, distributed <strong>and</strong> heterogeneous systems,<br />
remote <strong>and</strong> mobile control access.<br />
The main advantage of Web Services is to<br />
offer an open architecture for any type of client in<br />
a simple way. In fact, using Web Services any<br />
client can access the same environmental<br />
information independently from its platform,<br />
language, <strong>and</strong> above all device.<br />
This paper describes relevant studies <strong>and</strong><br />
experiments concerning a web services-based<br />
extension to the OASI (<strong>Environmental</strong> Observatory<br />
of Southern Switzerl<strong>and</strong>) project [Andretta et al.,<br />
2004] in order to support the access of the air,<br />
noise <strong>and</strong> traffic measures for Southern<br />
Switzerl<strong>and</strong> from any device. To this aim, the<br />
application context is firstly introduced, followed<br />
by a presentation of Web Services <strong>and</strong> a<br />
description of an open architecture for the<br />
management of the environmental contents using<br />
Web Services.<br />
2. OASI Project<br />
The OASI project originates from the risk of<br />
the environmental degradation caused by the<br />
increasing traffic or by the constructions of new<br />
buildings. In this situation, the Southern<br />
Switzerl<strong>and</strong> local government (Cantone Ticino)<br />
needs tools to monitor <strong>and</strong> control this risk.<br />
The main goals of the OASI project are:<br />
543
• To control the effects on the environment<br />
within 20 years;<br />
• To guarantee a management <strong>and</strong> a modern<br />
access to information;<br />
• To offer in real time important information to<br />
the authorities <strong>and</strong> the population.<br />
The system is based on the observation <strong>and</strong><br />
collection in a database of information about<br />
traffic, emission, <strong>and</strong> their effects.<br />
The previous solution is offering internet or<br />
intranet access to this information through an<br />
architecture consisting of a Java (st<strong>and</strong>-alone) or<br />
an Internet browser client, a web server <strong>and</strong><br />
application server (Tomcat [Tomcat]), <strong>and</strong> an<br />
Oracle database, as shown in Figure 1.<br />
Figure 1. Initial OASI Architecture.<br />
This architecture is presenting limitations on<br />
the data accessibility which depend on different<br />
user’s devices, interoperability issues of various<br />
distributed <strong>and</strong> heterogeneous systems <strong>and</strong> the<br />
need for remote <strong>and</strong> mobile control access.<br />
Within this context the Web Services<br />
technology has been considered.<br />
3. Previous Works<br />
Some research works related to this project<br />
support as well the demonstration of the feasibility<br />
of a Web Services based approach <strong>and</strong> the constant<br />
search of a st<strong>and</strong>ard architecture for client-server<br />
communication based on this technology.<br />
Three relevant projects can be mentioned<br />
within the context of the present study concerning<br />
environmental data management <strong>and</strong> Web<br />
Services. Dwyer <strong>and</strong> Clark [2002] in “Web<br />
Services Implementation: The Beta Phase of EPA<br />
Network Nodes” describe an introduction to XML<br />
<strong>and</strong> SOAP, <strong>and</strong> how they are used in the U.S.<br />
<strong>Environmental</strong> Protection Agency architecture to<br />
generate node services requests <strong>and</strong> to protect the<br />
data exchanges. The APNEE - TU project<br />
[Karatzas] has been designed to offer<br />
environmental information to different users. The<br />
information can be accesed from diverse channels<br />
like J2ME, WAP, SMS, internet, voice servers, etc.<br />
through an informatic portal that provides real-time<br />
information; <strong>and</strong> the MINNE project [MINNE] is a<br />
research project in the area of mobile computing<br />
for collecting, reporting, <strong>and</strong> delivering<br />
<strong>Environmental</strong> Information. The aim of the project<br />
is to find new ways to access the enviromental<br />
information, specially through mobile technology.<br />
4. Web Services<br />
Web Services are a technology that permits<br />
the integration of heterogeneous systems into a<br />
neutral platform. They can be considered as an<br />
interface which describes a set of operations made<br />
accessible on the network through st<strong>and</strong>ard XML<br />
messages [Sommaruga, 2003].<br />
Their main characteristics can be summarized<br />
in the following points:<br />
• Information interchanges with other web<br />
services<br />
• Accessible across many protocols such as<br />
HTTP, SMTP, etc.<br />
• Based on st<strong>and</strong>ard XML (Extensible<br />
Markup Language) languages<br />
• Neutral platform, i.e. they do not depend<br />
on any platform<br />
• Offering compatibility of heterogeneous<br />
systems.<br />
St<strong>and</strong>ard Technology Purpose<br />
[XML]<br />
[SOAP]<br />
[UDDI]<br />
[WSDL]<br />
Extensible<br />
Markup<br />
Language<br />
Simple Object<br />
Access Protocol<br />
Universal<br />
Discovery,<br />
Description <strong>and</strong><br />
Integration<br />
Web Services<br />
Description<br />
Language<br />
Content’s<br />
definition<br />
language<br />
Communication<br />
protocol based on<br />
XML<br />
Public directory<br />
service that offers<br />
information about<br />
web services<br />
Protocol that<br />
describe web<br />
service abilities<br />
Table 1. Web Services main st<strong>and</strong>ards.<br />
Other st<strong>and</strong>ards on which the Web Services<br />
technology is based are described in Table 1.<br />
In particular, SOAP is a protocol based on<br />
XML that defines the message format. It is<br />
independent of the platform, programming<br />
language, <strong>and</strong> device, allowing in this way a large<br />
interoperability.<br />
544
How a web service generally works is<br />
presented in the following Figure 2.<br />
An application via a SOAP client retrieves<br />
service information by asking to UDDI registry for<br />
Application<br />
Soap client<br />
1<br />
UDDI<br />
request<br />
2<br />
Figure 2. Web Services working flow.<br />
4<br />
response<br />
Soap<br />
3 Service<br />
processor<br />
On the basis of this information, the client can<br />
communicate with the SOAP service requesting by<br />
means of a SOAP message (2) the execution of an<br />
operation. Once the service is executed (3), the<br />
result is then passed back to the requesting client in<br />
the form of a SOAP XML message (4).<br />
5. A Web Services based open architecture<br />
One of the goals of the project was to offer a<br />
st<strong>and</strong>ard for the client-server communication <strong>and</strong><br />
to extend its use on wireless devices.<br />
To this aim, an architecture based on the web<br />
services <strong>and</strong> the tree-tiers model [Sun] has been<br />
adopted. In this architecture all the client-server<br />
communication is based on XML <strong>and</strong> SOAP. On<br />
this basis, any client that underst<strong>and</strong>s SOAP can be<br />
added as an interface (see below the Presentation<br />
level).<br />
The basic principle is the possibility to wrap<br />
the functionalities the OASI system has to offer<br />
into web services <strong>and</strong> to transform each client<br />
application into a web services SOAP client.<br />
In this way each actor in the system on the<br />
server <strong>and</strong> on the client, i.e. data providers <strong>and</strong><br />
data consumers, can “speak” the same language,<br />
underst<strong>and</strong>ing each other <strong>and</strong> therefore<br />
interoperate in an effective way.<br />
The three-tier model allows the functionalities<br />
of a distributed system to be separated into three<br />
layers or levels:<br />
• Data level; where data are stored for<br />
instance in a database server.<br />
• Logical level; where the web services are<br />
defined <strong>and</strong> operate. Here it is supported by<br />
an application server <strong>and</strong> a Web services<br />
SOAP listener.<br />
a necessary specific service; the UDDI registry<br />
searches for the required service <strong>and</strong> gives its<br />
description to the client (1).<br />
• Presentation level; where all the user<br />
interaction occurs. It may consist of any<br />
client that needs to access the information.<br />
In our system it has been developed in the<br />
form of a WML (Wap) user client, a more<br />
advanced Java (J2ME) user interface, <strong>and</strong><br />
an SMS client via a GSM line.<br />
The separation of these levels is detailed in<br />
Figure 3. From this picture emerges the clear<br />
separation of the data from its processing <strong>and</strong> use<br />
in the logical level <strong>and</strong> particularly from a number<br />
of different clients which can independently access<br />
the data. It is worth noting that the access to the<br />
data is controlled exclusively in the logical level<br />
<strong>and</strong> it is interfaced to the client in a uniform way<br />
by means of SOAP messages.<br />
For instance, the structure of a SOAP message<br />
for requesting some data from the “AirService”<br />
<strong>and</strong> getting a response are in a st<strong>and</strong>ard easy to<br />
read format, like in the following sample message<br />
excerpts where two measured values are returned<br />
(170 <strong>and</strong> 150), for the time 8:00:00 <strong>and</strong> 7:00:00<br />
respectively as underlined in the messages.<br />
Sample Request Message<br />
<br />
<br />
<br />
1<br />
21-07-03<br />
PM10<br />
ug/m3<br />
<br />
<br />
Sample Response Message<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
A002<br />
8:00:00<br />
170<br />
545
<br />
A001<br />
7:00:00<br />
150<br />
<br />
<br />
<br />
Figure 3. Web services based open architecture.<br />
The graphic corresponding to these request<br />
<strong>and</strong> response messages could be elaborated <strong>and</strong><br />
displayed in client, such as a Java enabled mobile<br />
device as later presented in Figure 5.<br />
6. Implementation<br />
This section describes how Web Services has<br />
been implemented in the OASI project prototype.<br />
The project has been primarily a feasibility<br />
experiment within the main OASI project for<br />
testing the use of the web services technology in<br />
order to provide more mobility <strong>and</strong> accessibility to<br />
its environmental data.<br />
A UML analysis has been firstly carried out<br />
for the specification of Web Services. This analysis<br />
allowed us to identify how the logical level can<br />
function. In this point, a number of Web Services<br />
are defined <strong>and</strong> developed. They identify a) a<br />
group h<strong>and</strong>ling the air information, b) a group<br />
implementing the access to the noise information<br />
<strong>and</strong> c) a group implementing the access to the<br />
traffic information using the OASI’s database. In<br />
addition, another group allows users to manage<br />
his/her own information, e.g. e-mail, OASI’s news,<br />
user’s group, etc.<br />
In a second phase, it has been developed <strong>and</strong><br />
implemented using real devices that supports<br />
J2ME <strong>and</strong> WAP technology.<br />
J2ME [J2ME] is a Sun’s Java developing<br />
platform for programming applications dedicated<br />
to mobile devices which support the java virtual<br />
machine. Such typical devices include many<br />
mobile phone models of the last generation.<br />
The J2ME language is based on the midlet<br />
concept. A Midlet is a portable programming code<br />
that consists of a particular java class. This code<br />
can be downloaded via a data-phone connection<br />
into the mobile device <strong>and</strong> can be run as a local<br />
application by the local virtual machine.<br />
In order to simplify the development phase,<br />
emulators for J2ME <strong>and</strong> WAP clients have been<br />
exploited in our project. A final testing has been<br />
accomplished on real mobile devices such as Nokia<br />
Java enabled phones.<br />
All users that have a J2ME phone or WAP<br />
phone can have access to the information via Web<br />
Services, although it is important to note that these<br />
services may have a cost, <strong>and</strong> these costs depend<br />
546
on the mobile operators in relation to the bytes<br />
transmitted.<br />
The present project can be helpful because it<br />
shows how the same information can be delivered<br />
on diverse formats (text, graphic, etc).<br />
Currently, this experimental phase has been<br />
completed <strong>and</strong> has allowed the architectural <strong>and</strong><br />
technological feasibility to be validated. In a future<br />
phase, a plan for the real deployment <strong>and</strong><br />
integration of the Web Services approach into the<br />
OASI environment will be evaluated by the main<br />
project’s responsible (Canton Ticino).<br />
The specific languages <strong>and</strong> technologies used<br />
for the system implementation are here<br />
summarized.<br />
A database Server Oracle 9i was used for<br />
retrieving the data in the Data level. The web<br />
services of the logical level were developed using<br />
the Soap Server Axis 1.1 on Apache Tomcat 4.1.27<br />
as the application server (i.e. servlet container) on<br />
Java SE 1.4.03 <strong>and</strong> also as the Wap Server. In the<br />
Presentation level, the Java MicroEdition (J2ME)<br />
<strong>and</strong> its extension for supporting Mobile<br />
Information Device Profile (MIDP 2.0) have been<br />
used on the Java enabled mobile clients<br />
additionally supported by KSoap on J2ME for the<br />
Soap Client. For the other type of clients<br />
considering Wap devices, the Wireless Application<br />
Language (WML) was used supported by Axis 1.1.<br />
The next pictures show samples of the<br />
interaction of both a J2ME <strong>and</strong> a WAP client<br />
within the OASI system.<br />
Figure 4. Selection of the type of data to be shown<br />
in a J2ME Client.<br />
Figure 5. Example of Air graphic in a J2ME<br />
client.<br />
In Figure 4, a menu in a J2ME Client is<br />
offering the possibility of a selection of the type of<br />
data to be shown. The result of this selection, <strong>and</strong> a<br />
following definition of a time range, is shown in<br />
Figure 5, where it is possible to observe a statistic<br />
graphic which permits users to analyze the air’s<br />
measurements in the Canton Ticino – Switzerl<strong>and</strong>.<br />
547
Figure 6. Air information measures displayed in a<br />
J2ME Client.<br />
Figure 7. Example of PM10 (ug/m 3 ) Air pollution<br />
variation in a J2ME Client.<br />
Figure 6, presents the graphic interface of a<br />
J2ME client for requesting air measurements in a<br />
given place at a specified date.<br />
An example of PM10 (ug/m 3 ) Air pollution<br />
variation over one day, previously specified, with a<br />
time interval of 1 hour, is displayed in a J2ME<br />
Client (Figure 7).<br />
Figure 8. One of the OASI functionality menus for<br />
the interaction in a Wap Client.<br />
An initial interaction in a Wap Client<br />
concerning the possibility of selection of various<br />
functionalities is presented in Figure 8. This<br />
selection permits the presentation of environmental<br />
information <strong>and</strong> graphics, user data management<br />
(profile information, login, <strong>and</strong> email information<br />
<strong>and</strong> settings), <strong>and</strong> some news.<br />
7. Conclusions<br />
A presentation of an innovative application of<br />
the Web Services technology to the environmental<br />
field has been presented. An open architecture for<br />
the management of the environmental contents<br />
using Web Services has been introduced. This<br />
architecture presents a number of advantages with<br />
respect to traditional systems <strong>and</strong> solutions,<br />
including:<br />
• The system can be accessed from anywhere,<br />
i.e. open <strong>and</strong> mobile access to data, both for<br />
accessing <strong>and</strong> for administering them.<br />
• The possibility of achieving a st<strong>and</strong>ard way<br />
for client-server communication based on<br />
SOAP <strong>and</strong> XML.<br />
• The possibility of integrating any kind of<br />
device into the architecture without<br />
touching or modifying the underlying<br />
logical <strong>and</strong> data level.<br />
This architecture could be exp<strong>and</strong>ed <strong>and</strong><br />
successfully applied to other similar domain<br />
problems, where it will be possible to easily<br />
integrate heterogeneous systems via Web Services<br />
abstractions. Using Web Services any client can<br />
access the same environmental information<br />
independently from its platform, language, <strong>and</strong><br />
above all device.<br />
The architecture proposed was implemented in<br />
particular for mobile clients that support J2ME or<br />
WAP technology, where the environment data can<br />
be displayed <strong>and</strong> presented to the end user in a<br />
textual or graphical format according to the<br />
specific device profile.<br />
The development of this system architecture<br />
can be summarized in three steps according to the<br />
three-tier levels:<br />
1) Data level: representing the information,<br />
i.e. the environmental data, for instance using a<br />
DB, as in our case.<br />
2) Logical level: building the main web<br />
service(s) for interfacing the data <strong>and</strong> exposing the<br />
accessible functionalities (WSDL’s )<br />
to be used by the clients.<br />
548
3) Presentation level: developing the various<br />
kinds of client, wherever <strong>and</strong> whenever necessary,<br />
as we did for J2ME, Wap <strong>and</strong> SMS.<br />
A number of issues also emerged in the<br />
project, including low capacity for processing the<br />
information or limited display in the mobile<br />
devices. However, we hope that in the future these<br />
technical problems will be resolved by the rapid<br />
evolution of technology.<br />
In conclusion, the OASI’s Mobile Web<br />
Services have demonstrated that the OASI’s team<br />
may use the Web Services Technology to exchange<br />
environmental data. Future research will allow us<br />
to extend the project to other devices like SMS<br />
phones through a gateway implementation.<br />
6. Acknowledgements<br />
The authors wish to thank the University of<br />
Applied Sciences of Southern Switzerl<strong>and</strong> <strong>and</strong><br />
Canton Ticino for their kind collaboration <strong>and</strong> for<br />
providing us with all available data.<br />
Special thanks to RETECA Foundation which<br />
allowed Engineer Erick Arauco to complete his<br />
Masters in Advanced Computer Science studies.<br />
The work described in this project has been<br />
accomplished in partial fulfilment of Arauco’s<br />
Master Diploma in Advanced Computer Science at<br />
SUPSI during year 2003 [Arauco].<br />
7. References<br />
ApacheWebServices, Apache Web Services<br />
Project, URI: http://ws.apache.org<br />
Arauco, E., Servizi Mobili per il Monitoraggio<br />
Ambientale del Progetto Oasi. University of<br />
Applied Sciences of Southern Switzerl<strong>and</strong><br />
Manno, Switzerl<strong>and</strong>, Oct. 2003.<br />
J2ME, Java 2 Platform, Micro Edition (J2ME),<br />
2003 URI: http://java.sun.com/j2me/<br />
Andretta, M., G. Bernasconi, G. Corti <strong>and</strong> R.<br />
Mastropietro, OASI: An Integrated<br />
multidomain Information System, iEMSs<br />
2004, The <strong>International</strong> <strong>Environmental</strong><br />
<strong>Modelling</strong> <strong>and</strong> <strong>Software</strong> Society Conference,<br />
14-17 June 2004, University of Osnabrück,<br />
Germany, URI:<br />
http://www.iemss.org/iemss2004/<br />
Dwyer, C., Clark Chris, Nobles M., Web Services<br />
Implementation: The Beta Phase of EPA<br />
Network Nodes. <strong>International</strong> Emission<br />
Inventory Conference "Emission Inventories -<br />
Partnering for the Future," Atlanta, GA,<br />
April 15-18, 2002, URI:<br />
http://www.epa.gov/ttn/chief/conference/ei11/<br />
datamgt/dwyer.pdf<br />
Karatzas, K., Moussiopoulos, N., Providing timely<br />
<strong>and</strong> valid air quality information via human –<br />
centered electronic services: The Apnee<br />
Project. Aristotle University of Thessaloniki,<br />
Greece, URI:<br />
http://www.edie.net/library/features/ENB046.<br />
HTML<br />
MINNE Mobile <strong>Environmental</strong> Information<br />
Systems project, University of Oulu, Finl<strong>and</strong>,<br />
URI: http://www.minne.oulu.fi<br />
SOAP, Web services SOAP, URI:<br />
http://ws.apache.org/soap/<br />
Sommaruga, L., Web Services <strong>and</strong> SOAP,<br />
Postgraduate Course, University of Applied<br />
Sciences of Southern Switzerl<strong>and</strong>, 2003 URI:<br />
http://www.macs.supsi.ch/<br />
Sun, Web-Tier Application Framework Design,<br />
2002, URI:<br />
http://java.sun.com/blueprints/guidelines/desi<br />
gning_enterprise_applications_2e/webtier/web-tier5.html<br />
Tomcat, The Jakarta Site – Apache Tomcat URI:<br />
http://jakarta.apache.org/tomcat<br />
UDDI, UDDI.org URI: http://www.uddi.org/<br />
W3CWebServices, W3C’s Web Services Activity,<br />
URI: www.w3.org/2002/ws/<br />
WSDL, Web Services Description Language<br />
(WSDL) 1.1, W3C Note 15 March 200,<br />
URI: http://www.w3.org/TR/wsdl<br />
XML, Extensible Markup Language (XML)URI:<br />
http://www.w3.org/XML/<br />
549
Concepts of Decision Support for River Rehabilitation<br />
P. Reichert, M. Borsuk, M. Hostmann, S. Schweizer, C. Spörri, K. Tockner <strong>and</strong> B. Truffer<br />
Swiss Federal Institute for <strong>Environmental</strong> Science <strong>and</strong> Technology (EAWAG)<br />
8600 Dübendorf, Switzerl<strong>and</strong><br />
Abstract: River rehabilitation decisions, like other decisions in environmental management, are often taken<br />
by authorities without sufficient transparency about how different goals, outcomes, <strong>and</strong> concerns were<br />
considered during the decision making process. This can lead to lack of acceptance or even opposition by<br />
stakeholders. In this paper, a concept is outlined for the use of techniques of decision analysis to structure<br />
scientist <strong>and</strong> stakeholder involvement in river rehabilitation decisions. The main elements of this structure<br />
are (i) an objectives hierarchy that facilitates explicit discussion of goals, (ii) an integrative probability<br />
network model for the prediction of the consequences of rehabilitation alternatives, <strong>and</strong> (iii) a mathematical<br />
representation of preferences for possible outcomes elicited from important stakeholders. This structure<br />
leads to transparency about expectations of outcomes by scientists <strong>and</strong> valuations of these outcomes by<br />
stakeholders <strong>and</strong> can be used (i) to analyse synergies <strong>and</strong> conflict potential between stakeholders, (ii) to<br />
analyse the sensitivity of alternative-rankings to uncertainty in prediction <strong>and</strong> valuation, <strong>and</strong> (iii) as a basis<br />
for communicating the reasons for the decision. These analyses can be expected to stimulate the creation of<br />
alternatives with a greater degree of consensus among stakeholders. The paper concentrates on the overall<br />
concept, the objectives hierarchy <strong>and</strong> the design of the integrative model. More details about the integrative<br />
model, the stakeholder involvement process, <strong>and</strong> the assessment of results will be published separately.<br />
Because many decisions in environmental management are characterized by a complex scientific problem<br />
<strong>and</strong> diverse stakeholders, the outlined methodology will be easily transferable to other settings.<br />
Keywords: decision analysis; stakeholder involvement; river rehabilitation.<br />
0. INTRODUCTION<br />
In many industrialized countries, river ecosystems<br />
have been strongly impacted over the past<br />
centuries, mainly by constraining their widths to<br />
gain agricultural l<strong>and</strong> <strong>and</strong> improve flood<br />
protection of cultivated <strong>and</strong> urban l<strong>and</strong>. River<br />
rehabilitation has the goal to reestablish part of<br />
these ecosystems. Decisions about measures of<br />
river rehabilitation are difficult because of the<br />
uncertainty about the outcomes, the number of<br />
stakeholders with partly conflicting objectives,<br />
<strong>and</strong> the difficult <strong>and</strong> time consuming governmental<br />
decision procedure.<br />
Decision analysis techniques [von Winterfeldt<br />
<strong>and</strong> Edwards, 1986; Clemen, 1996; Eisenführ und<br />
Weber, 2003] were originally developed to<br />
support individual decision makers. However,<br />
because these techniques are used to structure the<br />
decision problem <strong>and</strong> to make explicit expectations<br />
about outcomes <strong>and</strong> preferences, they can<br />
also be used to support group decisions or to<br />
structure stakeholder involvement <strong>and</strong> communication<br />
about reasons for decisions. The potential<br />
of these <strong>and</strong> other multiple criteria decision<br />
support methods is of interest for environmental<br />
decision making [Lahdelma et al, 2000].<br />
In this paper, we describe a general procedure of<br />
how decision analysis techniques can beneficially<br />
be used to support river rehabilitation decisions.<br />
The procedure is divided into seven steps:<br />
1. definition of the decision problem;<br />
2. identification of objectives <strong>and</strong> attributes;<br />
3. identification <strong>and</strong> pre-selection of alternatives;<br />
4. prediction of outcomes;<br />
5. quantification of preferences of stakeholders<br />
for outcomes;<br />
6. ranking of alternatives;<br />
7. assessment of results.<br />
These seven steps are briefly described in<br />
sections 1-7 of this manuscript in the context of<br />
decisions about river rehabilitation measures for a<br />
particular river reach. The problem of integrative<br />
planning of river rehabilitation in the context of<br />
the whole river basin is not addressed in this<br />
paper.<br />
550
1. DEFINITION OF THE DECISION<br />
PROBLEM<br />
Definition of the decision problem consists of<br />
identification of ecological deficits of the river<br />
reach <strong>and</strong> of stakeholders involved in or affected<br />
by the decision [Hostmann et al., 2004].<br />
2. OBJECTIVES AND ATTRIBUTES<br />
2.1 Objectives<br />
An objective is something a decision maker (or<br />
stakeholder) would like to achieve. Objectives<br />
can be divided into fundamental objectives<br />
(directly related to what a decision maker would<br />
like to achieve) <strong>and</strong> means objectives (lead to the<br />
accomplishment of fundamental objectives).<br />
Fundamental objectives are usually structured<br />
hierarchically according to their degree of concreteness<br />
[Clemen, 1996; Eisenführ <strong>and</strong> Weber,<br />
2003]. The objectives at each level of such a<br />
hierarchy should be mutually exclusive <strong>and</strong><br />
collectively exhaustive [Keeney, 1992].<br />
Figure 1 provides a hierarchy of fundamental objectives<br />
for a rehabilitated river reach which can<br />
serve as a guideline for value assessments in river<br />
rehabilitation projects. This hierarchy was developed<br />
by scientist involved in the multidisciplinary<br />
Rhone-Thur project for scientific support of<br />
river rehabilitation projects in Switzerl<strong>and</strong> [Peter<br />
et al. 2004]. It served as a basis for the value<br />
assessments by all stakeholder groups (there was<br />
no request for additional objectives when using a<br />
simplified version of this hierarchy for value<br />
assessments).<br />
At the first level, the overall objective is divided<br />
into the objectives of achieving l<strong>and</strong>scape integrity<br />
<strong>and</strong> socio-economic well-being.<br />
L<strong>and</strong>scape integrity is further divided into ecosystem<br />
integrity <strong>and</strong> hydrogeomorphic integrity.<br />
It is obvious that, due to the important influence<br />
of river hydrology <strong>and</strong> morphology on the development<br />
of the ecosystem, we run into difficulty<br />
distinguishing means objectives from fundamental<br />
objectives <strong>and</strong> with having mutually exclusive<br />
objectives in this branch of the objectives<br />
hierarchy. Alternatives would be to either concentrate<br />
on ecosystem integrity <strong>and</strong> treat hydrogeomorphic<br />
integrity as a means objective to<br />
achieve ecosystem integrity, or to concentrate on<br />
hydrogeomorphic integrity <strong>and</strong> assume that this is<br />
sufficient to guarantee ecosystem integrity.<br />
Neither of these approaches is satisfying. The<br />
first does not account for achieving hydrogeomorphic<br />
integrity as a fundamental objective,<br />
while the second omits ecosystem integrity as an<br />
important (or even the most important) fundamental<br />
objective. This does not imply that hydrogeomorphic<br />
attributes are not useful for quantifying<br />
the means objective of achieving ecosystem<br />
integrity. To account for the difficulties outlined<br />
above, we decided to use both ecosystem integrity<br />
<strong>and</strong> hydrogeomorphic integrity as fundamental<br />
objectives. The difficulty of this approach is that,<br />
when characterizing the preference structure, we<br />
have to assign values to hydrogeomorphic<br />
integrity excluding its benefits to ecosystem<br />
integrity, to keep the objectives mutually<br />
exclusive (otherwise we would double-count the<br />
value of ecosystem integrity).<br />
Ecosystem integrity is divided into natural<br />
ecosystem function <strong>and</strong> natural species diversity.<br />
At this level we again have problems of specifying<br />
mutually exclusive objectives as the species<br />
are a determinant of ecosystem function. Still, it<br />
seems necessary to distinguish between a function<br />
provided by a small number of species or by a<br />
diverse ecosystem.<br />
Hydrogeomorphic integrity is divided into<br />
natural river morphology, natural discharge<br />
regime, <strong>and</strong> good water quality.<br />
The branch socio-economic well-being is divided<br />
into ensuring ecosystem services, low<br />
implementation cost, <strong>and</strong> guaranteeing job<br />
opportunities. The objective of ensuring ecosystem<br />
services guarantees that society benefits<br />
from the ecosystem. Low implementation cost<br />
helps the society affording implementation of the<br />
measures. Guaranteeing job opportunities,<br />
particularly in agriculture, is an important<br />
objective of stakeholders.<br />
Further details are represented by the lower level<br />
objectives in Figure 1.<br />
2.2 Attributes<br />
An attribute is a measurable quantity that can be<br />
used to quantify the degree of fulfilment of an<br />
objective. The lowest level objectives of the<br />
hierarchy are characterized by the attributes listed<br />
at the right-h<strong>and</strong> side of Figure 1. In some cases,<br />
these attributes can easily be used to quantify the<br />
degree of fulfilment of the corresponding<br />
objective. However, in other cases, the chosen<br />
attributes are a compromise between a good<br />
characterisation of the objective <strong>and</strong> a reasonable<br />
expected prediction accuracy.<br />
3. ALTERNATIVES<br />
Important options for rehabilitation of river sections<br />
are widening the river bed, lowering the<br />
floodplains, <strong>and</strong> construction of retention basins<br />
or side channels. Decision alternatives typically<br />
consist of combinations of these measures. In<br />
many cases, loosening river width constraints is<br />
the most important measure for rehabilitation.<br />
551
4. PREDICTION OF OUTCOMES<br />
The outcomes of rehabilitation measures are difficult<br />
to predict. As rehabilitation measures usually<br />
directly affect the shape of the river bed, the most<br />
direct consequences consist of hydraulic <strong>and</strong><br />
morphological changes. These then have consequences<br />
on the benthic population, fish,<br />
vegetation, <strong>and</strong> shoreline community. In addition,<br />
they have social <strong>and</strong> economic consequences.<br />
These relationships are visualized in Fig. 2.<br />
Nat. ecosystem<br />
function<br />
Nat. population<br />
stability<br />
Functioning<br />
organic cycles<br />
Density of<br />
refugia<br />
Mean primary<br />
productivity<br />
Number of nat.<br />
tributaries p.r.l.<br />
Mean leaf decomposition<br />
rate<br />
Mean respiration<br />
rate<br />
Ecosystem<br />
integrity<br />
Natural benthic<br />
community div.<br />
Mean density<br />
of algae<br />
Mean density of<br />
grazers & collect.<br />
Mean density<br />
of shredders<br />
Mean density<br />
of predators<br />
Natural fish<br />
diversity<br />
Abundance<br />
of trout<br />
Abundance<br />
of barbel<br />
Nat. species<br />
diversity<br />
Abundance<br />
of nase<br />
Natural vegetation<br />
diversity<br />
Area of pioneer<br />
vegetation p.r.l.<br />
Area of soft wood<br />
vegetation p.r.l.<br />
Area of hard wood<br />
vegetation p.r.l.<br />
Area of gravel<br />
bars p.r.l.<br />
Natural shoreline<br />
community div.<br />
Mean density of<br />
carabid beetles<br />
Mean density<br />
of spiders<br />
L<strong>and</strong>scape<br />
integrity<br />
Mean density<br />
of ants<br />
Natural morphological<br />
type<br />
Morphological<br />
type<br />
Natural spatial<br />
variability<br />
Coef. of variation<br />
of water depth<br />
Coef. of variation<br />
of flow velocity<br />
Nat. river<br />
morphology<br />
High connectivity<br />
(long., lat., vert.)<br />
Length of<br />
connected reach<br />
Fraction of natural<br />
river banks<br />
Ratio of bank<br />
to river length<br />
Fraction of fine<br />
sediments<br />
Natural floodplain<br />
morphology<br />
Average<br />
floodplain width<br />
Natural flood<br />
characteristics<br />
Discharge of<br />
annual flood<br />
Number of delam.<br />
floods per season<br />
Hydrogeomorphic<br />
integrity<br />
Nat. discharge<br />
regime<br />
Natural low<br />
water periods<br />
Average<br />
discharge.<br />
5% fractile of<br />
discharge distrib.<br />
Rehabilitated<br />
river section<br />
Absence of artificial<br />
fluctuations<br />
Natural temperature<br />
regime<br />
Amplitude <strong>and</strong><br />
period of fluct.<br />
Long term/seasonal<br />
temp. change<br />
Rate of increase/decrease<br />
Amplitude of short<br />
term temp. fluct.<br />
Low suspended<br />
sediment conc.<br />
Mean susp. sed.<br />
conc. at low disch.<br />
Good water<br />
quality<br />
High oxygen<br />
concentration<br />
Min. dissolved<br />
oxygen conc.<br />
Natural nutrient<br />
concentrations<br />
Mean phosphate<br />
concentration<br />
Mean inorg.<br />
nitrogen conc.<br />
Low pollutant<br />
concentrations<br />
Mean metal<br />
concentrations<br />
Mean organic<br />
pollutant conc.<br />
High flood retention<br />
capacity<br />
Expected<br />
damage cost<br />
Retention<br />
volume p.r.l<br />
Ensuring ecosystem<br />
services<br />
Good w. quant./<br />
qual. at gw. wells<br />
High self-purification<br />
capacity<br />
Groundwater recharge<br />
rate p.r.l.<br />
Reaeration<br />
coefficient<br />
Groundwater<br />
infilt./transp. time.<br />
High recreational<br />
value<br />
Area of accessible<br />
gravel bars p.r.l.<br />
Socio-economic<br />
well-being<br />
Low implem. cost<br />
(constr., maint)<br />
Low construction<br />
cost<br />
Low maintenance<br />
cost<br />
Construction cost<br />
per year, p.r.l<br />
Maintenance cost<br />
per year, p.r.l.<br />
Guaranteeing job<br />
opportunities<br />
Guaranteeing<br />
agricultural jobs<br />
Guaranteeing<br />
non-agri. jobs<br />
Change in no. of<br />
agricult. jobs<br />
Change in no. of<br />
non-agri. jobs<br />
Figure 1.<br />
Objectives hierarchy (rectangular boxes <strong>and</strong> lines) <strong>and</strong> attributes (rhombic boxes) corresponding to the<br />
lowest-level objectives for a rehabilitated river section (p.r.l. = per unit river length).<br />
552
River hydraulics<br />
<strong>and</strong> morphology<br />
Figure 2.<br />
Fish<br />
Rehabilitation measures<br />
Benthic population<br />
Vegetation<br />
Socio-economic<br />
consequences<br />
Shoreline community<br />
Important relationships between conesquences<br />
of river rehabilitation measures.<br />
Prediction of consequences of rehabilitation<br />
measures requires a model of cause-effect relationships.<br />
Such a model must combine knowledge<br />
from all available sources such as basic<br />
scientific knowledge, specialized literature, more<br />
detailed models, measured data, <strong>and</strong> expert knowledge.<br />
Probability network models provide a very<br />
useful model structure to combine different types<br />
of knowledge, to divide a model into more easily<br />
tractable sub-models, <strong>and</strong> to explicitly consider<br />
prediction uncertainty [Pearl, 1988; Charniak,<br />
1991; Reckhow, 1999; Borsuk et al., 2004]. This<br />
is the reason why we recommend building the<br />
integrative model of cause-effect relationships as<br />
a probability network model. Fig. 3 visualizes the<br />
most important cause-effect relationships between<br />
external forcings <strong>and</strong> all attributes identified in<br />
Fig. 1 <strong>and</strong> how those relationships are divided<br />
into the six sub-models of hydraulics, benthic<br />
population, fish, vegetation, shoreline<br />
community, <strong>and</strong> economics. Brief descriptions of<br />
how these six sub-models are constructed are<br />
given in the following six subsections. More<br />
detailed descriptions of all sub-models will be<br />
published separately.<br />
4.1 Hydraulic Sub-Model<br />
The hydraulics sub-model predicts river morphology,<br />
gravel transport, velocity <strong>and</strong> depth distribution,<br />
<strong>and</strong> river bed clogging [Schweizer et al.,<br />
2004]. It is based on an analysis of natural channel<br />
morphology predicted by one of the relationships<br />
derived by Bledsoe <strong>and</strong> Watson [2001] <strong>and</strong><br />
considers width constraints with the aid of da<br />
Silva’s [1991] analyses. Prediction of velocity<br />
distributions are based on Lamouroux [1995], <strong>and</strong><br />
of river bed clogging on Schälchli [1993].<br />
4.2 Benthic Population Sub-Model<br />
The benthic population sub-model consists of a<br />
simplified approach relative to dynamic river<br />
benthos models [McIntire, 1973; Rutherford,<br />
1999]. It estimates seasonal benthic population<br />
densities based on the most important influence<br />
factors affected by rehabilitation measures.<br />
4.3 Fish Sub-Model<br />
In the fish sub-model, the dependence of the parameters<br />
of a fish population model on external<br />
influence factors is formulated, <strong>and</strong> then the fish<br />
population model is solved dynamically. The<br />
results are summarized by a probability network<br />
[Lee <strong>and</strong> Rieman, 1997; Borsuk et al., 2002].<br />
4.4 Vegetation Sub-Model<br />
The vegetation model maps the response surface<br />
of a mechanistic individual-based floodplain<br />
vegetation model [Prentice et al., 1993] using a<br />
probability network.<br />
4.5 Terrestrial Shoreline Fauna Sub-Model<br />
This sub-model is based on a simple quantification<br />
of the empirical relationship between environmental<br />
driving variables <strong>and</strong> population density<br />
<strong>and</strong> species identity of carabid beetles,<br />
spiders, <strong>and</strong> ants [Boscaini et al., 2000].<br />
4.6 Economic Sub-Model<br />
The economic sub-model quantifies the effects of<br />
the revitalisation work on the local economy, <strong>and</strong><br />
uses changes in the number of jobs as a proxy. It<br />
is built as an input-output model [Miller <strong>and</strong><br />
Blair, 1985] that is integrated into the probability<br />
network model formalism. This type of model<br />
uses an input-output table between different sectors<br />
of the economy to derive technical coefficients<br />
through division by the sectoral outputs. It<br />
then assumes that these technical coefficients do<br />
not change <strong>and</strong> calculates the change in sectoral<br />
activities <strong>and</strong> employment for the dem<strong>and</strong> change<br />
in the construction <strong>and</strong> other involved sectors<br />
during implementation of the rehabilitation<br />
measures. The underlying input-output table is<br />
constructed by adapting the national input-output<br />
table based on local employment statistics<br />
(location quotient method, Isard et al. [1998]).<br />
4.7 Integrative Model<br />
The complete model combining all sub-models<br />
can be used for decision support among alternatives.<br />
For detailed planning of river construction<br />
required for implementing the chosen alternative,<br />
more detailed investigations may be necessary.<br />
5. PREFERENCES FOR OUTCOMES<br />
Stakeholder preferences can be elicited in the<br />
form of value functions [von Winterfeldt <strong>and</strong><br />
Edwards, 1986; Eisenführ und Weber, 2003] as<br />
functions of the attributes. Often, such multiattribute<br />
value functions will be built as weighted<br />
sums of single-attribute value functions. To keep<br />
the value elicitation tractable, the objectives<br />
hierarchy may have to be simplified.<br />
553
Number of nat.<br />
tributaries p.r.l.<br />
Density of<br />
refugia<br />
Discharge of<br />
annual flood<br />
Average<br />
discharge.<br />
Amplitude <strong>and</strong><br />
period of fluct.<br />
Valley<br />
slope<br />
Number of delam.<br />
floods per season<br />
5% fractile of<br />
discharge distrib.<br />
Rate of increase/decrease<br />
Benthic<br />
Population<br />
Sub-Model<br />
Mean primary<br />
productivity<br />
Mean respiration<br />
rate<br />
Mean density<br />
of algae<br />
Mean density<br />
of shredders<br />
Mean leaf decomposition<br />
rate<br />
Mean density of<br />
grazers & collect.<br />
Mean density<br />
of predators<br />
Gravel size<br />
distribution<br />
River width<br />
constraints<br />
Hydraulics<br />
Sub-Model<br />
Morphological<br />
type<br />
Coef. of variation<br />
of water depth<br />
Length of<br />
connected reach<br />
Ratio of bank<br />
to river length<br />
Average<br />
floodplain width<br />
Coef. of variation<br />
of flow velocity<br />
Fraction of natural<br />
river banks<br />
Fraction of fine<br />
sediments<br />
Fish<br />
Sub-Model<br />
Vegetation<br />
Sub-Model<br />
Abundance<br />
of trout<br />
Abundance<br />
of nase<br />
Area of pioneer<br />
vegetation p.r.l.<br />
Area of hard wood<br />
vegetation p.r.l.<br />
Abundance<br />
of barbel<br />
Area of soft wood<br />
vegetation p.r.l.<br />
Area of gravel<br />
bars p.r.l.<br />
Gravel<br />
supply<br />
Long term/seasonal<br />
temp. change<br />
Mean susp. sed.<br />
conc. at low disch.<br />
Min. dissolved<br />
oxygen conc.<br />
Mean phosphate<br />
concentration<br />
Mean metal<br />
concentrations<br />
Accessibility<br />
of river section<br />
Amplitude of short<br />
term temp. fluct.<br />
Mean inorg.<br />
nitrogen conc.<br />
Mean organic<br />
pollutant conc.<br />
Shoreline<br />
Community<br />
Sub-Model<br />
Mean density of<br />
carabid beetles<br />
Mean density<br />
of ants<br />
Expected<br />
damage cost<br />
Groundwater recharge<br />
rate p.r.l.<br />
Reaeration<br />
coefficient<br />
Area of accessible<br />
gravel bars p.r.l.<br />
Mean density<br />
of spiders<br />
Retention<br />
volume p.r.l<br />
Groundwater<br />
infilt./transp. time.<br />
Construction cost<br />
per year, p.r.l<br />
Maintenance cost<br />
per year, p.r.l.<br />
Figure 3.<br />
Economics<br />
Sub-Model<br />
Change in no. of<br />
agricult. jobs<br />
Change in no. of<br />
non-agri. jobs<br />
Integrative model for the prediction of outcomes of decision alternatives for river rehabilitation. The<br />
rhombic nodes represent the attributes shown in Fig. 1, the round nodes are additional required inputs, <strong>and</strong><br />
the bold round nodes are the sub-models of the integrative river rehabilitation model. Nodes in the left<br />
column represent model inputs (some of them influenced by the decision alternative), nodes in the central<br />
column intermediate nodes, <strong>and</strong> nodes in the right column model outputs.<br />
As the l<strong>and</strong>scape integrity branch is resolved to a<br />
relatively high resolution in the hierarchy shown in<br />
Fig. 1, an option is to summarize ecological<br />
integrity <strong>and</strong> hydrogeomorphic integrity by a semiquantitative<br />
attribute scale visualized by a picture<br />
[Hostmann et al., 2004]. An alternative would be<br />
to elicit value functions for ecological <strong>and</strong><br />
hydrogeomorphic attributes from scientists <strong>and</strong> let<br />
the stakeholders only assess the weights of these<br />
branches based on a description of the range of<br />
possible outcomes.<br />
6. RANKING OF ALTERNATIVES<br />
The integrative model developed in section 4 leads<br />
to predictive probability distributions of the<br />
attributes. Applying the value functions elicited in<br />
section 5 to these attributes leads to a probability<br />
distribution of preference rankings of the<br />
alternatives for each stakeholder. Figure 4<br />
summarizes the results of such a ranking based on<br />
preliminary outcome predictions for a case study<br />
in Switzerl<strong>and</strong>.<br />
Rank<br />
Recreational<br />
organisations<br />
1<br />
2<br />
3<br />
4<br />
5<br />
Forest rangers<br />
Federal<br />
administration<br />
Stakeholder groups<br />
Industry<br />
<strong>Environmental</strong><br />
organisations<br />
Agricultural<br />
representatives<br />
Communities<br />
Cantonal<br />
administration<br />
Figure 4. Example of rankings of five river<br />
rehabilitation decision alternatives for<br />
different stakeholder groups according to<br />
Hostmann et al. (2004).<br />
0<br />
1<br />
2<br />
3<br />
4<br />
554
7. ASSESSMENT OF RESULTS<br />
The preference rankings of the alternatives derived<br />
from predictions <strong>and</strong> value assessments can be<br />
used to evaluate acceptance <strong>and</strong> conflict potential<br />
between stakeholders (alternative 4 in Fig. 4 was<br />
developed as a compromise alternative based on<br />
the results for alternatives 0-3). This can be used<br />
to structure stakeholder discussions, to develop<br />
compromise alternatives, <strong>and</strong> to make the basis for<br />
decisions transparent [Hostmann et al., 2004].<br />
Furthermore, the sensitivity of the results to<br />
uncertainty in prediction <strong>and</strong> valuation can be<br />
assessed.<br />
8. CONCLUSIONS<br />
River rehabilitation decisions can be controversial<br />
due to uncertain outcomes <strong>and</strong> conflicting interests<br />
of stakeholders. This paper demonstrates how<br />
decision analysis techniques can support such decisions<br />
by structuring the decision <strong>and</strong> stakeholder<br />
involvement processes <strong>and</strong> by making scientific<br />
assumptions <strong>and</strong> social preferences explicit.<br />
Nevertheless there are cases in which application<br />
of these techniques have been found to be poorly<br />
accepted [Hobbs et al., 1992]. Implementation<br />
aspects may responsible for these results.<br />
9. ACKNOWLEDGEMENTS<br />
This project was supported by the multidisciplinary<br />
Rhone-Thur project for scientific support of<br />
river rehabilitation projects in Switzerl<strong>and</strong> initiated<br />
<strong>and</strong> funded by the Swiss Federal Office for Water<br />
<strong>and</strong> Geology (BWG), the Swiss Federal Institute<br />
for <strong>Environmental</strong> Science <strong>and</strong> Technology<br />
(EAWAG) <strong>and</strong> the Swiss Federal Institute for<br />
Forest, Snow <strong>and</strong> L<strong>and</strong>scape Research (WSL)<br />
[Peter et al., 2004]. In addition, many project<br />
partners contributed through stimulating discussions,<br />
comments, <strong>and</strong> suggestions to this paper.<br />
10. REFERENCES<br />
Bledsoe, B.P. <strong>and</strong> Watson, C.C., Logistic analyis of<br />
channel pattern thresholds: me<strong>and</strong>ering, braiding,<br />
<strong>and</strong> incising, Geomorphology 38:281-300, 2001.<br />
Borsuk, M.E., P. Reichert, <strong>and</strong> P. Burkhardt-Holm, A<br />
Bayesian network for investigating the decline in<br />
fish catch in Switzerl<strong>and</strong>. In A.E. Rizzoli <strong>and</strong> A.J.<br />
Jakeman (Editors) Integrated Assessment <strong>and</strong><br />
Decision Support, Proc. of the iEMSs conference,<br />
Lugano, Switzerl<strong>and</strong>. Vol. 2, pp. 108-113, 2002.<br />
Borsuk, M.E., Stow, C.A., <strong>and</strong> Reckhow, K.H. Bayesian<br />
network of eutrophication models for synthesis,<br />
prediction, <strong>and</strong> uncertainty analysis. Ecological<br />
<strong>Modelling</strong>. In press, 2004.<br />
Boscaini, A., Franceschini,A. <strong>and</strong> Maiolini, B., River<br />
ecotones: carabid beetles as a tool for quality<br />
assessment, Hydrobiologia 422/423, 173-181, 2000.<br />
Charniak, E., Bayesian networks without tears, AI<br />
Magazine, 12(4), 50-63, 1991.<br />
Clemen, R.T., Making Hard Decisions, PWS-Kent,<br />
Boston, second edition, 1996.<br />
Da Silva A.M.A.F., Alternate bars <strong>and</strong> related alluvial<br />
processes. Thesis of Master of Science, Queen’s<br />
University, Kingston, Ontario, Canada, 1991.<br />
Eisenführ, F. <strong>and</strong> Weber, M., Rationales Entscheiden,<br />
Springer, Berlin, 4th edition, 2003.<br />
Hobbs, B.F., Chankong, V. And Hamadeh, W., Does<br />
choice of multicriteria method matter? An experiment<br />
in water resources planning, Wat. Resourc.<br />
Res. 28(7), 1767-1779, 1992.<br />
Hostmann, M., Truffer, B., Reichert, P. <strong>and</strong> Borsuk,<br />
M.E., Stakeholder values in decision support for<br />
river rehabilitation, submitted to Archiv für<br />
Hydrobiologie, Large River Supplement, 2004.<br />
Isard, W. et al., eds. Methods of Interregional <strong>and</strong><br />
Regional Analysis. Ashgate, 1998.<br />
Keeney, R.L., Value-Focused Thinking, Harvard<br />
University Press, Cambridge, 1992.<br />
Lahdelma, R., Salminen, P. <strong>and</strong> Hokkanen, J., Using<br />
multicriteria methods in environmental planning <strong>and</strong><br />
management, Env. Man. 26(6), 595-605, 2000.<br />
Lamouroux, N., Souchon, Y. <strong>and</strong> Herouin, E.,<br />
Predicting velocity frequency distributions in stream<br />
reaches, Wat. Resourc. Res. 31(9), 2367-2375, 1995.<br />
Lee, D.C. <strong>and</strong> Rieman, B.E., Population viability<br />
assessment of Salmonids by using probabilistic<br />
networks, North American J. of Fisheries Management<br />
17, 1144-1157, 1997.<br />
McIntire, C.D., Algal dynamics in laboratory streams: a<br />
simulation model <strong>and</strong> its implications, Ecological<br />
Monographs 43, 399-420, 1973.<br />
Miller, R.E. <strong>and</strong> Blair, P.E., Input-Output Analysis:<br />
Foundations <strong>and</strong> Extensions, Prentice-Hall, 1985.<br />
Pearl, J., Probabilistic reasoning in intelligent systems:<br />
networks of plausible inference, Morgan Kaufmann,<br />
San Mateo, California, 1988.<br />
Peter, A., Kienast, F. <strong>and</strong> Nutter, S., The Rhone-Thur<br />
River project: a comprehensive river rehabilitation<br />
project in Switzerl<strong>and</strong>, submitted.<br />
Prentice, I.C., Sykes, M.T. <strong>and</strong> Cramer, W., A<br />
simulation model for the transient effects of climate<br />
change on forest l<strong>and</strong>scapes, Ecological <strong>Modelling</strong><br />
65(1-2), 51-70, 1993.<br />
Reckhow, K.H., Water quality predictions <strong>and</strong><br />
probability network models, Can. J. Fish. Aquat.<br />
Sci. 56, 1150-1158, 1999.<br />
Rutherford, J.C., Scarsbrook, M.R. <strong>and</strong> Broekhuizen,<br />
N., Grazer control of stream algae: modeling<br />
temperature <strong>and</strong> flood effects, Journal of <strong>Environmental</strong><br />
Engineering 126(4), 331-339, 1999.<br />
Schälchli U.: Die Kolmation von Fliessgewässersohlen:<br />
Prozesse und Berechnungsgrundlagen, Mitteilungen<br />
der Versuchsanstalt für Wasserbau, Hydrologie und<br />
Glaziologie der ETH Zürich Nr. 124, 1993.<br />
Schweizer, S., Borsuk, M.E. <strong>and</strong> Reichert, P., Predicting<br />
the hydraulic <strong>and</strong> morphological consequences of<br />
river rehabilitation, submitted to iEMSs 2004.<br />
Von Winterfeldt, D. <strong>and</strong> Edwards, W., Decision<br />
Analysis <strong>and</strong> Behavioural Research, Cambridge<br />
University Press, 1986.<br />
555
Decision Making under Uncertainty in a Decision Support<br />
System for the Red River<br />
Inge A.T. de Kort <strong>and</strong> Martijn J. Booij<br />
Faculty of Engineering Technology, University of Twente, P.O. Box 217, 7500 AE Enschede, the Netherl<strong>and</strong>s<br />
(i.a.t.dekort@utwente.nl; m.j.booij@utwente.nl)<br />
Abstract: Decision support systems (DSSs) are increasingly being used in water management for the evaluation<br />
of impacts of policy measures under different scenarios. The exact impacts generally are unknown <strong>and</strong><br />
surrounded with considerable uncertainties. These uncertainties stem from natural r<strong>and</strong>omness, uncertainty in<br />
data, models <strong>and</strong> parameters, <strong>and</strong> uncertainty about measures <strong>and</strong> scenarios. It may therefore be difficult to make<br />
a selection of measures relevant for a particular water management problem. In order to support policy makers to<br />
make a strategic selection between different measures in a DSS while taking uncertainty into account, a<br />
methodology for the ranking of measures has been developed. The methodology has been applied to a pilot DSS<br />
for flood control in the Red River basin in Vietnam <strong>and</strong> China. The decision variable is the total flood damage<br />
<strong>and</strong> possible flood reducing measures are dike heightening, reforestation <strong>and</strong> the construction of a retention basin.<br />
For illustrative purposes, only parameter uncertainty is taken into account. The methodology consists of a Monte<br />
Carlo uncertainty analysis employing Latin Hypercube Sampling <strong>and</strong> a ranking procedure based on the<br />
significance of the difference between output distributions for different measures. The significance is determined<br />
with the Student test for Gaussian distributions <strong>and</strong> with the non-parametric Wilcoxon test for non-Gaussian<br />
distributions. The results show Gaussian distributions for the flood damage in all situations. The mean flood<br />
damage in the base situation is about 2.2 billion US$ for the year 1996 with a st<strong>and</strong>ard deviation due to parameter<br />
uncertainty of about 1 billion US$. Selected applications of the measures reforestation, dike heightening <strong>and</strong> the<br />
construction of a retention basin reduce the flood damage with about 5, 55 <strong>and</strong> 300 million US$ respectively. The<br />
construction of a retention basin significantly reduces flood damage in the Red River basin, while dike<br />
heightening <strong>and</strong> reforestation reduce flood damage, but not significantly.<br />
Keywords: Decision support systems; Water management; Uncertainty; Ranking methodology; Red River<br />
1 INTRODUCTION<br />
Decision support systems (DSSs) are increasingly<br />
being used in water management for the evaluation<br />
of impacts of policy measures under different<br />
scenarios. The exact impacts generally are unknown<br />
<strong>and</strong> surrounded with considerable uncertainties.<br />
These uncertainties stem from natural r<strong>and</strong>omness,<br />
uncertainty in data, models <strong>and</strong> parameters, <strong>and</strong><br />
uncertainty about measures <strong>and</strong> scenarios. It may<br />
therefore be difficult to make a selection of measures<br />
relevant for a particular water management problem.<br />
This paper describes a methodology for the ranking<br />
of measures in order to support policy makers to<br />
make a strategic selection between different<br />
measures in a DSS while taking uncertainty into<br />
account. The methodology is applied to a pilot DSS<br />
for flood control in the Red River basin in Vietnam<br />
<strong>and</strong> China. The decision variable is the total flood<br />
damage <strong>and</strong> possible flood reducing measures are<br />
dike heightening, reforestation <strong>and</strong> the construction<br />
of a retention basin. For illustrative purposes, only<br />
parameter uncertainty is taken into account.<br />
2 DESCRIPTION OF DSS<br />
2.1 Introduction<br />
The DSS consists of a hydrological, hydraulic <strong>and</strong><br />
socio-economic model. The hydrological model has<br />
a spatial resolution of 5 km for the complete river<br />
basin. The output of this model is input into the<br />
556
hydraulic model, which has a spatial resolution of 1<br />
km for the deltaic part of the river basin. The output<br />
of the hydraulic model is input into the socioeconomic<br />
model. This latter model has a spatial<br />
resolution of both 1 km <strong>and</strong> 5 km (see Figure 1. ).<br />
The temporal resolution of the DSS is one day <strong>and</strong><br />
the time period considered one year (1996). This<br />
year has been chosen, because it contains one of the<br />
major floods which have occurred in the river basin.<br />
The DSS is implemented in the GIS-based model<br />
environment PCRaster [Wesseling et al., 1996] as<br />
discussed in Booij [2003]. The three models are<br />
described in 2.2 <strong>and</strong> the flood control measures are<br />
considered in 2.3.<br />
Figure 1. Red River basin at a spatial resolution of 5 km (left, extent of area 770 km x 660 km) <strong>and</strong> delta of the<br />
Red River basin at a spatial resolution of 1 km (right, extent of area 154 km x 132 km).<br />
2.2 DSS model components<br />
The hydrological model is based on the HBV model<br />
concepts [Bergström <strong>and</strong> Forsman, 1973]. The HBV<br />
model is a conceptual hydrological model <strong>and</strong><br />
simulates basin discharge using precipitation <strong>and</strong><br />
evapotranspiration as input. The relevant routines<br />
used are a precipitation routine representing rainfall,<br />
a soil moisture routine determining actual<br />
evapotranspiration, overl<strong>and</strong> flow <strong>and</strong> subsurface<br />
flow, a fast flow routine representing storm flow, a<br />
slow flow routine representing subsurface flow <strong>and</strong> a<br />
transformation routine for flow delay <strong>and</strong><br />
attenuation.<br />
The simulated discharge serves as input into the<br />
hydraulic model. It is transformed into water depth<br />
using a stage-discharge relation derived from<br />
measured data. The water depth applies to the<br />
complete deltaic area. An additional water depth due<br />
to the tide is added to this water depth. The<br />
inundation depth in the flooded area is determined<br />
using this river water depth, the dike height <strong>and</strong> the<br />
elevation in the flooded area. A certain decrease of<br />
the inundation depth is assumed when in the flood<br />
wave is in its falling stage.<br />
The socio-economic model determines with simple,<br />
linear functions the flood damage <strong>and</strong> incomes for<br />
different economic sectors. The flood damage is<br />
dependent on the simulated inundation pattern <strong>and</strong><br />
the l<strong>and</strong> use type, while the incomes are dependent<br />
on the economic sector (through prices, costs etc.)<br />
<strong>and</strong> the l<strong>and</strong> use type. The decision variable is the<br />
total flood damage in the deltaic area of the Red<br />
River basin.<br />
2.3 Flood control measures<br />
The DSS can be used for the evaluation of impacts of<br />
policy measures under different scenarios. Three<br />
different measures are considered, namely dike<br />
heightening, reforestation <strong>and</strong> the construction of a<br />
retention basin. The impacts of these measures will<br />
be compared with the impacts in the base situation.<br />
The measures are briefly described below.<br />
The dike system is represented by a constant dike<br />
height relative to mean sea level, which obviously is<br />
a simplification of reality. Moreover, it is assumed<br />
that the dike system is of good quality, which may<br />
not hold in reality. For example Nghia [2000] states<br />
that the overall dike system is outdated, poor in<br />
repair <strong>and</strong> vulnerable to erosion. The measure dike<br />
heightening is achieved by increasing the constant<br />
dike height with 1 meter.<br />
557
Reforestation is a sustainable flood control measure<br />
<strong>and</strong> supports retainment of water in the soil <strong>and</strong><br />
prevents erosion. This is achieved by adapting the<br />
l<strong>and</strong> use pattern in the DSS, which subsequently will<br />
change the soil moisture function in the hydrological<br />
model <strong>and</strong> the damage <strong>and</strong> income estimates in the<br />
socio-economic model. Forest is r<strong>and</strong>omly attributed<br />
to areas in a certain elevation range <strong>and</strong> with some<br />
specific l<strong>and</strong> use types in the base situation.<br />
The construction of a retention basin is based on an<br />
existing retention basin. The main functions of the<br />
basin are flood control <strong>and</strong> power production. The<br />
water storage <strong>and</strong> release are dependent on several<br />
factors such as the inflow, the actual storage in the<br />
reservoir, the minimum <strong>and</strong> maximum storage <strong>and</strong><br />
the maximum outflow. More details about the<br />
implementation of the reservoir in the DSS can be<br />
found in De Kort [2003].<br />
3 RED RIVER BASIN<br />
The Red River basin is situated in China <strong>and</strong><br />
Vietnam <strong>and</strong> has a surface area of about 169 000<br />
km 2 . The delta covers about 15 000 km 2 <strong>and</strong> starts<br />
near Hanoi, the capital of Vietnam. The average<br />
annual precipitation strongly varies over the area<br />
between 700 <strong>and</strong> 4800 mm. About 80 % of the<br />
precipitation occurs in summer when the Southwest<br />
monsoon brings warm, moist air across in the Indo-<br />
Chinese peninsula. Most of the floods therefore<br />
occur in July <strong>and</strong> August. The average discharge of<br />
the Red River is about 3750 m 3 /s [Nghia, 2000].<br />
Similar to elsewhere in Southeast Asia, there is a<br />
marked contrast between the isolated <strong>and</strong> sparsely<br />
populated mountains <strong>and</strong> the densely populated<br />
delta. The delta is a low lying area mainly used for<br />
the cultivation of rice (about 88 % of the area) <strong>and</strong><br />
has one of the highest population densities (over<br />
1000 people per km 2 ) in the world. The upstream,<br />
mountainous area is more forested (about 42 % of<br />
the area) <strong>and</strong> grassl<strong>and</strong> forms the transition zone<br />
between the forest <strong>and</strong> rice areas.<br />
Daily precipitation <strong>and</strong> evapotranspiration data from<br />
15 stations <strong>and</strong> daily discharge data from 5 stations<br />
are used in this analysis. Furthermore, elevation data<br />
from a global digital elevation model <strong>and</strong> l<strong>and</strong> use<br />
data from a global l<strong>and</strong> cover database are employed.<br />
The spatial resolutions are 1 km for both the<br />
elevation <strong>and</strong> l<strong>and</strong> use data. This spatial resolution<br />
for elevation is assumed to be appropriate for<br />
inundation modelling taking into account the flatness<br />
of the study area <strong>and</strong> the research objective. Socioeconomic<br />
data include incomes, agricultural yields<br />
<strong>and</strong> flood damage in general at a provincial level <strong>and</strong><br />
on an annual basis. Further information about the<br />
Red River basin <strong>and</strong> the data resources can be found<br />
in Booij [2003] <strong>and</strong> De Kort [2003].<br />
4 RANKING METHODOLOGY<br />
4.1 Introduction<br />
A methodology for the ranking of measures in a DSS<br />
has been developed in order to support policy makers<br />
to make a strategic selection between different<br />
measures while taking uncertainty into account. The<br />
methodology consists of an uncertainty analysis <strong>and</strong><br />
a ranking procedure based on the significance of the<br />
difference between output distributions for different<br />
measures. These two steps are described below.<br />
4.2 Uncertainty analysis<br />
In an uncertainty analysis, the effect of different<br />
uncertainties (e.g. from data, models <strong>and</strong> parameters)<br />
on the output of interest (the decision variable) is<br />
determined. Two aspects are discussed, namely the<br />
type of uncertainty to be investigated <strong>and</strong> the choice<br />
of the uncertainty analysis method.<br />
For illustrative purposes, only the effect of parameter<br />
uncertainty on the total flood damage is taken into<br />
account. This uncertainty source is chosen, because<br />
it may have large effects on the output, is relatively<br />
easy to quantify <strong>and</strong> is interesting in the context of<br />
the DSS. Only the uncertainty of six dominant<br />
parameters is considered. These are two parameters<br />
in the fast flow routine of the hydrological model,<br />
two parameters in the stage-discharge relation <strong>and</strong><br />
one parameter in the inundation formulation of the<br />
hydraulic model, <strong>and</strong> one parameter in the flood<br />
damage function for rice of the socio-economic<br />
model. They have been selected on the basis of a<br />
first-order uncertainty analysis [see De Kort, 2003].<br />
The uncertainty analysis method has been chosen<br />
based on a multi criteria analysis. Criteria for the<br />
selection were the nature of the model, research<br />
purpose, previous comparisons <strong>and</strong> available<br />
resources [Morgan <strong>and</strong> Henrion, 1990; Booij, 2002].<br />
Based on this analysis, the Latin Hypercube<br />
Sampling (LHS) method has been chosen, which is a<br />
stratified sampling version of the Monte Carlo<br />
method <strong>and</strong> efficiently estimates the statistics of an<br />
output [Melching, 1995].<br />
558
4.3 Ranking procedure<br />
The ranking procedure is based on the significance<br />
of the difference between output distributions for<br />
different measures taking parameter uncertainty into<br />
account. Therefore, first the distribution type needs<br />
to be determined <strong>and</strong> second, the significance of the<br />
differences is required as described below.<br />
The hypothesis of output distributions being<br />
normally distributed is tested visually with quantilequantile<br />
plots <strong>and</strong> quantitatively with the<br />
Kolmogorov-Smirnov test [see e.g. Zar, 1996]. The<br />
nature of the output distribution (Gaussian or non-<br />
Gaussian) determines which test is used in the next<br />
step.<br />
The significance is determined with the Student test<br />
for Gaussian distributions <strong>and</strong> with the Wilcoxon test<br />
for non-Gaussian distributions. The Student test<br />
compares the means of two distributions, while<br />
taking the variance of both distributions into account.<br />
The specific Student test to be used depends on the<br />
homogeneity of the variances from both<br />
distributions. The Wilcoxon signed rank test [see e.g.<br />
Zar, 1996], also known as the Mann-Whitney test, is<br />
a non-parametric test that detects differences in the<br />
distribution of two situations by ranking the output<br />
in both situations <strong>and</strong> comparing the resulting,<br />
st<strong>and</strong>ardised ranks.<br />
5 RESULTS<br />
5.1 Uncertainty analysis<br />
Histogram basic situation<br />
Histogram reforestation<br />
35<br />
35<br />
30<br />
30<br />
25<br />
25<br />
Frequency<br />
20<br />
15<br />
Frequency<br />
20<br />
15<br />
10<br />
10<br />
5<br />
5<br />
0<br />
0<br />
1400<br />
1000<br />
600<br />
200<br />
Damage in million US$<br />
1400<br />
1000<br />
600<br />
200<br />
Damage in million US$<br />
5400<br />
5000<br />
4600<br />
4200<br />
3800<br />
3400<br />
3000<br />
2600<br />
2200<br />
1800<br />
Histogram dike heightening<br />
5400<br />
5000<br />
4600<br />
4200<br />
3800<br />
3400<br />
3000<br />
2600<br />
2200<br />
1800<br />
Histogram retention basin<br />
35<br />
35<br />
30<br />
30<br />
25<br />
25<br />
Frequency<br />
20<br />
15<br />
Frequency<br />
20<br />
15<br />
10<br />
10<br />
5<br />
5<br />
1400<br />
1000<br />
600<br />
200<br />
1400<br />
1000<br />
600<br />
200<br />
0<br />
0<br />
Damage in million US$<br />
Damage in million US$<br />
5400<br />
5000<br />
4600<br />
4200<br />
3800<br />
3400<br />
3000<br />
2600<br />
2200<br />
1800<br />
5400<br />
5000<br />
4600<br />
4200<br />
3800<br />
3400<br />
3000<br />
2600<br />
2200<br />
1800<br />
Figure 2. Histograms <strong>and</strong> fitted Gaussian curves for base situation <strong>and</strong> three flood control measures<br />
559
The results of the uncertainty analysis will be briefly<br />
described. First, some information about the<br />
dominant parameters <strong>and</strong> the implementation of LHS<br />
is given.<br />
Only the six dominant parameters contributing<br />
considerably to the output uncertainty are sampled in<br />
the LHS uncertainty analysis. For all six parameters<br />
uniform distributions are assumed, because no data<br />
were available <strong>and</strong> other studies [e.g. Yu et al., 2001]<br />
employed uniform distributions for similar analyses<br />
as well. A total number of 100 samples of parameter<br />
sets has been used to generate 100 output values for<br />
each situation (base situation <strong>and</strong> three measures).<br />
This number of samples is arbitrary chosen based on<br />
previous uncertainty analysis studies <strong>and</strong> the fact that<br />
this number corresponds to a reasonable number of<br />
about 1000 samples when employing Monte Carlo<br />
analysis [Yu et al., 2001].<br />
The results of the four sets of 100 LHS simulations<br />
are shown in Figure 2. The simulated mean flood<br />
damage for the base situation corresponds well with<br />
the observed one (not shown here) of about 2.2<br />
billion US$. It should be noted here that the flood of<br />
1996 was one of the five major floods in the 20 th<br />
century <strong>and</strong> thus the resulting damage was high. The<br />
measures reforestation, dike heightening <strong>and</strong> the<br />
construction of a retention basin reduce the<br />
simulated mean flood damage with about 5, 55 <strong>and</strong><br />
300 million US$ respectively. It should be noted that<br />
the extent to which the flood damage is reduced<br />
depends on the dimensions <strong>and</strong> the location of the<br />
flood control measure. The small effect of<br />
reforestation on the flood damage may be due to the<br />
fact that erosion <strong>and</strong> sedimentation processes are not<br />
taken into account in the DSS. These processes<br />
probably play an important role in realising the flood<br />
control function of reforestation. St<strong>and</strong>ard deviations<br />
for all four situations are high (up to 45 % of the<br />
mean value) indicating large uncertainties in the<br />
estimation of the total flood damage. Obviously, this<br />
results in large overlaps of the probability<br />
distributions shown in Figure 2.<br />
5.2 Ranking procedure<br />
The first step in the ranking procedure has been the<br />
determination of the distribution type. The quantilequantile<br />
plots showed reasonable straight lines with<br />
even in the tails only slight deviations from the<br />
expected normal value. This is confirmed<br />
quantitatively by the Kolmogorov-Smirnov test.<br />
Moreover, Large Lilliefors significance values (>><br />
0.05) indicated that the output results can be<br />
considered as normally distributed. The four normal<br />
distributions (gray line is under black line) <strong>and</strong> their<br />
statistical notation are shown in Figure 3.<br />
Black basic situation N(2235, 970)<br />
Black dot dike heightening N(2180, 885)<br />
Gray reforestation N(2230, 975)<br />
Gray dot retention basin N(1930, 770)<br />
Frequency<br />
Damage in million US$<br />
Figure 3. Normal distribution for base situation <strong>and</strong> three flood control measures (gray line is under black line).<br />
560
The second step has been the assessment of the<br />
significance of the difference between output<br />
distributions for different measures taking parameter<br />
uncertainty into account. The Student test is used for<br />
this purpose, because the model outputs were found<br />
to be normally distributed. According to this test, the<br />
construction of a retention basin is the only measure<br />
that significantly improves flood control for the Red<br />
River (two-tailed significance level < 0.05 <strong>and</strong> a<br />
mean difference of about 300 million US$). The<br />
other two flood control measures result in a smaller<br />
mean flood damage than in the base situation, but do<br />
not significantly improve the situation. The final<br />
ranking of the flood control measures is therefore: 1.<br />
construction of a retention basin; 2. dike heightening;<br />
3. reforestation.<br />
6 CONCLUSIONS<br />
A methodology for the ranking of measures in a DSS<br />
while taking uncertainty into account has been<br />
developed <strong>and</strong> applied to a pilot DSS for flood<br />
control in the Red River basin in Vietnam <strong>and</strong> China.<br />
The methodology consists of an uncertainty analysis<br />
<strong>and</strong> a ranking procedure based on the significance of<br />
the difference between output distributions for<br />
different measures.<br />
The mean flood damage in the base situation is about<br />
2.2 billion US$ for the year 1996 with a st<strong>and</strong>ard<br />
deviation due to parameter uncertainty of about 1<br />
billion US$. The measures reforestation, dike<br />
heightening <strong>and</strong> the construction of a retention basin<br />
reduce the flood damage with about 5, 55 <strong>and</strong> 300<br />
million US$ respectively. The construction of a<br />
retention basin significantly reduces flood damage in<br />
the Red River basin, while dike heightening <strong>and</strong><br />
reforestation reduce flood damage, but not<br />
significantly.<br />
Decision making on the basis of these results should<br />
be done with care. Several potentially important<br />
processes (e.g. erosion, sedimentation) are not taken<br />
into account yet, because of the pilot status of the<br />
DSS. Moreover, only six dominant parameters are<br />
considered in the uncertainty analysis. Other points<br />
which should be kept in mind are the dependency of<br />
the outcomes on the location <strong>and</strong> dimensions of the<br />
measures <strong>and</strong> the fact that implementation <strong>and</strong><br />
maintenance costs of measures are not considered<br />
yet. However, the methodology proved to be suitable<br />
for the ranking of measures <strong>and</strong> may support<br />
decision makers when dealing with uncertainty.<br />
7 ACKNOWLEDGEMENTS<br />
This study has been done within the context of the<br />
FLOCODS project which is funded under the EC<br />
contract number ICA4-CT2001-10035 within the<br />
Fifth Framework Program. This study benefited<br />
greatly from the discussions with Denie Augustijn en<br />
Jean-Luc de Kok of the University of Twente.<br />
8 REFERENCES<br />
Bergström, S., <strong>and</strong> A. Forsman, Development of a<br />
conceptual deterministic rainfall-runoff model,<br />
Nordic Hydrology, 4, 147-170, 1973.<br />
Booij, M.J., Appropriate modelling of climate<br />
change impacts on river flooding, Ph.D. thesis,<br />
University of Twente, Enschede, 2002.<br />
Booij, M.J., Decision support system for flood<br />
control <strong>and</strong> ecosystem upgrading in Red River<br />
basin, In: G. Blöschl, S. Franks, M. Kumagai, K.<br />
Musiake <strong>and</strong> D. Rosbjerg (Eds.), Water<br />
Resources Systems - Hydrological Risk,<br />
Management <strong>and</strong> Development, Proc.<br />
Symposium HS02b at IUGG 2003, 30 June-11<br />
July 2003, Sapporo, Japan, 115-122, 2003.<br />
De Kort, I.A.T., Decision making under uncertainty–<br />
Ranking measures in a decision support system<br />
for flood control in the Red River in Vietnam<br />
while taking uncertainty into account, M.Sc.<br />
thesis, University of Twente, Enschede, 2003.<br />
Melching, C.S., Reliability estimation, In: V.P.<br />
Singh (Ed.), Computer models of watershed<br />
hydrology, Water Resources Publications,<br />
Colorado, 1995.<br />
Morgan, M.G., <strong>and</strong> M. Henrion, Uncertainty: a<br />
guide to dealing with uncertainty in quantitative<br />
risk <strong>and</strong> policy analysis, Cambridge University<br />
Press, Cambridge, 1990<br />
Nghia, T., Flood control planning for Red River<br />
basin, Proceedings of <strong>International</strong> European-<br />
Asian Workshop Ecosystem & Flood, Hanoi,<br />
Vietnam, 2000.<br />
Wesseling, C.G., D.-J. Karssenberg, P.A. Burrough,<br />
<strong>and</strong> W.P.A. van Deurssen, Integrating dynamic<br />
environmental models in GIS: the development<br />
of a dynamical modelling language,<br />
Transactions in GIS, 1, 40-48, 1996.<br />
Yu, P.-S., T.-C. Yang, <strong>and</strong> S.-J. Chen, Comparison<br />
of uncertainty analysis methods for a distributed<br />
rainfall-runoff model, Journal of Hydrology,<br />
244, 43-59, 2001.<br />
Zar, J.H., Biostatistical analysis, Prentice Hall,<br />
Upper Saddle River, NJ, 1996.<br />
561
Development of a GIS-based Decision Support Tool for<br />
Integrated Water Resources Management in Southern<br />
Africa<br />
M. Märker a , K.Bongartz b <strong>and</strong> W.-A. Flügel b<br />
a<br />
Institute for Geoecology, Potsdam Universit, Germany<br />
b<br />
Geoinformatics, Geographical Institute, Friedrich Schiller University, Jena Germany<br />
Abstract: In about 30 years large semi-arid areas in Africa will have not enough water for sustainable food<br />
production. Basin water resources are to a large extend allocated to agriculture <strong>and</strong> forestry, thus, also<br />
groundwater frequently is overexploited. Irrigation agriculture will compete under increasing demographic<br />
pressure with claims by other powerful water users such as the timber industry. The environment also claims<br />
for enough water to fulfil its service functions. Because of the different stakeholders <strong>and</strong> their water needs a<br />
social accepted water resource governance must be implemented that ensure equitable water distribution to<br />
improve health, <strong>and</strong> alleviate poverty. With its objective to develop an GIS-based Decision Support Tool for<br />
Integrated Water Resources Management this study addresses the above stated challenges for water resources<br />
management in southern Africa. Therefore the following steps with their specific objectives <strong>and</strong> techniques<br />
have been conducted: (i) Delineation of key parameters <strong>and</strong> building a spatially enabled database. (ii)<br />
<strong>Modelling</strong> hydrological <strong>and</strong> geoecological system dynamics. (iii) Integration of socio-economic <strong>and</strong><br />
ecological requirements in order to develop the DSS. The overall scope of the project, collaboratively carried<br />
out by a transnational interdisciplinary research consortium, was an innovative computer based toolset<br />
designed as an assembly of tested, validated <strong>and</strong> well documented procedures comprising the above outlined<br />
key technologies.<br />
Keywords: South Africa, Integrated Water Resources Management, Decision Support.<br />
1. INTRODUCTION<br />
In Southern Africa the paramount issues affecting<br />
human beings are food security, poverty<br />
alleviation, improved health <strong>and</strong> environmental<br />
security. This problems are closely related to a<br />
sustainable <strong>and</strong> productive capacity of agriculture<br />
<strong>and</strong> to the availability of freshwater in an adequate<br />
quality for fisheries, irrigation, environmental,<br />
industrial <strong>and</strong> human consumption. These general<br />
problems <strong>and</strong> especially growing dem<strong>and</strong>s of<br />
society regarding use <strong>and</strong> protection of water<br />
bodies, call for a multidisciplinary approach in<br />
order to develop new views <strong>and</strong> strategies to policy<br />
for water resources management. Thus, water<br />
resources management (WRM) has become<br />
increasingly complex over the last decades.<br />
The complexity of WRM affects both the technical<br />
issues (e.g. computation concepts) as well as the<br />
requirements of different stakeholders. The<br />
challenge for WRM is to integrate state of the art<br />
technical methods <strong>and</strong> tools for water resources<br />
assessment considering the different impacts on<br />
water resources. These impacts consist in<br />
competing stakeholder dem<strong>and</strong>s such as<br />
environmental water requests <strong>and</strong> socio-economic<br />
strains on water resources.<br />
In this paper we are presenting an example for an<br />
integrated water resources management in the<br />
KwaZulu/ Natal province of the South African<br />
Republic. This example is part of an framework<br />
project financed by the EU entitled: “The<br />
development of an innovative computer based<br />
“Integrated Water Resources Management System<br />
(IWRMS)” in semiarid catchments for water<br />
resources analysis <strong>and</strong> prognostic scenario<br />
planning”.<br />
The current water resources management in<br />
Southern Africa is characterised by: (i) a spatially<br />
<strong>and</strong> temporally uneven <strong>and</strong> uncertain distribution,<br />
(ii) problematic water quality (e.g. ecoli bacteria),<br />
(iii) lack of information about resources <strong>and</strong> (iv) a<br />
new water legislation to be implemented.<br />
562
2. OBJECTIVES<br />
The overall objective of this project was to develop<br />
an Integrated Water Resources Management<br />
System (IWRMS) for semi-arid catchments <strong>and</strong><br />
validate it in the Southern African region. The<br />
IWRMS, an assembly of tested, validated <strong>and</strong><br />
documented procedures comprising database<br />
management, remote sensing, GIS, <strong>and</strong> physically<br />
based modelling, is designed to enable water<br />
managers <strong>and</strong> decision makers to improve the<br />
strategic planning of catchment water resources.<br />
Background for IWRMS was a preliminary<br />
evaluation of competing stakeholder dem<strong>and</strong>s in<br />
order to optimise the decision tool with respect to a<br />
sustainable water use <strong>and</strong> to protect water <strong>and</strong> l<strong>and</strong><br />
resources. In order to achieve this overall<br />
objective, three major work components were<br />
defined <strong>and</strong> carried out in the Mkomazi basin. The<br />
first step was to gain relevant information by<br />
developing <strong>and</strong> applying remote sensing methods<br />
as well as surveys of water dem<strong>and</strong>. Subsequently ,<br />
this information was classified <strong>and</strong> used as an input<br />
to catchment models, which were utilised to<br />
simulate “What-if?”-scenarios, such as l<strong>and</strong> use<br />
<strong>and</strong> climate change impacts on water resources.<br />
Moreover, GIS methods were applied to balance<br />
water supply <strong>and</strong> water dem<strong>and</strong> scenarios to<br />
identify areas under particular water competition.<br />
Furthermore new spatial disaggregation concepts<br />
regarding hydrological <strong>and</strong> soil-erosion processes<br />
were developed to improve the simulation <strong>and</strong><br />
scenario creation capabilities.<br />
One of the final activities of the IWRMS project<br />
was the development of a technical approach to<br />
integrate all components <strong>and</strong> its implementation as<br />
a prototype software system. Finally the<br />
implementation of the results into practice was<br />
carried out with identified stakeholders.<br />
3. STUDY AREA<br />
The Mkomazi catchment, situated in<br />
KwaZulu/Natal, South Africa comprises approx.<br />
4400 km², stretching from the Great Escarpment of<br />
the Drakensberg (3000 m a.s.l.) to sea level at its<br />
mouth at the Indian Ocean. The mean annual<br />
precipitation (MAP) varies between 500 - 1200<br />
mm. The study catchment reflects the<br />
subcontinent's uneven distribution of resources <strong>and</strong><br />
population in general. The catchment is<br />
characterized by communal l<strong>and</strong>s (mainly<br />
subsistence farming, densely populated, less<br />
developed) <strong>and</strong> large scale commercial farming<br />
<strong>and</strong> forestry (sugar cane, maize; exotic eucalypt,<br />
pine <strong>and</strong> acacia species; major water users,<br />
sparsely populated, economically important) <strong>and</strong><br />
small scale commercial farming <strong>and</strong> resettlement<br />
areas (positioned between the communal <strong>and</strong><br />
commercial l<strong>and</strong> uses, strong potential for<br />
redevelopment, but little access to resources).<br />
Fig. 1 Location of study area<br />
Besides agriculture <strong>and</strong> forestry, the other user<br />
sectors are of varying importance in terms of<br />
dem<strong>and</strong> amount. Urban centres, mining activities<br />
<strong>and</strong> paper industries, however, play a considerable<br />
role as water users.<br />
South Africa has a new water legislation which<br />
prioritises water use <strong>and</strong> protection in a similar<br />
way: Primary water use (drinking, cooking,<br />
washing, gardening, live stock watering) <strong>and</strong><br />
environmental dem<strong>and</strong> have the highest priority.<br />
Thereafter, all commercial water uses have to be<br />
licensed depending on the efficiency <strong>and</strong> socioeconomic<br />
impact. All current water uses have to<br />
undergo a thorough revision <strong>and</strong> are subject to reallocation.<br />
Water use conflicts are caused by large scale<br />
afforestation with dem<strong>and</strong>ing exotic species like<br />
eucalypt as well as a population growth making the<br />
rural water supply difficult in terms of water<br />
availability, water allocation <strong>and</strong> water quality. A<br />
proposed reservoir in the upper catchment was an<br />
additional dem<strong>and</strong>ing challenge to investigate.<br />
4. METHODS<br />
Core of the IWRMS is a physically-based<br />
hydrological model that is able to describe<br />
hydrological process dynamics within the<br />
catchment. Models with such a complex structure<br />
require many physiographic up-to-date data, such<br />
as, l<strong>and</strong>use, vegetation cover <strong>and</strong> topography,<br />
which are normally limited in availability,<br />
particularly in developing countries. A way to<br />
overcome this dilemma is to use state of the art<br />
remote sensing techniques. Besides conventional<br />
measurement of meteorological input at climate<br />
stations, the following information is crucial for<br />
analysing <strong>and</strong> modelling water resources <strong>and</strong> hence<br />
563
distributed hydrological model, the ACRU (Agro-<br />
Hydrological <strong>Modelling</strong> System, Schulze 1995),<br />
that accounts for the natural water cycle <strong>and</strong> is<br />
capability of simulating “What-if?” scenarios like<br />
climate or l<strong>and</strong> use changes or impoundment<br />
impacts (Tab. 1).<br />
Fig. 2 Concept <strong>and</strong> components of the IWRMS<br />
was objective in IWRMS:<br />
• Creation of different scale DEMs, accounting for<br />
scale transfer (hillslope, catchment, basin),<br />
• Catchment-wide l<strong>and</strong>use classifications, with<br />
improved hydrological relevance,<br />
• Detection of rural settlements to obtain<br />
information on water dem<strong>and</strong> <strong>and</strong> pollution<br />
sources,<br />
• Mapping of erosion features <strong>and</strong> l<strong>and</strong><br />
degradation as reference units for an enhanced<br />
erosion model,<br />
• Estimation of time series of climate <strong>and</strong><br />
vegetation variables,<br />
• Survey of water dem<strong>and</strong>, use, consumption <strong>and</strong><br />
need of the different user sectors.<br />
The structure of an integrated information <strong>and</strong><br />
management system for water resources in<br />
Southern Africa is shown in figure 2. By fusion of<br />
data of different spectral, temporal <strong>and</strong> spatial<br />
information as well as their integration with ground<br />
measurements, it is possible to combine the various<br />
advantages of the different sources.<br />
Tab. 1: IWRMS scenarios<br />
Scenario Description<br />
A “Present catchment state”<br />
B “Base line” potential natural vegetation<br />
according to Ackocks veld types (1998)<br />
C “Climate change”: implementation of<br />
GCM results with 2 x CO 2 (Murphy et al.<br />
1995)<br />
D “Reservoir Construction”: integration of<br />
proposed dam for water transfer out of<br />
system<br />
A GIS is coupled with the system <strong>and</strong> serves as a<br />
geodata management <strong>and</strong> pre-processing unit for<br />
the hydrological process model. GIS is also used as<br />
a post processing component, integrating the<br />
various model results <strong>and</strong> visualising them. The<br />
core of the system is made up of a physically-based<br />
4.1 <strong>Modelling</strong> hydrological <strong>and</strong> geoecological<br />
system dynamics<br />
The hydrological modelling was based on the<br />
distributed modelling approach of Response Units<br />
(RU’s) (Flügel 1996, Bongatz 2003). Therefore<br />
the test catchment was subdivided into<br />
homogeneous process entities (WRRUs; Water<br />
Resources Response Units) in<br />
Fig. 3 Spatial catchment representations by<br />
disaggregation into WRRUs<br />
order to enable a management based modelling<br />
(Fig. 3) (Taylor et al. 2000). These management<br />
issues are implemented by the formulation <strong>and</strong><br />
simulation of l<strong>and</strong> use <strong>and</strong> climate change<br />
scenarios using ACRU. GIS applications were used<br />
subsequently to combine the results of both the<br />
hydrological models <strong>and</strong> the water dem<strong>and</strong><br />
analysis. The objectives are to balance supply <strong>and</strong><br />
dem<strong>and</strong> under various runoff <strong>and</strong> dem<strong>and</strong><br />
scenarios <strong>and</strong> to analyse new allocation principles.<br />
4.2 <strong>Modelling</strong> geochemical system dynamics<br />
(Water Quality)<br />
The main factor influencing water quality in the<br />
Mkomazi catchment is related to sediment<br />
transport in the rivers. These sediments lead to<br />
“off-side damages” (e.g. degraded water quality,<br />
reservoir sedimentation). To assess the sediment<br />
dynamics an innovative approach was developed in<br />
order to integrate different relevant erosion<br />
processes such like interrill-rill erosion <strong>and</strong> gully<br />
erosion. The latter linear processes are often<br />
neglected in conventional models, despite their<br />
564
considerable contribution to the overall sediment<br />
yield in the region studied (see Märker et al. 2001).<br />
4.3 Integration of socio-economic <strong>and</strong> ecological<br />
requirements<br />
An appreciation of the broad structure of dem<strong>and</strong><br />
is necessary for the design <strong>and</strong> scaling of an<br />
effective overall approach. Coarse compilation of<br />
catchment-specific dem<strong>and</strong> has been carried out by<br />
analysing existing data from catchment<br />
management organisations. More challenging is the<br />
complex issue of primary water use at rural<br />
community level. In this context an attempt has<br />
been made to identify a generic structure of<br />
individual water use based on detailed householdlevel<br />
surveys in the catchment. The following<br />
sectors have been further investigated: (i) Primary<br />
water use (water for living) – rural <strong>and</strong> urban. (ii)<br />
<strong>Environmental</strong> flows. (iii) Commercial water use<br />
(agriculture, forestry, industry, mining). (iv) Other<br />
water uses, including recreational water use.<br />
5. RESULTS<br />
It has been recognised by IWRMS that there are<br />
significant distinctions between dem<strong>and</strong>, need <strong>and</strong><br />
consumption. In general, it is the aim of the new<br />
water legislation to meet water need, but in<br />
practice the estimation of dem<strong>and</strong> tends to be<br />
dominated by current consumption – which may<br />
overestimate need in sectors where water use is<br />
inefficient, <strong>and</strong> under-estimate need where the<br />
sector is ill-equipped to set <strong>and</strong> justify more<br />
realistic target needs (specifically in primary water<br />
use). It has been assumed up to this point that<br />
water consumption for primary purposes reflects<br />
simply human consumption, but it should be<br />
remembered that subsistence water use also<br />
requires stock watering. The data for households,<br />
household size, number of water journeys, <strong>and</strong><br />
number/size of water containers has been used to<br />
assess likely per capita primary water consumption.<br />
For the Mkomazi catchment the survey have shown<br />
that the average consumption per day <strong>and</strong> person is<br />
32.0 litres (1311 people) whereas the predominant<br />
water sources are community boreholes, streams as<br />
well as protected <strong>and</strong> unprotected springs.<br />
5.1 Model <strong>and</strong> scenario simulation results<br />
The models verify the proposed <strong>and</strong> applied<br />
methodology (ACRU with WRRUs) in general.<br />
This configuration was very useful for a quick<br />
flexible scenario development <strong>and</strong> modelling. The<br />
distributed nature of the catchment model facilitate<br />
(i) the distributed parameterisation of model<br />
scenarios to specific conditions in the respective<br />
sub-areas <strong>and</strong> (ii) the possibility of quick mapping<br />
of the model results to the WRRUs, hence<br />
obtaining a spatially disaggregated view of the<br />
problem.<br />
Fig. 4 Example of scenario comparison for entire<br />
Mkomazi catchment (low flow conditions)<br />
The scenarios applied to the Mkomazi catchment<br />
showed clear trends, which can be summarised as<br />
follows: (i) Compared to the potential natural<br />
status of the catchment, the most impacted areas<br />
are those which have experienced the greatest<br />
anthropogenic changes. Areas afforested with<br />
exotic tree species <strong>and</strong> irrigated areas planted with<br />
sugar cane are the most severely impacted WRRUs<br />
in terms of water balance (mean streamflow<br />
reductions 30 – 100 % for irrigated sugar cane).<br />
(ii) Grassl<strong>and</strong>, which is not heavily degraded,<br />
shows hardly any changes in hydrological response<br />
compared to the natural state. Areas with<br />
subsistence agriculture <strong>and</strong> degraded l<strong>and</strong>,<br />
however, generate higher streamflow (up to 25 %)<br />
than under baseline l<strong>and</strong> cover conditions due to<br />
higher direct runoff response. There are also<br />
indications of increased streamflow in the areas of<br />
invasion of thicket <strong>and</strong> bushl<strong>and</strong> into an area which<br />
under baseline conditions was coastal forest <strong>and</strong><br />
thornveld. (iii) The impacts of present l<strong>and</strong> use on<br />
low flows can be summarised such that areas<br />
already over-committed in terms of water<br />
allocation are those areas that have least of their<br />
annual streamflow regime represented as low flows<br />
(Fig. 4). The exceptions to this feature are that in<br />
the most commercially developed areas, the annual<br />
streamflow is impacted to a greater extent than the<br />
low flows. During periods of low flows in these<br />
WRRUs there is virtually no available water<br />
according to simulations, except for the seepage<br />
releases from dams. (iv) The climate change<br />
scenarios reveal a heavy impact on the water<br />
balance of virtually the entire catchment. Areas in<br />
which there is presently excess water that can be<br />
utilised by plants in transpiration are most affected.<br />
Areas supporting commercial afforestation as well<br />
as irrigation are amongst those most severely<br />
affected by potential climate change. However,<br />
where there is a greater spatial extent of irrigated<br />
l<strong>and</strong> compared with afforestation, the<br />
565
Fig. 5 Water availability/deficiency for chosen<br />
scenarios in the Mkomazi catchment<br />
percentage reduction from present l<strong>and</strong> use is<br />
negligible. This can be interpreted as no available<br />
water in those areas under present l<strong>and</strong> use<br />
conditions. (v) In the upper Mkomazi catchment<br />
there is substantial available water for allocation to<br />
inter-basin transfers from a reservoir proposed<br />
there (Fig. 4). The simulation of impacts on the<br />
water balance of this transfer shows that the<br />
greatest impact on accumulated streamflow is at<br />
the site of the proposed transfer where there is<br />
expected to be a 32% reduction in median<br />
streamflow compared with those from present l<strong>and</strong><br />
use. The impact diminishes downstream, but has<br />
still not fully recovered at the mouth of the river (-<br />
22 %). A scenario combining this water transfer<br />
with climate change conditions, shows basically<br />
the same impacts as with present climate<br />
conditions, however, to a greater extent (79 – 45 %<br />
reduction of mean streamflow). A number of<br />
dem<strong>and</strong> scenarios have been developed <strong>and</strong><br />
compared against the ACRU-based runoff<br />
scenarios. Results for the Mkomazi basin are<br />
displayed in figure 5. Scenario 3 refers to current<br />
sectoral dem<strong>and</strong> plus environmental reserve versus<br />
baseline runoff (low flow conditions); scenario 14<br />
represents deficiency based on future dem<strong>and</strong><br />
(current dem<strong>and</strong> plus inter basin transfer from<br />
proposed dam <strong>and</strong> increased (+10 %) forestry).<br />
One problem with dem<strong>and</strong>/deficiency modelling is<br />
to distinguish between water consumption <strong>and</strong> loss<br />
(e.g. by additional evapotranspiration from<br />
irrigated l<strong>and</strong>) <strong>and</strong> water that is used, but actually<br />
mostly returns to the streamflow after its use.<br />
Vegetative water consumption is modelled within<br />
the hydrological model, <strong>and</strong> thus must not be<br />
considered again by a GIS-based post-processing<br />
of supply <strong>and</strong> dem<strong>and</strong> information, regardless,<br />
whether this is carried out on spatial subunits or by<br />
a network approach.<br />
5.2 Water Quality modelling<br />
The analysis in the Mkomazi catchment show, that<br />
water quality is mainly influenced by the<br />
transported sediments. These sediments were<br />
produced by different erosion processes within the<br />
catchment. It was shown by a detailed survey <strong>and</strong><br />
by a subsequently erosion modelling that especially<br />
the linear gully erosion processes have to be<br />
included into the calculation of the sediment<br />
budget particularly where the lithology (colluvial<br />
materials of Masotcheni formation) is highly<br />
vulnerable to erosion. In this study the<br />
regionalization approach of ERUs (Erosion<br />
Response Units ) (Märker et al . 2001) was used to<br />
identify areas subject to different erosion processes<br />
<strong>and</strong> as modelling entities.<br />
5.3 Implementation of results<br />
The roll-out of the IWRMS approach has been<br />
influenced by the fact that the national/regional<br />
implementation agencies concerned in the<br />
development <strong>and</strong> piloting processes all have<br />
existing technical systems or procedures. Technical<br />
compatibility between IWRMS <strong>and</strong> existing or<br />
emerging local systems has thus been absolutely<br />
basic to implementation success, <strong>and</strong> has motivated<br />
the adoption of a three-level implementation<br />
strategy which has permitted valuable<br />
dissemination to be achieved with users who do not<br />
require the full software system (Tab. 2).<br />
Implementation has taken the form of creating <strong>and</strong><br />
updating an annotated listing of all IWRMS output<br />
(data, information products, software tools <strong>and</strong><br />
models) that is available to the end-user. This was<br />
used in discussions with the identified potential<br />
users, <strong>and</strong> was also the basis for compiling the<br />
IWRMS metadata. An inventory of IWRMS data,<br />
output <strong>and</strong> system metadata has been delivered to<br />
major stakeholders. This is the primary method of<br />
attracting new users, <strong>and</strong> is a tangible indication of<br />
implementation. The most likely <strong>and</strong> immediate<br />
major end-users for IWRMS were identified <strong>and</strong><br />
profiled, <strong>and</strong> have been alerted to the three-level<br />
outputs of IWRMS. The full level three IWRMS<br />
was implemented as demonstrations during the<br />
566
final phase of the project in the form of data,<br />
output <strong>and</strong> software tools. In the first instance this<br />
provided feedback, <strong>and</strong> subsequently lay the<br />
foundation for the long-term implementation.<br />
Tab. 2: IWRMS end user <strong>and</strong> dissemination levels<br />
IWRMS level<br />
Level 1<br />
Use of output: data,<br />
maps, reports (e.g. l<strong>and</strong><br />
use, erosion, water<br />
dem<strong>and</strong><br />
Level 2<br />
Use of software tools<br />
<strong>and</strong> applications (e.g.<br />
GIS applications;<br />
ACRU software GIS<br />
based other decision<br />
support tools)<br />
Level 3<br />
Full users of ACRU<br />
enhanced by IWRMS<br />
web solutions including<br />
access to remote<br />
dtabase<br />
6. CONCLUSIONS<br />
Sub-level<br />
Level 1a: Users or information<br />
provider taking a library<br />
reference set of paper materals<br />
without a specific application in<br />
mind<br />
Level 1b: User taking a<br />
particular data product for a<br />
particular application purpose<br />
Level 2a: Enduser<br />
commissioning output form an<br />
IWRMS service provider<br />
Level 2b: Enduser installing<br />
software tools <strong>and</strong> undertaking<br />
their own analysis<br />
No sublevel distinction<br />
This study showed very clearly the necessity of a<br />
close co-operation with end users right from the<br />
beginning of the project, particularly for the<br />
formulation of sensible planning scenarios. To<br />
assess current <strong>and</strong> future water quantity the<br />
hydrological model (ACRU) was integrated into<br />
IWRMS <strong>and</strong> can be considered as a very valuable<br />
tool. Regarding water quality, improvements were<br />
made in terms of sediment load. Other substances<br />
as well as groundwater aspects were not<br />
investigated during this project, but are of high<br />
interest to the local water managers <strong>and</strong> should be<br />
studied in subsequent projects.<br />
Water dem<strong>and</strong> aspects are as important as water<br />
supply, particularly in the light of the new water<br />
laws in the Southern African countries. IWRMS<br />
was able to contribute to these issues in a very<br />
timely way, which was appreciated by the end<br />
users. Optimisation of water allocation between<br />
competing sectors (including environmental issues)<br />
<strong>and</strong> individual users while taking social <strong>and</strong><br />
economical aspects into account is also a very<br />
crucial topic, which has been tackled by IWRMS,<br />
but should be further investigated in future. The<br />
implementation of such a complex management<br />
system requires thorough user analysis <strong>and</strong> a<br />
flexible system concept <strong>and</strong> technical architecture<br />
that is able to h<strong>and</strong>le the various components.<br />
IWRMS has presented a viable approach to fulfil<br />
these requirements <strong>and</strong> to be able to extend the<br />
management system with new tools as they become<br />
available.<br />
7. REFERENCES<br />
Acocks, J.P.H. (1988): Veld types of Southern<br />
Africa. Botanical Research Institute, Pretoria,<br />
RSA, Botanical Survey of South Africa,<br />
Memoirs 57: 1-146.<br />
Bongartz K. (2003) Applying different spatial<br />
Distribution <strong>and</strong> Modeling concepts in three<br />
nested mesoscale catchments of Germany.<br />
Physics <strong>and</strong> Chemistry of the Earth Vol. 28<br />
Issue 33-36, pp.1343-1349.<br />
Flügel, W.-A. (1996): Hydrological Response<br />
Units (HRU's) as modelling entities for<br />
hydrological river basin simulation <strong>and</strong> their<br />
methodological potential for modelling<br />
complex environmental process systems. -<br />
Results from the Sieg catchment., Die Erde,<br />
1996, 127: 43-62.<br />
Märker, M., Moretti, S. & G. Rodolfi (2001):<br />
Assessment of water erosion processes <strong>and</strong><br />
dynamics in semiarid regions of southern<br />
Africa (KwaZulu/Natal RSA; Swazil<strong>and</strong>)<br />
using the Erosions Response Units Concept.<br />
Geogr. Fis. Dinam. Quat., Vol. 24, 71-83.<br />
Murphy, M. & Mitchell, J.F.B. (1995): Transient<br />
response of the Hadley Centre coupled<br />
oceanatmosphere model to increase carbon<br />
dioxide. Part 2. Spatial <strong>and</strong> temporal structure<br />
of the response. Journal of Climate, 8, 57 -<br />
80.<br />
Schulze, R.E. (1995). Hydrology <strong>and</strong><br />
Agrohydrology : A Text to Accompany the<br />
ACRU 3.00 Agrohydrological <strong>Modelling</strong><br />
System. Water Research Commission,<br />
Pretoria, RSA. Report TT69/95.<br />
Taylor, V., Schulze, R.E., Jewitt, G., Pike, A. &<br />
Horan, M.J.C. (2000): An integrated<br />
approach to assessing the management needs<br />
of the water resources of the Mkomazi<br />
catchment, Phase 1: <strong>Modelling</strong> catchment<br />
streamflow generation to assess the impacts<br />
of l<strong>and</strong> use change on available water<br />
resources. ACRUcons Report 36. School of<br />
Bioresources Engineering <strong>and</strong> <strong>Environmental</strong><br />
Hydrology, University of Natal,<br />
Pietermaritzburg, South Africa. pp96.<br />
Further information under:<br />
http://www.geogr.uni-jena.de/index.php?id=1750<br />
567
Possible Courses: Multi-Objective <strong>Modelling</strong> <strong>and</strong><br />
Decision Support Using a Bayesian Network<br />
Approximation to a Nonpoint Source<br />
Pollution Model<br />
David Swayne, Jie Shi<br />
University of Guelph, Computing research Laboratory for the Environment,<br />
Guelph, N1G2W1, CANADA, dswayne@uoguelph.ca<br />
Abstract: <strong>Modelling</strong> systems frequently work in a single domain, such as physical or chemical<br />
process modelling, hydrology or combinations, to simulate process in nature such as pollution transport<br />
or the production of food or manufactured goods. Side-effects of agro-industrial processes, or gains /<br />
losses from production enterprises are separately modelled without the ability to examine trade-offs or<br />
alternatives. Multi-objective modeling attempts to combine "apples <strong>and</strong> oranges" through decision<br />
theoretical principles. Such treatments can couple production <strong>and</strong> waste systems to quantify the<br />
economic cost of remediation. We demonstrate such an application, from the data acquisition, model<br />
calibration through to the hypothesis testing, for a nonpoint source pollution model together with a<br />
yield / energy / revenue model based on corn / grain / meadow rotations typically found in Southern<br />
Ontario, Canada, using realistic economic data obtained from agricultural operations similar to those<br />
found in this region.<br />
Keywords: multicriteria decision support, Bayesian networks, nonpoint source pollution<br />
1. Introduction<br />
Agricultural nonpoint source pollution<br />
(AGNPSP) has been increasingly recognized as<br />
a major contributor to the declining quality of<br />
lakes <strong>and</strong> rivers in Canada (<strong>and</strong> elsewhere).<br />
Many modeling systems have been constructed<br />
to mimic the transport <strong>and</strong> fate of agricultural<br />
chemicals <strong>and</strong> nutrients in surface waters. They<br />
have been very successful, even quantitatively,<br />
but the problems persist. Our research aims to<br />
couple the AGNPSP problem to the economic<br />
side of agriculture, to develop a realistic<br />
estimate of the costs inherent in<br />
environmentally sustainable agricultural<br />
practices.<br />
In order to do this, we have to have a set of<br />
models <strong>and</strong> data for several different problem<br />
areas. For the moment we will consider the<br />
example of crop production instead of food or<br />
dairy animals. Some aspects of those studies<br />
may yield to pollution models that might be<br />
considered to be more like point source. These<br />
issues may eventually be incorporated into<br />
st<strong>and</strong>ard models, but for our demonstration<br />
project we restricted ourselves to st<strong>and</strong>ard crops<br />
that are usually grown in Southern Ontario,<br />
Canada, using st<strong>and</strong>ard means of production<br />
<strong>and</strong> a limited suite of conservation practices.<br />
Our inputs of yield <strong>and</strong> crop pricing were based<br />
on historical yield data, <strong>and</strong> in some cases on<br />
generating a distribution of prices for the<br />
particular commodities that was based on<br />
ranges containing historical prices. Our erosion<br />
<strong>and</strong> sediment data were obtained from<br />
experimental work in Canada, particularly on<br />
the GAMES model [Rudra 1986]. Our model<br />
strategy does not depend on the internal<br />
mechanics of the model chosen, i.e. the<br />
methodologies of model construction would not<br />
change if a new model were substituted that<br />
was capable of generating similar data.<br />
2. Background<br />
Sustainable development, is defined as<br />
economic development that meets the needs of<br />
the present without compromising the ability of<br />
future generations to meet their needs<br />
[Hermanides 1987; Cornelissen 2000]. In other<br />
words, a proper environmental policy objective<br />
should consider three dimensions: economic,<br />
568
social <strong>and</strong> environmental aspects. Conflict<br />
exists between them. There is no way to satisfy<br />
all of the criteria. For agriculture l<strong>and</strong> use<br />
management policy, the two major paradigms<br />
are: yield-oriented policy (conventional,<br />
production - driven), <strong>and</strong> environment -<br />
oriented policy (environmentally - driven)<br />
[Huylenbroeck 1996]. Recent research has<br />
investigated long-term sustainable l<strong>and</strong> use<br />
management [Bessembinder 1996; El-Swaify<br />
1996]. Most of the work is on the l<strong>and</strong> use<br />
arrangement/planning, or production analysis. It<br />
usually uses linear programming, logistic<br />
regression analysis, or dynamic programming<br />
techniques to optimize a criteria function<br />
[Bessembinder 1996; Hipel 1996] for resolving<br />
conflict. The trade-off process used to solve the<br />
conflicts between economic <strong>and</strong> environmental<br />
objectives in a rural planning project<br />
[Huylenbroeck 1996], is described as having<br />
three steps: 1) separate aggregation of the<br />
economic <strong>and</strong> ecological criteria; 2)<br />
visualization of trade-offs; <strong>and</strong>, 3) discrete<br />
compromise analysis to support the final choice<br />
A substantial amount of recent research has<br />
been carried out to develop decision support<br />
tools for the management of agro-forestry<br />
resources. Among these tools the Multiple<br />
Criteria Decision-Making (MCDM) approach<br />
plays a prominent role [Marangon 1998]. Most<br />
of these tools perform decision analysis based<br />
on calculating the result by assigning<br />
weights/scores [Parton, 1996; Yakowitz 1996;<br />
Noghin 1997; Wang 1998; Tan 1998; Bots<br />
2000; Costa 2001; Janssen, 2001] to the<br />
alternative criteria attributes, or by assigning a<br />
probabilistic proportion to a decision tree<br />
structure [Gratch 1995; Warburton 1998]. Some<br />
have linked their Expert System with a<br />
Geographical Information System (GIS)<br />
database [Abu-Zeid 1996; Crosetto 2000; Rao<br />
2001]. Since the 1980s, a probability-utility<br />
based approach using Influence Diagrams<br />
(Decision Networks) [Tatman 1990] has<br />
become accepted as an efficient alternative for<br />
many classes of models [Howard 1984;<br />
Shachter 1986].<br />
Influence Diagrams are a class of graphical<br />
modelling paradigm that can represent<br />
probabilistic inference <strong>and</strong> decision analysis<br />
models. The reasons that an Influence Diagram<br />
is an effective modelling framework for a<br />
diverse array of problems involving probability<br />
are: a) it captures both the structural <strong>and</strong><br />
qualitative aspects of the decision problem <strong>and</strong><br />
serves as the framework for an efficient<br />
quantitative analysis of the problem; b) it<br />
allows efficient representation <strong>and</strong> exploitation<br />
of the conditional independence in a decision<br />
model; <strong>and</strong>, c) it has proven to be an effective<br />
tool for not only communicating decision<br />
models among decision analysts <strong>and</strong> decision<br />
makers, but also for communicating between<br />
the analyst <strong>and</strong> the computer.<br />
3. Criteria<br />
Watershed pollution comes from many sources.<br />
We emphasize agriculture l<strong>and</strong> use activities<br />
because much of the contamination of surface<br />
waters is due to nonpoint Source (NPS)<br />
pollution. According to the USEPA [Osmond<br />
1996], approximately 60% of the total NPS<br />
pollution load on assessed surface waters is due<br />
to agricultural runoff. The primary agricultural<br />
pollutants are sediment, nutrients <strong>and</strong><br />
pesticides. For decision procedures these issues<br />
are the set of available options, the criteria <strong>and</strong><br />
the uncertainty on the outcomes of each option.<br />
Each watershed is unique in its physical<br />
characteristics, l<strong>and</strong> uses, water resources,<br />
socioeconomic status, <strong>and</strong> public concerns.<br />
Generally speaking, decision - making involves<br />
the need to evaluate a finite number of possible<br />
choices (alternatives/c<strong>and</strong>idates) based on a<br />
finite number of attributes (criteria). In our<br />
research, the decision alternatives are seven<br />
cropping tillage methods (Table 1) arranged<br />
into nine multi-year scenarios. The criteria<br />
attributes are soil erosion rate, sediment yield<br />
(local <strong>and</strong> total in the whole watershed),<br />
operating net revenue (price*yield – energy,<br />
labour <strong>and</strong> capital cost). These latter (cost)<br />
criteria estimates are defined as ranges (where,<br />
for example the energy inputs for a particular<br />
strategy are based on estimates from the<br />
literature, <strong>and</strong> sub-divided in our model to<br />
reflect variance within a particular range, <strong>and</strong><br />
increased for no-till versus normal cropping.<br />
Our construction combines environmental<br />
pollutant transport models with economic (crop<br />
yield, expense, revenue) models to investigate a<br />
sustainable tillage system in order to encourage<br />
the adoption of improved management systems.<br />
A user interface was developed to provide the<br />
communication tool which links the decision<br />
maker’s interaction with the graphic probability<br />
model. The application enables the user to view<br />
the impact of parameters (such as sediment<br />
yield, erosion rate) on decision alternative<br />
scenarios. It can also aid the user in making<br />
long term soil productivity predictions.<br />
4. Belief networks<br />
A belief network (also called Bayesian network,<br />
or probabilistic network) is a graphical<br />
presentation of probability combined with<br />
mathematical inference calculation. It is used to<br />
represent dependencies between r<strong>and</strong>om<br />
variables. Each variable represented as node, is<br />
connected by directed links, represented as<br />
569
arrows or arcs, with conditional probability<br />
table (CPT) values assigned to the variables<br />
making up a belief network. The nodes in a<br />
belief network are called chance nodes. Chance<br />
nodes represent uncertain events or variables.<br />
They can be a continuous or discrete r<strong>and</strong>om<br />
variable, or a set of events. A deterministic<br />
node is a special case of chance node, which<br />
operates deterministically on other nodes. The<br />
arrows are the directed links between variables<br />
(nodes) <strong>and</strong> this direction represents the<br />
conditional dependent relationship of these<br />
nodes. For example, an arrow entering a chance<br />
node means that the author’s probability<br />
assignment represented by the chance node is<br />
conditional on the node at the other end of the<br />
arrow (its input).<br />
interconnection (e.g. Figure 1). The conditional<br />
probability distributions for the probability<br />
model came from Monte Carlo simulations<br />
using GAMES. The model provides results at<br />
both field <strong>and</strong> watershed scales <strong>and</strong> examines<br />
soil loss <strong>and</strong> sediment transport on a seasonal<br />
basis. Figure 2 depicts a partial integration of<br />
GAMES into the combined model, for a<br />
particular cell or plot X from Figure 1.<br />
Figure 1. Approximate layout of the STRAT<br />
watershed. The actual layout has 471 nodes<br />
each representing a homogeneous plot.<br />
5. Influence Diagrams<br />
Influence Diagrams (ID) are the extension of<br />
belief networks. In addition to nodes for<br />
representing r<strong>and</strong>om variables provided by<br />
belief networks, Influence Diagram also<br />
provides decision nodes for modeling<br />
alternatives <strong>and</strong> utility nodes for the utility<br />
evaluation. A decision node indicates a decision<br />
facing the decision maker—similar to decision<br />
nodes in decision trees. An arrow entering a<br />
decision node means that the author’s decision<br />
is made with knowledge or the outcome of the<br />
uncertain quantity at the other end of the arrow.<br />
The utility nodes represent the utility function<br />
of the decision maker. Utilities are associated<br />
with each of the possible outcomes of the<br />
decision problem modelled by the Influence<br />
Diagram. Influence Diagrams are useful <strong>and</strong><br />
powerful tool for modelling a decision problem.<br />
They can be used to model both simple decision<br />
problems (only one decision node) <strong>and</strong><br />
sequential decision problems ( more than one<br />
decision nodes <strong>and</strong> utility nodes). The later is<br />
also known as dynamic decision modelling. In<br />
this case, the next decision always depends on<br />
the previous decision or states.<br />
While at Guelph, Dorner [2000] built a belief<br />
network based on simulation data of running<br />
the GAMES model. The nodes <strong>and</strong> the<br />
relationships between the nodes were obtained<br />
from GAMES. Based on the universal soil loss<br />
equation, GAMES requires as its input: a<br />
segmentation of the watershed into individual<br />
homogeneous plots of l<strong>and</strong> based on slope <strong>and</strong><br />
aspect ratio, area of plot, soil type, so-called<br />
cropping factor (the erodibility of soil for a<br />
particular crop – essentially a l<strong>and</strong> use factor),<br />
<strong>and</strong> precipitation information for the region.<br />
The plots are connected in a dendritic drainage<br />
network, from which the individual model<br />
components in the GAMES model inherit their<br />
Figure 2. GAMES layout of the plot X in<br />
Figure 1(above at lower right), with cells Y <strong>and</strong><br />
Z upslope <strong>and</strong> cell W downslope<br />
X<br />
Y<br />
Z<br />
Crop Rotation<br />
Factors<br />
Delivered<br />
Sediment from<br />
Upstream<br />
Delivered<br />
Sediment from<br />
Upstream<br />
Crop <strong>and</strong><br />
Management<br />
Practices<br />
Erosion<br />
Sediment Leaving<br />
the Field<br />
Incoming<br />
Sediment Leaving<br />
Field<br />
Labor Input<br />
Production<br />
Total Sum of<br />
Sediment Leaving<br />
Filed<br />
Cell Delivery<br />
Ratio<br />
The watershed examined in this the project was<br />
Stratford Avon Upper Watershed (Figure 1),<br />
which was part of a comprehensive study for<br />
the Great Lakes Pollution from L<strong>and</strong> Use<br />
Activities (PLUARG) in the late 1970s [Wall<br />
1978; Wall 1979]. STRAT is an upl<strong>and</strong><br />
watershed where erosion rates vary significantly<br />
over the entire watershed. It has an area of 537<br />
hectares with l<strong>and</strong> slopes up to 9% [Dickinson<br />
1990]. Most of the surface material is<br />
comprised of s<strong>and</strong>y hills.<br />
W<br />
570
The STRAT watershed is rural, primarily in<br />
cropl<strong>and</strong>. The primary rotations are<br />
hay(meadow) / grain / corn, grain / corn or hay /<br />
grain. The soil types are loamy soils or clay<br />
loam.<br />
6. Utility<br />
In order to determine the desirability of an<br />
outcome to the decision maker, a value is<br />
assigned to each outcome. The term ‘utility’ is<br />
used in sense of the quality of being useful. The<br />
following equation is used to calculate the<br />
expected utility EU(A|E) of action A given<br />
evidence E.<br />
EU(A|E) =<br />
= Σ i (P(Result i (A)|E, Do(A))*U(Result i (A) ))<br />
the extended formula for expected utility after<br />
new evidence E j is:<br />
EU(A|E,E j ) =<br />
= Σ i (P(Result i (A)|E, Do(A),E j )*U(Result i (A) ))<br />
Result i (A) are the possible outcome states after<br />
executing a nondeterministic action A.<br />
U(S) denotes the utility of state S. Do(A) is the<br />
proposition that action A is executed in the<br />
current state. An action A will have possible<br />
outcome states for Result i (A), where the index i<br />
ranges over the different outcomes. .<br />
Maximum Expected Utility (MEU) is the<br />
fundamental idea of decision theory. This<br />
means that a decision is rational if, <strong>and</strong> only if,<br />
it chooses the action that yields the highest<br />
expected utility averaged over all the possible<br />
outcomes of the action.<br />
MEU(A|E) =<br />
max AΣ i (P(Result i (A)|E,Do(A))*U(Result i (A) ))<br />
7. Experiments<br />
The principal experiments involving<br />
environmental effects are based on nine yearly<br />
crop tillage management rotation scenarios<br />
(Table 1) We analyzed the relationships<br />
between erosion on various management<br />
strategies with the long-term economic income.<br />
Briefly, CORN is the North American name for<br />
maize using st<strong>and</strong>ard non-conservation<br />
procedures, GRAINS are cereal crops (eg.<br />
wheat), conservation tillage (CONS_TILL) is<br />
the technique of planting without plowing,<br />
CORN_X_SLOPE refers to corn rows<br />
contouring the slope to minimize erosion, <strong>and</strong><br />
MEADOW refers to leaving a field untilled for<br />
a season.<br />
We conducted experiments on economic,<br />
environmental <strong>and</strong> integrated models. The<br />
analysis is broadly divided into three parts:<br />
local short-term policy; local long-term policy;<br />
<strong>and</strong>, whole watershed policy analysis. We have<br />
produced a prodigious set of results, so only a<br />
few examples are discussed here.<br />
Table 1. Crop Rotation Scenarios<br />
Scenario<br />
Number<br />
Scenario Name Abbreviation<br />
#1 CORN-CORN CornCorn<br />
#2 CORN-GRAINS CornGrains<br />
#3 CORN-MEADOW CornMeadow<br />
#4 CORN_X_SLOPE-GRAINS CrossGrains<br />
#5<br />
CORN_X_SLOPE-<br />
CORN_X_SLOPE<br />
CrossCross<br />
#6<br />
CORN_CONS_TILL-<br />
CORN_CONS_TILL<br />
TillTill<br />
#7<br />
CORN_CONS_TILL-<br />
GRAINS<br />
TillGrains<br />
#8 CORN_GRAINS-MEADOW CornGrainMeadow<br />
#9<br />
CORN_CONS_TILL-<br />
MEADOW<br />
TillMeadow<br />
For the first experiment, we chose to analyze<br />
the results for ten cells from the network<br />
selected for a range of to their erodibility. All<br />
471 cells were involved, but we extracted<br />
results for the selected cell numbers for detailed<br />
examination. We ran nine alternative tillage<br />
management scenarios.<br />
The TillTill management practice produces the<br />
lowest erosion rate for continuous cropping. If<br />
the erosion rate is in excess of a tolerable rate,<br />
however, even good policy cannot change the<br />
injury to the field. This result verified Pierce et<br />
al.[1983] that a field with an erodibility higher<br />
than 5.0 ton/ha should be put into fallow for<br />
recovery. Otherwise the injury is unrecoverable.<br />
Figure 3. Experiment 2, total number of years<br />
of productivity.<br />
The next set of experiments were more longterm,<br />
<strong>and</strong> looked at earning potential over long<br />
stretches, as well as years of soil productivity<br />
left in the fields tested. Figure 3 shows one part<br />
571
of that result, productive years remaining in the<br />
field under continuation of the scenarios.<br />
The third set of experiments, on the whole<br />
watershed, explored point policy versus range<br />
policy options. For point policy, erosion rates<br />
were set, <strong>and</strong> the system in the field <strong>and</strong><br />
upslope was adjusted to meet the policy<br />
objective. In the range policy options, erosion<br />
rates for several fields were left in an acceptable<br />
range, <strong>and</strong> the system made decisions only<br />
when the permissible range was exceeded. We<br />
determined that the point policy produced more<br />
satisfactory results, in general, when compared<br />
with the range policy.<br />
The economic effects of the various policy<br />
options are difficult to represent in the limited<br />
space of this paper. In general, field conditions<br />
of low erosion represent opportunities for<br />
economical production. Therefore the<br />
conversion cost for environmental conservation<br />
practices is very high, when compared to the<br />
same effort for plots with high erosion<br />
potential. This follows common sense, when<br />
more gain is realized from less effort.<br />
Immediate economic effects are evaluated by<br />
noting the difference in yield (translating into<br />
revenue less expenses). Long-term economic<br />
costs <strong>and</strong> benefits are calculated from pursuing<br />
policy year-to-year to extinction, calculating<br />
total differences over a range of revenue<br />
models. This latter analysis is still problematic,<br />
since the discounting into the future is anyone’s<br />
guess.<br />
8. Conclusions<br />
A decision tree does not have the necessary<br />
structure to contain all the probabilistic<br />
information needed to answer value-ofinformation<br />
questions. The joint probability<br />
distribution of all the uncertain variables is<br />
needed <strong>and</strong> this is more easily <strong>and</strong> naturally<br />
captured in Influence Diagram. The efficiency<br />
<strong>and</strong> speed using the ID is unchallenged. In our<br />
specific problem, the decision network had 12<br />
nodes for each sub-network (field) <strong>and</strong> a total of<br />
over 471 sub-networks for a whole watershed.<br />
The utility function had 5 variables <strong>and</strong> each<br />
variable had 3 to 20 attributes (discrete<br />
variables) or value ranges (continuous<br />
variables). The CPT (conditional probability<br />
table) of the utility had 181,440 entries.<br />
If we use a traditional optimization algorithm, it<br />
would be a long computation. With ID, the user<br />
can interact directly <strong>and</strong> adjust the complex<br />
parameters to get a better decision based on the<br />
specific criteria preference.<br />
Research using weighted multicriteria methods<br />
reaches conflicting conclusions regarding the<br />
alternatives’ effects on the economic returns<br />
<strong>and</strong> the environment. Heilman et al, [1996]<br />
shows that no-till tillage can improve economic<br />
returns to the farmer <strong>and</strong> have positive off-site<br />
benefits. But Prato <strong>and</strong> Fulcher [1996] report<br />
that no-till tillage can reduce the sediment yield<br />
but produces the lowest net return among all the<br />
alternatives. Our result suggests that the<br />
conservation tillage management might be a<br />
suitable choice of the alternatives for a longterm<br />
policy, because it satisfies both<br />
environmental <strong>and</strong> farm’s needs. Our analysis<br />
indicates that conservation tillage management<br />
technique has a limit to potential benefits. If the<br />
field’s soil condition is too bad (if the erosion<br />
rate is already in excess of an appropriate<br />
range), there is no technique available to help<br />
the field recover from the damage. The<br />
tolerance erosion rate in our simulations is 7.5<br />
ton/ha. This agrees with the reported result by<br />
Pierce [1983]. If conventional tillage<br />
management is used, the field can only last for<br />
less than 100 years of cropping with top soil<br />
depth 38.5 cm. But with conservation tillage<br />
management the profitable years can last more<br />
than three times that.<br />
Our research suggests also that we can build a<br />
comprehensive model that contains most of the<br />
relevant environmental <strong>and</strong> economic factors<br />
for environmentally responsible decisionmaking.<br />
Acknowledgements<br />
The authors are grateful for the hard work of the<br />
session organizers <strong>and</strong> the referees. The<br />
generous support of the Canadian Natural<br />
Sciences <strong>and</strong> Engineering Research Council is<br />
acknowledged.<br />
References<br />
Abu-Zeid, K.M., 1996. A multicriteria decision<br />
support system for evaluating cropping<br />
pattern strategies in Egypt. Proc. First Inter.<br />
Conf. on Multiple Objective Decision<br />
Support Systems (MODSS): 105-120.<br />
Bessembinder, J.J.E., M.K.V. Ittersum, R.<br />
Rabbinge <strong>and</strong> L.O. Fresco, 1996. Long-term<br />
explorations of sustainable l<strong>and</strong> use.<br />
MODSS: 437-445.<br />
Bots, P.W.G., Hukshof, J.A.M., 2000.<br />
Designing multi-criteria decision analysis<br />
processes for priority setting in health policy.<br />
J. Multi-Crit. Decis. Anal. 9:56-75<br />
Cornelissen, A.M.G., J. van den Berg, W.J.<br />
Koops, M. Grossman <strong>and</strong> H.M.J. Udo, 2000.<br />
Assessment of sustainable development: a<br />
novel approach using fuzzy set theory. ERIM<br />
Report Series Research in Management: 1-17.<br />
Costa, C.A.B. 2001. The use of multi-criteria<br />
decision analysis to support the search for<br />
less conflicting policy options in a multi-actor<br />
572
context: case study. J. Multi-Crit. Decis.<br />
Anal. 10: 111-125.<br />
Crosetto, M., S. Trarantola, <strong>and</strong> A. Saltelli,<br />
2000. Sensitivity <strong>and</strong> uncertainty analysis in<br />
spatial modeling based on GIS. Agriculture,<br />
Ecosystems <strong>and</strong> Environment 81: 71-79<br />
Dickinson, W.T., R.P. Rudra <strong>and</strong> G.J. Wall,<br />
1990. Targeting remedial measures to control<br />
non-point source pollution. Water Resources<br />
Bulletin, 26(3):499-507.<br />
Dorner, S., D.A. Swayne, R.P. Rudra, C. Pal<br />
<strong>and</strong> C. Newald, 2001. Integrating parametric<br />
uncertainty <strong>and</strong> modeling results into an<br />
advisory system for watershed management.<br />
Elsevier. advances in environmental research.<br />
5(4): 445-451.<br />
El-Swaify, S.A. & D.S. Yakowitz, 1996.<br />
<strong>Environmental</strong> Management: Concepts,<br />
Approaches, <strong>and</strong> Applications MODSS: 3-7.<br />
Gratch, J.M., 1995. Sequential inductive<br />
learning. In: Report (University of Illinois at<br />
Urbana-Champaign. Dept. of Computer<br />
Science) UIUCDCS-R-95-1908. Urbana, Ill.<br />
Heilman, P., J.L. Leonard, D.S. Yakowitz.<br />
1996. Quantifying economic incentives<br />
needed for control of nonpoint source<br />
pollution in agriculture. MODSS: 513-526.<br />
Hermanides ,G. <strong>and</strong> P. Nijkamp, 1998.<br />
Multicriteria evaluation of sustainable<br />
agricultural l<strong>and</strong> use: a case studey of Lesvos.<br />
In: Multicriteria Analysis for L<strong>and</strong>-Use<br />
Management. Kluwer: 61-78.<br />
Hipel, K.W., D.M. Kilgour, L.Fang <strong>and</strong> X.<br />
Peng, 1996. Using the decision support<br />
system GMCR for resolving conflict in<br />
resources management. MODSS'96: 23-46.<br />
Howard, R.A. <strong>and</strong> Matheson, J.E., 1984.<br />
Influence Diagrams. Readings on the<br />
Principles <strong>and</strong> Applications of Decision<br />
Analysis. Menlo Park, CA: Strategic<br />
Decision Group. Vol. II: 719-762<br />
Huylenbroeck, G.V., 1996. A multicriteria<br />
approach for trade-off analysis between<br />
economic <strong>and</strong> environmental objectives in<br />
rural planning. MODSS: 419-435.<br />
Janssen, R., 2001. On the use of multi-criteria<br />
analysis in environmental impact assessment<br />
in the Netherl<strong>and</strong>s. J. Multi-Crit. Decis.<br />
Anal. 10:101-109.<br />
Marangon, F. & Tempesta, T., 1998. Rural<br />
l<strong>and</strong>scape <strong>and</strong> economic results of the farm; a<br />
multi-objective approach. In: Multicriteria<br />
Analysis for L<strong>and</strong>-Use Management..<br />
Kluwer:49-60.<br />
Noghin, V., 1997. Relative importance of<br />
criteria: a quantitative approach. J. Multi-Crit.<br />
Decis. Anal. 6:355-363<br />
Parton. K.A <strong>and</strong> A. Nissapa, 1996. Use of goal<br />
programming in aquaculture policy decision<br />
making in southern Thail<strong>and</strong>. MODSS 1996,<br />
pp 641-653.<br />
Pierce, F.J., W.E. Larson, R.H. Dowdy, W.A.P.<br />
Graham, 1983. Productivity of soils:<br />
Assessing long term changes due to soil<br />
erosion. J. Soil Water Cons. 38(1):39-44.<br />
Prato, Tony, Chris Fulcher, Shunxiang Wu, <strong>and</strong><br />
Jian Ma. Multiple-Objective Decision<br />
Making for Agroecosystem Management.<br />
Agricultural <strong>and</strong> Resource Economics Review<br />
25: 2 (1996): 200-212.<br />
Rao, K.S. <strong>and</strong> R. Pant, 2001. L<strong>and</strong> use<br />
dynamics <strong>and</strong> l<strong>and</strong>scape change pattern in a<br />
typical micro watershed in the mid elevation<br />
zone of central Himalaya, India. Decision<br />
Sciences Web Journals. Agriculture,<br />
Ecosystems & Environment. 86:(2) 113-124.<br />
Rudra, P.R., W.T. Dickinson, D.J. Clark <strong>and</strong><br />
G.J. Wall. 1986. Games – A Screening Model<br />
of Soil Erosion <strong>and</strong> Fluvial Sedimentation in<br />
Agricultural Watersheds. Canadian Water<br />
Resources Journal, 11(4):58-71.<br />
Shachter, R. 1986. Evaluating influence<br />
diagrams. Operations Research, 43: 871-882.<br />
Swayne, D.A, 2003. Putting the decision in<br />
decision support system. Proc. IEMSs: 45-52.<br />
A. Rizzoli <strong>and</strong> A.J. Jakeman (eds).<br />
Tan, Y.S. <strong>and</strong> N.M. Fraser, 1998. The modified<br />
star graph <strong>and</strong> the petal diagram: two new<br />
visual aids for discrete alternative<br />
multicriteria decision making. J. Multi-Crit.<br />
Decis. Anal. 7: 20-33<br />
Tatman, J.A. <strong>and</strong> Shachter R.D., 1990.<br />
Dynamic Programming <strong>and</strong> Influence<br />
diagrams. IEEE Transactions on System,<br />
Man <strong>and</strong> Cybernetics, 20(2):365-379<br />
Wall, G.J., L.J.P.V. Vliet, <strong>and</strong> W.T. Dickinson,<br />
1978. <strong>International</strong> reference group on great<br />
lakes pollution from l<strong>and</strong> use activities:<br />
Contribution of sediment to the Great Lakes<br />
from agricultural activities in Ontario. IJC<br />
technical report re: PLUARG, <strong>International</strong><br />
Joint Commission, Windsor, Ontario.<br />
Wall, G.J., W.T. Dickinson <strong>and</strong> L.P.J. van<br />
Vliet, 1979. Agricultural sources of fluvial<br />
suspended sediments. Progress in Water<br />
Technology, 11(6):481-499<br />
Wang, M. <strong>and</strong> Yang, J., 1998. A multi-criterion<br />
experimental comparison of three multiattribute<br />
weight measurement methods. J.<br />
Multi-Crit. Decis. Anal. 7: 340-350<br />
Warburton, A., 1998. The decision tree<br />
polytope <strong>and</strong> its application to sequential<br />
decision problem. J. Multi-Crit. Decis. Anal.<br />
7: 330-339<br />
Yakowitz, D.S. A Multiattribute Tool for<br />
Decision Support: Ranking a Finite Number<br />
of Alternatives. MODSS: 205-215.<br />
573
A Spatial DSS for South Australia's Prawn Fisheries.<br />
Using Historic Knowledge Towards <strong>Environmental</strong> <strong>and</strong><br />
Economical Sustainability<br />
B. Ostendorf <strong>and</strong> N. Carrick<br />
School of Earth <strong>and</strong> <strong>Environmental</strong> Sciences, University of Adelaide, Glen Osmond, South Australia, 5064.<br />
Abstract: The Spencer Gulf Penaeid prawn fishery is an example of a sustainable fishery due to close<br />
collaboration between fishers, research <strong>and</strong> government. The fishery has undergone substantial increase in<br />
fishing efficiency due to improvement in gear, increase in crew skill, effective use of communication<br />
networks for monitoring catch, stock assessment, <strong>and</strong> for rapid response for change in harvest strategies.<br />
Spatial decision-making has reduced the fishing effort to around 60 days per year <strong>and</strong> less than 10% of the<br />
area of the Gulf is trawled, with increasing economic gain due to development of real time adaptive harvest<br />
strategies undertaken in collaboration with the fisheries industry. The reduced trawl time <strong>and</strong> fishery<br />
closures, which have been adapted, have important implications for the environment. This presentation<br />
describes the process of spatial decision-making <strong>and</strong> the utility of spatial information techniques using<br />
historic spatial data in conjunction with near real-time survey data <strong>and</strong> statistical risk assessment. The system<br />
is implemented linking an Oracle database to ArcGIS, Genstat <strong>and</strong> Splus <strong>and</strong> mobile phone technologies.<br />
Keywords: decision support systems; fishery; Oracle; GIS; Statistical models<br />
1. INTRODUCTION<br />
There are three prawn fisheries in South Australia<br />
namely Spencer Gulf, Gulf St Vincent <strong>and</strong> the<br />
West Coast all of which are based exclusively on<br />
the Western king prawn Melicertus latisulcatus<br />
(Penaeidae). (Figure 1).<br />
The Spencer Gulf fishery is the largest Australian<br />
producer of Western king prawn (Carrick 2002),<br />
<strong>and</strong> is one of 5 Australian commercial trawl<br />
fisheries that produce more than 1,500 tonnes per<br />
annum (Tab. 1).<br />
It has long been recognized by the industry that<br />
sustainability can only be maintained if the fishing<br />
effort is limited in space <strong>and</strong> time. Such limitations<br />
are often difficult define <strong>and</strong> personal interests may<br />
outweigh the interest for the broader good. This is<br />
where objective decision support is needed the<br />
most.<br />
Currently, the industry is self-regulated, the<br />
government bodies play a rather observing role but<br />
with strong rights to interfere if necessary. This has<br />
not been necessary since the fleet is relatively<br />
small with only 39 licenses (vessels) operating <strong>and</strong><br />
a personal contact between most fishermen.<br />
Government authorities have worked in close<br />
collaboration with the industry since it has been<br />
possible to demonstrate the benefits of reducing the<br />
fishing effort.<br />
Figure 1: Geographical location of South<br />
Australia’s prawn fisheries<br />
Yet discussions about how to limit the spatial<br />
extent (closures) <strong>and</strong> the periods of fishing are<br />
recurring. In this paper we give an overview of the<br />
574
management <strong>and</strong> describe the decision support tool<br />
that was developed as part of a project funded<br />
jointly by the Fisheries Research <strong>and</strong> Development<br />
Cooperation, the Industry, the University of<br />
Adelaide <strong>and</strong> the government authority<br />
(PIRSA/SARDI, Primary Industry <strong>and</strong> Resources<br />
South Australia / South Australian Research <strong>and</strong><br />
Development Institute)<br />
Table 1: Australian prawn fishery Catch statistics<br />
for Western king prawns (Melicertus latisulcatus)<br />
Fishery Vessels Latitude Longitude Catch<br />
(tonnes)<br />
Spencer Gulf 39 34° 00’ S 137° 30’ E 1,600 –<br />
2,500<br />
Gulf St Vincent 10 35° 00’ S 138° 10’ E 250 – 400<br />
West Coast 3 33° 30’ S 135° 45’ E 100 – 120<br />
Shark Bay,<br />
Western<br />
Australia<br />
Exmouth Gulf,<br />
Western<br />
Australia<br />
Broome,<br />
Western<br />
Australia<br />
Northern<br />
Prawn<br />
Nichol Bay,<br />
Western<br />
Australia<br />
27 25° 30’ S 114° 00’ E 1,150 –<br />
990<br />
13 22° 00’ S 114° 20’ E 400<br />
5 18° 00’ S 122° 00’ E 100<br />
150 15° 00’ S 136° 00’ E 41<br />
12 20° 20’ S 117° 00’ E 43<br />
2. Management Of The Fishery<br />
2.1 Scope of the management plan<br />
The Fisheries Act 1982 provides the statutory<br />
framework to ensure the management <strong>and</strong><br />
ecologically sustainable development of South<br />
Australia’s marine <strong>and</strong> freshwater fisheries<br />
resources. South Australia has management<br />
jurisdiction for western king prawns, from the low<br />
water mark out to 200 nautical miles in the waters<br />
adjacent to South Australia. The regulations that<br />
govern the management of the South Australian<br />
prawn fisheries are established in the Scheme of<br />
Management (Prawn Fisheries) Regulations 1991<br />
<strong>and</strong> the Fisheries (General) Regulations 2000.<br />
Management plans cover commercial fishing<br />
activity for prawns (Melicertus latisulcatus)<br />
undertaken within South Australian waters <strong>and</strong><br />
provide a statement of the policy framework <strong>and</strong><br />
management strategies.<br />
The prawn management plans do not form part of<br />
the Scheme of Management (Prawn Fisheries)<br />
Regulations 1991 <strong>and</strong> do not have any statutory<br />
basis. The powers contained in Section 14 of the<br />
Fisheries (Management Committees) Regulations<br />
1995 provide the legal basis for the preparation of<br />
management plans. Responsibility for the<br />
preparation of management plans rests with<br />
individual Fishery Management Committees<br />
(FMC’s).<br />
The Spencer Gulf <strong>and</strong> West Coast Prawn Fisheries<br />
Management Plan operates for a five-year period<br />
(from 1998 to 2003), subject to annual review <strong>and</strong><br />
amendments considered necessary by Minister for<br />
Agriculture, Food <strong>and</strong> Fisheries, the Prawn FMC<br />
or the Director of Fisheries.<br />
2.2 Objectives of the management plan<br />
The priority for management of the prawn fisheries<br />
is to ensure that the fishery is sustainable so that<br />
future generations may benefit from exploitation of<br />
the resource. Commensurate with this priority are<br />
a number of more specific biological, economic,<br />
environmental <strong>and</strong> social objectives that have been<br />
developed by the Prawn FMC to complement the<br />
broad directives of section 20 of the Fisheries Act<br />
1982.<br />
McDonald (1998) provides an outline of the<br />
Management plan for the Spencer Gulf Prawn<br />
Fishery. The primary management objectives for<br />
the Spencer Gulf fishery are:<br />
• To maintain the biomass within historical<br />
levels <strong>and</strong> eliminate risk of recruitment<br />
decline due to over-fishing;<br />
• To ensure harvesting procedures are directed<br />
towards optimising size at capture;<br />
• To maintain <strong>and</strong> enhance the profitability of<br />
the fishery by optimising prawn size, market<br />
timing, minimising the costs of fishing <strong>and</strong><br />
the administrative costs of managing the<br />
fishery; <strong>and</strong><br />
• To minimise bycatch <strong>and</strong> trawl impact to the<br />
benthos through development of more<br />
effective <strong>and</strong> efficient gear <strong>and</strong> harvesting<br />
strategies.<br />
The management plan provides a statement of the<br />
policy, objectives <strong>and</strong> strategies to be employed for<br />
the sustainable management of the Spencer Gulf<br />
prawn fishery.<br />
2.3 Reference Points <strong>and</strong> Performance<br />
Indicators<br />
Reference points provide a quantitative measure of<br />
a performance indicator that is used as a<br />
benchmark of performance against objectives, <strong>and</strong><br />
can be used to trigger a management response.<br />
They are agreed <strong>and</strong> quantitative measures used to<br />
assess performance of the fishery based on defined<br />
management objectives. Caddy <strong>and</strong> McMahon<br />
(1995) <strong>and</strong> others have provided detailed<br />
background into the conceptual <strong>and</strong> applied<br />
aspects of reference points for fisheries<br />
management. Reference points allow a decision<br />
575
framework to be developed, however, reference<br />
points <strong>and</strong> performance indicators need to be<br />
updated <strong>and</strong> refined regularly. There are two types<br />
of reference points for rational exploitation of<br />
fisheries namely:<br />
• Target reference points. These are indicators,<br />
which are considered the most desirable<br />
target from a fishery management<br />
perspective.<br />
• Limit reference points. These are threshold<br />
levels, which warn that action is needed to<br />
rectify the fishery before it suffers longerterm<br />
productivity decline.<br />
Morgan (1996) recommended a number of<br />
biological reference points for the Spencer Gulf<br />
prawn fishery, which have been adopted by the<br />
government (PIRSA). The biological reference<br />
points consist of the following environmental <strong>and</strong><br />
economic categories:<br />
Sustainability:<br />
• Maintain exploitation rates at present levels<br />
of effort. The target reference point for<br />
effective effort is between 70-80 fishing<br />
nights while the limit reference point is 80<br />
nights. Effective effort is a function of the<br />
amount of trawl effort (hours, days) <strong>and</strong> the<br />
fishing power (or catching efficiency) of the<br />
fleet.<br />
• Maintain at least 50 percent of the virgin<br />
spawning biomass. This target indicator is the<br />
level of the recruit of the year spawning<br />
biomass, which remains after fishing <strong>and</strong> is<br />
assessed in November-December. The limit<br />
reference point for protecting the resource is<br />
that exploitation should not reduce the stock<br />
to a level of 40 percent.<br />
• Maintain the recruitment index at a level,<br />
which ensures suitable recruitment to the<br />
fishery. The reference point is based on<br />
assessment of recruits to grounds in the<br />
period February to April of each year. The<br />
levels set by Morgan (1996) are the numbers<br />
of prawns (male <strong>and</strong> female)
the main port at Wallaroo. The fleet work in the<br />
night with information exchanged over night or<br />
early morning. The real time management system<br />
is in collaboration between Government<br />
(PIRSA/SARDI) <strong>and</strong> industry to ensure fishing<br />
operations are sustainable <strong>and</strong> economically<br />
efficient. The system utilises trawl survey data,<br />
information from a committee at sea, <strong>and</strong><br />
modelling to determine optimum utilisation of the<br />
resource. A coordinator at sea has an important<br />
role in discussion <strong>and</strong> communicating information<br />
to the fleet <strong>and</strong> coordinating operations with shorebase<br />
maintained by a PIRSA/SARDI fishery<br />
scientist who develops harvest strategies in<br />
collaboration with the Committee at Sea <strong>and</strong> the<br />
FMC.<br />
Figure 2: Historical trends in Western king prawn catch <strong>and</strong> nominal trawl effort (hours) in Spencer Gulf<br />
from 1978/79 to 2001/2002<br />
The DSS provides a powerful tool for<br />
management of the fishery as information<br />
obtained in real time can be evaluated <strong>and</strong><br />
subsequently discussed at FMC meetings.<br />
The key objectives of the DSS are:<br />
• To allow rapid information processing<br />
• To develop a link of spatial <strong>and</strong> non-spatial<br />
data with statistical analysis software,<br />
visualisation <strong>and</strong> communication tools<br />
• To allow fast analysis of ishery commercial<br />
logbook data <strong>and</strong> assess spatial <strong>and</strong> temporal<br />
pattern in catch <strong>and</strong> effort<br />
• To improve catch sampling <strong>and</strong> stock<br />
assessment by efficient information<br />
communication <strong>and</strong> improve analytical<br />
techniques<br />
• To allow comparison with historical<br />
harvesting strategies<br />
The key aspect of DSS is to allow instantaneous<br />
evaluation of real-time survey data through spatial<br />
visualisation <strong>and</strong> a statistical comparison with<br />
historical data. Visualisation is extremely<br />
important due to the multitude of error sources in<br />
the h<strong>and</strong>written raw data that is sometimes<br />
collected under very difficult conditions (i.e.<br />
rough sea). Both the visual analysis <strong>and</strong> the<br />
objective statistical evaluation will help<br />
management decisions to be made more<br />
objectively because they are based on the vast<br />
amount of knowledge <strong>and</strong> information from<br />
historic harvesting efficiencies <strong>and</strong> research<br />
projects.<br />
A further intention of the DSS is a long-term<br />
influence on the collection of data by suggesting<br />
changes to the data collection process.<br />
Technological means exist for streamlining the<br />
reporting process but implementation may be<br />
limited by legal as well as logistic constraints.<br />
However, the prawn DSS will only be used by a<br />
very limited number of people in a few<br />
committees (Committee at Sea <strong>and</strong> the FMC).<br />
Therefore, less effort is required for user<br />
friendliness. The two key design strategies were<br />
577
• efficiency <strong>and</strong><br />
• comprehensiveness.<br />
Efficiency was maximised by using most<br />
advanced database technology (ORACLE), a<br />
commercial GIS (ArcGIS) <strong>and</strong> statistical software<br />
packages (S-Plus <strong>and</strong> GENSTAT).<br />
Comprehensiveness was ensured by incorporating<br />
government <strong>and</strong> industry databases <strong>and</strong> placing<br />
much effort on entering historic data applying a<br />
very rigid quality control at the level of entering<br />
data into the database <strong>and</strong> by visualising spatial<br />
pattern.<br />
Technically, the linkages between the software<br />
components of the DSS were kept as simple as<br />
possible. Exchange is realised through ODBC<br />
(Open Database Connectivity) drivers of the<br />
Windows operating system or through text files.<br />
The database runs on a laptop <strong>and</strong> is hence<br />
transportable. The DSS will only be used by a<br />
very limited number of operators <strong>and</strong> dem<strong>and</strong>s<br />
substantial skills including excellent statistical<br />
knowledge, GIS skills, <strong>and</strong> also basic SQL<br />
knowledge. Using connectivity rather than hardwiring<br />
features is most advantageous for spatial<br />
data visualisation <strong>and</strong> statistical operations, since<br />
the operator is able to use a wide range of his<br />
favourite tools from a rapidly changing<br />
developing software market.<br />
4. APPLICATION OF THE DSS FOR<br />
ADAPTIVE<br />
REAL-TIME<br />
MANAGEMENT<br />
The data collection system incorporates<br />
information from stock assessment trawl surveys<br />
<strong>and</strong> adaptive surveys (also called “spot surveys”),<br />
which are undertaken to assess the stock <strong>and</strong><br />
improve the development of harvest strategies.<br />
The “spot” surveys are smaller research surveys<br />
undertaken over limited areas prior to the<br />
commencement of fishing in each period with<br />
independent research observers required to record<br />
catch <strong>and</strong> prawn size at strategic locations<br />
determined by research in collaboration with<br />
FMC members.<br />
The main stock assessment surveys provide vital<br />
information on stock, as well as develop harvest<br />
strategies. Information on biomass density, levels<br />
of recruits <strong>and</strong> the abundance of spawners are<br />
required as a feedback <strong>and</strong> adaptive control<br />
system, especially when depletion is high.<br />
Furthermore, size frequency <strong>and</strong> prawn density<br />
data from different regions (assemblages of trawl<br />
stations) are used to simulate optimal harvest<br />
periods to optimise egg production <strong>and</strong> economic<br />
return. Spatial harvest simulation models have<br />
been developed <strong>and</strong> incorporate a suite of input<br />
parameters including: prawn size frequency,<br />
densities, natural mortality, fishing mortalities,<br />
size-fecundity <strong>and</strong> maturation, seasonal<br />
catchability, growth parameters, <strong>and</strong> price<br />
structure.<br />
Once data are obtained from surveys, simulations<br />
are done to determine the optimum period to fish<br />
different areas. Figure 3 shows an example of the<br />
information processing <strong>and</strong> statistical modelling.<br />
Based on sampling in February 2002 the bio-value<br />
($/hr trawling) of a section of the Wallaroo<br />
ground would maximise value in May 2002,<br />
(Carrick 2002).<br />
Figure 3: Simulation of optimal period to harvest<br />
a segment of the Wallaroo prawn stock from<br />
survey sampling undertaken in February 2002<br />
The difficulty of using historic information in the<br />
database with newly collected data in statistical<br />
models implies that users of the DSS need to be<br />
highly experienced <strong>and</strong> skilled. Data are<br />
heterogeneous <strong>and</strong> the assumptions of the<br />
statistical models need carefully evaluated, which<br />
is best accomplished within statistical software<br />
packages rather than having statistical procedures<br />
hard-wired in a DSS.<br />
Throughout a fishing period, areas available to<br />
fishing can change as fishing progresses.<br />
Therefore, fishing areas are opened <strong>and</strong> closed<br />
based on the size of prawns, catch rates,<br />
depletion, spawning status <strong>and</strong> likely migration<br />
patterns of prawns. A key feature of this fishery is<br />
the use of real time monitoring <strong>and</strong> the<br />
corresponding changes to fishing strategies<br />
throughout the fishing periods in response to the<br />
daily movement of prawns <strong>and</strong> fleet depletion<br />
rates. Effective communication of “real time”<br />
information is critical to ensure that management<br />
is conducted in a sustainable way. The skippers<br />
of vessels have a major role in reporting real time<br />
information <strong>and</strong> a Committee at Sea assist in the<br />
development of harvest strategies. Fishery<br />
578
closures are an important “input” tool when<br />
orchestrated with effective real time adaptive<br />
management. The types of closures in Spencer<br />
Gulf consist of:<br />
• Permanent area closures - to protect small<br />
prawns <strong>and</strong> vulnerable discards.<br />
• Seasonal area Closures - variable <strong>and</strong> are<br />
used to protect small prawns, prevent<br />
reproductive depletion, <strong>and</strong> optimise value<br />
of catch for different levels of recruitment<br />
<strong>and</strong> stock size.<br />
• Total Gulf seasonal closures - December to<br />
March <strong>and</strong> June to November. To reduce<br />
trawl effort <strong>and</strong> to protect spawners.<br />
• Total Gulf moon closures – to reduce fishing<br />
inefficiency <strong>and</strong> maintain quality of catch.<br />
• Daylight closures – to reduce fishing<br />
inefficiency <strong>and</strong> further reduce impact of<br />
trawling on discards <strong>and</strong> habitat.<br />
A map illustrating the main spatial closures,<br />
implemented from April to June 2002, show that a<br />
sector of the Wallaroo ground was closed to<br />
trawling, owing to prawn size in the closed region<br />
being below optimal harvest size (Figure 4).<br />
surveys. The data is collected as part of the<br />
m<strong>and</strong>atory reporting of catch <strong>and</strong> effort. Most<br />
likely, without the regulations formulated in the<br />
1982 act we would not have such a<br />
comprehensive database. Yet the main deficiency<br />
of the m<strong>and</strong>atory data collection (catch averaged<br />
per day) is the lack of spatial reference. Location<br />
of best fishing grounds <strong>and</strong> risks is proprietary<br />
individual fishermen knowledge that is not readily<br />
shared. It may be assumed that the biggest hurdle<br />
for further advances in management is the lack of<br />
evidence that information sharing will increase<br />
individual fishermen income.<br />
The industry is sustainable <strong>and</strong> the autonomous<br />
decision making of the industry has not been<br />
harmful to the stock as this can unfortunately be<br />
readily observed in other fisheries. This can be<br />
attributed in part to the small <strong>and</strong> regionally<br />
isolated community of fishermen with a common<br />
interest in maintaining the productivity of their<br />
assets over the long term. In fact, a majority of the<br />
industry steering committee members has<br />
historically been advocating a reduced fishing<br />
effort <strong>and</strong> extent. The DSS provides objective<br />
arguments. An important aspect of the prawn<br />
fisheries, however, is the increase of the price<br />
with prawn size; hence economic return can be<br />
increased in spite of a reduced catch thereby<br />
contributing to the health of the industry.<br />
6. ACKNOWLEDGEMENTS<br />
The authors wish to thank the Fisheries Research<br />
<strong>and</strong> Development Cooperation (FRDC), South<br />
Australian Research <strong>and</strong> Development Institute<br />
(SARDI) <strong>and</strong> the University of Adelaide for their<br />
support of the development of the DSS.<br />
Figure 4: Main spatial closures implemented in<br />
Spencer Gulf prawn fishery over the period April<br />
to June 2002<br />
5. CONCLUSIONS:<br />
Decision making of the prawn industry has been<br />
much enhanced through the development of a<br />
comprehensive database that constitutes the heart<br />
of the DSS. Of lesser importance for the<br />
management of the industry has been a<br />
comfortable user interface. Two data sources are<br />
most important: catch <strong>and</strong> effort <strong>and</strong> targeted<br />
7. REFERENCES<br />
Caddy, J. <strong>and</strong> R. Mahon, Reference points for<br />
fisheries management. FAO Fisheries<br />
Technical Paper 347: 1-83, 1995.<br />
Carrick N., Report on the status of the Spencer<br />
Gulf Prawn fishery. Report to the South<br />
Australian Prawn Fishery Management<br />
Committee, March 2002<br />
MacDonald, N., Management plan for the<br />
Spencer Gulf <strong>and</strong> West Coast Prawn<br />
Fisheries. Internal document, Primary<br />
Industries <strong>and</strong> Resources, South Australia,<br />
1998.<br />
Morgan, G., Review of research <strong>and</strong> management<br />
of the Spencer Gulf Prawn fishery. South<br />
Australian Fisheries Management Series 20,<br />
1996.<br />
579
Optimum Sustainable Water Management in an<br />
Urbanizing River Basin in Japan, Based on Integrated<br />
<strong>Modelling</strong> Techniques<br />
E. KUDO a <strong>and</strong> M. OSTROWSKI a<br />
a Institute for Engineering Hydrology <strong>and</strong> Water Management, Darmstadt University of Technology,<br />
Darmstadt GERMANY (kudo@ihwb.tu-darmstadt.de)<br />
Abstract: In this research, a Decision Support System (“DSS”) was developed, using a combination of<br />
various existing models for Integrated Water Management (“IWM”). This DSS is then applied to a small<br />
urbanized basin, the Taguri-River basin in Japan. In developing the DSS, different existing dynamic <strong>and</strong><br />
steady state models were combined. These models include a rainfall-runoff analysis model, two river<br />
analysis models, a groundwater analysis model, <strong>and</strong> a geographical information system (“GIS”). The DSS<br />
was developed based on three basic elements: Database, model base, <strong>and</strong> tool base. A data exchange<br />
architecture was chosen <strong>and</strong> then exchange programs were written that are to act between different water<br />
analysis models in order to adequately translate the data format for each respective model. To improve the<br />
overall water condition in the basin, the DSS was used to simulate ten different measure-scenarios for the<br />
focus basin. These scenarios consider l<strong>and</strong> use, ground water level, allocation of drainage system, sewerage,<br />
water quality <strong>and</strong> quantity. During the research it became evident that a combination of measures is most<br />
effective for the basin, <strong>and</strong> accordingly such combination of measures was also simulated with the DSS.<br />
Finally, this paper describes the uncertainties of the DSS <strong>and</strong> discusses its further practical applicability.<br />
Keywords: Decision Support System; Integrated Water Management; Geographical Information System;<br />
Combined Model; Data Exchange Architecture<br />
1. INTRODUCTION<br />
Water problems consist of many different interrelated<br />
elements: Social, ecological, <strong>and</strong><br />
economical elements. Thus, effective water<br />
management requires a device that provides the<br />
decision maker with accurate descriptions of<br />
various water conditions, including surface water<br />
<strong>and</strong> groundwater, <strong>and</strong> considering water quality <strong>and</strong><br />
water quantity in dry <strong>and</strong> wet weather (Water<br />
management that takes into account all the<br />
abovementioned factors is called “Integrated Water<br />
Management: IWM”). Moreover, sustainable water<br />
management requires assistance from <strong>and</strong> decisionmaking<br />
involving all the different groups within the<br />
focus basin: Politicians, administration officers,<br />
civil engineers, <strong>and</strong> stakeholders. In order to ensure<br />
the participation <strong>and</strong> support of these groups, it is a<br />
prerequisite to inform them about the effects of<br />
intended measures <strong>and</strong> water management in an<br />
efficient <strong>and</strong> transparent way. Therefore, a device<br />
is required to gather <strong>and</strong> display the necessary<br />
information.<br />
A simulation model allows the user to evaluate<br />
various water conditions <strong>and</strong> their complex<br />
interaction as a whole, without generating high<br />
costs <strong>and</strong> long simulation times, <strong>and</strong> thus is useful<br />
for decision-making. Moreover, it is able to<br />
illustrate, which measure is most effective <strong>and</strong> in<br />
which way this measure influences upon different<br />
stakeholders, society, nature, <strong>and</strong> of cause the<br />
economy. Such a model that is able to simulate<br />
various water conditions in a basin <strong>and</strong> help in<br />
decision-making is the most important part of a<br />
Decision Support System (“DSS”).<br />
In this research, a DSS is developed using a<br />
combination of various existing models for IWM.<br />
It is then applied to the Taguri-River basin in Japan,<br />
a small urbanizing basin. 1<br />
2. THE FOCUS RIVER BASIN<br />
The Taguri-River basin is located 30-50 km<br />
northeast of Tokyo in the Chiba prefecture in Japan<br />
(Figure 1). The basin belongs to the Inba-Numa-<br />
Lake basin. The Taguri-River basin has an area of<br />
19 km 2 . The river flows through two local<br />
communities from south to north <strong>and</strong> into the Inba-<br />
Numa-Lake. The river system consists of one main<br />
river <strong>and</strong> two small tributaries. The entire basin can<br />
be divided into four sub-basins (Figure 2). The first<br />
sub-basin (A-basin) includes the upper stream of<br />
580
Japan<br />
Tokyo<br />
Taguri-River basin<br />
Rivers<br />
Sewer<br />
Sewage works<br />
D<br />
N<br />
Tone-River<br />
N<br />
Inba-Numa-Lake<br />
B<br />
Taguri-River basin<br />
Tokyo Bay<br />
Location <strong>and</strong> l<strong>and</strong> use of the<br />
four sub-basins of the Taguri-<br />
River basin<br />
Inba-Numa-<br />
Lake-Basin<br />
5km<br />
2 km<br />
Figure 1.<br />
Location of the basin of the<br />
Inba-Numa-Lake <strong>and</strong> the<br />
Taguri-River basin<br />
the main river that flows through a residential area<br />
<strong>and</strong> through rice fields. The second sub-basin (Bbasin)<br />
is located west of the first sub-basin <strong>and</strong><br />
includes a tributary, which flows through areas that<br />
contain many permeable areas (fields <strong>and</strong> forest,<br />
etc.). The third sub-basin (C-basin) is located<br />
northwest of the second basin <strong>and</strong> includes the<br />
other tributary. This river flows through a<br />
residential area that has not been fully developed<br />
yet. The fourth sub-basin (D-basin) includes the<br />
lower stream of the main river, which flows<br />
through rice fields <strong>and</strong> into the lake.<br />
The basin was further divided into 106 smaller subbasins<br />
that are defined according to l<strong>and</strong> use <strong>and</strong><br />
topography to analyse the basin in detail. Based on<br />
measured geological data, it is assumed that the<br />
geological system in the focus basin consists of<br />
three layers. The first layer is a permeable<br />
unconfined aquifer system that reaches from the<br />
surface to a depth of 50 m below surface. The<br />
second layer, which separates the first from the<br />
deeper aquifer, is 20 m thick silt. The third layer is<br />
a confined aquifer system reaching from a depth of<br />
70 m to 200 m below surface. Most of the<br />
groundwater is pumped up from the third layer.<br />
The focus term of the research is five years, from<br />
1995 through 1999. Some field investigation is<br />
done in 1997 to calibrate data between simulated<br />
<strong>and</strong> investigated data. Therefore, all results shown<br />
in the paper are the results based on the values<br />
found in 1997. The average humidity in 1997 was<br />
66 percent, the average temperature was 14.8 o C,<br />
<strong>and</strong> total evapo-transpiration was 788.3 mm. The<br />
planned sewerage (separate-sewerage system) area<br />
was 39 percent of the basin, where 90 percent of<br />
Figure 2.<br />
Table 1.<br />
Key figures for the sub basins<br />
Area (ha) Population No. of sub-basins<br />
A 651 15,597 44<br />
B 229 978 11<br />
C 593 34,746 29<br />
D 443 13,731 22<br />
total 1,916 65,053 106<br />
inhabitant lived. In the planned sewerage area the<br />
coverage of sewerage was 87 percent in 1997 (with<br />
13 percent still to be constructed). Key figures for<br />
the sub-basins are shown in Table 1.<br />
3 THE DSS FOR THE FOCUS BASIN<br />
Bernhard Hahn et al (2000) 2 have described that a<br />
DSS consists of four components: The model base,<br />
the tool base, the database, <strong>and</strong> the user interface.<br />
In this research, the DSS is developed for technical<br />
users. Therefore there is no strong focus on user<br />
interface. Accordingly, the developing process of<br />
the DSS mainly refers to the remaining three<br />
components of a DSS.<br />
In developing the DSS, the problems existing<br />
within the focus basin are determined through<br />
various field investigations. The major factors<br />
influencing the water circulation in the basin are<br />
wastewater from households, sewerage system,<br />
runoff from urban areas, pumping up groundwater,<br />
<strong>and</strong> resulting l<strong>and</strong> subsidence.<br />
3.1 Model Base<br />
In order to enable a smooth combination of the<br />
different models used for analysing these major<br />
581
factors in the basin, each model that will be<br />
integrated into the DSS has to meet the following<br />
five requirements:<br />
1) Each model must be capable of analysing both<br />
water quality <strong>and</strong> water quantity.<br />
2) Import / export of files from one model to<br />
another has to be simple, thus models using<br />
ASCII-files have been chosen.<br />
3) Input, output <strong>and</strong> temporary data files must be<br />
managed as separate files within each model.<br />
4) Time-interval for analysis must be adjustable.<br />
5) H<strong>and</strong>ling must be user-friendly.<br />
Considering these requirements, the following four<br />
water analysis models <strong>and</strong> a geographic information<br />
system (“GIS”) have been selected:<br />
1) SMUSI 4.0 (Schmutzfrachtsimulation: Pollution<br />
Load in Urban Drainage Systems) - rainfallrunoff<br />
analysis model (Darmstadt University)<br />
2) CE-QUAL-RIV1 (A dynamic, one-dimensional<br />
water quality model for streams) - dynamic river<br />
analysis model (Ohio University)<br />
3) QUAL2E (The enhanced stream water quality<br />
model) - steady state river analysis model<br />
(Texas Water Development Board <strong>and</strong> U.S.<br />
<strong>Environmental</strong> Protection Agency)<br />
4) PMWIN (A Simulation System for <strong>Modelling</strong><br />
Groundwater Flow <strong>and</strong> Pollution basis on<br />
MODFLOW) - groundwater analysis model<br />
(U.S. Geological Survey <strong>and</strong> Chiang et. al.)<br />
5) Arc View 3.2 - GIS (the <strong>Environmental</strong> Systems<br />
Research Institute: ESRI)<br />
The DSS uses many different physical, hydrologic,<br />
<strong>and</strong> hydraulic methods <strong>and</strong> equations. They include<br />
various coefficients (e.g. Manning coefficient, settle<br />
coefficients, decay rate, transmissivity, storage<br />
coefficient, hydraulics conductivity). These are<br />
defined either by means of a simple calculation or<br />
through trial <strong>and</strong> error using field investigation data.<br />
3.2 Tool Base<br />
The main function of the tool base is to control the<br />
interaction between the different models. The tool<br />
base had to be developed to provide for a<br />
combination of different models without changing<br />
original source codes, because most of the original<br />
source codes of the models are not available.<br />
Therefore, the data exchange architecture method is<br />
chosen. This type of integration consists of<br />
distributed systems <strong>and</strong> databases. It uses different<br />
models, with every model operating separately.<br />
Different exchange programs connecting different<br />
models transform output data into input data by<br />
using specific formats. If exchange data are given<br />
in linear or single form (e.g. water quality <strong>and</strong> water<br />
quantity), the exchange program is developed using<br />
FORTRAN. If exchange data are given in<br />
spreadsheet form (e.g. groundwater level <strong>and</strong><br />
topography data), MS Excel is used as exchange<br />
program. In consequence, nine programs using<br />
FORTRAN und two MS Excel programs were<br />
developed as exchange programs. Figure 3 shows<br />
the basic data flow in the DSS.<br />
Arc View 3.2<br />
Figure 3.<br />
3.3 Database<br />
E.P.E.1<br />
E.P.E.3<br />
No<br />
E.P.F.1<br />
QUAL2E<br />
E.P.E.2<br />
SMUSI 4.0<br />
Yes<br />
hydrodynamic<br />
PMWIN<br />
E.P.F.1<br />
CE-QUAL-RIV1<br />
E.P.F: exchange program with FORTRAN<br />
E.P.E: exchange program with MS Excel<br />
Data flow in the DSS model part<br />
The DSS requires user input of physical, social,<br />
geological <strong>and</strong> climatic data, which are used as<br />
independent variables in equations. Due to the<br />
necessity of parameter estimation for every model,<br />
the DSS also requires a wide range of various data<br />
measured over a long period of time. Regional<br />
information (e.g. climate, geological, <strong>and</strong><br />
population) <strong>and</strong> site-specific information (e.g. water<br />
quality <strong>and</strong> water quantity, l<strong>and</strong> use, <strong>and</strong> condition<br />
of sewerage system) are gathered through field<br />
investigations <strong>and</strong> from public offices or<br />
corporations that operate within the basin. Digital<br />
data for l<strong>and</strong> use <strong>and</strong> topography, as well as<br />
precipitation data are available in 10-minute steps.<br />
These data are prepared in different databases, such<br />
as MS Excel, ASCII files, or database for GIS.<br />
4 OPTIMIZATION OF THE WATER<br />
MANAGEMENT<br />
4.1 Water Conditions in Single Measure<br />
Scenarios<br />
Optimum sustainable water management aims for<br />
overall better water conditions in the entire basin<br />
(“Optimization”). The optimum water management<br />
plan for the focus basin was developed through<br />
582
Table 2.<br />
Indirect <strong>and</strong> direct evaluation of water management<br />
Groundwater River Construction<br />
Environment<br />
Scenario Quality Quantity Quality Quantity Flooding Total Cost Time Decision Running Cost<br />
1 0 0 2 -2 1 1 high long easy low low<br />
2 0 2 1 0 1 4 high long difficult middle low<br />
3 0 1 1 1 1 4 high long difficult middle low<br />
4 0 2 2 0 0 4 high long difficult high low<br />
5 0 1 -1 0 0 0 high middle difficult high low<br />
6 -1 1 -1 2 0 1 high middle difficult high relative high<br />
7 0 0 1 0 1 2 high long relative easy middle relative high<br />
8 -1 1 1 0 -1 0 middle middle easy low high<br />
9 -1 1 1 0 0 1 middle middle easy low high<br />
10 0 1 1 1 2 5 low short difficult low relative high<br />
DSS-simulation of ten scenarios using different<br />
single measures.<br />
Scenario 1: Completing the sewerage in planned<br />
area<br />
Scenario 2: Reusing rainwater on roofs for toilet<br />
<strong>and</strong> sprinkling<br />
Scenario 3: Reusing rainwater on roofs for rice<br />
fields<br />
Scenario 4: Reusing wastewater for toilet <strong>and</strong><br />
sprinkling<br />
Scenario 5: Reusing wastewater for rice fields<br />
Scenario 6: Reusing wastewater for preserving<br />
mean discharge in the river<br />
Scenario 7: Changing l<strong>and</strong> use (more permeable<br />
area)<br />
Scenario 8: Removing concrete from river bottom<br />
Scenario 9: Constructing small riverbed<br />
Scenario 10: Guiding runoff from non-urbanized<br />
areas into rice fields<br />
Table 2 shows the evaluation of the different<br />
scenarios. The Table shows the direct a <strong>and</strong><br />
indirect b evaluation of water management.<br />
The results displayed on the left h<strong>and</strong> in the Table<br />
are evaluated using a scale of five grades (- 2, - 1, 0,<br />
1, 2). The most effective measure (compared to the<br />
water condition without any measure) is allocated 2<br />
positive points, the measure with the most negative<br />
effect is allocated 2 negative points. A measure<br />
that does not show any effect is allocated 0 points.<br />
The sum of points allocated to each measure is<br />
shown in the middle column of the table. Scenario<br />
10 is allocated 5 points, scenarios 2, 3, <strong>and</strong> 4 are<br />
each allocated 4 points <strong>and</strong> scenario 5 <strong>and</strong> 8 are<br />
allocated zero points each.<br />
The grades on the right h<strong>and</strong> in the Table are<br />
expressed in different terms (e.g. high, low, easy,<br />
<strong>and</strong> difficult). When looking exclusively at the<br />
number of points, scenario 10 appears to be the<br />
single best measure of all scenarios introduced.<br />
However, this measure is not available in reality<br />
unless an underst<strong>and</strong>ing can be reached with the<br />
farmers owning the fields where the measure would<br />
a The implemented measure is evaluated depending on its effects<br />
upon water conditions.<br />
b The implemented measure is not evaluated depending on its<br />
effects, but based on the construction <strong>and</strong> management<br />
necessary for the measure (cost performance, construction<br />
time, ease of decision making, <strong>and</strong> impact upon nature).<br />
2: better, 1: good, 0: no effect, -1: bad, -2: worse<br />
have to be implemented. Moreover, this measure is<br />
not effective in improving the groundwater level in<br />
the third layer. With regard to the groundwater<br />
level in this layer, scenarios 2 or 4 appear to be the<br />
most effective measures.<br />
However, none of the single measures is<br />
satisfactorily improving the water condition in the<br />
entire basin. The aim has to be an overall<br />
improvement, taking into account the usual (dry)<br />
weather condition <strong>and</strong> long-term (yearly) water<br />
conditions.<br />
4.2 Optimization through Combined Measures<br />
In consequence, a combination of various measures<br />
is simulated in an additional scenario. The different<br />
measures have to be installed in different parts of<br />
the basin in order to improve water conditions. In<br />
order to optimize the water management, the<br />
measures have been combined with the aim to<br />
combine the positive effects of different measures<br />
in the most efficient way. Below is an outline of<br />
the main requirements for optimization, <strong>and</strong> the<br />
corresponding measures that were combined.<br />
1) Water that is used in the basin has to remain in<br />
the basin.<br />
Instead of constructing a large separate-sewerage<br />
system, all wastewater from households in planned<br />
separate-sewerage system areas is treated in several<br />
mid-sized sewage-works <strong>and</strong> is guided into the<br />
river in the upper areas of the main river in A-basin<br />
<strong>and</strong> a tributary in C-basin. The values for the<br />
treated water are below: BOD-level: 3.0 mg/l, NH 4 -<br />
N-level: 3.2 mg/l, PO 4 -P-level: 0.84 mg/l.<br />
2) Groundwater consumption has to be decreased<br />
through/replaced by reuse of treated wastewater<br />
from households for toilet <strong>and</strong> sprinkling.<br />
Rainwater on roofs <strong>and</strong> wastewater from<br />
households is reused for toilet <strong>and</strong> sprinkling.<br />
3) Peak flow in rainy times has to be decreased in<br />
order to minimize the risk of flooding.<br />
Permeable areas are increased by 30 percent.<br />
Rainwater from non-urbanized areas is guided into<br />
rice fields.<br />
4) Risk of contamination of food has to be<br />
minimized.<br />
583
Treated wastewater from households <strong>and</strong> runoff<br />
from urbanized areas is not reused in rice fields.<br />
5) <strong>Environmental</strong> effects have to be taken into<br />
consideration.<br />
A small riverbed is constructed. Concrete is<br />
removed from the river bottom.<br />
All results from this scenario (combined measure)<br />
are compared to the actual conditions in 1997. The<br />
effect of the implemented measures (increasing<br />
permeable areas <strong>and</strong> guiding rainwater into rice<br />
fields) on the water quantity <strong>and</strong> water quality at the<br />
lower point of D-basin in wet weather conditions is<br />
obvious (Figure 4). Peak flow decreases by 55<br />
percent. The water stored in rice fields drains out<br />
continuously <strong>and</strong> slowly after rainfall. The BODlevel<br />
also decreases by about 50 percent (Figure 5).<br />
Overall water quality is significantly better than in<br />
1997, because runoff decreases, thus also<br />
decreasing the amount of pollution source draining<br />
into the river.<br />
While the average annual discharge in the upper<br />
area of the main river increases in this scenario (due<br />
to the reuse of wastewater) (Figure 6), the annual<br />
average discharge in the lower part of the main<br />
river in this scenario is not different from that in<br />
1997. However, the average annual BOD-level is<br />
significantly better than it was in 1997 due to this<br />
combination of measures: Increasing coverage of<br />
sewerage <strong>and</strong> permeable area (Figure 7).<br />
Precipitation<br />
[mm/hr]<br />
Discharge [CMS]<br />
Precipitation<br />
[mm/hr]<br />
BOD [mg/l]<br />
0<br />
3<br />
6<br />
1.5<br />
1.0<br />
0.5<br />
With measures<br />
In 1997<br />
Precipitation<br />
0.0<br />
5.5 5.9 6.3 6.7 7.1 7.5<br />
Julian day (January 05, 1997, 12:00 - 07, 12:00)<br />
0<br />
3<br />
6<br />
6.0<br />
4.0<br />
2.0<br />
Figure 4.<br />
Discharge condition<br />
With measures<br />
In 1997<br />
Precipitation<br />
the total amount of discharge remains the same<br />
because of reusing wastewater for the river. The<br />
total amount of discharge <strong>and</strong> contaminants (BOD,<br />
NH 4 -N, PO 4 -P-level) under the combined measure,<br />
compared to the levels found in 1997, are illustrated<br />
in Figure 8.<br />
With the combined measure, pumping up of<br />
groundwater for water supply is decreased by 30<br />
percent (15 percent due to use of rainwater <strong>and</strong> 15<br />
percent due to reuse of wastewater for toilet <strong>and</strong><br />
sprinkling) <strong>and</strong> pumping up groundwater for rice<br />
fields is decreased by 25 percent through use of<br />
rainwater. Thus, a sufficient amount of water can<br />
be supplied for rice fields during the active season<br />
(from May until September). The groundwater<br />
level in the third layer is kept higher than in 1997.<br />
The maximum difference reaches about 30 cm in<br />
Discharge [CMS]<br />
BOD[mg/l]<br />
0.6<br />
0.4<br />
0.2<br />
With measures<br />
In 1997<br />
0.0<br />
0 2 4 6 8 10<br />
Junction with B-river<br />
Junction with C-river<br />
upper<br />
Distance [km]<br />
lower<br />
16<br />
12<br />
8<br />
4<br />
16.0<br />
12.0<br />
8.0<br />
4.0<br />
Figure 6.<br />
Annual discharge<br />
0<br />
0 2 4 6 8 10<br />
Junction with B-river<br />
Junction with C-river<br />
upper<br />
Distance [km]<br />
lower<br />
Figure 7.<br />
With measures<br />
In 1997<br />
Annual BOD-level<br />
With measures<br />
In 1997<br />
0.0<br />
5.5 6.0 6.5 7.0 7.5<br />
0.0<br />
*10 4 (m 3 ) *10 4 (kg)<br />
*10 4 (kg)<br />
*10 3 (kg)<br />
Discharge BOD<br />
NH 4 -N PO 4 -P<br />
Julian day (January 05, 1997, 12:00 - 07, 12:00)<br />
Figure 5.<br />
BOD-level condition<br />
In this scenario, the BOD-level is lower than in<br />
1997, which in itself is a positive effect. However,<br />
the NH 4 -N <strong>and</strong> PO 4 -P-levels have increased, while<br />
Figure 8.<br />
Total amount of discharge <strong>and</strong><br />
contaminants<br />
A-basin <strong>and</strong> the lower area of C-basin where much<br />
groundwater is pumped up. By removing the<br />
584
concrete from the river bottom <strong>and</strong> constructing a<br />
small riverbed in the upper areas of A- <strong>and</strong> C-rivers,<br />
the depth of the river can be kept at 10 cm in the<br />
upper area of A-river (3 cm in 1997) <strong>and</strong><br />
groundwater level in the first layer can be kept up<br />
to 1 m above the level in 1997.<br />
5 THE UNCERTAINTIES OF THE DSS<br />
Uncertainties from a model appear in different<br />
types / forms (C.S. Melching (1995) 3 ). In the DSS<br />
we can identify four types of uncertainties: Natural<br />
r<strong>and</strong>omness, data, model parameters <strong>and</strong> model<br />
structure. It is important for decision makers to<br />
underst<strong>and</strong> that each scenario holds the risk of a<br />
certain margin of error due to uncertainty. But it is<br />
equally important that the user underst<strong>and</strong>s that this<br />
does not render the DSS <strong>and</strong> its results useless. If<br />
the margin of error is kept in mind, the DSS is a<br />
useful tool for supporting the process of decisionmaking.<br />
In planning water management, the main source of<br />
uncertainties from natural r<strong>and</strong>omness is<br />
precipitation data. The complex r<strong>and</strong>omness from<br />
precipitation influences upon the results of all<br />
scenarios, because runoff from every sub-basin is<br />
aggregated in the simulation.<br />
Uncertainties from data may stem from outdated<br />
data. Thus, l<strong>and</strong> use, topographical, <strong>and</strong> geological<br />
data (impact upon runoff condition) may give rise<br />
to uncertainties. Digital maps are not updated on an<br />
annual basis in Japan. However, if large<br />
construction (e.g. urban renewal or developing a<br />
golf area) is done, the factor of l<strong>and</strong> use may differ<br />
significantly from the latest set of map data.<br />
Therefore, based on this data, l<strong>and</strong> use is modified<br />
by using field investigation <strong>and</strong> aerial photographs.<br />
For similar reasons, social data (e.g. population <strong>and</strong><br />
the coverage of the sewerage system) also give rise<br />
to uncertainties. It is very difficult to gather up-to<br />
date data annually, because the way the focus basin<br />
is divided into sub-basins does not correspond to<br />
the division into areas for which survey data from<br />
public offices are available.<br />
A good example for uncertainty from model<br />
parameters is the parameter for load rate from nonpoint<br />
sources. This parameter is defined by<br />
comparing the results of other research papers (for<br />
different basins) to field investigation data gathered<br />
in one part of the focus basin. The parameter is<br />
then applied to the entire focus basin. However, in<br />
reality, such parameters change depending on l<strong>and</strong><br />
use <strong>and</strong> weather conditions. Approximate<br />
parameters can be defined by means of long-term<br />
investigation <strong>and</strong> continual survey in sub-basins<br />
with different l<strong>and</strong> use <strong>and</strong> weather conditions.<br />
There are two types of uncertainty inherent to the<br />
model structure: Mistakes in programming <strong>and</strong><br />
inadequate model structure. This DSS has been<br />
developed using four models that have been used<br />
over a long time for various areas. Therefore, the<br />
probability of mistakes in programming in these<br />
four models can be rated as very small. Uncertainty<br />
from model structure can be limited through choice<br />
of models adequate to the focus basin <strong>and</strong> a<br />
comparison of data simulated by the model to field<br />
investigation data. Both have been taken into<br />
account in developing this DSS. Therefore,<br />
uncertainty of model structure is considered not to<br />
have a significant impact upon the results produced<br />
with it.<br />
6 DISCUSSION<br />
The DSS in this research is developed for an<br />
exemplary focus basin. To improve water<br />
conditions in the focus basin, various measures are<br />
simulated <strong>and</strong> subsequently combined, <strong>and</strong> thus the<br />
DSS shows different alternatives for integrated<br />
water management in the focus basin. Using a<br />
combination of various existing models,<br />
developing-cost <strong>and</strong> -time for a new model are<br />
reduced. Moreover, through using the data<br />
exchange architecture method, it holds the potential<br />
to be developed further for application in different<br />
basins <strong>and</strong> conditions. To that end, the DSS needs<br />
further enhancement by including an economic <strong>and</strong>,<br />
an ecological model. Furthermore, the<br />
implementation of functionality for automatic<br />
estimation of model parameters would be<br />
reasonable. In addition, the practical applicability<br />
of the DSS should be studied <strong>and</strong> discussed for a<br />
variety of river basins <strong>and</strong> water management<br />
problems.<br />
1 Kudo E., 2004. Optimum sustainable water<br />
management in an urbanizing river basin in<br />
Japan, based on integrated modelling<br />
techniques; Dissertation, Institute for<br />
Engineering Hydrology <strong>and</strong> Water<br />
Management, Darmstadt University of<br />
Technology, Darmstadt, Germany.<br />
2 Hahn B.; Engelen G., 2000. Concepts of decision<br />
support systems, ROKS – Research,<br />
Decision Support Systems (DSS),<br />
<strong>International</strong> Workshop 6 April 2000,<br />
Bundesanstalt für Gewässerkunde Koblenz.<br />
3 Melching, C. S.; Singh, V. P., 1995. Computer<br />
Models of Watershed Hydrology, Chapter 3,<br />
Reliability Estimation, Water Resources<br />
Publications, Littleton, Colorado, pp.71-85.<br />
585
Application of a GIS-based Simulation Tool to Analyze <strong>and</strong><br />
Communicate Uncertainties in Future Water Availability in<br />
the Amudarya River Delta<br />
Maja Schlüter a & Nadja Rüger b<br />
a<br />
Institute of <strong>Environmental</strong> Systems Research, University of Osnabrück, 49069 Osnabrück, Germany,<br />
mschluet@usf.uni-osnabrueck.de<br />
b<br />
Centre for <strong>Environmental</strong> Research, Dep. of Ecological <strong>Modelling</strong>, PF 500135, 04301 Leipzig, Germany<br />
Abstract: Simulation <strong>and</strong> decision support tools facilitate a process of reasoning about potential<br />
future development paths of a system, e.g. a river system, under alternative management strategies. Joint<br />
scenario development <strong>and</strong> analysis with river basin authorities <strong>and</strong> stakeholders can inform <strong>and</strong> structure<br />
discussions on management goals <strong>and</strong> major uncertainties affecting river basin management in future. Tools<br />
can support the determination of strategies that balance water allocation between multiple users, such as<br />
irrigation <strong>and</strong> the environment, <strong>and</strong> measures to cope with uncertainties. The GIS-based simulation tool<br />
TUGAI has been developed to facilitate exploration of alternative water management strategies for the<br />
degraded Amudarya river delta <strong>and</strong> to analyze their ecological implications. It combines a multi-objective<br />
water allocation model with simple models of l<strong>and</strong>scape dynamics <strong>and</strong> a fuzzy based assessment of habitat<br />
changes for riverine Tugai forests. The Tugai forests serve as an indicator of the ecological state of the delta<br />
region under a given water management scenario. In this contribution different sources of uncertainties in<br />
water availability for the Amudarya delta will be determined <strong>and</strong> the ecological implications of water supply<br />
uncertainty analyzed using the TUGAI tool. Scenario analysis provides an assessment of the range <strong>and</strong><br />
magnitude of the impact of those uncertainties on the ecological situation in the delta. Uncertainties inherent<br />
in system underst<strong>and</strong>ing <strong>and</strong> representation that influence model outcomes are presented <strong>and</strong> discussed in<br />
view of their role in the impact assessment. The application of simple simulation tools that integrate the up to<br />
date available knowledge of the system for identification of the type <strong>and</strong> range of relevant uncertainties<br />
affecting river basin management <strong>and</strong> their perception with managers <strong>and</strong> stakeholders, for analysis <strong>and</strong><br />
discussion of their potential impacts <strong>and</strong> development of cooping strategies will be discussed.<br />
Keywords: integrated models; ecological impact assessment; Amudarya delta; uncertainty; scenario analysis<br />
1. INTRODUCTION<br />
Simulation <strong>and</strong> decision support tools facilitate a<br />
process of reasoning about potential future development<br />
paths of a system, e.g. a river system,<br />
under alternative management strategies. Integrated<br />
models that formalize the up to date available<br />
knowledge <strong>and</strong> clarify important cause-effect relationships<br />
can help to communicate the complexity<br />
of the water management situation. The use of<br />
simulation tools in workshop settings with river<br />
basin authorities <strong>and</strong> stakeholders can support<br />
identification of major uncertainties affecting river<br />
basin management in future. Analyses of “what-if”<br />
scenarios provide an assessment of the impacts of<br />
those uncertainties on the human <strong>and</strong> natural<br />
system as a basis for the development of cooping<br />
strategies. Scenario analysis is central to climate<br />
change impact assessment, but is not widely used<br />
in water resource assessment so far [IPCC, 2001].<br />
Most decisions in river basin management have to<br />
be made under uncertainty. Uncertainty is defined<br />
as the impossibility of having all necessary<br />
information, knowledge <strong>and</strong> predictive capacity<br />
about a situation <strong>and</strong> its future development to describe<br />
<strong>and</strong> have full confidence in all possible<br />
outcomes of a decision. For analysis sources of uncertainty<br />
are identified <strong>and</strong> given a probability<br />
based on experience, knowledge or intuitive<br />
feelings. Potential impacts of those uncertainties<br />
can then be explored by scenario analysis.<br />
586
There are different sources of uncertainty in a river<br />
basin management context. Uncertainty can be<br />
caused by a lack of knowledge on the functioning<br />
of the system which is reflected in system representations.<br />
It can also originate from factors external<br />
to a given river management situation, such as<br />
socio-economic changes or global change [see e.g.<br />
Pahl-Wostl 1998, Carter 2001] or human decision<br />
making. There are many techniques to estimate uncertainties,<br />
especially those caused by insufficient<br />
system underst<strong>and</strong>ing <strong>and</strong> representation, but often<br />
scenario analysis testing a range of plausible futures<br />
remains the only possible means [Carter 2001].<br />
This paper presents the application of the<br />
integrated simulation tool TUGAI, which has been<br />
developed for ecological impact assessment in the<br />
Amudarya river delta, to determine <strong>and</strong> analyse<br />
management uncertainties in the river basin. As an<br />
example, scenario analysis of changes to the<br />
inflow to the delta is used to assess the potential<br />
impact of those water supply uncertainties on the<br />
ecological state of the delta. Uncertainty in inflow<br />
can be caused by a variety of factors, many of<br />
which are not reducible for water mangers in the<br />
delta region, such as political decisions, l<strong>and</strong> use<br />
changes upstream, or climate change. The potential<br />
<strong>and</strong> limits of applying the tool to analyze <strong>and</strong><br />
discuss those uncertainties with water managers<br />
<strong>and</strong> stakeholders will be discussed.<br />
2. WATER MANAGEMENT IN THE<br />
AMUDARYA RIVER DELTA<br />
The Amudarya delta is located in the semi-arid<br />
Turan lowl<strong>and</strong>s of Central Asia in Uzbekistan <strong>and</strong><br />
Turkmenistan (Figure 1). It is characterized by low<br />
precipitation (
3. THE TUGAI SIMULATION TOOL<br />
The TUGAI tool has been developed to facilitate a<br />
first quick assessment of the ecological effects of<br />
changes in the spatio-temporal water distribution<br />
in the delta of the Amudarya river. The aim of the<br />
tool is to foster underst<strong>and</strong>ing of interrelationships<br />
between human action <strong>and</strong> the state of the local<br />
environment, to initiate a goal finding process for<br />
ecological rehabilitation <strong>and</strong> a search for new,<br />
integrative strategies of water management. It can<br />
support the determination of water management<br />
strategies that balance water allocations between<br />
multiple users, including the environment.<br />
lake depression), have been chosen as important<br />
environmental variables for habitat suitability.<br />
Scenarios reflecting changes in inflow to the delta<br />
region or policy decisions for water management<br />
measures in the delta itself can be developed <strong>and</strong><br />
their implications for the ecological situation in the<br />
northern delta area (approximately 3000 km 2 )<br />
assessed. The tool allows qualitative comparison<br />
of alternative strategies, tradeoffs in spatiotemporal<br />
water allocation und the ecological<br />
implications of water supply uncertainties using<br />
colour coded maps, tables <strong>and</strong> graphs.<br />
Since the tool is based on assumptions concerning<br />
l<strong>and</strong>scape dynamics <strong>and</strong> their impact on deltaic<br />
ecosystems, deficient knowledge, uncertain data<br />
<strong>and</strong> parameter values, its “forecasts” are themselves<br />
subject to uncertainty. The range of outcomes<br />
achieved as results of scenario analysis<br />
should be critically reviewed in view of the uncertainties<br />
associated with the impact assessment<br />
[Kwadijk & Rotmans 1995].<br />
4. UNCERTAINTIES IN SYSTEM<br />
REPRESENTATION<br />
Figure 2: Flow chart of the Tugai tool with the<br />
three modules <strong>and</strong> their linkages. Dotted circles<br />
show measures or developments that can be<br />
expressed <strong>and</strong> analysed via scenarios.<br />
The tool combines an optimization model of water<br />
allocation (water management model) with simple<br />
GIS-based simulations of changes to major environmental<br />
variables (environmental models) <strong>and</strong><br />
an ecological assessment of the environmental<br />
situation based on the habitat quality for characteristic<br />
riverine Tugai forests (Habitat Suitability<br />
Index) (see Figure 2). The Tugai forests are used<br />
as an indicator of the ecological situation in the<br />
delta area. Tugai are mainly poplar-willow forests<br />
that occur along rivers <strong>and</strong> in river deltas in arid<br />
regions in Central Asia. They can tolerate extreme<br />
temperatures, drought <strong>and</strong> moderate soil salinity,<br />
but their establishment <strong>and</strong> viability depends on<br />
floods <strong>and</strong> “reachable” groundwater levels. A<br />
habitat suitability index indicates the suitability of<br />
a site for establishment, survival <strong>and</strong> reproduction<br />
under the given environmental conditions. Based<br />
on the habitat needs of the dominant poplar<br />
species, groundwater level (distance of groundwater<br />
table from soil surface), flooding regime<br />
(flooding frequency, timing <strong>and</strong> duration) <strong>and</strong><br />
geomorphology (classified into river bar, slope of<br />
river bar, floodplain or terrace, interfluve lowl<strong>and</strong>,<br />
In the following the three submodels of the<br />
TUGAI simulation tool are subject to a critical revision<br />
with respect to uncertainties in the representation<br />
of relevant processes <strong>and</strong> their impact on<br />
simulated habitat suitability for Tugai forests.<br />
Water management model<br />
The water management model uses a multi-criteria<br />
optimization algorithm to distribute the available<br />
inflow into the delta among the water users. The<br />
modelled water allocation necessarily differs from<br />
the real water distribution, because it optimizes the<br />
distribution in a deterministic way under “perfect<br />
knowledge” on future water availability. In reality<br />
however, management decisions are taken under<br />
high uncertainty. Nevertheless, given the complex<br />
river network <strong>and</strong> dem<strong>and</strong>s of the irrigation users,<br />
an optimization model appeared to be the best<br />
approach [see also Wardlaw <strong>and</strong> Barnes 1999].<br />
Calibration of the objective weights of the multiobjective<br />
function was carried out by fitting model<br />
results to observed data to achieve the best representation<br />
of current water allocation practices to be<br />
used as a reference scenario. When interpreting<br />
model results, they have to be seen as “best<br />
achievable” water allocation under the given<br />
supply <strong>and</strong> dem<strong>and</strong>s. Validation <strong>and</strong> sensitivity<br />
analysis have shown that the model manages to<br />
allocate available water quantities in the desired<br />
way, although spatio-temporal distribution within<br />
individual months might differ from the observed.<br />
The model reacts to changes in allocation priorities<br />
as expected [Schlüter et al. 2004].<br />
588
<strong>Environmental</strong> Models<br />
A statistical approach is applied to predict the<br />
groundwater level at 12 wells across the delta area<br />
depending on mean annual river runoff <strong>and</strong><br />
hydraulic gradient between the water level in the<br />
river <strong>and</strong> the well in the preceding year. Spatial<br />
interpolation between the wells is performed by<br />
triangulation. The main disadvantage of the<br />
statistical approach is that it is only valid within<br />
the range of values used to fit the regression<br />
equations. These limitations should be kept in<br />
mind when applying the tool, especially when<br />
defining future management scenarios that might<br />
create environmental conditions that go beyond<br />
these limits. Although, within the variable ranges<br />
dealt with in scenarios for the Amudarya river<br />
delta, the statistical approach is fully applicable.<br />
The occurrence of floods is simulated using a rulebased<br />
approach. If a certain runoff threshold at the<br />
gauging station at the entrance to the northern delta<br />
is exceeded a flood occurs. The threshold was<br />
determined by analysis of historic runoff data <strong>and</strong><br />
knowledge of years <strong>and</strong> months when floods had<br />
occurred [Schlüter et al. 2003]. This threshold<br />
certainly is a parameter the habitat suitability index<br />
will react sensitive to, because it determines how<br />
often floods occur (see 4.1 Example of scenario<br />
analysis). The flood extents were estimated based<br />
on satellite images of a large flood in 1998,<br />
assuming that all future floods will have the same<br />
extension. To some extent this approach can be<br />
justified by the existence of hydraulic structures<br />
(e.g. overflow dams) that assure that excess water<br />
is diverted into designated depressions.<br />
Nevertheless, the assumption of constant flooded<br />
area independent of the amount of water is<br />
certainly a strong limitation of the tool. Lack of<br />
time <strong>and</strong> data on channel geometry as well as a<br />
detailed digital elevation model for the flat<br />
territory made it impossible to construct a more<br />
realistic flood model. Here, future research <strong>and</strong><br />
modelling can significantly improve the impact<br />
assessment. At the time being, those limitations<br />
have to be taken into account in the interpretation<br />
of results of scenario analyses.<br />
Habitat Suitability Index<br />
Groundwater level, flooding frequency, flooding<br />
timing, flooding duration <strong>and</strong> geomorphology at a<br />
given site are assigned separate suitability values<br />
which are then combined to a single value – the<br />
habitat suitability index. Both, important habitat<br />
variables <strong>and</strong> the response of habitat suitability to<br />
changes of the variables have been determined on<br />
the basis of expert knowledge, because available<br />
data proved to be not suitable for statistical<br />
analysis. To test uncertainty in the structure of the<br />
model two different methods have been compared:<br />
a multi-criteria evaluation <strong>and</strong> a fuzzy rule-based<br />
system. Both use the same habitat variables, but<br />
combination to the composite index differs [Rüger<br />
2002]. Results of the two methods are similar with<br />
slight differences in their qualitative behaviour.<br />
The outcomes of the fuzzy rule-based system are<br />
more sensitive to flood events <strong>and</strong> low<br />
groundwater tables are judged more severe than<br />
the in the multi-criteria evaluation. However, it is<br />
not possible to predict how the introduction of<br />
other potentially important habitat variables (e.g.<br />
soil salinity) would alter the outcome of the habitat<br />
model. Nevertheless, we believe that both indices<br />
represent the to date available ecological knowledge<br />
on habitat requirements of Tugai forest.<br />
Expert evaluation <strong>and</strong> validation have shown that<br />
the simulation tool provides meaningful results<br />
despite the mentioned uncertainties in system<br />
underst<strong>and</strong>ing <strong>and</strong> representation. The tool has<br />
been tested with historic runoff data from the years<br />
1991-1999. Results of habitat suitability for Tugai<br />
forest do not contradict the data on Tugai forest<br />
distribution [Rüger et al. 2004] <strong>and</strong> regional<br />
experts confirm the qualitative simulation results.<br />
5. UNCERTAINTIES CAUSED BY POLI-<br />
TICAL AND ECONOMIC DEVELOP-<br />
MENTS IN THE RIVER BASIN<br />
The tool can be applied for scenario analysis with<br />
authorities <strong>and</strong> stakeholders in the Amudarya river<br />
basin to assess the implications of uncertainties in<br />
future water availability on the ecological situation<br />
in the delta. After identifying relevant uncertainties<br />
scenarios reflecting their expected impacts on the<br />
hydrological regime can be developed <strong>and</strong> analysed.<br />
This process guarantees that a wide range of<br />
views of relevant uncertainties among the stakeholders<br />
<strong>and</strong> priorities in policy making are represented<br />
in the analysis [Alcamo 2001].<br />
5.1. Example of scenario analysis<br />
A range of plausible scenarios reflecting a decrease<br />
in inflow to the delta from 1% up to 14 %<br />
of the inflow in the reference scenario has been<br />
developed to analyse the effect of further decrease<br />
in water supply on the riverine ecosystems. Inflow<br />
to the delta in the reference scenario is based on a<br />
14- year characteristic runoff time series (monthly<br />
time steps) at the gauge at the entrance to the delta.<br />
A decrease in inflow of up to 14% is realistic since<br />
it reflects the maximum amount of water Afghanistan<br />
is entitled to withdraw from the Amudarya<br />
river. In the past 30 years Afghanistan has not<br />
been able to use this water because of war. Instead<br />
these resources are used for irrigation in the downstream<br />
countries Turkmenistan <strong>and</strong> Uzbekistan.<br />
589
Figure 3 depicts the results of the scenario runs<br />
over a time period of 28 years (mean value). In all<br />
scenarios mean monthly inflow is reduced by the<br />
respective percentage, while irrigation water<br />
dem<strong>and</strong>s remain the same as in the reference<br />
scenario. It is expected that the ecological situation<br />
deteriorates in all scenarios because the water<br />
remaining after serving irrigation needs decreases.<br />
-2% -4% -6% -8%<br />
-10%<br />
-12%<br />
Scenario<br />
Inflow -1% 13 (5,6,7); 27 (5,6,7)<br />
Inflow -2% 13 (5,6,7); 27 (5,6,7)<br />
Inflow -3% 13 (5,6,7); 27 (5,6,7)<br />
Inflow -4% 13 (5,6,7); 27 (5,6,7)<br />
Inflow -5% 13 (5,6,7); 27 (5,6,7)<br />
Inflow -6% 13 (6,7); 27 (6,7)<br />
Inflow -7% 13 (6,7); 27 (6,7)<br />
Inflow -8% 13 (5,6,7); 27 (5,6,7)<br />
Inflow -9% 13 (6,7); 27 (6,7)<br />
Inflow -10% 13 (7); 27 (7)<br />
Inflow -11% -<br />
Inflow -12% 13 (6); 27 (6)<br />
Inflow -13% -<br />
Inflow -14% -<br />
Figure 3 Results of scenario analysis for a decrease in inflow of 2% to 14% of the inflow in the reference<br />
scenario over a time period of 28 years. The maps show the main river <strong>and</strong> major canals in the Northern delta<br />
region. Colour coding depicts the difference in mean habitat suitability between simulated <strong>and</strong> reference<br />
scenario over the entire simulation period. Dark grey indicates a decrease in habitat suitability, light grey an<br />
increase. The table lists the years <strong>and</strong> months (e.g. 5=Mai) when floods occur in the respective scenario.<br />
.<br />
Scenario analysis reveals a differentiated picture of<br />
the mean situation in the delta with areas where<br />
suitability has lowered <strong>and</strong> others where the<br />
situation for Tugai forests has improved. With a<br />
decrease of inflow by 2% the situation changes<br />
only insignificantly, thus indicating that the decrease<br />
in water availability has no severe ecological<br />
impact. In the scenarios with 4% to 8% less<br />
inflow habitat suitability for the Tugai forests<br />
visibly deteriorates along the river <strong>and</strong> on some<br />
river bars in the southern part of the study area.<br />
The area affected increases with decreasing inflow<br />
as expected, although differences to the reference<br />
are still small. The scenarios also show an increase<br />
in suitability in the northern former swampy part<br />
of the delta. This is caused by a lowering of the<br />
groundwater table to more suitable levels. Stakeholders<br />
<strong>and</strong> experts have to judge whether an<br />
establishment of forests in these regions would be<br />
feasible. In scenarios 10% <strong>and</strong> 12% there are areas<br />
in the southern part of the delta where suitability<br />
has significantly decreased. These are mainly<br />
formerly flooded areas, which now lack flooding.<br />
In the scenario of a decrease in inflow of 14%<br />
habitat suitability decreases in the entire delta with<br />
-14%<br />
Flood<br />
large areas strongly impacted. In this scenario no<br />
floods occur at all, a fact that strongly affects<br />
habitat suitability.<br />
Floods in the 28 year reference scenario occur in<br />
year 13 (Mai, June, July) <strong>and</strong> in the same months<br />
of year 27 (Fig 2). Up to a decrease in inflow by<br />
5% the flooding regime remains the same. Further<br />
decrease causes a shortening of the flooded period.<br />
The lack of one flooded month has no severe effect<br />
on suitability. With further decrease in inflow to<br />
the delta (Scenario 10% - 12%) the flood only<br />
occurs in July in year 13 <strong>and</strong> 27, which has a<br />
pronounced effect on habitat suitability as mentioned<br />
above.<br />
The analysis shows that there are thresholds in the<br />
decrease of monthly inflow beyond which habitat<br />
quality lowers significantly, mainly due to changes<br />
in the flooding regime. As long as floods still<br />
occur at least for one month changes in habitat<br />
suitability show a spatially differentiated picture<br />
with some areas deteriorating others improving.<br />
This information can support the planning of specific<br />
measures to cope with the uncertainty in inflow<br />
by e.g. guaranteeing a certain flow into a<br />
selected area. Although, when planning rehabilita-<br />
590
tion measures other factors than those accounted<br />
for in the TUGAI tool will also have to be considered,<br />
such as soil salinity or closeness of the site<br />
to human settlements. Joint scenario development<br />
<strong>and</strong> analysis with local stakeholders can point to<br />
human factors that are not explicitly taken into<br />
account in the models. The experience of the local<br />
stakeholders with their environmental system adds<br />
important aspects to the analysis that have not <strong>and</strong><br />
often cannot be included in the simulation tool.<br />
6. CONCLUSIONS<br />
Application of simple simulation tools, such as the<br />
TUGAI tool presented here, for scenario development<br />
<strong>and</strong> analysis with river basin authorities <strong>and</strong><br />
stakeholders facilitates discussions on major uncertainties<br />
that will affect river basin management<br />
in future. It can help to identify relevant uncertainties<br />
<strong>and</strong> to assess their implications from the point<br />
of view of the stakeholders. Different views might<br />
arise between various groups in the initial assessment<br />
<strong>and</strong> the interpretation of model results. These<br />
activities raise awareness <strong>and</strong> can support an ongoing<br />
process of determining goals <strong>and</strong> decision<br />
making [Pahl-Wostl 1998].<br />
The aim of the TUGAI tool is to facilitate such a<br />
process of analysis <strong>and</strong> decision making among<br />
authorities <strong>and</strong> stakeholders for the development of<br />
future water management strategies in the Amudarya<br />
river delta. Scenario outcomes reveal tendencies<br />
<strong>and</strong> magnitude of changes to the ecological<br />
condition in the delta that might result from<br />
changes to the hydrological regime. In order to<br />
provide scenario results real time in a workshop<br />
setting as input for discussions the tool has to be<br />
kept simple. Although uncertainties in model outcomes<br />
have been assessed <strong>and</strong> quantified as far as<br />
possible, a final judgment on the range of uncertainties<br />
inherent in the impact assessment versus<br />
the magnitude of expected future changes caused<br />
by external factors remains difficult. We believe<br />
that the tool is a good representation of the<br />
currently available knowledge <strong>and</strong> is simple<br />
enough that the user can underst<strong>and</strong> its dynamics<br />
<strong>and</strong> inherent uncertainties. It is crucial that<br />
uncertainties are evaluated at each stage of the<br />
impact assessment <strong>and</strong> that assumptions used in<br />
the representation of the system are clearly communicated<br />
[Carter 2001]. A dialogue with local<br />
experts which have large informal knowledge<br />
about ecological processes in the study area is important<br />
for further validation. In the given example<br />
the key role of floods is visible. The results can<br />
initiate discussions on the necessary measures to<br />
guarantee minimum amount of floods to maintain<br />
the forests <strong>and</strong> create awareness among those responsible<br />
for water allocation to the ecological<br />
effects of their measures. They might also catalyse<br />
a discussion on the appropriateness of the assumptions<br />
on the role of floods. Such interactive<br />
processes can also support specification of future<br />
research <strong>and</strong> data needs of a specific situation.<br />
7. REFERENCES<br />
Alcamo, J., Scenarios as tools for international environmental<br />
assessments. Env. Issue Rep.,<br />
24, European Environment Agency, 2001.<br />
Carter, T.R., Uncertainties in Assessing the Impacts<br />
of Regional Climate Change. In: M.<br />
B. India & D.L. Bonillo (eds.) Detecting<br />
<strong>and</strong> <strong>Modelling</strong> Regional Climate Change,<br />
Springer, Berlin, Germany, 441-469, 2001.<br />
Clark, M. J., Dealing With Uncertainty: Adaptive<br />
Approaches to Sustainable River Management.<br />
Aquatic Conservation -Marine <strong>and</strong><br />
Freshwater Ecosystems, 4, 347-363, 2002.<br />
IPCC Report, Hydrology <strong>and</strong> Water Resources.<br />
Implications of Climate Change for Water<br />
Management Policy. Climate Change 2001.<br />
Working Group II. http://www.grida.no/<br />
climate/ipcc_tar/wg2/188.htm, 2001.<br />
Kwadijk, J., Rotmans, J., The impact of climate<br />
change on the river Rhine–a scenario study.<br />
Climate Change, 30(4), 397-425, 1995<br />
Pahl-Wostl, C., Jaeger C.C., Rayner S., Schär, C.,<br />
van Asselt, M., Imboden, D. Vckovski, A.<br />
Regional Integrated Assessment <strong>and</strong> the<br />
Problem of Indeterminacy. In: P. Cebon, U.<br />
Dahinden, H. Davies, D.M. Imboden, . C.C.<br />
Jaeger (eds). Views from the Alps. Regional<br />
Perspectives on Climate Change. MIT<br />
Press, Cambridge, 435-497, 1998.<br />
Rüger, N., Habitat suitability for Populus<br />
euphratica in the Northern Amudarya delta<br />
– a fuzzy approach. Beiträge des Instituts<br />
für Umweltsystemforschung. 26. , 2002.<br />
http://www.usf.uni-osnabrueck.de/usf/<br />
beitraege/index.en.html<br />
Rüger N., Schlüter M., Matthies M. A fuzzy<br />
habitat suitability index for Populus<br />
euphratica in the Northern Amudarya delta.<br />
Submitted to Ecological <strong>Modelling</strong>, 2004.<br />
Schlüter M., Savitsky A.G., Rüger N., Lieth H.,<br />
Simulation der großräumigen Grundwasserund<br />
Überflutungsdynamik in einem degradierten<br />
Flussdelta als Basis für eine ökologische<br />
Bewertung alternativer Wassermanagementstrategien.<br />
In: J Strobl, T Blaschke,<br />
G Griesebner (eds) Angew<strong>and</strong>te Geogr.<br />
Informationsverarbeitung XV. Beiträge zum<br />
AGIT-Symposium Salzburg 2003,<br />
Wichmann, Heidelberg, 437-443, 2003.<br />
591
Schlüter M., Savitsky A.G., McKinney D.C., Lieth<br />
H.. Optimizing long-term water allocation<br />
in the Amudarya river delta - A water management<br />
model for ecological impact<br />
assessment. <strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong><br />
<strong>Software</strong>, in press 2004.<br />
Wardlaw, R.B., Barnes, J.M., Optimal Allocation<br />
of Irrigation Water Supplies in Real Time,<br />
Journal of Irrigation <strong>and</strong> Drainage<br />
Engineering 125 (6) 345-354, 1999.<br />
592
Integration of MONERIS <strong>and</strong> GREAT-ER in the<br />
Decision Support System for the German Elbe River<br />
Basin<br />
Jürgen Berlekamp a , Neil Graf a , Oliver Hess a , Sven Lautenbach a , Silke Reimer b <strong>and</strong> Michael Matthies a<br />
a<br />
Institute of <strong>Environmental</strong> Systems Research, University of Osnabrück, Osnabrück, Germany<br />
b<br />
Intevation GmbH Osnabrück,Germany<br />
Abstract: The Elbe-DSS is a tool for integrated river basin management of the German part of River Elbe<br />
basin. Various simulation models are used to assess the impact of measures such as reforestation, changes of<br />
agro-practices or efficiency of wastewater treatment plants <strong>and</strong> of external scenarios on a set of management<br />
objectives. For the assessment of nutrient <strong>and</strong> pollutant loads <strong>and</strong> impacts, MONERIS <strong>and</strong> GREAT-ER are<br />
integrated in the Elbe-DSS. MONERIS calculates nutrient inputs from diffuse <strong>and</strong> point sources on a sub<br />
catchment scale of about 1,000 km². GREAT-ER was developed as a tool for exposure assessment of point<br />
source emissions considering fate in sewage treatment plants as well as degradation <strong>and</strong> transport in rivers.<br />
Both models work on long-term scale but results are calculated for different spatial entities. GREAT-ER<br />
divides the whole river network into small segments that are linked through a routing algorithm. To integrate<br />
both models, diffuse nutrient inputs for the sub-catchments calculated from MONERIS were distributed to the<br />
river network in GREAT-ER, where further elimination <strong>and</strong> transport processes are calculated together with<br />
inputs from point sources. As an example for measures the effects of reforestation on phosphate loads <strong>and</strong><br />
concentrations in the river network were simulated. Results show a spatial heterogenic effect mainly<br />
influenced by the erosion pathway.<br />
Keywords: River Basin; Water quality; <strong>Modelling</strong>; Decision Support System<br />
1. INTRODUCTION<br />
Integrated river basin management involves all<br />
management issues related to supply, use,<br />
pollution, protection, rehabilitation <strong>and</strong> many<br />
others in a river basin. An integrated approach<br />
implies that relations between the abiotic <strong>and</strong> the<br />
biotic part of the various water systems, between<br />
ecological <strong>and</strong> economic factors <strong>and</strong> between<br />
various stakeholder interests are taken into<br />
consideration in decision processes. Over the last<br />
decades river basin management has become<br />
increasingly complex. Societal dem<strong>and</strong>s have<br />
increased the need for an improved ecological <strong>and</strong><br />
chemical quality of on use <strong>and</strong> protection of rivers<br />
<strong>and</strong> other water bodies. The pollution with a<br />
multitude of substances has lead to different views<br />
about strategies towards policy making for river<br />
basin management. The European Water<br />
Framework Directive consequently calls for a<br />
multidisciplinary approach of river basin<br />
management. A decision support system (DSS) for<br />
integrated river basin management of the German<br />
part of the Elbe river basin (Elbe-DSS) is currently<br />
under development, which involves taking into<br />
account chemical quality <strong>and</strong> ecological state of<br />
surface waters. Moreover, protection against flood<br />
<strong>and</strong> floodplain inundation as well as improvement<br />
of navigability is also part of the Elbe-DSS [BfG,<br />
2000]. The Elbe-DSS is designed to assist the<br />
competent authorities in their strategic planning for<br />
establishing programs of measures <strong>and</strong> also in the<br />
communication with stakeholders <strong>and</strong> the general<br />
public.<br />
This paper describes the integration of the models<br />
MONERIS [Behrendt et al., 1999] <strong>and</strong> GREAT-<br />
ER [Matthies et al., 2001] for the assessment of<br />
nutrient <strong>and</strong> pollutant loads <strong>and</strong> impacts.<br />
2. DESIGN OF ELBE-DSS<br />
2. 1 General structure<br />
The Elbe-DSS is the first project that covers<br />
strategic water policy issues of different spatial <strong>and</strong><br />
593
temporal scale for a large river basin. Starting from<br />
a feasibility study [BfG, 2001] user needs were<br />
identified by repeated discussion with<br />
representatives from international, national,<br />
regional <strong>and</strong> local authorities. Since many projects<br />
were carried out in the Elbe river basin after the<br />
German reunion in 1990 several simulation models<br />
<strong>and</strong> data sets have been readily available. This<br />
provided a current <strong>and</strong> comprehensive basis for the<br />
development of the Elbe DSS as a flexible <strong>and</strong><br />
user-friendly system.<br />
A feasibility study was conducted <strong>and</strong> lead to a<br />
preliminary system design, which was<br />
consequently stepwise refined. A set of external<br />
scenarios <strong>and</strong> measures for various management<br />
objectives were identified. Appropriate models<br />
were selected which deliver indicators to compare<br />
the impacts of specific measures <strong>and</strong> to support<br />
decisions to meet user requirements [Matthies et<br />
al., 2004]. Data sets of the catchment <strong>and</strong> river<br />
network were collected to support the model<br />
calculations.<br />
external scenario<br />
Catchment module<br />
catchment characteristics<br />
manag. objective<br />
measure<br />
discharge<br />
substance load<br />
River network module<br />
Main channel module<br />
charakteristics of<br />
river network<br />
river water<br />
flow<br />
river water<br />
quality<br />
Characteristics of main channel<br />
hydraulics flood risk<br />
water quality<br />
ecology<br />
Floodplain module<br />
floodplain hydraulics<br />
Flooplain<br />
characterics<br />
flood risk<br />
ecology<br />
To meet the requirements of the various spatial<br />
scales a design of four linked subsystems<br />
(modules) was chosen (Fig. 1). The catchment<br />
module assesses impacts of l<strong>and</strong> use <strong>and</strong> human<br />
activities on quantity <strong>and</strong> quality of drainage<br />
components. Hydrology <strong>and</strong> water quality of the<br />
river network with 33,500 km river length in the<br />
German Elbe river basin is simulated in the river<br />
network module. The main channel module<br />
describes water flow <strong>and</strong> ecology only in the main<br />
Elbe river. The floodplain module characterizes in<br />
more detail hydrology <strong>and</strong> ecology in a selected<br />
river stretch of 10 km length.<br />
Management objectives, external scenarios <strong>and</strong><br />
measures are specified at the level of single<br />
modules. A management objective describes the<br />
desired status to reach legal or other requirements,<br />
e.g. reduction of substance loads. External<br />
scenarios are defined as development pathways<br />
which are determined by climatic, hydrologic,<br />
Figure 1. General systems diagram<br />
socio-economic <strong>and</strong> ecological changes (e.g.<br />
climate change). A measure means a potential<br />
action which can be chosen to reach a management<br />
objective (e.g. reforestation, reduction of<br />
impervious areas or changes of agro practice).<br />
2.2 <strong>Software</strong> concept<br />
The Elbe-DSS is implemented using the DSSgenerator<br />
software Geonamica ® developed by<br />
RIKS [Hahn <strong>and</strong> Engelen, 2000]. It contains a<br />
GIS-based user interface, which allows flexible<br />
easy-to-use access to pre- <strong>and</strong> user-defined<br />
scenarios. Furthermore, a data base management<br />
system (DBMS), model base management system<br />
(MBMS) <strong>and</strong> a knowledge-based tool box are<br />
integrated under the graphical user interface.<br />
Evaluation tools have been provided for various<br />
kinds of decision-making, e.g. risk-based for<br />
hazardous pollutant concentrations, monetarybased<br />
for engineering measures or ecological<br />
services for floodplain restoration.<br />
594
Climate<br />
change<br />
external scenarios<br />
HBV / MONERIS HBV / MONERIS MONERIS / GREAT-ER<br />
L<strong>and</strong> use change<br />
Socio-economic<br />
change<br />
Catchment module<br />
System<br />
Statistiken oder<br />
Wetter Generator<br />
River basin -<br />
characteristics<br />
GTOPO30<br />
Topography<br />
rainfall<br />
Soil<br />
properties<br />
BGR<br />
Hydrogeology<br />
L<strong>and</strong>use<br />
CORINE<br />
measures<br />
Renaturation<br />
of floodplains<br />
Ecological<br />
drainage<br />
management<br />
reforestation<br />
Reduct. of impervious<br />
areas<br />
HBV<br />
HBV<br />
HBV<br />
MONERIS<br />
MONERIS<br />
HBV<br />
Evapotranspiration<br />
GREAT-ER / MONERIS<br />
Amount of<br />
discharges<br />
MONERIS<br />
Surface runoff<br />
Interflow<br />
Infiltration<br />
HBV<br />
Base flow<br />
River<br />
water<br />
flow<br />
Quality of<br />
discharge<br />
GREAT-ER<br />
Point<br />
sources<br />
Non-point<br />
sources<br />
MONERIS<br />
Changes in<br />
agro practice<br />
Changes of live<br />
stock sizes<br />
Treatment plant<br />
efficency<br />
MONERIS<br />
MONERIS<br />
GREAT-ER<br />
MONERIS<br />
reduction of<br />
substance loads<br />
R<strong>and</strong>streifen-<br />
Programme<br />
MONERIS<br />
objectives<br />
(indicators)<br />
To river network module<br />
To river network module<br />
Figure 2. System diagram of the catchment module<br />
2.3 Integrating GREAT-ER <strong>and</strong> MONERIS<br />
Nutrient loads (phosphorus, nitrogen) in the Elbe-<br />
DSS are calculated by the MONERIS model<br />
[Behrendt et al., 1999]. It is parameterised for 132<br />
sub catchments in the German Elbe river basin <strong>and</strong><br />
allows the average long-term simulation of P- <strong>and</strong><br />
N-loads from point <strong>and</strong> non-point sources.<br />
For the river network, GREAT-ER is integrated<br />
into the Elbe-DSS [Matthies et al., 2001; Matthies<br />
et al., 2003]. The whole digital river network is<br />
divided into reaches of about 2 km length giving a<br />
number of approximately 33,500 reaches in the<br />
German part of the Elbe River (without tide<br />
influenced coastal sub-catchments). GREAT-ER<br />
delivers concentrations of hazardous substances<br />
released by point sources, e.g. sewage treatment<br />
plants. The approach is similar to the CatchMODS<br />
system of Newham et al. [2004].<br />
Before integrating these models into the Elbe-DSS<br />
interfaces for communication of the models have to<br />
be defined. MONERIS as well as GREAT-ER are<br />
largely compatible at time <strong>and</strong> spatial scales. Both<br />
models calculate average long-term conditions<br />
without explicitly considering temporal dynamics.<br />
GREAT-ER is a steady state model <strong>and</strong> MONERIS<br />
calculates results for periods of many years. Both<br />
models focus on broad-scales <strong>and</strong> treat processes at<br />
a comparable spatial resolution.<br />
At the current state the Elbe-DSS is able to model<br />
the nutrients nitrogen <strong>and</strong> phosphorus from point<br />
<strong>and</strong> diffuse inputs. Pollutants modelled only from<br />
point sources are diclofenac, paracetamol (pharmaceuticals),<br />
EDTA (washing agent), HHCB<br />
(polycyclic musk fragrance) <strong>and</strong> boron.<br />
Industrial discharges<br />
Percentage<br />
degradation<br />
(substance<br />
independent)<br />
WWTP<br />
efficiency<br />
Discharge<br />
sewer system<br />
WWTP<br />
model<br />
River<br />
model<br />
Substance<br />
concentrations<br />
GREAT-ER<br />
domestic inputs<br />
Degredation<br />
rate<br />
(substance<br />
dependent)<br />
erosion<br />
Diffuse<br />
inputs<br />
impervious<br />
areas<br />
MONERIS<br />
surface runoff<br />
drainage<br />
model or<br />
modelled process<br />
input data or<br />
parameter<br />
model output<br />
atmospheric<br />
deposition<br />
groundwater<br />
Figure 3. Scheme of integrating data <strong>and</strong> processes<br />
from GREAT-ER <strong>and</strong> MONERIS<br />
595
Some model components such as processes<br />
describing the fate of nutrients from point sources<br />
exist in both models. While MONERIS only<br />
provides overall emissions per sub-catchment,<br />
GREAT-ER takes into account locations as well as<br />
technical st<strong>and</strong>ards. Due to this more detailed<br />
approach that allows measures on a basis of single<br />
treatment plants this input pathway is modelled by<br />
GREAT-ER. Other details of municipal waste<br />
water treatment like sewer systems overflows are<br />
not reflected in GREAT-ER <strong>and</strong> are calculated in<br />
MONERIS. Hence the models could not be used as<br />
entire packages <strong>and</strong> it was necessary to underst<strong>and</strong><br />
<strong>and</strong> separate components of both models (Fig. 3).<br />
The discharge of diffuse inputs into the river<br />
system calculated by MONERIS is realized by<br />
linking catchments to corresponding river reaches.<br />
The amount of diffuse emissions is distributed to<br />
the river reaches of a catchment by weighted<br />
length. In the river itself diffuse nutrient loads <strong>and</strong><br />
inputs from point sources are merged <strong>and</strong><br />
calculated by GREAT-ER concerning transport,<br />
decay <strong>and</strong> elimination processes.<br />
4. SIMULATED EFFECTS OF RE-<br />
FORESTATION ON PHOSPHORUS-<br />
LOADS AND -CONCENTRATIONS<br />
4.1 Calculation<br />
Phosphate is a major cause of eutrophication of<br />
fresh water bodys. Large parts of the Elbe River<br />
<strong>and</strong> its tributaries are still in a eutrophic or<br />
oligotrophic state although much effort has been<br />
made in the last decade to improve the Elbe river<br />
water quality.<br />
Figure 4. Effects of reforestation of agricultural crop l<strong>and</strong>. Converted areas (a), reduction of phosphorus<br />
emissions (b) <strong>and</strong> resulting reduction of phosphate concentration in the river network (c) are shown.<br />
As an example to illustrate a potential application<br />
of the DSS with the two models, the effect of a<br />
reforestation measure was analysed. For this<br />
purpose, agricultural cropping l<strong>and</strong> was evaluated<br />
regarding slope <strong>and</strong> soil properties <strong>and</strong> less<br />
suitable areas for agriculture were identified<br />
596
(Figure 4a) based on this evaluation. These areas<br />
suitable for reforestation were allocated to<br />
MONERIS catchments <strong>and</strong> phosphorus emissions<br />
were recalculated. Whilst not shown in this paper,<br />
the DSS would readily allow the analysis of other<br />
options such as intensifying agriculture in other<br />
areas to compensate for production losses or<br />
erosion potential <strong>and</strong> pathways. MONERIS<br />
reforestation effects are internally represented by<br />
reducing soil mobilisation <strong>and</strong> reduction of soil<br />
loss ratio In this scenario it was assumed that the<br />
loss of agricultural crop l<strong>and</strong> was not compensated<br />
by intensive agriculture in other areas.<br />
4.2 Results<br />
The low mountain range of Erzgebirge <strong>and</strong><br />
Voigtl<strong>and</strong> (south-east border of the Elbe river<br />
basin) show the strongest effect of the measure.<br />
Here, high percentages of converted areas correlate<br />
with high soil erosion caused by high relief energy<br />
(Figure 4b). Diffuse phosphorus emissions are<br />
decreased up to 60% for some catchments.<br />
The calculated changes of P-concentrations in the<br />
river network compared to the reference situation<br />
show similar results (Fig. 4c): high reductions up<br />
to 60 % occur in the streams of Erzgebirge <strong>and</strong><br />
Voigtl<strong>and</strong>. The pattern of calculated concentrations<br />
is similar to the P-emissions because variations are<br />
only caused by changes from diffuse sources <strong>and</strong><br />
these are calculated by MONERIS on a subcatchment<br />
scale. In more detail decreases of P-<br />
concentrations can be observed following the<br />
courses of the rivers Saale, Weiße Elster <strong>and</strong><br />
Spree.<br />
A profile along the courses of the rivers Saale <strong>and</strong><br />
Elbe shows relevant relative decreases of P-<br />
concentrations mainly in the upper reaches (Fig. 5).<br />
Confluences with rivers from regions with lower<br />
reduction of erosion (inflow from river Unstrut at<br />
Naumburg, mouth of river Saale near Magdeburg)<br />
weaken this effect.<br />
Comparisons with monitoring data allow<br />
estimating the uncertainty of the results. While<br />
monitoring data of river Elbe correspond with the<br />
model results very well (Fig. 6) differences exist<br />
for the rivers Havel, Spree <strong>and</strong> Mulde. Here model<br />
results differ from monitoring data by a relative<br />
error up to 30 % which may be caused by data<br />
lacks for waste water treatment parameters or<br />
known weaknesses of modelling hydrological flow.<br />
Figure 6. Comparison of simulated phosphorus<br />
concentrations with monitoring results along Elbe<br />
main channel.<br />
Figure 7. Comparison of simulated phosphorus<br />
concentrations with all available monitoring data.<br />
Figure 5. Effects of reforestation on the<br />
phosphorus concentration in Saale River <strong>and</strong> Elbe<br />
main channel. Relative changes on a profile from<br />
Saale headwaters to Geesthacht weir are shown.<br />
5. OUTLOOK<br />
The example of the reforestation measure<br />
demonstrates the general applicability of Elbe-<br />
DDS for sustainable water management. Here the<br />
Elbe-DSS can help the user to simulate the effect<br />
of measures on the management objectives. In this<br />
case it can be used to derive the river basin<br />
management plans of the EU Water Framework<br />
Directive.<br />
597
In the next step all indented measures <strong>and</strong><br />
scenarios will be implemented in Elbe-DSS. This<br />
also includes the main channel module <strong>and</strong><br />
floodplain module that are not described here in<br />
detail. Model results of the integrated system has to<br />
be checked carefully also by using uncertainty<br />
analysis. It is also strongly intended to<br />
communicate uncertainty of results of the Elbe-<br />
DSS to the users.<br />
While the prototype is based on mean long-term<br />
hydrological time series, the rainfall-runoff model<br />
HBV-D [Krysanova et al., 1999] will be integrated<br />
into the final Elbe-DSS.<br />
Tools for analysis <strong>and</strong> comparison of results will<br />
be integrated into the final software. Also tools for<br />
economical evaluation (e.g. cost-benefit analysis)<br />
will be available for selected measures <strong>and</strong><br />
scenarios after finishing the development of the<br />
Elbe-DSS. These tools will help the user to<br />
compare different possible alternatives <strong>and</strong><br />
prioritise management interventions. For instance it<br />
will be possible to compare measures like<br />
reforestation with changes of agro-practice or<br />
improvement of wastewater treatment plants<br />
regarding their effects on enhancing the ecological<br />
<strong>and</strong> chemical state of the rivers.<br />
As already done for the development of the whole<br />
Elbe-DSS all intended tools will be developed in<br />
close collaboration with the end users to reach their<br />
requirements.<br />
Hahn, B. <strong>and</strong> G. Engelen, Concepts of DSS<br />
Systems, in: BfG 2000, 9-44, 2000.<br />
Krysanova, V., A. Bronstert <strong>and</strong> D.-I. Wohlfeil:<br />
<strong>Modelling</strong> river discharge for large drainage<br />
basins: from lumped to distributed approach,<br />
Hydrological Sciences, 44(2), 313-331, 1999.<br />
Matthies, M., J. Berlekamp, F. Koormann <strong>and</strong> J. O.<br />
Wagner: Geo-referenced regional simulation<br />
<strong>and</strong> aquatic exposure assessment. Water<br />
Science <strong>and</strong> Technology, 43(7), 231-238,<br />
2001.<br />
Matthies, M., J. Berlekamp, S. Lautenbach, N.<br />
Graf <strong>and</strong> S. Reimer, Decision Support System<br />
for the Elbe River Water Quality Management.<br />
<strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong><br />
<strong>Software</strong>, 2004.<br />
Matthies, M., <strong>and</strong> J. Klasmeier, Geo-referenced<br />
stream pollution modeling <strong>and</strong> aquatic<br />
exposure assessment, in: D.A. Post (Ed),<br />
ModSim 2003, Integrative <strong>Modelling</strong> of<br />
Biophysical, Social, <strong>and</strong> Economic Systems<br />
for Resource Management Solutions,<br />
<strong>Modelling</strong> <strong>and</strong> Simulation Society of<br />
Australia <strong>and</strong> New Zeal<strong>and</strong> Inc., Canberra,<br />
666-671, 2003.<br />
Newham, L. T. H., R. A. Letcher, A. J. Jakeman<br />
<strong>and</strong> T. Kobayashi, A Framework for<br />
Integrated Hydrologic, Sediment <strong>and</strong> Nutrient<br />
Export <strong>Modelling</strong> for Catchment-Scale<br />
Management, <strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong><br />
<strong>Software</strong>, 2004.<br />
6. ACKNOWLEDGEMENTS<br />
The authors wish to thank the DSS development<br />
team for their kind collaboration, Bundesanstalt für<br />
Gewässerkunde (German Federal Institute for<br />
Hydrology), Federal Minister for Education <strong>and</strong><br />
Research <strong>and</strong> Federal <strong>Environmental</strong> Protection<br />
Agency for their financial support <strong>and</strong> data supply.<br />
7. REFERENCES<br />
Behrendt, H., P.-H. Huber, D. Opitz, O. Schmoll,<br />
G. Scholz <strong>and</strong> R. Uebe, Nährstoffbilanzierung<br />
der Flussgebiete Deutschl<strong>and</strong>s, UBA<br />
Texte 75/99, Berlin, Germany, 1999.<br />
BfG (Bundesanstalt für Gewässerkunde, German<br />
Federal Institute of Hydrology), Decision<br />
Support Systems (DSS) for river basin<br />
management. Koblenz, Germany, 2000.<br />
BfG (Bundesanstalt für Gewässerkunde, German<br />
Federal Institute of Hydrology), Towards a<br />
Generic Tool for River Basin Management -<br />
Feasibility study, Koblenz, Germany,2001.<br />
598
An integrated tool for water policy in agriculture<br />
G.M. Bazzani -National Research Council IBIMET, V.Gobetti 101, 40129 Bologna, Italy<br />
Mail: G.Bazzani@ibimet.cnr.it Fax: +39 051 6399204<br />
Abstract: The definition of proper tools to support the implementation of Water Framework Directive<br />
(WFD) is an urgent task in the European Union (EU). Agriculture deserves special attention since in most<br />
countries water consumption is higher than in other sectors <strong>and</strong> pollution due to irrigated agricultural activity<br />
is often a serious problem, while social <strong>and</strong> cultural issues are relevant. The paper presents a program called<br />
DSIRR designed to conduct an integrated analysis of water use in agriculture considering agronomic,<br />
hydraulic, economic <strong>and</strong> environmental aspects as well as complexity <strong>and</strong> uncertainty for decision making.<br />
The tool permits to analyze in great detail the relevant production systems existing in a catchment integrating<br />
stakeholders perspectives. The impact of markets, water <strong>and</strong> agricultural policies, climate, technological<br />
innovation can be assessed <strong>and</strong> the ex-ante analysis of economic instruments, suggested by WFD for cost<br />
recovery according with polluter-pays principle, conducted. Scenario analysis is used to cope with<br />
uncertainty. The paper presents a study conducted in the Po Basin in Italy comparing the impact of a water<br />
pricing <strong>and</strong> of the EU agricultural policy reform on annual <strong>and</strong> perennial crops systems. A set of indicators<br />
quantifies important differences in social, economical <strong>and</strong> environmental dimensions <strong>and</strong> suggests to adopt<br />
selective interventions. The results permit to appreciate the relevance of the tool to generate information to<br />
support the participatory policy process of basin plan implementation. A graphical user interface, a modular<br />
architecture, an open structure, a rich set of models, st<strong>and</strong>ardized database, make DSIRR a flexible <strong>and</strong><br />
powerful tool for a more sustainable agriculture <strong>and</strong> a sound water policy in agriculture.<br />
Keywords: Decision Support, Water, Agriculture, Economic analysis, Policy<br />
1 INTRODUCTION<br />
There is nowadays a strong agreement that water is<br />
a strategic resource which requires protection <strong>and</strong><br />
intervention. The 2000/60/EC Directive, known as<br />
Water Framework Directive (WFD), defines the<br />
basic principles of sustainable water policy in the<br />
European Union (EU). The Directive requires an<br />
integrated participative water resources policy,<br />
which simulation models <strong>and</strong> decision support DS<br />
should support. An impressive activity is currently<br />
observed in the field of integrated catchment<br />
modelling not only in EU. In fact the analysis <strong>and</strong><br />
modelling of human-technology-environment<br />
systems <strong>and</strong> the implications of complexity <strong>and</strong><br />
uncertainty for management concepts <strong>and</strong> decision<br />
making represent a promising approach which<br />
requires the contribute of scientists working in<br />
different fields <strong>and</strong> disciplines.<br />
This paper presents a program called “Decision<br />
Support for Irrigation” (DSIRR), which focuses on<br />
water use <strong>and</strong> policy in agriculture, integrating<br />
economic models with agronomic <strong>and</strong> engineering<br />
information. The contribution is organized as<br />
follows. First the policy context is briefly analyzed.<br />
The requirement for modelling irrigated agriculture<br />
for policy analysis is discussed in the second<br />
section, while the third one describes the tool in a<br />
non technical way. Results from an Italian case<br />
study focusing on water pricing in the Po Basin are<br />
illustrated in the next section. The final section<br />
presents conclusions <strong>and</strong> suggestions for further<br />
development based on the described model.<br />
2 THE POLICY CONTEXT<br />
2.1 Water policy in Europe<br />
WFD aims to reach within 2015 a “good status” for<br />
all water. Economic analysis <strong>and</strong> instruments<br />
receive great attention in the Directive. At this<br />
regard it is clearly pointed out that the principle of<br />
recovery of the costs of water services, including<br />
environmental <strong>and</strong> resource costs, should be<br />
adopted in accordance with the polluter-pays<br />
principle (Preamble 38, Articles 9 & 13 <strong>and</strong> Annex<br />
VII). It is well recognised that an economic analysis<br />
of water services based on long-term forecasts of<br />
supply <strong>and</strong> dem<strong>and</strong> in the river basin district is<br />
necessary for this purpose. Local specificities are<br />
considered <strong>and</strong> the subsidiarity principle is<br />
suggested to deal with them. Diversity in<br />
conditions <strong>and</strong> needs should be taken into account<br />
in the planning <strong>and</strong> execution of measures to<br />
ensure protection <strong>and</strong> sustainable use of water in<br />
the framework of the river basin. WFD asks member<br />
599
states to conduct a disaggregate analysis into at<br />
least the tree main economic sectors: industry,<br />
households <strong>and</strong> agriculture. But is some cases,<br />
particularly when social conflict due to water<br />
scarcity <strong>and</strong>/or environmental problems is high, the<br />
level of detail could be much higher. In those cases<br />
the comprehension of the mechanisms which<br />
determine water pattern uses <strong>and</strong> actors<br />
behaviours could be necessary to design proper<br />
interventions <strong>and</strong> policies.<br />
2.2 The EU Common Agricultural Policy<br />
The EU the Common Agricultural Policy (CAP) is<br />
currently experiencing a new reform, the so called<br />
Mid Term Review 2003. The reform aims to solve<br />
internal <strong>and</strong> external conflicts <strong>and</strong> proceeds along<br />
the path started in 1992 <strong>and</strong> reinforced in 1999 with<br />
Agenda 2000. CAP looks for social consensus, in a<br />
context of EU enlargement <strong>and</strong> market<br />
globalization, facing severe budgetary constraints.<br />
The current reform shifts the focus on a more<br />
sustainable agriculture with out giving up the farm<br />
income support. The reform moves in the direction<br />
of a decoupled policy with internal prices more in<br />
line with the world market, which means lower<br />
prices for most commodities, <strong>and</strong> farm income<br />
support in the form of direct payments to<br />
compensate for the previous reduction. Decoupling<br />
supports from production will reduce market<br />
distortions, while modulation <strong>and</strong> ecoconditionality<br />
of farm support will guarantee equity<br />
<strong>and</strong> environmental sustainability, respectively.<br />
3 MODELLING IRRIGATED AGRICULTURE<br />
FOR POLICY<br />
3.1 Water <strong>and</strong> agriculture<br />
The tool here presented tries to support a<br />
participatory planning process for water in the<br />
agricultural sector, that in most countries shows<br />
the higher water consumption. This is particularly<br />
true in southern Europe where irrigation itself<br />
represents over 50% of total dem<strong>and</strong>. In order to<br />
design policies capable to reduce consumption <strong>and</strong><br />
increase water quality the relation water-agriculture<br />
should be addressed in all its complexity (Ward et<br />
al., 2002). A good description of the processes,<br />
considering both the technical <strong>and</strong> the behavioral<br />
aspects, should be adopted. The scale of the<br />
analysis is therefore “micro” <strong>and</strong> representative<br />
actors, the farmers, should be considered.<br />
<strong>Environmental</strong> impacts are indirect effects of their<br />
activity which should be properly addressed when<br />
social welfare is considered <strong>and</strong> WFD is a case. In<br />
an economic context water represents a production<br />
factor, which enlarges substantially the farmers’ set<br />
of choices in terms of available crops <strong>and</strong><br />
processes. Irrigation have other important effects<br />
among which the increase of production in<br />
quantitative terms is not the main one. In many<br />
cases the higher quality of production <strong>and</strong> the<br />
reduction of risk due to uncertain <strong>and</strong> unstable<br />
climate conditions are prevalent, this is particularly<br />
true for vegetables <strong>and</strong> fruit. Furthermore water<br />
permits to st<strong>and</strong>ardize production over space <strong>and</strong><br />
time, <strong>and</strong> this is becoming a stringent requirement<br />
to access global markets. In many countries<br />
irrigated agriculture contributes to Gross Domestic<br />
Product (GDP) <strong>and</strong> export in a substantial way.<br />
The relation agriculture-environment is complex.<br />
On one side, natural environment is in developed<br />
countries an artifact <strong>and</strong> the agricultural sector is<br />
the main responsible for its creation <strong>and</strong><br />
preservation 1 . On the other h<strong>and</strong> pollution due to<br />
the agricultural activity is often a serious problem.<br />
There is a strong evidence that the use of water in<br />
agriculture favors more intensive practices which<br />
are often associated with a higher use of chemicals.<br />
But the relation irrigated agriculture environmental<br />
pollution is not linear, site specific conditions are<br />
determinant for the final environmental state; so<br />
great caution should be used to derive general<br />
conclusions.<br />
An aspect which deserves attention is how water is<br />
distributed at farm level, which means irrigation<br />
technology. Differences exist among techniques in<br />
terms of efficient use of water, but also in terms of<br />
farm income <strong>and</strong> labor requirements. Sound policies<br />
can increase water saving favoring technology<br />
innovation.<br />
3.2 Water pricing, an incentive economic<br />
instrument<br />
Water charges <strong>and</strong> prices are identified in the WFD<br />
as basic measures for achieving its environmental<br />
objectives, so a key issue is the assessment<br />
whether pricing policies provide appropriate<br />
incentives for users to reduce their water uses <strong>and</strong><br />
pollution. It is therefore essential to verify ex-ante if<br />
pricing can:<br />
• create the financial incentive to shift to<br />
technologies <strong>and</strong> practices that ensure a<br />
better use of available resources;<br />
• incentive users to shift to less polluting input<br />
<strong>and</strong> practices.<br />
Economic theory explains that in general price <strong>and</strong><br />
quantity are linked by an inverse relation. This is<br />
true also for water, but this function is not liner <strong>and</strong><br />
not constant, since price is only one of many<br />
variables which influence the amount of water used<br />
(Joahansson et al., 2002). The proportionality<br />
between water bill <strong>and</strong> water used <strong>and</strong> amount of<br />
pollution discharged is not enough. A key<br />
question is how do prices lead to changes in the<br />
dem<strong>and</strong> for water? The answer depends on the<br />
1 Appreciated l<strong>and</strong>scapes depend on water availability in<br />
agriculture, many examples can be found in Italy.<br />
Furthermore irrigation networks are often used to drain<br />
rain.<br />
600
price elasticity of dem<strong>and</strong> 2 , which can be easily<br />
calculated from water dem<strong>and</strong> curves. But to derive<br />
these functions is not an easy task since historical<br />
data are generally missing; models, including<br />
economic modules, represent a viable solution.<br />
4 DSIRR<br />
DSIRR is an interactive, flexible, transparent <strong>and</strong><br />
adaptable computer based decision support (DS)<br />
developed to support the recognition <strong>and</strong> the<br />
solution of complex strategic problems for<br />
improved decision making <strong>and</strong> policy design. The<br />
tool uses data <strong>and</strong> models, provides a graphical<br />
user-friendly interface, <strong>and</strong> can incorporate the<br />
decision makers’ own insights. The previous<br />
characteristics are relevant to favor stakeholders’<br />
involvement in the basin plan definition process<br />
requested by the WFD.<br />
4.1 Which support from the tool?<br />
Two reforms, in water <strong>and</strong> in agriculture, affect the<br />
primary sector. Their conjoint analysis is therefore<br />
essential, adopting a time horizon which should<br />
also consider other major sources of uncertainties<br />
like climate change <strong>and</strong> macroeconomic conditions<br />
of governance <strong>and</strong> markets. In this respect the<br />
support coming from DSIRR could be valuable<br />
since it permits to develop some of the economic<br />
analysis requested by the WFD, it aims to:<br />
• Conduct an economic analysis of water uses in<br />
agriculture at River Basin level but considering<br />
the relevant difference existing among the<br />
coexisting production systems;<br />
• To assess trends in water dem<strong>and</strong> according<br />
with different scenario for markets, agricultural<br />
policy <strong>and</strong> climate;<br />
• To assess the potential role of water pricing<br />
<strong>and</strong> its implications on cost-recovery;<br />
• To assess the impact of other water policies<br />
(e.g. environmental taxes, subsidies <strong>and</strong><br />
restriction in water supply);<br />
• To assess the impact of innovation in irrigation<br />
technology as well as in agriculture (e.g. new<br />
crops <strong>and</strong> varieties less water dem<strong>and</strong>ing or<br />
more resistant to plant diseases or water<br />
stress).<br />
In all these cases DSIRR permits a multidimensional<br />
assessment quantifying:<br />
• The sustainability of irrigated agriculture for<br />
farmers in terms of farm income;<br />
• The social implication in terms of employment;<br />
• The environmental pressure of the agricultural<br />
activity via selected indicators.<br />
4.2 The DS: a non technical description<br />
From the existing literature emerge that economic<br />
2 Elasticity is an index which reveals how the dem<strong>and</strong> is<br />
responsive to price change.<br />
models seem well suited to describe <strong>and</strong> analyze<br />
decision process <strong>and</strong> policy. A body of economic<br />
literature focuses on agriculture <strong>and</strong> irrigation. The<br />
consideration of stakeholders’ preferences <strong>and</strong><br />
their inclusion into models is an important<br />
requirement to predict the effect of policy<br />
intervention. Recent literature shows that<br />
multicriteria (MC) paradigm favors a good<br />
description of farmers’ behavior (Berbel et al., 1998;<br />
Gómez-Limón et al., 2000 <strong>and</strong> 2002). Following this<br />
approach DSIRR analyses the conjoint choice of<br />
crop mix, irrigation level, technology <strong>and</strong><br />
employment as an optimization problem <strong>and</strong> the<br />
problem is cast as constraint maximization <strong>and</strong><br />
solved using mathematical programming<br />
techniques (MPT) 3 . This methodology was applied<br />
in the EU research project aimed to assess the<br />
sustainability of irrigated agriculture in the EU<br />
(WADI), in this context the program was<br />
developed <strong>and</strong> tested. DSIRR presents some<br />
interesting innovative features.<br />
A first aspect which deserves attention is the<br />
presence of a Graphical User Interface (GUI) <strong>and</strong><br />
the definition of st<strong>and</strong>ardized dataset which can be<br />
distributed. This makes DSIRR a scenario manager<br />
for predefined agro-economic behavioral models.<br />
The present beta non commercial demo version<br />
operates as a 32 bit Windows application. A<br />
modular structure enables a continuous<br />
development of the program which can be easily<br />
linked to other models. For more information see<br />
Bazzani <strong>and</strong> Rosselli Del Turco (2003).<br />
A second aspect of interest is represented by the<br />
accurate description of the agricultural production<br />
<strong>and</strong> irrigation processes.<br />
• Agricultural practices <strong>and</strong> technologies are<br />
described on the basis of an input-output<br />
approach. Agronomic, financial, commercial,<br />
policy aspects are included. Different types of<br />
soil, seasonality, market conditions can be<br />
described.<br />
• Water supply is defined at farm gate distinctly<br />
for periods <strong>and</strong> supply systems considering<br />
different provision levels. This permits to<br />
analyze different tariff schemes.<br />
• Irrigation techniques are described on the basis<br />
of efficiency, energy <strong>and</strong> labour requirements,<br />
investment <strong>and</strong> operative costs <strong>and</strong> the surface<br />
covered.<br />
• Water-yield functions quantify the crop<br />
response to water in terms of production<br />
quantity 4 , their inclusion permits to identify the<br />
efficient irrigation volume by crop <strong>and</strong> type of<br />
soil on the basis of the decreasing marginal<br />
productivity of the resource.<br />
3 The models are solved using GAMS (General Algebraic<br />
<strong>Modelling</strong> System) (Brooke, 1992).<br />
4 Functions are derived via experimental research or other<br />
models.<br />
601
• Water dem<strong>and</strong> is quantified by periods on the<br />
basis of crop irrigation requirements, rain <strong>and</strong><br />
water tableau level.<br />
The user can decide case by case what is relevant<br />
<strong>and</strong> which aspects include. This option,<br />
introducing a great flexibility, makes the tool<br />
suitable for different situations.<br />
A third aspect deals with scale. A decomposition<br />
approach is adopted to reach the level adequate to<br />
the problem at h<strong>and</strong>. The spatial scale can be<br />
defined to describe in sufficient detail the<br />
complexity of the reality. Different types of farms,<br />
describing coexisting production systems (e.g.<br />
annual <strong>and</strong> perennial crops, family <strong>and</strong> industrial<br />
farms, etc. ), can be modeled <strong>and</strong> aggregated at<br />
basin scale.<br />
Scenario analysis is adopted to explore different<br />
states of the world related to macro-economic<br />
<strong>and</strong>/or climate conditions. Their use permits to deal<br />
with uncertainty in a practical way.<br />
The user can run the simulations without any<br />
specific knowledge of MPT <strong>and</strong> modelling thanks<br />
to the GUI, while some expertise in agriculture <strong>and</strong><br />
economics is requested. Utilities permit to access<br />
<strong>and</strong> modify internal databases, view reports <strong>and</strong><br />
tables, create charts. The present version can<br />
export the results to Excel in table <strong>and</strong> graphical<br />
form. Interfacing with other models <strong>and</strong> programs is<br />
easy. St<strong>and</strong>ard output includes: l<strong>and</strong> use (i.e. crop<br />
mix), agricultural practices, irrigation technologies<br />
<strong>and</strong> volumes plus a rich set of indicators. A first<br />
subset collects economic information covering<br />
private <strong>and</strong> public dimension (e.g. farm net income,<br />
contribution to GDP, etc.) A second deals with<br />
employment as social indicator (e.g. family <strong>and</strong><br />
external labor). A third assesses environmental<br />
pressures deriving from agriculture: (e.g. nitrate,<br />
chemicals, soil covering). Trade-off among<br />
conflicting objectives can be easily derived.<br />
4.3 The mathematical model of the farm<br />
Mono <strong>and</strong> multicriteria approaches are both<br />
available to represent the farmer’s decision<br />
process. In the former case the farmer acts as a<br />
profit maximizer, in the latter case the farmer’s<br />
objective function is composed of different<br />
components according to Multi Attribute Utility<br />
Theory (MAUT). The aggregate utility function<br />
assumed linear (1) requires normalization since<br />
different units are involved:<br />
+<br />
Z-Z o o<br />
o + -<br />
o o o<br />
U= w* (1)<br />
∑ Z-Z<br />
where: U represents the utility index, Z, Z + , Z -<br />
objectives values, ideal <strong>and</strong> nadir (ideal <strong>and</strong> nadir<br />
are respectively the best <strong>and</strong> worst case), w<br />
weights, o objectives.<br />
The selection of objectives <strong>and</strong> the estimate of the<br />
related weights can be derived via an interactive<br />
procedure with the decision makers, or via a noninteractive<br />
methodology proposed by Sumpsi et al.<br />
(1996), that minimizes the model results distance<br />
from observed farmers’ choices in a weighted goal<br />
programming. Income, risk, labour, technical<br />
difficulty can all be considered as possible<br />
attributes.<br />
In general the farmer’s problem is cast as a<br />
constraint maximization <strong>and</strong> in the simpler case can<br />
be formalized as 5 :<br />
max INC=<br />
{ XW , }<br />
∑∑∑{ Xc,i,s ⎡pc,i qc,i,s( wrc,i,s)<br />
+ suc-vcc,i,s⎤}<br />
c i s<br />
∑∑∑<br />
- W wp<br />
k l p<br />
k,l,p<br />
⎣<br />
k,l,p<br />
subject to:<br />
…<br />
X ir ≤ W ∀k,<br />
p<br />
s c i<br />
cis ,, c,i,s k,l,p<br />
l<br />
⎦<br />
(2)<br />
∑∑∑ ∑ (3)<br />
…<br />
where the indices represent: c crop, i irrigation<br />
level, s type of soil, k water source, l water<br />
provision level 6 , p period. To better readability<br />
variables, endogenously determined, are written in<br />
capital letters to distinguish from parameters<br />
,exogenously fixed. Symbols are: INC income (€),<br />
X c,i,s activities 7 (ha), p c,i crop market price (€/t),<br />
q c,i,s (wr c,i,s ) crop production as function of water (t),<br />
wr c,i,s crop water requirements (m 3 ), su c subsidies<br />
(€), vc c,i,s variable costs ( €), W k,l,p water<br />
consumption (m 3 ), wp k,l,p water price (€/m 3 ), ir c,i,s<br />
crop irrigation requirements (m 3 ).<br />
In equation 2, representing the farmers’ income<br />
objective function, production q is expressed as a<br />
function of water <strong>and</strong> irrigation costs are kept<br />
apart. This approach permits the derivation of<br />
water dem<strong>and</strong> function (4) via parametrization of<br />
price or quantity.<br />
W = f wp;<br />
Q<br />
( )<br />
(4)<br />
The function determines the quantity of water W<br />
dem<strong>and</strong>ed in a given district in a certain period as<br />
an inverse function of its price wp, given the farm<br />
production possibilities <strong>and</strong> characteristics Q. An<br />
upper limit is imposed on W to control water<br />
availability.<br />
5 A CASE STUDY<br />
The case study here presented considers a pricing<br />
policy in the Po Basin, the largest irrigated plain<br />
area in Italy, characterized by cold winters <strong>and</strong> hot<br />
5 This simplified formulation permits to appreciate the<br />
logic of the model. For a more complete presentation of<br />
the program see Bazzani (2004), IBIMET - Technical<br />
paper, in progress.<br />
6 Water provision levels permit to simulate an increasing<br />
pricing scheme, via blocked tariffs.<br />
7 An activity is a crop characterized by its production<br />
process, i.e. fertilization, irrigation, …; the same crop<br />
determines distinct activities if more production<br />
possibilities are considered.<br />
602
summers. The analysis compares two important<br />
cropping systems: the annual extensive (AE) <strong>and</strong><br />
intensive fruit (IF) which coexist in the region. Two<br />
agricultural regimes are analyzed: the existing CAP<br />
(A2000) <strong>and</strong> the incoming Mid Term Review<br />
(MTR). Under the current A2000, at the prevailing<br />
zero cost of water the observed crop mix is mainly<br />
given by maize, sugar beet, <strong>and</strong> soy been, all full<br />
irrigated, plus the set-aside requirement. The<br />
prevailing irrigation technique is represented by<br />
self moving gun. Calibrated the model to this<br />
situation, simulations were conducted for the two<br />
CAP regimes. Figure 2 shows water dem<strong>and</strong> (WD)<br />
<strong>and</strong> farm net income (NI) for the AE system in the<br />
water price (WP) range 0-20 8 . Consumption is on<br />
the right vertical axe, NI on the left one.<br />
NI (€/ha)<br />
600<br />
500<br />
400<br />
300<br />
200<br />
100<br />
0<br />
0 2 4 6 8 10 12 14 16 18 20<br />
WP (cents/m3)<br />
NI SB<br />
WQ SB<br />
NI MT<br />
WQ MT<br />
Figure 1. Water pricing on annual crops<br />
Rising of the WP determines three interlink<br />
adaptations regarding: crop mix, crop irrigation<br />
levels, irrigation technology, which represent<br />
endogenous variables of the models. A WP around<br />
8/10 cent €/m 3 splits the dem<strong>and</strong> curves (dotted<br />
lines) into two regions. Maize characterizes the first<br />
region with low WP but in the second leaves the<br />
field to rain fed wheat, this determines a sharp drop<br />
in the water dem<strong>and</strong>. The smaller jumps along the<br />
curve are due to the progressive decrease of crops<br />
irrigation levels. Water consumption becomes null<br />
at a WP of 20 cent €/m 3 under A2000. The impact of<br />
the MTR reduces WD in the first region, but has an<br />
opposite effect at higher prices. This depends on<br />
the relatively higher profitability of sugar beet<br />
which takes advantage of decoupled subsidies.<br />
The water saving is not at zero cost. The negative<br />
impact on NI can be visualized on the left vertical<br />
axe by the continuous lines. Under A2000 income<br />
decreases from 534 €/ha at zero price to 387 €/ha at<br />
WP 10 cent €/m 3 <strong>and</strong> to 329 €/ha at 20 cent €/m 3 .<br />
MTR function presents a similar pattern but lower<br />
values of about 5%. Water agency revenue (WAR)<br />
has a maximum at WP 8 cent €/m 3 where the entire<br />
surface is irrigated. Higher WP reduces WAR due<br />
to the reduced water consumption, this has<br />
important implication for cost recovery. Table 1<br />
reports for EA the main indicators for three price<br />
levels: the current situation (WP=0), a medium<br />
(WP=10) <strong>and</strong> a high price (WP=20).<br />
8<br />
All the figures describe main trends <strong>and</strong> should be<br />
interpreted more as probable path than exact numbers.<br />
2000<br />
1800<br />
1600<br />
1400<br />
1200<br />
1000<br />
800<br />
600<br />
400<br />
200<br />
0<br />
WQ (m3)<br />
PW NI SU GDP FL NIT PES SPR WQ<br />
Base<br />
0 534 340 1018 16 54 5068 0.44 1674<br />
10 387 340 871 15 52 4838 0.42 1372<br />
20 329 340 784 10 37 3849 0.30 0<br />
Mid Term Reform<br />
0 516 340 1003 18 64 5124 0.45 1838<br />
10 354 340 841 12 44 4135 0.32 386<br />
16 335 340 789 11 43 3863 0.29 0<br />
Table 1. Annual crops system indicators<br />
Subsidy (SU) keeps stable, since the per hectare<br />
value is in the region the same for irrigated maize<br />
<strong>and</strong> rain-fed wheat. GDP contribution <strong>and</strong><br />
employment decrease. <strong>Environmental</strong> indicators<br />
show a more articulated pattern. Nitrates (NIT) <strong>and</strong><br />
pesticides (PES) reveal decreasing pressures due to<br />
the extensivation process, which also determines a<br />
soil cover negative trend.<br />
The second production system analyzed is the fruit<br />
one which is relevant for added value <strong>and</strong><br />
employment. Figure 3 presents the impact of the<br />
same pricing policy on IF, format is unchanged.<br />
NI (€/ha)<br />
3000<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
NI SB<br />
WQ SB<br />
NI MT<br />
WQ MT<br />
0 2 4 6 8 10 12 14 16 18 20<br />
WP (cents/m3)<br />
Figure 2. Water pricing on fruit system<br />
The water dem<strong>and</strong> curves show now a completely<br />
different pattern mainly in the second region<br />
(WP>10 cent €/m 3 ) which is completely inelastic<br />
<strong>and</strong> stable at over 1300 m 3 /ha. This pattern<br />
depends on the higher marginal productivity of<br />
water in this system which is also captured by the<br />
economic indicators (NI <strong>and</strong> GDP). The reduced<br />
irrigation volume for a fruit system is due to the<br />
high efficiency of microirrigation largely adopted.<br />
Other important differences emerge in Table 2<br />
presenting the IF indicators.<br />
PW NI SU GDP FL NIT PES SPR WQ<br />
Base<br />
0 1754 44 3985 216 69 48656 0.86 1723<br />
10 1586 44 3789 216 69 48507 0.86 1485<br />
20 1371 51 3432 216 67 48811 0.84 1339<br />
Pac Reform<br />
0 1447 44 3667 216 67 48984 0.84 1615<br />
10 1298 44 3481 216 67 48811 0.84 1339<br />
20 1164 44 3347 216 67 48811 0.84 1339<br />
Table 2. Fruit system indicators<br />
Again most of the indexes show a decreasing trend<br />
in both scenario, but the magnitude are clearly<br />
1800<br />
1600<br />
1400<br />
1200<br />
1000<br />
800<br />
600<br />
400<br />
200<br />
0<br />
WQ (m3)<br />
603
much higher, confirming the intensity of the<br />
agricultural process.<br />
Comparing the results significant differences<br />
emerge. This information is relevant to design an<br />
efficient <strong>and</strong> effective water policy in the Basin.<br />
6 CONCLUSIONS<br />
DSIRR is an innovative program aimed to support<br />
water policy in agriculture via simulation behavioral<br />
models, integrating micro analysis at farm level with<br />
macro analysis at catchment scale.<br />
A Graphical User Interface permits a direct control<br />
of the simulation by the user; this feature along<br />
with flexibility, transparency <strong>and</strong> replicability,<br />
makes the tool suitable for a participative decision<br />
process. The integration of agronomic, engineering<br />
<strong>and</strong> economic aspects guarantee a good level of<br />
detail in the analysis. Farmer’s preferences are<br />
described following a multicriterial methodology<br />
which permits to integrate into the process<br />
stakeholders’ perspectives.<br />
The case study illustrated how the tool can be<br />
used to assess ex-ante the feasibility of a pricing<br />
policy in agriculture. Results point out that in the<br />
same Basin coexisting cropping systems exhibit<br />
very difference responses. In fact, while annual<br />
crops are quite sensible to a water price increase,<br />
fruit has a much more inelastic response. A pricing<br />
policy could therefore have positive effects in<br />
terms of water saving in the former but would result<br />
quite ineffective on the latter. A reduction of<br />
environmental pressures coming from agriculture is<br />
assessed but following a sensible contraction of<br />
farm income <strong>and</strong> agricultural employments. The<br />
impact of the new CAP reform decoupling<br />
subsidies from production <strong>and</strong> introducing ecosussidiarity<br />
seems to favor environmental<br />
objectives at expense of farm income <strong>and</strong><br />
employment. A trade off among conflicting<br />
environmental, socio <strong>and</strong> economic objectives<br />
emerges which the analysis can quantify leaving to<br />
the political process the final decision.<br />
DSIRR represents a practical <strong>and</strong> operational<br />
approach that could be applied by practitioners,<br />
dealing with the development of integrated river<br />
basin management plans, to assess ex-ante the<br />
effectiveness of individual <strong>and</strong> of combination of<br />
measures, when water use in agriculture were<br />
relevant. In fact it represents a bridge between<br />
science <strong>and</strong> policy, making operational economic<br />
methodologies <strong>and</strong> approaches. For its<br />
characteristics the program can be a useful tool to<br />
support discussion between experts <strong>and</strong><br />
stakeholders about alternative measures. This<br />
aspect is possibly more important than its exact<br />
predictions. Stakeholders integration into the<br />
economic analysis brings expertise <strong>and</strong><br />
information, it provides opportunities to discuss<br />
<strong>and</strong> validate key assumptions <strong>and</strong> finally it<br />
increases the ownership <strong>and</strong> acceptance of the<br />
results of the analysis. Hopefully, the program<br />
implementation in the next future will help to<br />
develop practical experience, will increase the<br />
knowledge base <strong>and</strong> will develop capacity in the<br />
integration of economics into water management<br />
<strong>and</strong> policy, favoring balanced solutions capable to<br />
achieve good water status in an efficient way with<br />
acceptable social impacts.<br />
7 REFERENCES<br />
Bazzani G., Di Pasquale S., Gallerani V., Morganti S.,<br />
Raggi M. E Viaggi D. (in print)<br />
The impact of EU water framework directive<br />
on irrigated agriculture in Italy: the case of<br />
the north east fruit district.<br />
Agricultural Economic Review.<br />
Bazzani, G., Rosselli Del Turco ,C., 2003. DSIRR: a<br />
Decision Support System for Irrigation <strong>and</strong><br />
Water Policy Design, in “Water<br />
management”, Brebbia C.A. Ed., WIT Press,<br />
Southampton-Boston, pp. 289-298.<br />
Berbel, J., Rodriguez, A.A., 1998. MCDM approach<br />
to production analysis: an application to<br />
irrigated farms in Southern Spain. European<br />
Journal of Operational Research 107, 108-<br />
118.<br />
Brooke, A., Kendrick, D., Meeraus, A., 1992. GAMS<br />
A user’s guide. The Scientific Press.<br />
European Council Directive 2000/60 Establishing a<br />
Framework for Community Action in the<br />
Field of Water Policy.<br />
Gómez-Limón, J.A., Arriaza, M., 2000. Socio-<br />
Economic <strong>and</strong> <strong>Environmental</strong> Impact of<br />
Agenda 2000 <strong>and</strong> Alternative Policy<br />
Choices for Market Liberalisation on an<br />
Irrigated Area In North-Western Spain.<br />
Agric. Economics Review 1, 18-30.<br />
Gómez-Limón, J.A., Arriaza, M., Berbel J., 2002.<br />
Conflicting Implementation of Agricultural<br />
<strong>and</strong> Water Policy in Irrigated Areas in the<br />
EU. J. of Agric. Economics 53, 259-281.<br />
Joahansson R. C., Tsur Y., Roe T.L., Doukkali R.,<br />
Dinar A., 2002. Pricing irrigation water: a<br />
review of theory <strong>and</strong> practice. Water Policy.<br />
4, 174-199.<br />
Sumpsi, J.M., Amador, F., Romero, C., 1996. On<br />
farmers’ objectives: A multi-criteria<br />
approach. European Journal of Operational<br />
Research 96, 64-71<br />
Varela-Ortega, C., Sumpsi, J. M., Garrido, A.,<br />
Blanco, M., Iglesias, E., 1998. Water Pricing<br />
Policies, Public Decision Making <strong>and</strong><br />
Farmers response: Implications for Water<br />
Policy. American Journal of Agricultural<br />
Economics 19, 193-202.<br />
Ward F.A., Michelsen A., 2002. The economic<br />
value of water in agriculture: concepts <strong>and</strong><br />
policy applications. Water policy 4, 423-446.<br />
604
Towards a Decision Support System for Real Time Risk<br />
Assessment of Hazardous Material Transport on Road<br />
D. Giglio a , R. Minciardi a,b , D. Pizzorni c , R.Rudari b,d , R. Sacile a,b , A. Tomasoni b , E. Trasforini b<br />
corresponding author: eva.trasforini@unige.it<br />
a DIST, Department of Communication, Computer <strong>and</strong> System Sciences, University of Genova, Italy<br />
b CIMA – Centro di ricerca Interuniversitario in Monitoraggio Ambientale, Italy<br />
c INTERMODE SpA, ENI Group, Genova, Italy<br />
d CNR-GNDCI, Italy.<br />
Abstract: Several emerging telematics technologies allow a new definition of the risk caused by the<br />
transport on road of hazardous material (hazmat), which can be more related to space <strong>and</strong> time varying<br />
factors. Since there are many varying factors (for example, the state of the road, of the weather, of the driver,<br />
of the hazardous material) which affects its degree, hazmat transport risk is by definition dynamic, so that<br />
conventional definitions of natural <strong>and</strong> industrial hazard <strong>and</strong> risk are not adequate. In addition, while the<br />
management systems related to the real-time planning of hazmat routing can be oriented towards the<br />
achievement of a minimum risk, on the other h<strong>and</strong>, with their decisions, they are the main actors that do<br />
influence the current value of risk on a territory. These aspects are discussed throughout this work <strong>and</strong> a<br />
proposal of decision support system integrating all these aspects is suggested, describing the necessary steps<br />
to develop it. Although the DSS is at a preliminary stage of development, an introductive demonstration of<br />
dynamic risk assessment is shown on a specific territory, namely the western districts of Liguria region (Italy)<br />
which are heavily interested by hazmat traffic, <strong>and</strong> characterized by transport infrastructures generally quite<br />
close to civil <strong>and</strong> industrial settlements.<br />
Keywords: Decision support systems, real time risk management, hazmat, logistics, transport.<br />
1. INTRODUCTION<br />
In Italy about 80% of road traffic is represented by<br />
the delivery of goods, <strong>and</strong> the overall trend in<br />
Europe seems to predict an increase of 30% within<br />
2010. About 18% of this is currently represented<br />
by hazardous material (hazmat) transportation.<br />
The current situation <strong>and</strong> the predicted trend of<br />
hazmat transportation require a particular attention<br />
as regards the definition of the risk, with attention<br />
to exposure of the population <strong>and</strong> the possible<br />
impact over the environment.<br />
Conventional approaches generally define the<br />
hazmat transport risk as an industrial risk with<br />
static components <strong>and</strong> with statistic components of<br />
the hazard. Since there are many factors (for<br />
example, the state of the road, of the weather, of<br />
the driver, of the hazardous material) which affects<br />
its degree,, hazmat transport risk is intrinsically<br />
dynamic, so that conventional definitions of natural<br />
<strong>and</strong> industrial hazard <strong>and</strong> risk are not adequate.<br />
In our work, a new approach is introduced <strong>and</strong><br />
discussed, in order to model <strong>and</strong> to assess properly<br />
the hazmat transport risk on road of petroleum<br />
products (class 3 of hazardous material<br />
classification made by Federal Motor Carrier<br />
Safety Administration, US [2001]) made by tank<br />
trucks. In particular, a dynamic evaluation of the<br />
likelihood that certain events (of assigned<br />
intensity) take place (i.e., the hazard) over the<br />
infrastructures of the considered territorial system<br />
is discussed. The preliminary activities towards the<br />
definition of a decision support system (DSS) for<br />
real time risk assessment of hazmat transport on<br />
road are also presented.<br />
This DSS, which can be also classified as an<br />
<strong>Environmental</strong> Decision Support System (EDSS)<br />
(Rizzoli <strong>and</strong> Young, 1997), is based on three<br />
modules:<br />
- the GIS based interface for the characterization<br />
of the problem <strong>and</strong> for the computation of the<br />
parameters involved in the formulation of the<br />
problem;<br />
- a real-time database where data characterizing<br />
the risk are stored;<br />
605
- the optimization module, defining the optimal<br />
tank truck routing.<br />
It is worthwhile to underline that, in the proposed<br />
DSS, the aim is to support dynamic route guidance<br />
of hazmat transport, <strong>and</strong> the problem is formalized<br />
as a mathematical programming problem, where<br />
the decisional variables are related to the routing of<br />
a fleet of trucks. In other words, the final goal of<br />
our research is to support decision makers in the<br />
hazmat transport planning, which, in a modern<br />
view, should be oriented not only to cost<br />
minimization but also towards risk minimization.<br />
Real-time hazmat transport risk assessment is so a<br />
pre-requisite to develop such a DSS, <strong>and</strong> the<br />
related methodological aspects are discussed in the<br />
next section.<br />
Finally, a preliminary demonstration of real-time<br />
hazmat transport risk assessment is presented over<br />
a particular area, namely western districts of<br />
Liguria region (Italy) which are heavily interested<br />
by hazmat traffic, <strong>and</strong> characterized by transport<br />
infrastructures generally quite close to civil <strong>and</strong><br />
industrial settlements.<br />
2. METHODS<br />
This work is part of a greater project (see<br />
acknowledgments) aiming to reduce the impact of<br />
hazmat transport on road by tank trucks of<br />
petroleum products. In this preliminary phase, a<br />
correct definition of hazmat transport risk is<br />
required. It should be observed that this definition<br />
could be related to the definition of both industrial<br />
<strong>and</strong> environmental risk. In fact, a tank truck<br />
transporting petroleum products can be viewed as a<br />
repository of a chemical product that, just as a<br />
conventional petroleum tank in a refinery, can<br />
represent a danger for the population. The main<br />
difference in this case, is that this danger is moving<br />
throughout a network of roads. So, while “static”<br />
pre-defined emergency plan <strong>and</strong> protocols are<br />
adequate to recover from an emergency in an<br />
industrial site that is settled in a precise geographic<br />
location with a clear definition of the possible<br />
interactions with the population <strong>and</strong> the<br />
environments, the same can not be assed for a<br />
moving hazard. In fact, the risk of hazmat transport<br />
dynamically changes in time <strong>and</strong> space as regards<br />
both the stress to which the hazmat transportation<br />
is subject to, to the possible impact an accident<br />
may cause. As an obvious example, a tank truck is<br />
subject to different stresses when traveling on a<br />
safe straight road with nice meteorological<br />
conditions or when traveling on a winding road<br />
during a storm. Similarly, as regards the possible<br />
consequences of an accident, obvious examples<br />
may be related to a tank truck moving either in a<br />
suburb area or on a bridge over a river, or moving<br />
over a flat l<strong>and</strong> with no important groundwater<br />
resources. In addition, a dynamic representation of<br />
impacts would be also necessary; for example, the<br />
risk released by a truck passing a residential area is<br />
certainly related to time: at night or at the weekend<br />
the potential damage will be greater than during<br />
hours of business.<br />
Dynamic risk assessment is needed since it is<br />
necessary to decide in a specific instant <strong>and</strong> in realtime<br />
the route minimizing the time/space varying<br />
risk. Dynamic risk assessment is also nowadays<br />
possible due to the several emerging telematics<br />
technologies allowing a more detailed definition of<br />
the dynamics components that affect in real-time<br />
the hazard.<br />
A proposal of dynamics definition of risk for<br />
hazmat transport is so required. In the following<br />
subsections these aspects are introduced.<br />
2.1 Conventional risk definition of industrial<br />
<strong>and</strong> environmental risk<br />
Throughout this work, the following general<br />
definitions of hazard <strong>and</strong> risk are taken into<br />
account (U.S. Department of Transportation<br />
Research [2000]). Hazard is related to the intrinsic<br />
characteristic of a material, condition, or activity<br />
that has the potential to cause harm to people,<br />
property, or the environment, <strong>and</strong> it is often<br />
defined in terms of a probability. Risk is related to<br />
the combination of the likelihood <strong>and</strong> the<br />
consequence of a specified hazard being realized.<br />
In the context of industrial hazards, risk is<br />
generally defined as a function (most frequently a<br />
product) of the likelihood frequency of a hazardous<br />
event <strong>and</strong> of its related magnitude in terms of<br />
damage on people, property or the environment.<br />
In the context of natural hazards, the definition by<br />
UNESCO [1972] is generally adopted, which<br />
allows computing the risk on a set of territorial<br />
elements that may be damaged by a natural hazard,<br />
as a function (specifically, a product) of the<br />
likelihood of the hazard, of the value of elements<br />
(people, property, or the environment) at risk, <strong>and</strong><br />
the so called vulnerability, that is the capacity of an<br />
element to resist to a hazard event.<br />
It is quite evident that these two definitions are<br />
somehow equivalent: both of them include a term<br />
related to the probability of the hazard, <strong>and</strong> both<br />
include a term related to the strength of the effects<br />
on the elements that are in the geographic <strong>and</strong><br />
temporal neighborhood of the event.<br />
2.2 Risk definition of hazmat transportation<br />
Both the previous definitions may be also adequate<br />
to the definition of hazmat transport risk taking<br />
into account that the probability of an event <strong>and</strong> its<br />
magnitude are time/space varying, since they are<br />
606
subject to several external/internal time/space<br />
varying factors. These varying factors represent the<br />
main difference with traditional environmental <strong>and</strong><br />
industrial risk definition. In environmental <strong>and</strong><br />
industrial risk definition the probabilistic<br />
component of the hazard is often the main relevant<br />
aspect, <strong>and</strong> since the occurrence of a catastrophic<br />
event can be hardly controlled. In hazmat transport<br />
risk definition the probabilistic evaluation of the<br />
hazard of a road tract is also important, but here<br />
the risk can be controlled in real-time, for example,<br />
by keeping hazmat trucks away from that road<br />
tract. In addition also the impact of the hazard,<br />
which is usually computed as a worst/mean<br />
scenario evaluation in environmental <strong>and</strong> industrial<br />
risks, in hazmat transport risk evaluation should be<br />
more properly monitored <strong>and</strong> assessed in real-time.<br />
In general, the transport routes can be taken into<br />
account as risk sources represented by segments<br />
that, in turn, are obviously made of an infinite<br />
number of points that are also a source of risk.<br />
There are several ways of quantifying hazmat<br />
transport risk, as shown by Erkut <strong>and</strong> Verter,<br />
[1998]. An accurate definition should include at<br />
least the characteristics of the following interacting<br />
factors: the transport network, the vehicles <strong>and</strong> the<br />
territory.<br />
A frequent approach in literature for hazmat<br />
transport risk analysis is based on three separate<br />
stages:<br />
1. to determine the probability of an accident<br />
involving the hazmat release;<br />
2. to estimate the level of potential exposure, given<br />
the nature of the event;<br />
3. to estimate the magnitude of the consequences<br />
(fatalities, injuries <strong>and</strong> property damage) given the<br />
level of exposure.<br />
In these stages, probability <strong>and</strong> conditional<br />
distributions are computed. In practice, due to lack<br />
of information, the three stages process is not<br />
completely developed, <strong>and</strong> a worst-case approach,<br />
taking into account the potentially impacted<br />
population, is often used (Zhang et al. [2000]).<br />
Therefore, the expected consequence risk<br />
associated with a road link l, is often expressed as<br />
(Zhang et al. [2000]):<br />
R = S P N<br />
(1)<br />
l<br />
l<br />
l<br />
l<br />
where, R l is the total risk from hazmat movement<br />
on link l, S l the number of shipments on link l, P l<br />
the probability of a release accident for a single<br />
shipment on link l <strong>and</strong> N l the total number of<br />
persons who will be affected by a release accident<br />
on link l. The rarity of hazmat accidents makes it<br />
very difficult to calculate empirical hazmat<br />
accident probabilities for each link; general truck<br />
accident rates are sometimes used to estimate these<br />
probabilities.<br />
For each relevant point of the route, a more<br />
detailed analysis can be performed to define the<br />
magnitude of the risk evaluating further aspects,<br />
such as for example, the area involved by the worst<br />
case of accident, the behavior of the emitted plume<br />
modeling it for instance as a Gaussian plume<br />
model.<br />
Several important works address these aspects.<br />
Among others Leonelli et al. [1999], two original<br />
detailed procedures for the evaluation of individual<br />
<strong>and</strong> societal risk, have been introduced, which can<br />
take into account, integrating them in the same<br />
approach, different transportation modes,<br />
hazardous materials, meteorological conditions <strong>and</strong><br />
seasonal situations, a non uniform wind probability<br />
density distribution <strong>and</strong> an accurate description of<br />
the indoor <strong>and</strong> outdoor population both on-route<br />
<strong>and</strong> off-route.<br />
Among the Geographic Information System (GIS)<br />
based approaches, the work by Zhang et al. [2000]<br />
adopts map algebra techniques to combine airborne<br />
contaminant concentrations mathematically with<br />
the population distribution to estimate risk, for a<br />
release at any point on a network, for all parts of<br />
the study area.<br />
2.2 The role of new emerging technologies in<br />
the risk definition of hazmat<br />
transportation<br />
There is no doubt that a relevant role in the risk<br />
definition of hazmat transportation is carried out by<br />
new emerging technologies. Several specific<br />
technologies are both ready <strong>and</strong> often already<br />
installed on board to monitor the state of the<br />
hazmat <strong>and</strong> of the overall travel <strong>and</strong> to record it in<br />
a sort of on-board “black box”. The most relevant<br />
innovation in this respect is the possibility to know<br />
in real-time the exact truck position, <strong>and</strong> to transfer<br />
this information jointly with the real-time “black<br />
box” records at low price wherever it may be<br />
requested.<br />
In this respect, wireless technologies, specifically,<br />
Global System for Mobile communication (GSM)<br />
<strong>and</strong> General Packet Radio Services (GPRS),<br />
coupled with Global Positioning Systems (GPS)<br />
represent well established <strong>and</strong> emerging<br />
technologies adopted to track in real-time a float of<br />
trucks. For example, the SIMAGE project by Di<br />
Mauro et al. [2002], which aims to develop a pilot<br />
system in some Italian district areas for the<br />
monitoring <strong>and</strong> control of the transport of<br />
dangerous substances mainly via road, is partially<br />
based on this technical solution.<br />
It is quite evident however that GPRS <strong>and</strong> GPS are<br />
not the solution of their own, but they should be<br />
inserted in adequate information system, including<br />
GIS (Contini et al. [2000]) <strong>and</strong> optimization tools<br />
(Beroggi <strong>and</strong> Wallace [1998]). These aspects are<br />
briefly discussed in subsections 2.4 <strong>and</strong> 2.5.<br />
607
2.3 A proposal of dynamic risk definition of<br />
hazmat transportation<br />
The real-time monitoring of the actual routing of<br />
vehicles transporting hazmat, as well as of the state<br />
of the factors that can affect the magnitude of the<br />
hazard, allow to introduce a dynamic definition of<br />
risk due to hazmat transport.<br />
In fact, an exhaustive definition of risk in this field<br />
should include both static <strong>and</strong> dynamic information<br />
affecting the definition of the current hazard of a<br />
transport. For instance, static information should<br />
include the basic features of the considered road<br />
segment (such as slope <strong>and</strong> turning) <strong>and</strong> territorial<br />
information (such as the proximity of a river basin,<br />
of urbanized areas, <strong>and</strong> so on). Instead, dynamic<br />
information should include traffic flow conditions<br />
(e.g., given by highway authorities), forecast or<br />
observed meteorological information, the current<br />
physical/chemical state (temperature, pressure<br />
etc...) of the transported hazmat, the current<br />
representation of impact <strong>and</strong> exposure <strong>and</strong> so on.<br />
The proposed approach would aim to evaluate all<br />
the four main elements, which the literature<br />
indicates as the most important in the evaluation of<br />
the likelihood of the hazard, which are: the driver,<br />
the truck, the hazmat, the road. All four of them<br />
have static <strong>and</strong> dynamic components the most<br />
important of which are resumed in table 1. In<br />
addition, to compute hazmat risk, a dynamic<br />
representation of impacts would be also necessary.<br />
One of the main objectives of our work is to take<br />
into account all of these important static <strong>and</strong><br />
dynamic components with the objective to<br />
minimize the risk when hazmat transport can be<br />
scheduled in real time.<br />
Table 1. The main factors affecting the hazard<br />
definition in hazmat transportation<br />
Factors<br />
influencing<br />
risk<br />
Evaluating<br />
static<br />
components<br />
Evaluating realtime/<br />
dynamics<br />
components<br />
Driver Training, Drive- Physiologic<br />
Test, Medical- conditions<br />
Test<br />
Truck Periodic checks Speed, wheel<br />
Hazmat Conditions at<br />
start of route<br />
Rout<br />
segment<br />
Type of road<br />
(municipal,<br />
highway etc...),<br />
characteristics<br />
(turning,<br />
slope, tunnels,<br />
bridges...)..<br />
conditions ...<br />
Chemical <strong>and</strong><br />
physical<br />
conditions<br />
traffic flow,<br />
speed, meteo<br />
conditions, roadbed<br />
conditions,<br />
men-at-work <strong>and</strong><br />
maintenance ...<br />
While the continuous evaluation of static<br />
components has been the goal of major hazmat<br />
logistic companies for many years, the availability<br />
of emerging technologies mentioned in the<br />
previous section represents an important challenge<br />
to improve the decrease of the hazard by the<br />
monitoring of the dynamic components.<br />
In this respect, a new definition of hazmat transport<br />
risk on road is required. In the work by Fabiano et<br />
al. [2002], a great emphasis is given to the<br />
evaluation of the expected frequency of an<br />
accident, which is at the basis of the computation<br />
of the hazard.<br />
Specifically, on a given road segment i of length L i<br />
on which n i vehicles (for example each year) are<br />
passing, the frequency f i of an accident can be<br />
computed by the following:<br />
f<br />
i<br />
= γ L n<br />
(2)<br />
i<br />
i<br />
6<br />
i<br />
∏<br />
γ i = γ 0 h i,<br />
j<br />
(3)<br />
j=<br />
1<br />
where γ i is the expected frequency of an accident,<br />
which can be computed on the basis of the basic<br />
frequency γ 0 (accident km −1 per vehicle) <strong>and</strong> h i,j<br />
are local amplifying/mitigating parameters.<br />
Specifically, for each road segment i, six<br />
parameters h i,j are proposed: four parameters<br />
related to intrinsic road characteristics (turns,<br />
slope, number of lanes, bridge/tunnel), meteo <strong>and</strong><br />
vehicle flow.<br />
Meteo <strong>and</strong> vehicle flow are two parameters that<br />
have even more relevance to estimate the<br />
likelihood of the hazard if they can be both<br />
monitored in real-time <strong>and</strong> predicted in their<br />
evolution. Other two parameters, which play a<br />
similar important role, are velocity <strong>and</strong> availability<br />
of service of the road segment (for example menat-work,<br />
that limit the service of the road segment,<br />
for example from two lanes to just one lane). So, in<br />
our approach, eight dynamics parameters affect the<br />
expected probability of an accident on a road<br />
segment.<br />
In addition the approach is limited to the analysis<br />
to the transport of petroleum products <strong>and</strong> to two<br />
scenarios: explosion <strong>and</strong> release. In the first case, it<br />
is important to evaluate the magnitude of the event<br />
with respect to people that can be involved in the<br />
explosion (mainly represented by the driver<br />
himself, the other drivers on the road <strong>and</strong> the<br />
people living in the neighborhood), while in the<br />
other case it is important to evaluate the possible<br />
damage on the infrastructure (for example loss of<br />
service of the road) <strong>and</strong> of the environment (for<br />
example, pollution of a water basin). In addition, a<br />
dynamic representation of these impacts would be<br />
also necessary, but, as a simplifying assumption<br />
either worst case or probabilistic models of the<br />
impact may be taken into account.<br />
608
2.4 A comprehensive decision support system<br />
architecture to support real time risk<br />
assessment <strong>and</strong> real time routing of<br />
hazardous material transport on road<br />
Having accepted that a dynamic risk definition of<br />
hazmat transportation is needed, another aspect<br />
should be put in evidence. The dynamics of this<br />
risk can be controlled in real time. To establish the<br />
elements supporting this idea, some considerations<br />
about how hazmat routing is commonly planned<br />
<strong>and</strong> controlled are introduced hereinafter.<br />
Static<br />
Planning <strong>and</strong><br />
characteristics<br />
Optimization of<br />
of routing Daily<br />
Orders Hazmat Routing<br />
routing plan<br />
Figure 1. Simplified representation of a<br />
planning system for hazmat routing definition.<br />
Figure 1 shows a simplified representation of a<br />
planning system, which supports the computation<br />
of the routing that should be performed by tank<br />
trucks in order to satisfy orders from customers.<br />
Specifically, referring to the particular case study<br />
treated within this work, in the case of distribution<br />
of petroleum products to the fuel stations<br />
distributed on a territory, each day a routing plan is<br />
defined according to the set of orders that have<br />
been received, to the characteristics of the road<br />
network, <strong>and</strong> with reference to the location of the<br />
dem<strong>and</strong>. Usually this plan is supported by<br />
optimization modules that are able to compute an<br />
optimal solution of a vehicle routing problem<br />
(Christofides et al. [1979]), achieving objectives<br />
such as for example the minimization of the overall<br />
time of delivery.<br />
3. PRELIMINARY RESULTS<br />
The aim of this research is to design <strong>and</strong> to<br />
implement a DSS for real time risk assessment of<br />
hazardous material transport on road. The project<br />
is currently evolving by steps. At the moment the<br />
modules shown in figure 3 have been implemented<br />
providing the possibility to track the transport<br />
within a GIS utility that can be accessed by a Web<br />
interface. In addition, preliminary computations of<br />
the real-time risk have been performed on<br />
historical data, since at the moment the DSS is not<br />
linked in real-time with meteo/traffic information.<br />
SIT<br />
Maps<br />
Planner / Users<br />
WEB/GIS Interface<br />
Evaluation of<br />
hazard <strong>and</strong> risk<br />
Real-time data from tank<br />
trucks<br />
Real-Time<br />
Database<br />
Meteo/Traffic Real-Time<br />
data<br />
Optimization<br />
planning<br />
module<br />
Figure 3. The preliminary modules of the DSS<br />
which have been implemented.<br />
Static<br />
characteristics of<br />
routing<br />
L<strong>and</strong> use<br />
information<br />
Meteo/Traffic<br />
Real-Time State<br />
Other Unknown<br />
Hazmat Traffic<br />
Orders<br />
Control <strong>and</strong><br />
Optimization of<br />
Hazmat Routing<br />
Real-time Hazmat<br />
Transport<br />
Monitoring<br />
Risk Assessment<br />
<strong>and</strong> Prediction of<br />
Hazmat Transport<br />
risk<br />
Figure 2. A modern view of risk based<br />
planning for hazmat routing definition.<br />
Minimizing risk requires different approaches that<br />
should include the minimization of risk (Zografos<br />
<strong>and</strong> Androutsopoulos [2004]), the equity in the<br />
distribution of risk in the territory (Current <strong>and</strong><br />
Ratick [1995]), <strong>and</strong> a real-time routing<br />
management (Beroggi [1994]). These aspects are<br />
the one on which we are currently investigating in<br />
order to make evolve the functionality of the<br />
planning system in figure 1, to the one shown in<br />
figure 2, which includes real-time decisional<br />
aspects.<br />
Figure 4. The WEB/GIS based interface of the<br />
DSS.<br />
An example of graphic user interface developed for<br />
the DSS is shown in figure 4, where the results of<br />
the computation of the risk are shown for some<br />
road segments over a particular area, namely the<br />
western districts of Liguria region (Italy) which are<br />
609
heavily interested by hazmat traffic, <strong>and</strong><br />
characterized by transport infrastructures generally<br />
quite close to civil <strong>and</strong> industrial settlements.<br />
4. CONCLUSIONS AND FUTURE<br />
DEVELOPMENTS<br />
The overall problem of hazmat risk transport is a<br />
complex interdisciplinary problem, that should be<br />
faced under many viewpoints one of which is the<br />
optimal risk-based planning of hazmat routing.<br />
Emerging telematics technologies <strong>and</strong> related new<br />
monitoring systems allow to improve the definition<br />
of risk, showing its real-time features <strong>and</strong> allowing<br />
to model its evolution. The proposed DSS aims to<br />
compute dynamic hazmat risk in real-time <strong>and</strong> it is<br />
based on a methodology starting from traditional<br />
static risk evaluations: dynamics components are<br />
introduced as a factor amplifying or reducing the<br />
accident expected frequency. In a comprehensive<br />
DSS, similar considerations should be taken into<br />
account in the evaluation of the magnitude, that is<br />
on the impact on vulnerable territorial elements.<br />
Once the DSS is verified on a greater set of<br />
historical <strong>and</strong> real-time information, it will be<br />
extended to be linked with or to enhance current<br />
route planning systems of hazmat transport.<br />
Future developments are both technological <strong>and</strong><br />
methodological. A technological aspect is related<br />
to the enhancement of the production of distributed<br />
information by the fleet of trucks adding sensors to<br />
monitor meteorological conditions (e.g.<br />
temperature <strong>and</strong> luminescence), information added<br />
by the driver (e.g. fog, accidents on the road), as<br />
well as information on the real-time health status of<br />
the driver himself. Methodological aspects will<br />
deal with the calibration of the model on a set of<br />
historical data <strong>and</strong> on the practical experience of<br />
drivers, <strong>and</strong> with the integration with route<br />
planning modules.<br />
5. ACKNOWLEDGEMENTS<br />
This work has been sponsored by INTERMODE<br />
S.p.A. with reference to a collaboration between<br />
DIST <strong>and</strong> INTERMODE S.p.A. INTERMODE<br />
S.p.A., ENI group, is a company that is in charge<br />
of the distribution of petroleum products to 8000<br />
Italian oil stations, with 2000 tank trucks per day<br />
moving more than 15000kton petroleum products<br />
per year on Italian roads.<br />
6. REFERENCES<br />
Beroggi G.E.G., A real-time routing model for<br />
hazardous materials, EJOR 75 (3), 508-520,<br />
1994.<br />
Beroggi G.E.G <strong>and</strong> W.A. Wallace, Routing of<br />
hazardous materials. In “Operational Risk<br />
Management”, Kluwer Academic Publishers,<br />
1998.<br />
Christofides N., A. Mingozzi, <strong>and</strong> P. Toth, The<br />
vehicle routing problem. In N. Christofides, A.<br />
Mingozzi, P. Toth, <strong>and</strong> C. S<strong>and</strong>i, editors,<br />
Combinatorial Optimization, Wiley,<br />
Chichester, UK, 315-338, 1979.<br />
Contini S., F. Bellezza, M. D. Christou <strong>and</strong> C.<br />
Kirchsteiger, The use of geographic<br />
information systems in major accident risk<br />
assessment <strong>and</strong> management, J. Hazardous<br />
Materials 78, 223-245, 2000.<br />
Current J. <strong>and</strong> S. Ratick, A model to assess risk,<br />
equity <strong>and</strong> efficiency in facility location <strong>and</strong><br />
transportation of hazardous materials, Location<br />
Science 3 (3), 187-201, 1995.<br />
Di Mauro C., J.P. Nordvik, <strong>and</strong> A.C. Lucia, Multicriteria<br />
decision support system <strong>and</strong> Data<br />
Warehouse for designing <strong>and</strong> monitoring<br />
sustainable industrial strategies - an Italian case<br />
study, IEMSS, vol.1, 216-220, Lugano, 2002.<br />
Erkut, E. <strong>and</strong> V. Verter, V., Modeling of transport<br />
risk for hazardous materials. Operations<br />
Research 46 (5), 625-642, 1998.<br />
Fabiano B., F. Currò, E. Palazzi <strong>and</strong> R. Pastorino,<br />
A framework for risk assessment <strong>and</strong> decisionmaking<br />
strategies in dangerous good<br />
transportation, Journal of Hazardous Materials<br />
93, 1–15, 2002.<br />
FMCSA, Comparative Risks of Hazardous<br />
Materials <strong>and</strong> Non-Hazardous Materials Truck<br />
Shipment Accidents/Incidents, Washington,<br />
DC, available at http://www.fmcsa.dot.gov,<br />
2001.<br />
Leonelli P., S.Bonvicini <strong>and</strong> G. Spadoni, New<br />
detailed numerical procedures for calculating<br />
risk measures in hazardous materials<br />
transportation . J. Loss Prev. Process Ind. 12,<br />
507-515, 1999.<br />
Rizzoli, A.E. <strong>and</strong> W.J. Young, Delivering<br />
environmental decision support systems:<br />
software tools <strong>and</strong> techniques, <strong>Environmental</strong><br />
<strong>Modelling</strong> <strong>and</strong> <strong>Software</strong> 12, n.2-3, 237-249,<br />
1997.<br />
UNESCO, Report of consultative meeting of<br />
experts on the statistical study of natural hazard<br />
<strong>and</strong> their consequences, Document<br />
SC/WS/500, 1972.<br />
U.S. DOT, Risk Management Framework for<br />
Hazardous Materials Transportation, available<br />
at http://hazmat.dot.gov, 2000.<br />
Zhang, J. Hodgson <strong>and</strong> E. Erkut, Using GIS to<br />
assess the risks of hazardous materials transport<br />
in networks, EJOR 121, 316-329, 2000.<br />
Zografos K.G. <strong>and</strong> K. N. Androutsopoulos, A<br />
heuristic algorithm for solving hazardous<br />
materials distribution problems, EJOR 152,<br />
507–519, 2004.<br />
610
Appropriate <strong>Modelling</strong> in DSSs for River Basin<br />
Management<br />
Yueping Xu <strong>and</strong> Martijn J. Booij<br />
Water Engineering <strong>and</strong> Management, Faculty of Engineering, University of Twente, Enschede, the Netherl<strong>and</strong>s<br />
(email: y.p.xu@ctw.utwente.nl; m.j.booij@ctw.utwente.nl)<br />
Abstract: There is increasing interest in the development of decision support systems (DSSs) for river basin<br />
management. Moreover, new ideas <strong>and</strong> techniques such as sustainability, adaptive management, Geographic<br />
Information System, Remote Sensing <strong>and</strong> participations of new stakeholders have stimulated their development. A<br />
DSS often encompasses a number of sub-models, such as models for flood risk, ecology, tourism, recreation <strong>and</strong><br />
navigation. These models are fundamental in supporting the whole decision-making process. However, often<br />
complicated <strong>and</strong> sophisticated models are used which are difficult to underst<strong>and</strong> <strong>and</strong> operate for decision-makers.<br />
Moreover, these models may be not necessary for some specific-purpose DSSs, such as those for preliminary<br />
planning purposes. The aim of this paper is therefore to find appropriate models by applying a proposed<br />
appropriateness framework. An appropriate system is defined as ‘a system which can produce outputs enabling<br />
decision makers to distinguish different river management actions under uncertainty according to the current<br />
problem’. The proposed framework is applied to a sub-model of a DSS — a flood risk model to illustrate the idea of<br />
appropriateness. The results show that the framework proposed is applicable. It helps distinguish the management<br />
actions <strong>and</strong> find the appropriate models for the DSSs.<br />
Keywords: decision support system; flood risk model; appropriate modelling; Latin Hypercube Simulation; Morris’<br />
method<br />
with respect to accuracy.<br />
1. INTRODUCTION<br />
There is increasing interest in the development of<br />
decision support systems (DSSs) for river basin<br />
management. Moreover new ideas <strong>and</strong><br />
techniques like sustainability, adaptive<br />
management, Geographic Information System<br />
(GIS), Remote Sensing (RS) <strong>and</strong> participations<br />
of new stakeholders have stimulated their<br />
development [Smits et al. 2000]. A DSS for river<br />
basin management often encompasses a number<br />
of sub-models, such as models for flood risk,<br />
ecology, tourism, recreation <strong>and</strong> navigation.<br />
These models are fundamental in supporting the<br />
whole decision-making process. However, often<br />
complicated <strong>and</strong> sophisticated models are used<br />
which are difficult to underst<strong>and</strong> <strong>and</strong> operate for<br />
decision-makers. Moreover, these models may<br />
be not necessary for some specific-purpose<br />
DSSs, such as those for preliminary planning<br />
purposes. In case of data insufficiency, simple<br />
models could be preferable if they can satisfy the<br />
requirements from the decision makers, e.g.,<br />
In the field of river basin management,<br />
uncertainty studies have been an essential part to<br />
support the decision making. In case a ranking<br />
of the river management actions based on<br />
particular decision variables is required,<br />
uncertainty will be one of the main obstacles. In<br />
order to make a sound decision, uncertainty<br />
reduction is often the first solution the analysts<br />
can provide.<br />
An appropriateness framework is proposed in<br />
this paper. An appropriate system is defined as ‘a<br />
system which can produce outputs enabling<br />
decision makers to distinguish different river<br />
management actions under uncertainty according<br />
to the current problem’. The framework employs<br />
uncertainty analysis to analyze the<br />
appropriateness of models used in the DSSs. As<br />
an example, a sub-model of a DSS — a flood<br />
risk model will be used to illustrate the use of the<br />
proposed approach.<br />
611
2. APPROPRIATENESS<br />
FRAMEWORK<br />
Figure 1 shows the general appropriateness<br />
framework proposed in this paper. This<br />
framework is used to find appropriate models in<br />
the DSSs with an aim to distinguish (rank) the<br />
river management actions. According to this<br />
figure, there are three important aspects (after<br />
inputs <strong>and</strong> quantitative modelling) involved in<br />
this framework. They are uncertainty analysis,<br />
appropriateness analysis <strong>and</strong> model<br />
improvements through uncertainty reduction<br />
respectively.<br />
Amplitude Sensitivity Test (FAST), <strong>and</strong><br />
Response Surface Methods [Morgan <strong>and</strong><br />
Henrion 1990]. They can be used to study how<br />
the uncertainty in the inputs <strong>and</strong> parameters are<br />
propagated into the model outputs (decision<br />
variables in the DSSs). Here one of the Monte<br />
Carlo Simulation methods, namely Latin<br />
Hypercube Simulation (LHS) method, will be<br />
used.<br />
2.2 Appropriateness analysis<br />
As introduced in Section 1, the appropriateness<br />
is defined under the concept of decision making<br />
under uncertainty. The appropriateness is<br />
quantified by a criterion, defined as the risk of<br />
making a wrong decision (R). The risk is the<br />
product of the mean difference (D) of the model<br />
outputs resulting from each combination of<br />
management actions <strong>and</strong> the probability of<br />
making a wrong decision (P) for each<br />
combination of management actions. This<br />
criterion can be used to determine whether the<br />
models in the DSSs are appropriate or not after<br />
uncertainty analysis. The mathematical equation<br />
of the risk is<br />
R = D * P<br />
(1)<br />
Figure 1: An appropriateness framework (R is<br />
the calculated risk <strong>and</strong> R* is the acceptable risk)<br />
2.1 Uncertainty analysis<br />
From a modeler’s point of view, there are three<br />
types of uncertainty: uncertainty in model<br />
quantities, uncertainty about model form <strong>and</strong><br />
uncertainty about the completeness/adequacy of<br />
the model [Van Asselt 2000]. In this paper, only<br />
the uncertainties in model quantities are<br />
considered. Uncertain model quantities include<br />
model inputs <strong>and</strong> parameters. The uncertainty<br />
caused by the model form <strong>and</strong> model<br />
completeness has not been studied although it is<br />
known to be important [Cardwell <strong>and</strong> Ellis 1996;<br />
Perrin et al. 2001].<br />
To investigate the effects of uncertainty on the<br />
decision variables, many uncertainty analysis<br />
methods are available, for example the first order<br />
method, Monte Carlo Simulation, Fourier<br />
Here the probability of making a wrong decision<br />
(P) is the probability that one measure<br />
outperforms another measure based on particular<br />
decision variables. According to the definition,<br />
there is one risk value for each of the k (k-1)/2<br />
combinations of management actions. k is the<br />
number of management actions. So R can be<br />
regarded as a set of risk value.<br />
Assume that the decision makers’ acceptable risk<br />
is R * , then the models are determined to be<br />
appropriate if all members of the risk set R are<br />
smaller than R * , that is<br />
R < R<br />
*<br />
(2)<br />
for all combinations of management actions.<br />
Else, the models are determined to be<br />
inappropriate.<br />
2.3 Model improvements through<br />
uncertainty reduction<br />
If the models are determined as inappropriate,<br />
they need to be improved in order to reduce the<br />
risk by reducing the uncertainty in the model<br />
612
outputs. There are several techniques available<br />
for reducing the uncertainty, for example by<br />
obtaining more measurement data. In this paper,<br />
reducing uncertainty in the model outputs is<br />
completed by reducing uncertainty in the inputs<br />
<strong>and</strong> parameters in the models, as indicated in<br />
Figure 1.<br />
In order to reduce the uncertainty, a screening<br />
sensitivity analysis method, named the Morris’<br />
method [Morris 1991] will be used. This method<br />
is used to investigate the importance of all inputs<br />
<strong>and</strong> parameters in the models. The most<br />
important inputs <strong>and</strong> parameters will be<br />
identified by the Morris’ method <strong>and</strong> uncertainty<br />
will be reduced in those quantities. The most<br />
important inputs <strong>and</strong> parameters are those that<br />
contribute most to the uncertainty in the final<br />
model outputs. In this way, the most efficient<br />
reduction of uncertainty in the model outputs can<br />
be achieved.<br />
The models will be improved until the<br />
uncertainty in the model outputs is tolerable to<br />
the decision makers according to the acceptable<br />
risk. Alternatively the efforts (costs <strong>and</strong> time) to<br />
reduce the uncertainty are not worthwhile<br />
compared to the amount of uncertainty reduced<br />
or it is impossible to reduce the uncertainty<br />
because of the nature of the uncertainty.<br />
3. CASE STUDY<br />
A sub-model of a developed DSS for the Dutch<br />
Meuse River — a flood risk model — is used to<br />
apply the appropriateness framework introduced<br />
in Section 2. This sub-model calculates the net<br />
present value (NPV) for different river<br />
management actions. The NPV is used as a<br />
decision variable to determine the<br />
appropriateness of models in the DSS.<br />
There are several components in this flood risk<br />
model, namely a flood frequency model, a<br />
hydraulic model, an inundation model, <strong>and</strong> a risk<br />
model.<br />
The primary objective of the flood frequency<br />
model is to relate the magnitude of extreme<br />
events (flood flows) to their frequency of<br />
occurrence through the use of probability<br />
distributions. In this analysis, the Gumbel<br />
Extreme Value distribution is used.<br />
The hydraulic model calculates water levels in<br />
the river channel for different flood flows.<br />
Stepwise steady non-uniform flow simulation is<br />
used for this purpose [Van Rijn 1994]. Assume<br />
there are no lateral flows.<br />
The inundation model is employed to calculate<br />
the inundation depths in the flood plains. The<br />
inundation depths are the differences between<br />
water levels <strong>and</strong> l<strong>and</strong> heights.<br />
The objective of the risk model is to calculate the<br />
NPV value for each management action. The net<br />
present value (NPV) is defined as the sum of<br />
expected annual damage [Shaw 1994], costs of<br />
management actions, <strong>and</strong> benefits from s<strong>and</strong> <strong>and</strong><br />
gravel extractions [Van Leussen et al. 2000].<br />
Here only the direct damage is considered (for<br />
example no damage to the ecological value) [De<br />
Blois 1996]. For floods of different probabilities,<br />
corresponding value of flood damage can be<br />
calculated. The economic damage in the<br />
floodplains is determined by the inundation<br />
depth, l<strong>and</strong> use type <strong>and</strong> the number of units of<br />
that l<strong>and</strong> use type. The damage is given in<br />
monetary values per unit (in euros). The<br />
expected annual damage is the expected annual<br />
value of these damages.<br />
Three management actions are formulated in this<br />
paper to investigate how they affect the NPV<br />
value. They are:<br />
• The base situation (M 1 ).<br />
• Deepening the summer bed by 1 meter<br />
(M 2 ).<br />
• Spatial planning, for example relocation<br />
of valuable capital from the floodplains<br />
to higher l<strong>and</strong> (M 3 ).<br />
4. RESULTS<br />
4.1 Uncertainty analysis: the NPV value<br />
As stated before, only uncertainty in the inputs<br />
<strong>and</strong> parameters will be considered. In this case<br />
study, there are a total of 112 inputs <strong>and</strong><br />
parameters in the models. A sample size of 100<br />
will be selected in LHS simulation.<br />
The two parameters in the flood frequency model<br />
are assumed to be normally distributed. For the<br />
hydraulic parameters, a questionnaire has been<br />
employed to investigate how uncertain these<br />
parameters are. The distributions of all the other<br />
613
inputs <strong>and</strong> parameters are arbitrarily set uniform<br />
in shape, because there are insufficient data<br />
available to infer any particular type of<br />
distribution for these inputs <strong>and</strong> parameters.<br />
Ranges of variability have been selected either<br />
according to the information available, or in<br />
absence of such information, assuming 20% of<br />
uncertainty is involved in the inputs <strong>and</strong><br />
parameters (nominal value ± 20%<br />
).<br />
The uncertainties in the inputs <strong>and</strong> parameters<br />
are propagated into the model outputs, here NPV<br />
in million euros. The fitted normal distributions<br />
for three management actions are shown in<br />
Figure 2 (x- axis<br />
exist among the model outputs, which make it<br />
difficult to rank these three management actions.<br />
4.2 Appropriateness analysis: risk calculation<br />
Table 1, Table 2 <strong>and</strong> Table 3 present the mean<br />
differences, the probabilities of making a wrong<br />
decision <strong>and</strong> the risks of making a wrong<br />
decision (‘Case 0’, bold numbers in three tables)<br />
for each combination of management actions.<br />
Table 1 <strong>and</strong> Table 2 show that, as expected,<br />
small mean differences correspond to large<br />
probabilities of making a wrong decision. This<br />
means the mean differences <strong>and</strong> the probabilities<br />
have counteracting effects on each other. The<br />
risks are actually combined effects of both<br />
aspects.<br />
Commonly the acceptable risk is determined by<br />
decision makers. However, in this case study a<br />
value of six million euros is chosen for a<br />
preliminary analysis. The appropriateness of the<br />
models is judged based on this acceptable risk.<br />
The bold numbers in Table 3 indicate that the<br />
models used in this case are inappropriate<br />
because one of the risks calculated (6.60 million<br />
euros) is higher than the acceptable risk.<br />
Figure 2: Fitted normal distributions for model<br />
outputs from three management actions<br />
is the natural logarithm (LN) of the NPV value).<br />
This figure shows that large areas of overlap<br />
4.4 Uncertainty reduction: model<br />
improvements<br />
As described in Section 4.3, the models are<br />
judged as inappropriate because of the failure of<br />
satisfying the acceptable risk defined.<br />
Table 1: The mean differences (million euros)<br />
Management actions Case 0 Case 1 Case 2<br />
compared<br />
M 2 & M 1 5.61 6.88 6.93<br />
M 1 & M 3 23.27 22.31 22.83<br />
M 2 & M 3 28.88 29.19 29.76<br />
Table 2: The probabilities of making a wrong decision<br />
Management actions<br />
Case 0 Case 1 Case 2<br />
compared<br />
M 2 & M 1 0.42 0.41 0.39<br />
M 1 & M 3 0.28 0.24 0.22<br />
M 2 & M 3 0.20 0.17 0.12<br />
614
Table 3: The risks of making a wrong decision (million euros)<br />
Management actions<br />
Case 0 Case 1 Case 2<br />
compared<br />
M 2 & M 1 2.38 2.85 2.70<br />
M 1 & M 3 6.60 5.38 5.10<br />
M 2 & M 3 5.69 4.97 3.58<br />
The Morris’ method identified that the most<br />
important inputs <strong>and</strong> parameters in the flood risk<br />
model are river slope, bed level coefficients,<br />
depths of the summer bed <strong>and</strong> Nikuradse<br />
coefficients in the flood plains. They are all<br />
parameters in the hydraulic model. The Morris’<br />
method also concluded that all the parameters in<br />
the hydraulic model appear to be more important<br />
than the parameters in the flood frequency model<br />
<strong>and</strong> the parameters in the damage functions of<br />
the risk model. These parameters contribute<br />
more to the uncertainty in the model outputs than<br />
the others. Therefore the idea is to try to reduce<br />
the uncertainty in the parameters from the<br />
hydraulic model.<br />
In this paper, the modelers are not interested in<br />
how the uncertainties are reduced although it is<br />
important. To investigate how the uncertainty<br />
reduction in the most important inputs <strong>and</strong><br />
parameters affects the risks, two cases are<br />
considered based on different assumptions (for<br />
illustration only):<br />
• Case 1: assume a reduction of<br />
uncertainty in river slope, bed level<br />
coefficients, depths of the summer bed<br />
<strong>and</strong> Nikuradse coefficients in the flood<br />
plains<br />
• Case 2: assume deterministic<br />
parameters in the hydraulic model<br />
In order to study the effects of uncertainty<br />
reduction, the original system without<br />
improvement is represented here as ‘Case 0’.<br />
The calculated mean differences, the<br />
probabilities of making a wrong decision <strong>and</strong> the<br />
risks of making a wrong decision after<br />
uncertainty reduction are again shown in Table<br />
1, Table 2 <strong>and</strong> Table 3 respectively.<br />
Most of the mean differences in Table 1 show an<br />
increase of value except the combination for M 1<br />
<strong>and</strong> M 3 . For this combination, the mean<br />
difference first decreases <strong>and</strong> then increases. The<br />
increase of the mean difference shows an<br />
indication of more easily distinguishing the<br />
management actions. The unstable change of the<br />
mean differences maybe a result of the nonlinearity<br />
of the models <strong>and</strong> insufficient<br />
simulation runs (r<strong>and</strong>om). The effects of nonlinearity<br />
<strong>and</strong> simulation runs have not been<br />
investigated in this paper. The probabilities of<br />
making a wrong decision presented in Table 2<br />
show a decrease of value because of the<br />
reduction of uncertainties, in turn, helping reduce<br />
the value of risks calculated.<br />
For both cases, the risks calculated are smaller<br />
than the predefined acceptable risk of six million<br />
euros. Based on this, it is concluded that, under<br />
both cases the models used in the DSS are<br />
appropriate.<br />
5. CONCLUSIONS<br />
In the case study presented in this paper, the high<br />
uncertainty in the model outputs produced<br />
indistinguishable situations for some<br />
combinations of management actions. This is<br />
often the case for DSSs in general. The models<br />
were determined to be inappropriate by<br />
comparing the value of risk of making a wrong<br />
decision for each combination of management<br />
actions with the acceptable risk. After<br />
improving the models by reducing the<br />
uncertainty in the most important inputs <strong>and</strong><br />
parameters, the models became appropriate. The<br />
analysis in this section gives a good idea of how<br />
the proposed appropriate framework worked in<br />
this case study.<br />
A key point in this paper is the definition of the<br />
criterion that is used to determine the<br />
appropriateness of models used in the DSS. This<br />
criterion, defined as the risk of making a wrong<br />
decision for each combination of management<br />
actions, combines two interesting aspects. These<br />
aspects are the mean difference for each<br />
combination of management actions <strong>and</strong> the<br />
probability of making a wrong decision. They<br />
are both important for the risks of making a<br />
615
wrong decision <strong>and</strong> have counteracting effects<br />
on each other. The criterion is proved to be a<br />
reasonable one for analyzing the appropriateness<br />
of models used in this DSS.<br />
Due to the non-linearity of the models <strong>and</strong> the<br />
r<strong>and</strong>om of the simulation, one of the mean<br />
differences showed an unstable change when the<br />
uncertainty in inputs <strong>and</strong> parameters was reduced.<br />
This can be partly solved by increasing the runs<br />
of the LHS simulations or by calculating the<br />
confidence intervals of the risks. Else this<br />
situation could be an obstacle in finding the<br />
appropriate models <strong>and</strong> results in more efforts<br />
necessary in reducing the uncertainty.<br />
Van Asselt, M.B.A., Perspectives on uncertainty<br />
<strong>and</strong> risk. The PRIMA approach to<br />
decision support, Kluwer Academic<br />
Publishers. the Netherl<strong>and</strong>s, 2000.<br />
Van Leussen, W., Boot, U. <strong>and</strong> Verkerk, H.,<br />
Civil engineering aspects of the “Meuse<br />
Works”: a restoration program of the<br />
River Meuse’. In: Proceedings of<br />
<strong>International</strong>es Wasserbau-Symposium.<br />
Aachen, Germany, 2000.<br />
Van Rijn, L.C., Principles of fluid flow <strong>and</strong><br />
surface waves in rivers, estuaries, seas<br />
<strong>and</strong> oceans, AQUA publications, the<br />
Netherl<strong>and</strong>s, 1994<br />
6. REFERENCES<br />
Cardwell, H. <strong>and</strong> Ellis, H., Model uncertainty<br />
<strong>and</strong> model aggregation in<br />
environmental management, Applied<br />
Mathematics <strong>Modelling</strong>, 20, 121-134,<br />
1996.<br />
De Blois, C.J. EMBRIO: evaluation of methods<br />
for the assessment of flood damage <strong>and</strong><br />
flood risk in floodplains. Phase 2: Nonmonetary<br />
damage, discounting, <strong>and</strong><br />
economic growth, University of<br />
Twente, Enschede, the Netherl<strong>and</strong>s,<br />
1996.<br />
Morgan, M.G., <strong>and</strong> Henrion, M., Uncertainty: A<br />
Guide to Dealing with Uncertainty in<br />
Quantitative Risk <strong>and</strong> Policy Analysis,<br />
Cambridge University Press,<br />
Cambridge, U.K., 1990.<br />
Morris, M.D., Factorial sampling plans for<br />
preliminary computational experiments,<br />
Technometrics, 33, 161-174, 1991.<br />
Perrin, C., Michel, C. <strong>and</strong> Andréassian, V., Does<br />
a large number of parameters enhance<br />
model performance? Comparative<br />
assessment of common catchment<br />
model structures on 429 catchments,<br />
Journal of Hydrology, 242, 275-301,<br />
2001.<br />
Shaw, E.M., Hydrology in Practice, T.J.<br />
<strong>International</strong> Ltd, Padstow, Cornwall,<br />
Great Britain, 1994.<br />
Smits, A.J.M., Nienhuis, P.H. <strong>and</strong> Leuven,<br />
R.S.E.W., New Approaches to River<br />
Management, Backhuys Publishers, the<br />
Netherl<strong>and</strong>s, 2000.<br />
616
Water Management, Public Participation <strong>and</strong> Decision<br />
Support Systems: the MULINO Approach<br />
J. Feás a , C. Giupponi a,b , P. Rosato a,c<br />
a Fondazione Eni Enrico Mattei, Venice, Italy<br />
b Università degli Studi di Milano, Italy<br />
c Università di Trieste, Italy<br />
Abstract: One of the main issues in the environmental decision making field is the necessity, sometimes<br />
obligation imposed by the legislation, to communicate the decision process <strong>and</strong> make it more<br />
comprehensible. In other words, the objective is to increase the transparency of the decision making available<br />
all the relevant information related to the decision process for all interested actors. For this reason, many<br />
tools have been developed over the last decades: indicators, conceptual frameworks, <strong>and</strong> impact assessment<br />
studies are examples. However, many of these tools try to represent the environmental situation or<br />
hypothetical future states without any explicit reference to how decisions are taken or should be taken. Some<br />
environmental decision support systems are developed for that specific purpose. One critical point in the<br />
development of such a DSS is the connection between the representation of reality <strong>and</strong> the elicitation of<br />
preferences of the decision makers. Moreover, environmental decision making requires that preferences <strong>and</strong><br />
value judgments refer to technical <strong>and</strong> scientific information that is not easy to communicate to people in<br />
general. The European project, MULINO (contract no. EVK1-2000-22089), completed at the end of 2003,<br />
has focused on connecting environmental tools <strong>and</strong> decision support methods, by combining the DPSIR<br />
approach with multicriteria analysis methods in a decision support system called mDSS. The DPSIR is a<br />
conceptual framework developed by the EEA through which environmental problems can be structured <strong>and</strong><br />
explored in a heuristic way. This process may be undertaken in a group (e.g. the decision makers <strong>and</strong> the<br />
stakeholders together) using the framework to structure discussion between those who decide <strong>and</strong> those who<br />
are involved in the problem. In this paper, we describe the MULINO approach, focusing on the experience<br />
gained with the end users involved in the project in applying the mDSS software. In particular, we present the<br />
use of the DPSIR approach to structure <strong>and</strong> communicate their decis ion context <strong>and</strong> the potentials for<br />
stakeholders’ involvement.<br />
Keywords: <strong>Environmental</strong> tools, decision support system, DPSIR, Water Framework Directive<br />
1. INTRODUCTION<br />
Since 1992 <strong>and</strong> the signing of the Rio Declaration<br />
on Environment <strong>and</strong> Development <strong>and</strong> Agenda 21,<br />
where a plan of action to achieve the sustainable<br />
development into the 21st Century was set out, the<br />
concepts of public participation <strong>and</strong> stakeholder<br />
involvement have had a growing influence on<br />
policy formation <strong>and</strong> decision making processes.<br />
There are still large knowledge gaps <strong>and</strong> culture<br />
clashes, which make the realization of<br />
participatory processes problematic for most<br />
governing bodies. The increase in the number of<br />
actors, both public <strong>and</strong> private, affirms the need for<br />
capacity building to define mediation techniques<br />
<strong>and</strong> co-operative approaches appropriate for active<br />
stakeholder involvement. At the present time<br />
however, the situation is complicated for<br />
authorities that are obliged to execute participative<br />
planning procedures.<br />
Like environmental planning in general, Integrated<br />
Water Resources Management (IWRM) is usually<br />
characterized by the involvement of numerous<br />
decision-makers operating at different levels <strong>and</strong><br />
the large number of stakeholders with conflicting<br />
preferences <strong>and</strong> different value judgments<br />
[Lahdelma et al. 2000)] This makes the<br />
development of policy implementation strategies<br />
<strong>and</strong> decision making in the context of IWRM a<br />
very complex issue, also because it requires a<br />
broader integration with other sectors such as<br />
environment, energy, industry, agriculture,<br />
tourism.<br />
Adequate methodologies <strong>and</strong> tools become<br />
therefore necessary in order to measure how a<br />
specific policy meets the objectives established by<br />
the various actors, to identify <strong>and</strong> underst<strong>and</strong> the<br />
possible conflicts that may arise between these<br />
actors <strong>and</strong>, finally, to design possible paths <strong>and</strong><br />
courses of action to arrive at a sustainable solution.<br />
The need for adequate methodologies <strong>and</strong> tools<br />
calls for a strong role to be played by science <strong>and</strong><br />
research. The commitments made by the scientific<br />
617
community of the 2002 World Summit on<br />
Sustainable Development was in fact to make<br />
science more policy relevant [ICSU, 2002].<br />
<strong>Environmental</strong> problems <strong>and</strong> in particular those<br />
related to water resources are usually very<br />
complex <strong>and</strong> therefore the decision making process<br />
requires high background in environmental,<br />
economic <strong>and</strong> social disciplines. Moreover, there is<br />
quite often a dramatic gap between those who<br />
analyse <strong>and</strong> provide disciplinary expertises <strong>and</strong><br />
those who decide, not only in the knowledge but<br />
also in the aims <strong>and</strong> the way of thinking <strong>and</strong> the<br />
language [Luiten, 1999].<br />
The European Water Framework Directive (WFD)<br />
[EC, 2000] specifically addresses public<br />
information <strong>and</strong> consultation in Article 14. It is<br />
obligatory for the Member States to involve the<br />
public in the implementation of the Directive by<br />
publishing specific information relevant to the<br />
River Basin Plans <strong>and</strong> to be open to comments<br />
made by the public about the planning process.<br />
Member States are also to encourage the active<br />
involvement of all interested parties, which would<br />
require more than the publication of information.<br />
The WFD is laying the groundwork for social<br />
sustainability by establishing public involvement<br />
in planning procedures as common practice. Even<br />
if the level of obligatory participation is the most<br />
basic, for some European countries this is a<br />
necessary first step as it may be that citizens have<br />
had no legitimate role in the management of water<br />
before.<br />
The participation of a range of stakeholders in the<br />
planning process might take on a number of forms,<br />
including public forums, focus groups, <strong>and</strong> the use<br />
of specialized workshop techniques or software for<br />
group decision making. All of these alternatives<br />
however have social implications for the<br />
underst<strong>and</strong>ing of how rights <strong>and</strong> responsibilities<br />
are distributed with society.<br />
The amount of decision control that is devolved to<br />
the community for the management of natural<br />
resources <strong>and</strong> the role that public authorities play<br />
determine to a great extent the socio-political<br />
character of a society. For some Member States,<br />
Article 14 may represent a “business as usual”<br />
scenario in that this kind of information exchange<br />
between the citizens <strong>and</strong> the public authorities<br />
already takes place in some form. This means that<br />
the communication infrastructures are already in<br />
place <strong>and</strong> that both individuals <strong>and</strong> stakeholder<br />
groups expect the opportunity to comment on<br />
planning proposals. For other Member States, it is<br />
possible that there is little history of such<br />
exchanges, making the implementation of this<br />
Article more difficult. It may be costly to establish<br />
new lines of communication <strong>and</strong> the facilities for<br />
collecting <strong>and</strong> recording public opinions, <strong>and</strong> such<br />
procedures may be incompatible with current<br />
planning approaches. Moreover, there may be<br />
resistance to what may seem like a step towards a<br />
redistribution of power that threatens the freedom<br />
of individuals or organizations to make decisions<br />
in a non-transparent way.<br />
2. DECISION MAKING IN IWRM<br />
2.1 The role of public participation in IWRM<br />
In a decision making process, it is possible to<br />
identify people, groups or institutions that can play<br />
a meaningful role in the final decision. In general,<br />
we can classify these main actors as decision<br />
makers, people <strong>and</strong> groups affected, <strong>and</strong> analysts .<br />
But normally in the real life, not all of these actors<br />
are always involved in the decision making<br />
process.<br />
The decision maker is situated in the centre of the<br />
decision making process <strong>and</strong> is the one who has<br />
the institutional power <strong>and</strong> responsibility to select<br />
<strong>and</strong> implement a solution for a specific problem.<br />
People affected are all those whom will be<br />
influenced by the consequences of the solution<br />
adopted <strong>and</strong> implemented by the decision maker.<br />
The analyst is the person/group that helps <strong>and</strong><br />
guides the decision maker to analyse <strong>and</strong> represent<br />
their preference structures <strong>and</strong> those from other<br />
interested groups.<br />
One of the main issues in the field of<br />
environmental decision making is the need,<br />
sometimes the obligation imposed by the<br />
legislation, to communicate the decision process<br />
<strong>and</strong> make it more comprehensible <strong>and</strong> transparent.<br />
For the reasons described above, there is no doubt<br />
that public participation has become a major issue<br />
in IWRM. In order to facilitate the active<br />
involvement of all the stakeholders in water<br />
decision problems there is a challenge that has to<br />
be faced: the integration of scientific knowledge<br />
<strong>and</strong> public participation. This is not an easy task.<br />
Facing water problems, decision makers find<br />
public participation important for various reasons,<br />
first of all because it is required by legislation (e.g.<br />
the WFD). Moreover, decision makers are<br />
responsible of the selected decision <strong>and</strong> also its<br />
acceptance, for which public participation is<br />
essential. Nevertheless, major problems in IWRM<br />
like the lack of available information, the<br />
uncertainty about future effects or the incomplete<br />
knowledge of experts, create more difficulties on<br />
obtaining these goals . Decsion makers have in<br />
general, little experience in sustainable water<br />
management. Because of this inexperience <strong>and</strong> the<br />
uncertainty inherent to these decision problems ,<br />
public preferences need to be included in a more<br />
direct way by sharing part of the responsibility <strong>and</strong><br />
trying to find compromise solutions that facilitate<br />
acceptance.<br />
618
Another reason for public participation is the role<br />
that water plays in our society. Water can be<br />
considered an important primary good, <strong>and</strong> is<br />
closely related to social <strong>and</strong> economic<br />
development. In addition, environmental<br />
sustainability is critical. One possibility to better<br />
underst<strong>and</strong> <strong>and</strong> implement common interests is<br />
public participation.<br />
In contrast, some disadvantages have to be also<br />
taken into account <strong>and</strong> to be solved. Public<br />
administrations, that normally have the<br />
responsibility to make decisions in IWRM, are not<br />
always experienced applying public participation.<br />
In addition, the public involvement could represent<br />
a problem to the restrictions in cost <strong>and</strong> time that<br />
normally guides administrative procedures.<br />
2.2 Integration of public participation in<br />
IWRM<br />
Once the crucial importance of the public<br />
participation in the decision making process in<br />
IWRM has been recognized, the next step must be<br />
to clarify the way public participation, decision<br />
making <strong>and</strong> science knowledge can be integrated.<br />
For this integration, all the meaningful information<br />
has to be collected, structured <strong>and</strong> presented in an<br />
underst<strong>and</strong>able way to help decision makers to<br />
integrate all the actors involved in the decision<br />
making process <strong>and</strong> all the scientific knowledge<br />
available. Several decision support systems have<br />
been developed in the last years to satisfy this<br />
need, for specific water resource planing activities<br />
such as prevention of water shortages (drought),<br />
surpluses (floods) <strong>and</strong> water impairment<br />
(pollution). Examples of such DSS are<br />
WATERWARE [Fedra, 1994], [Jamieson <strong>and</strong><br />
Fedra, 1996a; Jamieson <strong>and</strong> Fedra, 1996b],<br />
AQUATOOL [Andreu et al., 1996], NELUP<br />
[O’Callaghan, 1995], [Dunn et al., 1996],<br />
FLOODSS [Catelli et al., 1998], DSSIPM [da<br />
Silva et al., 2001], STEEL-GDSS [Ostrowski,<br />
1997].<br />
A decision making process normally implies the<br />
following steps (Figure 1): identification of the<br />
alternatives that can solve the problem; the<br />
selection of the criteria against which alternatives<br />
are going to be compared; the estimation of the<br />
performances of the alternatives related to the<br />
criteria; the selection of the aggregation procedures<br />
of the information derived from performances <strong>and</strong><br />
the relative importance of the criteria in the final<br />
decision.<br />
As described above, decision makers do not have<br />
information enough about the perceptions of the<br />
problems by the society due to the complexity of<br />
water problems. The role of public participation at<br />
this level could be helpful to identify the main<br />
relevant criteria <strong>and</strong> their societal targets in the<br />
decision process. However, the general public also<br />
has problems to identify these criteria , normally<br />
represented by physical, social <strong>and</strong> economic<br />
issues, out of specific <strong>and</strong> comprehensive data. For<br />
this reason, indicators available from scientific<br />
knowledge can provide crucial guidance for<br />
decision-making. They can translate physical <strong>and</strong><br />
social science knowledge into manageable units of<br />
information that can facilitate the public<br />
participation in the decision-making process.<br />
Indicators may provide a means of measuring,<br />
monitoring <strong>and</strong> reporting on progress towards<br />
societal goals (e.g. quality of life, welfare, etc.) . It<br />
may be thus possible to assess effectiveness of<br />
policy measures by analysing causality between a<br />
policy <strong>and</strong> its impacts in terms of changes in<br />
indicator values. Still, getting the public to<br />
underst<strong>and</strong> such scientific information is daunting.<br />
INDICATORS &<br />
MEASURES<br />
CONCEPTUAL<br />
FRAMEWORK<br />
ANALYSIS<br />
ASSESSMENT<br />
ALTERNATIVES<br />
AGGREGATION<br />
PROCEDURE<br />
ANALYSIS<br />
MATRIX<br />
EVALUATION<br />
MATRIX<br />
DECISION<br />
DECISION<br />
ANALYSIS<br />
CRITERIA<br />
ASSESING<br />
WEIGHTS<br />
PUBLIC<br />
PARTICIPATION<br />
Figure 1: Knowledge <strong>and</strong> decision making for<br />
IWRM <strong>and</strong> sustainable development.<br />
2.3 Using conceptual frameworks for public<br />
participation<br />
In order to assess whether policies will be working<br />
<strong>and</strong> to fine-tune them in order to reach the ultimate<br />
objective, conceptual frameworks are needed.<br />
They facilitate the underst<strong>and</strong>ing <strong>and</strong> exchange of<br />
information between policy-makers, stakeholders<br />
<strong>and</strong> technical <strong>and</strong> scientific support.<br />
Public participation could be also involved in the<br />
identification of alternatives. But as political<br />
decision makers, they need an overall view of the<br />
problems. Frameworks that structure collections of<br />
indicators <strong>and</strong> that communicate their application<br />
are being developed, at different analytical levels.<br />
For the purposes of IWRM, the frameworks for<br />
environmental reporting <strong>and</strong> monitoring may play<br />
a positive role. A relevant example is the DPSIR<br />
619
framework (Driving Force – Pressure – State –<br />
Impact – Response), developed by European<br />
institutions: the EEA <strong>and</strong> Eurostat [EA, 1999].<br />
This conceptual framework applied to water<br />
management is reported in Figure 2 <strong>and</strong> presented<br />
in more detail.<br />
The DPSIR framework is widely used to structure<br />
indicators to allow for a holistic <strong>and</strong> multidimensional<br />
view of causal relationships in<br />
human-environmental systems . Within the DPSIR<br />
framework, indicators are used to assess different<br />
states of the interaction between man <strong>and</strong> his<br />
environment. The integrated set of indicators is<br />
assumed to simplify for the decision-maker <strong>and</strong><br />
stakeholders the comprehension of the complex<br />
interlinkages between multisectoral human action<br />
<strong>and</strong> the coevolutions of ecological, economical <strong>and</strong><br />
social states.<br />
AGRICULTURE<br />
- Irrigation<br />
- Intensification<br />
INDUSTRY<br />
- Water supply<br />
- Refrigeration<br />
ENERGY<br />
- Hydropower<br />
- Dams<br />
TOURISM<br />
- Water supply<br />
- Recreational<br />
HOUSEHOLDS<br />
- Water supply<br />
- Treatment<br />
EUROPEAN ECONOMY<br />
PRODUCTION TECHNOLOGY CONSUMPTION<br />
WATER<br />
ABSTRACTION<br />
- Surface water<br />
- Groundwater<br />
EMISSIONS<br />
- Localised<br />
- Diffused<br />
BASIN<br />
ACTIVITIES<br />
- Agriculture<br />
- Forestry<br />
BASIN<br />
MANGEMENT<br />
- Dams<br />
- Canals<br />
PHYSICAL STATE<br />
- Hydrology<br />
- L<strong>and</strong>scape<br />
- Availability<br />
CHEMICAL STATE<br />
- Air quality<br />
- Water quality<br />
- Soil quality<br />
BIOLOGICAL<br />
STATE<br />
- Marine waters<br />
- Surface waters<br />
- Groundwater<br />
POLICIES, PLANS, PROGRAMS AND PROJECTS<br />
EUROPEAN ENVIRONMENT<br />
DRIVING FORCES PRESURES STATE IMPACT<br />
MACRO-ECONOMIC<br />
& SECTOR<br />
POLICY<br />
ENVIRONMENTAL<br />
POLICY<br />
RESPONSES<br />
SETTING<br />
TARGETS<br />
ENVIRONMENTAL<br />
IMPACTS<br />
- Biodiversity<br />
- Wetl<strong>and</strong>s<br />
- Ecosystems<br />
ECONOMIC<br />
IMPACTS<br />
- Abatement costs<br />
- Scarcity<br />
SOCIAL<br />
IMPACTS<br />
- Cultural heritage<br />
- Welfare<br />
PRIORITISING<br />
Figure 2: DPSIR framework applied to water<br />
management [adapted from NERI 1997]<br />
As the example shown in Figure 2, conceptual<br />
frameworks could help to identify the decision<br />
level related with the specific problem <strong>and</strong> the<br />
range of alternatives that could solve it. This<br />
conceptual framework allows to have a common<br />
underst<strong>and</strong>ing of the problem that is a basic step<br />
for an effective decision making process <strong>and</strong> the<br />
basis to propose.<br />
In order to obtain the analysis matrix, decision<br />
makers have to reflect their value judgements <strong>and</strong><br />
preferences by the public utility functions. As in<br />
the selection of the criteria, decision makers have<br />
the problem of lack of information about this<br />
point. That is why at this point public participation<br />
is needed. By public participation, asking directly<br />
all the actors involved in the decision process<br />
about their individual preferences, the general<br />
form of the public utility function for each<br />
criterion previously selected can be obtained.<br />
Public participation is also needed in the selection<br />
of the aggregation procedure. Several aggregation<br />
methods are available <strong>and</strong> the analyst should help<br />
to select the most suitable method based on the<br />
preferences of the actors involved <strong>and</strong>, depending<br />
on the problem faced. Not all the problems are the<br />
same <strong>and</strong> each specific context requires a specific<br />
method.<br />
The last point where public participation could<br />
play an important role in the decision process is in<br />
the assessing of weights to aggregate all the<br />
information. In this step, some conflict may arise<br />
because of the different interest of the actors<br />
involved in the process. Public participation could<br />
increase the acceptance of the final decisions,<br />
making clear the individual preferences <strong>and</strong> giving<br />
the basis for possible compromise solutions<br />
We believe that public participation could play an<br />
important role in the decision making process<br />
related to IWRM, where the environmental tools<br />
could be also helpful. There is not a consensus<br />
about the involvement of public in the decision<br />
process. Different levels of public participation<br />
have their advantages <strong>and</strong> disadvantages <strong>and</strong> they<br />
must be clearly established for each particular type<br />
of problem.<br />
3. MULINO DSS<br />
A methodological approach <strong>and</strong> a DSS tool have<br />
been developed within the MULINO Project<br />
[Giupponi et al., 2004] for integrating the four<br />
steps described above, in the context of decision<br />
making in IWRM. The next paragraphs describe<br />
how indicators <strong>and</strong> indices are managed within a<br />
conceptual framework <strong>and</strong> how they can be<br />
utilized in specific forms of analysis, for the<br />
implementation of IWRM principles in decision<br />
making.<br />
3.1 The conceptual framework <strong>and</strong> the role<br />
<strong>and</strong> management of indicators<br />
Within the IWRM context the initial task of<br />
decision makers is usually that of acquiring or<br />
consolidating knowledge about the territory they<br />
manage by collecting information about human<br />
activities <strong>and</strong> their relationships with the<br />
environmental systems. This may be based upon<br />
the identification of suitable indicators, which may<br />
provide concise quantification <strong>and</strong> temporal<br />
monitoring of the main human <strong>and</strong> environmental<br />
variables interacting within the given territorial<br />
systems, typically a river basin.<br />
The whole informative <strong>and</strong> decision process<br />
should be then formalized within a conceptual<br />
network, in this case based upon the DPSIR<br />
approach. In such a conceptual framework related<br />
to natural resources management <strong>and</strong>, in particular<br />
to IWRM, the Impacts describe the existing<br />
problems arising from the change detected in State<br />
variables, which affects their economic value,<br />
environmental function <strong>and</strong> social role (either in<br />
quantitative, or qualitative terms), thus allowing to<br />
support decision making within the perspective of<br />
sustainable development. Such a conceptual<br />
620
structure can support the establishment of new<br />
lines of communication between different actors<br />
<strong>and</strong> help to facilitate the collection <strong>and</strong> recording<br />
of public opinions..<br />
The level of the responses has to be related to the<br />
magnitude of the impacts. These different<br />
responses need different planning processes <strong>and</strong><br />
different decision makers could be involved. The<br />
different planning levels could be policies, plans,<br />
programs <strong>and</strong> projects, from macro to micro level.<br />
A crucial aspect of implementing the DPSIR<br />
approach in a methodology for implementing the<br />
principles of IWRM in decision making is the<br />
transformation of a static reporting scheme in a<br />
dynamic framework for integrated analysis <strong>and</strong><br />
assessment. The next two paragraphs present how<br />
Integrated Assessment <strong>Modelling</strong> (IAM) combined<br />
with a Multi-Criteria Analysis (MCA) methods<br />
can provide methodological support for analysis<br />
<strong>and</strong> assessment procedures.<br />
3.2 Analysis methods: modelling <strong>and</strong><br />
evaluation<br />
The implementation of IAM in the DPSIR<br />
framework is approached in the proposed<br />
methodology by focusing on the DPS part of the<br />
conceptual framework. These three elements were<br />
considered as explicit formalizations of driving<br />
variables, model parameters <strong>and</strong> outputs,<br />
respectively. In the case of water pollution models,<br />
for instance, D’s represent the forcing variables<br />
ruling the behaviour of the simulated system (i.e.<br />
the catchment). P’s may be represented by<br />
parameters that express the rate of pollution<br />
processes <strong>and</strong> S’s are the output variables<br />
quantifying the dynamic evolution of the<br />
catchment system, as affected by the considered<br />
pollution sources <strong>and</strong> processes. Integration of<br />
models may occur at various levels <strong>and</strong> in different<br />
ways <strong>and</strong> thus relationships along the chains could<br />
be expressed by parallel one-to-one flows, or oneto-many<br />
(e.g. one activity affecting various<br />
environmental compartments), or many-to-one<br />
(e.g. various sectors affecting the same<br />
environmental indicator), or even many-to-many,<br />
in the case of multi-sector integrated models.<br />
In the context of environmental decision making,<br />
IAM can support the identification of the correct<br />
Responses by providing sets of indicator values.<br />
These values are derived from subsequent<br />
simulation runs in which model(s) are<br />
parameterised to represent the expected<br />
consequences of a set of possible alternative<br />
responses. The development of a set of evaluation<br />
indices is a crucial step. It should be targeted to<br />
evaluate Impacts deriving from the State indicators<br />
provided by IAM. Evaluation procedures may be<br />
implemented by focusing on the link between S<br />
<strong>and</strong> I <strong>and</strong> between I <strong>and</strong> R by adapting concepts<br />
<strong>and</strong> methods derived from MCA literature, Multi-<br />
Attribute methods in particular [Hwang <strong>and</strong> Yoon,<br />
1981]. Within this disciplinary context a<br />
preliminary phase of Problem Structuring is<br />
targeted to the identification of the criteria to be<br />
considered for choosing among previously defined<br />
options. These factors are expressed as indicators<br />
deriving from output variables of IAMs or<br />
monitoring activities. The step between the<br />
quantification of State variables <strong>and</strong> the<br />
identification of Impact evaluation indices can be<br />
conceptualised according to MCA theory as the<br />
conversion of the Analysis Matrix into an<br />
Evaluation Matrix (EM), which expresses the<br />
estimated impacts.<br />
Having identified the impacts as they vary under<br />
the effects of alternative response options, the<br />
decision maker has to apply a decision rule to<br />
aggregate the values stored in the EM to identify<br />
the preferred option, filling therefore the gap<br />
between I <strong>and</strong> R. In the simplest case, the rule can<br />
be expressed by the weighted sum of values stored<br />
in the columns of the EM. Various iterations are<br />
possible <strong>and</strong> needed at this step to refine the<br />
weights, or apply alternative decision rules by<br />
considering the results of the sensitivity analysis to<br />
select a robust response. Parallel procedures are<br />
also possible in multi-stakeholders group decision<br />
making.<br />
3.3 Assessment methods: a dynamic <strong>and</strong><br />
integrated DPSIR-DSS tool<br />
A DSS is ideally suited to answering questions<br />
arising from policy changes on water resources by<br />
providing the underst<strong>and</strong>ing of the processes<br />
involved, evaluating the consequences <strong>and</strong><br />
delivering advice. Moreover, communication about<br />
how decisions are reached is greatly facilitated<br />
using a DSS in which effects of alternative options<br />
can be explained <strong>and</strong> their impacts assessed in a<br />
form which can be comprehended by the nonexpert.<br />
In accordance with the WFD, the DSS<br />
developed by the MULINO project adopts the<br />
DPSIR as a well known intuitive graphical<br />
interface <strong>and</strong> integrates hydrological <strong>and</strong> socioeconomic<br />
approaches in order to assist water<br />
authorities in the management of water resources.<br />
From a practical viewpoint, mDSS manages social,<br />
economic <strong>and</strong> environmental criteria, by<br />
formalising them as D, P, or S indicators <strong>and</strong> then<br />
by considering them as decision factors within the<br />
AM.<br />
4. CONCLUSIONS<br />
There is a clear need for methodologies <strong>and</strong> tools<br />
to put IWRM principles into practice, in an<br />
application context in which decisions <strong>and</strong> choices<br />
621
are assessed in terms of their sustainability not<br />
only over the long term but also with regards to<br />
their day-to-day contribution to the perspective of<br />
sustainable development.<br />
The need mentioned above may also be described<br />
in terms of the implementation of an integrated<br />
methodological framework allowing decision<br />
makers to choose first <strong>and</strong> then to monitor the<br />
process induced by their decisions.<br />
Various methods <strong>and</strong> tools, such as modelling,<br />
environmental impact assessment <strong>and</strong> decision<br />
support, have shown to provide rational insight in<br />
the system’s behaviour <strong>and</strong> the problems<br />
addressed. However, integration remains a difficult<br />
issue.<br />
The conceptual framework briefly described above<br />
may contribute to provide methodological support<br />
to cope with the general problem of IWRM<br />
implementation, by supporting in particular:<br />
- the management of the complexity of decision<br />
context s typical of IWRM;<br />
- the management of large amounts of multisectoral<br />
<strong>and</strong> multidisciplinary information;<br />
- the commu nication between the scientific <strong>and</strong><br />
the policy sector <strong>and</strong> between decision makers<br />
<strong>and</strong> the involved stakeholders.<br />
5. REFERENCES<br />
Andreu, J., Capilla, J., <strong>and</strong> Sanchis, E.<br />
AQUATOOL - a generalized DSS for waterresources<br />
planning <strong>and</strong> operational<br />
management, Journal of Hydrology, 177:269-<br />
291, 1996.<br />
Catelli, C., Pani, G., <strong>and</strong> Todini, E., FLOODSS,<br />
Flood operational DSS, Balabanis, P.,<br />
Bronstert, A., Casale, R., <strong>and</strong> Samuels, P.<br />
(eds): Ribamod: River basin modelling,<br />
management <strong>and</strong> flood mitigation, 1998.<br />
da Silva, L. M., Park, J. R., Keatinge, J. D. H.,<br />
Pinto, <strong>and</strong> P.A., The use of the DSSIPM in<br />
the Alentejo region of the southern Portugal,<br />
Agricultural Water Management,51:203-215,<br />
2001.<br />
Dunn, S. M., Mackay, R., Adams, R., <strong>and</strong><br />
Oglethorpe, D. R., The hydrological<br />
component of the NELUP decision support<br />
system: an appraisal, Journal of Hydrology,<br />
177:213-235, 1996.<br />
EC (European Communities) , Directive<br />
2000/60/EC of the European Parliament <strong>and</strong><br />
of the Council Establishing a Framework for<br />
Community Action in the Field of Water<br />
Policy (OJ L 327, 22.12.2000), 2000.<br />
EEA (European <strong>Environmental</strong> Agency),<br />
<strong>Environmental</strong> Indicators: Typology <strong>and</strong><br />
Overview, European Environment Agency,<br />
Copenhagen, 1999.<br />
Fedra, K., GIS <strong>and</strong> environmental modeling.<br />
Goodchild, M. F., Parks, B. O., <strong>and</strong><br />
Steyaert, L. T. (eds): <strong>Environmental</strong><br />
modeling with GIS, Oxford university<br />
press, 1994.<br />
Giupponi C., Mysiak J., Fassio A. <strong>and</strong> V. Cogan ,<br />
MULINO-DSS: a computer tool for<br />
sustainable use of water resources at the<br />
catchment scale, Mathematics <strong>and</strong> Computers<br />
in Simulation, 64 (1), 2004.<br />
Hwang, C.L. <strong>and</strong> K. Yoon, Multiple Attribute<br />
Decision Making: Method <strong>and</strong> Applications,<br />
Springer-Verlag, Berlin, 1981.<br />
ICSU (<strong>International</strong> Council for Science), ‘Making<br />
Science for Sustainable Development More<br />
Policy Relevant: New Tools for Analysis’,<br />
ICSU Series on Science for Sustainable<br />
Development, No. 8, 2002<br />
Jamieson, D. G. <strong>and</strong> Fedra, K., The 'WaterWare'<br />
decision-support system for river-basin<br />
planning. 1. Conceptual design, Journal of<br />
Hydrology, 1977:163-175, 1996a.<br />
Jamieson, D. G. <strong>and</strong> Fedra, K. The 'WaterWare'<br />
decision-support system for river-basin<br />
planning. 3. Example applications, Journal of<br />
Hydrology, 177:199-211, 1996b.<br />
Lahdelma R., P.Salminen <strong>and</strong> J. Hokkanen, Using<br />
multicriteria methods in environmental<br />
planning <strong>and</strong> management, <strong>Environmental</strong><br />
Management, 26 (6): 595-605, 2000.<br />
Luiten, H. (1999), ‘A legislative view on science<br />
<strong>and</strong> predictive models’, <strong>Environmental</strong><br />
Pollution, 100: 5-11, 1999.<br />
NERI (National <strong>Environmental</strong> Research Institute)<br />
(1997), Integrated <strong>Environmental</strong> Assessment<br />
on Eutrophication, Technical Report nº207,<br />
Denmark.<br />
O’Callaghan, NELUP: An introduction, Journal<br />
<strong>Environmental</strong> Planning <strong>and</strong> Management,<br />
38(1):5-20, 1995.<br />
Ostrowski, M. W., Improving sustainability of<br />
water resources systems using the group<br />
decision support system STEEL-DSS,<br />
Research Report at Darmstadt University of<br />
Technology, 1997.<br />
622
A Dual-scale <strong>Modelling</strong> approach to Integrated Resource<br />
Management in East <strong>and</strong> South-east Asia: Challenges <strong>and</strong><br />
Potential solutions<br />
Reimund Rötter 1 , Marrit van den Berg 2,3 , Huib Hengsdijk 4 , Joost Wolf 1 , Martin van Ittersum 2 , Herman van<br />
Keulen 2,4 , Epifania O. Agustin 5 , Tran Thuc Son 6 , Nguyen Xuan Lai 7 , Wang Guanghuo 8 , <strong>and</strong> Alice G.<br />
Laborte 9<br />
1<br />
2<br />
3<br />
4<br />
5<br />
6<br />
7<br />
8<br />
9<br />
Alterra, Soil Science Centre, Wageningen UR, P.O. Box 47, 6700 AA Wageningen, The Netherl<strong>and</strong>s<br />
E-mail: Reimund.Roetter@wur.nl<br />
Plant Production Systems, Wageningen University, P.O. Box 430, 6700 AK Wageningen, The Netherl<strong>and</strong>s<br />
Development Economics, Wageningen University, P.O. Box 8130, 6700 EW Wageningen, The Netherl<strong>and</strong>s<br />
Plant Research <strong>International</strong>, Wageningen UR, P.O. Box 16, 6700 AA Wageningen, The Netherl<strong>and</strong>s<br />
Mariano Marcos State University, Batac, Ilocos Norte, Philippines<br />
National Institute for Soils <strong>and</strong> Fertilizer, Hanoi, Vietnam<br />
Cuu Long Delta Rice Research Institute, Omon, Cantho, Vietnam<br />
Zhejiang University, Hangzhou, P.R. China<br />
<strong>International</strong> Rice Research Institute (IRRI), DAPO, Box 7777, Metro Manila, Philippines<br />
Abstract: Currently, in many of the highly productive lowl<strong>and</strong> areas of E <strong>and</strong> SE Asia a trend to further<br />
intensification <strong>and</strong> diversification of (agricultural) l<strong>and</strong> use can be observed. Growing economies <strong>and</strong> urbanization<br />
also increase the claims on l<strong>and</strong> <strong>and</strong> water by non-agricultural uses. As a result, decisions related to the management<br />
<strong>and</strong> planning of scarce resources become increasingly complex. Technological innovations at the field/farm level are<br />
needed but not sufficient – changes in resource use at regional scale will also be essential. To support decisionmaking<br />
in such situations, we advocate a multi-scale modelling approach embedded in a solid participatory process.<br />
To this end, the Integrated Resource Management <strong>and</strong> L<strong>and</strong> use Analysis (IRMLA) Project is developing an<br />
analytical framework <strong>and</strong> methods for resource use analysis <strong>and</strong> planning, for four sites in Asia. In the envisaged<br />
multi-scale approach, integration of results from field, farm, district <strong>and</strong> provincial level analysis is based on<br />
Interactive multiple goal linear programming (), Farm Household <strong>Modelling</strong> (FHM), production ecological concepts<br />
<strong>and</strong> participatory techniques. The novel approach comprises the following steps: (i) Inventory/quantification of<br />
current l<strong>and</strong> use systems, resource availability, management practices <strong>and</strong> policy views, (ii) Analysis of alternative,<br />
innovative l<strong>and</strong> use systems/technologies, (iii) Exploration of the opportunities <strong>and</strong> limitations to change resource<br />
use at regional scale under alternative future scenarios, (iv) <strong>Modelling</strong> decision behavior of farmers <strong>and</strong><br />
identification of feasible policy interventions, <strong>and</strong> (v) Synthesis of results from farm to regional level for negotiation<br />
of the most promising options by a stakeholder platform. In the current paper, the operationalization of a dual-scale<br />
approach is illustrated by the outputs (development scenarios, promising policy measures <strong>and</strong> innovative production<br />
systems) from various component models for the case study Ilocos Norte, Philippines. A procedure is discussed for<br />
the integration of results from the different model components at two different decision making levels (farm <strong>and</strong><br />
province).<br />
Keywords: l<strong>and</strong> use conflicts; scenario analysis; bio-economic models; rice-based farming, Philippines.<br />
1. INTRODUCTION<br />
Agricultural systems in East <strong>and</strong> South-east Asia<br />
are being challenged by the simultaneous<br />
requirements for increased productivity, more<br />
diversified products <strong>and</strong> reduced environmental<br />
impact, creating potential conflict situations in<br />
l<strong>and</strong> use objectives among various stakeholder<br />
groups. Current l<strong>and</strong> use policies in general<br />
inadequately take into consideration multiple<br />
objectives <strong>and</strong> the increased complexity of<br />
current resource management decisions [Walker,<br />
2002; Lu et al., 2004]. In such situations,<br />
effective systems analysis tools at different<br />
623
scales are required to identify conflicts <strong>and</strong><br />
design sustainable l<strong>and</strong> use systems <strong>and</strong><br />
supportive policy options [Van Ittersum et al.,<br />
1998].<br />
Since the early 1980s, a range of complementary<br />
analytical frameworks <strong>and</strong> operational tools have<br />
been developed [Stoorvogel & Antle, 2001]. On<br />
the basis of their objectives we can distinguish<br />
explorative <strong>and</strong> predictive tools. Explorative<br />
tools analyse the potential (im)possibilities of<br />
strategic natural resource use configurations,<br />
often at regional or farm scale. To this purpose, a<br />
frequently used procedure is interactive multiple<br />
goal linear programming () (De Wit et al., 1988).<br />
models generate optimal l<strong>and</strong> use options under<br />
different sets of objectives <strong>and</strong> constraints.<br />
Regional models as operationalized in the<br />
SysNet project [Roetter et al., 2004] form one of<br />
the major building blocks of the IRMLA<br />
approach to multi-scale analysis.<br />
So-called predictive tools are required to analyse<br />
the likely l<strong>and</strong> use changes in the short term as a<br />
result of introducing alternative agricultural<br />
policies <strong>and</strong> technologies [Bouman et al., 2000].<br />
For example, the technique of farm household<br />
modelling (FHM) is applied for simulating the<br />
impact of feasible changes in policy <strong>and</strong><br />
technology choice for different (model) farm<br />
groups in a study area [Kruseman <strong>and</strong> Bade,<br />
1998]. FHM is the second major tool in the<br />
IRMLA project.<br />
In most cases, the various modelling approaches,<br />
whether exploratory or predictive have been<br />
applied separately at a single scale. This can only<br />
shed partial light on solutions to agricultural <strong>and</strong><br />
environmental policy problems which are<br />
essentially of a multi-scale nature. Policy makers<br />
at the provincial level, for instance, have only a<br />
limited number of variables that they can control.<br />
Variables such as choice of crop, area cultivated<br />
<strong>and</strong> fertilizer <strong>and</strong> pesticide rates are decided by a<br />
huge number of other decision makers, i.e.<br />
farmers, which apply different criteria. C<strong>and</strong>ler<br />
et al. [1981] addressed this problem <strong>and</strong><br />
examined the potential contribution of multilevel<br />
programming to solve two-level (public – private<br />
interest) conflicts. They detected a range of<br />
algorithmic problems in multilevel<br />
programming. Solutions were only found for<br />
special cases. To make things even more<br />
complicated, public interest at one level (e.g.<br />
province) may be in conflict with the public<br />
interest at another level (e.g. municipality).<br />
Integration of results from different scales<br />
remains a rsearch challenge [Bouman et al.,<br />
2000]. In this paper we do not intend to resolve<br />
the problems inherent to multilevel<br />
programming. Rather we want to demonstrate<br />
that, as a first step, combination of farm<br />
household modelling <strong>and</strong> regional multiple goal<br />
linear programming embedded in participatory<br />
processes can overcome shortcomings of singlelevel<br />
modelling. This will be illustrated by<br />
confronting results from regional level<br />
explorations with farm household level analysis<br />
of the best l<strong>and</strong> use strategy. The result from this<br />
dual-scale analysis will help to identify the<br />
options for promoting more resource-use<br />
efficient production technologies than presently<br />
practiced in Ilocos Norte province, Philippines.<br />
2. CASE STUDY CHARACTERIZATION<br />
2.1 L<strong>and</strong> use issues <strong>and</strong> agricultural<br />
development perspectives for Ilocos Norte<br />
According to current local government views,<br />
agriculture will maintain its central role in the<br />
economic development of Ilocos Norte province.<br />
However, agriculture will increasingly compete<br />
for l<strong>and</strong> with for instance industrialization,<br />
recreation parks <strong>and</strong> tourism areas. Competition<br />
for scarce natural resources, particularly l<strong>and</strong> <strong>and</strong><br />
water, is evident in the most recent provincial<br />
development plan, which includes conversion of<br />
some agricultural areas into other uses. Such<br />
conversion will not spare the strategic agriculture<br />
<strong>and</strong> fisheries development zones identified in<br />
earlier plans, such as Dingras municipality. The<br />
provincial plan sets boundary conditions on<br />
future availability on agricultural l<strong>and</strong> resources.<br />
Recent dialogues between scientists <strong>and</strong> Ilocano<br />
stakeholders on agricultural l<strong>and</strong> use issues<br />
revealed that assessment of trade-offs between<br />
rice production, diversification of production <strong>and</strong><br />
farmers’ income was a major issue for the Ilocos<br />
Norte province as well as for Dingras<br />
municipality. <strong>Environmental</strong> issues, such as<br />
nitrate pollution <strong>and</strong> excessive pesticide residues<br />
needed to be addressed as well [Roetter <strong>and</strong><br />
Wolf, 2003].<br />
2.2 Site description<br />
Ilocos Norte Province, in northwestern Luzon,<br />
Philippines, has a population of nearly 0.5<br />
million people <strong>and</strong> a total l<strong>and</strong> resource of 0.34<br />
million ha, of which 46% is covered by forests.<br />
Mean annual rainfall ranges between 1650 mm<br />
624
in the southwest to more than 2,400 mm in the<br />
eastern mountain ranges. On average, 6-7<br />
typhoons per year cross the province (mostly<br />
between August <strong>and</strong> November). About 38% of<br />
the total area is classified as agricultural l<strong>and</strong><br />
[Roetter et al., 2000]. Rice-based production<br />
systems prevail. Rice is grown in the wet season<br />
(June-October), whereas diversified cropping<br />
(tobacco, garlic, onion, maize, sweet pepper <strong>and</strong><br />
tomato) is practiced in the dry season, using<br />
irrigation (mainly) from groundwater. A welldeveloped<br />
marketing system facilitates this<br />
relative intensive production system of rice <strong>and</strong><br />
cash crops [Lucas et al., 1999]. Dingras has a<br />
population of 33,300 persons <strong>and</strong> a total l<strong>and</strong><br />
resource of 17,310 ha, of which 55% is<br />
agricultural l<strong>and</strong>.<br />
3. DEVELOPMENT AND IMPLEMENTA-<br />
TION OF MODELS AND DATABASES<br />
3.1 Regional level<br />
A total of 200 l<strong>and</strong> units were defined by overlaying<br />
biophysical characteristics (irrigated<br />
areas, annual rainfall <strong>and</strong> distribution, slope <strong>and</strong><br />
soil texture) <strong>and</strong> administrative units, comprising<br />
22 municipalities <strong>and</strong> one township. The total<br />
area available for agriculture for the year 2010<br />
was estimated at 119,850 ha (assuming an<br />
overall l<strong>and</strong> use conversion rate of 7% from<br />
agriculture to non-agricultural uses) [Roetter et<br />
al., 2000]. The l<strong>and</strong> use types (LUTs) included<br />
in this study comprise (i) single cropping of root<br />
crops, sugarcane, <strong>and</strong> rice followed by fallow;<br />
(ii) double cropping: two rice crops, rice in<br />
rotation with (yellow or white) corn, garlic,<br />
mungbean, peanuts, tomato, tobacco, cotton,<br />
potato, onion, sweet pepper, eggplant, <strong>and</strong><br />
vegetables; (iii) triple cropping: three rice crops,<br />
<strong>and</strong> rice in rotation with garlic <strong>and</strong> mungbean,<br />
with (white or yellow) corn <strong>and</strong> mungbean, <strong>and</strong><br />
with water melon <strong>and</strong> mungbean. The available<br />
resources for agriculture such as l<strong>and</strong>, laborforce<br />
<strong>and</strong> irrigation water were quantified per<br />
l<strong>and</strong> unit <strong>and</strong> per month. Provincial dem<strong>and</strong> for<br />
agricultural products was assessed on the basis<br />
of information on per capita dem<strong>and</strong> <strong>and</strong><br />
projected population from the Provincial<br />
Planning Office. Details on the procedures<br />
applied to assess resource availability <strong>and</strong><br />
constraints have been described in previous<br />
studies [Roetter et al., 2000; Laborte et al.,<br />
2002].<br />
An model developed for Ilocos Norte province<br />
[Roetter et al., 2004] was applied. Four major<br />
agricultural development goals as identified by<br />
stakeholders were included: Maximizing<br />
farmers’ income <strong>and</strong> rice production, <strong>and</strong><br />
minimizing nitrogen fertilizer <strong>and</strong> biocide use<br />
(while maintaining a minimum level of income<br />
<strong>and</strong>/or crop production). The specific ‘what-if<br />
question’ addressed is: how does goal attainment<br />
(rice production, income, etc.) <strong>and</strong> l<strong>and</strong> use<br />
allocation change, if under given resource<br />
availability <strong>and</strong> a set of available production<br />
activities, the production techno-logies change.<br />
Three basic model runs were performed for<br />
analysing effects of changes in production<br />
technologies on the different l<strong>and</strong> use objectives.<br />
3.2 Farm level<br />
Dingras, one of the 22 municipalities of the<br />
province, was chosen for analysis at the<br />
household level. For Dingras, six l<strong>and</strong> units were<br />
distinguished based on drainage conditions <strong>and</strong><br />
the presence <strong>and</strong> duration of surface irrigation.<br />
Twenty-two major cropping systems were<br />
identified on the basis of an extensive farm<br />
household survey. These cropping systems do<br />
not all match with cropping systems identified at<br />
the provincial level. Dingras is located in the<br />
inner lowl<strong>and</strong>s of Ilocos Norte about 10 to 15 km<br />
to the East from the main road connecting the<br />
provincial capital Laoag with Ilocos Sur. Dingras<br />
has a relatively high incidence of triple cropping<br />
systems that are less important at the provincial<br />
level. One hundred fifty households were<br />
covered in an extensive survey <strong>and</strong> classified<br />
into four relatively homogeneous groups based<br />
on their l<strong>and</strong>, labor <strong>and</strong> capital resources. The<br />
average resource endowments of each group<br />
were used to define representative households.<br />
The major characteristics of these households<br />
are:<br />
• Medium farm, well drained: 0.92 ha of<br />
cultivated l<strong>and</strong>, 64% groundwater irrigation,<br />
74% sharecropped.<br />
• Medium farm, poorly drained: 1.07 ha, 76%<br />
surface irrigation, 80% sharecropped.<br />
• Large farm: 1.63 ha, well-drained, 85%<br />
surface irrigation, 86% sharecropped.<br />
• Small irrigated farm: 0.83 ha, well-drained,<br />
100% surface irrigation, 94% sharecropped.<br />
A linear programming model was developed for<br />
each household type. The models maximize<br />
income above subsistence, given the household<br />
specific endowment of resources, minimum<br />
625
consumption requirements, limits on off-farm<br />
employment <strong>and</strong> credit, <strong>and</strong> generic input-output<br />
coefficients for crop production <strong>and</strong> prices for<br />
inputs <strong>and</strong> outputs.<br />
3.3 Three alternative production technologies<br />
Three production technology levels were<br />
evaluated in both the regional <strong>and</strong> farm models:<br />
(technology 1) ‘average farmers’ practice,<br />
(technology 2) high yield/high input <strong>and</strong><br />
(technology 3) ‘high yield/improved practice’.<br />
The relevant input-output coefficients for<br />
technology 1 were derived from farm surveys in<br />
the province, <strong>and</strong> average values for these farms<br />
were applied [Roetter et al., 2000]. For<br />
technology 2, the mean of the values with a yield<br />
level between the 90 th <strong>and</strong> 95 th percentile of the<br />
survey data was applied. Fertilizer <strong>and</strong> pesticide<br />
use were assumed 100% higher <strong>and</strong> labour 70%<br />
higher, other inputs remaining identical to those<br />
in the average practice. For the ‘improved<br />
practice’ (technology 3), the same, high, yields<br />
as in technology 2 were assumed, but labour <strong>and</strong><br />
biocide inputs remained identical to those in<br />
‘average farmers’ practice. We assumed higher<br />
fertilizer use efficiency than in the first 2<br />
technologies. In comparison to technology 1,<br />
average applications of N, P, <strong>and</strong> K were<br />
reduced by 20% for non-rice crops. For rice, a<br />
more balanced NPK application was assumed:<br />
N was reduced by 40%, of P by 15% <strong>and</strong> of K<br />
increased by 20%.<br />
The relevant input-output coefficients were<br />
established by applying the technical coefficient<br />
generator TechnoGIN-3 [Ponsioen et al., 2004].<br />
4. RESULTS<br />
4.1 Regional level<br />
We consider two scenarios for presentation: (A)<br />
‘maximize farmers’ income’, <strong>and</strong> (B) ‘minimize<br />
N fertilizer use’. For both scenarios, the<br />
satisfaction of provincial dem<strong>and</strong> for agricultural<br />
products, <strong>and</strong> available labour <strong>and</strong> water were<br />
introduced as constraints. Results for scenario A<br />
(Table 1) show, among others, that if all farmers<br />
in Ilocos Norte would apply technology 2, their<br />
income would be considerably higher than with<br />
technology 1. However, if all farmers would<br />
apply the improved, more resource-efficient<br />
practice (technology 3), even higher income<br />
levels than with technology 2 could be achieved<br />
at about 30% lower inputs of fertilizers <strong>and</strong><br />
pesticides. When the water constraint was<br />
removed, farmers’ income could further increase<br />
by more than 50% [Laborte et al., 2002].<br />
Table 1: Results of the regional explorations<br />
(year 2010): A. Maximize Farmers’ Income <strong>and</strong><br />
B. Minimize Nitrogen Fertilizer Use (constraints:<br />
l<strong>and</strong>+water+labor <strong>and</strong> provincial dem<strong>and</strong> for<br />
important food crops satisfied)<br />
MAXIMIZE FARMERS’INCOME<br />
Variable Unit<br />
tech1 tech2 tech3<br />
tech1 tech2 tech3<br />
Income 10 9 Pesos 15.3 30.4 36.6<br />
Rice 10 3 ton 119 226 241<br />
Employment<br />
10 6 labdays 9.5 17.8 12.1<br />
Biocide 10 3 kg a.i. 75 161.6 79.5<br />
N Fertilizer 10 3 ton 13.5 33.8 15.9<br />
L<strong>and</strong> % 100 91 96<br />
used<br />
MINIMIZE N FERTILIZER USE<br />
Income 10 9 Pesos 1.0 1.0 1.3<br />
Rice 10 3 ton 113 113 113<br />
Employment<br />
10 6 labdays 2.7 3.5 2.0<br />
Biocide 10 3 kg a.i. 7.2 7.7 4.3<br />
N Fertilizer 10 3 ton 3.1 3.9 1.2<br />
L<strong>and</strong><br />
used<br />
% 22 17 17<br />
Results for scenario B (Table 1) indicate that<br />
application of technology 3 could reduce<br />
nitrogen fertilizer use by almost 70% as<br />
compared to technology 2, while still meeting<br />
the local dem<strong>and</strong> for agricultural products.<br />
Income from farming would be slightly higher<br />
than for technologies 1 <strong>and</strong> 2. In scenario B,<br />
however, for all technologies only about one<br />
fifth of the available l<strong>and</strong> would be used <strong>and</strong><br />
income from farming would be marginal as<br />
compared to scenario A.<br />
For all technologies, in scenario A, total rice<br />
production would exceed the current production<br />
levels. Site-specific, <strong>and</strong> more-balanced nutrient<br />
<strong>and</strong> pest management practices could lead to<br />
considerably higher incomes at reduced<br />
environmental costs, while still satisfying local<br />
dem<strong>and</strong> for the main food crop: a clear win-win<br />
situation.<br />
A regional explores the ultimate consequences<br />
of optimally allocating l<strong>and</strong> to different uses for<br />
a given set of objectives at provincial scale.<br />
Objectives of decision makers at lower scales are<br />
assumed subject to the provincial objectives. In<br />
reality, there are many resource managers with<br />
different objectives <strong>and</strong> resource endowments,<br />
626
<strong>and</strong> groups using different sets of criteria for<br />
guiding their decisions. The possible impact of<br />
various alternative policy interventions at farm<br />
scale is presented in the following section for<br />
different farm types in Dingras municipality.<br />
4.2 Farm level<br />
Table 2 presents the results of base run<br />
simulations <strong>and</strong> 4 development scenarios for<br />
three of the four representative households. The<br />
results for the medium farm-poorly drained are<br />
not listed, as they are very similar to those of the<br />
medium farm-well drained.<br />
In the base run, all model farmers use technology<br />
2 on irrigated l<strong>and</strong> <strong>and</strong> technology 3 on most of<br />
their dryl<strong>and</strong>. The small farmer also uses<br />
(average) farmer technology (technology 1) due<br />
to credit constraints. Income is clearly highest<br />
for the large farmer, who also uses most biocides<br />
<strong>and</strong> nitrogen fertilizer. Income of the small<br />
farmer is 16% higher than for the medium<br />
farmer, whereas the small farmer uses more than<br />
four times as much nitrogen fertilizer <strong>and</strong> almost<br />
three times as much biocides. This difference is<br />
explained by the use of the high input technology<br />
on irrigated l<strong>and</strong> <strong>and</strong> the more sustainable<br />
improved technology on dryl<strong>and</strong>.<br />
The first policy scenario simulates the removal<br />
of all credit constraints, which potentially leads<br />
not only to an increase in income but also in the<br />
use of agrochemicals. Only the large <strong>and</strong> the<br />
small farmer were credit constrained in the base<br />
run. The impact of increased credit availability is<br />
low for the large farmer, but high for the small<br />
farmer, who uses the additional credit to<br />
substitute high-input technology for average<br />
farmer technology. This results in an increase of<br />
income by 15%, <strong>and</strong> of nitrogen <strong>and</strong> biocide use<br />
by 23% <strong>and</strong> 19%, respectively.<br />
At present, there is little off-farm employment,<br />
which makes sustainable, labor-intensive<br />
production technologies relatively attractive.<br />
This could change in the future. Simulations<br />
show that the unlimited availability of off-farm<br />
employment would lead to an increase in income<br />
of 12-16% for all farmers, but to limited changes<br />
in the sustainability of agricultural production<br />
except for the medium farmer, who increases his<br />
biocide use by 40%.<br />
Finally, we evaluated two price-change scenarios<br />
to assess the potential of increasing agricultural<br />
sustainability through input price policies. The<br />
large farmer is affected most by this policy.<br />
Changing biocide prices is most effective: a 10%<br />
increase in biocide prices results in a 7%<br />
decrease in both the use of fertilizers <strong>and</strong><br />
biocides, while the same increase in fertilizers<br />
results in a decrease of 5% for both types of<br />
inputs. The other changes are minor.<br />
Table 2: Results of the household simulations for<br />
Dingras municipality<br />
Income Rice N fertilizer<br />
Biocides<br />
(10 3 (ton)<br />
(kg a.i.)<br />
Pesos) (kg)<br />
Medium farm-well drained<br />
Base run 301.1 5.0 47 187<br />
Unlimited credit 0% 0% 0% 0%<br />
Unlimited off-farm<br />
employment<br />
10% increase in fertilizer<br />
prices<br />
10% increase in biocide<br />
prices<br />
Large farm<br />
16% 0% 0% 40%<br />
0% 0% 0% 0%<br />
0% 0% 0% 0%<br />
Base run 656.0 11.8 376 1013<br />
Unlimited credit 3% 6% 3% 1%<br />
Unlimited off-farm<br />
employment<br />
10% increase in fertilizer<br />
prices<br />
10% increase in biocide<br />
prices<br />
Small irrigated farm<br />
12% -8% 1% -1%<br />
-1% -4% -5% -5%<br />
-2% -7% -7% -7%<br />
Base run 348.4 5.1 203 540<br />
Unlimited credit 15% 41% 23% 19%<br />
Unlimited off-farm<br />
employment<br />
10% increase in fertilizer<br />
prices<br />
10% increase in biocide<br />
prices<br />
6% -1% 0% 0%<br />
-1% -1% -1% -1%<br />
-2% -3% -2% -2%<br />
5. DISCUSSION AND OUTLOOK<br />
There is a need for tremendous agricultural<br />
productivity increases in the countries with high<br />
population densities in E <strong>and</strong> SE Asia, such as<br />
the Philippines. Such increase can only be<br />
achieved sustainably by judicious use of external<br />
inputs <strong>and</strong> natural resources, <strong>and</strong> supportive<br />
policies. Model results for the province show the<br />
high potential of the new technologies to<br />
improve income <strong>and</strong> sustainability at the same<br />
time, implicitly suggesting that investment in<br />
agricultural research <strong>and</strong> extension is the answer.<br />
However, there are some constraints to<br />
optimizing resource use efficiency (such as<br />
limited access to credit), which cannot be<br />
analysed using the regional model. Here, FHM<br />
comes in for analysing the constraints <strong>and</strong><br />
627
possibilities to adoption of sustainable<br />
technologies at the farm level.<br />
Analysis of the effectiveness of different policy<br />
instruments (public investment to improve<br />
access to credit, off-farm employment <strong>and</strong> price<br />
instruments) in contributing to regional<br />
development goals was performed for<br />
representative farm types in Dingras<br />
municipality. The policy instruments resulted in<br />
variable trade-offs between income, rice<br />
production <strong>and</strong> ecological sustainability of<br />
agricultural production depending on farm type.<br />
Thus, in addition to regional level explorations,<br />
FHM analysis shows that increased availability<br />
of off-farm employment <strong>and</strong> capital are likely to<br />
hamper adoption of sustainable technologies,<br />
while increased prices of agrochemicals appear<br />
quite effective in stimulating adoption of these<br />
technologies at a only limited decrease in<br />
household income. Hence, FHM <strong>and</strong> regional<br />
modelling complement each other. When<br />
developed <strong>and</strong> applied in close interaction with<br />
stakeholders, such multi-scale modelling<br />
approach can provide valuable information for<br />
policy development in relation to natural<br />
resource management (van Ittersum et al., 2004).<br />
Such process is currently underway in the case<br />
study regions of the IRMLA project.<br />
6. REFERENCES<br />
Bouman, B.A.M., Jansen, H.G.P., Schipper, R.A.,<br />
Hengsdijk, H., Nieuwenhuyse, A. (Eds), Tools for<br />
l<strong>and</strong> use analysis on different scales. With<br />
examples from Costa Rica. Kluwer Academic<br />
Publishers, Dordrecht, The Netherl<strong>and</strong>s, 2000.<br />
C<strong>and</strong>ler, W., Fortuny-Amat, J., MacCarl, B., The<br />
Potential Role of Multilevel Programming in<br />
Agricultural Economics. Amer. J. Agr. Econ 63,<br />
521-531,1981.<br />
De Wit, C.T., Van Keulen, H., Seligman, N.G.,<br />
Spharim, I., Application of interactive multiple<br />
goal programming techniques for analysis <strong>and</strong><br />
planning of regional agricultural development.<br />
Agricultural Systems 26, 211-230, 1988.<br />
Kruseman, G. & Bade, J., Agrarian Policies for<br />
Sustainable L<strong>and</strong> Use: Bio-economic modelling to<br />
assess the effectiveness of policy instruments.<br />
Agricultural Systems 58, 465-481, 1998.<br />
Kuyvenhoven, A., J. Bouma <strong>and</strong> H. Van Keulen (Eds),<br />
Policy analysis for sustainable l<strong>and</strong> use <strong>and</strong> food<br />
security. Agricultural Systems, Special Issue 58,<br />
281-481,1998.<br />
Laborte AG, Roetter R, Hoanh CT, The l<strong>and</strong> use<br />
planning <strong>and</strong> analysis system of the systems<br />
research network in Asia, paper presented at<br />
MODSS’99, 1999 Brisbane, Australia Aug 1-6. In:<br />
P.A. Lawrence & Robinson, eds. Proceedings<br />
Report QNRM02143, Queensl<strong>and</strong> Department of<br />
Natural Resources <strong>and</strong> Mines. Australia, 2002.<br />
Lu, C.H., Van Ittersum, M.K., Rabbinge , R., A<br />
scenario exploration of strategic l<strong>and</strong> use options<br />
for the Loess Plateau in Northern China.<br />
Agricultural Systems, 79, 145-170, 2004.<br />
Lucas, M.P., P<strong>and</strong>ey, S., Vilano, R.A., Culanay, D.R.,<br />
Obien, S.R., Characterization <strong>and</strong> economic<br />
analysis of intensive cropping systems in rainfed<br />
lowl<strong>and</strong>s of Ilocos Norte. Experimental Agriculture<br />
35, 1-14, 1999.<br />
Ponsioen, T.C., Van Oort, P.A.J., Laborte, A.G.,<br />
Roetter, R.P., TechnoGIN: A Technical coefficient<br />
generator for cropping systems in Ilocos Norte<br />
Province. Quantitative Approaches to Systems<br />
Analysis No. 26, Wageningen University, The<br />
Netherl<strong>and</strong>s, 2004.<br />
Roetter, R.P., Van Keulen, H., Van Laar, H.H. (Eds.),<br />
Synthesis of methodology development <strong>and</strong> case<br />
studies. SysNet Research Paper Series No. 3,<br />
<strong>International</strong> Rice Research Institute, Los Baños,<br />
Philippines, 94 pp, 2000.<br />
Roetter, R.P. <strong>and</strong> J. Wolf, Second Annual Scientific<br />
Report. IRMLA Project Report No. 4,<br />
Wageningen UR, The Netherl<strong>and</strong>s, 102 pp.<br />
(+Annexes), 2003.<br />
Roetter, R.P., Hoanh, C.T., Laborte, A.G., Van<br />
Keulen, H., Van Ittersum, M.K., Dreiser, C., Van<br />
Diepen, C.A., De Ridder, N., Van Laar, H.H.,<br />
Integration of Systems Network (SysNet) tools for<br />
regional l<strong>and</strong> use scenario analysis in Asia.<br />
<strong>Environmental</strong> <strong>Modelling</strong> & <strong>Software</strong> (in press),<br />
2004.<br />
Stoorvogel, J.J. & Antle, J.M., Regional l<strong>and</strong> use<br />
analysis : the development of operational tools.<br />
Agricultural Systems 70, 623-640, 2001.<br />
Van Ittersum, M.K., Rabbinge, R., Van Latensteijn,<br />
H.C., Exploratory l<strong>and</strong> use studies <strong>and</strong> their role in<br />
strategic policy making. Agricultural Systems 58,<br />
309-330, 1998.<br />
Van Ittersum, M.K. Roetter, R.P., Van Keulen, H., De<br />
Ridder, N., Hoanh, C.T., Laborte, A.G., Aggarwal,<br />
P.K., Ismail, A.B., Tawang, A., A systems<br />
network (SysNet) approach for interactively<br />
evaluating strategic l<strong>and</strong> use options at subnational<br />
scale in South <strong>and</strong> South-east Asia. L<strong>and</strong><br />
Use Policy 21, 101-113, 2004.<br />
Walker, D.H., Decision support, learning <strong>and</strong> rural<br />
resource management. Agricultural Systems 73,<br />
113-127, 2002.<br />
628
The role of Multi-Criteria Decision Analysis in a<br />
DEcision Support sYstem for REhabilitation of<br />
contaminated sites (the DESYRE software)<br />
C. Carlon a , S. Giove b , P. Agostini c , A. Critto a,d , <strong>and</strong> A. Marcomini d<br />
a<br />
Consorzio Venezia Ricerche, Marghera-Venice, Italy (cla.cvr@vegapark.ve.it)<br />
b<br />
Dept of Applied Mathematics, University of Venice, Italy (sgiove@unive.it)<br />
c<br />
Interdepartmental Centre IDEAS, University of Venice, Italy (paola.agostini@unive.it)<br />
d<br />
Dept of <strong>Environmental</strong> Sciences, University of Venice, Italy (marcom@unive.it)<br />
Abstract: The rehabilitation of contaminated sites involves several considerations in terms of environmental,<br />
technological <strong>and</strong> socio-economic aspects. A decision support system becomes therefore necessary in order to<br />
manage problem complexity <strong>and</strong> to define effective rehabilitation interventions. DESYRE (Decision Support<br />
sYstem for Rehabilitation of contaminated sites) is a software system which integrates risk assessment with<br />
socio-economic analysis <strong>and</strong> technological assessment in order to provide decision-makers with different<br />
remediation scenarios to be evaluated. The structure of the system allows a subsequent analysis, from socioeconomic<br />
analysis <strong>and</strong> site characterization, to risk assessment before <strong>and</strong> after remediation technologies<br />
selection, until the definition of remediation scenarios. The system integrates several analytical tools, such as<br />
geostatistics, Fuzzy logic, risk assessment <strong>and</strong> geographical information systems (GIS). The present paper<br />
focuses on the role of the Multi-Criteria Decision Analysis (MCDA), which represents the core of the DSS. In<br />
the DESYRE framework, MCDA is applied for the definition of the pool of the suitable remediation<br />
technologies. The analytic hierarchy process is applied to rank technologies <strong>and</strong> develop alternative<br />
remediation scenarios. The scenarios are described by a set of indices which can be aggregated by decision<br />
makers to rank alternative options. Future research developments suggest the MCDA application also for the<br />
evaluation of the remediation scenarios by different stakeholders, in a Group Decision Making (GDM)<br />
context.<br />
Keywords: Contaminated sites, Multi-criteria Decision Analysis, Risk Assessment, Remediation technologies,<br />
Decision Support Systems, GIS.<br />
1. INTRODUCTION<br />
Decision-making on environmental issues is often a<br />
process characterized by complexity, uncertainty,<br />
multiple <strong>and</strong> sometimes conflicting management<br />
objectives, as well as integration of numerous <strong>and</strong><br />
different data types.<br />
In the case of contaminated sites, additional<br />
problematic aspects arise: heterogeneity of site<br />
contamination, high costs for remediation<br />
activities, presence of multiple stakeholders,<br />
crucial integration of risk assessment with socioeconomic<br />
<strong>and</strong> technological valuations.<br />
Contaminated sites management requires to<br />
perform specialistic judgements <strong>and</strong> to translate<br />
them in alternatives of rehabilitation interventions.<br />
Therefore, a decision support system is needed in<br />
order to provide coherent <strong>and</strong> realistic management<br />
scenarios, by linking all the interested issues in a<br />
transparent <strong>and</strong> reproducible way.<br />
629
Several attempts in this direction have been made<br />
in recent years (Bardos et al., 2001). At the same<br />
time, Geographical Information Systems (GIS)<br />
have been recognised as valid <strong>and</strong> effective<br />
instruments in supporting decision-making, due to<br />
the possibility of spatial elaborations of different<br />
information (Eastman et al., 1993).<br />
The DESYRE (Decision Support System for<br />
rehabilitation of contaminated sites) software is a<br />
successful example of these efforts of integrating<br />
risk analysis procedures with socio-economic<br />
evaluations <strong>and</strong> technology assessment in a<br />
supporting GIS-based tool for decision-making. In<br />
a first phase, DESYRE provides assessment<br />
modules for a multi-disciplinary team of experts,<br />
composed of risk assessors, socio-economists <strong>and</strong><br />
technology engineers. The experts are supported<br />
along all the analytical steps, from site<br />
characterization to socio-economic valuation <strong>and</strong><br />
technologies ranking, until the definition of<br />
different remediation scenarios to be presented to<br />
the final decision-makers. In the last phase,<br />
DESYRE provides decision makers with tools for<br />
comparing alternative remediation scenarios.<br />
During one of the analytical phases, DESYRE<br />
implements a Multi-Criteria Decision Analysis<br />
(MCDA). The importance of the application of a<br />
MCDA procedure in a decision process has been<br />
widely recognised. In fact, given the high level of<br />
complexity of environmental decision problems,<br />
MCDA represents a fundamental help for the<br />
decision maker in the presence of possibly<br />
conflicting targets (Munda, 1994). Moreover, this<br />
methodology assures great transparency to the<br />
whole decisional process.<br />
The paper will first present the general<br />
organization of the DESYRE software. The second<br />
part will be focused on the Multi-Criteria Decision<br />
Analysis, encompassing what has been already<br />
implemented <strong>and</strong> future developments.<br />
2. DESYRE GENERAL PRESENTATION<br />
DESYRE software is the result of a three-year<br />
project funded by the Italian Ministry for<br />
University <strong>and</strong> Scientific Research. DESYRE main<br />
objective is the creation <strong>and</strong> comparison of<br />
different remediation scenarios in terms of residual<br />
risk, technological choices <strong>and</strong> socio-economic<br />
benefits.<br />
Addressing the cited main objective, DESYRE<br />
software allows the user to perform subsequent<br />
analysis of site characterisation, risk assessment,<br />
technologies selection <strong>and</strong> scenarios construction.<br />
By applying different <strong>and</strong> specific tools provided<br />
in the system, such as Fuzzy logic <strong>and</strong> Multicriteria<br />
Decision Analysis, geostatistics methods<br />
<strong>and</strong> GIS tools, the system allows to investigate<br />
each aspect of the contaminated site remediation<br />
process in a stepwise procedure. Application is<br />
facilitated by the user-friendly interface <strong>and</strong> the<br />
clear guideline, where all parameters, assumptions<br />
<strong>and</strong> data, in addition to final results, are clearly<br />
visualized <strong>and</strong> highlighted. For instance, it is<br />
possible to create chemical databases <strong>and</strong> GISbased<br />
risk maps. Moreover, transparency for the<br />
decision process is guaranteed by the use of Multicriteria<br />
Decision Analysis methodologies <strong>and</strong><br />
effective analysis is assured by the active role of<br />
experts within the whole process.<br />
As stated above, DESYRE organisation through a<br />
stepwise procedure assists the analytical <strong>and</strong><br />
decisional process development (Figure 1).<br />
First stage is the socio-economic analysis. Since<br />
remediation objectives are strictly related to socioeconomic<br />
drivers <strong>and</strong> constraints, the provided<br />
analysis, based on a Fuzzy expert system, allows to<br />
select the most attractive l<strong>and</strong> use to be considered<br />
in the risk assessment (Facchinetti et al., 2003).<br />
Subsequently, a site characterization is performed,<br />
which provides the analysis of spatial distribution<br />
of contaminants by using geostatistical methods (in<br />
particular, variography <strong>and</strong> Kriging), in order to<br />
define areas of homogenous contamination.<br />
Analysis is carried out both on soil <strong>and</strong><br />
groundwater.<br />
Risk assessment (US-EPA, 1989; ASTM, 1998) is<br />
then performed twice during the process, before<br />
<strong>and</strong> after the simulation of treatment performances<br />
of selected technologies. The first assessment<br />
provides a site zoning according to risk levels; the<br />
second one allows to evaluate the residual risk<br />
after the application of a technological set. In both<br />
cases, exposure pathways (such as ingestion,<br />
dermal contact or inhalation) as well as interested<br />
receptors like humans or waters are considered. Six<br />
classes of chemicals are identified, related to<br />
technological treatment capabilities:<br />
- non halogenated volatile organic compounds,<br />
- halogenated volatile organic compounds,<br />
- non halogenated semivolatile organic<br />
compounds,<br />
- halogenated semi-volatile organic compounds,<br />
- fuels<br />
- inorganics.<br />
Between the first <strong>and</strong> the second risk assessment<br />
phase, the selection of technologies is proposed to<br />
the user. A first collection of suitable technologies<br />
is made considering the general characteristics of<br />
the whole site <strong>and</strong> contaminants of concern. Then,<br />
a more focused selection is performed by assigning<br />
specific technologies to identified risk areas.<br />
Technologies are ranked on the basis of keycriteria<br />
such as cost, time, efficiency, reliability,<br />
public acceptability. For the technologies ranking,<br />
630
a Multi-criteria Decision Analysis is performed.<br />
Experts are called in this phase to provide<br />
knowledge <strong>and</strong> expertise by assigning technologies<br />
to each homogenous risk area, with the option of<br />
creating several sets of technologies applied<br />
differently in space <strong>and</strong> time. The system allows<br />
then to run the risk assessment procedure again in<br />
order to evaluate risk reduction <strong>and</strong> residual risk<br />
levels.<br />
Finally, on the basis of results from previous<br />
investigations, remediation scenarios are identified.<br />
In this decisional phase, alternative scenarios can<br />
be compared on the basis of a set of indices<br />
derived from the technological selection, the risk<br />
assessment procedure <strong>and</strong> the socio-economic<br />
analysis. A comparative matrix is therefore<br />
presented to the stakeholders involved in the<br />
decision-making process.<br />
DESYRE software has been tested in two areas<br />
(450 <strong>and</strong> 43 hectares wide, respectively) of the<br />
mega-site of Porto Marghera, Italy. Porto<br />
Marghera is a 3,600 hectares industrial (mainly<br />
petro-chemical) zone, located at the border of the<br />
Venice lagoon. Common contaminants are PAHs,<br />
amines, dioxins, halogenated organic compounds<br />
<strong>and</strong> metals (such as As, Cd, Pb, Zn).<br />
According to the Italian Law 426/98, Porto<br />
Marghera is the largest contaminated site of<br />
national interest in Italy. For this reason, it has<br />
represented a challenging opportunity for the<br />
application of the DESYRE software, which aims<br />
at the integration of risk assessment with socioeconomic<br />
<strong>and</strong> technological valuations in large<br />
contaminated sites with conflicting social <strong>and</strong><br />
economic drivers <strong>and</strong> pressures.<br />
Moreover, the application has been instrumental<br />
for evaluating technological indications provided<br />
within the Master Plan developed for the same<br />
area, <strong>and</strong> for comparing them with the software<br />
elaborations.<br />
The scenarios provided by the application of the<br />
DESYRE software have highlighted the usefulness<br />
of DESYRE in supporting decision making<br />
through the constitutions of different alternatives.<br />
Proposed solutions have been characterized by<br />
great variety in the different considered<br />
parameters, from costs to time duration, technology<br />
performance <strong>and</strong> environmental impacts. DESYRE<br />
outlined advantages <strong>and</strong> limitations of each option,<br />
as a necessary basis for the creation of a broad<br />
consensus on a final choice among multiple<br />
stakeholders.<br />
The verification of the DESYRE DSS through the<br />
experimental application to the two case-studies is<br />
the object of a specific manuscript in preparation,<br />
to be submitted for publication.<br />
4. MULTI-CRITERIA DECISION ANALYSIS<br />
WITHIN THE DESYRE FRAMEWORK<br />
The Multi Criteria Decision Analysis (MCDA) is<br />
the core of the DSS. In the considered MCDA<br />
problem, the decision scenario is represented by a<br />
two-entries table, where each row corresponds to<br />
an alternative, <strong>and</strong> each column to a criterion. Each<br />
alternative can then be represented by the vector of<br />
its criteria values. After having discharged the<br />
dominated alternatives (the ones whose criteria<br />
values are equal or worst than other alternatives)<br />
the decision maker needs to solve the problem of<br />
selecting the best alternatives, or of ranking all the<br />
remaining ones. Various approaches exist in the<br />
literature on MCDA problems to solve possible<br />
conflicts among criteria. A feasible classification<br />
consists of multiple attribute utility theory,<br />
outranking, or interactive methods, but even other<br />
classification are possible, for instance, based on<br />
compensatory <strong>and</strong> non-compensatory methods<br />
(Chen et al., 1992; Vincke, 1992). Anywise, a<br />
complete scenario of the available methods is<br />
beyond the purpose of this contribution. Among<br />
the most appealing ones, we limit to quote the<br />
PROMETHEE (Brans et al., 1986), the TOPSIS<br />
(Chen, 2000), the AHP (Saaty, 1980), the<br />
ELECTRE (Roy, 1989), the rough set approach,<br />
the aggregation operators (like the family of OWA<br />
introduced by Yager (1988)), <strong>and</strong> the Fuzzy<br />
ranking methods. One of the most diffuse approach<br />
is the simple additive weight method (SAW), in<br />
which all the criteria values are weighted by a<br />
suitable real number measuring the importance of<br />
the weights, <strong>and</strong> subsequently added. Although its<br />
simplicity, the SAW method is characterised by a<br />
serious drawback: no interaction among the<br />
attributes is admitted, since the preferential<br />
independence axiom is required. Moreover, some<br />
difficulty exists for the weights assignment. To this<br />
purpose, some methods like AHP can be suggested<br />
(Saaty, 1980), <strong>and</strong> other tools such as Fuzzy logic,<br />
the Choquet integral (Murofushi <strong>and</strong> Sugeno,<br />
1989), <strong>and</strong> the theory of aggregation operators,<br />
(Chen et al., 1992). Another characterisation<br />
regards the question if the problem needs to be<br />
approached by a single decision maker, or by a<br />
group of Experts or decision makers. In the latter<br />
case, we speak about Group Decision Theory, for<br />
which the consensus measures are an important<br />
item, showing how much the group of decision<br />
makers agree or disagree about the alternative<br />
ranking (Carlsson et al., 1992).<br />
In the DESYRE framework, the MCDA tools are<br />
applied first in the definition of the pool of suitable<br />
technologies <strong>and</strong> second for the comparison of<br />
alternative scenarios.<br />
631
With concern to the technologies selection, a score<br />
is assigned to each technology on the basis of keycriteria,<br />
like cost, development time, efficiency (or<br />
performance), reliability, flexibility, public<br />
acceptability <strong>and</strong> so on. This method is applied to<br />
each set of technologies chosen by the Expert, <strong>and</strong><br />
it is similar to the SMART approach (Lootsma,<br />
1997; Lootsma, 2000; Triantaphyllou, 1999).<br />
Afterward, to each criterion a weight is assigned,<br />
with the aim of enhancing its relative importance.<br />
The weight assignment phase is performed using<br />
the AHP approach, both with the Saaty method <strong>and</strong><br />
with the multiplicative approach. In both the<br />
approaches, the decision maker (in this case the<br />
environmental engineering expert) is asked for a<br />
comparison among each couple of criteria. In the<br />
Saaty method, those values belong to the integer<br />
scale (1, 2, 3, 4, 5, 6, 7, 8, 9) <strong>and</strong> their reciprocal,<br />
while in the multiplicative approach the<br />
geometrical scale is used. The two approaches also<br />
differ for other items, like the computation of the<br />
aggregated judgments. The interested reader can<br />
refer to the quoted references. Moreover, some<br />
algorithms are applied to compute the consistency<br />
of the judgments (inconsistency appears where the<br />
transitivity property between three judgments is<br />
violated), thus helping the decision maker to revise<br />
its judgments about the comparison between a<br />
couple of criteria. Note that the AHP approach is<br />
applied only to the weights assignment phase, <strong>and</strong><br />
not to the scoring of the criteria values. This in<br />
mainly due to the fact that the number of possible<br />
alternatives (decontamination technologies) can be<br />
quite large, <strong>and</strong> the number of required<br />
comparisons could be unacceptable. Then, we<br />
preferred to directly evaluate the alternatives by the<br />
Expert on the basis of some lower lever subcriterion.<br />
The second application of the MCDA within the<br />
DESYRE software regards the definition of the<br />
remediation scenarios to support decision-makers<br />
(in this case stakeholders for the remediation of the<br />
site). A remediation scenario is characterised by:<br />
- the rehabilitation of the contaminated l<strong>and</strong> to a<br />
specific l<strong>and</strong> use, which is related to socioeconomic<br />
benefits,<br />
- a set of remediation technologies, which are<br />
related to costs, time duration of interventions,<br />
performance reliabilities <strong>and</strong> environmental<br />
impacts,<br />
- the reduction of contaminant concentrations in<br />
soils <strong>and</strong> groundwater, which is related to a<br />
reduction of human health risk.<br />
Therefore, a set of indices can be used to describe<br />
each scenario encompassing socioeconomic,<br />
technological <strong>and</strong> human health risk. These indices<br />
are automatically calculated by the socioeconomic,<br />
risk assessment <strong>and</strong> technological<br />
modules during the creation of each alternative<br />
scenario (Figure 1). In the case of the risk <strong>and</strong> the<br />
technological indices, they are derived by<br />
aggregation of micro-indices. The technological<br />
index is based on three micro-indices: (1) rank of<br />
technologies applied, (2) number of technologies<br />
(a low number is preferred), (3) performance of the<br />
overall technological set. The risk index is based<br />
on four micro-indices: (1) residual risk in terms of<br />
magnitude, (2) residual risk in terms of surface, (3)<br />
risk benefit, (4) risk uncertainties. A detailed<br />
description of indices is the object of a specific<br />
manuscript in preparation.<br />
The aggregation of micro-indices into the<br />
technological <strong>and</strong> risk indices are performed by<br />
experts through SAW methodologies. While SAW<br />
has been adopted in this preliminary stage, the use<br />
of OWA (Ordered Weighted Average) operators,<br />
or Choquet integrals or AHP can be evaluated in<br />
future implementations.<br />
All indices are normalised in a 0-1 scale <strong>and</strong><br />
provided to decision makers for ranking alternative<br />
scenarios. The optimal scenario is always a<br />
compromise among socio-economic <strong>and</strong> risk<br />
benefits, technological reliability, times <strong>and</strong> costs<br />
<strong>and</strong> environmental impacts. The set of indices can<br />
be aggregated into one index according to SAW<br />
methodologies in order to rank alternative<br />
scenarios. Decision makers can adjust the weight<br />
of each index according to their preferences: e.g.,<br />
local authorities can be more interested in socioeconomic<br />
benefits, while l<strong>and</strong> owners can be<br />
concerned with remediation costs <strong>and</strong> environment<br />
association may push for a minimum risk<br />
objective. DESYRE outpoints advantages <strong>and</strong><br />
disadvantages of each scenario: e.g., a drawback of<br />
heavy remediation interventions is that<br />
environmental impacts may overcome the benefit<br />
of risk reduction. Indices can be displayed by<br />
means of histograms such as the ones showed in<br />
figure 2.<br />
632
Socio - economic analysis<br />
Socio - economic Index<br />
Risk Assessment (pre)<br />
Tech . rank<br />
microindex<br />
Tech . number<br />
microindex<br />
SAW<br />
Technological Index<br />
Tech . performance<br />
microindex<br />
Cost Index<br />
Technologies selection<br />
Residual risk magnitude<br />
microindex<br />
Residual risk surface<br />
microindex<br />
Time Duration Index<br />
<strong>Environmental</strong> Impact<br />
Index<br />
SAW<br />
Scenarios<br />
ranking<br />
Risk Assessment (post)<br />
Residual risk uncertainty<br />
microindex<br />
SAW<br />
Risk Index<br />
Risk net reduction<br />
microindex<br />
Figure 1. Derivation of indices <strong>and</strong> micro-indices in the assessment phase <strong>and</strong> their aggregation for ranking<br />
alternative scenarios.<br />
Comparison of scenarios by macro-indexes<br />
Aggregated indexes comparison<br />
1<br />
0.9<br />
Index<br />
value<br />
0.8<br />
1<br />
index<br />
value<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
Socio-economic index<br />
Technologic index<br />
Risk index<br />
Env. Impact index<br />
Cost<br />
Duration<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
Scenario 1 Scenario 2 Scenario 3<br />
0.2<br />
0.1<br />
0<br />
Scenario 1 Scenario 2 Scenario 3<br />
Figure 2. Example of histograms for comparing hypothetical scenarios based on socio-economic, technologic<br />
<strong>and</strong> risk indices <strong>and</strong> ranking scenarios based on the aggregated index.<br />
5. FUTURE DEVELOPMENTS<br />
Multi-criteria Decision Analysis has demonstrated<br />
its high potential in supporting experts in the<br />
definition of the pool of remediation technologies<br />
within the DESYRE framework.<br />
Future research developments can be planned, in<br />
order to further implement the MCDA analysis at<br />
the decisional phase. We propose to consider the<br />
presence of multiple decision makers, thus each<br />
possible remediation scenario will be evaluated in<br />
a Group Decision Making (GDM) context, using<br />
the multiplicative AHP (Ramanathan et al., 1994;<br />
Van Den Honert et al., 1996). Some consensus<br />
measures can be easily introduced in this<br />
framework, <strong>and</strong> the degree of importance of each<br />
Expert can be automatically defined by the<br />
procedure itself using a devoted session. In this<br />
phase, all the Experts assign a pair-wise<br />
comparison of all the couples of criteria, <strong>and</strong><br />
subsequently the AHP methodology provides the<br />
computation of the importance weights. Moreover,<br />
an interactive phase helps the Expert to insert or<br />
delete some alternatives during the process.<br />
Another future implementation regards the<br />
possibility for decision makers to evaluate<br />
rehabilitation scenarios on the basis of additional<br />
items beyond that provided by the experts, since<br />
633
also political <strong>and</strong> economic impact factors need to<br />
be considered. At methodological level this<br />
objective does not pose substantial differences<br />
from what proposed so far. Finally, we intend to<br />
implement a modified version of the classical<br />
TOPSIS method, the so-called BB-TOPSIS<br />
(Rebai, 1993) since both numerical <strong>and</strong> logical<br />
data appear in the criteria definition, <strong>and</strong> this<br />
(simple <strong>and</strong> intuitive) method does not require a<br />
common measurement scale, nor the use of<br />
transforming functions (see the quoted references).<br />
6. ACKNOWLEDGEMENTS<br />
DESYRE was funded by the Italian Ministry for<br />
University <strong>and</strong> Scientific Research. A special<br />
acknowledge to Stefano Silvoni <strong>and</strong> Stefano<br />
Soriani (University of Venice), Manuela Samiolo<br />
<strong>and</strong> Stefano Foramiti (CVR), Gianni Antonio<br />
Petruzzelli (CNR) <strong>and</strong> Emiliano Ramieri <strong>and</strong> Luca<br />
Dentone (Thetis SpA) for their contribution in the<br />
development <strong>and</strong> testing of the DESYRE software.<br />
7. REFERENCES<br />
ASTM-American Society for Testing <strong>and</strong><br />
Materials, St<strong>and</strong>ard Provisional Guide for<br />
Risk-Based Corrective Action, Final report,<br />
PS 104-98, ASTM, Philadelphia, 1998.<br />
Bardos, R.P., Mariotti, C., Marot, F., Sullivan, T.,<br />
Framework for Decision Support used in<br />
Contaminated L<strong>and</strong> Management in Europe<br />
<strong>and</strong> North America, L<strong>and</strong> Contaminated &<br />
Reclamation, 9, 149-163, 2001.<br />
Brans, J.P., Vincke, Ph., Mareschal, B., How to<br />
select <strong>and</strong> how to rank projects: the<br />
Promethee method, European Journal of<br />
Operational Research 24, pp. 228-238, 1986.<br />
Carlsson, C., Ehrenber,g D., Eklund, P., Fedrizzi<br />
M., Gustafsson, P., Lindholm, P.,<br />
Merkuryeva G., Riissanen T., Ventre A.G.S.,<br />
Consensus in distribuited soft environments,<br />
European Journal of Operational Research,<br />
61, 165-185, 1992.<br />
Chen, C.T., Extensions of the TOPSIS for group<br />
decision-making under fuzzy environment,<br />
Fuzzy Sets <strong>and</strong> Systems, 114, 1-9, 2000.<br />
Chen, S.J., Hwang C.-L., Hwang F.P., Fuzzy<br />
multiple attribute decision making, Springer-<br />
Verlag, Berlin, 1992.<br />
Eastman, J.R., Kyem, P.A.K., Toledano, J., Jin,<br />
W., GIS <strong>and</strong> Decision Making, UNITAR<br />
Explorations in GIS Technology - Vol.4,<br />
United Nations Institute for Training <strong>and</strong><br />
Research, Geneva, 1993.<br />
Facchinetti, G., Mastroleo, G., Mannino, I.,<br />
Soriani, S.. A fuzzy expert system for the<br />
socio-economic analysis of contaminated<br />
sites. Extended Proceedings ACS 2003, 10th<br />
<strong>International</strong> Multi-Conference on Advanced<br />
Computer Systems, 22-24 October,<br />
Miedzyzdroje, Pol<strong>and</strong>, 2003.<br />
Lootsma F.A., Fuzzy logic for planning <strong>and</strong><br />
decision making, Kluwer Academic<br />
Publishers, 1997.<br />
Lootsma F.A., Multi-criteria Decision Analysis via<br />
ratio <strong>and</strong> difference judgement, Kluwer<br />
Academic Publishers, 2000.<br />
Munda, G., Fuzzy information in multi-criteria<br />
environmental evaluation models, Report<br />
EUR 15602EN of Joint Research Center of<br />
European Commission, 1994.<br />
Murofushi, T., Sugeno M., An interpretation of<br />
fuzzy measures <strong>and</strong> the Choquet integral as<br />
an integral with respect to a fuzzy measure,<br />
Fuzzy Sets <strong>and</strong> Systems 29, pp. 201-227,<br />
1989.<br />
Ramanathan, R., Ganesh, L.S., Group preference<br />
aggregation methods employed in AHP: an<br />
evaluation <strong>and</strong> an intrinsic process for<br />
deriving members’s weightages, European<br />
Journal of Operational Research, 79, 249-<br />
265, 1994.<br />
Rebai, A., BB-TOPSIS: a bag based technique for<br />
order preference by similarities to ideal<br />
solution, Fuzzy Sets <strong>and</strong> Systems 60, pp.143-<br />
162, 1993.<br />
Roy, B., The outranking approach <strong>and</strong> the<br />
fundations of Electre methods, Document du<br />
Lamsade 53, Universitè Paris-Dauphinc,<br />
1989.<br />
Saaty, T.L., The analytic hierarchy process,<br />
McGraw Hill, New York, 1980.<br />
Triantaphyllou E., Multi-criteria decision making<br />
methods: a comparative study, Kluwer<br />
Academic Publishers, 1999.<br />
US-EPA, Risk assessment guidance for superfund<br />
Vol 1, Human health evaluation manual,<br />
Final Report, EPA/540//1-89/002,<br />
Washington DC, 1989.<br />
Van Den Honert, R.C., Lootsma F.A., Group<br />
preference aggregation in the multiplicative<br />
AHP. The model of the group decision<br />
process <strong>and</strong> Pareto optimality, European<br />
Journal of Operational Research, 96, 363-<br />
370, 1996.<br />
Vincke, P., Multi-criteria decision aid, Wiley &<br />
Sons, Baffins Lane, Chichester, Engl<strong>and</strong>,<br />
1992.<br />
Yager, R.R., On ordered weighted averaging<br />
aggregation operators in multi-criteria<br />
decision making, IEEE Trans. Systems Man<br />
Cybernet., 18, 183-190, 1988.<br />
634
ICT Requirements for an ‘evolutionary’ development of<br />
WFD compliant River Basin Management Plans<br />
M. Blind a<br />
a<br />
RIZA Institute for Inl<strong>and</strong> Water Management <strong>and</strong> Waste Water Treatment, The Netherl<strong>and</strong>s,<br />
m.blind@riza.rws.minvenw.nl<br />
Abstract: The Water Framework Directive (WFD) poses an immense challenge to integrated water management<br />
in Europe. Aiming at a “good ecological status” of all water resources in 2015, integrated river basin<br />
management plans need to be in place by 2009, <strong>and</strong> need to be broadly supported by stakeholders. Cost<br />
effective programmes of measures must be put in place to meet the objective of “good ecological status”.<br />
These measures reach beyond the direct water domain <strong>and</strong> touch on fields such as spatial planning, public<br />
participation <strong>and</strong> socio-economics. Much information <strong>and</strong> knowledge needs to be available to create these<br />
plans. Information & Communication Technology (ICT) tools, such as computational models, are potentially<br />
very helpful in designing river basin management plans (rbmp-s). Based on a vision on an evolutionary<br />
development of Decision Support Systems in a collaborative planning process, this paper elaborates some<br />
key requirements for modelling <strong>and</strong> ICT. The EU-funded cluster of projects “CatchMod”, including the concerted<br />
action “Harmoni-CA”, is discussed from the viewpoint of these requirements.<br />
Keywords: Water Framework directive, ICT, modelling, collaborative planing<br />
1. INTRODUCTION<br />
In 2000 the European Parliament <strong>and</strong> Council<br />
passed the ambitious directive 2000/60/EC establishing<br />
a framework for Community action in the<br />
field of water policy, known as the Water Framework<br />
Directive (WFD). The key objective of this<br />
law is to achieve ‘good ecological status of<br />
Europe’s water resources by 2015.<br />
A key aspect of the WFD is integration. The WFD<br />
aims at integrating amongst others: i) environmental<br />
objectives, combining quality, ecological<br />
<strong>and</strong> quantity objectives; ii) all water resources,<br />
combining fresh surface water <strong>and</strong> groundwater<br />
bodies, wetl<strong>and</strong>s, coastal water resources at the<br />
river basin scale; iii) all water uses, functions <strong>and</strong><br />
values into a common policy framework; iv) disciplines,<br />
analyses <strong>and</strong> expertise, combining hydrology,<br />
hydraulics, ecology, chemistry, soil sciences,<br />
technology engineering <strong>and</strong> economics; v) stakeholders<br />
<strong>and</strong> the civil society in decision making,<br />
etc [1].<br />
To achieve the WFD’s objectives a number of<br />
activities need to be carried out, leading to an Integrated<br />
River Basin Management Plan (RBMP) in<br />
2009 (figure 1). Programmes of measures, leading<br />
to the desired state of the water resources need to<br />
be set. Measures may range from straightforward<br />
actions such as sewage treatment to financial incentives<br />
such as emission taxes for industry. The<br />
programme of measures should achieve the objectives<br />
in a cost-effective manner.<br />
The WFD requires involvement of stakeholder,<br />
such as the environmental or agricultural interest<br />
groups, <strong>and</strong> the general public. Besides informing<br />
these stakeholders through consultation, active<br />
participation in developing objectives <strong>and</strong> programmes<br />
of measures is strived for. Reaching the<br />
overall objective thus will be a collaborative effort<br />
in which tailored information is of uttermost importance.<br />
All this requires a huge effort in the design of<br />
River Basin Management Plans: effects of measures<br />
need to be evaluated in an integrated context,<br />
involving all the aspects mentioned above, <strong>and</strong><br />
635
Submit interim report on<br />
the implementation to the<br />
EC (Art. 15)<br />
Revised<br />
overview of<br />
significant water<br />
issues<br />
Update<br />
RBMP<br />
Assess current<br />
status, analyse<br />
preliminary gaps<br />
(Art. 5-8)<br />
Set up<br />
environmental<br />
objectives (Art. 4)<br />
Implement the<br />
programme of<br />
measures for<br />
RBD<br />
Evaluate the first <strong>and</strong><br />
prepare the second<br />
period.<br />
2012<br />
2004<br />
2013 2015<br />
2006<br />
Establish<br />
monitoring<br />
programmes<br />
(Art. 8)<br />
Public<br />
Participation<br />
(Art. 14)<br />
2009<br />
Gap analysis<br />
Develop River Basin<br />
Management Plan<br />
(RBMP) (Art.13-25,<br />
App.VII)<br />
information needs to be accessible in the way that<br />
all different types of stakeholders achieve a common<br />
underst<strong>and</strong>ing of the problems, objectives <strong>and</strong><br />
solutions.<br />
This paper aims at identifying some major ICT <strong>and</strong><br />
modelling issues from the perspective of collaborative<br />
planning <strong>and</strong> the limitations of integrated<br />
modelling systems. It builds on the author’s view<br />
on an evolutionary development of Decision Support<br />
Systems during the WFD implementation.<br />
The paper provides global insight in research carried<br />
out in the EC supported catchment-modelling<br />
cluster (CatchMod).<br />
2. THE WFD COLLABORATIVE<br />
PLANNING PROCESS<br />
A simple schematisation of the collaborative planning<br />
process is presented in figure 2. In general,<br />
such a process consists of a closely interlinked<br />
‘planning process’ path <strong>and</strong> an ‘information delivering’<br />
path. The planning process part consists of<br />
‘start’, ‘problem definition’, ‘solution selection’<br />
<strong>and</strong> ‘implementation’. Of course, this is a simplified<br />
representation: in a real-life situation the process<br />
is more continuous as new problems emerge,<br />
redefinition of problems is required <strong>and</strong>/or new<br />
solutions become available during the planning<br />
process (etc.). At all stages of the planning process<br />
stakeholders need to be involved. Furthermore, all<br />
steps require information that is tailored to the<br />
needs of the collaborative process, thus towards<br />
Set up the programme<br />
of measures for RDB<br />
(Art. 11)<br />
Figure 1: Visualisation of the time line of the WFD <strong>and</strong> its required activities <strong>and</strong> deliverables [1].<br />
different types of stakeholders with different levels<br />
of knowledge. In complex situations such as integrated<br />
river basin planning, this means that very<br />
specific, expert knowledge needs to be integrated<br />
<strong>and</strong> translated into underst<strong>and</strong>able information for<br />
non-specialists, amongst whom the general public.<br />
To achieve this, multi-disciplinary teams of scientists<br />
need to collaborate <strong>and</strong> integrate different<br />
sources of information <strong>and</strong> knowledge, such as<br />
observation data, results of state assessment models<br />
<strong>and</strong> predictive models.<br />
3. DECISION SUPPORT TOOLS AND<br />
THEIR LIMITATIONS FOR THE<br />
WFD IMPLEMENTATION<br />
In the past many tools have been developed to<br />
support water management. Especially in hydrology,<br />
computer modelling has been carried out for<br />
several decades. Integration of different domains<br />
in water modelling has lead to a broad availability<br />
of frequently large, advanced modelling suites.<br />
Specialists generally use such models <strong>and</strong> modelling<br />
suites.<br />
In the last decade systems have been developed<br />
that integrate more <strong>and</strong> more domains, <strong>and</strong> can be<br />
used by non-specialist users. These developments<br />
often supported planning processes similar to the<br />
process described in the previous section. The systems<br />
emerged from linking existing models, expert<br />
rules, databases <strong>and</strong> other tools <strong>and</strong> developing the<br />
636
Planning process<br />
(e.g. WFD)<br />
Start<br />
Interaction &<br />
communication<br />
Information:<br />
Data & <strong>Modelling</strong><br />
Selection of building blocks<br />
Problem definition<br />
Solution<br />
Data system building<br />
Tuning the system<br />
Application<br />
Quality assurance<br />
Implementation<br />
Adaptation<br />
Figure 2: Simplified representation of the participatory planning process<br />
means to calculate or visualize (pre-calculated)<br />
effects of different management options (measures).<br />
In such systems additions such as multicriteria<br />
tools <strong>and</strong> cost effectiveness analysis tools<br />
provide means to achieve some optimisation during<br />
the selection of solutions. Though individual<br />
domain models also support decision-making the<br />
author reserves the word Decision Support System<br />
(DSS) for such integrated systems.<br />
In the eyes of the author, the problem of the current<br />
DSS-s is that they have been developed for<br />
quite specific issues <strong>and</strong> do not cover the broadness<br />
of the WFD. The information path is often<br />
detached from the planning path, meaning that the<br />
information path is not closely following the dem<strong>and</strong>s<br />
from the planning path. Though the systems<br />
are of high quality, adapting them to new situations,<br />
e.g. changing <strong>and</strong> adding models, changing<br />
the geographic area they apply to, etc, is far from<br />
easy. It often requires much effort by both model<br />
& tool specialists <strong>and</strong> software developers. It is a<br />
major challenge for DSS developers (software<br />
developers <strong>and</strong> modelling specialists) to match the<br />
dem<strong>and</strong>s <strong>and</strong> the speed of the planning process.<br />
The DSS development nevertheless has the distinct<br />
purpose of focussing discussion <strong>and</strong> gaining (mutual)<br />
underst<strong>and</strong>ing of all participants in a collaborative<br />
planning process. A DSS is therefore frequently<br />
called a Discussion Support System as<br />
opposed to Decision Support System.<br />
A relatively new branch of software tools supporting<br />
the collaborative process are gaming <strong>and</strong> learning<br />
tools. These tools are extremely useful when<br />
aiming at common underst<strong>and</strong>ing between different<br />
stakeholders, each with their own backgrounds<br />
<strong>and</strong> interests. Gaming tools can be used to get<br />
common underst<strong>and</strong>ing of problems in river basins,<br />
but also to achieve underst<strong>and</strong>ing of (conflicting)<br />
interests, effects of behaviour patterns <strong>and</strong><br />
decision making processes. They are thus very<br />
usefull in the early stages of a planning process.<br />
Gaming tools share similar problems as DSS-s –<br />
adapting them to new situations, issues <strong>and</strong> river<br />
basins is quite elaborate.<br />
Today, we find ourselves facing the immense challenge<br />
to integrate more domains in water management,<br />
include all different types of stakeholders<br />
<strong>and</strong> develop cost effective programmes of measures<br />
as to meet the objectives of the Water Framework<br />
Directive. We need to find effective combinations<br />
of technical measures <strong>and</strong> socio-economic<br />
incentives to achieve good ecological status of<br />
Europe’s water resources. Responsible River Basin<br />
Authorities all over Europe are working on the<br />
current requirements of the WFD, such as lists of<br />
protected areas, assessments of states <strong>and</strong> human<br />
impacts, setting preliminary objectives, etc. Soon,<br />
their focus of attention will move towards setting<br />
up monitoring programmes <strong>and</strong> programmes of<br />
measures.<br />
4. MODELS AND TOOLS IN THE WFD<br />
AND ITS GUIDANCE DOCUMENTS<br />
Models <strong>and</strong> tools are addressed at several points in<br />
both the legal WFD document <strong>and</strong> several guidance<br />
documents. It would be too far-reaching to<br />
provide a full overview within the scope of this<br />
paper, but for illustrative purposes some information<br />
is presented in this section.<br />
In the legal document it states under section 1.3.<br />
Establishment of type-specific reference conditions<br />
for surface water body types it states ‘Typespecific<br />
biological reference conditions based on<br />
modelling may be derived using either predictive<br />
models or hindcasting methods.’ In paragraph 1.5:<br />
Assessment of Impact it states: ‘Member States<br />
shall use the information collected above, <strong>and</strong> any<br />
other relevant information including existing environmental<br />
monitoring data, to carry out an assessment<br />
of the likelihood that surface waters bod-<br />
637
ies within the river basin district will fail to meet<br />
the environmental quality objectives set for the<br />
bodies under Article 4. Member States may utilise<br />
modelling techniques to assist in such an assessment.’<br />
The guidance document on the planning<br />
process [1] explicitly states that it does not include<br />
‘Specific methodologies for the planning process:<br />
hydrologic modelling, decision support systems,<br />
etc.’ It does however acknowledge the usefulness<br />
of models: ‘Although the systems approach to water<br />
resources planning is not restricted to mathematical<br />
modelling, models do exemplify the approach.<br />
They can represent in a fairly structured<br />
<strong>and</strong> ordered manner the important interdependencies<br />
<strong>and</strong> interactions among the various control<br />
structures <strong>and</strong> users of a water resources system.<br />
Models permit an evaluation of the economic <strong>and</strong><br />
physical consequences of alternative engineering<br />
structures, of various operating <strong>and</strong> allocating<br />
policies, <strong>and</strong> of different assumptions regarding<br />
future flows, technology, costs, <strong>and</strong> social <strong>and</strong><br />
legal requirements. Although this systems methodology<br />
cannot define the best objectives or assumptions,<br />
it can identify good decisions, given those<br />
objectives <strong>and</strong> assumptions.’ And ‘Thus, the role<br />
models may be viewed as that of tools from which<br />
to derive answers to well-posed questions about<br />
the performance or behaviour of the system that is<br />
being planned. However, because of the dynamics<br />
of the planning process, it may happen that the<br />
answers derived from the models will suggest that<br />
the original questions were not well conceived <strong>and</strong><br />
need to be reformulated. Hence, the role of models<br />
is iterative. They are used to produce information<br />
that may be fed forward to aid in decision-making<br />
(i.e., plan formulation). With equal value, they<br />
may produce information that is fed back to aid in<br />
redefining the problem.’<br />
The guidance ‘Public Participation in relation to<br />
the Water Framework Directive’ [2] <strong>and</strong> the guidance<br />
on impacts <strong>and</strong> pressures [3] provide numerous<br />
examples of the use of tools, mainly in its annex.<br />
The Guidance Document on Implementing<br />
the GIS Elements of the WFD [4] specifically<br />
deals with information systems <strong>and</strong> provides a<br />
data-model. It does not concern modeling <strong>and</strong> decision<br />
support systems.<br />
Though the above does not provide a full analysis<br />
on ICT <strong>and</strong> modeling of the WFD <strong>and</strong> its guidances,<br />
it leads to the conclusion that only little<br />
guidance is provide on ICT <strong>and</strong> model requirements.<br />
This is supported by an analysis of WFD<br />
guidance documents on data aspects carried out by<br />
Blind <strong>and</strong> de Blois [5]. Though the WFD legal text<br />
<strong>and</strong> the guidances do not oblige the use of models<br />
<strong>and</strong> tools, the benefit of modeling <strong>and</strong> the use of<br />
tools is clearly recognized in the different guidances.<br />
What the factual role of models <strong>and</strong> tools<br />
will be during the WFD implementation is however<br />
yet unclear. This poses a problem for the development<br />
of Decision Support Systems.<br />
5. THE AUTHOR’S VISION<br />
In the author’s view, it is necessary to integrate<br />
science, ICT technology, communication means in<br />
a very flexible, but scientifically sound manner to<br />
efficiently <strong>and</strong> effectively develop sound WFD<br />
compliant River Basin Management Plans. It is<br />
necessary to bring the DSS development much<br />
closer to the WFD planning process. In early<br />
stages of this process simple models <strong>and</strong> tools are<br />
required which allow the participants of a collaborative<br />
process to gain insight in the water system<br />
<strong>and</strong> achieve some common underst<strong>and</strong>ing <strong>and</strong> a<br />
basis for discussion. Based on the discussions on<br />
pressures, impacts, responses, measures [etc.]<br />
more detailed tools need to be incorporated. Since<br />
the time to develop the WFD compliant River Basin<br />
Management Plan is limited adding more detail<br />
to the DSS must be a simple <strong>and</strong> quick process. As<br />
the collaborative planning process progresses, the<br />
DSS will need to gradually evolve towards a dedicated<br />
DSS for the river basin at h<strong>and</strong>.<br />
The key characteristic of this vision lies in the<br />
‘evolution’ of the DSS. The author firmly believes<br />
that developing a single DSS from the beginning,<br />
either at a European, National or basin scale is not<br />
the way forward, since:<br />
1) Such a system will need to incorporate all<br />
domains, problems <strong>and</strong> possible measures, for<br />
all different stages of the planning process,<br />
making it too large to develop from scratch,<br />
use it, <strong>and</strong> maintain it into the future. Differences<br />
in data-availability will add to this problem:<br />
a single system will need to work with<br />
low <strong>and</strong> high data-availability.<br />
2) Each river basin has its own characteristics<br />
<strong>and</strong> problems, which requires local knowledge<br />
to be incorporated <strong>and</strong> dedicated development.<br />
The characteristics <strong>and</strong> problems are not<br />
limited to the natural sciences, but also include<br />
cultural, institutional <strong>and</strong> linguistic issues.<br />
3) Scientific robustness, validity <strong>and</strong> transparency<br />
will be difficult, if not impossible to<br />
achieve.<br />
4) Support from the research community will be<br />
lacking. On one h<strong>and</strong> because new insights<br />
will be difficult to incorporate, reducing the<br />
motivation of scientists to contribute, <strong>and</strong> on<br />
638
the other h<strong>and</strong> because due to the fact that the<br />
selected tools <strong>and</strong> models will exclude alternative<br />
models <strong>and</strong> tools, practically excluding<br />
science <strong>and</strong> scientific debate from the DSS<br />
<strong>and</strong> widening the gap between research <strong>and</strong><br />
practical application. The system becomes an<br />
‘institution’ itself.<br />
5) Creating a single system will (possibly) lead<br />
to exclusiveness, reducing competition, interfering<br />
markets <strong>and</strong> rendering past investments<br />
obsolete.<br />
6) …<br />
The main drawback of creating a single system is<br />
however that during the collaborative process unforseen<br />
questions will arise which cannot be supported.<br />
Subsequently, adaptations will be required.<br />
Adapting fully integrated systems is usually a<br />
complex endeavour given the complexity of the<br />
interrelations. The single system thus poses the<br />
great danger of being leading to the discussions in<br />
the collaborative process. In the collaborative<br />
process the planning process should lead the development<br />
of the information system.<br />
The author believes that even on a river basin or<br />
national scale it will be very difficult to develop<br />
one system that answers all (yet unknown) questions.<br />
6. ICT, MODEL AND TOOL NEEDS<br />
As concluded in section 4 the WFD <strong>and</strong> its guidance<br />
documents do not provide direct guidance on<br />
particular tools <strong>and</strong> models, but do acknowledge<br />
the benefit of their use. Following from the vision<br />
of the author it is also clear that creating a single<br />
Decision Support System which supports the collaborative<br />
planning process <strong>and</strong> the development<br />
of river basin management plans is (in the author’s<br />
view) not desirable, let alone feasible. The key<br />
ICT <strong>and</strong> modelling requirements should therefore<br />
lie on a more abstract or generic level, which supports<br />
the ‘evolutionary’ development of decision<br />
support systems. The key requirement to achieve<br />
this is a modular approach, in which models, databases<br />
<strong>and</strong> other tools are independent (small)<br />
units. Modularity alone, however, does not result<br />
in the flexibility <strong>and</strong> speed required for the collaborative<br />
process: the modules need a common<br />
interface, which allows information to pass from<br />
one model to another, to tools <strong>and</strong> user interfaces.<br />
Such an interface is required to allow quick linkages<br />
of modules to integrated systems, preferably<br />
without additional programming. The interface<br />
also allows swapping models, for example when<br />
more complex models are required. The st<strong>and</strong>ard<br />
should include the means to underst<strong>and</strong> what data<br />
can be exchanged, either by providing a st<strong>and</strong>ard<br />
data-dictionary or self-descriptive methods (st<strong>and</strong>ard<br />
meta-data dictionary). Currently there is no<br />
broadly accepted interface <strong>and</strong> there are only few<br />
models <strong>and</strong> tools that share the same (IT) interface.<br />
Developing <strong>and</strong> agreeing on an interface<br />
st<strong>and</strong>ard is thus urgently needed.<br />
If such a st<strong>and</strong>ard is developed <strong>and</strong> agreed upon<br />
models <strong>and</strong> tools need to be adapted to comply<br />
with this st<strong>and</strong>ard. The collection of models <strong>and</strong><br />
tools should form a repository of modules, which<br />
can be flexibly linked. Besides obvious modules<br />
such as hydrological, ecological, economical (etc.)<br />
models, the repository must also include tools for<br />
multi criteria analysis, uncertainty analysis, gaming,<br />
etc. With respect to (non-specialist) end-users,<br />
exchanging information <strong>and</strong> data is not limited to<br />
passing numbers – the information must be useful<br />
to the recipients, thus information processing, filtering,<br />
translation of information need to be part of<br />
the repository as well.<br />
In the author’s view models <strong>and</strong> tools are readily<br />
available, <strong>and</strong> many alternatives exist in most scientific<br />
domains. Currently an extensive <strong>and</strong> comprehensive<br />
overview on available tools <strong>and</strong> models<br />
is lacking.<br />
Structuring models <strong>and</strong> tools in a repository will<br />
allow gap analysis, <strong>and</strong> (cost) efficient further developments.<br />
To further support the evolutionary approach to<br />
DSS development guidance is required to select<br />
‘the right tools for the right purpose at the right<br />
time’. This requires that for each model <strong>and</strong> tool<br />
sufficient meta-data is available to determine the<br />
usefulness. Of particular interest is the scientific<br />
soundness of a model or tool when linked with<br />
other tools. This requires scientific research resulting<br />
in practical guidances. Tool <strong>and</strong> model selection<br />
criteria should not be limited to ‘content’: the<br />
quality of the software should also be considered<br />
when integrating different models <strong>and</strong> tools.<br />
Much of the time required to build dedicated decision<br />
support systems lies in the collection of data<br />
<strong>and</strong> populating the models. In modular, integrated<br />
systems using the same base datasets is often a<br />
problem. Though the three-tier approach (user<br />
interfaces, models <strong>and</strong> data-layer) is well known<br />
<strong>and</strong> agreed upon, many (legacy) tools require<br />
dedicated input. Improving this situation can be<br />
obtained through a st<strong>and</strong>ard interface as well. Furthermore<br />
a common (high-level multilingual meta-<br />
) data model is required. Given the anticipated<br />
639
complexity of WFD Decision Support Systems<br />
<strong>and</strong> need for flexibility much more effort is required<br />
to quickly link data <strong>and</strong> models. [Note: One<br />
should be aware that collecting the data for WFD<br />
reporting does not deliver a dataset that is sufficient<br />
for (advanced) modelling! <strong>Modelling</strong> will<br />
require much more detailed data.]<br />
Harmoni-<br />
RiB<br />
Harmoni-<br />
QuA<br />
Clime<br />
TransCat<br />
EuroHarp<br />
The foreseeable complexity of WFD related modelling<br />
<strong>and</strong> Decision Support Systems, the need for<br />
transparency of the collaborative process <strong>and</strong> the<br />
ambition to achieve some comparable quality in<br />
the (development of) River Basin Management<br />
Plans requires guidance <strong>and</strong> tools to develop, use,<br />
<strong>and</strong> record complex integrated systems. Such<br />
methods <strong>and</strong> tools should also support working in<br />
multidisciplinary teams <strong>and</strong> increase the trust in<br />
modelling results by, amongst others, the public.<br />
Finally, one of the key requirements to achieve the<br />
vision of the author is improving the accessibility<br />
of models, tools <strong>and</strong> data. Legal <strong>and</strong> practical barriers<br />
prohibiting quick <strong>and</strong> easy use of tools need<br />
to be resolved, e.g. by harmonized access rights<br />
<strong>and</strong> technologies such as web services. This does<br />
not mean that software should be free of charge.<br />
The above points form the basis need for an evolutionary<br />
approach to WFD Decision Support System<br />
development. Other tools related challenges<br />
are also very important <strong>and</strong> require attention:<br />
• The scientific linkage between freshwater <strong>and</strong><br />
coast <strong>and</strong> sea.<br />
• Integrated uncertainty assessment (data models,<br />
planning process)<br />
• Multilingual support <strong>and</strong> support tools in<br />
transboundary regions<br />
• Integration of earth observation technology<br />
• …<br />
In the view of the author, the issues raised above<br />
are very important for developing the River Basin<br />
Management Plans, but it is certainly not a complete<br />
list of issues.<br />
7. THE CATCHMOD INITIATIVE<br />
The European Commission’s Research Directorate<br />
General supports a number of research projects<br />
<strong>and</strong> a concerted action that focus on supporting the<br />
WFD implementation using computational models<br />
<strong>and</strong> other computational tools. These projects are<br />
clustered in CatchMod, the catchment-modelling<br />
cluster (figure 3, table 1). In the previous sections<br />
the vision of the author <strong>and</strong> subsequent requirements<br />
have been elaborated. In this section the<br />
Temp-<br />
QSim<br />
Harmoni-<br />
CoP<br />
Harmoni-<br />
-CA<br />
Harmon-<br />
IT<br />
Tisza<br />
River<br />
BMW<br />
Figure 3: The CatchMod projects (acronyms).<br />
CatchMod projects are introduced in the light of<br />
the requirements.<br />
The HarmonIT project is developing a st<strong>and</strong>ard<br />
interface for data-exchange. On a meta-level it<br />
defines structures for data description. The BMW<br />
project develops benchmark criteria for models,<br />
facilitating the proper selection. Euroharp compares<br />
a suite of models for nutrient emissions,<br />
which is also helpful for model selection. Many<br />
other projects will research the applicability of<br />
models in different situations; for example, in<br />
TempQSim the specific requirements for water<br />
quality models in temporary waters are researched,<br />
including the aspects of data availability. In Clime,<br />
the linkage between climate change <strong>and</strong> ecology is<br />
under investigation. Databases including uncertainty<br />
information <strong>and</strong> being able to hold many<br />
different types of data from all WFD relevant domains,<br />
<strong>and</strong> methodologies for uncertainty propagation<br />
in integrated modelling are researched in<br />
HarmoniRiB. HarmoniQuA elaborates guidance<br />
on the proper setting up <strong>and</strong> use of integrated<br />
modelling systems. It develops tools, which help<br />
the modellers, both by providing advice <strong>and</strong> structure,<br />
as in providing reporting structures <strong>and</strong><br />
communication facilities to non-modellers. In the<br />
HarmoniCoP project the use of tools for collaborative<br />
planning, including gaming <strong>and</strong> DSS are researched,<br />
leading to guidance on collaborative<br />
planning including these aspects. Transboundary<br />
modelling, data issues, multilingual problems <strong>and</strong><br />
transboundary communications are key points of<br />
attention in the TransCat <strong>and</strong> Tisza River projects.<br />
So all the above projects are in part of the same<br />
cluster, their time-lines limit the possibilities to reuse<br />
each other’s results ‘on the fly’. The concerted<br />
action Harmoni-CA’s task is to facilitate the synthesis,<br />
for example by supporting the benchmarking<br />
of all models using the BMW criteria. Harmoni-CA<br />
should further facilitate <strong>and</strong> synthesize<br />
640
HarmonIT IT Frameworks (2002-2005) Hwww.harmonit.comH<br />
BMW Benchmark Models for the Water Framework Directive (2002-2004)<br />
Hhttp://www.ymparisto.fi/default.asp?node=11687&lan=enH<br />
EUROHARP Towards Harmonised Procedures for Quantification of Catchment Scale Nutrient Losses from<br />
European Catchments (2002-2005) Hwww.euroharp.orgH<br />
CLIME Climate <strong>and</strong> lake impacts in Europe (2003-2005) Hhttp://www.water.hut.fi/climeH<br />
TempQSim Evaluation <strong>and</strong> improvement of water quality models for application to temporary waters in southern<br />
European catchments (2002-2005) Hwww.tempqsim.netH<br />
TISZA RIVER Real-life scale integrated catchment models for supporting water- <strong>and</strong> environmental management<br />
decisions (2002-2004) Hwww.tiszariver.comH<br />
HarmoniCoP Harmonizing Collaborative Planning (2002 -2005) Hwww.harmonicop.infoH<br />
TRANSCAT Integrated water management of transboundary catchments (2003-2006) Hhttp://transcat.isq.pt/H<br />
HarmoniQuA Harmonising Quality Assurance in model based catchment <strong>and</strong> river basin management (2002-<br />
2005) Hwww.harmoniqua.orgH<br />
HarmoniRiB Harmonised techniques <strong>and</strong> representative river basin data for assessment <strong>and</strong> use of uncertainty<br />
information in integrated water management (2002-2006) Hwww.harmoniRIB.comH<br />
Harmoni-CA Concerted action on Harmonised <strong>Modelling</strong> Tools for Integrated Basin Management<br />
Hwww.harmoni-ca.infoH<br />
Table 1: The CatchMod projects<br />
discussions on the use of models <strong>and</strong> tools in general,<br />
the science-policy interface, the modellingmonitoring<br />
relationship <strong>and</strong> develop a broadly<br />
supported overall methodology, in which all methodologies<br />
developed by the scientific community<br />
get a clear place. Harmoni-CA also works on improving<br />
the accessibility of models <strong>and</strong> data. A<br />
communication services centre is set up to facilitate<br />
to improve the linkage between the WFD dem<strong>and</strong><br />
side <strong>and</strong> the supporting side of science <strong>and</strong><br />
technology. It speaks for itself that all CatchMod<br />
projects have many more objectives than described<br />
above. All projects apply a range of models in realife-cases<br />
<strong>and</strong> discuss with end-user groups.<br />
8. SUMMARY & CONCLUSIONS<br />
The difficulty in making the ICT dem<strong>and</strong>s of the<br />
WFD tangible lies in the fact that the WFD legal<br />
text <strong>and</strong> guidance documents do not provide guidance<br />
on the use <strong>and</strong> requirements of models <strong>and</strong><br />
tools. As a result a list of tools <strong>and</strong> tool characteristics<br />
cannot simply be elicited from these documents.<br />
It should be clear that it is not the intention<br />
of the WFD to be a straightjacket, <strong>and</strong> there is<br />
common agreement that the implementation is<br />
requires tailored approaches.<br />
Discussions at the Harmoni-CA conference [6]<br />
between people involved in the implementation<br />
process (WFD managers) <strong>and</strong> scientists / modellers<br />
did not result in a clear-cut view on ICT /<br />
modelling requirements.<br />
Instead of waiting for requests, it is the author’s<br />
view to anticipate the potential need for modelling<br />
<strong>and</strong> Decision Support Systems in the WFD phase<br />
‘development of River Basin Management Plans’.<br />
The modelling <strong>and</strong> ICT world needs to be ready to<br />
deliver quickly, as soon as the questions are<br />
emerging from the planning process. The author<br />
advocates some key requirements which together<br />
form an ‘infrastructure’: a set of basic st<strong>and</strong>ards<br />
<strong>and</strong> guidances which support an ‘evolutionary’<br />
approach of DSS development. The reasoning<br />
originates from the assumption that modelling <strong>and</strong><br />
information will be an important aspect in implementing<br />
the WFD. However, different views on<br />
the necessity <strong>and</strong> use of advanced tools exist, <strong>and</strong><br />
only time will show how much use will be made of<br />
models <strong>and</strong> ICT.<br />
Obviously, a gap remains between the ‘infrastructure’<br />
requirements advocated by the author, <strong>and</strong><br />
practical DSS systems required for implementing<br />
the WFD. This gap will be closed as tangible requirements<br />
for support emerge. If the key requirements<br />
are met, the integrated modelling community<br />
can quickly deliver<br />
The CatchMod Cluster of projects delivers potential<br />
solutions to many of the issues addressed. The<br />
results of the projects will require harmonisation<br />
<strong>and</strong> future support. It is the task of Harmoni-CA to<br />
facilitate both aspects of CatchMod.<br />
CatchMod is ‘just a cluster’ of modelling <strong>and</strong> ICT<br />
related projects <strong>and</strong> represents just a fraction of<br />
research going on in this particular field. In other<br />
EC-research <strong>and</strong> in national projects ICT issues<br />
such as distributed databases, distributed modelling,<br />
metadata st<strong>and</strong>ards <strong>and</strong> web-based applications<br />
are developed. Of course, also issues addressed<br />
by CatchMod projects are addressed in<br />
other projects. Synthesizing available knowledge<br />
must include these initiatives – Harmoni-CA<br />
should facilitate this process.<br />
641
7. REFERENCES<br />
[1] CIS (Common Strategy on the Implementation<br />
of the Water Framework Directive), 2003, Best<br />
practices in river basin planning - Work Package<br />
2:Guidance on the planning process.<br />
[2] CIS (Common Strategy on the Implementation<br />
of the Water Framework Directive), 2003, Guidance<br />
On Public Participation In Relation To The<br />
Water Framework Directive: Active involvement,<br />
Consultation, <strong>and</strong> Public access to information.<br />
[3] CIS (Common Strategy on the Implementation<br />
of the Water Framework Directive), 2003, Analysis<br />
of Pressures <strong>and</strong> Impacts - The Key Implementation<br />
Requirements of the Water Framework Directive.<br />
[4] CIS (Common Strategy on the Implementation<br />
of the Water Framework Directive), 2003, Strategy<br />
Guidance Document on Implementing the GIS<br />
Elements of the WFD (Working Group GIS).<br />
[5] Blind, M. <strong>and</strong> Ch de Blois, 2003,The Water<br />
Framework Directive <strong>and</strong> its guidance documents<br />
– Review of data aspects, In: Harmonised techniques<br />
<strong>and</strong> representative river basin data for assessment<br />
<strong>and</strong> use of uncertainty information in<br />
integrated water management (HarmoniRiB): Requirements<br />
report, Refsgaard, J. (ed.), chapter 5.<br />
[6] Anonymous, 2004, Summary Report: Harmoni-Ca<br />
Forum And Conference, Underst<strong>and</strong>ing<br />
each other’s dem<strong>and</strong> <strong>and</strong> support for implementing<br />
the WFD, Brussels, 18 & 19 February 2004, Results<br />
of sessions (in preparation).<br />
642
DAWN: A platform for evaluating water-pricing policies<br />
using a software agent society<br />
I. N. Athanasiadis ab , P. Vartalas b <strong>and</strong> P. A. Mitkas ab<br />
a Informatics <strong>and</strong> Telematics Institute, Centre for Research <strong>and</strong> Technology Hellas, GR-57001 Thermi, Greece<br />
b Department of Electrical <strong>and</strong> Computer Engineering, Aristotle University of Thessaloniki, GR-54124<br />
Thessaloniki, Greece<br />
ionathan@ee.auth.gr<br />
Abstract: Lately there is a transition in water management: policy makers leave aside traditional methods<br />
focused on additional-supply policies <strong>and</strong> focus on water conservation using dem<strong>and</strong> control methods. Water<br />
Agencies use water-pricing policies as an instrument for controlling residential water dem<strong>and</strong>. However, design<br />
<strong>and</strong> evaluation of a water-pricing policy is a complex task, as economic, social <strong>and</strong> political constraints have<br />
to be incorporated. In order to support policy makers in their tasks, we developed DAWN, a software tool<br />
for evaluating water-pricing policies, implemented as a multi-agent system. DAWN simulates the residential<br />
water dem<strong>and</strong>-supply chain <strong>and</strong> enables the design, creation, modification <strong>and</strong> execution of different scenarios.<br />
<strong>Software</strong> agents behave as water consumers, while econometric <strong>and</strong> social models are incorporated into them<br />
for estimating future consumptions. Scenarios <strong>and</strong> models can be parameterized through a friendly graphical<br />
user interface <strong>and</strong> software agents are instantiated at runtime. DAWN’s main advantage is that it supports<br />
social interaction between consumers, which is activated using agent communication. Thus, variables affecting<br />
water consumption <strong>and</strong> associated with consumer’s social behavior can be included into DAWN scenarios.<br />
In this paper, DAWN’s agent architecture is detailed <strong>and</strong> agent communication using ontologies is discussed.<br />
Focus is given on the econometric <strong>and</strong> social simulation models used for agent reasoning. Finally, the platform<br />
developed is presented along with real-world results of its application at the region of Thessaloniki, Greece.<br />
Keywords: Agent-based simulation; Water-pricing policy; Decision Support Systems; Autonomous Agents<br />
1 INTRODUCTION<br />
1.1 Background<br />
Policy making for water management is in general<br />
a dem<strong>and</strong>ing task. It involves in-depth study of all<br />
factors <strong>and</strong> conditions including water consumption<br />
dem<strong>and</strong>s <strong>and</strong> water assets availability. Even more,<br />
it requires critical ability <strong>and</strong> expertise for identifying<br />
the consequences of a certain policy in effect.<br />
Support to this procedure is provided by scientific<br />
methodologies <strong>and</strong> tools that simulate the water<br />
management cycle. Simulation tools are used to<br />
model the water management dynamics on a certain<br />
system. Their use supports policy-makers to estimate<br />
future effects of a certain policy on the system.<br />
The overall goal of a simulator is not to forecast<br />
the exact state of the modeled system, but to<br />
explore how the system will evolve because of a<br />
specific policy. In this work, we focus on water<br />
management in urban areas. Pricing water in urban<br />
areas is a complicated task, as economic, social<br />
<strong>and</strong> political parameters have to be considered.<br />
Water-pricing policy design involves the investigation<br />
of water-dem<strong>and</strong> <strong>and</strong> its correlation with water<br />
price. In studying the residential water dem<strong>and</strong>,<br />
researchers have utilized a variety of statistical <strong>and</strong><br />
econometric techniques, <strong>and</strong> they have focused on<br />
finding the appropriate dem<strong>and</strong> management policies<br />
that offer incentives in saving water [Mylopoulos<br />
et al., 2004].<br />
In order to support policy makers in such tasks,<br />
we developed DAWN, a software tool for evaluating<br />
water-pricing policies, implemented as a multiagent<br />
system. Policy-makers <strong>and</strong> analysts, the actual<br />
users of DAWN, can specify water-pricing scenarios<br />
<strong>and</strong> evaluate them through a software environment.<br />
In this paper, we present the DAWN platform<br />
<strong>and</strong> demonstrate its capabilities to simulate social<br />
interactions among consumers <strong>and</strong> to explore<br />
how the total water consumption may be affected.<br />
Quantitative results, obtained through DAWN simulations,<br />
can be used by analysts for further evalua-<br />
643
tion of the pricing policies.<br />
1.2 Agent modeling for water management<br />
DAWN employs an agent-based approach to model<br />
the residential water-supply chain. Agent modeling<br />
has been used for water-management simulations.<br />
Consider the FIRMA (Freshwater Integrated<br />
Resource Management with Agent) project, where<br />
agent models are used for the simulation of natural,<br />
hydrological, social <strong>and</strong> economic dimensions<br />
of water resources management at water basins.<br />
[Gilbert, 2003; Moss et al., 2000]. FIRMA is a<br />
decision support tool for the integrated design of<br />
water management. Agents in FIRMA are used<br />
to model all stakeholders involved in a water basin<br />
dem<strong>and</strong>-supply cycle. A second example is the SI-<br />
NUSE project, which employs an agent-based approach<br />
for integrated management of a water table<br />
system. Agent models in SINUSE are used to represent<br />
interactions between a water table <strong>and</strong> its users<br />
while taking into account the social behavior of the<br />
farmers [Feuillette et al., 2003]. In addition, Ducrot<br />
et al. [2002] are working on the NEGOWAT project,<br />
which is expected to deliver a integrated methodology<br />
using both agent models <strong>and</strong> role-playing<br />
games to address conflict resolution <strong>and</strong> negotiation<br />
for sustainable water management at a water catchment<br />
in the Metropolitan Region of Sao Paulo.<br />
The aforementioned agent-based decision support<br />
systems have been successfully applied for<br />
medium-level water management regions, such as<br />
river-basins, water-tables or water catchments. Advancing<br />
on the way earlier research work has dealt<br />
with agent-based decision support systems for water<br />
management, we present DAWN, a software tool<br />
for simulating something different, i.e. the residential<br />
water dem<strong>and</strong>-supply chain, <strong>and</strong> to a local scale.<br />
2 THE DAWN PLATFORM<br />
2.1 Functional requirements<br />
Residential water dem<strong>and</strong>-supply chain involves<br />
two main stakeholders. One is the Water Utility,<br />
which constantly supplies water to an urban area<br />
<strong>and</strong> puts into effect the water pricing policy. The<br />
other is the area residents who consume water <strong>and</strong><br />
pick up its cost. A generic scenario of this system<br />
is:<br />
1. The Water Utility initiates a pricing policy for<br />
managing water dem<strong>and</strong>.<br />
2. The Consumers Society reacts to the selected<br />
pricing policy. Individual consumer reconsider<br />
their own water consumption with respect<br />
to water price, social influence, weather<br />
conditions <strong>and</strong> other fixed parameters.<br />
3. The Water Utility revises its water-pricing<br />
policy in a timely fashion.<br />
The main principle, upon which DAWN was designed,<br />
is that actual consumers interact with each<br />
other. Social activity definitely affects water consumption<br />
behavior. Thus, DAWN’s core is a society<br />
of agents, serving as a sample of water consumers.<br />
This society reacts to the employment of specific<br />
water-pricing policies, simulating the dynamic behavior<br />
of the actual consumers.<br />
Water consumer agents follow an econometric<br />
model for estimating their consumption, while a social<br />
interaction model was incorporated to define<br />
consumer social behavior. Configuration of this hybrid<br />
model is on the discretion of the DAWN user.<br />
Scenario evaluation <strong>and</strong> DAWN’s agent model may<br />
be easily modified by the user, who can define all<br />
parameters required for the simulation through a<br />
GUI.<br />
The main purposes of DAWN are to:<br />
a. Support the evaluation of water-pricing scenarios,<br />
constituting a flexible, easy-to-use<br />
simulation tool.<br />
b. Provide reliable results, to support the<br />
decision-making process.<br />
c. Model the social behavior of water consumers<br />
<strong>and</strong> provide a methodology for incorporating<br />
it into state-of-the-art consumption models.<br />
2.2 DAWN scenarios<br />
The main objective of DAWN is to simulate the<br />
residential water dem<strong>and</strong>-supply chain in order to<br />
facilitate the evaluation of water pricing scenarios.<br />
DAWN users (scientists, analysts, decision-makers)<br />
relying on their expertise <strong>and</strong> the available data need<br />
to follow a certain procedure to realize this objective.<br />
An abstract description of the DAWN simulation<br />
procedure involves the following steps:<br />
1. Data collection <strong>and</strong> scenario design. A user<br />
prepares the simulation scenario by specifying<br />
a set of parameters for the water-pricing<br />
policy <strong>and</strong> the water consumption model. The<br />
scenario is input to DAWN through a GUI.<br />
2. System self–configuration. DAWN processes<br />
the scenario entered by the user. The<br />
644
simulation procedure is initialized <strong>and</strong> autonomous<br />
agents consuming water, within the<br />
society of agents, are instantiated.<br />
3. Scenario simulation. The simulation procedure<br />
is launched. Simulation is performed in<br />
iterative steps. Each step simulates a time interval,<br />
during which water consumption is estimated.<br />
4. Result presentation. While the simulation<br />
runs, the total <strong>and</strong> individual consumption are<br />
presented to the user. When the simulation is<br />
terminated, all results are saved.<br />
5. Evaluation of the scenario results. DAWN<br />
scenario quantitative results are evaluated by<br />
the user. Comparison <strong>and</strong> study of the results<br />
can lead to valuable conclusions.<br />
The aforementioned procedure is schematically represented<br />
in Figure 1.<br />
2.3 DAWN Architecture<br />
The DAWN functionality described in the previous<br />
section has been realized through an agent-based<br />
architecture, shown in Figure 2. All actors in the<br />
residential water dem<strong>and</strong>-supply chain simulated<br />
have been implemented using agents. The Water<br />
Consumers, the Water Utility <strong>and</strong> the Meteorological<br />
Office are represented by autonomous agents,<br />
who undertake the simulation of the water dem<strong>and</strong>supply<br />
chain. In addition, the Simulator Agent (SA)<br />
is utilized to moderate <strong>and</strong> synchronize the simulation<br />
procedure. Simulator Agent is also responsible<br />
for capturing user-defined scenario specifications<br />
through a GUI.<br />
When a new experiment is started by the user, SA<br />
sets up the whole simulation process <strong>and</strong> instantiates<br />
all agents required for the simulation. These<br />
agents include a Water Supplier Agent (WSA), a<br />
Meteorologist Agent (MOA), <strong>and</strong> a set of Water<br />
Consumer Agents (CA). During the instantiation<br />
process, SA appoints to all agents the user-specified<br />
parameters, required to realize their respective models.<br />
Having all agents instantiated, the simulation process<br />
starts. SA facilitates the overall procedure. A<br />
simulation step starts when SA asks from WSA the<br />
total consumption for the respective time-interval.<br />
WSA contacts all CAs, informing each one about<br />
the cost of water it consumed in the previous step.<br />
Each CA utilizes the econometric <strong>and</strong> social models<br />
to estimate its new consumption. Social activity<br />
between CAs is realized via agent messaging. Water<br />
consumption econometric <strong>and</strong> social models are<br />
discussed in the next section.<br />
Each CA reports its water consumption dem<strong>and</strong>s to<br />
the WSA, which calculates the total dem<strong>and</strong>. The<br />
latter is communicated to SA <strong>and</strong> presented to the<br />
user, signing the termination of the simulation cycle.<br />
Each simulation cycle corresponds to a certain time<br />
interval, defined by the user.<br />
MOA is responsible for suppling all CAs with meteorological<br />
conditions. As weather conditions are<br />
communitywide, this information is passed to CAs<br />
through WSA. Note that the MOA intervenes in the<br />
process only if meteorological parameters are used.<br />
The simulation process is repeated for a certain<br />
number of iterations, defined by the user. When<br />
all iterations are performed, the simulation ends. In<br />
DAWN, agent lifecycle is equal to the simulation<br />
time. When the simulation is ended, all agents are<br />
terminated.<br />
Scenario<br />
Specs<br />
Simulator Agent<br />
SA<br />
Simulation<br />
Results<br />
Data<br />
collection<br />
DAWN User<br />
(Scientist, Analyst, Decision-maker)<br />
Scenario<br />
Design<br />
DAWN<br />
(<strong>Software</strong> platform)<br />
Scenario<br />
Model<br />
Agent-based<br />
Simulation<br />
System<br />
self-configuration<br />
Iterative<br />
process<br />
Meteorologist<br />
Agent<br />
Water Supplier<br />
Agent<br />
CA<br />
CA<br />
CA<br />
MOA<br />
WSA<br />
CA<br />
DAWN platform<br />
CA<br />
Scenario<br />
evaluation<br />
Result<br />
presentation<br />
CA<br />
CA<br />
CA<br />
CA<br />
Society of Water<br />
Consumer Agents<br />
Figure 1: Functional flow diagram<br />
Figure 2: DAWN platform architecture<br />
645
3 DAWN MODELS<br />
3.1 Econometric model<br />
Neighbors<br />
}<br />
SightLimit=1<br />
Water dem<strong>and</strong> estimation is usually formed as a<br />
generic model of the form Q = f(P, Z), which relates<br />
water consumption Q to some price measures<br />
P <strong>and</strong> other factors Z [Arbues et al., 2003].<br />
Neighborhood<br />
CA(3,3)<br />
The econometric model, adopted in DAWN, specifies<br />
the water consumption Q for time interval T in<br />
the form:<br />
Figure 3: CA society distributed over a 2-d grid<br />
ln Q T = ∑ i<br />
e i ln(f i [P T −1 ]) + ∑ j<br />
e j ln(f j [Z T ])<br />
where P is the price, Z are other factors influencing<br />
water dem<strong>and</strong>, e is the corresponding elasticity.<br />
This model is highly reconfigurable by the system<br />
user as all variables, functions <strong>and</strong> elasticities can<br />
be specified.<br />
3.2 Social model<br />
In DAWN, the water consumer society is simulated<br />
as a set of Consumer Agents. Each CA represents a<br />
single consumer or a consumer group having common<br />
needs. Communication among CAs simulates<br />
the social interaction among consumers. CAs are<br />
situated on a square grid. Each CA is determined by<br />
its position on the grid. So, a single CA is identified<br />
as CA(x,y), where (x, y) are its coordinates on the<br />
grid. Social interaction between CAs is limited to a<br />
neighborhood, in analogy with the actual social interaction,<br />
which is not communitywide. Therefore,<br />
a CA neighborhood is specified as the square area<br />
on the grid, whose center is the specific CA <strong>and</strong> its<br />
radius is defined by the SightLimit parameter. All<br />
agents residing in the neighborhood are supposed to<br />
be CA’s Neighbors.<br />
As an example, consider a 2-d grid of side equal to<br />
six, shown in Figure 3. Let the SightLimit parameter<br />
be one. For each CA, a neighborhood square of side<br />
three is defined. The neighborhood area of CA(3,3)<br />
is shown in Figure 3. The social model is realized<br />
in the neighborhood area, so CA(3,3) consumption<br />
is affected only by its two neighbors, CA(2,2) <strong>and</strong><br />
CA(2,4). Note that each agent may reside in more<br />
than one neighborhoods. For example, CA(2,2) is<br />
neighbor of both CA(3,3) <strong>and</strong> CA(1,1).<br />
All CAs operate simultaneously in two distinct, yet<br />
cooperative roles. CAs behave as both consumers<br />
<strong>and</strong> neighbors. The neighbor role is analogous with<br />
the actual consumer behavior to influence its neighbors,<br />
through social activities. In the agent field,<br />
this activity is modeled as an agent communication.<br />
Each CA, acting as neighbor, sends an agent message,<br />
containing its influence in the form of a social<br />
weight. On the other h<strong>and</strong>, each CA acting as consumer<br />
is influenced by its neighbors. It receives all<br />
social weights sent by its neighbors, <strong>and</strong> based on<br />
their sum, readjusts its water dem<strong>and</strong>.<br />
3.3 DAWN hybrid model<br />
In DAWN the social model is incorporated into the<br />
generic econometric model, forming a hybrid model<br />
formed as Qd = f(P, S, Z), where S is the social<br />
model factor. The latter is specified as a function of<br />
the social weights communicated.<br />
The water quantity Q consumed by a CA at time<br />
interval T is specified by the function<br />
ln Q T = ∑ e i ln(f i [P T −1 ]) + ∑ e j ln(f j [Z T ])<br />
i<br />
j<br />
+ ∑ e ki ln(f k [ ∑ SW T ])<br />
k<br />
N<br />
where the social model is incorporated as the sum<br />
of all social weights SW, communicated during the<br />
T -th cycle in the neighborhood N. In this way, the<br />
social interaction among neighboring agents is incorporated<br />
into the classic econometric model.<br />
In the real world, consumers exhibit dissimilar behaviors.<br />
Each individual promotes water conservation<br />
practices at a different degree. In a similar way,<br />
changes in water dem<strong>and</strong>, as a reflection of social<br />
influence, are different. Thus, in DAWN, the user<br />
is enabled to specify distinct CA types. Each type<br />
corresponds to a consumer group behaving in a similar<br />
way, thus sharing the same hybrid model. Consumer<br />
Agents of various types enable the user to<br />
simulate nonidentical social behaviors. In this way,<br />
consumers having dissimilar social behaviors can be<br />
represented by distinct consumer type agents.<br />
646
Application<br />
User<br />
Designed<br />
Scenario<br />
Scenario<br />
Consumer Types<br />
Simulation Duration<br />
Community<br />
Population<br />
Grid Dimension<br />
Simulation Step<br />
Pricing Policy<br />
Meteorological Data<br />
Request scenario<br />
simulation<br />
/ Request consumer<br />
status<br />
/ Display results<br />
Launch personal<br />
GUI<br />
/ Failure<br />
Start<br />
simulation step<br />
/ Sends step result<br />
/ Failure<br />
Simulation<br />
Agent<br />
Simulated<br />
Scenario<br />
Failure<br />
Consumers Community<br />
Neighbour<br />
Social<br />
Function<br />
Ask / Sends<br />
social<br />
weights<br />
Consumer<br />
Agent<br />
Dem<strong>and</strong><br />
Curve<br />
Asks / Sends<br />
personal<br />
consumption<br />
Water<br />
Supplier<br />
Agent<br />
Pricing<br />
Policy<br />
Asks / Sends<br />
meteorogical<br />
data<br />
Met Office<br />
Agent<br />
Meteorological<br />
Data<br />
Dem<strong>and</strong> Curve<br />
Elasticities<br />
Parameters<br />
Functions<br />
Pricing Policy<br />
Price Blocks<br />
Stategy<br />
Base year<br />
data<br />
CPI<br />
Meteorological<br />
Data<br />
Temperature<br />
Rainfall<br />
Figure 4: AORML External Agent Diagram<br />
4 DAWN IMPLEMENTATION<br />
Concept<br />
4.1 <strong>Software</strong> design <strong>and</strong> ontology<br />
DAWN model has been implemented using software<br />
agents, which were designed using the GAIA<br />
methodology [Wooldridge et al., 2000]. The software<br />
agent interaction has been specified using<br />
the Agent-Object-Relationship modeling language<br />
(AORML), introduced by Wagner [2003]. In Figure<br />
4, the AORML external agent diagram is depicted.<br />
Each Consumer Agent realizes two distinct<br />
roles, the role of a Water Consumer <strong>and</strong> the role of<br />
a Consumer Neighbor. Each CA, in order to calculate<br />
its own dem<strong>and</strong>, communicates with its neighbors,<br />
asking for their Social Weights, <strong>and</strong> when it is<br />
asked for its social influence replies with the appropriate<br />
weight. Additional functional activities of all<br />
agents involved in DAWN are incorporated, along<br />
with the administrative functions of the platform,<br />
such as GUI updates <strong>and</strong> user inputs.<br />
Agent messages follow a generic ontology developed<br />
using the Protégé-2000 ontology editor [Noy<br />
et al., 2001]. Part of the WDS ontology developed<br />
is shown in Figure 5, containing the concepts of the<br />
system, along with agent Actions <strong>and</strong> Predicates.<br />
The slots of the various concepts have been configured<br />
in order to contain the appropriate information<br />
communicated by the agents. For example, Social<br />
Parameter corresponds to the Parameter concept<br />
having two slots: name, <strong>and</strong> value for describing<br />
name <strong>and</strong> value of social weights.<br />
PriceBlock<br />
limitUp int<br />
limitDown int<br />
no int<br />
price Float<br />
WaterConsumption<br />
quantity Float<br />
AskForWeights<br />
HasMetData<br />
Predicate<br />
Parameter<br />
name String<br />
weight Float<br />
Consumes<br />
TotalConsumption<br />
StepAttr<br />
Id int<br />
MetData<br />
temperature Float<br />
rainfall Float<br />
Start<br />
AgentAction<br />
LaunchGUI<br />
SaveResults<br />
Figure 5: Water Dem<strong>and</strong>-Supply Ontology: Concepts,<br />
Predicates <strong>and</strong> AgentActions<br />
4.2 Implementation details<br />
The DAWN platform was implemented in Java,<br />
while the development <strong>and</strong> utilization of agents<br />
confronts to the FIPA specifications [FIPA, 2002].<br />
JADE platform has been used for agent development<br />
[Bellifemine et al., 2001].<br />
The DAWN user is required neither to have any<br />
programming skills, nor to underst<strong>and</strong> the internal<br />
functionalities of the platform. The advantages<br />
of the implemented system are its fast performance,<br />
user-friendly interface <strong>and</strong> its “open”, easyto-parameterize<br />
implementation.<br />
647
5 DEMONSTRATION<br />
The DAWN platform has been used for the estimation<br />
water dem<strong>and</strong> in the urban area of Thessaloniki,<br />
Greece. The metropolitan area of Thessaloniki<br />
comprises more than one million consumers.<br />
Aggregate data taken from the Municipal Water<br />
Supply <strong>and</strong> Sewerage Utility records have been<br />
used. Econometric models developed by experts<br />
[Mylopoulos et al., 2004; Kolokytha et al., 2002]<br />
have been extended using DAWN. A model with<br />
one hundred consumers groups clustered in four<br />
types has been simulated, evaluating a set of elective<br />
pricing policies. The four consumer agent types<br />
has been induced empirically, based on questionnaire<br />
data, acquired from a field survey on 1,356<br />
households. DAWN hybrid model has been calibrated<br />
for the region of Thessaloniki, based on prior<br />
field studies, covering period 1994-2000. The calibrated<br />
model has been used for evaluating water<br />
pricing scenarios for the period 2004-2010. The<br />
main task was to explore the quantitative results of<br />
implementing an information <strong>and</strong> education policy<br />
in the direction of controlling water dem<strong>and</strong>.<br />
6 CONCLUSIONS<br />
In this paper, we presented DAWN, a water management<br />
decision support system, dedicated to estimate<br />
urban water dem<strong>and</strong>s <strong>and</strong> evaluate waterpricing<br />
policies. DAWN can be used to explore<br />
the community’s response to a water-pricing policy.<br />
It makes a step ahead in the estimation models<br />
applied in the residential water dem<strong>and</strong> sector, as<br />
it considers consumers behavior <strong>and</strong> social interactions.<br />
DAWN usage has been evaluated for assisting<br />
water decision makers to estimate future water dem<strong>and</strong>s<br />
in the region of Thessaloniki. In particular,<br />
DAWN was used to explore social behavior <strong>and</strong> its<br />
connections with water consumption. DAWN has<br />
been welcomed by decision makers <strong>and</strong> the water<br />
experts for underst<strong>and</strong>ing the quantitative implications<br />
of an information <strong>and</strong> education policy.<br />
Future efforts will build on the current framework<br />
to extend its competence. Our intention is to include<br />
other aspects of the urban water management<br />
problem, as the water availability, quality <strong>and</strong> supply<br />
costs, through the development of an adaptive<br />
behavior for WSA.<br />
ACKNOWLEDGMENTS<br />
Authors would like to express their gratitude to<br />
Prof. Y. Mylopoulos, Dr. A. Mentes <strong>and</strong> Ms. D.<br />
Vagiona for their help during DAWN development,<br />
<strong>and</strong> b. the two anonymous reviewers for their valuable<br />
comments.<br />
REFERENCES<br />
Arbues, F., M. A. Garcia-Valinas, <strong>and</strong> R. Martinez-<br />
Espineira. Estimation of residential water dem<strong>and</strong>:<br />
A state-of-the-art review. Journal of<br />
Socio-Economics, 251:1–22, 2003.<br />
Bellifemine, F., A. Poggi, <strong>and</strong> G. Rimassa. Jade:<br />
a FIPA2000 compliant agent development environment.<br />
In Proceedings of the 5th <strong>International</strong><br />
Conference on Autonomous Agents, pages 216–<br />
217, Montreal, Canada, 2001. ACM.<br />
Ducrot, R., M. L. R. Martins, P. Jacobi, <strong>and</strong> B. Reydon.<br />
Water management at the urban fringe in<br />
metropolitan cathment: Example of the sao paolo<br />
upstream cathment (brasil). In Proceedings of<br />
the 5th <strong>International</strong> Eco-City Conference, Shenzhen,<br />
China, 2002.<br />
Feuillette, S., F. Bousquet, <strong>and</strong> P. L. Goulven. SI-<br />
NUSE: A multi-agent model to negotiate water<br />
dem<strong>and</strong> management on a free access water table.<br />
<strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong> <strong>Software</strong>, 18:<br />
413–427, 2003.<br />
FIPA. Agent Management Specification. Doc.<br />
No. SC00023J, Foundation of Physical Intelligent<br />
Agents, Geneva, Switzerl<strong>and</strong>, 2002.<br />
Gilbert, N. The firma project: An overview. University<br />
of Surrey, United Kingdom, Available at<br />
http://firma.cfpm.org, 2003.<br />
Kolokytha, E., Y. Mylopoulos, <strong>and</strong> A. Mentes. Evaluating<br />
dem<strong>and</strong> management aspects of urban water<br />
policy - A field survey in the city of Thessaloniki<br />
- Greece. Urban Water, 4:391–400, 2002.<br />
Moss, S., T. Downing, <strong>and</strong> J. Rouchier. Demonstrating<br />
the role of stakeholder participation: An<br />
agent based social simulation model of water dem<strong>and</strong><br />
policy <strong>and</strong> response. Technical Report<br />
CPM-00-76, Centre for Policy <strong>Modelling</strong>, The<br />
Business School, Manchester Metropolitan University,<br />
2000.<br />
Mylopoulos, Y. A., A. K. Mentes, <strong>and</strong> I. Theodossiou.<br />
Modeling residential water dem<strong>and</strong> using<br />
household data: A cubic approach. Water <strong>International</strong>,<br />
29(1), 2004.<br />
Noy, N. F., M. Sintek, S. Decker, M. Crubezy, R. W.<br />
Fergerson, <strong>and</strong> M. A. Musen. Creating semantic<br />
web contents with protege-2000. IEEE Intelligent<br />
Systems, 16(2):60–71, 2001.<br />
Wagner, G. The Agent–Object–Relationship metamodel:<br />
Towards a unified conceptual view of<br />
state <strong>and</strong> behavior. Information Systems, 28(5):<br />
475–504, 2003.<br />
Wooldridge, M., N. R. Jennings, <strong>and</strong> D. Kinny.<br />
The Gaia methodology for agent-oriented analysis<br />
<strong>and</strong> design. Autonomous Agents <strong>and</strong> Multi-<br />
Agent Systems, 3(3):285–312, 2000.<br />
648
Empirical Evaluation of Decision Support Systems:<br />
Concepts <strong>and</strong> an Example for<br />
Trumpeter Swan Management<br />
Richard S. Sojda<br />
Northern Rocky Mountain Science Center, United States Department of the Interior - Geological Survey, 212<br />
AJM Johnson Hall - Ecology Department, Montana State University,<br />
Bozeman, Montana 59717, USA (sojda@usgs.gov)<br />
Abstract: Decision support systems are often not empirically evaluated, especially the underlying modelling<br />
components. This can be attributed to such systems necessarily being designed to h<strong>and</strong>le complex <strong>and</strong><br />
poorly structured problems <strong>and</strong> decision making. Nonetheless, evaluation is critical <strong>and</strong> should be focused<br />
on empirical testing whenever possible. Verification <strong>and</strong> validation, in combination, comprise such<br />
evaluation. Verification is ensuring that the system is internally complete, coherent, <strong>and</strong> logical from a<br />
modelling <strong>and</strong> programming perspective. Validation is examining whether the system is realistic <strong>and</strong> useful<br />
to the user or decision maker, <strong>and</strong> should answer the question: “Was the system successful at addressing its<br />
intended purpose?” A rich literature exists on verification <strong>and</strong> validation of expert systems <strong>and</strong> other<br />
artificial intelligence methods; however, no single evaluation methodology has emerged as preeminent.<br />
Under some conditions, modelling researchers can test performance against a preselected gold st<strong>and</strong>ard.<br />
Often in natural resource issues, such a st<strong>and</strong>ard does not exist. This is particularly true with near real-time<br />
decision support that is expected to predict <strong>and</strong> guide future scenarios while those scenarios are, in fact,<br />
unfolding. When validation of a complete system is impossible for such reasons, examining major<br />
components can be substituted, recognizing the potential pitfalls. I provide an example of evaluation of a<br />
decision support system for trumpeter swan (Cygnus buccinator) management that I developed using<br />
interacting intelligent agents, expert systems, <strong>and</strong> a queuing model. Predicted swan distributions over a 13<br />
year period were tested against observed numbers. Finding such data sets is key to empirical evaluation.<br />
Keywords: Decision support system; Verification; Validation; Empirical evaluation; Model; Trumpeter swan<br />
1. INTRODUCTION<br />
Decision support systems use a combination of<br />
models, analytical techniques, <strong>and</strong> information<br />
retrieval to help develop <strong>and</strong> evaluate appropriate<br />
alternatives [Adelman 1992; Sprague <strong>and</strong> Carlson<br />
1982]. Because such systems h<strong>and</strong>le complex<br />
<strong>and</strong> poorly structured problems, they are difficult<br />
to empirically evaluate. However, it is still easy<br />
to argue that evaluation of all decision support<br />
systems is important. For example, in the case of<br />
trumpeter swans, there are ecological <strong>and</strong> public<br />
policy reasons that increase the importance of<br />
ensuring that the right system has been built <strong>and</strong><br />
been built correctly. In this paper, I focus on the<br />
modelling components of decision support<br />
systems <strong>and</strong> the integration of those components.<br />
Evaluation of the overall acceptance among<br />
natural resource managers of decision support<br />
systems, or other socioeconomic measures of<br />
their success <strong>and</strong> failure, are important but are<br />
not addressed.<br />
2. DISCERNING DIFFERENCES BETWEEN<br />
VERIFICATION AND VALIDATION<br />
Definitions of verification <strong>and</strong> validation in<br />
relation to computer software <strong>and</strong> modelling<br />
[Fishmann <strong>and</strong> Kiviat 1968; Mihram 1972;<br />
Adrion et al. 1982] have changed little over the<br />
years. These definitions are not absolute, but<br />
their use is becoming more definite over time.<br />
The following are from O’Keefe et al. [1987] <strong>and</strong><br />
were adapted from Boehm [1981]: “Validation<br />
means building the right system. Verification<br />
means building the system right.” These have<br />
been frequently referenced by others [e.g.,<br />
D’Erchia 2001; Mosqueira-Rey <strong>and</strong> Moret-<br />
Bonillo 2000; Plant <strong>and</strong> Gamble 2003; Santos<br />
649
2001]. A combined definition of verification <strong>and</strong><br />
validation of software, provided by Wallace <strong>and</strong><br />
Fujii [1989], was the analysis <strong>and</strong> testing “to<br />
determine that it performs its intended functions<br />
correctly, to ensure that it performs no unintended<br />
functions, <strong>and</strong> to measure its quality <strong>and</strong><br />
reliability.” Verification has been defined<br />
[Adrion et al. 1982] as “demonstration of<br />
consistency, completeness, <strong>and</strong> correctness of the<br />
software.” The simplicity <strong>and</strong> completeness of<br />
Mihram’s [1972] definition of validation in<br />
relation to simulation is attractive: “…the<br />
adequacy of the model as a mimic of the system<br />
which it is intended to represent.” There is a<br />
plethora of discussions about the semantics of<br />
evaluating models, <strong>and</strong> Johnson [2001] provides a<br />
summary related to natural resource management.<br />
My specifications for verification <strong>and</strong> validation<br />
in reference to decision support systems draw<br />
almost entirely from the above authors.<br />
Verification is ensuring that the system is<br />
internally complete, coherent, <strong>and</strong> logical from a<br />
modelling <strong>and</strong> programming perspective. Have<br />
the algorithm, knowledge, <strong>and</strong> other structures<br />
been correctly encoded? Validation is examining<br />
whether the system achieved the project’s stated<br />
purpose related to helping the user(s) reach a<br />
decision(s). Validation of a particular model can<br />
also have the more limited meaning of whether<br />
the model is an adequate representation of the<br />
system it represents. This is sometimes described<br />
as black-box testing: do the inputs result in<br />
correct <strong>and</strong> useful outputs? Whether model or<br />
decision support system is being tested, I agree<br />
with Mihram [1972] that verification must occur<br />
before validation. This avoids the inadvertent<br />
situation where software provides expected<br />
outputs simply via calibration <strong>and</strong> correlation of<br />
input <strong>and</strong> outputs rather than via logical<br />
relationships. I use the term evaluation to<br />
encompass both verification <strong>and</strong> validation, but<br />
distinguish between them when used<br />
independently. I agree with Adelman [1992] that<br />
both should be part of the development process,<br />
<strong>and</strong> evaluators should specifically be part of the<br />
development team to foster iterative<br />
improvements. This is not to ignore the need for<br />
independent verification <strong>and</strong> validation of models<br />
<strong>and</strong> systems to ensure that the development team<br />
does not inadvertently err in their work.<br />
3. POTENTIAL METHODS FOR<br />
EMPIRICAL EVALUATION<br />
3.1 An Overview<br />
Stuth <strong>and</strong> Smith [1993] followed the ideas of<br />
Eason [1988] <strong>and</strong> recommended iterative<br />
prototyping methods for decision support system<br />
development. Verification <strong>and</strong> validation are part<br />
of that iterative process. Verification should be<br />
performed prior to any delivery of a working<br />
system, even if a prototype. General validation<br />
might be done at this stage as well, with detailed<br />
efforts performed later. If one agrees that<br />
software development can be a living process,<br />
then verification <strong>and</strong> validation are part <strong>and</strong><br />
parcel to that process <strong>and</strong> need to continue as<br />
system refinements <strong>and</strong> redeployments continue<br />
[Carter et al. 1992; Stuth <strong>and</strong> Smith 1993].<br />
Sprague <strong>and</strong> Carlson [1982] recommend that<br />
organizations building their first decision support<br />
system recognize that it essentially is a research<br />
activity, <strong>and</strong> that evaluation should center on a<br />
general, “value analysis”. Since then, it has<br />
become imperative that analytic <strong>and</strong> quantitative<br />
rigor be added beyond “soft testimonials”<br />
[Adelman 1991; Adelman 1992; Andriole 1989;<br />
Cohen <strong>and</strong> Howe 1989]. Sensitivity analysis can<br />
be a validation tool, especially for heuristic-based<br />
systems, <strong>and</strong> for systems where few or no test<br />
cases are available for comparison [Bahill 1991;<br />
O’Keefe et al. 1987]. Whenever validation is<br />
conducted, it is important to recognize to where,<br />
in space <strong>and</strong> time, inferences can be drawn from<br />
the validation data set. Another issue is the need<br />
to show not only how well a system performs, but<br />
also that it can avoid a catastrophic<br />
recommendation [Rushby 1988]. This is<br />
important in many natural resource venues<br />
because of the great concern for irretrievable <strong>and</strong><br />
long term ecological changes.<br />
It is my sense that validation is often the more<br />
neglected part of evaluation, so I will focus there.<br />
However, I do not wish to slight verification as it<br />
is critical to build decision support systems based<br />
on sound cause-effect relationships <strong>and</strong> not on<br />
poorly understood relationships between input<br />
<strong>and</strong> output.<br />
3.2 Analogous Concepts From Artificial<br />
Intelligence<br />
Successful implementation of decision support<br />
<strong>and</strong> expert systems hinges on incorporating three<br />
evaluation procedures [Adelman 1992]: (1)<br />
examining the logical consistency of system<br />
algorithms (verification), (2) empirically testing<br />
the predictive accuracy of the system (validation),<br />
<strong>and</strong> (3) documenting user satisfaction.<br />
Verification <strong>and</strong> validation of knowledge-based<br />
<strong>and</strong> other decision support systems are known to<br />
be more problematic than in general modelling<br />
for many reasons [Gupta 1991]. A few<br />
difficulties in verifying multiagent systems<br />
[O’Leary 2001] are noteworthy, such as rule<br />
650
conflict, circularity, non-used or unreachable<br />
antecedents, <strong>and</strong> agent isolation. Plus, not only is<br />
it important for a system to h<strong>and</strong>le common cases,<br />
it ought to be able to deal with extreme events.<br />
This latter ability is one characteristic often only<br />
found with human experts. However, extreme<br />
events are not only common in, but often drive,<br />
ecological systems.<br />
Wallace <strong>and</strong> Fujii [1989] provide a matrix of 41<br />
techniques <strong>and</strong> tools that can be applied to 10<br />
software verification <strong>and</strong> validation issues.<br />
Cohen <strong>and</strong> Howe [1989] take a slightly different<br />
approach specific to artificial intelligence<br />
methods, <strong>and</strong> they, too, discuss evaluation from<br />
the perspective of the software development life<br />
cycle. They emphasize empirical studies for such<br />
evaluation, whether focusing on verification or<br />
validation. For testing knowledge-based systems,<br />
Murrell <strong>and</strong> Plant [1997] provide a categorization<br />
of 145 automated techniques.<br />
3.3 Alternative Validation Methods<br />
3.3.1 Gold St<strong>and</strong>ard<br />
Under some conditions, modelling researchers<br />
can test performance against a preselected gold<br />
st<strong>and</strong>ard. Mosqueira-Rey <strong>and</strong> Moret-Bonillo<br />
[2000] describe this for intelligent systems as<br />
having test cases with known, prior outcomes.<br />
Virvou <strong>and</strong> Kabassi [2004] actually had such a set<br />
of cases based on expert opinion that they used<br />
for testing an intelligent graphical user interface.<br />
Often in natural resource issues, such a st<strong>and</strong>ard<br />
does not exist. This is particularly true with near<br />
real-time decision support that is expected to<br />
predict <strong>and</strong> guide future scenarios while those<br />
scenarios are, in fact, unfolding. Although this<br />
approach is theoretically desirable, I am not aware<br />
of an actual implementation in an environmental<br />
decision support system. This is not surprising in<br />
a domain where problems tend to be ill-defined<br />
<strong>and</strong> the associated knowledge uncertain.<br />
3.3.2 Real-time <strong>and</strong> Historic Data Sets<br />
In an ideal world, one could construct a decision<br />
support system <strong>and</strong> test its performance against<br />
actual scenarios as they unfold. This is not often<br />
possible because implementation of systems may<br />
need to be immediate. One alternative is to build<br />
the system using data, information, <strong>and</strong><br />
knowledge from one set of situations <strong>and</strong> validate<br />
using an independent set, as done for crop yields<br />
[Priya <strong>and</strong> Shibasaki 2001], for a bass<br />
bioenergetics model [Rice <strong>and</strong> Cochran 1984],<br />
<strong>and</strong> for timber harvest [Wang <strong>and</strong> LeDoux 2003].<br />
Prior versus post testing is another example of<br />
this, <strong>and</strong> a decision support system for credit<br />
management was so validated by Kanungo et al.<br />
[2001]. When a data-driven model is a<br />
significant part of the decision support system,<br />
sometimes the data can be r<strong>and</strong>omly separated<br />
into two parts, one for model development <strong>and</strong><br />
one for validation. Pretzch et al. [2002] illustrate<br />
this using an extensive data set with a forest<br />
management simulator. Haberl<strong>and</strong>t et al. [2002]<br />
also took this approach for water quality<br />
assessments in river basins. A third option, when<br />
the decision support system is not data-based but<br />
rather knowledge-based, is to empirically evaluate<br />
predictions (outputs) from the system against a<br />
historic data set. This does assume that the logic<br />
underlying the system is constant over time. An<br />
example of this latter case is more fully developed<br />
in Section 4. (See tests 1 <strong>and</strong> 3A in Table1.)<br />
3.3.3 Panel of Experts<br />
It is sometimes possible to test performance<br />
against an independent panel of experts [O’Keefe<br />
et al. 1987]. This is a relatively common<br />
technique in the field of artificial intelligence <strong>and</strong><br />
recent examples include multiagent web mining<br />
[Chau et al. 2003] <strong>and</strong> graphical user interface<br />
development [Virvou <strong>and</strong> Kabassi 2004]. Two<br />
concerns must be addressed, however. First, the<br />
panel of experts needed for such an evaluation<br />
must not be connected to system development.<br />
To do so would be so confounding that no<br />
reasonable experimental design would be feasible.<br />
Second, one of the basic tenets of using decision<br />
support systems for complex issues is that such<br />
questions can be beyond the capability of single<br />
persons to conceptualize <strong>and</strong> solve [Bol<strong>and</strong> et al.<br />
1992; Brehmer 1991].<br />
3.3.4 Sensitivity Analysis<br />
Sensitivity analysis is often more important in<br />
model validation than decision support system<br />
evaluation. This stems from the typical decision<br />
support system being highly complex, <strong>and</strong> it<br />
being difficult to isolate individual inputs, or<br />
small enough groups of inputs, to perform<br />
sensitivity analysis. Plus, some sort of gold<br />
st<strong>and</strong>ard or data set is still needed with which to<br />
work. (See test 7A in Table1.)<br />
3.3.5 Component Testing<br />
Sometimes it is not possible to validate a<br />
complete system, but one can test individual<br />
components. It is not uncommon, for example, to<br />
have multiple expert systems embedded in one<br />
651
decision support system. When one validates<br />
each component separately, however, the<br />
interactions of the components <strong>and</strong> evolutionary<br />
behavior of the full system are not known. When<br />
testing of components is the only option, it is<br />
important to acknowledge this shortcoming.<br />
Often, when separate components of a system are<br />
validated, it can be argued that this is a form of<br />
system verification, as described by Rusu [2003].<br />
(See test 6A in Table1.)<br />
4. AN EXAMPLE: DECISION SUPPORT<br />
SYSTEM FOR TRUMPTER SWAN<br />
MANAGEMENT<br />
4.1 Background<br />
A multiagent system of interacting intelligent<br />
agents [Weiss 1999], expert systems, <strong>and</strong> a<br />
queuing model was developed to assist waterfowl<br />
managers simulate the effect of management<br />
actions on swan distributions [Sojda 2002]. This<br />
decision support system was evaluated at three<br />
levels: (1) verification of individual components,<br />
as well as the overall system, (2) soft validation of<br />
the expert systems, <strong>and</strong> (3) validation of the<br />
whole system.<br />
It was decided not to evaluate the system against<br />
a team with expertise in flyway management of<br />
swans, primarily because it was not feasible to<br />
assemble such a panel that was independent of the<br />
people used in knowledge engineering. This was<br />
true for two related reasons. First, the total<br />
number of workers in the domain is small.<br />
Second, the cadre of such workers is closely<br />
interrelated institutionally <strong>and</strong> academically.<br />
4.2 Verification of Components <strong>and</strong> Whole<br />
System<br />
A key part of designing the individual expert<br />
systems was developing flowcharts of the<br />
ecological logic <strong>and</strong> using them to consult with<br />
experts for changes <strong>and</strong> refinement. Similarly,<br />
the “planeditor” facility in the multiagent<br />
software, DECAF, [Graham <strong>and</strong> Decker 2000;<br />
Graham 2001] allowed me to develop graphical<br />
representations of the logic underlying each agent<br />
<strong>and</strong> consult with specialists in multiagent system<br />
design. When running the multiagent system,<br />
DECAF provided information about how each<br />
agent was functioning <strong>and</strong> about failed<br />
communications among agents. Utilities within<br />
the expert system development shell were used<br />
for verification of logical consistency of each<br />
expert system, including a static check for<br />
problems such as incomplete rules <strong>and</strong> trees. For<br />
example, an error would be detected if more than<br />
one rule tried to set a value for a single-valued<br />
variable, or if the consequent portion of a rule was<br />
inadvertently not provided. The utilities also<br />
dynamically checked the system with stochastic<br />
runs, <strong>and</strong> the final system was checked using<br />
500,000 simulated runs with no problems<br />
detected.<br />
4.3 Soft Validation of the Expert System<br />
Components<br />
Demonstrations of each expert system were made<br />
to waterfowl managers, biologists, <strong>and</strong><br />
researchers. This involved meetings <strong>and</strong><br />
telephone consultations where individuals ran<br />
actual scenarios <strong>and</strong> provided comments. In<br />
addition, the expert systems were available in<br />
st<strong>and</strong>-alone fashion on the World Wide Web, both<br />
in prototype <strong>and</strong> final versions. Such validation<br />
targeted the underlying ontologies, knowledge,<br />
<strong>and</strong> problem solving logic, but was not empirical.<br />
4.4 Validation Using An Historic Data Set<br />
Based on queuing theory [Dshalalow 1995;<br />
Hillier <strong>and</strong> Lieberman 1995], the DSS begins by<br />
using an observed number of swans at each of 27<br />
areas for the breeding season of one year, <strong>and</strong><br />
then simulates the number at each of those areas<br />
for the four subsequent seasons, concluding with<br />
a simulated number for the breeding season of the<br />
subsequent year. The system simulates breeding<br />
swan numbers in one year increments. It was a<br />
comparison of the simulated number for the<br />
subsequent year versus the observed number for<br />
that same year that was the basis of my empirical<br />
652
Test<br />
MVPTMP<br />
p-value<br />
Interpretation from rejecting the null hypothesis<br />
1 .0001 output from base queuing model similar to observed numbers<br />
3A .0001 output using all expert systems (3) <strong>and</strong> activating all (7) refuge agents similar to observed numbers<br />
6A - output using 3 expert systems identical to that with only the flyway expert system<br />
7A - output using alternate breeding threshold of 0.4 identical to that using the st<strong>and</strong>ard, 0.6<br />
Table 1. Interpretation of MVPTMP analyses from 4 of 34 experimental runs of the decision support system<br />
for trumpeter swan management. Null hypotheses were developed a priori [Sojda 2002]. No p-value is<br />
reported when output between the two groups was identical.<br />
testing. An observed number of swans was<br />
available only for the breeding season, <strong>and</strong> not the<br />
other seasons, so analysis was limited to data for<br />
that season. Comparisons of simulated <strong>and</strong><br />
observed data could be made for 13 years, 1988-<br />
2000. Observed numbers were those collected by<br />
the member agencies of the Pacific Flyway<br />
Council <strong>and</strong> informally reported by the United<br />
States Fish <strong>and</strong> Wildlife Service on an annual<br />
basis [e.g., Reed 2000].<br />
4.5 Data Analysis<br />
Although all 27 areas were always used in the<br />
queuing model, swans had never been observed in<br />
seven areas during the breeding season <strong>and</strong> those<br />
areas were excluded from statistical analysis. In<br />
all such cases, the system did not simulate swans<br />
in those areas. This ensured that the consistent<br />
simulation of no swans where none were expected<br />
did not artificially inflate the evaluated accuracy<br />
<strong>and</strong> precision of the system.<br />
Thirty-four black-box experiments were<br />
conducted to empirically validate the decision<br />
support system’s ability to predict swan<br />
distributions in the flyway [Sojda 2002]. The<br />
results from four of the experiments are provided<br />
in Table 1. Multivariate Matched-Pairs<br />
Permutation Test (MVPTMP) statistical<br />
procedures [Mielke <strong>and</strong> Berry 2001] were used<br />
for the analyses. The first of the pair is simulated<br />
data, the second is either observed data or<br />
simulated data from a run of the system with a<br />
different configuration. To test the base model (a<br />
queuing system), predicted numbers of swans for<br />
20 areas were compared against observed<br />
numbers for a series of 13 years. In such<br />
analyses, a small p-value is evidence of similarity<br />
of distributions of swans over both space <strong>and</strong> time<br />
between the two groups of data forming a pair.<br />
Because of the multidimensional structure of such<br />
comparisons of spatial data over time, it was<br />
difficult to provide visualizations. Accompanying<br />
departures of the simulated from the observed<br />
numbers of swans were simply graphed [Sojda<br />
2002].<br />
5. DISCUSSION AND CONCLUSIONS<br />
Validation is the process of determining whether<br />
the stated purpose of the system was achieved. I<br />
conclude that multiagent systems were an<br />
effective way to model movement of waterfowl in<br />
a flyway. Because models are abstractions of<br />
reality, it is inherent that they will have<br />
shortcomings from not being able to accurately<br />
represent all knowledge, logical relationships, <strong>and</strong><br />
probabilistic intricacies. Overall, the evidence<br />
was strong that the base model (in the decision<br />
support system for trumpeter swan management)<br />
mimicked the observed pattern of swan<br />
distributions over time, as does the system run<br />
with the default configuration. Almost all<br />
experimental runs of the decision support system<br />
showed the same pattern.<br />
It seems irresponsible to deliver a decision<br />
support system that has not been adequately<br />
evaluated, including both verification <strong>and</strong><br />
validation. Empirical evaluation in some form is<br />
critical, <strong>and</strong> can range from experiments run<br />
against a preselected gold st<strong>and</strong>ard to more simple<br />
testing of system components. It is imperative to<br />
underst<strong>and</strong>, from an experimental <strong>and</strong> logical<br />
perspective, to what extent inferences can be<br />
made as a result of the validation. In the end, the<br />
question to answer is: Was the system successful<br />
at addressing its intended purpose? Often,<br />
searching for the right database for empirical<br />
evaluation can be as important as adequate<br />
decision support system development, itself.<br />
6. ACKNOWLEDGEMENTS<br />
I recognize A. Howe, D. Dean, P. Mielke, <strong>and</strong> S.<br />
Stafford for their encouragement <strong>and</strong> for<br />
introducing many of the key concepts found in<br />
this paper. R. Jachowski stimulated thought<br />
about objectivity <strong>and</strong> practical application of<br />
653
model evaluation. F. D’Erchia provided a review<br />
of an early manuscript. Funding was provided by<br />
the U.S. Department of Interior: the Geological<br />
Survey-Biological Resources Division <strong>and</strong> the<br />
Fish <strong>and</strong> Wildlife Service. This research was part<br />
of Geological Survey, Biological Resources<br />
Division Project Number 915.<br />
7. REFERENCES<br />
Adelman, L., Experiments, quasi-experiments,<br />
<strong>and</strong> case studies: a review of empirical<br />
methods for evaluating decision support<br />
systems, IEEE Transactions on Systems,<br />
Man, <strong>and</strong> Cybernetics, 21(2), 293-301, 1991.<br />
Adelman, L., Evaluating decision support <strong>and</strong><br />
expert systems, John Wiley <strong>and</strong> Sons, New<br />
York, New York, 1992.<br />
Andriole, S. J., H<strong>and</strong>book of decision support<br />
systems, TAB Professional <strong>and</strong> Reference<br />
Books, Blue Ridge Summit, Pennsylvania,<br />
248 pages, 1989.<br />
Adrion, W. R., M. A. Branstad, <strong>and</strong> J. C.<br />
Cherniavsky, Validation, verification, <strong>and</strong><br />
testing of computer software, ACM<br />
Computing Surveys, 14(2), 159-192, 1982.<br />
Bahill, T. A., Verifying <strong>and</strong> validating personal<br />
computer-based expert systems, Prentice-<br />
Hall, Inc., Englewood Cliffs, New Jersey,<br />
205 pages, 1991.<br />
Boehm, B. W., <strong>Software</strong> engineering economics,<br />
Prentice-Hall, Inc., Englewood Cliffs, New<br />
Jersey, 767 pages, 1981.<br />
Bol<strong>and</strong>, R. J., A. K. Mahewshwari, D. Te'eni, D.<br />
G. Schwartz, <strong>and</strong> R. V. Tenkasi, Sharing<br />
perspectives in distributed decision making,<br />
pages 306-313, in: Proceedings of the<br />
Conference on Computer-Supported<br />
Cooperative Work, Association for<br />
Computing Machinery, New York, New<br />
York, 1992.<br />
Brehmer, B. Distributed decision making: some<br />
notes on the literature, pages 3-14 in:<br />
Rasmussen, J., B. Brehmer, <strong>and</strong> J. Leplat,<br />
eds., Distributed decision making: cognitive<br />
models for cooperative work, John Wiley <strong>and</strong><br />
Sons, Chichester, Engl<strong>and</strong>, 1991.<br />
Carter, G. M., M. P. Murray, R. G. Walker, <strong>and</strong><br />
W. E. Walker, Building organizational<br />
decision support systems, Economic Press<br />
Inc., San Diego, California, 358 pages, 1992.<br />
Chau, M., D. Zeng, H. Chen, M. Huang, <strong>and</strong> D.<br />
Hendriawan, Dwsign <strong>and</strong> evaluation of a<br />
multiagent collaborative Web mining system,<br />
Decision Support Systems, 35, 167-183,<br />
2003.<br />
Cohen, P. R., <strong>and</strong> A. E. Howe, Toward AI<br />
research methodology: three case studies in<br />
evaluation, IEEE Transactions on Systems,<br />
Man, <strong>and</strong> Cybernetics, 19(3), 634-646, 1989.<br />
Dshalalow, J. H., An anthology of classical<br />
queuing models, pages 1-42 in: Dshalalow, J.<br />
H., ed., Advances in queuing: theory,<br />
methods, <strong>and</strong> open problems, CRC Press.<br />
Boca Raton, Florida, 1995.<br />
D’Erchia, F., C. Korschgen, M. Nyquist, R. Root,<br />
R. Sojda, P. Stine, A framework for<br />
ecological decision support systems: building<br />
the right systems <strong>and</strong> building the systems<br />
right., U.S. Geological Survey, Biological<br />
Resources Division, Information <strong>and</strong><br />
Technology Report USGS/BRD/ITR-2001-<br />
0002, Washington, DC, 50 pages, 2001.<br />
Eason, K., Information technology <strong>and</strong><br />
organizational change, Taylor <strong>and</strong> Francis<br />
Publishing, London, United Kingdom, 247<br />
pages, 1988.<br />
Fishmann, G. S., <strong>and</strong> P. J. Kiviat, The statistics of<br />
discrete-event simulation, Simulation, 10,<br />
185-195, 1968.<br />
Graham, J. R., Real-time scheduling in distributed<br />
multi agent systems, PhD Dissertation,<br />
University of Delaware. Newark, Delaware,<br />
166 pages, 2001.<br />
Graham, J. R. <strong>and</strong> Decker, K. S., Towards a<br />
distributed, environment-centered agent<br />
framework, pages 290-304 in: Jennings,<br />
Nicholas R. <strong>and</strong> Y. Lesperance, eds.,<br />
Proceedings of the Sixth <strong>International</strong><br />
Workshop on Agent, Theories, Architectures,<br />
<strong>and</strong> Languages (ATAL-99), Springer-Verlag,<br />
Berlin, Germany, 2000.<br />
Gupta, U., Validating <strong>and</strong> verifying knowledgebased<br />
systems, IEEE Computer Society<br />
Press, Washington, DC, 423 pages, 1991.<br />
Haberl<strong>and</strong>t, U., V. Krysanova, <strong>and</strong> A. Bardossy,<br />
Assessment of nitrogen leaching from arable<br />
l<strong>and</strong> in large river basins - Part II:<br />
regionalisation using fuzzy rule based<br />
modelling, Ecological <strong>Modelling</strong>, 150, 277-<br />
294, 2002.<br />
Hillier, F. S. <strong>and</strong> G. J. Lieberman, Introduction to<br />
operations research, McGraw-Hill, Inc., New<br />
York, New York, 998 pages, 1995.<br />
Johnson, D. H., Validating <strong>and</strong> evaluating<br />
models, pages 105-119 in: Shenk, T. M., <strong>and</strong><br />
A. B. Franklin, eds., <strong>Modelling</strong> in Natural<br />
Resource Management: Development,<br />
Interpretation, <strong>and</strong> Application, Isl<strong>and</strong> Press.<br />
Washington, DC, 2001.<br />
Kanungo, S., S. Sharma, <strong>and</strong> P. K. Jain,<br />
Evaluation of a decision support system for<br />
credit management decisions, Decision<br />
Support Systems, 30, 419-436, 2001.<br />
Mielke, P. W. <strong>and</strong> K. J. Berry, Permutation<br />
methods: a distance function approach.<br />
654
Springer-Verlag, New York, New York, 352<br />
pages, 2001.<br />
Mihram, G. A., Some practical aspects of the<br />
verification <strong>and</strong> validation of simulation<br />
models, Operational Research Quarterly,<br />
23(1), 17-29, 1972.<br />
Mosqueira-Rey, E, <strong>and</strong> V. Moret-Bonillo,<br />
Validation of intelligent systems: a critical<br />
study <strong>and</strong> a tool, Expert Systems with<br />
Applications, 18, 1-16, 2000.<br />
Murrell, S. <strong>and</strong> R. T. Plant, A survey of tools for<br />
the validation <strong>and</strong> verification of knowledgebased<br />
systems: 1985-1995, Decision Support<br />
Systems, 21, 307-323, 1997.<br />
O’Keefe, R. M., O. Balci, <strong>and</strong> E. P. Smith,<br />
Validating expert system performance, IEEE<br />
Expert, 2(4), 81-90, 1987.<br />
O’Leary, D. O. Verification of multiple agent<br />
knowledge-based systems, <strong>International</strong><br />
Journal of Intelligent Systems, 16, 361-376,<br />
2001.<br />
Plant, R., <strong>and</strong> R. Gamble, Methodologies for the<br />
development of knowledge-based systems,<br />
1982-2002, The Knowledge Engineering<br />
Review, 81(1), 47-81, 2003.<br />
Pretzsch, H., P. Biber, <strong>and</strong> J. Dursky, The single<br />
tree-based st<strong>and</strong> simulator SILVA:<br />
construction, application <strong>and</strong> evaluation,<br />
Forest Ecology <strong>and</strong> Management, 162,3-21,<br />
2002.<br />
Priya, S., <strong>and</strong> R. Shibasaki, National spatial crop<br />
yield simulation using GIS-based crop<br />
production model, Ecological <strong>Modelling</strong>,<br />
135, 113-129, 2001.<br />
Reed, T., 2000 Fall trumpeter swan survey.<br />
Unpublished report, U.S. Fish <strong>and</strong> Wildlife<br />
Service, Lakeview, Montana, 28 pages, 2000.<br />
Rice, J. A., <strong>and</strong> P. A. Cochran, Independent<br />
evaluation of a bioenergetics model for<br />
largemouth bass, Ecology, 65(3), 732-739,<br />
1984.<br />
Rushby, J., Validation <strong>and</strong> testing of knowledgebased<br />
systems: how bad can it get?, pages 77-<br />
83 in: Gupta, U., ed., Validating <strong>and</strong><br />
Verifying Knowledge-Based Systems, IEEE<br />
Computer Society Press, Los Alamitos,<br />
California, 1988.<br />
Rusu, V., Combining formal verification <strong>and</strong><br />
conformance testing for validating reactive<br />
systems, <strong>Software</strong> Testing, Verification <strong>and</strong><br />
Reliability, 13, 157-180, 2003.<br />
Santos, E., Jr., Verification <strong>and</strong> validation of<br />
Bayesian knowledge-bases, Data <strong>and</strong><br />
Knowledge Engineering, 37, 307-329, 2001.<br />
Sojda, R. S., Artificial intelligence based decision<br />
support for trumpeter swan management,<br />
PhD Dissertation, Colorado State University,<br />
Fort Collins, Colorado, 193 pages, 2002.<br />
Sprague, R. H. Jr., <strong>and</strong> E. D. Carlson, Building<br />
effective decision support systems, Prentice-<br />
Hall, Englewood Cliffs, New Jersey, 329<br />
pages, 1982.<br />
Stuth, J. W. <strong>and</strong> M. S. Smith, Decision support<br />
for grazing l<strong>and</strong>s: an overview, pages 1-35 in<br />
J.W. Stuth <strong>and</strong> B.G. Lyons, eds., Decision<br />
support systems for the management of<br />
grazing l<strong>and</strong>s, Man <strong>and</strong> the Biosphere Series<br />
<strong>Volume</strong> 11: Papers from the <strong>International</strong><br />
Conference on Decision Support Systems for<br />
Resource Management, The Parthenon<br />
Publishing Group, Pearl River, New York,<br />
1993.<br />
Virvou, M., <strong>and</strong> K. Kabassi, Evaluating an<br />
intelligent graphical user interface by<br />
comparison with human experts, Knowledge-<br />
Based Systems, 17, 31-37, 2004.<br />
Wallace, D. R. <strong>and</strong> R. U. Fujii, <strong>Software</strong><br />
verification <strong>and</strong> validation: an overview,<br />
IEEE <strong>Software</strong>, 6(3), 10-17, 1989.<br />
Wang, J., <strong>and</strong> C. B. LeDoux, Estimating <strong>and</strong><br />
validating ground-based timber harvesting<br />
production through computer simulation,<br />
Forest Science, 49(1), 64-76, 2003.<br />
Weiss, G., Prologue: multiagent systems <strong>and</strong><br />
distributed artificial intelligence, pages 1-23<br />
in: Weiss, G., ed., Multiagent systems: a<br />
modern approach to distributed artificial<br />
intelligence, MIT Press, Cambridge,<br />
Massachusetts, 1999.<br />
655
An integrated modelling approach to conduct multifactorial<br />
analyses on the impacts of climate change on<br />
whole-farm systems.<br />
M. Rivington a , G. Bellocchi b , K.B. Matthews a , K. Buchan a <strong>and</strong> M. Donatelli b<br />
a<br />
Macaulay Institute, Craigiebuckler, Aberdeen, UK. (m.rivington@macaulay.ac.uk)<br />
b Research Institute for Industrial Crops, via di Corticella 133, 40128 Bologna, Italy.<br />
Abstract: Climate change impact studies on whole-farm systems require a holistic approach due to the<br />
complexities of biophysical processes, management <strong>and</strong> inter-relationships of l<strong>and</strong> use within a single farm.<br />
This paper details the process of utilising a multiple-objective, strategic l<strong>and</strong> use planning tool to conduct<br />
multi-factorial analyses on the impacts of climate change at the farm scale. Two example sites are given to<br />
illustrate the flexibility of the method: an upl<strong>and</strong> mixed sheep <strong>and</strong> suckler cow farm in Scotl<strong>and</strong>, with cold<br />
wet winters <strong>and</strong> cool moist summers; <strong>and</strong> a combined cropping <strong>and</strong> indoor reared beef farm in Italy, with<br />
cool moist winters <strong>and</strong> warm dry summers. The approach allows the additional risk that climate change<br />
may introduce to the farm system to be quantified. Model output facilitates the development of adaptation<br />
<strong>and</strong> amelioration strategies. This Integrated Assessment (AI) approach employs the L<strong>and</strong> Allocation<br />
Decision Support System (LADSS), a framework which permits a wide range of counter-factual<br />
assessments of financial, social <strong>and</strong> environmental impacts of changes to policy, management <strong>and</strong><br />
biophysical conditions. The framework contains a Geographical Information System (GIS) <strong>and</strong> relational<br />
database linked with l<strong>and</strong> use models, impact assessments <strong>and</strong> planning tools. Crop based l<strong>and</strong> uses are<br />
represented by the CropSyst cropping systems model <strong>and</strong> livestock by a Livestock Production Model<br />
(LPM). The framework provides an opportunity to explore the linkages between sub-components of the<br />
farm system <strong>and</strong> demonstrates the diversity of possible climate change impacts. The paper indicates the<br />
importance of management decisions in determining amelioration of the impacts of climate change on the<br />
farm system. Farms constitute one of the fundamental units within the agri-ecosystem, hence it is important<br />
to underst<strong>and</strong>ing the impacts of change <strong>and</strong> the subsequent requirements for management adaptation. This<br />
underst<strong>and</strong>ing can then be used to better inform policy makers.<br />
Keywords: Climate change impacts, multi-factorial analyses, farm systems, LADSS, CropSyst.<br />
1. INTRODUCTION<br />
The impacts of climate change (CC) may manifest<br />
themselves on whole-farm systems in many ways,<br />
ranging from subtle, small-scale cause <strong>and</strong> effects,<br />
to extreme events, with both occurring<br />
simultaneously. It is desirable to know which is<br />
more influential in determining a farm’s viability<br />
<strong>and</strong> capacity to adapt. The intricacies of both scales<br />
of impacts require a detailed modelling framework,<br />
which should be able to represent the complexities<br />
of biophysical processes, l<strong>and</strong> use interrelationships<br />
<strong>and</strong> management regimen. Such a<br />
framework enables a holistic IA approach to<br />
determine the impacts of CC throughout the farm<br />
system. Previous studies have identified potential<br />
CC impacts for a range of farm components, e.g.<br />
individual crops at the national scale (Holden et al<br />
2003), site-specific cropping systems (Tubiello et<br />
al 2000), milk yield <strong>and</strong> dairy herds (Topp <strong>and</strong><br />
Doyle 1996), <strong>and</strong> crop yields <strong>and</strong> ecosystem<br />
processes (Izaurralde et al 2003). Studies carried<br />
out on cropping systems in Europe include: Bindi<br />
et al (1999); Bellocchi et al (2002); Donatelli et al<br />
(2002). However, to complete a holistic study of<br />
CC impacts, it is necessary to include the<br />
consequences on the biophysical components,<br />
inter-relationships between l<strong>and</strong> uses, <strong>and</strong> the<br />
subsequent financial <strong>and</strong> social aspects. It is<br />
desirable to know whether there is sufficient<br />
656
flexibility <strong>and</strong> resilience within a farm system to<br />
cope with the impacts of CC, <strong>and</strong> what adaptation<br />
<strong>and</strong> amelioration strategies will be required.<br />
Strategies to cope with CC are most likely to be<br />
facilitated through changes in management<br />
practises, to both individual l<strong>and</strong> uses <strong>and</strong> the<br />
overall farm system. Using the framework output, a<br />
soft-systems appraisal approach can be taken (e.g.<br />
Matthews et al 2002), where practitioners identify<br />
potential adaptation <strong>and</strong> amelioration strategies.<br />
These strategies can then be created as new<br />
scenarios run within the framework to determine<br />
how they respond to CC.<br />
This paper describes the structure of a framework<br />
consisting of the L<strong>and</strong> Allocation Decision Support<br />
System (for full details see LADSS website,<br />
http://www.mluri.sari.ac.uk/LADSS/ladss.shtml)<br />
which has been adapted to enable studies of the<br />
impacts of CC. Two contrasting locations <strong>and</strong> farm<br />
systems are given as examples of the flexibility of<br />
the framework: one a mixed sheep <strong>and</strong> suckler cow<br />
system in central Scotl<strong>and</strong> (Hartwood); the other is<br />
a combined cropping <strong>and</strong> indoor reared beef farm<br />
in Tuscany, Italy (Montepulciano).<br />
2. FRAMEWORK STRUCTURE<br />
LADSS (Matthews et al 1999) enables farm-scale<br />
multi-objective l<strong>and</strong> use planning, considering<br />
financial, social <strong>and</strong> environmental constraints. The<br />
system determines a range of optimal l<strong>and</strong> use<br />
patterns for a given set of constraints, e.g. financial<br />
return versus l<strong>and</strong> use diversity. Additional outputs<br />
include financial, social <strong>and</strong> environmental impacts<br />
for each optimised l<strong>and</strong> use pattern. This enables<br />
detailed counter-factual analysis across a broad<br />
range of inter-relationships. The framework<br />
consists of a Geographical Information System<br />
(GIS) linked to an Oracle Relational Data Base<br />
Management System (RDBMS). L<strong>and</strong> use systems<br />
are represented by simulation models: CropSyst<br />
cropping systems simulation model (Stöckle et al<br />
2003) <strong>and</strong> a Livestock Production Model (LPM)<br />
which simulates sheep <strong>and</strong> cattle systems. A simple<br />
forestry model represents trees grown on the farm.<br />
L<strong>and</strong> use system models estimate a range of<br />
variables including productivity for areas within a<br />
farm that have the biophysical attributes required to<br />
support them. The models are linked to a package<br />
of integrated impact assessments: financial; social<br />
<strong>and</strong> environmental <strong>and</strong> a genetic algorithm based<br />
l<strong>and</strong> use planning tool. The social impact<br />
assessment contains a Resource Scheduling Tool<br />
(RST) (Matthews et al 2003) which is linked to the<br />
RDBMS <strong>and</strong> l<strong>and</strong> use models. The timing of<br />
management events is then used to schedule labour<br />
<strong>and</strong> machinery requirements, with associated cost<br />
being calculated by the RST <strong>and</strong> financial<br />
assessment. Data are supplied from the RDBMS to<br />
the l<strong>and</strong> use systems models <strong>and</strong> impact<br />
assessments. Spatial data detailing the farm layout<br />
are supplied from the GIS. The database <strong>and</strong> GIS<br />
also perform analysis on the stored data <strong>and</strong> those<br />
returned by the l<strong>and</strong> use model components.<br />
Spatial Data<br />
Capture<br />
Geographic<br />
Information<br />
System<br />
SW-Oracle<br />
Interface<br />
Oracle<br />
RDBMS<br />
Global &<br />
Management<br />
Parameters<br />
G2-SW Bridge<br />
G2-Oracle Interface<br />
L<strong>and</strong><br />
Use Planning<br />
Tools<br />
Integrated<br />
Impact<br />
Assessments<br />
L<strong>and</strong>-Use<br />
Systems<br />
Models<br />
CropSyst<br />
Graphical Communication Interface<br />
Decision makers, consultants,<br />
or policy analysts<br />
Figure 1. LADSS component structure with input elements. SW refers to SmallWorld GIS <strong>and</strong> G2 is a<br />
knowledge base development software.<br />
657
2.1 Framework requirements<br />
The framework requires a detailed set of inputs<br />
describing the variability of biophysical<br />
characteristics, the management regimen <strong>and</strong><br />
associated costs <strong>and</strong> revenues within the farm.<br />
Spatial data are captured from aerial photography<br />
which is also used to define in-field soil sampling<br />
strategies. This facilitates cost effective soil data<br />
collection <strong>and</strong> permits identification of<br />
homogeneous areas of soil. In modelling crop<br />
production, daily precipitation, air maximum <strong>and</strong><br />
minimum temperature <strong>and</strong> solar radiation data are<br />
used within CropSyst, hence a requirement for siterepresentative<br />
data. The LPM requires information<br />
about the livestock management <strong>and</strong> feed regimen.<br />
2.2 Spatial configuration<br />
The spatial configuration of farm resources is<br />
important as it serves to identify the biophysical<br />
<strong>and</strong> practical constraints on management. These<br />
help determine the boundaries of adaptation <strong>and</strong><br />
amelioration strategies. A farm within the<br />
framework exists in four object classes in a<br />
hierarchy:<br />
• L<strong>and</strong> block fragment polygons (LBFP) are the<br />
smallest unit <strong>and</strong> are used to calculate spatial<br />
geometry within the GIS.<br />
• The l<strong>and</strong> block fragment (LBF) are areas of<br />
biophysically homogeneous l<strong>and</strong>, which also<br />
have a uniform management regime. Soils exist<br />
in multiples of layers within an LBF. The LBF is<br />
the lowest level at which financial analysis is<br />
conducted.<br />
• L<strong>and</strong> blocks (LB) are areas of homogeneous l<strong>and</strong><br />
use <strong>and</strong> the scale at which the l<strong>and</strong> use system<br />
models are applied.<br />
• An Enterprise (farm) is made up of l<strong>and</strong> blocks<br />
<strong>and</strong> forms the top level of the hierarchy, for<br />
financial, social <strong>and</strong> environmental analysis <strong>and</strong><br />
auditing.<br />
maximum <strong>and</strong> minimum mean monthly<br />
temperatures are 17 <strong>and</strong> -2º C respectively.<br />
Elevation ranges between 150 to 300 m a.s.l. with<br />
south facing slopes. Soils are shallow, poorly<br />
drained gleys which rarely fall much below field<br />
capacity, making the farm difficult to work in wet<br />
conditions <strong>and</strong> susceptible to poaching. Prolonged<br />
wet periods prevent machine access to fields. The<br />
soil shows high within-field spatial variability.<br />
Field boundaries are irregular shapes <strong>and</strong> nonsystematic,<br />
being artefacts of previous ownership<br />
(Fig. 2). Winter wheat <strong>and</strong> spring barley is grown<br />
as whole crop fodder. A farm such as Hartwood<br />
would typically be family run, equivalent to<br />
employing 3 staff.<br />
The farm in Montepulciano, Italy, is a 300 ha<br />
combined cropping <strong>and</strong> indoor reared beef system,<br />
with cool moist winters <strong>and</strong> warm dry summers.<br />
Elevation is about 300 m a.s.l., with no slope.<br />
Maximum air temperatures are often higher than 35<br />
ºC. Average annual rainfall is about 700 mm,<br />
mostly concentrated in spring <strong>and</strong> autumn. Field<br />
boundaries are uniform <strong>and</strong> systematic (Fig. 2.).<br />
Crops grown include: cereals (durum wheat),<br />
forages (alfalfa, triticale), oil-seed (sunflower) <strong>and</strong><br />
horticultural (capsicum, tomato) root (sugarbeet)<br />
<strong>and</strong> leaf crops (tobacco). Livestock activity is the<br />
breeding of Chianina cattle (c. 300 animals), reared<br />
for meat production <strong>and</strong> reproduction, organised as<br />
partially free housing in stalls. Livestock manure is<br />
applied directly to the fields. Deep soils have been<br />
artificially created as a result of the saturation of<br />
the antecedent wetl<strong>and</strong> sediments from flooding.<br />
Semi-permanent <strong>and</strong> permanent water bodies make<br />
up 10 hectares of the farm, providing reservoirs for<br />
crop irrigation. The volume of such reservoirs can<br />
be estimated to determine the irrigation capacity.<br />
There is a high irrigation dem<strong>and</strong> in the spring <strong>and</strong><br />
summer. The farm is family run with 6 permanent<br />
staff, 2 of which are devoted to livestock<br />
management. Temporary staff are employed at<br />
harvesting <strong>and</strong> transplanting times.<br />
2.3 Farming systems coverage<br />
The flexibility of the framework can be illustrated<br />
by detailing two contrasting farm systems in<br />
diverse locations which are currently represented<br />
by LADSS. Hartwood Farm in Scotl<strong>and</strong> is a 350 ha<br />
mixed sheep (c. 500) <strong>and</strong> cattle (c. 200) system,<br />
representative of upl<strong>and</strong> farms in marginal<br />
production areas in upl<strong>and</strong> central Scotl<strong>and</strong>. It is<br />
characterised by cold wet winters <strong>and</strong> cool moist<br />
summers. Mean annual rainfall is about 1200 mm,<br />
2.4 Excluded considerations <strong>and</strong> limitations<br />
There are limits to the considerations made in a<br />
holistic study. In the examples given it is not<br />
possible to assess animal welfare <strong>and</strong> conseqential<br />
labour requirements, crop quality etc. External<br />
influences acting on the farm, such as new policies,<br />
changes in commodity prices etc. although able to<br />
be represented, are currently considered as fixed<br />
for this type of study.<br />
658
Hartwood<br />
Montepulciano<br />
Homogeneous soil /<br />
LBF<br />
L<strong>and</strong> Block (field boundary)<br />
Figure 2. Relationship of soil variability <strong>and</strong> field boundaries between Hartwood <strong>and</strong> Montepulciano farms<br />
2.5 System adaptations<br />
It is important to know the variability of soil<br />
resource distribution given the requirements to<br />
estimate the impact of CC on farm productivity <strong>and</strong><br />
consequences on management. There are<br />
substantial differences in the spatial configuration<br />
of soil variability in relation to the shape <strong>and</strong> size<br />
of LBF between the two example farms. At<br />
Hartwood the soils are highly variable within large<br />
irregular shaped fields (therefore fields with<br />
multiple LBFs), whereas at Montepulciano LBFs<br />
are small, regular shaped <strong>and</strong> can exist within<br />
relatively uniform soil conditions (Fig 2). The<br />
differences between the two sites required that<br />
changes be made to the GIS <strong>and</strong> RDBMS structure.<br />
By making the LB to LBF a ‘many-to-many’<br />
relationship within the database allows an LBF to<br />
be part of many LB, reducing the number of l<strong>and</strong><br />
use model simulations made per LBF. In the<br />
example of Fig 2, l<strong>and</strong> use models are applied to<br />
each LBF: for Hartwood the results of l<strong>and</strong> use<br />
model simulations are aggregated between the LBF<br />
to give a single set of values for the LB; at<br />
Montepulciano the simulation is applied for a<br />
single LB <strong>and</strong> the results replicated for other LBs<br />
within the same LBF.<br />
3. EXAMPLE CC SCENARIOS<br />
The framework is able to use daily CC scenario<br />
weather data from a wide range of sources<br />
representative of different scales. For example<br />
from general circulation models (GCM), statistical<br />
downscaling, or weather generators. For case<br />
studies, a basic analytical approach entails the<br />
introduction of estimates of CC induced alterations<br />
<strong>and</strong> examination of how the framework estimates<br />
differs from the a base solution without climate<br />
change. Simulations can be run for transient<br />
scenarios drawn from two GCMs, the Canadian<br />
Climatic Centre (CCC) model <strong>and</strong> the Hadley<br />
Centre (HAD) model. Although impact analysis<br />
can be based on these transient scenarios, it is<br />
preferable to use average climate conditions for<br />
periods around the year 2030 or 2090. Long-term<br />
observed daily weather records form the basis to<br />
generate 50 years of baseline <strong>and</strong> altered climate<br />
data sets, by using the ClimGen (Stöckle et al<br />
2001) stochastic generator. Average productivity<br />
<strong>and</strong> resource use can then be estimated over this 50<br />
year period for the baseline <strong>and</strong> altered climate<br />
scenarios. Use of CropSyst to estimate crop<br />
biomass enables atmospheric CO 2 concentration<br />
levels to be set that represent the time period <strong>and</strong><br />
CC scenario being simulated, e.g. 350 ppmv for the<br />
baseline; 445 ppmv for 2030, <strong>and</strong> 660 ppmv for<br />
2090 IS92a scenario of future emissions (IPCC<br />
2000).<br />
4. FRAMEWORK OUTPUTS<br />
For a no CC scenario (baseline) the framework can<br />
produce a range of optimised patterns of l<strong>and</strong> use<br />
for a given set of constraints. Maintaining these<br />
constraints <strong>and</strong> then using weather data from the<br />
CC scenarios permits between-scenario<br />
comparisons. Responses to CC can be observed<br />
for each individual l<strong>and</strong> use (e.g. change in yield),<br />
biophysical entity (e.g. soil water <strong>and</strong> nitrogen<br />
balances) <strong>and</strong> whole farm (change in l<strong>and</strong> use<br />
659
patterns). Outputs also need to indicate changes in<br />
the levels of risk associated with CC per l<strong>and</strong> use<br />
<strong>and</strong> their role within the whole farm system. For<br />
each of the patterns of l<strong>and</strong> use per scenario,<br />
associated labour <strong>and</strong> machinery requirements<br />
(from the RST), financial <strong>and</strong> environmental<br />
impact assessment outputs are produced. Outputs<br />
can be visualised in both the GIS (maps of the<br />
spatial allocation of l<strong>and</strong> uses) <strong>and</strong> G2 interface<br />
(RST, social, environmental <strong>and</strong> financial impacts),<br />
allowing practitioner appraisal. This permits<br />
identification of potential management adaptations<br />
<strong>and</strong> amelioration strategies.<br />
4.1 Potential biophysical impacts of CC<br />
Weather is the primary variable determining the<br />
soil/crop state <strong>and</strong> influences the ability of farmers<br />
to respond with appropriate management. Weather<br />
determines biophysical relationships such as soil<br />
<strong>and</strong> crop water <strong>and</strong> nitrogen balances, phenological<br />
development <strong>and</strong> ultimately biomass accumulation<br />
e.g. as modelled by CropSyst. Changes in climate<br />
will lead to altered weather inputs resulting in<br />
different responses of the biophysical relationships,<br />
e.g. evapotranspiration <strong>and</strong> rainfall input, thermal<br />
time accumulation etc. <strong>and</strong> subsequent levels of<br />
crop production.<br />
There are numerous pathways in which CC affects<br />
the soil/crop/livestock processes <strong>and</strong> overall farm<br />
system. Analysis of effects such as impacts on crop<br />
yields, water dem<strong>and</strong>, water supply, <strong>and</strong> livestock<br />
production, using biophysical models can inform<br />
us of why a particular climate scenario causes<br />
yields to rise or fall <strong>and</strong> suggest directions for<br />
adaptation at a range of scales within the farm.<br />
Within the framework, it is possible to analyse at<br />
least three principal direct effects of CC:<br />
• Crop yields<br />
• Water supply <strong>and</strong> irrigated crops, soil water<br />
balance<br />
• Livestock performance <strong>and</strong> grazing / pasture<br />
supply<br />
Impacts on each of these determine the new set of<br />
resource use levels. These elements can be<br />
investigated via the frameworks output.<br />
4.2 Potential CC impacts on whole farm systems<br />
Observation of the impacts on crop phenological<br />
development (due to altered air temperature<br />
conditions <strong>and</strong> impact on thermal time<br />
accumulation) indicate the shifts that can occur in<br />
the timing of management events. The impacts of<br />
this are picked up by analysis of the RST output.<br />
For the dry Italian farm, determining the changes in<br />
soil water balance <strong>and</strong> crop water uptake for all<br />
crops enable estimates to be made of irrigation<br />
dem<strong>and</strong>. This amount compared with the estimated<br />
total irrigation capacity permits identification of<br />
potential shortages. Such analysis permits the<br />
identification of potential requirements to invest in<br />
increased irrigation capacity. For the wetter<br />
Scottish farm, changes in days when the l<strong>and</strong> is<br />
workable due to soil water-logging can be<br />
identified.<br />
The relationship between the ability of a farm to<br />
produce fodder for the livestock system determines<br />
the herd size. Changes in farm produced fodder<br />
crop quantity can be modelled within the<br />
framework, but not the impacts on feed quality.<br />
The consequences in altered fodder crop biomass<br />
production can therefore be partially identified<br />
through analysis of the outputs from the LPM.<br />
Utilising the l<strong>and</strong> use pattern optimisation<br />
capability of LADSS permits GIS maps to be<br />
drawn of potential future patterns of l<strong>and</strong> use<br />
within a farm. This enables the visualisation of<br />
how patterns of l<strong>and</strong> may change over a period of<br />
time.<br />
5. DISCUSSION<br />
The framework described is able to establish the<br />
relationship between CC impacts on biophysical<br />
processes functioning at the within-field scale <strong>and</strong><br />
the consequences on l<strong>and</strong> use productivity <strong>and</strong><br />
subsequent farm-scale financial, social <strong>and</strong><br />
environmental values. The framework is<br />
sufficiently flexible to allow a wide range of farm<br />
systems in diverse locations to be represented.<br />
Holistic IA approaches to studying CC at the farmscale<br />
need to consider the points at which the<br />
weather impacts manifest themselves. Changes to<br />
patterns in weather <strong>and</strong> quantities of variables can<br />
impact on soil water <strong>and</strong> chemical balances <strong>and</strong><br />
subsequent crop growth characteristics. Alterations<br />
to livestock feed capabilities, access to l<strong>and</strong> <strong>and</strong><br />
timing of turn-out dates determine herd sizes,<br />
nutrient <strong>and</strong> energy flows <strong>and</strong> overall productivity.<br />
Changes in areas that can support l<strong>and</strong> uses within<br />
a farm at a given threshold of productivity can be<br />
identified, as can l<strong>and</strong> uses that are subject to<br />
increased risk. The output from the system is<br />
strongly influenced by the input weather data. The<br />
660
emphasis then falls on the quality of the CC<br />
scenario weather data. However the flexibility of<br />
the framework permits a wide range of CC<br />
scenarios to be represented. The manipulation of<br />
input weather data provides the opportunity to<br />
simulate a wide range in the frequency <strong>and</strong><br />
magnitude of extreme events <strong>and</strong> their impacts.<br />
This permits the identification of a scale of<br />
importance of the impacts, e.g. the relationship<br />
between infrequent extreme events <strong>and</strong> subtle<br />
continuous alterations to the biophysical<br />
environment. Impacts of extreme events may be<br />
beyond a farm’s capacity to adapt, but the<br />
framework output may indicate how the frequency<br />
of these events affects its overall viability due to<br />
the increased risk.<br />
6. CONCLUSSION<br />
This type of IA approach permits the pre-emptive<br />
identification <strong>and</strong> testing of appropriate CC impact<br />
adaptation <strong>and</strong> amelioration strategies. Farmers,<br />
l<strong>and</strong> managers <strong>and</strong> policy makers can then be<br />
better informed in constructing management<br />
methods <strong>and</strong> policies that will assist in coping with<br />
CC. The flexibility of the framework, as illustrated<br />
by the two farms detailed here, will enable the<br />
identification of the variability in impact scales for<br />
a diverse range of farm systems <strong>and</strong> locations.<br />
7. ACKNOWLEDGEMENTS<br />
The authors gratefully acknowledge the funding<br />
support of the Scottish Executive Environment <strong>and</strong><br />
Rural Affairs Department, the British-Italian<br />
Partnership Programme for Young Researchers<br />
(British Council, Ministero dell’Istruzione<br />
dell’Università e della Ricerca, CRUI).<br />
8. REFERENCES<br />
Bellocchi G., C. Maestrini, G. Fila, F. Fontana.<br />
Assessment of the effects of climate change <strong>and</strong><br />
elevated CO 2 : a case study in Northern Italy. VII<br />
European Society for Agronomy Congress, 15-<br />
18 July, Cordoba, Spain, 763-764. 2002.<br />
Bindi M., M. Donatelli, L. Fibbi, C.O.Stöckle.<br />
Estimating the effect of climate change on<br />
cropping systems at four European sites.<br />
Proceedings of the 1st <strong>International</strong> Symposium<br />
<strong>Modelling</strong> Cropping Systems, 21-23 June,<br />
Lleida, Spain, 147-148. 1999.<br />
Donatelli M., F. Tubiello, U. Peruch, C.<br />
Rosenzweig. Scenarios of climate change effects<br />
on sugar beet in Northern <strong>and</strong> Central Italy. Ital.<br />
J. Agron., 6, 133-142. 2002.<br />
Holden, N.M., A.J. Brereton, R. Fealy, <strong>and</strong> J.<br />
Sweeney. Possible change in Irish climate <strong>and</strong> its<br />
impact on barley <strong>and</strong> potato yields. Agric. For.<br />
Meterol. 116, 181-196. 2003.<br />
IPCC, Special Report on Emissions Scenarios,<br />
http://www.grida.no/climate/ipcc/emission/index<br />
.htm). 2000.<br />
Izaurralde, C.E., N.J. Rosenberg, R.A. Brown,<br />
A.M. Thomson. Integrated assessment of Hadley<br />
Centre (HADCM2) climate-change impacts on<br />
agricultural productivity <strong>and</strong> irrigation water<br />
supply in the conterminous United States Part II.<br />
Regional agricultural production in 2030 <strong>and</strong><br />
2095. Agric. For. Meteor. 117, 97-122. 2003.<br />
Matthews, K.B., A.R. Sibbald <strong>and</strong> S. Craw.<br />
Implementation of a spatial decision support<br />
system for rural l<strong>and</strong> use planning: integrating<br />
GIS <strong>and</strong> environmental models with search <strong>and</strong><br />
optimisation algorithms. Computers <strong>and</strong><br />
Electronics in Agriculture 23, 9-26. 1999.<br />
Matthews, K.B., K. Buchan, A.R. Sibbald <strong>and</strong> S.<br />
Craw. Using soft-systems methods to evaluate<br />
the outputs from multi-objective l<strong>and</strong> use<br />
planning tools. Integrated Assessment <strong>and</strong><br />
Decision Support: Proc. Int. <strong>Environmental</strong><br />
<strong>Modelling</strong> <strong>and</strong> <strong>Software</strong> Society, 24 - 27 June,<br />
University of Lugano, Switzerl<strong>and</strong>, Vol. 3, 247 –<br />
252. 2002.<br />
Matthews, K.B., K. Buchan <strong>and</strong> A. Dalziel.<br />
Evaluating labour requirements within a multiobjective<br />
l<strong>and</strong>-use planning tool. MODSIM 2003<br />
Int. Congress on <strong>Modelling</strong> <strong>and</strong> Simulation, 14-<br />
17 th July, Townsville, Australia. 2003.<br />
Stöckle, C.O., R. Nelson, M. Donatelli, <strong>and</strong> F.<br />
Castellvi. ClimGen: a flexible weather<br />
generation program. Proc. 2 nd Int. Symp.<br />
<strong>Modelling</strong> Cropping Systems, 16-18 July,<br />
Florence, Italy. European Society of Agronomy.<br />
2001.<br />
Stöckle, C.O., M. Donatelli, <strong>and</strong> R. Nelson.<br />
CropSyst, a cropping systems simulation model.<br />
Europ. J. Agronomy 18, 289-307. 2003.<br />
Topp C.F.E., <strong>and</strong> C.J. Doyle. Simulating the<br />
impact of global warming on milk <strong>and</strong> forage<br />
production in Scotl<strong>and</strong>: 2. The effects on milk<br />
yields <strong>and</strong> grazing management of dairy herds.<br />
Agric. Syst. 52, 243-270. 1996.<br />
Tubiello, F.N., M. Donatelli, C. Rosenzweig <strong>and</strong><br />
C.O. Stöckle. Effects of climate change <strong>and</strong><br />
elevated CO2 on cropping systems: model<br />
predictions at two Italian locations. Eur. J.<br />
Agron. 13, 179-189. 2000.<br />
661
Some Methodological Concepts to Analyse the Role of<br />
IC-tools in Social Learning Processes<br />
Pierre Maurel a , Flavie Cernesson a , Nils Ferr<strong>and</strong> a , Marc Craps b , Pieter Valkering c<br />
a<br />
Cemagref/ENGREF, Montpellier, France<br />
b<br />
Centre for Organisational <strong>and</strong> Personnel Psychology, Katholieke Universiteit Leuven, Belgium<br />
c<br />
ICIS, University of Maastricht, the Netherl<strong>and</strong>s<br />
Abstract: The Water Framework Directive requires to include public besides the water experts <strong>and</strong> policy<br />
makers into development <strong>and</strong> implementation of River Basin Management (RBM) plans (see Article 14). In such<br />
a context, the EU research project HarmoniCOP, studies a method to improve Public Participation based on<br />
Social Learning (SL) concepts. SL refers to the growing capacity of a social network to develop <strong>and</strong> perform<br />
collective actions. The different stakeholder groups in a basin are supposed to realize that a complex issue such<br />
as RBM can be better resolved in a collective way, taking account the diversity of interests, of mental frames, of<br />
knowledge <strong>and</strong> relying on disseminated information <strong>and</strong> knowledge. Information <strong>and</strong> Communication tools (ICtools)<br />
can play an important role to support the Social Learning dimension of the Public Participation. This paper<br />
presents a HarmoniCOP project synthesis of the definition of different concepts <strong>and</strong> proposes a framework of<br />
analysis. A first part consists in a preliminary qualitative characterization of the role of the IC-tools stemming<br />
from a bibliography analysis. Twenty IC-tools are already inventoried <strong>and</strong> four criteria are proposed:<br />
communication direction, public size, usage purpose (management of information <strong>and</strong> knowledge, elicitation of<br />
perspectives, interaction support <strong>and</strong> simulation), phases in the PP process. A second part presents a framework<br />
of analysis based on a joint approach of psychologists <strong>and</strong> engineering sciences experts. This framework will be<br />
tested in a number of empirical investigations to assess the tools used in historical <strong>and</strong> real-time case studies<br />
from three perspectives: their technical characteristics, their impact on PP <strong>and</strong> SL <strong>and</strong> their usability as perceived<br />
by the users. Finally, we present some perspectives concerning expected outcomes of the HarmoniCOP project.<br />
Keywords: IC-tools; Public participation; Social Learning; Water Framework Directive<br />
1. INTRODUCTION<br />
1.1. The Water Framework Directive <strong>and</strong> Public<br />
Participation<br />
In Europe, the Water Framework Directive (WFD)<br />
2000/60/EC of 23 October 2000 established a<br />
framework for Community action in the field of<br />
water policy. The key objective of the directive is to<br />
achieve by 2015 “good water status” for all<br />
European surface <strong>and</strong> underground waters. One of<br />
the five main instruments that will be used to reach<br />
this objective is Public Participation (PP).<br />
The main article concerning Public Participation is<br />
Article 14 stating: “River basin management plans<br />
Member States shall encourage the active<br />
involvement of all interested parties in the<br />
implementation of this Directive, in particular in the<br />
production, review <strong>and</strong> updating of the river basin<br />
management plans.”<br />
First of all, we have to define the terms “public<br />
participation” <strong>and</strong> more precisely “active<br />
involvement of interested parties”.<br />
PP can generally be defined as allowing people to<br />
influence the outcome of plans <strong>and</strong> working<br />
processes. Several benefits but also drawbacks can<br />
be expected from PP, as described in a recent<br />
synthesis [Drafting Group 2002, Mostert 2003]. This<br />
synthesis also shows that PP is necessary but it has<br />
to be organized in order to make it work, especially<br />
in term of level of PP <strong>and</strong> type of public to involve.<br />
Different levels of PP may be considered, based on<br />
[Arstein 1969]’s “ladder of citizen participation”:<br />
1- Information: the public gets/has access to<br />
information, which is a basic condition for all<br />
levels of PP.<br />
2- Consultation: the views of the public are sought.<br />
3- Discussion: real interaction takes place between<br />
the public <strong>and</strong> the government.<br />
4- Co-designing: the public takes an active part in<br />
developing policy or designing projects.<br />
5- Co-decision-making: The public shares decisionmaking<br />
powers with the government.<br />
6- Decision-making: the public performs public<br />
tasks independently.<br />
“Active involvement” integrates here levels 4, 5 <strong>and</strong><br />
6.<br />
662
Several types of public can be distinguished among<br />
the broad term “public”:<br />
The WFD refers to the term “public” with respect to<br />
information <strong>and</strong> consultation levels of PP. In this<br />
case, the definition given by Art. 2(d) of the<br />
2001/42/EC SEIA Directive (European Union, the<br />
European Parliament, The Council 2001) is<br />
applicable: “One or more natural or legal persons,<br />
<strong>and</strong>, in accordance with national legislation or<br />
practice, their associations, organisations or<br />
groups.” Government bodies are usually not<br />
considered to be part of the "public”.<br />
The terms “stakeholder” or “interested party” are<br />
used concerning the active involvement level. This<br />
category of actor integrates any person, group or<br />
organisation with an interest or “stake” in an issue<br />
either because they will be affected or because may<br />
have some influence on its outcome. The guidance<br />
document for PP related to the WFD proposes a<br />
typology of stakeholders involved in River Basin<br />
Management (RBM): professionals, authorities <strong>and</strong><br />
elected people, local groups <strong>and</strong> non-professional<br />
organised entities <strong>and</strong> finally, individual citizens,<br />
farmers <strong>and</strong> companies representing themselves. We<br />
can also add to this typology the “experts”<br />
(government <strong>and</strong> water authorities experts,<br />
academics, private consultants).<br />
For the “public”, levels of PP 1 <strong>and</strong> 2 only are<br />
required by the WFD <strong>and</strong> levels 3 <strong>and</strong> 4 may be<br />
considered as best practice <strong>and</strong> should be<br />
encouraged. For the “stakeholders”, level 4 is the<br />
minimum required by the WFD <strong>and</strong> levels 5 <strong>and</strong> 6<br />
have to be promoted.<br />
1.2. Social Learning in the HarmoniCOP Project<br />
Considering the interest as well as the limits of<br />
traditional PP, the EU research project<br />
HarmoniCOP 1 studies a new approach of PP called<br />
Social Learning (SL) which promotes collective<br />
actions within social networks [Craps et al. 2003a].<br />
This concept is represented in figure 1.<br />
RBM is considered as a social-relational activity<br />
[part 2.2 of figure 1] (interests, water practices,<br />
information, knowledge, funds spread over many<br />
actors) <strong>and</strong> a complex technical task [2.3], both<br />
cannot be separated. SL corresponds both to this<br />
participatory social/technical process [2] as well as<br />
to the outcomes of this process [3]. It takes place in<br />
a specific context [1] in terms of the governance<br />
structure (actors, regulation <strong>and</strong> cultural norms) <strong>and</strong><br />
the river basin environment. This context can be<br />
affected in turn by the outcomes [4]. This collective<br />
1 HarmoniCOP - Harmonising COllaborative<br />
Planning - http://www.harmonicop.info/<br />
problem solving approach requires that the actors<br />
meet each other, develop relational practices [2.1].<br />
The quality of these relational practices is<br />
fundamental from a SL perspective: The different<br />
stakeholder groups in a river basin learn to take into<br />
account the diversity of interests, of mental frames,<br />
of knowledge <strong>and</strong> relying on disseminated<br />
information <strong>and</strong> knowledge, <strong>and</strong> may be realize that<br />
complex issues like RBM are better resolved then.<br />
4.<br />
Feedback<br />
1.1.<br />
Governance structure<br />
2.2.<br />
Social<br />
involvement<br />
3.1.<br />
Relational qualities<br />
1. Context<br />
2. Process<br />
2.1.<br />
Relational<br />
practices<br />
3. Outcomes<br />
Figure 1. Graphical framework of the Social<br />
Learning concept in HarmoniCOP<br />
2. IC-TOOLS AS FACILITATING<br />
MECHANIMS FOR PP AND SL<br />
1.2.<br />
Natural environment<br />
,<br />
2.3.<br />
Content<br />
management<br />
3.2.<br />
Technical qualities<br />
This context raises the crucial issue of information<br />
design, storage <strong>and</strong> retrieval <strong>and</strong> communication<br />
between stakeholders in ways that are relevant for<br />
them <strong>and</strong> that allows collective learning [Rool 2004,<br />
Woodhill 2004]. Effective communication is all the<br />
more essential as PP is highly time-consuming due<br />
to the increasing number of interactions <strong>and</strong> the<br />
difficulties to combine expert <strong>and</strong> non-expert<br />
knowledge, even if this process is fruitful [Pahl-<br />
Wostl 2002].<br />
2.1. Definition of IC-tools in the context of SL<br />
Within HarmoniCOP project, an Information <strong>and</strong><br />
Communication Tool (IC-tool) is defined as a<br />
material artefact, device or software, that can be<br />
seen <strong>and</strong>/or touched, <strong>and</strong> which is used in a<br />
participatory process to facilitate Social Learning. It<br />
supports interaction between stakeholders through<br />
two-way communication processes.<br />
The term “information” is used here in a more<br />
general meaning than its strict definition. It also<br />
includes data, knowledge <strong>and</strong> points of view that are<br />
exchanged between actors on a given issue.<br />
663
The term “communication” can be defined here as<br />
social interaction through messages [Fisker 1990].<br />
This is much more than the exchange of information,<br />
but also a mean to reflect <strong>and</strong> reinforce social<br />
relations or "communities". New communication<br />
patterns can help to build up new communities.<br />
Within these communities, new representations of<br />
reality <strong>and</strong> new "meanings" can develop.<br />
2.3. List of IC-tools<br />
After a literature review <strong>and</strong> a comparison of usage<br />
situation in the different countries involved in<br />
HarmoniCOP, a list of tools has finally been<br />
established (see table 1).<br />
Artefacts<br />
Info System / <strong>Software</strong><br />
- Questionnaire * - Information system<br />
- Maps * , photos, images - GIS *<br />
- 3D scale model * - Scenario tool *<br />
- Conceptual models - Multicriteria analysis tool *<br />
For data base<br />
- Simulation tool<br />
For systems dynamic - Spreadsheet<br />
- Cognitive mapping tool * - Decision Support System<br />
- Actors mapping tool * - IA models *<br />
- Management of comments - Internet<br />
- Role playing game * - Web information<br />
Devices<br />
- Forum communities<br />
- Interactive white board * - CSDM<br />
- Board game * - Web mapping<br />
- Group Support System *<br />
Table 1. List of IC-tools<br />
*<br />
an IC-tool index card is presented in [Maurel 2003]<br />
3. CRITERIA OF CATEGORIZATION<br />
To categorize IC-tools, four main criteria have been<br />
identified as useful for those who will have to<br />
organize in practice the WFD PP process.<br />
3.1. Communication direction<br />
This criterion allows to determine the attractiveness<br />
of the IC-tool according to the direction of<br />
communication: top-down (from the leading team to<br />
the stakeholders <strong>and</strong> the general public), bottom-up<br />
or both (bi-directional).<br />
3.2. Public size<br />
We have distinguished two types of public size<br />
where IC-tools can be used to support<br />
communication. The first type corresponds to small<br />
working groups (single or multiparty) where people<br />
generally meet face to face or exchange through<br />
specified tools. The second type corresponds to the<br />
general public. Most of the time, the relational<br />
events are space-time distributed.<br />
3.3. Usage purpose<br />
Four main purposes have been identified :<br />
- Management of information <strong>and</strong> knowledge<br />
The corresponding IC-tools aim to store, retrieve,<br />
analyse, display <strong>and</strong> disseminate information.<br />
This is one of the traditional substantive functions of<br />
some IC-tools but in the context of SL <strong>and</strong> PP, it<br />
raises important issues. How does one deal with the<br />
sharing of information between actors belonging to<br />
different communities of knowledge <strong>and</strong> of practice<br />
with multiple perspectives, points of view,<br />
vocabulary, skills ? How are uncertainties addressed<br />
? How to keep the memory of relational events <strong>and</strong><br />
make it accessible <strong>and</strong> underst<strong>and</strong>able to nonparticipants<br />
? How to respect the confidentiality<br />
rules that have been adopted ? How to assure well<br />
balanced, or at least well accepted informational<br />
power <strong>and</strong> resources among the actors ?<br />
- Perspective elicitation<br />
Here, the IC-tools help to elicit perspectives <strong>and</strong><br />
behaviours of stakeholders, to make them explicit to<br />
the others. This may be the most challenging <strong>and</strong><br />
innovative relational function of IC-tools to<br />
contribute to SL. However this function depends not<br />
only on the intrinsic properties of the tool but also<br />
on the way it is designed <strong>and</strong> used within<br />
“transitional spaces” [Craps et al 2004] that cross the<br />
boundaries between communities of knowledge <strong>and</strong><br />
of practice. To be able to fulfil this function, an ICtool<br />
should have all or part of the properties of what<br />
[Star et al 1989] call boundary objects <strong>and</strong> [Vinck et<br />
al 1995] call intermediary objects:<br />
• common point of reference for conversations.<br />
• support <strong>and</strong> reveal different representations of<br />
the reality, meanings, points of views.<br />
• means of translation between individuals or<br />
groups belonging to different communities of<br />
knowledge. Even if a full translation seems<br />
utopic, the structure of a boundary object is<br />
shared enough to work together.<br />
• means of coordination <strong>and</strong> alignment.<br />
• working arrangements, adjusted as needed <strong>and</strong><br />
not imposed by one community or by outside<br />
st<strong>and</strong>ards.<br />
• plastic enough to be transformable (an “open”<br />
object <strong>and</strong> not a “closed” object) during the<br />
interaction process.<br />
• trace of the collaborative process (successive<br />
proposals of transformation, successive states of<br />
the final output, comments, etc).<br />
• help to manage uncertainties (through larger<br />
number of solutions found <strong>and</strong> evaluated,<br />
development of trust, increase of knowledge).<br />
- Interaction support<br />
The objectives of using IC-tools are to support the<br />
interactions between actors, to improve<br />
communication <strong>and</strong> bring the individuals together.<br />
664
This function complements the previous one <strong>and</strong><br />
raises also central issues related to SL. It depends<br />
also on the way the tool is implemented <strong>and</strong> used by<br />
the participants.<br />
- Simulation<br />
The scope of IC-tools here is to simulate the<br />
dynamics of RB systems for environmental, <strong>and</strong>/or<br />
technical <strong>and</strong>/or economical aspects.<br />
This is also a function expected traditionally of ICtools<br />
such as DSS, Integrated Assesment models,<br />
qualitative modelling techniques.<br />
3.4. Phases in the PP process<br />
We have chosen to comply with the four phases<br />
proposed in the EU guidance document for Public<br />
Participation: 1) starting organisation, 2) actors <strong>and</strong><br />
context analysis, 3) diagnosis of the situation, 4)<br />
search for solutions, <strong>and</strong> two additional phases: 5)<br />
implementation <strong>and</strong> 6) follow-up <strong>and</strong> feed-back.<br />
A first qualitative classification of IC-tools using the<br />
four criteria previously described <strong>and</strong> a three level<br />
scale (0: low interest, 1: medium interest, 2: high<br />
interest) is presented in [Maurel 2003].<br />
4. FRAMEWORK OF ANALYSIS<br />
The following framework of analysis is based on a<br />
joint approach of psychologists <strong>and</strong> engineering<br />
sciences experts. It will be tested in 2004 <strong>and</strong> 2005<br />
in a number of empirical investigations to assess the<br />
tools used in historical <strong>and</strong> real-time case studies<br />
(HarmoniCOP WP5).<br />
The evaluation criteria are derived from<br />
HarmoniCOP discussions <strong>and</strong> from literature on the<br />
evaluation of PP [Webler 2001], on the evaluation of<br />
tools [Ubbels et al 2000], on the factors of<br />
technology acceptance <strong>and</strong> usability [Legris 2003],<br />
<strong>and</strong> on participation in integrated assessment <strong>and</strong><br />
modelling for the environment [Pahl-Wostl 2002].<br />
Based on these criteria, a list of questions <strong>and</strong> their<br />
underlying assumptions have been produced <strong>and</strong><br />
included in an instrument called “Social Learning<br />
Pool of Questions” (PoQ) [Craps et al 2003b].<br />
The PoQ consists of three layers:<br />
• What : A list of general questions, summarizing<br />
the main issues that have to be considered in<br />
relation to SL in RBM. The structural order of<br />
the questions follows the conceptual framework<br />
that is demonstrated in figure 1.<br />
• Why : A short explanation of the underlying<br />
assumptions for these questions.<br />
• How : Examples of concrete <strong>and</strong> clear questions<br />
that can be used during the interview of<br />
stakeholders.<br />
Within the PoQ, IC-tools will be analyzed from<br />
three perspectives: their technical characteristics <strong>and</strong><br />
usage situation, their impact on PP <strong>and</strong> SL <strong>and</strong> their<br />
usability as perceived by the users.<br />
4.1. IC-tools characteristics <strong>and</strong> usage situation<br />
A charting procedure, included in the PoQ, has been<br />
established to facilitate the collection <strong>and</strong> analysis of<br />
information [Ferr<strong>and</strong> et al. 2004].<br />
A first series of factual criteria concerns the usage<br />
situation of IC-tools for each relational event in the<br />
PP process :<br />
• list of ICtools that have been used ;<br />
• phase(s) in the process ;<br />
• main usage purposes (both for relational <strong>and</strong><br />
substantive tasks) ;<br />
• relations between the actors <strong>and</strong> the IC-tool :<br />
who promoted or prevented the use of the tool,<br />
who manages it, who provides the<br />
data/information/knowledge, who has access to<br />
it or to its informational content ?<br />
Then, for each IC-tool that has been identified, a<br />
second series of criteria addresses the technical<br />
characteristics of the tool. These criteria are<br />
synthetized in an IC-tool index card divided in 5<br />
main sections:<br />
- General characteristics: Each tool is<br />
characterized by its type, its complexity, its<br />
availability, <strong>and</strong> its current stage of development.<br />
- Usage purposes: The IC-tool uses are defined<br />
according to the context of the participatory<br />
process <strong>and</strong> the relational <strong>and</strong>/or substantive tasks<br />
to be performed. Four main usage purposes (with<br />
the corresponding functionalities <strong>and</strong> conditions of<br />
use) are a priori proposed: information <strong>and</strong><br />
knowledge management, interaction support,<br />
perspective elicitation, simulation (see chapter 3).<br />
These functionalities represent the potentials of<br />
the tool. It will be possible to observe a difference<br />
between the potentials of the tool <strong>and</strong> the effective<br />
use: restrictive use, or use for other purposes.<br />
- Sustainability: Some conditions are necessary to<br />
guarantee a minimal sustainability of the tool:<br />
direct or indirect use by the actors, availability of<br />
use support, degree of openness, <strong>and</strong> management<br />
of the monitoring/reporting or tracability.<br />
- Informational output description: Content <strong>and</strong><br />
formal aspects.<br />
- Uncertainties management: The information is<br />
rarely an original quantitative data set. There are<br />
numerous sources of uncertainty, particularly in<br />
ecosystem management, linked to variability (of<br />
natural processes, human behaviour, social<br />
dynamics, etc.) <strong>and</strong> to limited knowledge (lack of<br />
observations, practically immeasurable data, etc.)<br />
Therefore, an important function of IC-tools is to<br />
be able to h<strong>and</strong>le <strong>and</strong> to communicate uncertainty.<br />
665
The stake is to convince all participants that the<br />
decision process is at least as important as the<br />
decision output, because the output will have to be<br />
modified in the future due to uncertainty.<br />
4.2. Impact of IC-tools on PP <strong>and</strong> SL<br />
The sharing of informational resources among<br />
the participants<br />
A first issue concerns the analysis of the allocation<br />
of IC-tools resources (tools, skills, facilitators,<br />
training, data, information, time, money) among the<br />
participants during the RBM PP process. The<br />
assumption is that a certain degree of equality<br />
among the parties concerning their informational<br />
resources is necessary for a credible PP process. A<br />
related point is to analyse whether there is a gradual<br />
emergence of formal or informal agreements<br />
between stakeholders concerning the sharing of<br />
resources to participate, as an indicator of SL.<br />
Influence of IC-tools on the relational quality<br />
among the participants<br />
Our assumption is that IC-tools can help improve the<br />
communication between the participants at different<br />
organizational scales (within a working group,<br />
between working groups, between a representative<br />
<strong>and</strong> his constituencies, between the project team <strong>and</strong><br />
the general public, between institutions).<br />
Another point is that some IC-tools or some specific<br />
tasks related to a tool may help share the same<br />
language or underst<strong>and</strong> each other or at least, make<br />
explicit the differences of representation among the<br />
participants (i.e. thesaurus, database dictionary, ...).<br />
The last assumption is that participating in the codesign<br />
of an IC-tool facilitates the acknowledgement<br />
of both expert <strong>and</strong> local knowledge <strong>and</strong> offers a<br />
positive context for bi-directional communication<br />
<strong>and</strong> mutual underst<strong>and</strong>ing. A distinction will have to<br />
be made between tools that are imposing <strong>and</strong><br />
structuring certain interaction characteristics, <strong>and</strong><br />
tools that leave more freedom among participants.<br />
Influence of IC-tools on the technical quality of<br />
the PP process outcomes<br />
The assumption is that IC-tools may help the<br />
involved actor network to resolve better the<br />
substantive river basin issues through different<br />
ways:<br />
• by improving the amount <strong>and</strong> quality of<br />
knowledge on the river basin thanks to better<br />
access to information, to a mutual enrichment<br />
between expert <strong>and</strong> local knowledge;<br />
• by allowing to test more alternatives during the<br />
“search of solutions” phase;<br />
• by allowing a better ranking of alternatives (e.g.<br />
through the multi-criteria analysis process);<br />
• by integrating better the different components of<br />
a complex river basin system (e.g. models able<br />
to link surface <strong>and</strong> subsurface water issues, ...).<br />
The interest of co-designed activities developed in<br />
the previous section is still relevant for the technical<br />
quality issue.<br />
We also expect that the quality of the relations<br />
among the actors is reflected in an enhanced quality<br />
<strong>and</strong> satisfaction with the technical outcomes of the<br />
process; <strong>and</strong> the other way around: the better the<br />
joint technical solutions, the more the actors get<br />
motivated to invest in their interaction.<br />
4.3. Perceived usability of IC-tools<br />
By perceived usability, we refer to the degree to<br />
which the user expects the tool to fit a given purpose<br />
in a given context (characteristics of the physical,<br />
organisational <strong>and</strong> social environment).<br />
Four components of usability have been selected :<br />
• The learnability: amount of things that have to<br />
be learnt before using a tool.<br />
• The effectiveness: accuracy <strong>and</strong> completeness<br />
with which users achieve specific goals.<br />
• The efficiency: amount of resources consumed<br />
in performing a task.<br />
• The satisfaction: users’ subjective reactions in<br />
performing a task (absence of discomfort,<br />
positive attitudes towards the use).<br />
The perceived usability predicts “attitude toward<br />
using” the tool, defined as the user’s desirability of<br />
her or his using the system. This attitude itself<br />
influences the individual’s behavioral “intention to<br />
use the tool”.<br />
People perceive the usability of a tool through<br />
indirect sources (‘peers’ or champions opinions,<br />
technical documentation) or practical experiences. In<br />
this second case, the level of usability for a given<br />
tool will depend on its performances to fulfil a<br />
substantive <strong>and</strong>/or relational task in a specific<br />
context. This will influence the decision to use or<br />
not to use these IC-tools again in the future.<br />
5. PERSPECTIVES<br />
A first perspective concerns the lessons that will be<br />
learned from the national studies (HarmoniCOP<br />
WP4) <strong>and</strong> the historical <strong>and</strong> real-time case studies<br />
(HarmoniCOP WP5) analysed through the Pool of<br />
Questions. The results will shown which IC-tools<br />
have been used, their usage situation as well as the<br />
relational <strong>and</strong> substantive outputs perceived by the<br />
users or observed by WP5 teams. They will help to<br />
assess the gap between the potentials of the tools,<br />
the real uses <strong>and</strong> the perceived usability. Our<br />
preliminary qualitative categorization of IC-tools<br />
will be updated according to these results. A crosscomparison<br />
between the different case studies will<br />
also contribute to better underst<strong>and</strong> the economical,<br />
technical, institutional <strong>and</strong> cultural factors that might<br />
affect the usability of the tools. Finally, the case<br />
666
studies will allow to verify our hypothesis on the<br />
importance of sharing informational resources <strong>and</strong><br />
of co-designing IC-tools.<br />
Our major expectation is to be able through these<br />
findings to make more explicit the relational<br />
functions of the IC-tools <strong>and</strong> their impact on SL.<br />
A second more practical perspective derived from<br />
the previous one concerns the production of a<br />
h<strong>and</strong>book. It will allow the WFD practitioners to<br />
tailor a participatory RBM process to<br />
regional/regional conditions. Concerning IC-tools, it<br />
will help the SL facilitators to answer concrete<br />
questions such as : What are the relational <strong>and</strong><br />
substantive functions of a tool ? How should it be<br />
used ? Which resources <strong>and</strong> skills are required ?<br />
What is its applicability in the different phases of the<br />
PP process ? When was it used <strong>and</strong> who might be<br />
contacted for additional information ?<br />
This h<strong>and</strong>book is considered by HarmoniCOP as a<br />
mean to make underst<strong>and</strong>able the concept of<br />
“learning together for managing together” <strong>and</strong> to put<br />
it effectively into practice.<br />
6. ACKNOWLEDGEMENTS<br />
Research conducted under financial contribution of<br />
EC under the contract n° EVK1-CT-2002-00120.<br />
The contributors to HarmoniCOP WP3 are:<br />
• Cemagref: O. Barreteau A. Boutet, F.<br />
Cernesson, N. Ferr<strong>and</strong>, , P. Garin, P. Maurel.<br />
• Colenco Power Engineering Ltd: L.<br />
Schlickenrieder.<br />
• Delft University of Technology: L. Carton, B.<br />
Enserink, D. Kamps, E. Mostert.<br />
• ICIS, University of Maastricht: P. Valkering.<br />
• K.U.Leuven - COPP: M. Craps.<br />
• Seecon Deutchl<strong>and</strong> GmbH : M. Hare<br />
• USF, University of Osnabrück: C. Pahl-Wostl.<br />
• WL Delft Hydraulics: P. Gijsbers, H. van der<br />
Most.<br />
7. REFERENCES<br />
Arnstein, S., A ladder of citizen participation in the<br />
USA. Journal of the American Institute of<br />
Planners. p. 216-224, 1969.<br />
Craps, M., ed., Social Learning in River Basin<br />
Management, WP2 report of the HarmoniCOP<br />
project, 70p, 2003a.<br />
Craps, M., <strong>and</strong> P. Maurel, ed., Social Learning Pool<br />
of Questions. An instrument to diagnose Social<br />
Learning <strong>and</strong> IC-tools in European River Basin<br />
Management, Combined WP2/WP3 report of<br />
the HarmoniCOP project, 65p, 2003b.<br />
Craps, M., A. Dewulf, M. Mancero, E. Santos, R.<br />
Bouwen, Generating transitional space between<br />
professional <strong>and</strong> indigeneous communities for<br />
sustainable water management in the Andes,<br />
Journal of Community <strong>and</strong> Applied Social<br />
Psychology, to be published, 2004.<br />
Drafting Group, Guidance on public participation in<br />
relation to the Water Framework Directive -<br />
Active involvement, consultation, <strong>and</strong> public<br />
access to information, EU Report, 66 pp. +<br />
annexes, 2002.<br />
Ferr<strong>and</strong>, N., F. Cernesson, P. Maurel, Comparative<br />
charting of IC Tools usage in Social Learning<br />
processes, <strong>International</strong> <strong>Environmental</strong><br />
<strong>Modelling</strong> <strong>and</strong> <strong>Software</strong> (iEMSs) conference,<br />
Osnabrück, Germany, June 14-17, 2004.<br />
Fisker, J., Introduction to Communication Studies,<br />
2 nd ed., Routledge, London-New York, 1990.<br />
Legris P., J.Ingham, P.Collerette, Why do people use<br />
information technology ? A critical review of<br />
the technology acceptance model. Information<br />
& Management 40 (2003) 191-204, 2003.<br />
Maurel, P. ed., Public participation <strong>and</strong> the European<br />
Water Framework Directive. Role of<br />
Information <strong>and</strong> Communication Tools, WP3<br />
report of the HarmoniCOP project, 94p, 2003.<br />
Mostert, E., The challenge of Public Participation.<br />
Water Policy, 5 (2003) pp. 179-197, 2003.<br />
Pahl-Wostl, C., Participative <strong>and</strong> Stakeholder-based<br />
policy design, evaluation <strong>and</strong> modeling<br />
processes, Integrated Assessment 3:3–14, 2002.<br />
Roll, G., Generation of usable knowledge in<br />
implementation of the European water policy,<br />
In: Langaas, S., <strong>and</strong> J.G. Timmerman, (eds.),<br />
The role <strong>and</strong> use of environmental information<br />
in European transboundary river basin<br />
management, IWA, 258 pp., London, 2004.<br />
Star, S.L., <strong>and</strong> J.R. Greisemer, Institutional Ecology,<br />
“Translations,” <strong>and</strong> boundary objects:<br />
Amateurs <strong>and</strong> professionals in Berkeley’s<br />
Museum of Vertebrate Zoology, Social studies<br />
of Science, 19, 387-420, 1989.<br />
Ubbels, A., J.M. Verhallen, Suitability of decision<br />
support tools for collaborative planning<br />
processes in water resources management.<br />
RIZA, Wageningen university, 47p, 2000.<br />
Vinck, D. <strong>and</strong> A. Jeantet, Mediating <strong>and</strong><br />
Commissioning Objects in the Sociotechnical<br />
Process of Product Design: A Conceptual<br />
Approach,. In MacLean D., P. Saviotti <strong>and</strong> D.<br />
Vinck (eds.): Management <strong>and</strong> New<br />
Technology: Design, Networks <strong>and</strong> Strategy,<br />
COST Social Science Series, Bruxelles, 1995.<br />
Webler, T., Tuler, S., Krueger, R. 2001. What is a<br />
good Public Participation process? Five<br />
perspectives from the public. <strong>Environmental</strong><br />
management 27(3), pp. 435-450.<br />
Woodhill A.J., Dialogue <strong>and</strong> transboundary water<br />
resources management: towards a framework<br />
for facilitating social learning. In: Langaas, S.,<br />
<strong>and</strong> J.G. Timmerman, (eds.), The role <strong>and</strong> use<br />
of environmental information in European<br />
transboundary river basin management, IWA,<br />
258 pp., London, 2004.<br />
667
Tools to Think With? Towards Underst<strong>and</strong>ing the Use<br />
<strong>and</strong> Impact of Model-Based Support Tools<br />
B.S. McIntosh a , R.A.F. Seaton b <strong>and</strong> P.Jeffrey<br />
a<br />
School of Water Sciences, Cranfield University, College Road, Cranfield, Bedfordshire MK43 0AL, UK<br />
b<br />
Seaton Associates, 9 Park Road, Stevington, Bedfordshire MK43 7QD, UK<br />
Abstract: Formal models are an established technology for research in the environmental sciences. For<br />
several years now there has been an effort to enhance the re-usability of computer models for research<br />
purposes <strong>and</strong> to transfer the perceived benefits of formal modelling to environmental planning <strong>and</strong> policymaking.<br />
These efforts have resulted in the creation of a variety of support tools including DSS <strong>and</strong> modelling<br />
frameworks. However, there are a number of issues which may pose barriers to the uptake <strong>and</strong> use of such<br />
tools. We contend that new technologies <strong>and</strong> new techniques for exploring <strong>and</strong> manipulating them have to be<br />
translated into the pre-existing knowledge of user communities before they can be effectively employed. To<br />
explore this proposition we report on research currently being undertaken to gain a better underst<strong>and</strong>ing of<br />
the knowledge processes that influence the response of potential users to model-based support tools in the<br />
context of policy-relevant science research. Importantly we distinguish between conceptual, model <strong>and</strong><br />
software technology - between the approach of ‘Time Geography’, the case-study models, <strong>and</strong> Time<br />
Geographical model / database analysis software being developed. Using this separation, the impact of Time<br />
Geography is being researched as an innovation with potential to influence both problem conceptualisation<br />
<strong>and</strong> formal analysis. We propose that taking a knowledge dynamics perspective on the use of formal models<br />
in environmental policy yields useful insights into their potential benefits <strong>and</strong> limitations. Through this<br />
perspective we seek to explore what might make a support tool ‘good to think with’.<br />
Keywords: support tools; models; knowledge transfer; re-use; receptivity; Time Geography.<br />
1. INTRODUCTION<br />
Formal mathematical <strong>and</strong> computable models are a<br />
well-established method in the natural sciences. As<br />
tools for underst<strong>and</strong>ing they provide a valuable<br />
means of knowing about the world <strong>and</strong> about<br />
theories of the world. Further, they are valuable<br />
sources of knowledge in their own right [Morrison<br />
& Morgan 1999].<br />
The potential of model-based methods as a source<br />
of advice for tackling management problems is<br />
also well-established with the concept of the<br />
decision support system [Sage 1991]. This<br />
potential has been recognised within the<br />
environmental planning <strong>and</strong> policy research<br />
communities [Engelen et al. 1997, van Daalen et<br />
al. 2002, Jakeman & Letcher 2003]. The apparent<br />
success of formal models as epistemological<br />
devices combined with pressures to perform<br />
environmental research in a cost-effective <strong>and</strong><br />
productive way has given rise to a need to re-use<br />
models [Argent 2004, Oxley et al. in press] rather<br />
than having multiple environmental model<br />
developers ‘reinventing the wheel’ with regards to<br />
common issues <strong>and</strong> phenomena.<br />
As a result numerous ‘integrated models’,<br />
‘modelling frameworks’ <strong>and</strong> ‘environmental<br />
decision support systems’ have been produced by<br />
the research community over the past decade both<br />
to facilitate the business of environmental research<br />
<strong>and</strong> to provide information for planning <strong>and</strong> policy<br />
[Rizzoli & Young 1997, Argent 2004]. We shall<br />
collectively term such computer-based devices<br />
‘support tools’ for the purposes of this paper<br />
regardless of whether they are designed to support<br />
research by reducing model development<br />
redundancy or to support planning through<br />
providing a means of accessing, exploring <strong>and</strong><br />
applying scientific knowledge.<br />
Many support tool technologies are based upon or<br />
involve re-using the conceptual structures; the<br />
mathematical, rule-based or algorithmic<br />
formalisations, or; the software implementations of<br />
668
existing formal models to address (i) new instances<br />
of previously encountered (sets of) issues; (ii)<br />
previously encountered (sets of) issues in a<br />
different way, or; (iii) newly encountered (sets of)<br />
issues. As such, re-use may entail a requirement to<br />
integrate formal models <strong>and</strong> software tools that<br />
have never previously been integrated. More<br />
importantly from the perspective of this paper, reuse<br />
may need to be undertaken by groups other<br />
than the original model developer(s). This will be<br />
particularly true if researchers are to use modelling<br />
frameworks or model libraries developed by other<br />
teams to aid their work, or if planning / policy endusers<br />
are to use DSS to select <strong>and</strong> use appropriate<br />
models <strong>and</strong>/or databases to address management<br />
issues as they evolve.<br />
Technical issues aside (for these are outside the<br />
scope of this paper), re-use of models by groups<br />
not originally involved in the design of those<br />
models can be considered as a process of<br />
innovation, of knowledge transfer from one group<br />
(the designers) to another (the users). We use<br />
groups in a broad sense, as being differentiated in<br />
various ways from organisational affiliation <strong>and</strong><br />
purpose to disciplinary background, from social<br />
norms <strong>and</strong> cultural preferences to access to<br />
computers, training <strong>and</strong> support.<br />
Both conceptual <strong>and</strong> software technologies are<br />
elements that are involved in <strong>and</strong> will influence the<br />
process of re-use, <strong>and</strong> in doing so affect the way in<br />
which support tools are used <strong>and</strong> the impact that<br />
they may exert on tasks performed by end-users.<br />
Our view of technology is therefore also broad <strong>and</strong><br />
includes concepts <strong>and</strong> methods as well as physical<br />
or software artefacts, for all can be used as tools to<br />
assist with particular tasks.<br />
This paper reports on research currently being<br />
undertaken to elicit, explore <strong>and</strong> underst<strong>and</strong> the<br />
processes involved in re-using conceptual <strong>and</strong><br />
software-based model technologies <strong>and</strong> the way in<br />
which re-use influences the process of performing<br />
policy-relevant research. The approach being taken<br />
provides a novel contribution to the debate on how<br />
best to develop <strong>and</strong> deploy support tools in<br />
environmental research <strong>and</strong> policy.<br />
2. ISSUES WITH SUPPORT TOOL<br />
(RE)USE<br />
We recognise the positive roles that formal models<br />
can play within <strong>and</strong> between different groups such<br />
as providing a common language for dialogue, as<br />
tools for supporting argument <strong>and</strong> for performing<br />
analyses <strong>and</strong> for raising issue awareness [Morrison<br />
& Morgan 1999, van Daalen et al. 2002]. However<br />
when individuals use support tools it is pertinent to<br />
consider a number of issues.<br />
Research into human-computer interactions (HCI)<br />
has produced various conceptual models of<br />
computer-based tool use. Two relevant HCI<br />
models are the socio-technical systems model of<br />
Eason [1991] <strong>and</strong> the factor model of Preece et al.<br />
[1994]. Both emphasise the variety of influences<br />
which affect the way in which computers are used<br />
to accomplish tasks. Further, both models indicate<br />
that to underst<strong>and</strong> the relationships between users,<br />
computer tools <strong>and</strong> task performance, that the<br />
relationships between factors including task nature,<br />
task constraints, tool characteristics, organisational<br />
setting <strong>and</strong> the user must first be understood.<br />
Within mechanical engineering, Busby [1999]<br />
notes that the traditional views of design re-use<br />
problems are technical in orientation. The<br />
motivations for re-using designs are clear –<br />
reduction of time <strong>and</strong> money spent, avoiding<br />
redundancy, avoiding error <strong>and</strong> providing greater<br />
consistency in product functionality <strong>and</strong><br />
maintenance. However Busby [1999] finds that<br />
design re-use is often problematic for a variety of<br />
reasons including inhibited transfer caused by the<br />
need to extensively modify existing components<br />
during re-use (noted in the context of<br />
environmental decision support by Oxley et al. in<br />
press); error arising from incorrect interpretation of<br />
existing designs; idiosyncratic designer or user<br />
preferences, <strong>and</strong>; a preference for innovation <strong>and</strong><br />
novelty that may be culturally embedded in the<br />
design or user organisation.<br />
In the physical sciences the objective, in general<br />
terms, is to progressively hypothesise <strong>and</strong> test with<br />
the aim of producing a convergence of<br />
underst<strong>and</strong>ing over time leading to the production<br />
of robust, accurate <strong>and</strong> precise models. Questions<br />
of ontology <strong>and</strong> whether the right variables are<br />
included in a model are perhaps less contentious.<br />
However in the environmental sciences this is not<br />
the case. Given the pressure to better integrate<br />
human <strong>and</strong> environmental issues [Costanza 2003]<br />
for research <strong>and</strong> planning / policy, environmental<br />
science is no longer concerned simply with the<br />
‘natural’, it is now concerned with linking these<br />
phenomena to social, economic, infrastructure <strong>and</strong><br />
governance structures <strong>and</strong> processes.<br />
However, our underst<strong>and</strong>ing of how social systems<br />
interact with natural or semi-natural systems<br />
(water, air, soil) is poor. The emerging field of<br />
socio-natural systems science [Winder 2000, van<br />
der Leeuw 2001] aims to address these<br />
relationships but issues of complexity, evolutionary<br />
change [Funtowicz & O’Connor 1998] <strong>and</strong><br />
‘wicked problems’ [Rittel & Webber 1973] can be<br />
669
expected to prevent our knowledge approaching<br />
the precision <strong>and</strong> accuracy of the physical sciences.<br />
Under these conditions questions of ontology <strong>and</strong><br />
epistemology, of representation, method <strong>and</strong><br />
meaning become central <strong>and</strong> can be the source of<br />
great controversy. The danger lies in assuming that<br />
because we can formally represent, manipulate <strong>and</strong><br />
re-use models of the world through software<br />
technology that we are indeed exploring the same<br />
meanings <strong>and</strong> ‘genuinely’ re-using scientific<br />
knowledge. Problems of definition already appear<br />
<strong>and</strong> are acknowledged in research looking to<br />
integrate <strong>and</strong> re-use relatively simple bio-physical<br />
models made by different developers [Argent<br />
2004]. We can expect such problems to get worse<br />
as environmental science continues to exp<strong>and</strong> into<br />
the social <strong>and</strong> socio-natural domains.<br />
Despite these difficulties, ongoing model <strong>and</strong><br />
support tool research in the environmental sciences<br />
appears focussed primarily on technical, often<br />
software-oriented concerns [Rizzoli & Young<br />
1997, Argent 2004]. Little attention is being paid<br />
to the contextual issues that accompany the use of<br />
support tools, except as motivating factors (e.g.<br />
improving the applicability of science to<br />
management at minimal cost). Neither is much<br />
attention being paid to the impact that using these<br />
tools may have on the tasks performed by different<br />
end-user groups. It does not necessarily follow that<br />
re-using a piece of software equates to effective<br />
transfer <strong>and</strong> re-use of the underst<strong>and</strong>ing intended<br />
by a model or support tool developer. Knowledge<br />
is more than computer code.<br />
In the following sections we shall describe a<br />
programme of research intended to both better<br />
underst<strong>and</strong> the impacts of specific conceptual <strong>and</strong><br />
support tool innovations on policy-relevant<br />
environmental science research, <strong>and</strong> to explore the<br />
utility of methods inspired by work in knowledge<br />
dynamics for assessing innovation impact in the<br />
context of support tool use.<br />
3. TIME GEOGRAPHY AS AN<br />
INNOVATION<br />
The particular innovation that we are assessing is<br />
called Time-Geography (TG), <strong>and</strong> it represents an<br />
approach that became influential in the 1950's due<br />
to the pioneering work of the Swedish geographer<br />
Torsten Hägerstr<strong>and</strong> [Hägerstr<strong>and</strong> 1985, Lenntorp<br />
1999, Winder 2003]. Time-Geography provides a<br />
coherent ontological framework within which to<br />
explore the effects of spatio-temporal constraints<br />
on the behaviour of individuals <strong>and</strong> to underst<strong>and</strong><br />
how new socio-economic structures <strong>and</strong><br />
environmental dynamics may emerge at higher<br />
scales as a result.<br />
The project within which this research exists has<br />
set out to apply <strong>and</strong> empirically evaluate the impact<br />
<strong>and</strong> utility of TG methods for analysing <strong>and</strong><br />
interpreting three policy-relevant case-studies. The<br />
case-studies are looking at regional l<strong>and</strong>-use <strong>and</strong><br />
water supply infrastructure planning in the UK;<br />
European inter-urban migration, <strong>and</strong>; l<strong>and</strong>-use <strong>and</strong><br />
sustainable agriculture in Spain. The project is<br />
concerned with transferring knowledge from one<br />
domain (Time Geography) to each case-study <strong>and</strong><br />
in doing so of re-using concepts <strong>and</strong> methods.<br />
A major part of the work is to produce a generic<br />
tool for analysing TG data from both simulation<br />
experiments <strong>and</strong> empirical studies. This tool,<br />
named TiGS (‘Time Geographic Analysis System’)<br />
will be used to analyse the output from the models<br />
/ databases produced by each case-study – a<br />
st<strong>and</strong>ardised interface will be used to link models /<br />
databases to TiGS. Once integrated, each casestudy<br />
research team will have access to the<br />
facilities of TiGS to explore their models <strong>and</strong> data.<br />
Each case-study research is therefore being<br />
exposed in parallel to the conceptual innovation of<br />
TG <strong>and</strong> to the support tool innovation of TiGS.<br />
The research programme reported here exists to<br />
identify the extent to which a TG perspective<br />
provides additional insight or emergent knowledge<br />
to each of the policy-relevant case studies. How<br />
will TG <strong>and</strong> TiGS be perceived by, used by <strong>and</strong><br />
exert an impact on the research tasks undertaken?<br />
The next two sections describe how we are<br />
structuring the research <strong>and</strong> our interpretation of<br />
results.<br />
4. RECEPTIVITY – A FRAMEWORK FOR<br />
UNDERSTANDING<br />
Technology assessment <strong>and</strong> knowledge transfer<br />
research articulates response to innovation options<br />
in terms of receptivity [Seaton & Cordey-Hayes<br />
1993, Trott et al. 1995]. The receptivity of<br />
recipients to most innovation is highly variable but<br />
generally poor with high failure rates for uptake.<br />
The two main reasons are inappropriate design of<br />
the innovation <strong>and</strong> the limitations of recipient<br />
adaptivity. New conceptual technologies <strong>and</strong> new<br />
techniques for exploring <strong>and</strong> manipulating them<br />
(like support tools) have to be translated into the<br />
pre-existing knowledge of user communities before<br />
they can be effectively employed. If the knowledge<br />
pre-supposed by a model developer is not<br />
possessed by, does not map well onto, or is<br />
disagreed over in some way by a potential model<br />
670
user then receptivity to the innovation, the model<br />
may be low.<br />
A useful distinction in innovation studies was made<br />
by Seaton & Cordey-Hayes [1993] between how<br />
the innovation looks to the proponent<br />
(Accessibility), how it is made available (Mobility)<br />
<strong>and</strong> how a potential recipient sees it within their<br />
world (Receptivity) – the ‘AMR model’.<br />
Proponents of an innovation tend to emphasise its<br />
beneficial attributes <strong>and</strong> tend thus to assume people<br />
or organisations will readily take advantage of it.<br />
More pro-active proponents will “push” the<br />
innovation towards perceived clients, however the<br />
actual uptake <strong>and</strong> benefit of any particular<br />
innovation to a recipient can vary widely <strong>and</strong> is a<br />
result of a complex set of processes.<br />
A further, complementary four stage model of<br />
receptivity in technology, or more broadly<br />
knowledge, transfer termed the ‘4A model’ was<br />
developed by Trott et al. [1995]. The 4A model is<br />
composed of four stages each representing a set of<br />
processes that contribute to the overall process of<br />
knowledge transfer – awareness, association,<br />
assimilation <strong>and</strong> application. As with the AMR<br />
model, the 4A model does not constitute an<br />
operational description; rather it provides a<br />
conceptual framework for grouping <strong>and</strong> examining<br />
the processes involved in knowledge transfer.<br />
The argument we put forward is that there is a<br />
complementary need to underst<strong>and</strong> innovation<br />
from the recipients’ point of view. The notion of<br />
receptivity is based on the idea that innovation is<br />
not primarily about physical objects or software<br />
artefacts but about new knowledge, <strong>and</strong> about how<br />
an organisation or group of people can adapt<br />
(adaptive capacity) this new knowledge around<br />
their existing knowledge <strong>and</strong> activities. Receptivity<br />
can be defined as the ability of an organisation,<br />
community or individual to be aware of, to identify<br />
<strong>and</strong> to take effective advantage of a technology.<br />
Based upon Seaton & Cordey-Hayes [1993] <strong>and</strong><br />
Trott et al. [1995] we propose here that an<br />
underst<strong>and</strong>ing of the processes that affect how<br />
innovations are perceived, translated for particular<br />
purposes within a new context <strong>and</strong> eventually<br />
applied by recipients is essential both for assessing<br />
<strong>and</strong>, crucially, explaining innovation impact. This<br />
in turn may become the source of valuable design<br />
advice for those seeking to transfer knowledge,<br />
perhaps through the re-use of formal scientific<br />
models <strong>and</strong> support tools.<br />
5. RESEARCHING INNOVATION<br />
IMPACT<br />
Taking Time Geography as an innovation with<br />
potential conceptual <strong>and</strong> methodological impact,<br />
the main aim of the research is to examine <strong>and</strong><br />
explain the use <strong>and</strong> impact of TG ‘technology’<br />
using a framework based upon receptivity models<br />
of knowledge transfer.<br />
TiGS is just one way through which TG, as an<br />
innovation, can be transferred to <strong>and</strong> used by a<br />
recipient group. Indeed, to properly underst<strong>and</strong> the<br />
impact of TiGS it is necessary to have a broader<br />
appreciation of the process of TG knowledge<br />
transfer during the project. The impact of TiGS<br />
will be dependent upon the way in which TG has<br />
been received prior to use of TiGS.<br />
This conclusion leads us to a central element of the<br />
research being carried out - the separation of the<br />
impact of the conceptual technology that is TG<br />
from the support tool technology that is TiGS.<br />
TiGS represents one particular interpretation of TG<br />
but it is not the only one <strong>and</strong> will not be the only<br />
one that project participants are exposed to or use.<br />
Separating TiGS <strong>and</strong> TG may permit us to come to<br />
some initial conclusions on the comparative impact<br />
of different knowledge transfer mechanisms.<br />
Figure 1 illustrates the conceptual model of basic<br />
knowledge interactions <strong>and</strong> relationships within the<br />
project derived from observation of project<br />
meetings over the course of the first year <strong>and</strong> from<br />
interpreting the project work programme<br />
description. The formation of an underst<strong>and</strong>ing of<br />
these knowledge interactions was an initial<br />
objective of the research <strong>and</strong> will provide a vital<br />
framework for articulating <strong>and</strong> interpreting the<br />
overall process of knowledge transfer.<br />
Figure 1 Basic knowledge interactions (the TG<br />
ontology is a particular articulation of TG<br />
prepared for the project consortium)<br />
Within the framework shown in Figure 1 the<br />
specific research questions being addressed are:<br />
671
1. What impact at the work-package level<br />
does TG have on case-study research? The project<br />
is composed of three nested levels of operation –<br />
whole project, work-package <strong>and</strong> within workpackage<br />
(i.e. the task level).<br />
2. What impact at the task level does TG<br />
have on case-study research? It will be through<br />
affecting the way in which research tasks are<br />
performed that TG, or any other innovation, may<br />
impact the whole work-package level.<br />
3. What are the major factors at the task<br />
level that influence the receptivity of case-study<br />
researchers to TG? The aim of this question is to<br />
uncover the reasons behind any impact that TG has<br />
on the way in which research tasks are carried out.<br />
In addressing this question it should be possible to<br />
move towards a more process-based underst<strong>and</strong>ing<br />
of TG impact in policy-relevant research.<br />
4. What are the major factors at the task<br />
level that influence the receptivity of case-study<br />
researchers to particular design options <strong>and</strong>, in<br />
particular, to the selection of one option over<br />
another? Here an option may be a model or<br />
database design option, the inclusion or exclusion<br />
of a topic or variable of study, a scale of study<br />
decision, a methodology etc. The aim of this<br />
question is to assess the other factors which<br />
influence receptivity to TG within the context of<br />
the work-package research process <strong>and</strong> to better<br />
underst<strong>and</strong> the reasons behind the selection of<br />
particular options during the research process.<br />
Our main method of research is through observing,<br />
reporting <strong>and</strong> interpreting the research process in<br />
formal working meetings. This method will focus<br />
on knowledge <strong>and</strong> how the different knowledgebased<br />
roles that project participants fulfil in terms<br />
of project function, disciplinary background,<br />
experience, skills <strong>and</strong> underst<strong>and</strong>ing influence<br />
receptivity to TG through their interactions in the<br />
context of tasks. Project participants will remain<br />
anonymous; they will only be identified in terms of<br />
their knowledge roles. No detailed sociological<br />
analysis is being undertaken.<br />
However to ensure that the research process can be<br />
adequately understood, particularly in terms of the<br />
processes that influence receptivity, <strong>and</strong> also to<br />
ensure that the impact of TiGS is assessed<br />
adequately, two other distinct research activities<br />
are being undertaken. The full set of activities is:<br />
• Observing <strong>and</strong> interpreting the research<br />
process during formal work-package<br />
meetings.<br />
• Interviews to unpack the work-package<br />
research process <strong>and</strong> verify activity 1.<br />
• Workshop evaluation of TiGS using pre- <strong>and</strong><br />
post- use interviews.<br />
There is no clear precedent for the kind of work we<br />
are doing although we owe much to the research<br />
reported by Lemon [1999]. Consequently, the<br />
research will be exploratory <strong>and</strong> inductive in tone<br />
rather than hypothetico-deductive. The aim will be<br />
to identify the major factors <strong>and</strong> processes that<br />
influence receptivity to TG. We anticipate that<br />
these may feed into future research as testable<br />
recommendations regarding the structure of<br />
research activities that involve the transfer of<br />
conceptual <strong>and</strong> methodological innovations.<br />
6. CONCLUSIONS<br />
A variety of software <strong>and</strong> modelling technologies<br />
are emerging in the form of ‘support tools’ to<br />
better h<strong>and</strong>le issues of model-based scientific<br />
knowledge integration <strong>and</strong> re-use. These<br />
technologies are motivated by legitimate concerns<br />
about increasing the efficiency <strong>and</strong> costeffectiveness<br />
of environmental research <strong>and</strong><br />
ensuring that science can be effectively <strong>and</strong> easily<br />
transferred to management application. However,<br />
the current technical <strong>and</strong> software oriented<br />
research agenda in environmental modelling does<br />
not address the plethora of non-technical issues<br />
that may impact the receptivity of environmental<br />
modellers, scientists <strong>and</strong> policy-maker end-user<br />
groups to these emerging technologies. Failure to<br />
address these issues may mean that emerging<br />
technologies are not taken up by end-user groups<br />
or are used in ways which are ignorant of the<br />
sensitivities <strong>and</strong> complexities required of<br />
environmental research as it extends into the study<br />
<strong>and</strong> management of socio-natural interactions.<br />
It is proposed that splitting the conceptual <strong>and</strong><br />
technical elements of model-based support tools<br />
provides a useful starting point for assessing their<br />
impact on research <strong>and</strong> policy analysis /<br />
formulation tasks. From here the framework<br />
suggested by receptivity models of innovation <strong>and</strong><br />
knowledge transfer is proposed as a means of<br />
structuring research <strong>and</strong> interpreting results in<br />
terms of the factors <strong>and</strong> processes that influence<br />
the way in which end-users perceive, modify <strong>and</strong><br />
apply model-based innovation. Why are some<br />
concepts <strong>and</strong> tools used <strong>and</strong> others not for<br />
particular tasks? How can concepts <strong>and</strong> tools be<br />
transferred more successfully into different task,<br />
user <strong>and</strong> organisational contexts?<br />
Although too early to give empirical results or to<br />
indicate whether the research framework described<br />
can provide normative ‘good practice’ guidance,<br />
the work reported here should provide a means of<br />
672
initially assessing the utility of the receptivity<br />
research programme <strong>and</strong> innovation / knowledge<br />
perspective. This method may provide a way of<br />
informing future research in environmental model<br />
<strong>and</strong> support tool development in terms of design,<br />
deployment <strong>and</strong> patterns of use. It may be possible<br />
through better underst<strong>and</strong>ing of how different<br />
groups use support tools to design tools that are<br />
indeed ‘good to think with’.<br />
7. ACKNOWLEDGEMENTS<br />
The authors would like to acknowledge the<br />
financial support of the EC through the TiGrESS<br />
(Contract #EVG3-2001-00024) <strong>and</strong> Virtualis<br />
(Contract # IST-2000-28121) projects.<br />
8. REFERENCES<br />
Argent, R.M., An overview of model integration<br />
for environmental applications – components,<br />
frameworks <strong>and</strong> semantics, <strong>Environmental</strong><br />
<strong>Modelling</strong> <strong>and</strong> <strong>Software</strong>, 19: 219-234, 2004<br />
Busby, J.S., The Problem with Design Reuse: An<br />
Investigation into Outcomes <strong>and</strong> Antecedents,<br />
Journal of Engineering Design, 10: 277-296,<br />
1999.<br />
Costanza, R., A vision of the future of science:<br />
reintegrating the study of humans <strong>and</strong> the rest<br />
of nature, Futures, 35:651-671, 2003.<br />
Eason, K.D., Ergonomic perspective on advances<br />
in human-computer interaction, Ergonomics,<br />
34: 721-741, 1991.<br />
Engelen, G., White, R. & Uljee, I., Integrating<br />
constrained cellular automata models, GIS<br />
<strong>and</strong> decision support tools for urban planning<br />
<strong>and</strong> policy-making, In: Timmermans, H. (ed.),<br />
Decision Support Systems in Urban<br />
Planning, E & FN Spon, 1997.<br />
Funtowicz, S. & O’Connor, M., The Passage from<br />
Entropy to Thermodynamic Indeterminacy: a<br />
Social <strong>and</strong> Science Epistemology for<br />
Sustainability, In: Mayumi K. <strong>and</strong><br />
J.M.Gowdy (eds), Bioeconomics <strong>and</strong><br />
sustainability: essays in honor of Nicholas<br />
Georgescu-Rogen, Edward Elgar,<br />
Cheltenham, USA, 1998.<br />
Hägerstr<strong>and</strong> T., Time geography: focus on the<br />
corporeality of man, society <strong>and</strong> environment.<br />
The science <strong>and</strong> praxis of complexity. The<br />
United Nations University pp. 193-216, 1985.<br />
Jakeman, A.J. & R.A. Letcher, Integrated<br />
assessment <strong>and</strong> modelling: features,<br />
principles <strong>and</strong> examples for catchment<br />
management, <strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong><br />
<strong>Software</strong>, 18: 491-501, 2003.<br />
Lemon, M. (ed.), Exploring <strong>Environmental</strong><br />
Change Using an Integrative Method,<br />
Gordon <strong>and</strong> Breach, Amsterdam, ISBN 90-<br />
5699-193-0, ISSN 1027-2607, 1999.<br />
Lenntorp, B., Time-geography – at the end of its<br />
beginning, GeoJournal, 48: 155-158, 1999.<br />
Morgan, M.S. & Morrison, M. (eds.), Models as<br />
Mediators, Perspectives on Natural <strong>and</strong><br />
Social Science, Cambridge University Press,<br />
Cambridge, 1999.<br />
Oxley, T., McIntosh, B.S., Winder, N., Mulligan,<br />
M. & Engelen, G., Integrated <strong>Modelling</strong> &<br />
Decision Support Tools: A Mediterranean<br />
Example, <strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong><br />
<strong>Software</strong>, in press.<br />
Preece, J, Rogers, Y., Sharp, H., Benyon, D.,<br />
Holl<strong>and</strong>, S. & Carey, T., Human-computer<br />
interaction, Addison-Wesley, 1994.<br />
Rittel, H.W.J. & Webber, M.W., Dilemmas in a<br />
General Theory of Planning, Policy Sciences,<br />
4: 155-169, 1973.<br />
Rizzoli, A.E. & Young, W.J., Delivering<br />
environmental decision support systems:<br />
software tools <strong>and</strong> techniques, <strong>Environmental</strong><br />
<strong>Modelling</strong> <strong>and</strong> <strong>Software</strong>, 12: 237-249, 1997.<br />
Sage, A.P., Decision Support Systems<br />
Engineering, John Wiley <strong>and</strong> Sons Inc.,<br />
London, 1991.<br />
Seaton R.A.F. & Cordey-Hayes M., The<br />
Development <strong>and</strong> Application of Interactive<br />
Models of Industrial Technology Transfer,<br />
Technovation, 13: 45-53, 1993.<br />
Trott P, Seaton R.A.F. & Cordey-Hayes M.,<br />
Inward Technology Transfer as an interactive<br />
process: a case-study of ICI, Technovation,<br />
15: 25-43, 1995.<br />
Van Daalen, E., Dresen, L. & Janssen, M.A., The<br />
roles of computer models in the<br />
environmental policy life-cycle,<br />
<strong>Environmental</strong> Science <strong>and</strong> Policy, 5: 221-<br />
231, 2002.<br />
van der Leeuw, S. E., Vulnerability <strong>and</strong> the<br />
Integrated Study of Socio-natural<br />
Phenomenon. IHDP Update, 2(1): 6-7, 2001.<br />
Winder, N., The path-finder as historian: an assay<br />
of modernism, of the emerging science of<br />
cultural ecodynamics <strong>and</strong> the mathematics of<br />
history, SMC, Kiruna, ISBN 91-973922-0-0,<br />
2000.<br />
Winder, N., ‘What is Time-Geography? An<br />
Approach to Human Space-Time Behaviour<br />
due to Torsten Hagerstr<strong>and</strong>’, Presentation to<br />
the First TiGrESS Project meeting, EC grant<br />
number EVG3-2001-00024, www.tigress.ac,<br />
2003.<br />
673
Uncertainty in the Water Framework Directive:<br />
Implications for Economic Analysis<br />
J. Myšiak, K. Sigel<br />
UFZ Centre for <strong>Environmental</strong> Research, Permoserstrasse 15, 04318 Leipzig, Germany<br />
(jaroslav.mysiak@ufz.de)<br />
Abstract: The Water Framework Directive (WFD) imposes a new approach to water resource management in<br />
the EU states. Uncertainty surrounding its implementation, however, could badly affect the achievement of<br />
the objectives set by the Directive. Although not directly linked to a set of techniques to deal with it, the WFD<br />
<strong>and</strong> accompanying guideline documents identify uncertainty as a factor likely to play a significant role in<br />
assessing the risk of failing to achieve the objectives <strong>and</strong> setting up the required programmes of measures. In<br />
this paper, by addressing the initial description of a river basin we analyse uncertainty in socioeconomic<br />
descriptors such as demographic <strong>and</strong> water-use data. Socioeconomic data, models <strong>and</strong> evaluation techniques<br />
supporting the economic analysis of water uses are crucial parts of a Decision Support System (DSS) aimed at<br />
facilitating the WFD implementation.<br />
Keywords: Water resources; Decision Support System; Catchment; <strong>Modelling</strong>; Economic Analysis<br />
1. INTRODUCTION<br />
The Water Framework Directive (WFD) is a piece<br />
of environmental legislation which is<br />
unprecedented in the history of the EU. As well as<br />
imposing environmental objectives to be achieved,<br />
the WFD also lays down a set of instruments <strong>and</strong><br />
procedures to analyse the socioeconomic <strong>and</strong><br />
environmental impacts of current water uses <strong>and</strong> to<br />
help implement measures to achieve the objectives.<br />
To support the implementation of the WFD, a<br />
series of guideline documents have been developed<br />
under the Common Implementation Strategy of the<br />
WFD (CIS) to explain the novel concepts <strong>and</strong><br />
guide the application of the WFD’s instruments.<br />
However, none of them – nor indeed WFD itself –<br />
addresses the issue of uncertainty. However,<br />
uncertainty is likely to be an important factor in<br />
guiding activities designed to achieve the WFD’s<br />
objectives <strong>and</strong> effectively allocate the resources<br />
available. Indeed, ignoring uncertainty could, in<br />
many cases, result in the desired status of water<br />
resources not being achieved because the available<br />
information (uncertainty being a piece of<br />
information) has not been sufficiently exploited.<br />
Excessive dem<strong>and</strong>s for certainty on the other h<strong>and</strong><br />
can lead to unnecessarily expensive measures<br />
being implemented while valuable resources,<br />
which could be more effectively allocated to other<br />
catchment locations, are wasted.<br />
Economic analysis in connection with the WFD is<br />
designed to analyse the importance of water to the<br />
economy <strong>and</strong> the socioeconomic development of<br />
river basins (WATECO 2003). Several Decision<br />
Support Systems (DSS) have been developed to<br />
facilitate the economic analysis of water resources,<br />
especially to (i) analyse the socioeconomic drivers<br />
which exert pressures on water resources <strong>and</strong> are<br />
thus responsible for the water’s current status; (ii)<br />
investigate the dynamics of water uses <strong>and</strong><br />
contributes to the development of a baseline<br />
scenario; (iii) assess the cost recovery level of<br />
water services; (iv) select the most cost-effective<br />
programme of measures to achieve the WFD’s<br />
objectives.<br />
In all these tasks, uncertainty is likely to play an<br />
important role. As a conclusion drawn from the<br />
initial description of a river basin, the analysis of<br />
current water uses <strong>and</strong> the prediction of future<br />
development (baseline scenario), the ‘risk’ (in the<br />
sense of likelihood) of the WFD’s objectives not<br />
being met needs to be assessed. This is crucial,<br />
because once the likelihood of failure is known,<br />
suitable measures can be adopted. Uncertainties<br />
from different sources are summed up in this<br />
assessment, e.g. uncertainty in data collection,<br />
transformation (from the original spatial units for<br />
674
which they are collected to the hydrological<br />
boundaries where they are required), <strong>and</strong> forecast<br />
models. These uncertainties may vary <strong>and</strong> interact<br />
differently at various spatial levels: e.g. the<br />
transformation of demographic data to a river basin<br />
district is normally less uncertain than to sub-basin<br />
survey areas.<br />
In this paper, we analyse the uncertainty in the<br />
assessment of key economic drivers likely to<br />
influence pressures on water resources. We focus<br />
on demographic development (<strong>and</strong> domestic water<br />
supply) as a representative socioeconomic data set<br />
for several reasons: (i) demographic development<br />
is regarded as one of the main driving forces<br />
behind the pressures on water resources<br />
(IMPRESS 2003); (ii) the population size <strong>and</strong><br />
especially age structure determine a number of<br />
other economic indicators such as inflation,<br />
national saving rates, investment rates, gross<br />
domestic product growth rates, etc. (Lindh <strong>and</strong><br />
Malberg, 2000); <strong>and</strong> (iii) these data are best<br />
available from the data required to perform<br />
economic analysis, meaning a number of<br />
uncertainty sources which are common to any other<br />
socioeconomic data may be demonstrated. The<br />
analysis <strong>and</strong> the case study presented in the paper<br />
were developed to aid the development of a DSS<br />
for the White Elster River to analyse pressures <strong>and</strong><br />
impacts <strong>and</strong> subsequently to compile a programme<br />
of measures designed to achieve the WFD’s<br />
objectives.<br />
2. UNCERTAINTY IN THE WFD<br />
Although the WFD recognises uncertainty as a<br />
relevant factor, it does not contain a<br />
comprehensive framework for describing <strong>and</strong><br />
h<strong>and</strong>ling it. In fact, the term ‘uncertainty’ is not<br />
used by the WFD; instead, two other expressions in<br />
the context of uncertainty can be found: “Adequate<br />
level of confidence <strong>and</strong> precision” <strong>and</strong> “risk”. The<br />
former is used in relation to: (i) the process of<br />
establishing the reference conditions for surface<br />
water body types; (ii) monitoring the ecological<br />
<strong>and</strong> chemical status of surface waters <strong>and</strong> (iii) the<br />
identification of trends in groundwater pollution.<br />
Presumably these three domains should be<br />
regarded as representative because the problem of<br />
uncertainty also arises in other domains. Instead of<br />
the term ‘adequate’ (as applied to the level of<br />
confidence <strong>and</strong> precision), the WFD uses the<br />
expressions ‘sufficient’ <strong>and</strong> ‘acceptable’. The<br />
simultaneous employment of the terms<br />
‘confidence’ <strong>and</strong> ‘precision’ expresses the<br />
subjective (confidence) <strong>and</strong> objective (precision)<br />
character of uncertainty.<br />
The term ‘risk’ is used in two different meanings in<br />
the WFD. In the context of “risk to or via the<br />
aquatic environment”, ‘risk’ is used in the sense of<br />
danger (hazardous substances). One common<br />
approach for dealing with this kind of risk is to<br />
establish a link between the negative outcomes <strong>and</strong><br />
the likelihood of these outcomes occurring. In the<br />
case of hazardous substances, the WFD dictates<br />
that two strategies be followed: scientific risk<br />
assessment <strong>and</strong> the precautionary principle. In the<br />
context of “risk (of water bodies) failing to meet<br />
the environmental quality objectives” the term<br />
‘risk’ could firstly be interpreted as ‘possibility’.<br />
However, from the context it can be concluded that<br />
the WFD here implicitly refers to the sum of<br />
pressures affecting the water body. Hence ‘risk’<br />
becomes a negative meaning increased by the<br />
negative wording (‘failing to achieve’).<br />
Concerning strategies for dealing with uncertainty,<br />
the WFD states that the “level of confidence <strong>and</strong><br />
precision” has to be “estimated” <strong>and</strong> has to be<br />
“adequate”. These two steps, estimating <strong>and</strong><br />
evaluating uncertainty, can be designated as central<br />
components of any kind of strategy for dealing<br />
with uncertainty. In addition, the WFD contains<br />
several elements which may play an important role<br />
for dealing with uncertainty as they influence the<br />
way in which information <strong>and</strong> (imperfect)<br />
knowledge are generated <strong>and</strong> h<strong>and</strong>led in the<br />
implementation process of the WFD, e.g. designed<br />
<strong>and</strong> targeted monitoring programmes,<br />
participation, adaptation <strong>and</strong> review of the WFD.<br />
These elements, although not explicitly linked to<br />
uncertainty, are very important as they focus on a<br />
multitude of types <strong>and</strong> sources of uncertainty.<br />
3. UNCERTAINTY IN ECONOMIC<br />
ANALYSIS<br />
A variety of socioeconomic descriptors is required<br />
at some stage of the WFD implementation process.<br />
A comprehensive list of socioeconomic descriptors<br />
has been produced by the WATECO (2003),<br />
LAWA (2002) <strong>and</strong> the Economics Sub-Group of<br />
the <strong>International</strong> Commission for the Protection of<br />
the Danube River (ICPDR). In the latter, the<br />
descriptors are structured into (i) general<br />
socioeconomic indicators (e.g. population, gross<br />
domestic product, rate of economic growth,<br />
employment), (ii) characteristics of water services<br />
(e.g. total water production, water supply, water<br />
dem<strong>and</strong>, wastewater treatment, irrigation water<br />
supply), <strong>and</strong> (iii) characteristics of water uses (e.g.<br />
agriculture, industry, hydropower). In addition, the<br />
forecast of future development with regard to these<br />
descriptors has to be integrated into the baseline<br />
scenario, which describes the dynamics of the river<br />
675
asin without any additional provisions resulting<br />
from the WFD.<br />
Many general socioeconomic descriptors are<br />
collected by statistical offices which guarantee<br />
(albeit not totally, as shown below) the uniform<br />
methodology of data acquisition, data<br />
comparability, the constancy of data upgrade, <strong>and</strong><br />
the basic assessment of uncertain components of<br />
the data. These data are normally non-confident<br />
<strong>and</strong> available at the municipal or higher aggregated<br />
district level. On the other h<strong>and</strong>, some other data<br />
(e.g. water abstraction payments, waste water<br />
charges) not normally collected by statistical<br />
bureaux is either not available at all or (at least)<br />
partly confidential, available only on dem<strong>and</strong> <strong>and</strong><br />
accessible at a higher aggregation level (Interwies<br />
et al., 2003).<br />
The socioeconomic data have an uncertain<br />
component, the magnitude of which depends on a<br />
variety of factors including (i) the quality of<br />
measurements, (ii) the quality of models from<br />
which they are derived, (ii) the scale at which the<br />
data are collected or made publicly available (data<br />
confidence issue), (iv) the upgrade frequency <strong>and</strong><br />
(v) the quality of metadata documenting the<br />
descriptors, to name but a few. Generally speaking,<br />
uncertainty in economic analysis is caused (<strong>and</strong><br />
accumulated) through (i) the conceptualisation of<br />
the phenomena analysed; (ii) measurement <strong>and</strong><br />
representation; <strong>and</strong> (iii) data conversion <strong>and</strong><br />
analysis (Fig. 1). Below we address the issues<br />
related to the quality of the general socioeconomic<br />
descriptors using the example of the demographic<br />
data <strong>and</strong> domestic water supply.<br />
3.1 Uncertainty due to Conceptualisation <strong>and</strong><br />
Measurement<br />
The demographic data are collected by statistical<br />
offices relatively infrequently, normally once every<br />
5 or 10 years. The last census took place in the<br />
states that were previously part of the GDR (East<br />
Germany) in 1981 <strong>and</strong> were technically updated in<br />
1991. Since then the number of inhabitants has<br />
been updated with data from registry offices. In<br />
demographic analysis the population size may be<br />
the subject of uncertainty analysis as there is a<br />
variety of quantities which may be referred to (i.e.<br />
who are counted). Statistical bureaux normally<br />
count inhabitants according to their main place of<br />
residence rather than the inhabitants actually living<br />
in a local authority. Especially in the bigger cities<br />
with a university (e.g. Leipzig with about 30,000<br />
students <strong>and</strong> Halle with about 17,000 students, the<br />
former being nearly completely included <strong>and</strong> the<br />
latter included by 1/3 in the river basin district)<br />
where many students are registered under their<br />
second place of residence, the actual number of<br />
inhabitants is understated. For example, for the city<br />
of Leipzig the difference between the estimated<br />
total population in Leipzig <strong>and</strong> the number of<br />
inhabitants with their main residence in the city<br />
accounts for 28,500 inhabitants (~6%). In addition,<br />
a high number of commuters (different place of<br />
residence <strong>and</strong> place of work) represent another<br />
source of uncertainty. People commute to work<br />
between local authorities, districts or even federal<br />
states. Depending on the region, these migrations<br />
may account for quite large uncertainty, especially<br />
when the demographic data is used as an input to<br />
predict the future water supply dem<strong>and</strong>. In Saxony<br />
the share of commuters accounts for 10%.<br />
3.2 Uncertainty due to Data Analysis <strong>and</strong><br />
Transformation<br />
Figure 1: Uncertainty <strong>and</strong> error propagation.<br />
The case study area is the River White Elster<br />
catchment, a tributary of the River Saale that<br />
eventually flows into the River Elbe. Most of the<br />
White Elster area (5200 km2) is in Germany,<br />
although a small upl<strong>and</strong> part is situated in the<br />
Czech Republic. The structural diversity of the<br />
local government units for which socioeconomic<br />
data is normally collected makes the case study an<br />
ideal place to investigate the availability <strong>and</strong><br />
quality of socioeconomic data.<br />
The socioeconomic data, unlike environmental<br />
data like l<strong>and</strong> cover, are collected <strong>and</strong> aggregated<br />
for spatial units which are not readily compatible<br />
with river basins districts. The data are often<br />
available for statistical <strong>and</strong>/or administrative units<br />
such as local authorities, districts, federal states or<br />
the national level. Other data are collected<br />
primarily for different spatial units such as water<br />
supply districts, wastewater disposal districts, areas<br />
of high population concentrations, etc. To perform<br />
the economic analysis, this data have to be<br />
restructured to hydrological spatial units such as<br />
river basin districts or even more detailed water<br />
sub-basin survey areas. In Germany, federal states’<br />
statistical bureaux assigned the water management<br />
relevant data (e.g. abstraction) to river basin<br />
districts according to the statistical units’ centre of<br />
676
gravity. A more precise algorithm is based on a<br />
weighted average of the geographic <strong>and</strong> settlement<br />
shares of the local authorities’ segment covered by<br />
the river basin district (LAWA 2002).<br />
Data transformation is a considerable source of<br />
uncertainty whose importance increases with the<br />
number of administrative units intersected by the<br />
boundary of the river basin district. The White<br />
Elster river basin passes through four German<br />
federal states, four Government regions, 22<br />
districts <strong>and</strong> 334 local authorities. Since each of<br />
the four states has its own Department of Statistics,<br />
there are four different data providers for basic<br />
statistical information about the river basin district.<br />
The river basin completely contains 194<br />
municipalities (with a total area of about 3200<br />
km 2 ) <strong>and</strong> intersects another 140 municipalities<br />
(with a total area of about 3800 km 2 ). The local<br />
authorities completely contained within the river<br />
basin districts are on average smaller (mean ~16<br />
km 2 , st<strong>and</strong>ard deviation ~18 km 2 ) <strong>and</strong> more<br />
homogeneous than the intersected local authorities<br />
(mean ~27 km 2 , st<strong>and</strong>ard deviation ~31 km 2 ),<br />
which means a rather high uncertain component for<br />
example in the assessed number of inhabitants<br />
living in the river basin district. Indeed, the total<br />
population living in the intersected local authorities<br />
(<strong>and</strong> who are thus more problematic for assigning<br />
to the river basin districts) accounts for 1.6 million<br />
(in 2001), which is more than double the<br />
population living in the local authorities<br />
completely within the hydrological boundary of the<br />
river basin district (0.76 million).<br />
Different approaches to restructuring the<br />
demographic data to the hydrological boundaries<br />
of the river basin yield different results. For<br />
example, the transformation of the local authority<br />
based population data among the respective river<br />
basin districts is often based on the percentage of<br />
populated area concerned. To calculate the<br />
perceptual share of the settled area in each local<br />
authority, the CORINE L<strong>and</strong> Cover (CLC) data<br />
recommended by LAWA (reference date 1997,<br />
reference scale 1:100,000) <strong>and</strong> a more precise<br />
biotope map (based on CIR images, reference date<br />
1992-93, reference scale 1: 10 000, see Rosenberg<br />
2003) were used. While at the river district level<br />
the CLC <strong>and</strong> biotope map based assessments<br />
performed equally, at the more disaggregated<br />
(district) level the differences between them ranged<br />
from –23% to 15%. Although both data sets differ<br />
in terms of their resolution <strong>and</strong> quality of<br />
classification, the transformation based on them<br />
assumes a perfect correlation between the<br />
inhabited area <strong>and</strong> the number of inhabitants which<br />
does not hold, as the following example shows. For<br />
other socioeconomic descriptors such as the<br />
number of households or age structure, this<br />
relationship is even lower or non-existent.<br />
In the main urban centres, transformation based on<br />
settlement shares can cause higher uncertainty as<br />
the different population density (dwelling houses<br />
with different numbers of floors) in different parts<br />
of the city cannot be considered by using the<br />
settlement l<strong>and</strong> cover data. To assess this<br />
uncertainty we analysed the population <strong>and</strong><br />
settlement l<strong>and</strong> cover data for the city of Leipzig.<br />
We found that the difference between the estimated<br />
<strong>and</strong> the actual population size in our test area<br />
ranged from –300 % (in the periphery of the city<br />
where the actual population size is lower than a<br />
proportional share derived from the settlement<br />
area) to +70% with the lowest difference (~0)<br />
being close to the city centre (Fig. 2).<br />
Figure 2: Differences between actual <strong>and</strong><br />
calculated population size in the boroughs of the<br />
city of Leipzig when transformation is based on the<br />
CLC data.<br />
Administrative reforms are another source of (in<br />
some cases considerable) uncertainty. In Saxony,<br />
for instance during the reforms carried out between<br />
1991– 2001, the number of local authorities was<br />
reduced from 1623 to 539, while the number of<br />
districts decreased from 48 to 29. Although this<br />
caused only little uncertainty in population data<br />
(which has been correspondingly adjusted for the<br />
past periods), this makes it largely impossible to<br />
compare data about water services <strong>and</strong> analyse past<br />
trends. Water management data are collected for<br />
spatial units corresponding to the areas supplied by<br />
an enterprise. Originally corresponding to the local<br />
government areas, due to the administrative reform<br />
this data differs from the current administrative<br />
units <strong>and</strong> must be restructured to fit the river basin<br />
districts. The magnitude of uncertainty caused by<br />
administrative reforms varies considerably among<br />
the data required for economic analysis, being<br />
lower for data aggregated at district level <strong>and</strong><br />
higher for data available at the lower level. Finally,<br />
the uncertainty of boundaries of river basin/<br />
subbasins <strong>and</strong> administrative boundaries<br />
exacerbate the above-described uncertainties.<br />
677
3.3 Uncertainty due to Prediction – Base Line<br />
Scenario<br />
The Directive stipulates the development of a<br />
baseline scenario which frames the forecast <strong>and</strong><br />
assessment in key economic drivers likely to<br />
influence water status until 2015. Forecasting<br />
demographic development is a task for each state’s<br />
statistical bureau. The current forecasts are<br />
available for 2002–20 in Saxony, 1997–2015 in<br />
Thuringia <strong>and</strong> 2000–15 in Saxony-Anhalt. These<br />
forecasts are based on different models <strong>and</strong><br />
assumptions <strong>and</strong> are thus only partly comparable.<br />
The smallest administrative unit for which the<br />
forecast is available is district. For Saxony <strong>and</strong><br />
Thuringia there are two different scenarios of<br />
further demographic development available, based<br />
on different assumptions for (i) the mortality rate<br />
<strong>and</strong> migration between the German states (in<br />
Saxony <strong>and</strong> (ii) immigration from other European<br />
countries, especially EU Associated States (in<br />
Thuringia). In Saxony-Anhalt <strong>and</strong> Thuringia, the<br />
available forecasts are being updated as the current<br />
forecasts have performed poorly in predicting the<br />
demographic development of past years. Another<br />
demographic prediction (INKAR) available at<br />
district level has been developed by the Federal<br />
Department of Construction <strong>and</strong> Regional Planning<br />
(BBR) for all German states. This prediction<br />
consists of just one scenario. The differences<br />
between the predictions of states’ statistical<br />
bureaux <strong>and</strong> INCAR predictions are up to 49 <strong>and</strong><br />
47 per cent (scenario 1 <strong>and</strong> scenario 2) in Saxony,<br />
19 <strong>and</strong> 16 per cent in Thuringia, <strong>and</strong> 43 per cent in<br />
Saxony-Anhalt. An analysis of variance revealed<br />
significant differences between how the Inkar<br />
prediction fits the states’ forecasts (p < 0.01, Fig.<br />
3). The different scenarios differ not only with<br />
regard to the absolute numbers of expected<br />
inhabitants but also in the sign of the expected<br />
trend in development. The differences between<br />
available forecasts differ considerably across the<br />
districts (spatial variability).<br />
One considerable source of uncertainty in the<br />
predictions considered is the fact that future<br />
economic development in the region is largely<br />
neglected. This is an important factor considering<br />
the vast economically motivated emigration from<br />
the states in eastern Germany in the early 1990s. In<br />
those years, for example, Saxony lost some 11% of<br />
its population. This emigration is still continuing,<br />
albeit less significantly, <strong>and</strong> Saxony’s population is<br />
expected to decline by 14–17% by 2020. Although<br />
the long-term predictions may include a significant<br />
uncertainty which increases towards later periods,<br />
frequent updates help to keep the actual level of<br />
uncertainty manageable. For example, the<br />
differences between the different predictions <strong>and</strong><br />
scenarios does not exceed 10% in the first five<br />
predicted years. In Saxony, the latest demographic<br />
forecast tallies well with current development<br />
(differences < 1%) at the state level despite the<br />
mismatch (>10%) in the individual parameters of<br />
the model.<br />
Difference (2010)<br />
20<br />
15<br />
10<br />
5<br />
0<br />
-5<br />
-10<br />
Differences between the demographic prognoses sumarised by state<br />
mean value 95% confidence interval<br />
S Saxony; SA Saxony-Anhalt; TH Thuringia<br />
V1 Variant 1; V2 Variant 2<br />
S_V1 S_V2 SA TH_V1 TH_V2<br />
Prognosis<br />
Figure 3: Differences between the demographic<br />
prognoses in the districts (above) <strong>and</strong> summarised<br />
by state (below).<br />
Socioeconomic drivers predicted in the baseline<br />
scenario have to be linked to water dem<strong>and</strong> <strong>and</strong><br />
wastewater disposal to assess their impact on water<br />
resources. Demographic data (population size,<br />
number of households, age structure, etc.) alone is<br />
not sufficient to explain the water consumption<br />
pattern as documented by low (<strong>and</strong> non-significant)<br />
correlations between population size <strong>and</strong> water<br />
consumptions, or by the reduction of water<br />
consumption per capita in Saxony by 6% (1995–<br />
98). A significant source of uncertainty in water<br />
dem<strong>and</strong> prediction, besides the uncertainties<br />
discussed above, is the fact that factors such as<br />
technological development, shifts in social values,<br />
globalisation <strong>and</strong> also climate change (on the<br />
supply side) are largely neglected. Although<br />
LAWA recognises these uncertainty factors as<br />
potentially significant for the future development<br />
of water uses <strong>and</strong> services, instruments for<br />
assessing these uncertainties are lacking <strong>and</strong> river<br />
basin authorities are not expected to address these<br />
issues (LAWA 2002). Unlike the demographic<br />
678
prediction, the forecast of future water dem<strong>and</strong> is<br />
not being pursued by the statistical bureaux. A<br />
practical impediment to the development of such<br />
predictions is a lack of larger time series data for<br />
the calibration of the forecast models. Currently<br />
this data about water consumption <strong>and</strong> wastewater<br />
disposal are available for four previous periods<br />
with a collecting interval of three years.<br />
Additionally, the records available are not<br />
comparable because of the complex administrative<br />
reform carried out in the past ten years.<br />
application of economic appraisal <strong>and</strong> multicriteria<br />
decision methods may be surrounded by<br />
uncertainty resulting for example (i) from the<br />
choice of a method, (ii) from restricting the number<br />
of participants <strong>and</strong> thus the preferences modelled,<br />
(iii) from monetising non-marked goods (e.g.<br />
wetl<strong>and</strong> values); <strong>and</strong> (iv) from aggregating<br />
preferences about a multitude of conflicting<br />
objectives. Although not addressed here, they are<br />
the subject of another paper being prepared by<br />
Mysiak et al. (in preparation).<br />
4. CONCLUSIONS AND DISCUSSION<br />
Although a considerable amount of uncertainty was<br />
identified in the demographic development <strong>and</strong><br />
future water dem<strong>and</strong> requirements, these<br />
assessments must be considered cautiously. It is<br />
not solely the level of uncertainty that indicates the<br />
significance of driving forces <strong>and</strong> pressures.<br />
Negative developments of pressures (albeit with an<br />
uncertain magnitude of decline) <strong>and</strong>/or nonsignificant<br />
impacts of the pressure considered may<br />
lead to the assessment of a pressure as not being<br />
significant for river <strong>and</strong> riverine ecosystems. In the<br />
White Elster case study, we conclude that (i) little<br />
to moderate uncertainty is included in the<br />
description of current population (which is higher<br />
in large cities <strong>and</strong> in the areas with a high<br />
proportion of commuters), (ii) high uncertainty<br />
accompanies the prediction of further demographic<br />
development (uncertainty being larger for more<br />
distant periods), <strong>and</strong> (iii) moderate to high<br />
uncertainty is caused by restructuring the<br />
population data to the river basin district<br />
(depending on the structural diversity of the local<br />
authorities intersected by the river basin district<br />
boundary). Although because of the progressing<br />
population decline in the river basin district, the<br />
population may be regarded as a non-significant<br />
driving force behind pressures on water resources,<br />
because of the high spatial variability of the data<br />
analysed, higher caution is advised at the level of<br />
sub-basin survey areas.<br />
Furthermore, the predictions of future water<br />
dem<strong>and</strong> <strong>and</strong> wastewater disposal are moderately to<br />
highly uncertain in the long term because of the<br />
relevant uncertainty sources such as climate change<br />
or technological development <strong>and</strong> innovation, in<br />
addition to uncertain demographic prediction.<br />
Other places where uncertainty plays an important<br />
role include estimating the current level of cost<br />
recovery for water services <strong>and</strong> selecting<br />
programmes of measures to achieve the WFD<br />
objectives. In both cases, the estimation of<br />
environmental <strong>and</strong> resource costs may include<br />
large uncertain components. In the latter case, the<br />
5. ACKNOWLEDGMENTS<br />
This research was kindly supported by the EC<br />
under contract no. EVK1-2000-22089 <strong>and</strong> HPMD<br />
-CT-2001-00117. The authors gratefully<br />
acknowledge the help of Matthias Rosenberg in<br />
analysing the settlement areas using a biotope map.<br />
6. REFERENCES<br />
EC (European Commission), Directive 200/60/EC<br />
of the European Parliament <strong>and</strong> of the Council<br />
establishing a framework for Community<br />
action in the field of water policy, Official<br />
Journal of the European Communities, L 372<br />
(43), 1-73, 2000.<br />
Interwies, E, Pielen, B, Strosser, P, The Economic<br />
Analysis according to the Water Framework<br />
Directive in the Danube River Basin, Ecologic,<br />
Institute for <strong>International</strong> <strong>and</strong> European<br />
<strong>Environmental</strong> Policy, 2003.<br />
IMPRESS, Guidance Document on Analysis of<br />
Impacts <strong>and</strong> Pressures, Common<br />
Implementation Strategy, 2003.<br />
WATECO, Guidance Document on Economic<br />
Analysis on Water Uses, Common<br />
Implementation Strategy, 2003.<br />
LAWA, Länderarbeitsgemeinschaft Wasser,<br />
Guidance Document for the implementation of<br />
the EC Water Framework Directive, 2002.<br />
Lindh, T. <strong>and</strong> Malberg, B., Can age structure<br />
forecast inflation trends? Journal of<br />
Economics <strong>and</strong> Business, 52, 31-49, 2000.<br />
Mysiak, J. <strong>and</strong> Sigel, K., Role <strong>and</strong> implication of<br />
uncertainty for economic analysis according to<br />
the WFD, (in preparation).<br />
BBR, Bundesamt für Bauwesen und Raumordnung<br />
Raumordnungsprognose INKAR Pro.<br />
Rosenberg, M., Development of a unified biotope<br />
map for Saxony, Saxony-Anhalt <strong>and</strong><br />
Thuringia, UFZ Umweltforschungszentrum,<br />
2003.<br />
679
.<br />
An Interactive Spatial Optimisation Tool for Systematic<br />
L<strong>and</strong>scape Restoration<br />
B.A. Bryan a , L.M. Perry b , D.Gerner b , B. Ostendorf b , <strong>and</strong> N.D. Crossman 2<br />
a Policy <strong>and</strong> Economic Research Unit, CSIRO L<strong>and</strong> <strong>and</strong> Water, Glen Osmond, South Australia, 5064.<br />
b School of Earth <strong>and</strong> <strong>Environmental</strong> Sciences, University of Adelaide, Glen Osmond, South Australia, 5064.<br />
Email: Bertram.Ostendorf@adelaide.edu.au<br />
Abstract: This paper describes the design <strong>and</strong> construction of a prototype spatial decision support system<br />
(SDSS) for an interactive evaluation of integrated l<strong>and</strong>scape restoration planning using spatial information<br />
technology. L<strong>and</strong>scape planning involves spatially explicit decisions about the types of l<strong>and</strong>uses allowable,<br />
<strong>and</strong> the extent <strong>and</strong> location of these l<strong>and</strong>uses. This decision-making needs to be supported by accurate <strong>and</strong><br />
detailed information about the spatial distribution of numerous parameters affecting the distribution of<br />
l<strong>and</strong>use. The SDSS that we present in this paper comprises a geographic information system (GIS) tightly<br />
coupled with an analytical optimisation module by means of an interactive interface. The GIS is used for<br />
storage, manipulation <strong>and</strong> visualisation of spatial data, <strong>and</strong> for assessing the results of the analytical module<br />
computing optimal spatial pattern. Several user-selectable parameters allow consideration of management<br />
objectives related to planning for l<strong>and</strong>scape restoration.<br />
Keywords: decision support systems; integer programming; GIS; l<strong>and</strong>scape restoration; priority setting.<br />
1. INTRODUCTION<br />
1.1 L<strong>and</strong>scape Planning <strong>and</strong> Optimisation<br />
Typically, <strong>and</strong> with some notable exceptions,<br />
l<strong>and</strong>scape restoration efforts tend to occur on the<br />
scale of the individual property/l<strong>and</strong>owner. As<br />
such, the restoration efforts are rarely planned so<br />
as to be of maximum benefit to the regional<br />
ecology <strong>and</strong> biodiversity. Systematic conservation<br />
planning (SCP) [Margules <strong>and</strong> Pressey, 2000]<br />
involves selecting the areas <strong>and</strong> environments to<br />
conserve in order to maximise the chances for<br />
biodiversity sustainability. SCP is a difficult<br />
problem [Margules et al., 2002] <strong>and</strong> involves<br />
consideration of an established suite of principles<br />
such as comprehensiveness, adequacy,<br />
representativeness, efficiency, flexibility,<br />
irreplaceability, <strong>and</strong> complementarity [Margules<br />
<strong>and</strong> Pressey, 2000]. Using SCP principles <strong>and</strong> with<br />
the coupling of Integer Programming (IP) <strong>and</strong><br />
Geographic Information Systems (GIS) the<br />
potential now exists for l<strong>and</strong>scape restoration<br />
activities to be systematically planned using a<br />
range of spatial databases. Thereby, maximum<br />
ecological value can be gained from current<br />
restoration efforts. Whilst the principles of<br />
systematic conservation planning are reasonably<br />
well established, the methods for implementing<br />
these principles are many <strong>and</strong> varied. The methods<br />
can be classed according to whether they can<br />
guarantee an optimal solution or not.<br />
The nature of spatial problems amenable to<br />
solution by optimisation approaches is diverse. So<br />
too are the models used for their solution. An<br />
optimisation paradigm used in spatial planning is<br />
integer or zero-one (0-1) programming. The major<br />
advantage of this technique is that it guarantees the<br />
optimal solution [Haight et al., 2000] (if the<br />
problem is tractable of course), thereby removing<br />
ambiguity about just how good the solution is. The<br />
biggest drawback to IP problems is that they are<br />
NP-complete [Karp, 1972]. In other words, the<br />
time taken for the models to run is a polynomial<br />
function of the number of inputs. Previous studies<br />
exploring problems of only modest size have<br />
proven to be intractable. Studies of spatial<br />
phenomena, especially those using GIS, typically<br />
involve large databases covering wide areas often<br />
at high resolution. It is not uncommon to work<br />
with raster databases of 20 million cells or more.<br />
The data-intensive nature of GIS has been<br />
680
fundamentally at odds with the data-restrictive<br />
nature of the IP paradigm. However, new<br />
proprietary algorithms have greatly increased the<br />
tractability of IP problems [Rodrigues <strong>and</strong> Gaston,<br />
2002a]. Thereby, fast algorithms have bridged the<br />
data requirements gap between IP <strong>and</strong> GIS, <strong>and</strong><br />
opened up these techniques to widespread<br />
application in the spatial domain.<br />
Many studies have used IP for conservation<br />
planning, particularly reserve selection [Cocks <strong>and</strong><br />
Baird, 1989; Church et al., 1996; Williams <strong>and</strong><br />
ReVelle, 1996; Haight et al., 2000; ReVelle et al.,<br />
2002; Rodrigues <strong>and</strong> Gaston, 2002a], but IP has<br />
not been used for systematic l<strong>and</strong>scape restoration.<br />
Several other methods that do not guarantee an<br />
optimal solution [Underhill, 1994] have been used<br />
in systematic reserve design including scoring<br />
approaches [Pressey <strong>and</strong> Nicholls, 1989a],<br />
heuristic algorithms [Pressey <strong>and</strong> Nicholls, 1989b;<br />
Csuti et al., 1997], <strong>and</strong> simulated annealing<br />
[Possingham, et al. 2000]. Optimality is not<br />
everything in reserve design of course [Csuti et al.,<br />
1997], but it does provide certainty when<br />
negotiating for conservation in areas of high<br />
l<strong>and</strong>use dem<strong>and</strong>.<br />
2 METHODS<br />
The Carrickalinga Creek catchment forms the<br />
study area for this analysis. The study area covers<br />
5,586 ha <strong>and</strong> is located in the southern Mt. Lofty<br />
Ranges, some 60 km south of Adelaide, the capital<br />
city of South Australia (Figure 1). The Mt. Lofty<br />
Ranges is a highly fragmented agricultural region<br />
with less than 10% of the native forests <strong>and</strong><br />
woodl<strong>and</strong>s remaining. Remnant vegetation is<br />
mostly located in the upper reaches of the<br />
catchment. The remaining area is cleared l<strong>and</strong><br />
under mixed use, predominantly agriculture <strong>and</strong><br />
grazing (Figure 1).<br />
1.2 Spatial Decision Support Systems<br />
A spatial decision support system (SDSS) is an<br />
intelligent information system that reduces<br />
decision making time as well as improving the<br />
consistency <strong>and</strong> quality of the decisions [Cortes et<br />
al., 2000]. A SDSS can be either problem specific<br />
or situation <strong>and</strong> problem specific [Rizzoli <strong>and</strong><br />
Young, 1997]. Both are tailored to a specific<br />
problem, but the latter is limited to one specific<br />
spatial location.<br />
Amongst Rizzoli <strong>and</strong> Young’s [1997] six desirable<br />
features of an SDSS is the ability to deal with<br />
spatial data <strong>and</strong> ability to be used effectively for<br />
diagnosis, planning, management <strong>and</strong><br />
optimisation.<br />
In this paper we describe the design <strong>and</strong><br />
construction of a prototype SDSS combining IP<br />
<strong>and</strong> GIS to solve a l<strong>and</strong>scape planning problem.<br />
This SDSS is not location specific <strong>and</strong> can be<br />
applied to any area of interest at any spatial scale.<br />
We present a brief demonstration of the SDSS<br />
with the aim of identifying high priority areas for<br />
the restoration of an adequate <strong>and</strong> representative<br />
l<strong>and</strong>scape ecological system in the Carrickalinga<br />
Creek catchment, South Australia.<br />
Figure 1: Location of the Carrickalinga Creek<br />
catchment in the Mt. Lofty Ranges, South<br />
Australia.<br />
Topography of the catchment is undulating to hilly<br />
with elevation ranging from sea-level at the mouth<br />
of the creek to 420m ASL toward the upper<br />
reaches of the catchment. The climate of the<br />
catchment is a typical coastal Mediterranean<br />
regime characterised by a strong seasonal<br />
demarcation of moderate to warm dry summers<br />
<strong>and</strong> cool, wet winters.<br />
2.1 The Data<br />
This optimisation analysis is based on six physical<br />
environmental variables <strong>and</strong> a mapped Soil<br />
L<strong>and</strong>scape Units (SLUs) variable. These variables<br />
act as surrogates for species distributions. The use<br />
of surrogates is preferable when there has been<br />
removal of extensive areas of native habitat. The<br />
environmental variables (Table 1) are a subset of<br />
681
those available in BIOCLIM (variables 1 to 4)<br />
[Nix, 1986] <strong>and</strong> the TAPES-G (variables 5 <strong>and</strong> 6)<br />
suite of topographic modelling tools [Gallant <strong>and</strong><br />
Wilson, 1996]. Bryan [2003] should be consulted<br />
for a detailed description of methods used to derive<br />
the variables. Each of the six continuous physical<br />
environmental variables was categorised into 5<br />
classes.<br />
Table 1. List of variables used in this study.<br />
1. Annual Mean Temperature<br />
2. Temperature Annual Range<br />
3. Annual Mean Precipitation<br />
4. Annual Mean Moisture Index<br />
5. Net Radiation<br />
6. Steady-state Wetness Index<br />
7. Soil L<strong>and</strong>scape Units<br />
The soils data was derived from a long-term soil<br />
survey by the South Australian Department of<br />
Primary Industries <strong>and</strong> Resources (PIRSA).<br />
Interpretation of aerial photography <strong>and</strong> field<br />
surveys are used to identify polygons representing<br />
homogeneous areas of soil. These homogenous<br />
areas are termed Soil L<strong>and</strong>scape Units. The<br />
Carrickalinga Creek study area is comprised of 36<br />
Soil L<strong>and</strong>scape Units. All data were converted to<br />
50m resolution grid layers. All GIS analyses were<br />
performed in ESRIs ArcGIS suite of tools.<br />
2.2 Integer Programming<br />
The classic set-covering/minimum representation<br />
IP model is used in this study to identify the<br />
minimum number of sites required to meet the<br />
conservation targets defined by proportional <strong>and</strong><br />
area constraints. The model was written in ILOG’s<br />
Optimisation Programming Language (OPL), a<br />
high-level scripting language part of the<br />
OPLStudio software. OPLStudio uses the CPLEX<br />
optimiser to solve linear IP problems. CPLEX has<br />
been found to be efficient in its solution of linear<br />
IP problems in conservation planning [Ando et al.,<br />
1998; Church et al., 1996; Rodrigues <strong>and</strong> Gaston,<br />
2002b]. The software comes with its own<br />
application programming interface (API), thus<br />
allowing the solvers to be accessed through a<br />
variety of programming languages. The setcovering/minimum<br />
representation model is<br />
described below [adapted from Possingham et al.,<br />
2000].<br />
The number of grid cells or sites (m) of 50m<br />
resolution in the Carrickalinga Creek catchment<br />
study area totalled 22,336. The total number of<br />
classes (including 5 classes of each environmental<br />
variable <strong>and</strong> the Soil L<strong>and</strong>scape Units) (n)<br />
equalled 66. An m x n matrix A (22,336 rows x 66<br />
columns) was created whose elements a ij were<br />
attributed a binary value according to the class of<br />
each site. Sites are given a value of one if they<br />
exhibit a particular environmental class or soil<br />
group, zero otherwise such that:<br />
1 if site i occurs in class j<br />
a ij =<br />
{<br />
for i = 1…m <strong>and</strong> j = 1…n<br />
Next, a variable is defined that reflects whether or<br />
not a site is selected for restoration, as the vector X<br />
with dimension m <strong>and</strong> elements x i , given by<br />
1 if site i is selected for restoration<br />
x i = { 0 otherwise<br />
for i = 1…m<br />
In words, the set-covering/minimum representation<br />
problem strives to minimise the number of sites in<br />
the reserve system subject to areal <strong>and</strong><br />
proportional constraints for each class (c j ). Areal<br />
constraints are a function of the area of the class,<br />
the proportional target ((p), the minimum<br />
percentage of each class to be restored), <strong>and</strong> the<br />
minimum area target ((t), the minimum number of<br />
sites in each class to be restored). For each class,<br />
the areal constraint is equal to the proportional<br />
target multiplied by the number of sites in the class<br />
if this value is greater than or equal to the<br />
minimum area target. Otherwise, the areal<br />
constraint for the class equals either the total<br />
number of sites in the class or the specified<br />
minimum area target, whichever is the lesser<br />
value. Mathematically, the optimisation techniques<br />
attempt to [adapted from Possingham et al., 2000]:<br />
m<br />
minimise ∑<br />
subject to<br />
0 otherwise<br />
i=<br />
1<br />
m<br />
i=1<br />
x i<br />
∑ a ij x<br />
where a ij , x i ∈ {0,1},<br />
i<br />
≥ c j<br />
for j = 1…n<br />
m<br />
Aj = ∑ aij<br />
i=<br />
1<br />
pA j if pA j ≥ t<br />
<strong>and</strong> c j = { min(Aj, t) otherwise<br />
682
2.3 Spatial Decision Support System<br />
Development<br />
Our SDSS, the Conservation Reserve Evaluation<br />
<strong>and</strong> Design Optimisation System (CREDOS;<br />
Figure 2), is formed by the combination of the<br />
GIS, the CREDOS interface, <strong>and</strong> the IP analytical<br />
module. The interface provides the coupling<br />
between the GIS <strong>and</strong> the analytical module, <strong>and</strong><br />
was written in Microsoft Visual Basic 6.0 (VB)<br />
using an ActiveX Dynamic Link Library (DLL)<br />
project. Functionality for manipulation of the<br />
spatial datasets was incorporated by means of<br />
ESRI ArcObjects, the development platform for<br />
the ArcGIS family of applications. ESRI is a<br />
proponent of the interoperability protocols<br />
expounded by the OpenGIS consortium (OGC),<br />
<strong>and</strong> ArcObjects is therefore built using Microsoft<br />
Component Object Model (COM) technology that<br />
allows applications using such technology to be<br />
written in any COM compliant programming<br />
language.<br />
The ActiveX project was compiled into an<br />
executable file (a DLL), thereby allowing<br />
portability between GIS sessions. Because<br />
CREDOS is a spatial analysis tool, the comm<strong>and</strong><br />
to execute the CREDOS DLL was seamlessly<br />
included as an additional toolbar in the GIS.<br />
majority of this time (90%) was allocated to<br />
CREDOS data preparation (binary conversion of<br />
input variables) in the GIS.<br />
Figure 2: The prototype Spatial Decision Support<br />
System, CREDOS.<br />
3 RESULTS<br />
CREDOS (Figure 2) consists of a set of input<br />
windows that allow the user to select the working<br />
directory, sites (zone) layer, input variables,<br />
optimisation model, constraints <strong>and</strong> outputs. These<br />
can be changed any time prior to, or after, running<br />
the model. During run time the user is informed of<br />
progress via a window that is updated as each<br />
CREDOS modelling procedure is completed. This<br />
assist the user in debugging input data <strong>and</strong> in<br />
determining the existence of any related procedural<br />
problems. Final output of CREDOS is a grid layer<br />
of sites identified as an optimal solution to the<br />
imposed proportional <strong>and</strong> areal constraints (Figure<br />
3), <strong>and</strong> a tabular summary of identified sites <strong>and</strong><br />
the corresponding values of the input variables.<br />
The tabular summary can be used to validate the<br />
model by confirming that solutions meet the areal<br />
<strong>and</strong> proportional targets.<br />
In the demonstration presented here, output<br />
consists of 4,495 50m cells (1,123.8 ha; 20% of<br />
the study area), providing at least 10 cells of each<br />
physical environmental type <strong>and</strong> soil class. The<br />
complete process (data preparation <strong>and</strong> IP problem<br />
solving) took approximately 15 minutes to solve<br />
on a P4, 3.0 GHz, 1.0Gb RAM. However the<br />
Figure 3: Optimal sites for revegetation in the<br />
study area based on the 20% proportion <strong>and</strong> 10 cell<br />
area constraints.<br />
4 DISCUSSION AND CONCLUSION<br />
The IP optimisation models implemented in this<br />
study were successful in finding efficient, adequate<br />
<strong>and</strong> representative combinations of sites for<br />
l<strong>and</strong>scape restoration given the specified<br />
parameters. The prototype SDSS facilitated <strong>and</strong><br />
simplified the modelling procedure by providing a<br />
user-friendly interface to find optimal solutions.<br />
Our prototype SDSS can be applied across any<br />
683
study area <strong>and</strong> any scale to solve user-selected<br />
optimisation constraints for l<strong>and</strong>scape planning.<br />
The solutions found by the IP models are<br />
maximally efficient. Maximally efficient solutions<br />
simply strive to find the fewest cells capable of<br />
satisfying conservation targets – in this case a<br />
minimum area <strong>and</strong> proportional targets. If more<br />
area is required, then it is a simple task to increase<br />
either of the areal targets in the SDSS. Restoration<br />
can always be increased beyond that recommended<br />
if required, or preferred, by l<strong>and</strong>owners. This<br />
simply requires re-application of the SDSS with<br />
different choices of constraints.<br />
There are many agricultural regions in Australia<br />
that have been subjected to extensive clearing <strong>and</strong><br />
fragmentation of the native biological<br />
communities. In these regions, reserve selection,<br />
alone, will not facilitate the conservation of the<br />
natural biodiversity, <strong>and</strong> restoration is required<br />
[Bryan, 2002]. L<strong>and</strong> in these regions is usually in<br />
high dem<strong>and</strong> from a variety of l<strong>and</strong> uses, <strong>and</strong><br />
restoration effort is precious. Hence, areas <strong>and</strong><br />
environments must be judiciously planned <strong>and</strong><br />
prioritised for restoration to gain maximum<br />
ecological benefit, whilst having minimal adverse<br />
economic impact through conversion from<br />
productive l<strong>and</strong>use. The major benefit of<br />
systematic l<strong>and</strong>scape restoration is that it can be<br />
used to coordinate <strong>and</strong> gain maximum ecological<br />
benefit from all restoration initiatives within a<br />
region. The restoration initiatives may come from<br />
the local l<strong>and</strong>holder, major regional scale<br />
government programs, or anywhere within this<br />
spectrum. Such planning is often in the h<strong>and</strong>s of<br />
natural resource management professionals who<br />
are often not technically proficient in complex GIS<br />
<strong>and</strong> modelling procedures. The prototype SDSS<br />
presented in this paper bridges the gap between<br />
those professionals <strong>and</strong> the modelling community.<br />
IP models have considerable potential in l<strong>and</strong>scape<br />
restoration <strong>and</strong> other conservation planning<br />
problems. The study presented here is a proof of<br />
concept. Significant advances in our SDSS<br />
functionality <strong>and</strong> model algorithm sophistication<br />
are currently under development to make our<br />
results truly useful in planning for l<strong>and</strong>scape<br />
restoration. If, for whatever reason, a site cannot<br />
be restored, the network of sites will no longer<br />
meet conservation targets. There are possibly very<br />
many optimal solutions <strong>and</strong> very many slightly<br />
sub-optimal solutions to the problems. Given the<br />
short run time for the models, it is relatively<br />
simple to modify the SDSS so each model can be<br />
processed many times, each time adding the<br />
previous solution as a constraint, <strong>and</strong> thereby<br />
finding many options <strong>and</strong> providing flexibility in<br />
restoration design.<br />
The underlying IP model is naïve to current<br />
l<strong>and</strong>use <strong>and</strong> economic cost except for the<br />
assumption that each site costs the same amount<br />
<strong>and</strong> the objective is to minimise the total cost of<br />
the system. Inclusion of l<strong>and</strong>use <strong>and</strong> cadastral<br />
information in the models will enhance the<br />
applicability of the model because the assumptions<br />
made become more realistic. We are investigating<br />
other improvements in the CREDOS by<br />
incorporating spatial effects. Such spatial effects<br />
are being integrated into the model to improve the<br />
l<strong>and</strong>scape structure of the resultant habitats. For<br />
example, sites are weighted that are close to<br />
existing reserves, riparian habitats, <strong>and</strong>/or transport<br />
corridors. Alternatively, constraints are set that<br />
force the model to select n replications of classes,<br />
separated by a certain distance, for replication <strong>and</strong><br />
enhanced insurance against local catastrophes. The<br />
results to this work will become available at a later<br />
date.<br />
The spatially-explicit, GIS-based IP optimisation<br />
approach taken in this research is an innovative<br />
approach to l<strong>and</strong>scape restoration. The<br />
development of a prototype SDSS is not novel in<br />
itself. However, the application of CREDOS<br />
facilitates solving of complex optimisation<br />
algorithms by non-modelling professionals. The<br />
case study presented in this paper demonstrates the<br />
utility of IP in planning for l<strong>and</strong>scape restoration.<br />
Ecological restoration is essential in many<br />
fragmented agricultural l<strong>and</strong>scapes to sustain<br />
ecological, environmental <strong>and</strong> human systems.<br />
Geographic priorities are required to guide<br />
restoration activities that are based on sound<br />
science to gain the maximum benefit from these<br />
activities for the conservation of biodiversity.<br />
Systematic l<strong>and</strong>scape restoration can be of great<br />
benefit in planning for the long term ecological,<br />
environmental, economic <strong>and</strong> social sustainability<br />
of other fragmented agricultural regions in<br />
Australia <strong>and</strong> overseas. The success of IP in this<br />
application reflects its potential in many allied<br />
areas. Current work is adding functionality to the<br />
prototype SDSS, thus allowing more complex<br />
optimisation problems to be solved by nontechnical<br />
professionals.<br />
5 REFERENCES<br />
Ando, A., J. Camm, S. Polasky <strong>and</strong> A. Solow,<br />
Species distributions, l<strong>and</strong> values, <strong>and</strong><br />
efficient conservation, Science 279(5359),<br />
2126-2128, 1998.<br />
684
Bryan, B.A., Reserve selection for nature<br />
conservation in South Australia: past,<br />
present <strong>and</strong> future, Australian Geographical<br />
Studies, 40(2), 196 – 209, 2002.<br />
Bryan, B.A., Physical environmental modelling,<br />
visualization <strong>and</strong> query for supporting<br />
l<strong>and</strong>scape planning decisions, L<strong>and</strong>scape<br />
<strong>and</strong> Urban Planning, 65, 237-259, 2003.<br />
Church, R.L., D.M. Stoms <strong>and</strong> F.W. Davis,<br />
Reserve selection as a maximal covering<br />
location problem, Biological Conservation,<br />
76(2), 105-112, 1996.<br />
Cocks, K.D. <strong>and</strong> I.A. Baird, Using mathematical<br />
programming to address the multiple<br />
reserve selection problem: an example from<br />
the Eyre Peninsula, South Australia,<br />
Biological Conservation, 49, 113-130,<br />
1989.<br />
Cortes, U., M. Sanchez-Marre, L. Ceccaroni, I. R-<br />
Roda <strong>and</strong> M. Poch, Artificial intelligence<br />
<strong>and</strong> environmental decision support<br />
systems, Applied Intelligence, 13, 77-91,<br />
2000.<br />
Csuti, B., S. Polasky, P.H. Williams, R.L. Pressey,<br />
J.D. Camm, M. Kershaw, A.R. Kiester, B.<br />
Downs, R. Hamilton, M. Huso <strong>and</strong> K. Sohr,<br />
A comparison of reserve selection<br />
algorithms using data on terrestrial<br />
vertebrates in Oregon, Biological<br />
Conservation, 80(1), 83-97, 1997.<br />
Gallant, J.C. <strong>and</strong> J.P. Wilson, TAPES-G: A gridbased<br />
terrain analysis program for the<br />
environmental sciences, Computers <strong>and</strong><br />
Geosciences, 22(7), 713-722, 1996.<br />
Haight, R.G., C.S. ReVelle <strong>and</strong> S.A Snyder, An<br />
integer optimization approach to a<br />
probabilistic reserve site selection problem,<br />
Operations Research, 48(5), 697-708, 2000.<br />
Karp, R.M., Reducibility among combinatorial<br />
problems, in Complexity of Computer<br />
Computations (Proceedings of a symposium<br />
at the IBM Thomas J. Watson Research<br />
Center), Yorktown Heights, NewYork.<br />
Plenum, 85-103, 1972<br />
Margules, C.R. <strong>and</strong> R.L. Pressey, Systematic<br />
conservation planning, Nature, 405(6783),<br />
243-253, 2000.<br />
Margules, C.R., R.L. Pressey <strong>and</strong> P.H. Williams,<br />
Representing biodiversity: data <strong>and</strong><br />
procedures for identifying priority areas for<br />
conservation, Journal of Bioscience, 27(4),<br />
Supplement 2, 309-326, 2002.<br />
Nix, H.A., A biogeographic analysis of the<br />
Australian elapid snakes, in R. Longmore<br />
(ed) Atlas of Elapid Snakes 7, Australian<br />
Government Publishing Service, Canberra,<br />
4-15, 1986.<br />
Possingham, H.P., I. Ball <strong>and</strong> S. Andelman,<br />
Mathematical methods for identifying<br />
representative reserve networks, in S.<br />
Ferson <strong>and</strong> M. Burgman (eds.), Quantitative<br />
Methods for Conservation Biology.<br />
Springer-Verlag, New York, 291-305, 2000.<br />
Pressey, R.L. <strong>and</strong> A.O. Nicholls, Efficiency in<br />
conservation evaluation: scoring versus<br />
iterative approaches, Biological<br />
Conservation, 50, 199-218, 1989a.<br />
Pressey, R.L. <strong>and</strong> A.O. Nicholls, Application of a<br />
numeric algorithm to the selection of<br />
reserves in semi-arid New South Wales,<br />
Biological Conservation, 50, 263-278,<br />
1989b.<br />
ReVelle, C.S., J.C. Williams <strong>and</strong> J.J. Bol<strong>and</strong>,<br />
Counterpart models in facility location<br />
science <strong>and</strong> reserve selection science,<br />
<strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong> Assessment,<br />
7(2), 71-80, 2002.<br />
Rizzoli, A.E. <strong>and</strong> W.J. Young, Delivering<br />
environmental decision support systems:<br />
software tools <strong>and</strong> techniques,<br />
<strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong> <strong>Software</strong>,<br />
12(2-3), 237-249.<br />
Rodrigues, A.S.L. <strong>and</strong> K.J. Gaston, Optimisation<br />
in reserve selection procedures - why not?,<br />
Biological Conservation, 107, 123-129,<br />
2002a.<br />
Rodrigues, A.S.L. <strong>and</strong> K.J. Gaston, Rarity <strong>and</strong><br />
conservation planning across geopolitical<br />
units, Conservation Biology, 16(3), 674-<br />
682, 2002b.<br />
Underhill, L.G., Optimal <strong>and</strong> suboptimal reserve<br />
selection algorithms, Biological<br />
Conservation, 70, 85-87, 1994.<br />
Williams, J.C. <strong>and</strong> C.S. Revelle, A 0-1<br />
programming approach to delineating<br />
protected reserves, Environment &<br />
Planning B - Planning & Design, 23(5),<br />
607-624, 1996.<br />
685
Assessing the Feasibility of Using Radar Satellite Data to<br />
Detect Flood Extent <strong>and</strong> Floodplain Structures<br />
Ms. Edith Stabel 1<br />
1<br />
Department of Physical Geography, Saarl<strong>and</strong> University, P.B. 15 11 50, 66041 Saarbrücken, Germany,<br />
E.Stabel@mx.uni-saarl<strong>and</strong>.de<br />
ABSTRACT: River dynamics <strong>and</strong> hydrological behaviour are strongly influenced by human activities both in the<br />
catchment areas <strong>and</strong> in the floodplains. The knowledge of recent <strong>and</strong> historical river dynamics <strong>and</strong> related<br />
morphological <strong>and</strong> structural changes on the l<strong>and</strong> surface (e.g. sedimentation, accumulation, river bed movement) is<br />
essential in assessing the flood risk <strong>and</strong> the vulnerability of human resources <strong>and</strong> structures. Earth Observation (EO)<br />
systems provide data to monitor <strong>and</strong> to analyse both, the river dynamics <strong>and</strong> small surface changes. Especially, radarbased<br />
systems <strong>and</strong> interferometric data analysis are of high interest. Along selected sites in the River Odra area, we<br />
analysed the potential of radar-based EO-applications for the detection of structural changes, validated by fieldwork.<br />
It is shown that the coherence information is of great significance: On the one h<strong>and</strong>, it could be used to eliminate<br />
misclassifications of the flood extent caused by double bounce scattering, corner reflection or smooth surfaces. On the<br />
other h<strong>and</strong> the production of RGB’s type Interferometric Signatures (coherence, average, difference of t<strong>and</strong>em pairs)<br />
proofed to be a powerful tool to visualise the flood dynamics in space <strong>and</strong> time but also the morphologic structure in the<br />
floodplain. As conclusion, it is shown that the combined analysis of radar backscatter <strong>and</strong> coherence information will be<br />
very useful in the flood application domain, especially with respect to risk assessment <strong>and</strong> vulnerability mapping. In<br />
addition, the methods described will support the collection of relevant base data claimed by the EU water framework<br />
directive.<br />
Keywords: Spaceborne Earth Observation, SAR Interferometry, Coherence Analysis, River Dynamics, Flood,<br />
Floodplain Structures, Floodplain Management, River Odra<br />
1. INTRODUCTION<br />
The impacts of human activities <strong>and</strong> water regulation<br />
on rivers <strong>and</strong> floodplains are well known. The more<br />
intensive river basins are used by man <strong>and</strong> the lesser<br />
user functions are adapted to natural river<br />
characteristics, the bigger the damages will be if a flood<br />
crisis happens. Floods of great magnitude cannot be<br />
prevented, but flood damages can be limited. In order<br />
to take successful measures studies of the spatial <strong>and</strong><br />
temporal flood distribution are essential. Floodplain<br />
management requires also a characterisation of<br />
floodplain structures as well as information about flood<br />
extent <strong>and</strong> river dynamics. Remote Sensing <strong>and</strong><br />
Geographic Information Systems (GIS) are important<br />
tools for the analysis <strong>and</strong> visualisation of geographical<br />
entities of river systems <strong>and</strong> for decision support for<br />
management measures [1].<br />
The main aim of the research described in this paper is<br />
to assess the feasibility of using radar satellite imagery<br />
for the floodplain management. ERS-SAR data <strong>and</strong><br />
interferometric products were used to document the<br />
pattern of floodplain inundation, floodplain structures<br />
<strong>and</strong> morphological changes due to flooding (e.g.<br />
erosion, break of a me<strong>and</strong>er). The Odra River basin in<br />
Pol<strong>and</strong> <strong>and</strong> Germany, where large-scale flooding<br />
occurred in summer 1997, was chosen as the focus of<br />
the study.<br />
The investigation was done subsequently to the Odra<br />
flood event. The high data availability, - ERS-1/2<br />
t<strong>and</strong>em data as well as different GIS-data -, but also the<br />
geographic dimension of the flooded areas are<br />
important preconditions to study river dynamics with<br />
radar EO-methods. Selected sites in the floodplain of<br />
the river Odra were analysed regarding the flood risk<br />
686
estimation <strong>and</strong> the vulnerability of resources <strong>and</strong><br />
structures.<br />
2. THE REGION OF INTEREST (ROI)<br />
The Odra River with its source in the Czech Republic<br />
has a length of 850 km. The river catchment area of<br />
about 124.000 km 2 plays an important role for the<br />
water economy of the western part of Pol<strong>and</strong> <strong>and</strong> the<br />
northeastern part of Germany.<br />
Due to regulation works, - which started already at the<br />
end of the 18 th century <strong>and</strong> continued up to the early<br />
decades of the 20 th century -, the course of the river<br />
Odra was shortened by 154 km [2]. The construction of<br />
a regulation infrastructure in the upper, middle <strong>and</strong><br />
lower course of the Odra has caused large-scale<br />
degradation of the river bed (erosion <strong>and</strong> deepening of<br />
the river bed up to 3m) <strong>and</strong> in some areas a lowering of<br />
the ground water table.<br />
On the other h<strong>and</strong>, recent hydraulic measures were not<br />
taken along the Odra. The river has maintained part of<br />
its natural, unregulated character, comprising<br />
floodplain forests <strong>and</strong> associated wildlife, mainly in the<br />
middle <strong>and</strong> lower sections of the river. Therefore some<br />
authors consider the Odra as the most natural large<br />
river in Europe [3].<br />
The greatest flood in the last century caused by the<br />
river Odra occurred in summer 1997. After strong<br />
rainfall in the middle <strong>and</strong> upper catchment areas floods<br />
appear in summer typically with short <strong>and</strong> steep flood<br />
waves. Exceptionally strong-rain falls, which occurred<br />
in three places of origin from 4. -8. July 1997 <strong>and</strong> from<br />
17. -21. July 1997, brought a huge flood disaster to the<br />
Odra <strong>and</strong> most of its tributaries with extensive flooding.<br />
All littoral states along the river Odra were affected by<br />
the flood disaster in July/August 1997. The flood was<br />
particularly strong in Pol<strong>and</strong> <strong>and</strong> the Czech Republic.<br />
Partially, the flood was influenced by human impact.<br />
Above all, the importance of natural retention areas was<br />
shown dramatically: Bursting of dikes in Pol<strong>and</strong> with<br />
an overall length of 40 km brought an inundation area<br />
of 55000 km 2 but also a noticeable reduction of the<br />
flood peak at the German-Polish section of the Odra.<br />
As a result of the persistent high water level two<br />
breaches in the levees of the German-polish Odra<br />
section occurred with a flooded area of 6000 ha in<br />
“Ziltendorfer Niederung”. The decline of water from<br />
the inundated areas took some weeks.<br />
Fig. 1: Research areas in Pol<strong>and</strong> <strong>and</strong> Germany<br />
The following test sites were selected do detect<br />
different river <strong>and</strong> floodplain characteristics:<br />
♦<br />
♦<br />
♦<br />
Floodplain near Krosno Odra skie (mapping of<br />
floodplain structures)<br />
Ziltendorfer Niederung situated at the German-<br />
Polish border south of Frankfurt/Oder (mapping<br />
of inundation pattern)<br />
Odra at the Polish-Czech border (mapping of the<br />
breached me<strong>and</strong>er near Chałupki)<br />
This selection was also done in view of the ground<br />
resolution of the EO instruments. The phenomena to be<br />
analysed must have a minimum spatial dimension in<br />
order to make sense of the use of satellite remote<br />
sensing methods.<br />
3. EARTH OBSERVATION DATA<br />
PROCESSING<br />
3. 1 Floodplain structures<br />
Floodplain structures are important for the potential of<br />
flood retention. In the work described, the term<br />
floodplain structures is not used in the hydraulical<br />
687
sense but in an geographical sense. Therefore, the<br />
structures detection includes different types of l<strong>and</strong><br />
cover like floodplain forests, herbs <strong>and</strong> bushes, thin<br />
woodl<strong>and</strong> vegetation, meadows <strong>and</strong> agricultural field as<br />
well as other flood plain related features such as<br />
rivulets, canals, ditches <strong>and</strong> different types of former<br />
river structures.<br />
In the floodplains of River Odra over 15 ecotopes<br />
could be distinguished from the Atlas of Odra<br />
Floodplain [4]. Because the morphological <strong>and</strong><br />
vegetative structures of each ecotope is known, these<br />
ecotopes can be labelled with a specific hydraulic<br />
roughness factor according to the studies of [5] who<br />
combine ecological <strong>and</strong> hydraulic roughness data. The<br />
preliminary results provide good perspectives for<br />
determining the hydraulic roughness of entire river<br />
sections.<br />
The hydraulic characteristics of river sections vary with<br />
time. It is laborious <strong>and</strong> expensive to adequately<br />
monitor any changes, which occur in the floodplains by<br />
conventional techniques (aerial photographs <strong>and</strong> field<br />
studies). In the future, faster <strong>and</strong> cheaper <strong>and</strong> more<br />
efficient techniques are needed to monitor abundance<br />
<strong>and</strong> structure of vegetation in large parts of a river<br />
basin. Airborne laser-altimetry is used in some studies<br />
[6]. However, in this context, a method involving<br />
spaceborne radar data also seems promising.<br />
Using the derived data as input for water flow models<br />
may provide quick <strong>and</strong> cheap monitoring of the<br />
continuously changing conditions in floodplains, <strong>and</strong><br />
may enable the river manager to ensure sufficient water<br />
flow capacity in a dynamic river bed.<br />
Fig. 2: Floodplain Krosno Odra skie: normal situation (7.5.1996 ERS-1)<br />
This research makes use of spaceborne radar data, -<br />
backscatter & coherence t<strong>and</strong>em data -, to obtain<br />
information on vegetation structure. ERS-1/2 t<strong>and</strong>em<br />
pairs with acquisition dates before <strong>and</strong> duringe the<br />
flood were analysed to determine different structural<br />
types. The well structured floodplains of Krosno<br />
Odra skie, Pol<strong>and</strong> served as test site.<br />
A better visibility of such structures was achieved when<br />
analysing radar images with an acquisition date during<br />
flood events. The reason is that slighthly flooded areas<br />
are increasing the contrast of radar backscatter of<br />
different l<strong>and</strong> units, for instance . due to doublebounce-scattering.<br />
Most of these structures can not be<br />
identified when using satellite imagery taken during<br />
normal water level.<br />
688
Fig. 3: Floodplain Krosno Odra skie: flooded (6.8.1997 ERS-2)<br />
Especially the pattern of former river structures,<br />
floodplain forest <strong>and</strong> thin woodl<strong>and</strong> vegetation can be<br />
identified much better on the flooded backscatter<br />
image.<br />
3. 2 Floodplain inundation pattern<br />
Discussing vulnerability with respect to flood hazards,<br />
the first step is the knowledge of the river dynamics, the<br />
inundation pattern <strong>and</strong> the maximum extent of the flood<br />
in the affected areas. Therefore, the flood lines were<br />
extracted from the images at different acquisitions<br />
during the phase of maximum flooding. For this part of<br />
the study the “Ziltendorfer Niederung” in, Germany<br />
was taken as test site.<br />
Fig. 4: Backscatter information, Ziltendorfer Niederung (Germany, 6.8.1997)<br />
Mapping the flood extent by using only backscatter<br />
information can lead to significant misclassifications.<br />
At the image shown above, there are streets <strong>and</strong><br />
settlements with a strong radar return (white pixels)<br />
despite the fact that they were flooded the time of<br />
image acquisition. Especially flooded tree-lined<br />
avenues produce a signal similar to that of a nonflooded<br />
situation. This phenomenon is related to an<br />
important principle of radar backscatter: the corner<br />
reflection <strong>and</strong> the double-bounce scattering. These<br />
689
misclassifications can cause supply bottlenecks <strong>and</strong><br />
problems in the decision making process in the flood<br />
hazard <strong>and</strong> crisis management.<br />
Two or three dimensional corner reflection is caused by<br />
the existence of buildings. Scattering from a forest<br />
canopy or from tree-lined streets can present a complex<br />
case of volume scattering. Double-bounce scattering<br />
between trunks <strong>and</strong> the ground is one important effect<br />
in volume scattering. This can give a strong return if<br />
the ground is covered with water. Buildings <strong>and</strong> trees<br />
are able to redirect a radar beam which was<br />
backscattered from a smooth water surface back to the<br />
radar sensor. This is why some flooded settlements <strong>and</strong><br />
tree-lined streets can look even brighter than not<br />
flooded areas [7].<br />
Calculating the coherence information of this area<br />
(Fig.5) provides additional data about the surface<br />
conditions. In this case the coherence map supplied<br />
information on flood conditions for the tree-lined<br />
avenues <strong>and</strong> the village.<br />
However, very flat meadows can also be responsible<br />
for a wrong assessment of the flood situation. The<br />
backscatter signal provides no radar return as meadows<br />
react as smooth surfaces like a water body. Water<br />
surfaces without waves act as a smooth surface. When<br />
the radar sensor transmits a beam of radar energy<br />
towards this smooth surface, the result is no backscatter<br />
return to the radar sensor but rather the scattering of the<br />
radar energy away from the sensor. The meadows areas<br />
can not easily be distinguished from flooded areas. The<br />
coherence map shows a very high coherence for the<br />
meadows, which means that this area was not flooded.<br />
The combination of coherence information derived by<br />
t<strong>and</strong>em radar data with backscatter data can avoid<br />
wrong interpretations of the flood situation.<br />
The two examples show that any automated<br />
unsupervised classification performed without an<br />
additional visual interpretation can result in serious<br />
misinformation.<br />
Fig. 5: Coherence information, Ziltendorfer Niederung (Germany, 6.8.1997)<br />
3.3 Flood Dynamics<br />
The detection of morphologic activity, - such as the<br />
bursting of a me<strong>and</strong>er -, provides information about the<br />
risk <strong>and</strong> the vulnerability of special sites in the<br />
floodplain. Visualising the morphodynamic activity can<br />
be done by processing a multitemporal image (RGB),<br />
type “Interferometric Signatures” (red = coherence,<br />
green = average, blue = difference). RGB images are<br />
created with SAR images of the PRI product level.<br />
The multitemporal image is a system of producing<br />
colour imagery that is based upon the additive<br />
properties of primary colours. The multitemporal<br />
technique uses black <strong>and</strong> white radar images taken at<br />
different dates <strong>and</strong> adds them to the red, green <strong>and</strong> blue<br />
colour channels. The resulting multitemporal image<br />
(RGB) reveals changes in the l<strong>and</strong> surface by the<br />
presence of colour on the image. The hue of the colour<br />
indicates the date of the change <strong>and</strong> the intensity of the<br />
colour the degree of change. The reason for change<br />
690
may be the growth of crops, a change in soil moisture<br />
or in soil structure, or the presence of floodwater in one<br />
image when it was absent before.<br />
Morphological changes such as the break of a me<strong>and</strong>er<br />
can be shown using a multitemporal colour composite<br />
with two images taken before <strong>and</strong> after flood <strong>and</strong> using<br />
coherence, average <strong>and</strong> difference information of these<br />
images. The detection of disturbed radar information<br />
requires a validation <strong>and</strong> specification by fieldwork. In<br />
the area of Chałupki the resolution of the ERS-SAR<br />
instrument is at its limit; the Odra is only 25m wide.<br />
3.4 Vulnerability Mapping<br />
The basic problem concerning floodplains is the<br />
conflict between human uses of river environments on<br />
the one h<strong>and</strong> <strong>and</strong> floodplain resources <strong>and</strong> natural<br />
functions one the other. All natural <strong>and</strong> cultural<br />
resources <strong>and</strong> functions of floodplains are subjected to<br />
threats, the most significant of which are related to<br />
human use <strong>and</strong> development.<br />
The permanent location of settlements, industrial<br />
plants, infrastructures as well as agricultural activities<br />
within floodplain are the most common infringements<br />
in contemporary times <strong>and</strong> result annually in ever<br />
increasing damages, risk for human life, personal<br />
inconveniences, <strong>and</strong> material loss world-wide, when<br />
floodwaters reclaim these l<strong>and</strong>s. Natural hazards are<br />
having an increased impact on human settlements,<br />
probably because of the greater number of settlements<br />
<strong>and</strong> their increased vulnerability due to their<br />
uncontrolled extension to high risk areas. The response<br />
<strong>and</strong> policy options to counteract are wise l<strong>and</strong> use <strong>and</strong><br />
emergency planning to reduce the impacts of floods <strong>and</strong><br />
other hazards <strong>and</strong> their interactions with human<br />
activities.<br />
The delineation of floodplains <strong>and</strong> socio-economic<br />
characteristics on maps to derive vulnerability<br />
information is a basic necessity for floodplain<br />
management.<br />
In the case of the Odra floodplain the information<br />
derived by radar satellite data will be integrated with<br />
different sources of Geodata to produce maps for the<br />
floodplain management.<br />
The derived risk <strong>and</strong> vulnerability maps will be used<br />
for flood risk estimation in sensitive areas. In addition,<br />
these maps will yield vital information to support<br />
decisions concerning the siting of flood-control works<br />
such as reservoirs <strong>and</strong> levees, or floodplain zoning<br />
provisions.<br />
In terms of describing characteristics for floodplain<br />
management by using radar satellite imagery the<br />
investigations permit the following conclusions:<br />
• Mapping floodplain structures with radar<br />
imagery with acquisition data during<br />
flood events improve the detection of<br />
different structural types, such as former<br />
river structures, floodplain forest <strong>and</strong> thin<br />
woodl<strong>and</strong> vegetation.<br />
• Combining backscatter <strong>and</strong> coherence<br />
information improves the mapping of the<br />
flood pattern <strong>and</strong> mitigates the errors<br />
derived by double-bounce-scattering <strong>and</strong><br />
corner-reflection occurring in backscatter<br />
data. This improved assessment of the<br />
flood situation is important for the<br />
decision making process in flood hazard<br />
management.<br />
• The detection of morphological changes,<br />
- the break of a me<strong>and</strong>er near Chalupki -,<br />
was taken at the limits of spatial<br />
resolution of ERS-1/2 due to the<br />
narrowness of the River Odra of 25 m in<br />
this area. For larger rivers it should be<br />
possible to map this kind of activity by<br />
using RGB type “Interferometric<br />
Signatures”. The detection of a<br />
morphodynamic process must be verified<br />
by field studies.<br />
However, floods are natural events <strong>and</strong> turn into a<br />
threat only through uncontrolled use <strong>and</strong> settlement in<br />
the potentially targeted areas. The vulnerability <strong>and</strong> risk<br />
need to be carefully estimated to optimise future<br />
planning of living in floodplains.<br />
There is also a necessity to combine all sources of data<br />
available for rivers <strong>and</strong> floodplains, not only by<br />
remotely sensed data but also using other Geodata, to<br />
improve floodplain management. Nevertheless,<br />
important progress can be achieved through<br />
interpreting <strong>and</strong> estimating existing data sets.<br />
5. ACKNOWLEDGEMENTS<br />
The authors wish to thank the German Federal<br />
Environment Foundation for funding the project <strong>and</strong><br />
the European Space Agency (ESA) providing the ERS-<br />
SAR data within the AO3-program.<br />
4. CONCLUSIONS<br />
691
6. REFERENCES<br />
[1] Leuven, R.S.E.W. & Poudevigne, I. & Teeuw,<br />
R.M. [Eds.] (2000): Application of GIS <strong>and</strong> Remote<br />
Sensing in river studies. – Backhuys Publishers,<br />
Leiden.<br />
[2] Parzonka, W. & Bartnik, W. (1998): Degradation<br />
of Middle Odra caused by regulation works. – In:<br />
Gayer, J. & Scheuerlein, H. & Starosolsky, O.<br />
[Eds.], Proceed. Intern. Confer. River Development,<br />
Budapest, Hungary, April 16-18, 1998, VITUKI,<br />
Budapest, pp. 345-352.<br />
[3] Nienhuis, P.H. & Chojnacki, J.C. & Harms, O. &<br />
Majewski, W. & Parzonka, W. & Prus, T. (2000):<br />
Elbe, Odra <strong>and</strong> Vistula: Reference Rivers for the<br />
restoration of biodiversity <strong>and</strong> habitat quality. – In:<br />
Smits, A.J.M. & Nienhuis, P.H. & Leuven,<br />
R.S.E.W. [Eds.]: New Approaches to River<br />
Management, Leiden, The Netherl<strong>and</strong>s.<br />
[4] WWF (2000): Der Oder-Auen-Atlas. – Auen-<br />
Institut des WWF, Rastatt.<br />
[5] Smits, A.J.M. & Havinga, H. & Marteijn, E.C.L.<br />
(2000): New concepts in river <strong>and</strong> water<br />
management in the Rhine river basin: how to live<br />
with the unexpected? – In: Smits, A.J.M. &<br />
Nienhuis, P.H. & Leuven, R.S.E.W. [Eds.]: New<br />
Approaches to River Management, Leiden, The<br />
Netherl<strong>and</strong>s.<br />
[6] Asselmann, N. (1999): Laseraltimetrie an<br />
vegetatieruwheid. – Report Q2577. WL Delft<br />
Hydraulics, Delft.<br />
[7] Halounova, Lena (1998): Integration of various<br />
types of remote sensing data for the Morava River<br />
catchment evaluation. – In: <strong>International</strong> Archives<br />
of Photogrammetry <strong>and</strong> Remote Sensing, Vol. 32,<br />
Part 7, Budapest.<br />
692
Optimal Groundwater Exploitation <strong>and</strong> Pollution<br />
Control<br />
Andrea Bagnera c , Marco Massabò a , Riccardo Minciardi a,b ,<br />
Luca Molini a , Michela Robba a,b,* , Roberto Sacile a,b<br />
a CIMA- Interuniversity Center of Research in <strong>Environmental</strong> Monitoring<br />
b DIST- Department of Communication, Computer <strong>and</strong> System Sciences<br />
c AMGA-Azienda Mediterranea Gas e Acqua<br />
* Corresponding author: michela.robba@unige.it<br />
DIST - University of Genova, via Opera Pia 13, 16100, Genova, Italy<br />
Abstract: Aquifer management is a complex problem in which various aspects should be taken into account.<br />
Specifically, there are conflicting objectives that should be achieved. On one side, there is the necessity to<br />
satisfy the water dem<strong>and</strong>, on the other the resource water should be protected by infiltration of pollutants or<br />
substances that could reduce its availability in terms of short term <strong>and</strong> long term management. The aim of<br />
this paper is to develop a management model that is able to define the optimal pumping pattern for p<br />
(p=1,…,P) wells that withdraw water from an aquifer (characterized by pollutant contamination) <strong>and</strong><br />
hydraulically interact, with the objectives of satisfying an expressed water dem<strong>and</strong> <strong>and</strong> control pollution. In<br />
order to formalize <strong>and</strong> solve the management problem, it is necessary to consider the equations governing<br />
flow <strong>and</strong> mass transport of the biodegradable pollutants characterizing the aquifer. Such equations may be<br />
solved by using a finite-difference numerical scheme. In this work, the numerical scheme is embedded in the<br />
management model. The decision (control) variables that are considered in the optimisation problems are the<br />
water flows pumped at each well p, at time interval t. Such flows influence the state variables of the system,<br />
that is, the hydraulic head <strong>and</strong> the pollutant concentrations in the aquifer. The objective function to be<br />
minimized in the optimisation problem includes three terms: water dem<strong>and</strong> dissatisfaction, pollutant<br />
concentrations in the extracted water, <strong>and</strong> pollutant concentrations in all cells of the discretized aquifer.<br />
Finally, the optimisation problem has been solved for a specific case study (Savona District, Italy), relevant<br />
to a confined aquifer affected by nitrate pollution deriving from agriculture activities.<br />
Keywords: Groundwater management, optimisation, pollution, decision support system, optimal pumping<br />
pattern.<br />
1. INTRODUCTION<br />
Water is essential for human life <strong>and</strong> its protection<br />
<strong>and</strong> sustainable exploitation are crucial tasks.<br />
Specifically, it is necessary to identify the possible<br />
water bodies that could be exploited (surface<br />
water, groundwater, reservoirs, etc.) <strong>and</strong>,<br />
according to water dem<strong>and</strong> needs, it is vital to<br />
define strategies that preserve the water resource<br />
from depletion <strong>and</strong> pollution <strong>and</strong> that are<br />
environmentally sustainable. The application of<br />
optimization techniques in groundwater quantity<br />
<strong>and</strong> quality management has been deeply<br />
investigated by Das <strong>and</strong> Datta (2001). In that<br />
work, they present a complete state of the art of the<br />
different optimisation approaches that have been<br />
applied to groundwater management. Specifically,<br />
the combined use of simulation <strong>and</strong> optimisation<br />
techniques is shown to be a powerful <strong>and</strong> useful<br />
method to determine planning <strong>and</strong> management<br />
strategies for optimal design <strong>and</strong> operation of<br />
groundwater systems. The simulation model can<br />
be combined with the management model either by<br />
using the system state equations as binding<br />
constraints in the optimisation model or by using a<br />
response matrix or an external simulation model.<br />
In literature, different techniques may be found to<br />
help in finding solution to the various management<br />
problems. Katsifarakis et al. (1999) combine the<br />
boundary element method (BEM) <strong>and</strong> genetic<br />
693
algorithms (GAs) to find optimal solution in three<br />
classes of commonly encountered groundwater<br />
flow <strong>and</strong> mass transport problems: determination<br />
of transmissivities in zoned aquifers, minimization<br />
of pumping cost from any number of wells under<br />
various constraints, hydrodynamic control of a<br />
contaminant plume by means of pumping <strong>and</strong><br />
injection wells. Psilovikos (1999) analyses the<br />
possibility of solving two management problems<br />
formulated as linear programming <strong>and</strong> mixed<br />
integer linear programming through the integration<br />
of simulation <strong>and</strong> optimization packages.<br />
The aim of this paper is to develop a management<br />
model that is able to define the optimal pumping<br />
pattern for p (p=1,…,P) wells that withdraw water<br />
from an aquifer, characterized by a point source<br />
pollutant contamination, with the objective of<br />
satisfying the requested water dem<strong>and</strong> <strong>and</strong> control<br />
pollution. Specifically, three different objectives<br />
(minimization of water dem<strong>and</strong> dissatisfaction,<br />
minimization of pollution in the aquifer <strong>and</strong><br />
minimization of pollution in the extracted water)<br />
have been considered. The state equations that<br />
describe the physical behaviour of the system are<br />
embedded as constraints in the optimisation model.<br />
2. THE PHYSICAL-CHEMICAL MODEL<br />
The overall model of the considered system may<br />
be decomposed into a hydraulic component <strong>and</strong> a<br />
chemical one. As regards the hydraulic component,<br />
The adopted model is drawn by Theim (1906) <strong>and</strong><br />
particularly focuses on the behaviour of the<br />
piezometric head at local scale, <strong>and</strong> specifically on<br />
the interaction among the various wells. The<br />
pollutant mass transport equation is solved using a<br />
finite difference scheme. The hypotheses under<br />
which our model is applied are:<br />
1. confined, homogeneous <strong>and</strong> isotropic aquifer;<br />
2. source terms represented by pumping wells<br />
with Q p (t) discharge pattern for p=1,…P ;<br />
3. wells completely penetrating <strong>and</strong> located in<br />
(x p , y p ), p=1,…P.<br />
The third hypothesis means that the fluid flow in<br />
the aquifer is only bi-dimensional, since the<br />
vertical component of the velocity field is close to<br />
zero when the wells pump from all the aquifer<br />
thickness. The flow equation with the relative<br />
initial <strong>and</strong> boundary conditions are:<br />
<br />
P<br />
∂ h ∂ h ∂ h<br />
K<br />
+ K = S s − δ − −<br />
∂ ∂ ∂<br />
∑ Q p(<br />
t)<br />
( x x p,<br />
y y p)<br />
x y t<br />
p = 1<br />
<br />
h<br />
( x,<br />
y,<br />
t = 0) = H<br />
(1)<br />
<br />
2 2<br />
h<br />
( x,<br />
y,<br />
t)<br />
= H if {(<br />
x,<br />
y)<br />
x + y = R }<br />
<br />
<br />
where H is the undisturbed piezometric level, R is<br />
the influence radius, K is the hydraulic<br />
conductivity, h is the piezometric head in the<br />
aquifer, S s is the specific storativity, <strong>and</strong> δ is the<br />
Kronecker Delta .<br />
The characteristic time scale of equation (1) is:<br />
S S R<br />
Tt =<br />
K<br />
It represents the time scale of the transition<br />
behaviour of the piezometric head within a<br />
regulation interval T R of the pumping flow. When<br />
the transition time scale T t is negligible with<br />
respect to the regulation time step T R it is possible<br />
to consider the flow equation under steady state in<br />
each regulation time interval. In this work we<br />
consider steady state conditions for successive<br />
regulation time steps.<br />
Integrating eq. (1), it is possible to evaluate the<br />
piezometric head, in stationary condition:<br />
h(<br />
x,<br />
y,<br />
t)<br />
= H +<br />
P<br />
∑<br />
p=<br />
1<br />
QP<br />
( t)<br />
ln<br />
2πT<br />
( x − x<br />
P<br />
)<br />
2<br />
+ ( y − y<br />
R<br />
P<br />
)<br />
(2)<br />
where T=KB is the trasmissivity of the<br />
homogeneous aquifer <strong>and</strong> B is its thickness.<br />
Deriving equation (2) <strong>and</strong> using the Darcy law, it<br />
is possible to write an analytical expression for the<br />
velocity field due to P pumping wells spread in the<br />
domain <strong>and</strong> having a different pumping rate Q w .<br />
Let n the soil porosity, <strong>and</strong> u <strong>and</strong> v the pore scale<br />
velocities of the fluid flowing in the aquifer along<br />
x <strong>and</strong> y directions, respectively. The pore scale<br />
velocities may be expressed as follows:<br />
P<br />
−1<br />
Q p ( t)<br />
( x − x p )<br />
u( x,<br />
y,<br />
t)<br />
=<br />
2 n π B<br />
∑<br />
(3)<br />
2<br />
y )<br />
2<br />
p = 1(<br />
x − x p ) + ( y −<br />
P<br />
Qp(<br />
t)<br />
( y −<br />
∑<br />
2<br />
p = 1(<br />
x − xp)<br />
+ ( y −<br />
−1<br />
y p)<br />
v( x,<br />
y,<br />
t)<br />
=<br />
(4)<br />
2<br />
2 n π B<br />
y )<br />
The knowledge of the velocity field is needed in<br />
order to solve the mass transport equation. In this<br />
work, a contaminant transport simulation model is<br />
used, which is able to predict the concentration<br />
behaviour in the aquifer for a biodegradable<br />
p<br />
p<br />
2<br />
694
pollutant. Since in many application concerning<br />
the monitoring of groundwater quality, the only<br />
concentration measures that are often available are<br />
the mean value over the thickness of the sampling<br />
well, the averaged mass transport equation is taken<br />
into account in this work. These equations can be<br />
obtained by vertically averaging the classical<br />
advection-dispersion equation over the thickness<br />
of the aquifer system (Willis et al.,1998; Bear,<br />
1972). These authors have found out the results<br />
under the following conditions:<br />
− horizontal flow<br />
− porosity <strong>and</strong> dispersion coefficients are<br />
constant in all the aquifer<br />
− the source <strong>and</strong> sink terms are represented by<br />
pumping wells<br />
− recharging phenomena are negligible because<br />
the aquifer is confined<br />
− the bio-degradation coefficient is constant in<br />
all the aquifer.<br />
The partial differential equation for the averaged<br />
concentration C is<br />
∂ C<br />
∂ t<br />
∂<br />
= −<br />
2<br />
∂ C<br />
+ D ±<br />
2<br />
∂ y<br />
( u C )<br />
∂ x<br />
P<br />
∑<br />
p=1<br />
∂<br />
−<br />
( v C ) ∂ C<br />
+ D +<br />
2<br />
∂ y ∂ x<br />
C Qp(<br />
t)<br />
δ(<br />
x−x<br />
B<br />
p,<br />
2<br />
y−y<br />
) − k C<br />
p<br />
(5)<br />
where k is the bio-degradation coefficient for the<br />
pollutant concentration, considering a first order<br />
kinetics.<br />
The boundary <strong>and</strong> initial conditions needed to<br />
solve equation (5) are:<br />
Co<br />
if ( x,<br />
y)<br />
= ( xo,<br />
yo<br />
)<br />
C(<br />
x,<br />
y,<br />
t)<br />
= <br />
0<br />
otherwise<br />
∂<br />
C<br />
<br />
∂ x<br />
<br />
∂<br />
C<br />
<br />
∂ y<br />
<br />
x=<br />
0, L<br />
y=<br />
0, L<br />
= 0<br />
= 0<br />
(6a)<br />
(6b)<br />
(6c)<br />
where x 0 , y 0 is the point corresponding to the<br />
pollutant source.<br />
The mass transport equation (5) can be solved by<br />
using the classical central finite difference scheme<br />
in space, <strong>and</strong> an implicit method in time (Fletcher,<br />
1991). The stability of the methods is controlled by<br />
the dispersion <strong>and</strong> advection Currant number,<br />
defined as<br />
v ∆t<br />
D ∆t<br />
Cadv<br />
= <strong>and</strong> Cdisp<br />
= .<br />
2<br />
∆ L<br />
∆ L<br />
The finite difference representation of equation (5)<br />
is for any point i,j, (a generic point (x,y) on the<br />
grid) at any time t) is :<br />
C<br />
t+<br />
1<br />
i,<br />
j<br />
( C<br />
= D<br />
−<br />
P<br />
∑<br />
p=<br />
1<br />
− C<br />
∆t<br />
C<br />
t<br />
i,<br />
j<br />
t+<br />
1<br />
i+<br />
1, j<br />
t+<br />
1<br />
i,<br />
j<br />
+ u<br />
t+<br />
1<br />
p<br />
t+<br />
1<br />
i , j<br />
− 2 C<br />
Q<br />
B∆x<br />
∆y<br />
∆x<br />
( C<br />
t+<br />
1<br />
i,<br />
j<br />
2<br />
∗<br />
t+<br />
1<br />
i+<br />
1, j<br />
+ C<br />
δ ( i − i<br />
p,<br />
− C<br />
2∆x<br />
t+<br />
1<br />
i−1,<br />
j<br />
j − j<br />
t+<br />
1<br />
i−1,<br />
j<br />
) ( C<br />
+ D<br />
p,<br />
)<br />
+ v<br />
) − k C<br />
t+<br />
1<br />
i , j<br />
t+<br />
1<br />
i,<br />
j+<br />
1<br />
t+<br />
1<br />
i,<br />
j<br />
( C<br />
t+<br />
1<br />
i,<br />
j+<br />
1<br />
− 2 C<br />
∆y<br />
− C<br />
2∆x<br />
t+<br />
1<br />
i,<br />
j<br />
2<br />
t+<br />
1<br />
i,<br />
j−1<br />
+ C<br />
(7)<br />
where (i p , j p ) is the location of the wells on the<br />
grid.<br />
3. THE MANAGEMENT MODEL<br />
The main purpose of this paper is to present a<br />
decision model able to manage groundwater<br />
resources, satisfying the water dem<strong>and</strong> <strong>and</strong><br />
controlling the aquifer pollution. Specific control<br />
<strong>and</strong> state variables have been defined in order to<br />
formalize suitable objective functions <strong>and</strong><br />
constraints. The control variables that characterize<br />
the system are the quantity of water that is<br />
extracted in each well p in time interval t. These<br />
quantities influence both the hydraulic head <strong>and</strong><br />
the concentration distributions in the aquifer. The<br />
state variables of the system correspond to the<br />
pollutant concentration to the hydraulic head in the<br />
aquifer. Let Q p,t be the control variable that<br />
represents the quantity of water that is extracted in<br />
each well p at time t. These quantities influence<br />
both the hydraulic head <strong>and</strong> the concentration<br />
distributions in the aquifer. The evolution in time<br />
<strong>and</strong> space of pollutant <strong>and</strong> hydraulic head in the<br />
aquifer, that are the system state variables, has to<br />
be modelled as proved by (2) <strong>and</strong> (7). Moreover let<br />
C i, j,<br />
t represent the pollutant concentration in the<br />
aquifer at time t in point (i,j). In this work, the<br />
pollutant concentration in the extracted water from<br />
wells corresponds to the pollutant concentration<br />
C i j,<br />
t<br />
, in the nodes of the grid where the wells are<br />
located. Specifically, C i, p represents the pollutant<br />
concentration in well p (p=1,…,P) at time t<br />
(t=1,…,T), where p=(i p ,j p ).<br />
The objective function considered in this paper is<br />
composed by three terms: minimization of water<br />
)<br />
t+<br />
1<br />
i,<br />
j−1<br />
)<br />
695
dem<strong>and</strong> dissatisfaction, minimization of pollutant<br />
concentration in extracted water, minimization of<br />
pollutant concentration in all nodes of the<br />
discretized aquifer. Every objective is weighed<br />
with specific coefficients in an overall objective<br />
function. The optimisation problem turns out to be<br />
non linear.<br />
3.1 Minimization of water dem<strong>and</strong><br />
dissatisfaction<br />
The water dem<strong>and</strong> dissatisfaction corresponds to<br />
the difference between the requested water <strong>and</strong> the<br />
extracted water from the wells, when such a<br />
difference is positive or zero. Thus this objective<br />
function (to be minimized) among this difference<br />
<strong>and</strong> zero, can be expressed as<br />
⌈<br />
⌉<br />
<br />
N T<br />
max Q ( )<br />
<br />
REQ<br />
− ∑ ∑ Q p,<br />
t , 0 (8)<br />
<br />
⌊<br />
p = 1t<br />
= 1 ⌋<br />
where:<br />
• Q REQ represents the overall requested water<br />
flow, expressed in l/s, over the whole decision<br />
horizon;<br />
• N is the number of available wells;<br />
• T is the planning horizon.<br />
3.2 Minimization of pollutant presence in<br />
extracted water<br />
Another objective of the optimization problem is<br />
to minimize the impact of the pollutant in the<br />
water extracted from wells. Let C ( p,<br />
t)<br />
be the<br />
pollutant concentration, expressed in mg/l, of the<br />
water extracted from well p in the t-th time<br />
interval. This objective function can be formalized<br />
as follows:<br />
N T<br />
∑ ∑ Q(<br />
p,<br />
t)<br />
F C<br />
p = 1t<br />
= 1<br />
where [ C ( p,<br />
t)<br />
]<br />
[ ( p,<br />
t)<br />
]<br />
(9)<br />
F is a function of pollutant<br />
concentration <strong>and</strong> has been considered to be<br />
2<br />
[ C ( p,<br />
t)<br />
] C ( p,<br />
t)<br />
F = (10)<br />
3.3 Minimization of pollutant concentration in<br />
the aquifer<br />
The aquifer pollution should be limited for two<br />
important reasons: the preservation of the water<br />
resource <strong>and</strong> the possibility to satisfy water<br />
dem<strong>and</strong> for a longer time in the future. Indicating<br />
with C ( i,<br />
j,<br />
T ) the pollutant concentration [mg/l]<br />
at node (i,j) at the end of the optimisation period,<br />
the objective function to be minimized is<br />
I J<br />
∑ ∑ C ( i,<br />
j,<br />
T )<br />
(11)<br />
i = 0 j = 0<br />
where i <strong>and</strong> j are the coordinates of the nodes of<br />
the grid representing the aquifer.<br />
3.4 The overall objective function<br />
The overall objective function to be minimized is<br />
given by the weighted sum of functions (8), (9),<br />
<strong>and</strong> (11), each one multiplied by a specific<br />
weighting factor. Then, the overall objective<br />
function is the minimization of by<br />
⌈<br />
min imize<br />
<br />
⌊<br />
N T<br />
β⋅ ∑ ∑ Q(<br />
p,<br />
t)<br />
F C<br />
p = 1t<br />
= 1<br />
{ α ⋅max<br />
<br />
Q − ∑ ∑ Q( p,<br />
t)<br />
REQ<br />
N<br />
p = 1t<br />
= 1<br />
I<br />
[ ( p,<br />
t)<br />
] + γ ⋅ ∑ ∑ C ( i,<br />
j,<br />
T ) }<br />
T<br />
J<br />
i = 0 j = 0<br />
⌉<br />
<br />
, 0<br />
+<br />
<br />
⌋<br />
(12)<br />
where α , β , <strong>and</strong> γ are suitable weighting<br />
coefficients.<br />
3.5 The constraints<br />
There are different kinds of constraints that should<br />
be considered in the model. The first class of<br />
constraints represents the state equations that<br />
represent the dynamics of the pollutant<br />
concentrations <strong>and</strong> of the hydraulic head, as driven<br />
by the control variables.<br />
The other constraints are: the hydraulic head<br />
limitations due to hydraulic conditions that must<br />
be respected, the wells capacity, <strong>and</strong> the<br />
constraints that avoid to extract water from wells<br />
when the pollutant concentration exceeds the one<br />
imposed by regulations.<br />
Besides, one can make that the equation on which<br />
the physical model is based hold only under<br />
specific hypothesis. One of them is that the aquifer<br />
is “in pressure”, that is to say:<br />
696
h(i,j,t) > B<br />
(13)<br />
where B is the aquifer thickness.<br />
Besides the water flow extracted from a well must<br />
be less or equal to its capacity, namely<br />
Q ≤ W p = 1,<br />
…,P<br />
t = 1,<br />
…,T<br />
(14)<br />
p,t p<br />
Finally, the water extracted must have a<br />
concentration of pollutant not exceeding a specific<br />
bound defined by regulations. In other words, this<br />
means that:<br />
C<br />
*<br />
p,t<br />
> C ⇒ Q<br />
p,t<br />
= 0 p=1,…,P t=1,…,T (15)<br />
where C* is the maximum pollutant concentration<br />
allowed by regulation.<br />
4. THE CASE STUDY<br />
The model has been applied to a study area of<br />
50mx50m in which three wells pump water from a<br />
confined aquifer that is affected by nitrate<br />
pollution. The spatial location of the pumping<br />
wells respect to the source of pollution sees well 1<br />
as the nearest to the pollutant source, while well 3<br />
is the most far. The case study is located within the<br />
Ceriale Municipality (Savona, Italy), <strong>and</strong> the<br />
confined aquifer is affected by nitrate pollution<br />
due to agricultural practices. The well field is used<br />
to extract water for drinking use, but it is<br />
periodically closed because of the pollution due to<br />
nitrates infiltration. The application of the<br />
optimisation model allows finding the optimal<br />
pumping pattern in order to satisfy the water<br />
dem<strong>and</strong> needs <strong>and</strong> to control the advancing of the<br />
pollutants in the aquifer.<br />
The optimisation problem has been solved over a<br />
three months period. The aquifer has been<br />
discretized in space (1 m), <strong>and</strong> in time (10 hours).<br />
The total water dem<strong>and</strong> is 60 l/min, while the<br />
pollutant concentration is 150 mg/l. The initial<br />
value of hydraulic head is 20 m, while the aquifer<br />
thickness is equal to 15 m. The problem has been<br />
solved for two different cases: Case1 (each well is<br />
able to pump the total amount of the water dem<strong>and</strong><br />
10 l/s), <strong>and</strong> Case2 (the three wells can pump at<br />
maximum the same quantity of water (3.33 l/s)).<br />
The management problem formalized in the<br />
previous section has been solved, in both cases,<br />
over a time horizon of three months.<br />
In Case 1, only well 1 (the nearest to the pollution<br />
source) overcomes the law limit (50 mg/l) reaching<br />
a concentration value of about 106 mg/l. Well 2<br />
<strong>and</strong> well 3 (the farthest from the pollution source)<br />
reach a maximum concentration of 40 <strong>and</strong> 21 mg/l,<br />
respectively, far below the threshold for the whole<br />
length of time horizon. Figure 1 shows the pattern<br />
of the concentration over time, for the three wells.<br />
Figure 1. Pollutant concentration in the extracted<br />
water (first case)<br />
C<br />
[m g /l]<br />
1 2 0<br />
5 0<br />
0<br />
T [ h ] 2 1 6 0<br />
W e ll 1<br />
C lim = 5 0<br />
W e ll 2<br />
W e ll 3<br />
In Case 2, see Figure 2, no management policy can<br />
be applied, as, in order to satisfy the water<br />
dem<strong>and</strong>, every well has to pump the maximum<br />
flow for the whole management horizon. Note that<br />
water pumped bywell 1 overcomes the limit after<br />
17 days, whereas, well 2 reaches such a limit after<br />
60 days. Only well 3 can work over 3 months.<br />
Figure 2. The pollutant concentration in the<br />
extracted water (Case 2)<br />
8 0<br />
C<br />
[m g /l]<br />
5 0<br />
0<br />
T [h ] 2 1 6 0<br />
W ell 1<br />
W ell 2<br />
W ell 3<br />
Finally, a sensitivity analysis has been performed<br />
on the weight coefficients of the objective<br />
function. Specifically, the value of the weighting<br />
factors is changed in order to make one objective<br />
more significant respect to the others.<br />
Figure 3 reports the results, assuming the<br />
weighting factor α , relevant to water dem<strong>and</strong><br />
satisfaction objective (equation (8)), as prevailing.<br />
This assumption forces the system to maximize the<br />
extracted water quantity, setting each well<br />
pumping rate to the maximum <strong>and</strong> stopping<br />
extraction when pollutant concentration is over the<br />
limit. This strategy causes a very quick<br />
overcoming of the fixed potability threshold in the<br />
extracted water: well 1 is over the limit in 7 days,<br />
well 2 in about 46 days, well 3 in 59.<br />
697
1 10<br />
C<br />
8 0<br />
[mg/l]<br />
5 0<br />
3 0<br />
1 0<br />
0<br />
Figure 3. The pollutant concentration in the<br />
extracted water (maximum extraction case)<br />
In order to see how solution varies when the<br />
minimization of the pollutant concentration in the<br />
extracted water is taken as the primary objective,<br />
the weighting factor β relevant to equation (9) is<br />
increased in order to be predominant respect to the<br />
other coefficients. Figure 4 reports the results for<br />
the optimisation problem. Specifically, this<br />
strategy can satisfy the quality of the pumped<br />
water (see Figure 4) but it turns out that in the<br />
most part of the time interval the overall pumped<br />
water is far below the request.<br />
9 0<br />
T [h] 2 16 0<br />
W ell 1<br />
W ell 2<br />
W ell 3<br />
minimization of pollutant concentration in all<br />
nodes of the discretized aquifer). Every objective<br />
is weighted by a factor whose value is set by the<br />
decision maker. Optimal solutions, which may<br />
support the decision makers in the evaluation of an<br />
extraction strategy, can be obtained solving the<br />
related mathematical programming problem that<br />
embeds a simplified simulation model of the<br />
aquifer dynamics. A preliminary sensitivity<br />
analysis on the parameters representing the<br />
weighting factors is reported.<br />
Future developments may regard the definition of<br />
other decision variables that can give the<br />
possibility of considering the possibility of the<br />
installation of a treatment plant or of the<br />
introduction of wells for the injection of water in<br />
the aquifer (in order to control the direction of the<br />
contaminant plume). Moreover, the physical model<br />
complexity may be increased: the most restrictive<br />
assumption is the homogeneity of the hydraulic<br />
conductivity of the aquifer (the management model<br />
can be adapted to this case using an appropriate<br />
solution for the velocity field). Finally, a different<br />
approach for groundwater management might be<br />
the identification of empirical models (both from<br />
simulation runs <strong>and</strong> real data collection) able to<br />
describe the response of the aquifer in every grid<br />
point (in terms of hydraulic head <strong>and</strong> pollutant<br />
concentration) to the pumping from the different<br />
wells.<br />
C<br />
[m g /l]<br />
5 0<br />
0<br />
Figure 4. Pollutant concentrations in pumped<br />
water<br />
5. CONCLUSIONS<br />
T [h ] 2 1 6 0<br />
W e ll 1<br />
W e ll 2<br />
W e ll 3<br />
The management of an aquifer is a very complex<br />
task since it is necessary to link together<br />
optimisation <strong>and</strong> simulation models in order to<br />
find strategies that are able to take into account<br />
several aspects (physical, economic,<br />
environmental, etc.). In this work, a mathematical<br />
formulation of the management problem has been<br />
presented, with reference to the extraction of water<br />
form wells, in order to satisfy three conflicting<br />
objectives (namely, the minimization of water<br />
dem<strong>and</strong> dissatisfaction, the minimization of<br />
pollutant concentration in extracted water, <strong>and</strong> the<br />
6. REFERENCES<br />
Das, A., Datta, B, Application of optimisation<br />
techniques in groundwater quantity <strong>and</strong><br />
quality management, Sadhana, 26(4) pp.<br />
293-316, 2001<br />
K.L. Katsifarakis, D.K. Karpouzos, N. Theossiou,<br />
Combined use of BEM <strong>and</strong> genetic<br />
algorithms in groundwater flow <strong>and</strong> mass<br />
transport problems, Engineering analysis<br />
with boundary elements, 23, pp.555-<br />
565,1999<br />
Fletcher, C. A. J, Computational Techniques for<br />
Fluid Dynamics 1. Springer Series in<br />
Computational Physics. Springer-Verlag,<br />
2st edition, pp. 253, 1991<br />
Psilovikos, A.A. Optimization models in<br />
groundwater Management, Based on Linear<br />
<strong>and</strong> Mixed Integer Programming. An<br />
application to a Greek Hydrogeological<br />
Basin. Phys.Chem. Earth (B) 24(1-2)<br />
pp.139-144, 1999<br />
Theim, G. Hydrologische Methoden, Gebhardt,<br />
Leipzig, pp. 56, 1906 (in German)<br />
698
¡<br />
Towards an <strong>Environmental</strong> DSS based on<br />
Spatio-Temporal Markov Chain Approximation<br />
Gérard Balent ,Marc Deconchat , Sylvie Ladet , Roger Martin-Clouaire ¡ <strong>and</strong> Régis Sabbadin ¡<br />
INRA-DYNAFOR,BP 27, 31326 Castanet-Tolosan Cedex, France<br />
INRA-BIA, BP 27, 31326 Castanet-Tolosan Cedex, France<br />
Abstract: The aim of this paper is to provide a mathematical model for assessing the influence of forest<br />
fragmentation on the dynamics of animal biodiversity in a changing l<strong>and</strong>scape. The model is based on a<br />
stochastic, spatially explicit population dynamics model which takes both temporal <strong>and</strong> spatial dynamics of<br />
biological processes into account. Unfortunately, this model is not tractable, so we will use a Monte Carlo<br />
simulation method in order to approximate the multidimensional r<strong>and</strong>om variables involved.<br />
The main strength of our approach is its ability to model generic biological <strong>and</strong> socio-economic dynamic<br />
processes, which are both explicitly spatial <strong>and</strong> stochastic. In order to demonstrate the usefulness of our biodiversity<br />
dynamics modeling tool we use available spatial data on the presence/absence of Erithacus Rubecula<br />
(robin) at different time points in the “Vallée de la Nère”, an area of fragmented forest located in the southwest<br />
of France, near Toulouse.<br />
Keywords: L<strong>and</strong>scape fragmentation, population dynamics, spatio-temporal Markov Chain approximation.<br />
1 INTRODUCTION<br />
Conservation <strong>and</strong> biodiversity management are important<br />
issues, especially in places where global<br />
climatic or l<strong>and</strong>scape changes (fragmentation) may<br />
drastically transform the ecosystem, with positive<br />
or negative influences upon human activities Huston<br />
[1994]. Underst<strong>and</strong>ing <strong>and</strong> anticipating these<br />
changes requires assessment of large regions in a<br />
quick <strong>and</strong> reliable way, but most predictive models<br />
of biodiversity operate at fine-grained spatial scale<br />
Deconchat <strong>and</strong> Balent [2001], or require a great<br />
amount of information Conroy <strong>and</strong> Noon [1996].<br />
Remote sensed data provide a unique way to obtain<br />
habitat description over large areas, provided that<br />
less precise prediction is accepted Williams [1996].<br />
The main difficulty is to establish a good statistical<br />
relationship between a set of species occurrence observations<br />
<strong>and</strong> the data sensed from space.<br />
Such a model has been proposed by Lauga <strong>and</strong><br />
Joachim [1992], which can be applied over large areas<br />
to produce a map of presence probabilities for<br />
a given species. However, this “static” approach<br />
neither takes into account the dynamics of the processes<br />
involved, nor the uncertainty pervading them.<br />
The aim of this paper is to tackle these aspects by<br />
providing a model of the influence of forest fragmentation<br />
on the dynamics of animal biodiversity in<br />
a changing l<strong>and</strong>scape. More precisely, we provide a<br />
mathematical model for studying the effect of l<strong>and</strong>scape<br />
use change on biodiversity. The main strength<br />
of our approach is its ability to model generic biological<br />
<strong>and</strong> socio-economic dynamic processes,<br />
which are both explicitly spatial <strong>and</strong> stochastic.<br />
Our method will be illustrated by a study of the<br />
dynamics of robins (Erithacus Rubecula) in a fragmented<br />
area of the southwest of France. However,<br />
our point is not to contribute to the knowledge of<br />
robin’s biology, but rather to propose a generalpurpose<br />
modeling tool. This is why we will just<br />
use well-known data on robin’s biology <strong>and</strong> an empirical<br />
study of the “static” response of the bird’s<br />
presence to the forest index. The missing parameters<br />
of the dynamic model (mainly dispersion) will<br />
be given “plausible” values or will be adjusted so<br />
that the long-term probabilities of presence of birds<br />
computed by our dynamic model will converge to<br />
the ones computed by an existing “static” response<br />
model.<br />
699
In other words, given that the only data that we have<br />
at the present time were collected in order to build a<br />
response of robin’s presence to an index called Forest<br />
Index (FI) Ladet [2000], we will consider that<br />
the output of this static model can be seen as the set<br />
of equilibrium (long-term) robin presence probabilities.<br />
So, when calibrating our dynamic model we<br />
will try to make them converge in the long run to<br />
the output of the static model. Of course, it should<br />
be precised that the interest of our model is not in its<br />
ability to mimic the output of the static model, but<br />
its ability to model short-term, out-of-equilibrium<br />
changes in the biological processes, in response to<br />
expected socio-economic changes! However, we<br />
have not sufficient data yet for assessing the quality<br />
of our tool with respect to observed changes in<br />
l<strong>and</strong>scape-use <strong>and</strong> presence probabilities. These are<br />
difficult to acquire since they need following individual<br />
for several years, on a significantly large<br />
area. However, we hope that our modeling tool we<br />
help to i) give some general ideas on the impact of<br />
socio-economic expected changes (forest fragmentation,<br />
agricultural l<strong>and</strong> desertion...) on birds presence<br />
<strong>and</strong> ii) help focusing costly data collection by<br />
underlying which parts of the biological model are<br />
important for assessing such an impact, by using<br />
sensitivity analysis on the range of plausible values<br />
of parameters, for example.<br />
We will present the static Forest Index response<br />
model in the following section. Then, we will point<br />
out some limitations of the static model, <strong>and</strong> present<br />
an improvement in Section 3, consisting in modeling<br />
the dynamics of robin through the use of a<br />
Markov chain on a multidimensional r<strong>and</strong>om variable,<br />
approximated by a set of pseudo independent<br />
mono-dimensional r<strong>and</strong>om variables. The obtained<br />
model is finally tuned <strong>and</strong> validated through comparisons<br />
with the static model <strong>and</strong> with on field<br />
measures (Sections 4 <strong>and</strong> 5).<br />
of French plain regions Joachim et al. [1997]. For<br />
each plot, experienced observers recorded all bird<br />
species contacted visually or by their vocal manifestation<br />
during 20 minute periods between sunrise <strong>and</strong><br />
up to 4 hours after sunrise. The bird census was performed<br />
during the month of May 1990 <strong>and</strong> included<br />
676 points scattered over the area. For the present<br />
study, we retained only the presence/absence information<br />
of robin. The SPOT satellite images cover<br />
a region of 60 x 60 km centered on N43 latitude<br />
<strong>and</strong> E1 longitude. The picture has been windowed<br />
on a study zone of approximately 2100 km2, with a<br />
20m resolution. As we know that Robin is strongly<br />
influenced by forest density <strong>and</strong> fragmentation, we<br />
classified the images with supervision to produce a<br />
binary map (forest/not forest).<br />
According to previous works in the region Lauga<br />
<strong>and</strong> Joachim [1992], Ladet [2000], we compute for<br />
each points of the map an index of forest influence<br />
(FI). The FI of a given point lies between 0 <strong>and</strong> 1,<br />
0 in an open area <strong>and</strong> 1 in a completely forested<br />
area. In order to compute the FI in a cell of coordinates<br />
¡£¢¥¤§¦©¨ , we take into account the presence or<br />
absence of forested cells within a radius around<br />
the cell. Furthermore, the influence of forested cells<br />
is smaller when cells are further away. So, cells<br />
are weighted according to their distance to the cell<br />
in which the FI is computed, the weight decreasing<br />
with the distance. In the case of robins, the value of<br />
the radius has been set to 100m Lauga <strong>and</strong> Joachim<br />
[1992]: cells further than 100m from the considered<br />
cell do not influence the FI. Let now be the binary<br />
matrix of forested <strong>and</strong> non forested cells (resolution<br />
of 20m). Let ¡¨¡ ¨<br />
where "!£#%$&¡£')(+*%¨-,¡ /. *0¨<br />
be the (decreasing)<br />
weight of distant cells.<br />
Then if a given cell has coordinates ¡1¢¥¤§¦©¨ ,<br />
2 STATIC MODEL OF PRESENCE / AB-<br />
SENCE OF ROBIN<br />
The study area lies between the Garonne <strong>and</strong> Gers<br />
rivers, in South-western France (lat.: N43 , long.:<br />
W1 ). It is a hilly region (200-400m a.s.l.), dissected<br />
by north-south valleys, within a sub-Atlantic<br />
climate with Mediterranean <strong>and</strong> mountain influences.<br />
The forests are fragmented <strong>and</strong> cover 15%<br />
of the area . Oaks Quercus robur <strong>and</strong> Q. sessiflora,<br />
often in association with chestnut Castanea sativa<br />
in coppice, cherry Prunus avium <strong>and</strong> wild service<br />
trees Sorbus torminalis are the main tree species in<br />
the area. The avian fauna is rather poor, <strong>and</strong> typical<br />
¡£¢-¤5¦©¨6 7 243 ¡£¢5K¤§¦KH¨ML ¡£¢5K¤§¦KNÖ(<br />
8:9+;£;:?5;H?1CGIJ-C<br />
The forest influence combines information on both<br />
forest patch size <strong>and</strong> isolation. These two variables<br />
have effects on birds occurrence Lescourret<br />
<strong>and</strong> Genard [1993], Villard et al. [1999]. A logistic<br />
regression linked the FI values <strong>and</strong> the presence/absence<br />
of Robin measured on the sampling<br />
points. The maximum of likelihood estimated the<br />
quality of the model. We cannot determine a priori<br />
the range of forest influence on Robin: despite<br />
it is a small species, we know that long-range influences<br />
can occur, because of a source effect of large<br />
forests for example Monteil et al. [2004]. In order<br />
to find the best model of occurrence we tested several<br />
radiuses for FI <strong>and</strong> produced a logistic model<br />
700
for each of them. Then, we chose the radius providing<br />
the model with the lowest maximum of likelihood<br />
(100m). We applied this model of occurrence<br />
on the whole studied area to predict robin’s distribution.<br />
The frequency of presence increased with<br />
the forest influence (FI), allowing using a simple logistic<br />
regression for modeling its response. Indeed<br />
robin is a bird of forest interior <strong>and</strong> forest edge in<br />
the area under study.<br />
3 SPATIALLY EXPLICIT STOCHASTIC<br />
DYNAMICS MODEL<br />
The problem with the robin’s presence model we<br />
have just presented is twofold: i) it is purely static<br />
<strong>and</strong> cannot take any colonization effect into account<br />
<strong>and</strong> can hardly model the short-term impact of clearcutting<br />
highly densely populated forest areas <strong>and</strong> ii)<br />
it is not really “spatial” insofar as, for example, two<br />
equally fragmented areas will have the same probability<br />
of presence of robin, even if they are located<br />
in different parts of the forest. In this Section we try<br />
to remedy these problems by coupling with the FI<br />
model a stochastic, individual based, spatially explicit<br />
population dynamics model in order to take<br />
into account the temporal <strong>and</strong> spatial dynamics of<br />
robin. Unfortunately, this model is not tractable,<br />
due to the high dimensionality of the r<strong>and</strong>om variables<br />
involved. So, in Section 4 we will use a Monte<br />
Carlo simulation method in order to approximate<br />
the probabilities of presence of robin across the area<br />
under study.<br />
3.1 Robin’s presence in the area<br />
The area under study is represented by an array of<br />
cells, each of which representing<br />
L¢¡<br />
a surface<br />
of . The cell surface is chosen with<br />
respect to the usual individual territory (1ha), since<br />
robin is a territorial bird <strong>and</strong> there are in general no<br />
more than one nest in each cell. So, 20m resolution<br />
* ' ' L * ' ' <br />
cells are grouped by 25 (5 5) <strong>and</strong> an average FI is<br />
computed for each group. £¥¤ of dimensions<br />
is the r<strong>and</strong>om variable representing the spatial configuration<br />
of L L¦¡<br />
¡1¢¥¤5¦¨<br />
robin’s nests over the whole territory.<br />
is a variable<br />
¤ stating whether<br />
<br />
a<br />
¡£¢-¤5¦©¨ *<br />
nest is present<br />
in cell . if<br />
¡1¢¥¤§¦©¨<br />
there is a ¤ nest in cell<br />
at time step ¤ ¡1¢¥¤§¦©¨ '<br />
(i.e. year ) § <strong>and</strong> if § ¡1¢¥¤§¦©¨ ¤ not. ¨ ')¤*©<br />
So the £ state space of is . We<br />
make the £ assumption<br />
¤<br />
that<br />
M¤(+(:(+¤£<br />
is<br />
D ¤ ')¤ *© <br />
an homogeneous<br />
¡<br />
Markov chain:<br />
D § ¤(+(:(+¤ ¤ ¨ <br />
,<br />
¨ <br />
¤ ¤ £ D£ ¤ ¤ £<br />
¡ £ ¤ D ¤ D£ ¤ ¤ ¨ <br />
£¤ So only depends £¤ A on , i.e. on the spatial<br />
configuration of the nests the previous year.<br />
In what follows, we approximate £ ¤ by a product<br />
of “independent” mono-dimensional r<strong>and</strong>om variables<br />
£<br />
¤ where £<br />
91
£<br />
male have only 40% of chance of generating a new<br />
nest. Taking these data into account we chose to<br />
fix the following probabilities on the number of<br />
;<br />
successor nests for each<br />
¡ ¢¤£¥£<br />
¡£'¨ ' ( ¦<br />
§<br />
nest:<br />
¡E*%¨ ')('©¦<br />
; § ¢¨£¥£<br />
¡ ¨ ')(+*<br />
; § ¢¤£¥£<br />
¡¨ <br />
')(+*<br />
¢¨£¥£<br />
; ¢¨£¥£<br />
¡ ¨ ')('©¦ §<br />
. These probabilities are arbitrary<br />
but can be adjusted by on site studies.<br />
Mortality. We first assumed a fixed mortality rate<br />
of 60%. The figures above give a fixed expectation<br />
of the number of successors for any given nest.<br />
<br />
')( ¦ L ' ')('©¦ L *<br />
')('©¦ L *<br />
Namely,<br />
. So, the expected number of nests<br />
may stay constant when we neglect any effect<br />
*<br />
of<br />
, the<br />
population would gradually decline to 0, while if<br />
the diffusion. If we had chosen ! #"<br />
(( (<br />
$&%<br />
*<br />
after a while the whole territory would<br />
<br />
be populated... These three possible evolutions do<br />
not fit the reality in which, when the l<strong>and</strong>scape is<br />
modified in one way or another, the nests population<br />
varies until it reaches a fixed point close to the<br />
FI response. This is why we chose to vary the mortality<br />
rate as a function of the forest index FI: Nests<br />
that are located in open areas are more subject to<br />
predation (since they are more visible) than nests<br />
located in completely forested areas. So, we define<br />
the mortality '<br />
¡1¢¥¤§¦©¨<br />
rate for the cell of<br />
¡1¢¥¤§¦©¨<br />
coordinates<br />
as:<br />
¡1¢¥¤5¦¨6 )(+*¤,-<br />
D<br />
,¥.0/ 132<br />
8:91?<br />
*4<br />
* ,-<br />
D<br />
,5./ 132<br />
8:9£?<br />
76<br />
'<br />
(1)<br />
This is the definition of a logistic response of mortality<br />
to forest index. , 6 , 6 <strong>and</strong> 6 are parameters<br />
which will be tuned in Section 5<br />
in cell ¡1¢ ¤5¦ 0¨ is:<br />
9?> ¡1¢¥¤§¦ ¤>¢ ¤§¦ %¨ A@B(DC ¡1¢¥¤§¦ ¤>¢ ¤§¦ %¨¡(<br />
¡1¢¥¤§¦©¨<br />
=<br />
¡1¢¥¤5¦¨ * -<br />
D £<br />
. / 1&2<br />
8+91?<br />
*4<br />
* £<br />
-<br />
D £<br />
./ 132<br />
8:9£?<br />
<br />
where<br />
where @ is a normalization factor, <br />
<br />
, <strong>and</strong> 8 are<br />
parameters <strong>and</strong> C ¡1¢¥¤§¦ ¤E¢ ¤5¦ %¨ is the 2D Gaussian of<br />
parameters ¡1¢ . ¢ M¤5¦ . ¦ ¨<br />
computed above. Finally,<br />
in order to take the presence of other nests into account<br />
we modified further the probability of diffusion<br />
by making it 0 in cells where a nest already<br />
exists.<br />
8 (2)<br />
So, we now have a stochastic, spatially explicit<br />
model of robin’s nests dynamics. With this model<br />
we could imagine to h<strong>and</strong>le the -dimensional<br />
r<strong>and</strong>om variable £ ¤ representing the possible spatial<br />
configurations of nests in the territory. However,<br />
this is impossible in practice, due to the high<br />
L ¡ £ ¤ dimensionality of . This is why we introduce in<br />
the following section an £ ¤ approximation of by a<br />
set of “independent” probabilities presence£ of<br />
which evolution over time will be computed through<br />
simulation Sabbadin [2003].<br />
91
£<br />
¢<br />
*<br />
¢<br />
M ¤ D ¡1¢¥¤5¦¨>¨ 9?F0GHGHG
1<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1<br />
Figure 3: Limit presence probabilities as functions<br />
of FI .<br />
<br />
for ( ¤ £)*<br />
, , . ')(H<br />
, ¨ .<br />
<br />
, ¨ ¦<br />
<strong>and</strong> 8<br />
' (' '§¦<br />
<br />
')( £©<br />
6 CONCLUDING REMARKS<br />
, 6 <br />
In this paper we have provided a model which can<br />
be used for the study of spatio-temporal dynamics<br />
of biological processes in spatial, possibly dynamic<br />
l<strong>and</strong>scapes. This model explicitly h<strong>and</strong>les spatial<br />
features, as well as stochasticity of events. The main<br />
originality of the model is to use a crude approximation<br />
of a multidimensional r<strong>and</strong>om variable, which<br />
evolution (otherwise impossible to model) over time<br />
is estimated through Monte Carlo Simulation. It has<br />
been illustrated on the example of robin in a large<br />
valley of the southwest of France.<br />
This work is only preliminary <strong>and</strong> we are considering<br />
the following extensions:<br />
i) First, the use of the model on the example of<br />
robins need further experiment to be validated. ii)<br />
Other birds species with different habitat requirements<br />
have been studied with respect to their relation<br />
to FI (44 species) in Ladet [2000]. The dynamics<br />
model we propose here should be adapted to<br />
these species as well, in order to have a clear view<br />
of the biodiversity evolution in the area.<br />
iii) In the dynamic model, only animal dynamics is<br />
considered, not l<strong>and</strong>scape evolution. However, the<br />
forest cover of the area under study has known deep<br />
changes in the past <strong>and</strong> should encounter even more<br />
changes in the near future. It is clear that our objective<br />
is to measure the impact of such changes on<br />
biodiversity. This could be easily studied within our<br />
framework, by considering l<strong>and</strong>scape dynamics in<br />
addition to animal dynamics. In this way, the model<br />
could help decision makers in assessing the impact<br />
of large scale decisions (deforestation, l<strong>and</strong> consolidation)<br />
on biodiversity.<br />
REFERENCES<br />
M. Conroy <strong>and</strong> B. Noon. Mapping of species richness<br />
for conservation of biological diversity: conceptual<br />
<strong>and</strong> methodological issues. Ecological<br />
Applications, 6:763–773, 1996.<br />
M. Deconchat <strong>and</strong> G. Balent. Vegetation <strong>and</strong><br />
bird community dynamics in fragmented coppice<br />
forests. Forestry, 74:105–118, 2001.<br />
M. Huston. Biological diversity. The coexistence of<br />
species on changing l<strong>and</strong>scapes. Cambridge University<br />
Press, Cambridge, UK, 1994.<br />
P. Isenmann. Le rougegorge. Belin-Eveil Nature,<br />
Paris, 2003.<br />
J. Joachim, J. Bousquet, <strong>and</strong> C. Faure. Atlas des<br />
oiseaux nicheurs de Midi-Pyrénées. Années 1985<br />
à 1989. AROMP, Toulouse, France, 1997.<br />
S. Ladet. Modélisation de la réponse des espèces<br />
d’oiseaux à la frgmentation forestière dans les<br />
coteaux du sud-ouest. Master’s thesis, Université<br />
Paul Sabatier. DEA Ecologie des systèmes continentaux,<br />
2000.<br />
J. Lauga <strong>and</strong> J. Joachim. <strong>Modelling</strong> the effects of<br />
forest fragmentation on certain species of forestbreeding<br />
birds. L<strong>and</strong>scape Ecology, 6:183–193,<br />
1992.<br />
F. Lescourret <strong>and</strong> M. Genard. Habitat, l<strong>and</strong>scape<br />
<strong>and</strong> birds composition in mountain forest fragments.<br />
Journal of <strong>Environmental</strong> Management,<br />
40:317–328, 1993.<br />
C. Monteil, M. Deconchat, <strong>and</strong> G. Balent. Interest<br />
of neural network for l<strong>and</strong>scape ecology- example<br />
of bird species richness models in fragmented<br />
forests. L<strong>and</strong>scape Ecology, 2004.<br />
R. Sabbadin. Approximating spatial markov decision<br />
processes for environmental management.<br />
In 15th <strong>International</strong> Congress on <strong>Modelling</strong> <strong>and</strong><br />
Simulation (MODSIM’03), volume vol. 4, pages<br />
1868–1873, Townsville, Australia, July 2003.<br />
M. Villard, M. Trzcinski, <strong>and</strong> G. Merriam. Fragmentation<br />
effects on forest birds: relative influence<br />
of woodl<strong>and</strong> cover <strong>and</strong> configuration on<br />
l<strong>and</strong>scape occupancy. Conservation Biology, 13:<br />
774–783, 1999.<br />
B. Williams. ssessment of accuracy in the mapping<br />
of vertebrate biodiversity. Journal of <strong>Environmental</strong><br />
Management, 47:269–282, 1996.<br />
704
L<strong>and</strong> Use <strong>and</strong> Hydrological Management:<br />
ICHAM, an Integrated Model at a Regional Scale<br />
in Northeastern Thail<strong>and</strong><br />
N.Hall a , R. Lertsirivorakul b , R. Greiner c , S. Yongvanit d , A. Yuvaniyama e , R. Last f <strong>and</strong><br />
W. Milne-Home f,g<br />
a Integrated Catchment Assessment <strong>and</strong> Management (iCAM), The Australian National University,<br />
Canberra ACT 0200, Australia, nhall@effect.net.au<br />
b Faculty of Technology, Khon Kaen University, Khon Kaen 40002, Thail<strong>and</strong><br />
c CSIRO Sustainable Ecosystems, Townsville, Qld 4810, Australia<br />
d Faculty of Humanities <strong>and</strong> Social Science, Khon Kaen University, Khon Kaen 40002, Thail<strong>and</strong><br />
e L<strong>and</strong> Development Department, Ministry of Agriculture & Cooperatives, Bangkok 10900, Thail<strong>and</strong><br />
f National Centre for Groundwater Management, University of Technology Sydney, NSW 2007, Australia<br />
g Institute for Water <strong>and</strong> <strong>Environmental</strong> Resource Management. University of Technology Sydney, NSW 2007,<br />
Australia,<br />
Abstract: Soil salinity is a major problem in Northeastern Thail<strong>and</strong> as a result of the interaction of<br />
groundwater flow systems with widespread deposits of rock salt. Successful salinity management would<br />
involve changing l<strong>and</strong> use <strong>and</strong> water balances at a regional scale with a time scale of 30 to 50 years. The<br />
scientific issue requires multidisciplinary cooperation including hydrologists, hydrogeologists, agronomists<br />
<strong>and</strong> economic <strong>and</strong> social researchers. A major issue is the real complexity of the quantitative relationships<br />
driving salinity under different environments <strong>and</strong> the uncertainty resulting from data limitations. This<br />
requires that modelling frameworks be open <strong>and</strong> accessible to a range of disciplines as well as allowing<br />
flexibility in coefficient values. This paper reports on interdisciplinary research in progress on salinity <strong>and</strong><br />
l<strong>and</strong> use in Northeastern Thail<strong>and</strong> using a combination of bio-economic modelling to assess the socioeconomic<br />
impacts of changing l<strong>and</strong> uses, including the use of agroforestry to manage salinity, <strong>and</strong><br />
groundwater modelling. Models that were derived originally to support the investigation of salinity in the<br />
Liverpool Plains of New South Wales, Australia have been redeveloped for application to Northeastern<br />
Thail<strong>and</strong>. The earlier models used the GAMS language but the current modelling is being developed in<br />
EXCEL <strong>and</strong> MODFLOW for ease of use. Preliminary results of the modelling indicate that the saline l<strong>and</strong><br />
area will increase under a “do nothing” scenario, from the present 13% of l<strong>and</strong> area to 24% in 30 years. The<br />
optimal l<strong>and</strong> use would include more rice cultivation <strong>and</strong> plantation forestry, with less cassava growing<br />
compared to present l<strong>and</strong> use.<br />
Keywords:<br />
Integrated modelling; Agroforestry; Salinity; Integrated Catchment Management.<br />
1. INTRODUCTION<br />
1.1 Salinity issues<br />
Ghassemi, Jakeman <strong>and</strong> Nix [1995] showed that<br />
soil salinisation is a l<strong>and</strong> degradation process<br />
which is a major problem in Northeastern<br />
Thail<strong>and</strong> as a result of the interaction of<br />
groundwater flow systems with widespread<br />
deposits of rock salt. It has been estimated that an<br />
area of 6 million hectares, is already affected by<br />
salt <strong>and</strong> that the problem is becoming more<br />
widespread. The salinity problem in North-east<br />
Thail<strong>and</strong> is of national importance because salt<br />
affected l<strong>and</strong> reduces crop <strong>and</strong> forage yields.<br />
Arunin [1987] stated that the reason for the<br />
spread of salinisation is primarily the removal of<br />
forest cover leading to increased groundwater<br />
recharge. This factor has been exacerbated by<br />
anthropogenic activities including dam<br />
construction, low technology salt extraction <strong>and</strong><br />
irrigation. The source of the salt is primarily the<br />
705
dissolution of rock salt in the Mahasarakham<br />
Formation, which underlies most of the Khorat<br />
Plateau in North-east Thail<strong>and</strong> <strong>and</strong> parts of the<br />
Lao PDR. Groundwater recharge on deforested<br />
upl<strong>and</strong>s allows deep groundwater flow systems to<br />
dissolve <strong>and</strong> transport the salt towards lowl<strong>and</strong><br />
discharge areas. Another source of salt is the<br />
shallow interflow in the regolith forming local<br />
flow systems.<br />
1.2 Salinity management<br />
Successful salinity management would involve<br />
changing l<strong>and</strong> use <strong>and</strong> water balances at a<br />
regional scale over 30 to 50 years as discussed by<br />
Pannell [2001]. Salinity management requires<br />
multidisciplinary cooperation including<br />
hydrologists, hydrogeologists, agronomists <strong>and</strong><br />
economic <strong>and</strong> social researchers.<br />
In Thail<strong>and</strong> the monsoon rain fills the soil profile<br />
in the wet season so that there is a layer of fresh<br />
water resting on saline groundwater <strong>and</strong> in some<br />
areas pressing it down. The plant roots live in the<br />
fresh water <strong>and</strong> use it up over the dry season. At<br />
the end of the dry season there will be little fresh<br />
water in the profile <strong>and</strong> rivers may well carry<br />
salty water flowing from the groundwater layer.<br />
The next monsoon rains wash the salt out of the<br />
rivers <strong>and</strong> presses the salt groundwater back into<br />
the profile allowing a further season of growth.<br />
This process is described by Lertsirivorakul <strong>and</strong><br />
Milne-Home [2000].<br />
Despite the differences, between Southeast<br />
Australia <strong>and</strong> Thail<strong>and</strong> there are similarities in the<br />
effects of l<strong>and</strong> use change. In both countries<br />
salinity has become a problem after deforestation<br />
<strong>and</strong> increased cropping. Limited reafforestation<br />
has been proposed as a management approach.<br />
This paper discusses modelling of salinity<br />
management through vegetation change. The<br />
research is helping the development of Thai tree<br />
planting policy <strong>and</strong> is linked with the L<strong>and</strong><br />
Development Department’s program of<br />
reforestation of recharge areas of the Northeast.<br />
1.3 Study regions<br />
Two regions have been the subject of modelling<br />
studies in northeast Thail<strong>and</strong> for this project. A<br />
catchment in Kalasin near Khon Kaen <strong>and</strong> one<br />
near Khorat. Both are intensively cropped with<br />
rice <strong>and</strong> cassava. Forest <strong>and</strong> treed areas are<br />
limited. Pigs, poultry, buffalo <strong>and</strong> cattle are kept.<br />
Each region approximates a catchment <strong>and</strong> is<br />
quite large (1245 sq km in Kalasin <strong>and</strong> 553 in<br />
Khorat). The regions are divided into subregions<br />
that are broadly homogeneous in terms of soils,<br />
hydrology <strong>and</strong> topography.<br />
This paper concentrates on the Khorat model. The<br />
study region covers 553 square kilometres. The<br />
l<strong>and</strong> slopes from east to west <strong>and</strong> is divided into<br />
three subregions. The Khorat region is divided,<br />
for this modelling exercise, into three subregions;<br />
the upper, middle <strong>and</strong> lower catchment.<br />
Subregion 1 is upl<strong>and</strong> along the east side of the<br />
modelling area with extensive forest cover. This<br />
is a recharge area. Subregion 2, in the middle of<br />
the catchment, is predominantly rice-growing<br />
country while subregion 3 grows rice with<br />
cassava <strong>and</strong> other crops.<br />
2. MODELLING ISSUES<br />
2.1 Context<br />
Quantitative analytical models are important<br />
means of testing hypotheses in relation to<br />
salinisation <strong>and</strong> its management but the<br />
complexity poses a series of methodological<br />
challenges.<br />
Greiner <strong>and</strong> Parton [1995] describe soil<br />
salinisation <strong>and</strong> its management as a complex<br />
systems problem that is characterised by issues of<br />
geographical <strong>and</strong> temporal scale in relation to<br />
process description as well as multiple nonlinearities<br />
<strong>and</strong> interdisciplinarity. This complexity<br />
poses challenges for the quantification of the<br />
system, in terms of capturing relevant factors,<br />
formalising systems relationships, accounting for<br />
risk <strong>and</strong> developing the data necessary to support<br />
the model. These challenges apply across both the<br />
socio-economic <strong>and</strong> hydro-geological domains of<br />
the system.<br />
2.2 Normative modelling approach <strong>and</strong><br />
operating system<br />
Greiner [1997] developed a bio-economic model<br />
(SMAC) for the Liverpool Plains catchment in<br />
Australia, which succeeded in solving the<br />
conceptual <strong>and</strong> methodological challenges outline<br />
above. The model was based on the approach of<br />
Baumol [1977] using regional, or catchment<br />
level, optimisation that applies spatial equilibrium<br />
modelling theory. However, the model required a<br />
large amount of data, based on a vast body of<br />
other research that had been completed for that<br />
catchment.<br />
Applying the approach to northeast Thail<strong>and</strong><br />
meant not just re-parameterising the model. The<br />
different agronomy <strong>and</strong> hydrology <strong>and</strong> the<br />
706
comparative shortage of data <strong>and</strong> of pre-existing<br />
catchment research required a different<br />
description of catchment relationships.<br />
The SMAC model was implemented in the<br />
GAMS language. GAMS is an excellent <strong>and</strong><br />
flexible modelling language but requires skilled<br />
users <strong>and</strong> is not widely accessible in Thail<strong>and</strong>.<br />
EXCEL, in contrast, has a straightforward <strong>and</strong><br />
comprehensible structure that is self-documenting<br />
<strong>and</strong> has a reasonably powerful solver for<br />
optimisation.<br />
It has been found that the spreadsheet approach is<br />
a useful communication tool in taking the model<br />
<strong>and</strong> its results to a wider audience including<br />
administrators. An example of this is Last, Hall,<br />
Anuluxtipun, Lertsirivorakul, Yongvanit, Milne-<br />
Home <strong>and</strong> Luangjame [2003].<br />
2.3 Hydrological <strong>and</strong> socio-economic data<br />
The hydrological data used in the bio-economic<br />
model is derived from on estimation of<br />
underground water movements beneath the<br />
modelled using a MODFLOW hydrological<br />
model run on an annual time basis. MODFLOW<br />
is a widely used model system for quantifying<br />
groundwater movements described by Harbaugh<br />
[1992]. MODFLOW models have already been<br />
applied in northeast Thail<strong>and</strong> by Lertsirivorakul<br />
<strong>and</strong> Milne-Home [2000].<br />
The MODFLOW model is not directly<br />
incorporated into the bio-economic model.<br />
Instead it is assumed that marginal changes in<br />
water movements, such as are likely to be<br />
produced by moderate l<strong>and</strong> use change, will<br />
change flows between sub-areas in the model in<br />
proportion to the change in inflows to the<br />
subregion. This is an approximation; if resources<br />
had been available it would have been better to<br />
simulate a set of scenarios of changing water<br />
accessions by sub-area <strong>and</strong> derive response<br />
functions that could have been used within the<br />
bio-economic model.<br />
The socio-economic modelling is based on farm<br />
management data collected as part of the wider<br />
project. Data was collected at a village level on<br />
crops, prices, farming costs <strong>and</strong> yields <strong>and</strong> on<br />
salinity <strong>and</strong> its effects. Data on family incomes<br />
was also collected including debt <strong>and</strong> off-farm<br />
incomes.<br />
Both formal <strong>and</strong> informal methods of collection<br />
were used <strong>and</strong> a variety of informants were<br />
contacted in each village. Village data was<br />
collated <strong>and</strong> checked then averaged to produce<br />
survey estimates at the appropriate level of<br />
aggregation for the models.<br />
2.4 Decision variables<br />
L<strong>and</strong> use activities are the major decision<br />
variables that influence the hydrological balance<br />
because different l<strong>and</strong> uses have different levels<br />
of recharge. As different l<strong>and</strong> use options use<br />
different amounts of available water, their impact<br />
on accessions to the groundwater system <strong>and</strong> on<br />
soil salinisation will differ. In turn, emerging<br />
salinity affects soil productivity <strong>and</strong> the l<strong>and</strong> use<br />
options that are potentially available to farmers.<br />
Yields of crops are reduced on saline l<strong>and</strong>.<br />
Representative farms embody the characteristics<br />
of each area. Each farm has l<strong>and</strong> use options<br />
associated with yields, recharge <strong>and</strong> runoff, <strong>and</strong><br />
flows of ground <strong>and</strong> surface water.<br />
The long-term catchment-scale optimisation<br />
ensures that externalities, that is costs <strong>and</strong><br />
benefits that arise from l<strong>and</strong>-use in one part of the<br />
catchment can be tracked through time <strong>and</strong> across<br />
the l<strong>and</strong>scape. Scenario analysis makes it possible<br />
to draw out temporal <strong>and</strong> spatial trade-offs of l<strong>and</strong><br />
management <strong>and</strong> l<strong>and</strong>-use change.<br />
2.5 Model implementation<br />
The agronomic <strong>and</strong> economic data are brought<br />
together with the outcomes of the hydrological<br />
modelling in the ICHAM bio-economic model<br />
(Isaan Catchment Hydrogeological <strong>and</strong><br />
707
YIELD<br />
PRICE<br />
RECEIPTS<br />
AREA<br />
CASH<br />
COST<br />
SUMMARY<br />
WATERUSE<br />
SALT<br />
YIELD<br />
COST<br />
MODFLOW<br />
DATA<br />
WATER<br />
MOVEMENT<br />
FARM<br />
INCOME<br />
SUMMARY<br />
MARGIN<br />
SETTINGS<br />
SALTING<br />
Figure 1. ICHAM flowchart<br />
Agricultural Model). ICHAM consists of a series<br />
of interconnected worksheets, which cover<br />
different aspects of the salinity management<br />
system.<br />
The operation of the model is described in the<br />
users manual Hall [2004]. The model is operated<br />
from the ‘Summary’ sheet that shows summaries<br />
of l<strong>and</strong> use, incomes <strong>and</strong> salinity over 30 years.<br />
All the other worksheets depend on Summary,<br />
Settings, which holds the data, <strong>and</strong> MODFLOW,<br />
which bring in the hydrology.<br />
The linkage from Salting, which estimates the<br />
area salted each year to Area allows the increase<br />
in salinisation in a year to affect future costs <strong>and</strong><br />
cropping <strong>and</strong> so the net present value of the<br />
farming operations.<br />
3. RESULTS<br />
3.1 Salinity <strong>and</strong> economic outcomes<br />
The results presented are for the Khorat<br />
catchment model. The analysis examines the<br />
differences between the base solution (triangles)<br />
for which current l<strong>and</strong> use is projected to continue<br />
for the next 30 years <strong>and</strong> an optimal solution<br />
(squares), which maximises farm profits at a<br />
catchment level. The left h<strong>and</strong> axis measures the<br />
percent of the catchment, which is saline, while<br />
the bottom axis shows years. The optimal solution<br />
takes an objective of maximising the net present<br />
value of farm incomes, taking into account the<br />
costs of salinity.<br />
Table 1 shows the effect of discount rate on the<br />
optimal solution. Two discount rates were used:<br />
four percent as an indicator of social time<br />
preference, <strong>and</strong> ten percent as an indicator of a<br />
commercial rate of time preference, such as that<br />
of a poor <strong>and</strong> indebted farmer<br />
Table 1. Percentage of area saline after 30 years<br />
%<br />
Current level of salinity 13.3<br />
Base solution after 30 years 21.3<br />
Optimal solution after 30 years<br />
- at 4% discount rate 15.3<br />
- at 10% discount rate 15.7<br />
The current saline area is 13 per cent of the<br />
catchment. The base solution results show that if<br />
current l<strong>and</strong> use continued for the next 30 years<br />
then salinity would be expected to increase from<br />
the present 13 per cent to 21 per cent of the<br />
catchment. However, if the catchment were<br />
managed for maximum profit, taking account of<br />
salinity costs (optimal solution), then the area<br />
saline would be only 15 per cent at a four percent<br />
discount rate. Taking account of future salinity<br />
costs would lead to less salinity than would occur<br />
with current l<strong>and</strong> use practice.<br />
Using the higher commercial discount rate of ten<br />
percent, rather than the social discount rate of<br />
four per cent, would lead to almost 16 per cent<br />
salinisation after 30 years, a half per cent more<br />
708
¢¡ ¢¢<br />
¡¢£¤¥<br />
¨©<br />
©<br />
30 year planning period<br />
¢¦ §¡ ¢¡ ¦ ¡ ¦ ¡<br />
than at the social rate. Hence, heavily indebted<br />
poor farmers may rationally decide to allow more<br />
salinisation than society considers desirable.<br />
trees while the acacias would be planted along the<br />
bunds around rice paddies so that their deep roots<br />
dry out the profile in the dry season. Experimental<br />
plantings of eucalypts, other trees <strong>and</strong> acacias are<br />
Figure 2. Area saline: base <strong>and</strong> optimal solutions<br />
currently underway in the catchment.<br />
If current l<strong>and</strong> use continues unchanged, the net<br />
present value of the cost of salinity over the next<br />
30 years is estimated to be 1026 million Baht,<br />
twelve per cent of net present value of farm<br />
incomes (at a four percent discount rate). The<br />
optimal solution at the same discount rate has a<br />
net present value of salinity cost of 756 million<br />
Baht, nine per cent of net present value of farm<br />
incomes. It is significant that the optimal solution<br />
does not eliminate all saline areas.<br />
3.2 L<strong>and</strong> use change<br />
Table 2 shows the l<strong>and</strong> use for the base <strong>and</strong><br />
optimal solutions of ICHAM for Khorat expresses<br />
in rai. One rai is equivalent to a 40-metre square<br />
so that there are 6.25 rai to a hectare. The optimal<br />
solution has more rice, other crops, fruit trees,<br />
other trees, <strong>and</strong> acacia <strong>and</strong> less cassava.<br />
The changes are not large as a proportion of the<br />
whole catchment, suggesting that stabilising<br />
salinity by changing l<strong>and</strong> use is feasible. The trees<br />
envisaged are plantation eucalypts <strong>and</strong> native<br />
Larger plantings of trees <strong>and</strong> acacias at the<br />
expense of food <strong>and</strong> cash crops would be needed<br />
to reduce the current levels of salinisation. These<br />
might have major consequences for the economic<br />
<strong>and</strong> social wellbeing of people farming in the<br />
catchment.<br />
Table 2. L<strong>and</strong> use in Khorat model: base <strong>and</strong><br />
optimal solutions at 4% discount rate<br />
Base Optimal<br />
Rai<br />
Rai<br />
Rice 210336 223567<br />
Cassava 111217 88549<br />
Other crops 9447 10809<br />
Fruit trees 3113 4381<br />
Other trees 36 6843<br />
Acacia 0 6732<br />
Forest 2454 2454<br />
Total 336603 336603<br />
709
4. DISCUSSION AND CONCLUSIONS<br />
ICHAM is a simplified representation of a very<br />
complex <strong>and</strong> partly unknown reality. It brings<br />
together our current knowledge of hydrogeology,<br />
agronomy <strong>and</strong> farm economics but further ground<br />
studies are needed before making confident<br />
recommendations about changing l<strong>and</strong> use in<br />
particular areas.<br />
The simulation results presented show that for a<br />
catchment in northeast Thail<strong>and</strong>, salinity will<br />
almost double in the next 30 years if current l<strong>and</strong><br />
use is maintained. The optimisation results show<br />
that it would be economically rational to<br />
implement l<strong>and</strong> use changes to reduce the rate of<br />
salinisation.<br />
The lower levels of salinity under optimisation<br />
show that there is market failure in the<br />
management of salinity on l<strong>and</strong> in northeast<br />
Thail<strong>and</strong>. This is not unusual in management of<br />
salinity because changes in water balances are<br />
affected by l<strong>and</strong> use in the whole catchment,<br />
while salinity normally affects only some areas.<br />
Hence there is no direct incentive for all farmers<br />
to change l<strong>and</strong> use for the benefit of farmers in<br />
affected areas [Greiner <strong>and</strong> Cacho [2001]. Also,<br />
new crops such as trees are different to field crops<br />
<strong>and</strong> often need a significant wait for income. This<br />
may not be an option for farmers who rely on the<br />
rice crop to feed their families.<br />
ICHAM shows that salinisation can be managed,<br />
the general direction to be taken <strong>and</strong> the<br />
approximate magnitude of l<strong>and</strong> use change<br />
needed. Changing l<strong>and</strong> use, without affecting the<br />
livelihood of farmers, requires social interactions<br />
between governments, at all levels, farmers,<br />
extension services <strong>and</strong> technical experts. This is a<br />
formidable undertaking but ICHAM shows that<br />
the cost of doing nothing will be a big increase in<br />
the area salinised <strong>and</strong> increasing losses caused by<br />
salinisation.<br />
ICHAM has been developed through cooperative<br />
effort in Australia <strong>and</strong> Thail<strong>and</strong> between agencies<br />
<strong>and</strong> disciplines. The modelling approach has been<br />
successfully applied <strong>and</strong> a training program is<br />
under development. An extension of the<br />
modelling into the Lao People’s Republic has also<br />
been discussed with Lao government agencies.<br />
5. REFERENCES<br />
Management of Saline/Alkaline Soils,<br />
Bangkok: FAO Regional Office for Asia<br />
<strong>and</strong> the Pacific, FAO, 1987.<br />
Baumol, WJ, Economic Theory <strong>and</strong> Operations<br />
Analysis, Prentice Hall, Englewood<br />
Cliffs, 1977.<br />
Ghassemi, F, A Jakeman <strong>and</strong> H Nix, Salinisation<br />
of L<strong>and</strong> <strong>and</strong> Water Resources. Human<br />
causes, extent, management <strong>and</strong> case<br />
studies, CAB <strong>International</strong>, Oxford,<br />
1995.<br />
Greiner, R, Integrated catchment management for<br />
dryl<strong>and</strong> salinity control in the Liverpool<br />
Plains catchment, L<strong>and</strong> <strong>and</strong> Water<br />
Resources Research <strong>and</strong> Development<br />
Corporation, Canberra, 1997.<br />
Greiner, R <strong>and</strong> O Cacho, On the efficient use of a<br />
catchment’s l<strong>and</strong> <strong>and</strong> water resources:<br />
whether <strong>and</strong> how to control dryl<strong>and</strong><br />
salinisation, Ecological Economics 38(3)<br />
441-58, 2001.<br />
Greiner, R <strong>and</strong> K Parton, Analysing dryl<strong>and</strong><br />
salinity management on a catchment<br />
scale with an economic ecological<br />
modelling approach, Ecological<br />
Engineering 4 191-8, 1995.<br />
Hall, N, Isaan Catchment Hydrological <strong>and</strong><br />
Agronomic Model ICHAM Users<br />
Manual, UTS, Sydney, 2004.<br />
Harbaugh, AW, A generalized finite-difference<br />
formulation for the U.S.Geological<br />
Survey modular three-dimensional finitedifference<br />
ground-water flow model,<br />
Open-File Report 91-494,<br />
U.S.Geological Survey, 1992.<br />
Last, R, N Hall, Y Anuluxtipun, R<br />
Lertsirivorakul, S Yongvanit, W Milne-<br />
Home <strong>and</strong> J Luangjame, Bio-economic<br />
modelling towards optimising l<strong>and</strong> use<br />
<strong>and</strong> salinity management in South-<br />
Eastern NSW <strong>and</strong> North-Eastern<br />
Thail<strong>and</strong>. 9th National Conference on<br />
the Productive Use <strong>and</strong> Rehabilitation of<br />
Saline L<strong>and</strong>s (PUR$L), Yeppoon, 2003.<br />
Lertsirivorakul, R <strong>and</strong> W Milne-Home, <strong>Modelling</strong><br />
the upward flux of saline groundwater at<br />
Si Than Pond, Khon Kaen University,<br />
Khon Kaen, N.E.Thail<strong>and</strong>. Proceedings<br />
4th Internat. Conf. on Diffuse Pollution,<br />
Bangkok, Thail<strong>and</strong>, 2000.<br />
Pannell, DJ, Dryl<strong>and</strong> salinity: economic<br />
scientific, social <strong>and</strong> policy dimensions,<br />
Australian Journal of Agricultural <strong>and</strong><br />
Resource Economics 45(4) 517-46,<br />
2001.<br />
Arunin, S, Management of saline <strong>and</strong> alkaline<br />
soils in Thail<strong>and</strong>. Paper presented at the<br />
Regional Expert Consultation on the<br />
710
Forecasting Municipal Solid Waste Generation in Major<br />
European Cities<br />
P. Beigl, G. Wassermann, F. Schneider <strong>and</strong> S. Salhofer<br />
Institute of Waste Management, BOKU – University of Natural Resources <strong>and</strong> Applied Life Sciences,<br />
Vienna, Austria<br />
Abstract: An underst<strong>and</strong>ing of the relationships between the quantity <strong>and</strong> quality of environmentally relevant<br />
outputs from human processes <strong>and</strong> regional characteristics is a prerequisite for planning <strong>and</strong> implementing<br />
ecologically sustainable strategies. Apart from process-related parameters, continuous <strong>and</strong> discontinuous<br />
socio-economic long-term trends often play a key role in the assessment of environmental impacts. This paper<br />
describes the development of a prognosis model for municipal solid waste (MSW) generation in European<br />
regions. The objective is to assess future municipal waste streams in major European cities. We therefore<br />
focussed on cities, which face significant social <strong>and</strong> economic changes, e.g. in central <strong>and</strong> east European<br />
(CEE) countries. The investigations covered waste-related data <strong>and</strong> a broad set of potential influencing<br />
parameters that contained commonly used social, economic <strong>and</strong> demographic indicators as well as previously<br />
proved waste generation factors. An extensive database was created with an annual time series up to 32 years<br />
from 55 European cities <strong>and</strong> 32 countries. The evaluation of this historic time series <strong>and</strong> the cross-sectional<br />
data by means of multivariate statistical methods has unveiled significant relationships between the status of<br />
regional development <strong>and</strong> municipal solid waste generation. We identified a core set of significant indicators,<br />
which can describe a long-term development path that predetermines the level of waste generation. These<br />
findings concerning this analogy have been integrated in an econometric model for European cities.<br />
Keywords: Waste management; Municipal solid waste; Waste generation; <strong>Modelling</strong>; Forecasting<br />
1. INTRODUCTION<br />
The development of waste management models<br />
over the last decades can be characterised by an<br />
increasing level of integration of related processes<br />
with consideration of environmental, economic <strong>and</strong><br />
social aspects. The genesis of these decision<br />
support tools [Björklund, 2000] is reflected by<br />
extending system boundaries, which are shown in<br />
Figure 1. In early models, attention was paid to the<br />
problems in subsystems, e.g. routing of vehicles<br />
<strong>and</strong> location of treatment <strong>and</strong> disposal facilities,<br />
with a focus on only a few criteria (e.g. costs).<br />
Recently, waste management models have started<br />
to evaluate entire waste management systems,<br />
considering broad sets of quantitative <strong>and</strong><br />
qualitative criteria.<br />
Up to now, most of these decision support tools for<br />
waste management planning use the amount of<br />
waste generation as the given input parameter<br />
[Björklund, 2000; White et al., 1999]. Thus the<br />
impacts of demographic, social <strong>and</strong> economic<br />
dynamics as well as other factors (e.g. consumption<br />
patterns or waste prevention) are not taken into<br />
consideration for the accurate assessment of future<br />
waste generation.<br />
Products<br />
Energy<br />
Costs<br />
Socioeconomic<br />
changes<br />
Demographic<br />
dynamics<br />
Waste management<br />
system<br />
Subsystem<br />
1<br />
Municipal solid<br />
waste generation<br />
Subsystem<br />
2<br />
Consumption<br />
patterns<br />
. . . .<br />
Subsystem<br />
x<br />
Waste<br />
prevention<br />
Products<br />
Energy<br />
Emissions<br />
Figure 1. System boundaries in waste management<br />
models with different level of integration.<br />
This paper aims to identify parameters which help<br />
to explain the present situation <strong>and</strong> to assess the<br />
future amount of municipal solid waste (MSW)<br />
generated per capita in different European cities.<br />
This study is part of the European Commission<br />
project “The Use of Life Cycle Assessment Tools<br />
for the Development of Integrated Waste<br />
Increasing model integration<br />
711
Management Strategies for Cities <strong>and</strong> Regions with<br />
Rapidly Growing Economies.” The goal of this<br />
project is to develop a highly integrated decision<br />
support tool for cities in southern, central <strong>and</strong> east<br />
European countries which helps to evaluate waste<br />
management options with regard to environmental,<br />
economic <strong>and</strong> social criteria. In this context, MSW<br />
forecasts should arouse the consciousness of the<br />
municipal decision makers to implement<br />
ecologically sustainable measures (e.g. increasing<br />
recycling quotas).<br />
2. METHODOLOGICAL<br />
CONSIDERATIONS<br />
Due to the high heterogeneity of municipal solid<br />
waste streams <strong>and</strong> the diversity of their ways<br />
through the economy, the identification of<br />
parameters is a highly complex problem. An<br />
overview of studies in this scientific field by Beigl<br />
et al. [2003] describes previous approaches, which<br />
can be classified by the type of model:<br />
• Input-output models: Here the input of the<br />
waste generator is assessed by using<br />
production, trade <strong>and</strong> consumption data about<br />
products related to the specific waste streams.<br />
• Factor models: These models focus on<br />
analyses of the factors, which describe the<br />
processes of waste generation. Examples of<br />
proved parameters are e.g. the income of<br />
households, dwelling types or the type of<br />
heating.<br />
Based on this comparative study, only a few<br />
methodological procedures came into<br />
consideration for application within the aimed<br />
forecasting model for cities. This was due to the<br />
following reasons:<br />
• Level of aggregation: The identification of<br />
parameters has to be based on a database,<br />
which describes regional peculiarities. The<br />
exclusive use of national aggregates in inputoutput<br />
models [Patel et al., 1998] is not<br />
appropriate for explaining regional dynamics.<br />
Therefore preference was given to factor<br />
models that focus on socio-economic <strong>and</strong><br />
demographic indicators available at a regional<br />
level [Bach et al., 2003].<br />
• Predictability of parameters: The selection<br />
of model parameters has to prioritise<br />
parameters at the city level, which can be<br />
forecasted with a relatively high accuracy <strong>and</strong><br />
a long forecasting horizon. Examples of such<br />
parameters with high inertia are the<br />
population age structure, household size or<br />
infant mortality rate [Lindh, 2003].<br />
• Applicability refers to the user-friendliness<br />
of the aimed forecasting tool. Therefore<br />
methods that provide easily available,<br />
st<strong>and</strong>ardised secondary data have to be<br />
favoured over elaborate <strong>and</strong> time-consuming<br />
qualitative approaches such as the Delphi<br />
method [Karavezyris, 2001].<br />
Based on these considerations, the amount <strong>and</strong><br />
composition of municipal solid waste were<br />
hypothesised in this approach dependent upon<br />
easily available socio-economic <strong>and</strong> demographic<br />
parameters. Under the assumption of an analogous<br />
development of regional characteristics <strong>and</strong> MSW<br />
generation, this could explain regional differences<br />
between cities as well as long-term changes<br />
concerning a city by means of an ex post analysis.<br />
3. DATA<br />
Due to their relevance for waste management<br />
planning, the waste potentials of the main materials<br />
-- such as organic material, paper <strong>and</strong> cardboard,<br />
plastics <strong>and</strong> compounds, glass or metals -- were<br />
defined as variables to be explained. As the waste<br />
potential of a certain waste stream (apart from<br />
illegally dumped waste) contains the separately<br />
collected material <strong>and</strong> a partial stream of the<br />
residual waste, data about the collected quantities<br />
of residual waste <strong>and</strong> the separately collected<br />
materials as well as about the composition of<br />
residual waste (derived from sorting analyses) were<br />
both defined in order to be collected as data.<br />
The investigation covered the collection <strong>and</strong><br />
inspection of the mentioned waste-related data as<br />
well as the data in terms of economic, demographic<br />
<strong>and</strong> social indicators in all major European cities<br />
with more than 500,000 inhabitants. Together with<br />
six regional partners we co-operated with local city<br />
representatives who provided us with waste-related<br />
<strong>and</strong> socio-economic data at the city level.<br />
Additionally, national data were obtained from<br />
international organisations, such as the United<br />
Nations or OECD.<br />
To enable the analysis of regional dynamics, the<br />
collection of data covered the years from 1970 to<br />
2001. This formed the basis for the hypothesis of<br />
the existence of a long-term development path,<br />
which is based on the analogous time-shifted<br />
changes of different regions.<br />
Finally, it was possible to collect data about the<br />
municipal solid waste quantities (including general<br />
data at the city <strong>and</strong> country level) in 55 major cities<br />
(out of a total of 65) in the EU-15 <strong>and</strong> 10 CEE<br />
countries with an average time-series length of ten<br />
years. In terms of the waste potential of the main<br />
712
materials, only 45 data sets from 31 cities were<br />
available.<br />
4. STATISTICAL EVALUATION<br />
There are remarkable differences in the MSW<br />
generation rates as well as in the growth rates in<br />
European cities. As an example, a comparison of<br />
economic areas in the year 2000 shows that major<br />
EU-15 cities were characterised by far higher<br />
MSW generation rates (510 kg/cap/yr) than the<br />
CEE cities (354 kg/cap/yr), while from 1995 to<br />
2001 annual growth in CEE cities is more than<br />
twice as high (4.3%) as in cities of EU-15<br />
countries (1.8%).<br />
Therefore several bivariate <strong>and</strong> multivariate<br />
statistical analyses were carried out to identify<br />
indicators with a significant impact on MSW generation<br />
<strong>and</strong> composition. Table 1 shows the considered<br />
socio-economic indicators, which were<br />
available as a time-series at both the city <strong>and</strong><br />
national level.<br />
Table 1. Available socio-economic indicators.<br />
• Population<br />
Available indicators at city <strong>and</strong> national level<br />
• Population age structure<br />
(0 to 14 years / 15 to 59<br />
years / 60 <strong>and</strong> more years)<br />
• Gross domestic product<br />
• Overnight stays<br />
• Average household size<br />
• Population density<br />
• Sectoral employment<br />
(Agriculture / Industry /<br />
Services)<br />
• Infant mortality rate<br />
• Life expectancy at birth<br />
• Unemployment rate<br />
Firstly single regression analyses (using Kolmogorov-Smirnov<br />
tests) proved that the single parameters<br />
at the city level explain only an<br />
unsatisfactory part of the intertemporal <strong>and</strong><br />
interregional variance. The infant mortality rate<br />
(R²=0.37, n=86) as the most significant indicator<br />
performed even better than the commonly used<br />
gross domestic product (R²=0.22, n=86) as well as<br />
all other mentioned urban indicators. This is<br />
mainly due to the fact that the (usually available)<br />
mean urban gross domestic product is a less<br />
meaningful indicator of the social st<strong>and</strong>ard than the<br />
infant mortality rate as it does not reflect the social<br />
<strong>and</strong> economic inequality, which is especially high<br />
within CEE cities [Förster et al., 2002].<br />
Secondly the hypothesis of a general long-term<br />
development path concerning urban waste<br />
generation was tested by means of multivariate<br />
analyses. A hierarchical cluster analysis was<br />
therefore carried out to categorise each case<br />
(representing a city in a certain year) within<br />
homogeneous groups with a similar social <strong>and</strong><br />
economic st<strong>and</strong>ard, here defined as the ‘prosperity<br />
level.’ Four national development indicators (see<br />
Table 2) were selected as cluster criteria in order to<br />
explain this latent ‘prosperity level’ variable.<br />
Table 2 shows the assumed prosperity indicators<br />
<strong>and</strong> the MSW generation rates, which prove the<br />
hypothesised relationship. Low MSW generation<br />
rates coincide with low gross domestic products,<br />
high infant mortality rates <strong>and</strong> agriculturally dominated<br />
economy <strong>and</strong> vice versa. Similar results were<br />
obtained by a principal component analysis. An<br />
analysis of variance using One-way-ANOVA confirmed<br />
the rejection of the null hypothesis, which<br />
states the identity of the group means, with an F-<br />
value of 61.7 <strong>and</strong> an F-significance of below 0.1%.<br />
Table 2. Municipal solid waste generated <strong>and</strong><br />
development indicators.<br />
National Prosperity<br />
development level<br />
indicators <strong>and</strong><br />
MSW generation<br />
Low Medi<br />
um<br />
High<br />
Very<br />
high<br />
Gross domestic 5841 11400 19418 21317<br />
product per capita 1<br />
Infant mortality rate 2 15.0 8.7 7.6 5.5<br />
Labour force in<br />
agriculture (%)<br />
Labour force in<br />
services (%)<br />
Municipal solid<br />
waste (kg/cap/yr)<br />
24.0 18.7 4.8 3.2<br />
44.4 52.2 59.4 66.2<br />
287 367 415 495<br />
1 USD Purchasing power parities at 1995 prices<br />
2 Per 1,000 births<br />
Based on this prosperity-related classification, the<br />
analysis of the waste potential data unveiled longterm<br />
trends in the municipal solid waste streams<br />
(Figure 2). While the generation of paper <strong>and</strong><br />
cardboard, glass, plastics <strong>and</strong> compounds in<br />
MSW [kg/cap/yr]<br />
500<br />
400<br />
300<br />
200<br />
100<br />
0<br />
low medium high very high<br />
Prosperity level<br />
Other<br />
fractions<br />
Metals<br />
Plastics <strong>and</strong><br />
compounds<br />
Glass<br />
Paper <strong>and</strong><br />
cardboard<br />
Organic<br />
waste<br />
Figure 2. Municipal solid waste streams at<br />
different prosperity levels.<br />
prosperous cities is significantly higher both in<br />
absolute (per capita) <strong>and</strong> relative (mass<br />
percentage) figures, the amount of organic waste<br />
713
generated is very similar in these four city groups.<br />
Additionally the impact of the remaining MSW<br />
indicators (see Table 1) was tested <strong>and</strong> is closely<br />
related to the model development described below.<br />
5. MODEL<br />
5.1 Approach<br />
The developed model was designed for repeated<br />
use by municipal representatives to appropriately<br />
assess the future municipal waste streams of major<br />
European cities. In order to support decisions<br />
concerning waste management strategies, the<br />
planning horizon was defined by 15 years.<br />
Due to the background of these addressed users<br />
(municipal representatives), a model concept was<br />
created in order to enable a suitable compromise<br />
between ease of use <strong>and</strong> forecasting accuracy.<br />
Usability primarily refers to the preference for<br />
model input parameters that are adequately<br />
predictable at a city level, such as demographic or<br />
social indicators.<br />
The selection of the applied approach was based<br />
on recent forecasting methodology [Armstrong,<br />
2001]. The size <strong>and</strong> type of database as well as the<br />
existing knowledge of relationships were the<br />
criteria for the method selection.<br />
Thus different methodological approaches were<br />
selected for the following two modules: an ‘MSW<br />
generation module’ forecasting the total MSW<br />
generation rate <strong>and</strong> an ‘MSW composition module’<br />
assessing the future mass percentage of main waste<br />
streams within the municipal solid waste. The<br />
extensive database with time series <strong>and</strong> crosssectional<br />
data from the total MSW quantities<br />
allowed the implementation of an econometric<br />
model. The small database concerning waste<br />
potentials enabled the application of only the<br />
comparably simple extrapolation method.<br />
Figure 3 shows the elements of the overall model.<br />
The first calculation step is similar for both<br />
modules <strong>and</strong> is based on the findings for the longterm<br />
analogies between prosperity <strong>and</strong> MSW<br />
generation. A given city in a defined future year<br />
will be assigned to one of four prosperity groups<br />
using forecasts or assumptions for developmentrelated<br />
indicators, such as the gross domestic<br />
product or infant mortality rate.<br />
The MSW generation module is an econometric<br />
model, which consists of a system of multiple<br />
linear equations for each city group. Due to the<br />
short length of the available time series, the<br />
classical econometric approach, rather than the<br />
vector autoregression (VAR) approach, was<br />
implemented.<br />
The MSW composition module is based on the<br />
assumption that prosperity-related changes in the<br />
waste composition are similar. Due to the limited<br />
waste potential data, the objective lies on a rough<br />
trend estimation represented as default values.<br />
Model input<br />
Long-term<br />
analogies<br />
Model output<br />
Development<br />
indicators<br />
Stage of<br />
prosperity?<br />
Extrapolation<br />
MSW<br />
composition by<br />
main fractions<br />
MSW-related<br />
parameters<br />
Equation - group 1<br />
Equation - group 2<br />
:<br />
Equation - group x<br />
MSW<br />
growth rate<br />
per capita<br />
Figure 3. MSW generation model.<br />
5.2 Implementation issues<br />
The main issues concerning the implementation<br />
refer to the estimation of the econometric model<br />
using socio-economic data, which are represented<br />
as cross-sectional time series. The following<br />
potential problems were considered to avoid<br />
misspecifications of the model [Armstrong, 2001]:<br />
Econometric model<br />
• Collinearity of parameters: Social <strong>and</strong><br />
economic indicators often are highly<br />
correlated. The inclusion of too many<br />
variables within multiple regression models<br />
typically causes collinearity problems leading<br />
to ill-conditioned models. A suitable measure<br />
for the identification of collinear variables is<br />
the variance inflation factor that was therefore<br />
used.<br />
• Autocorrelations between residuals of<br />
neighbouring cases often occur during<br />
analyses of time series or structured data.<br />
This can depend on wrongly assumed<br />
functional relationships or on measurement<br />
errors during the data collection. Thus<br />
residual analyses including the use of Durbin-<br />
Watson coefficients were applied.<br />
• Outliers: The full consideration of outliers,<br />
which can occur due to measurement errors (a<br />
714
well-known problem in waste management),<br />
deteriorates the accuracy of regression model<br />
estimations. Hence the median absolute<br />
percentage error (MdAPE) [Armstrong,<br />
2001] was used to avoid the<br />
overrepresentation of unrealistic values.<br />
5.3 Procedures<br />
The attribution of the cases according the<br />
prosperity level was based on the classification<br />
mentioned in Chapter 4. The initial specification<br />
included the indicators listed in Table 1. The final<br />
model was selected by backward regression using<br />
the ordinary least squares method. To avoid<br />
autocorrelation <strong>and</strong> collinearity problems, Durbin-<br />
Watson statistics, collinearity tests <strong>and</strong> residual<br />
analyses were carried out.<br />
6. RESULTS<br />
The estimated equations of the final MSW<br />
generation model for cities are represented by:<br />
MSW<br />
t<br />
= 014<br />
t<br />
359 .5 + 0. ⋅ GDP<br />
(1)<br />
t<br />
−197.1⋅<br />
log( INF urb<br />
)<br />
for cities with very high prosperity,<br />
MSW<br />
t<br />
= 016<br />
t<br />
276 .5 + 0. ⋅ GDP<br />
(2)<br />
t<br />
−126.5<br />
⋅ log( INF urb<br />
)<br />
for cities with high prosperity, <strong>and</strong><br />
MSW<br />
t<br />
t<br />
( INF )<br />
= −360.7<br />
− 375.6 ⋅ log<br />
(3)<br />
nat<br />
t<br />
t<br />
+ .93 ⋅ POP<br />
−<br />
−123.<br />
9 ⋅ HHSIZE<br />
8<br />
15 59<br />
t<br />
+ 11. 7 ⋅ LIFEEXP<br />
for cities with low or medium prosperity,<br />
where MSW t is the municipal solid waste<br />
generated per capita <strong>and</strong> year, GDP t is the national<br />
gross domestic product per capita at 1995<br />
purchasing power parities, INF is the infant<br />
mortality rate per 1,000 births in the city (INF urb )<br />
t<br />
or in the country (INF nat ), POP 15-59 is the<br />
percentage of the population aged 15 to 59 years,<br />
HHSIZE t is the average household size <strong>and</strong><br />
LIFEEXP t is the life expectancy at birth <strong>and</strong> t is the<br />
year.<br />
Concerning the definition of prosperity levels,<br />
Table 3 shows the approximate threshold values<br />
for three national indicators among the city groups.<br />
For exact calculations, a set of canonical<br />
discriminant functions was determined to allocate a<br />
city with a given (present or future) characteristic.<br />
All parameters are significant at the 5% error level.<br />
Only the ‘infant mortality rate’ parameter is log<br />
transformed because of its obviously exponential<br />
nature. The model explains 65% of the variation of<br />
the MSW generation rate per capita between cities<br />
<strong>and</strong> in time. The model was validated with a holdout<br />
sample which included 59% of all cases. The<br />
out-of-sample error represents a median absolute<br />
percentage error of 8.0%, providing a useful model<br />
for waste management planning.<br />
Table 3. Approximate threshold values between<br />
city groups (national indicators).<br />
Prosperity Gross domestic Infant Labour force<br />
level product 1 mortality rate 2 in agriculture (%)<br />
Low<br />
7,100 12.0 21.4<br />
Medium<br />
13,800 8.1 10.5<br />
High<br />
20,200 6.3 4.0<br />
Very high<br />
1<br />
USD Purchasing power parities per capita at 1995 prices<br />
2<br />
Per 1,000 births<br />
In the following, the factors are described which<br />
were found to have a significant impact on the<br />
amount of municipal solid waste:<br />
• Gross domestic product: This commonly<br />
used indicator proved to be a significant<br />
factor in cities with a high prosperity, but not<br />
for cities with a lower economic output. This<br />
is due to the fact that the high regional<br />
income inequality in CEE countries [Förster<br />
et al., 2002] causes a big gap between the<br />
usually available mean values compared with<br />
the clearly lower median values, which is a<br />
more meaningful, but rarely observed<br />
indicator for the social well-being <strong>and</strong> living<br />
st<strong>and</strong>ard.<br />
• Social indicators: Previous studies never<br />
used the infant mortality rate <strong>and</strong> life<br />
expectancy parameters to indicate MSW<br />
generation, but they showed a remarkable<br />
ability to serve as an additional or alternative<br />
variable for the gross domestic product. The<br />
advantages of their use are the high<br />
explanatory power for regional welfare, the<br />
good availability of data, the high quality of<br />
data (due to easy compilation without<br />
complicated definitions) <strong>and</strong> the relatively<br />
good predictability on the part of city<br />
municipalities.<br />
• Age structure: The positive relationship<br />
between the percentage of the medium age<br />
group <strong>and</strong> MSW generation confirmed the<br />
previous studies [Sircar et al., 2003; Lindh,<br />
2003].<br />
• Household size: As in the study of Dennison<br />
et al. [1996], the significantly negative<br />
715
elationship between the average household<br />
size <strong>and</strong> MSW generation was analysed on a<br />
regional scale.<br />
7. CONCLUSION AND OUTLOOK<br />
At present environmental models, such as in waste<br />
management, are characterised by a high level of<br />
sophistication in terms of the technical <strong>and</strong><br />
environmental optimisation within the drawn<br />
system boundaries, but also by an almost complete<br />
lack of consideration of social, demographic <strong>and</strong><br />
economic border conditions. In waste management<br />
the usual municipal <strong>and</strong> regional waste forecasts<br />
(as the key starting point for the development of<br />
waste management plans) regard the total<br />
population as the only input parameter<br />
[Karavezyris, 2001]. As a consequence these<br />
strongly simplified assumptions can favour false<br />
estimations concerning future waste treatment<br />
capacities <strong>and</strong> dimensioning of restructuring<br />
activities.<br />
This approach aimed to integrate the impacts of<br />
regional socio-demographic <strong>and</strong> economic<br />
dynamics on the municipal solid waste generation.<br />
A qualitative, analogy-related approach was<br />
combined with the econometric methodology in<br />
order to assess these relationships for 55 very<br />
heterogeneous European cities in a long-term<br />
perspective. The results showed that this model can<br />
be implemented as a helpful decision support tool<br />
for municipal waste planners.<br />
In future work, this model will be statistically<br />
refined (especially the composition module) <strong>and</strong><br />
realised in aJava environment. The beta version<br />
will be tested in six European cities from regions<br />
with rapidly growing economies to verify its<br />
practicability.<br />
6. ACKNOWLEDGEMENTS<br />
This study is part of a project initiated within the<br />
European Commission's Fifth Framework<br />
Programme (EVK4-CT-2002-00087).<br />
The authors wish to thank the following people for<br />
their dedicated work during the course of the<br />
investigation (in alphabetical order): Heinrich<br />
Riegler (BOKU – University of Natural Resources<br />
<strong>and</strong> Applied Life Sciences, Austria); Emilia den<br />
Boer, Jan den Boer (Darmstadt University of<br />
Technology, Germany); Panagiotis Panagiotakopoulos,<br />
Nantia Tsilemou (Democritus University<br />
of Thrace, Greece); Pouwel Inberg, Marjakke van<br />
der Sluis, Wim van Veen (De Straat Milieuadviseurs,<br />
the Netherl<strong>and</strong>s); Lothar Schanne<br />
(novaTec, Luxembourg); Francesc Castells,<br />
Montse Meneses, Julio Rodrigo (Universitat<br />
Rovira i Virgili, Spain); Iwona Mackow, Pawel<br />
Mrowinski, Marta Sebastian <strong>and</strong> Ryszard Szpadt<br />
(Wroclaw University of Technology, Pol<strong>and</strong>).<br />
8. REFERENCES<br />
Armstrong, J.S., Principles of forecasting: a<br />
h<strong>and</strong>book for researchers <strong>and</strong> practitioners,<br />
Kluwer Academic Publishers, Boston, 2001.<br />
Bach, H., A. Mild, M. Natter, <strong>and</strong> A. Weber,<br />
Combining socio-demographic <strong>and</strong> logistic<br />
factors to explain the generation <strong>and</strong><br />
collection of waste paper, Resources,<br />
Conservation <strong>and</strong> Recycling, In Press, 2003.<br />
Beigl, P., G. Wassermann, F. Schneider, <strong>and</strong> S.<br />
Salhofer, Municipal solid waste generation<br />
trends in European countries <strong>and</strong> cities, paper<br />
presented at the 9th <strong>International</strong> Waste<br />
Management <strong>and</strong> L<strong>and</strong>fill Symposium, CISA,<br />
S. Margherita di Pula (Cagliari), Sardinia,<br />
Italy, October 6-10, 2003.<br />
Björklund, A., <strong>Environmental</strong> systems analysis of<br />
waste management – Experiences from<br />
applications of the ORWARE model, Ph.D.<br />
thesis, Royal Institute of Technology,<br />
Stockholm, 2000.<br />
Dennison, G.J., V.A. Dodd, <strong>and</strong> B. Whelan, A<br />
socio-economic based survey of household<br />
waste characteristics in the city of Dublin,<br />
Irel<strong>and</strong>, II. Waste quantities. Resources,<br />
Conservation <strong>and</strong> Recycling, 17, 245-257,<br />
1996.<br />
Förster, M., D. Jesuit <strong>and</strong> T. Smeeding, Regional<br />
poverty <strong>and</strong> income inequality in central <strong>and</strong><br />
eastern Europe: Evidence from the<br />
Luxembourg income study, Syracuse<br />
University, New York, 2002.<br />
Karavezyris, V., Prognose von Siedlungsabfällen:<br />
Untersuchungen zu determinierenden<br />
Faktoren und methodischen Ansätzen, TK<br />
Verlag, Neuruppin, 2001.<br />
Lindh, T., Demography as a forecasting tool,<br />
Futures, 35, 37-48, 2003.<br />
Patel, M.K., E. Jochem, P. Radgem <strong>and</strong> E.<br />
Worrell, Plastic streams in Germany – an<br />
analysis of production, consumption <strong>and</strong><br />
waste generation, Resources, Conservation<br />
<strong>and</strong> Recycling, 24, 191-215, 1998.<br />
Sircar, R., F. Ewert, <strong>and</strong> U. Bohn, Ganzheitliche<br />
Prognose von Siedlungsabfällen, Müll und<br />
Abfall, 1, 7-11, 2003.<br />
White, P., M. Franke <strong>and</strong> P. Hindle, Integrated<br />
waste management: A lifecycle inventory,<br />
Aspen Publishers, Maryl<strong>and</strong>, USA, 1999.<br />
716
Real Time Optimal Resource Allocation in Natural<br />
Hazard Management<br />
P. Fiorucci a,b , F. Gaetani a,b , R. Minciardi a,b , R. Sacile a,b , E. Trasforini a<br />
eva.trasforini@unige.it<br />
a CIMA – Centro Interuniversitario di ricerca in Monitoraggio Ambientale, Università di Genova e della<br />
Basilicata, Italy.<br />
b DIST – Dipartimento di Informatica, Sistemistica e Telematica, Università degli Studi di Genova, Italy.<br />
Abstract: In the emergency management phase relevant to the occurrence of a catastrophic natural event,<br />
the efficiency of the emergency system can be deeply influenced by a correct assignment of the available<br />
resources to the “dem<strong>and</strong>” centers, i.e., those elements of the territory that are directly involved in the event.<br />
In this paper, a general formulation of the real time optimal resource allocation problem is presented, <strong>and</strong> it is<br />
formalized as a mathematical programming problem. The dynamics relevant both to the resources over the<br />
territory <strong>and</strong> to the prediction of the behavior of the natural phenomenon are suitably taken into account. A<br />
graph model is introduced in order to describe the territory under consideration, <strong>and</strong> resources can be located<br />
either in the nodes or are in transit over the direct links. The phenomenon dynamics can be associated to the<br />
nodes on the territory, <strong>and</strong> specifically to the dem<strong>and</strong> centers. The general approach has then been applied to<br />
the forest fire hazard. A simple model aiming to describe the spread of a fire over the territory has been<br />
developed, <strong>and</strong> it has been used for the formulation of the mathematical programming problem. A specific<br />
case study, relevant to the forest fire hazard in Liguria region (Italy), has been here defined <strong>and</strong> tested, <strong>and</strong> the<br />
main results are briefly reported <strong>and</strong> discussed.<br />
Keywords: Decision support systems, real time natural hazard management, environmental risk, forest fires,<br />
resource allocation problems.<br />
1. INTRODUCTION<br />
When a catastrophic natural event occurs, the<br />
efficiency of the emergency system can be deeply<br />
influenced by a correct allocation of the available<br />
resources to the “dem<strong>and</strong>” centers, i.e., those<br />
elements of the territory that are directly involved<br />
in the event.<br />
The optimal allocation of resources is a wellknown<br />
problem in Operation Research, which has<br />
been faced following many approaches. In the field<br />
of catastrophic event management, some specific<br />
approaches can be found: for example, in the<br />
resource management module of the TRACE<br />
system an allocation procedure is present to define<br />
an allocation plan [Paggio et al., 1999], while the<br />
work by Friedrich et al. [2000] is relevant to the<br />
optimal resource allocation after an earthquake.<br />
However, a modern view of civil protection<br />
comm<strong>and</strong> centers [Wybo, 1992] would require an<br />
integrated approach of the several natural risks,<br />
which may occur on a territory, with a clear<br />
formulation of the cost/benefit of a resource<br />
allocation plan. In this connection, strategic<br />
importance is assumed by the operative phase,<br />
when the event is occurring or about to occur; it is<br />
worth mentioning that no specific approach can be<br />
found in the literature as regards this issue.<br />
The main goal of this work is to define the<br />
structure of a decision support system (DSS), to<br />
aid decision makers in the optimal allocation of the<br />
resources required to manage an emergency due to<br />
a catastrophic natural event. The main module of<br />
the DSS is represented by a mathematical<br />
programming problem, whose general formulation<br />
is provided in this paper, in order to provide a<br />
common framework to treat all kinds of natural<br />
hazards. However, each kind of natural hazard<br />
requires a special formulation of the dynamics of<br />
the event <strong>and</strong> of the resources that are necessary to<br />
cope with the emergency. A case study is described<br />
in this work, referring to the real-time management<br />
of forest fire hazard.<br />
2. THE PROPOSED APPROACH<br />
In a resource allocation problem, first of all, the<br />
following issues are to be classified:<br />
717
− the representation of resources either via<br />
continuous or via integer variables;<br />
− the service dem<strong>and</strong> that may be distributed or<br />
concentrated over the territory.<br />
As a matter of fact, every alternative in both issues<br />
could be a reasonable modeling approach in the<br />
emergency management of natural hazards:<br />
however, for the sake of brevity, an exhaustive<br />
analysis of such issues is left to forthcoming<br />
papers.<br />
In this work, which is oriented towards a forest fire<br />
management hazard application, it seemed<br />
reasonable to choose a continuous representation<br />
for the resources, <strong>and</strong> a concentrated one for the<br />
service dem<strong>and</strong>; these features are shown in the<br />
next section.<br />
A general formalization for the real time resource<br />
allocation problem, with continuous resources <strong>and</strong><br />
concentrated dem<strong>and</strong>, is so hereinafter introduced.<br />
Then, the specific problem relevant to forest fires<br />
hazard is considered, <strong>and</strong> a detailed formulation of<br />
the resulting mathematical programming problem<br />
is provided.<br />
2.1 The general formalization<br />
It is assumed that both dem<strong>and</strong> centers <strong>and</strong><br />
resource location centers are represented as nodes<br />
of the directed graph G(V,L), where V is the set of<br />
nodes, <strong>and</strong> L is the set of the links among those<br />
nodes.<br />
If j is a generic node belonging to V, let us indicate<br />
by P(j) (resp. S(j)) the set of nodes predecessor<br />
(successor) of node j.<br />
The general problem formalization refers to a time<br />
horizon (of a suitable length) of T time intervals. It<br />
is assumed that each link belonging to the set L is<br />
characterized by a unitary transit time, that is the<br />
time required by the resources in order to transit<br />
over the link. This assumption can be generalized<br />
to include links with transit time greater than one,<br />
by introducing a suitable number of dummy nodes,<br />
each one characterized by null service dem<strong>and</strong>.<br />
The primary decision variables of the problem are:<br />
U j (t): amount of resources assigned to node j<br />
during time interval (t, t+1), t=0,…,T-1,<br />
j∈V;<br />
w jl (t): amount of resources that during time<br />
interval (t, t+1), t=0,…,T-1, move from<br />
node j to node l.<br />
Further variables necessary for the formalization of<br />
the problem are:<br />
D j (t)<br />
service dem<strong>and</strong> in j during time interval<br />
(t, t+1), t=0,…,T-1 (it may result from<br />
dynamic model whose behaviour is<br />
influenced by the amount of resources<br />
assigned to that particular node).<br />
Besides, the parameters U ~<br />
j indicate the amount of<br />
resources located at each node of the considered<br />
graph at the initial time instant of the optimization<br />
horizon.<br />
The cost function of the proposed general<br />
formulation is composed by three terms: the first<br />
one is relevant to the estimated damage, the second<br />
one is relevant to the inadequate assignment of<br />
resources to the nodes of the graph, <strong>and</strong> the last<br />
one relevant to the transfer cost between the nodes.<br />
The first term has to be written in order to penalize<br />
the total amount of estimated damage in each node<br />
<strong>and</strong> in each time interval, <strong>and</strong> can be expressed in<br />
the general form<br />
P<br />
Pj ( t) = f j ( D j ( t)<br />
).<br />
The second term is introduced in order to take into<br />
account the possible partial incompatibility among<br />
the resources <strong>and</strong> a location center (e.g., the base<br />
may be not well equipped for a particular kind of<br />
resources); the functions expressing the cost of<br />
inadequate assignment of resources to nodes O j (t)<br />
can be expressed in the general form<br />
O<br />
O j ( t) = f j ( U j ( t)<br />
).<br />
Finally, the transfer cost s jl (t) between nodes is<br />
introduced in order to penalize the resource<br />
movements among the nodes. Such a cost takes<br />
into account the total amount of resources moving<br />
on each link for each time interval, <strong>and</strong> can be<br />
expressed as:<br />
s<br />
s jl ( t) = f jl ( w jl ( t)<br />
).<br />
The general formalization of the dynamic (realtime)<br />
resource allocation problem is the following<br />
min T t<br />
−1<br />
∑ ∑<br />
= 0 j∈V<br />
T −<br />
+ 1 t= 0 j∈V<br />
l∈S<br />
s.t.<br />
∑ ∑ ∑<br />
f<br />
P<br />
j<br />
T −1<br />
O<br />
( D ( t)<br />
) + ∑ ∑ f U ( t)<br />
s<br />
j<br />
( w ( t)<br />
)<br />
t= 0 j∈V<br />
j<br />
( )<br />
f jl jl<br />
(1)<br />
( j)<br />
D<br />
j<br />
[ D ( t),<br />
U ( t ] j ∈V<br />
D ( t + 1) = f<br />
)<br />
j<br />
j<br />
t = 0,...,<br />
T −1<br />
(2)<br />
U<br />
j<br />
~<br />
0 = j ∑ jl j ∈V<br />
(3)<br />
l∈S<br />
( j)<br />
( ) U − w ( 0)<br />
( t + 1) = U ( t) − ∑ w ( t + )<br />
U j<br />
j<br />
jl 1 + wlj<br />
t<br />
l∈S<br />
( j)<br />
l∈P( j)<br />
j ∈V<br />
t = 0,...,<br />
T −1<br />
(4)<br />
∑U<br />
j∈V<br />
j<br />
~ + jl − ∑U<br />
j = 0<br />
j∈V<br />
l∈S<br />
( j)<br />
j∈V<br />
( t) ∑ ∑ w ( t)<br />
j<br />
j<br />
∑<br />
+<br />
( )<br />
t = 0,...,<br />
T −1<br />
(5)<br />
U j<br />
( t) j ∈V<br />
≥ 0 t = 0,...,<br />
T −1<br />
(6)<br />
718
w jl<br />
( t) j ∈V<br />
l ∈ S( j)<br />
≥ 0 t = 0,...,<br />
T −1<br />
(7)<br />
As a matter of fact, natural hazard management is<br />
in general characterized by two different dynamics:<br />
the first one is relevant to the physical process that<br />
characterizes the specific event (flood, forest fires,<br />
etc.), whereas the second one is relevant to the<br />
resources intervention. The two different dynamics<br />
nay be mutually related; for example considering<br />
an earthquake phenomenon, one can observe that<br />
after the physical event, the main goal of the<br />
resources is that of rescue the maximum number of<br />
injured people as early as possible; besides, the<br />
survival possibility of such people could be<br />
characterized by a dynamics strongly correlated<br />
with the dynamics of the intervention resources. In<br />
case of forest fire, as it will be described in the<br />
next section, risk is related with the fire spread<br />
dynamics, which is strongly correlated with the fire<br />
attack provided by the resources able to face the<br />
propagation of the fire.<br />
Coming back to the above formalization, it is<br />
worth observing explicitly that: constraints (2) are<br />
relevant to the dynamics of the process for a single<br />
node; constraints (3) <strong>and</strong> (4) are related to resource<br />
kinematics; constraints (5) are related to the<br />
conservation of resources over the territory;<br />
constraints (6) <strong>and</strong> (7) ensure the non-negativity of<br />
the decision variables.<br />
2.2 Application of the general formalization<br />
to the forest fire hazard management<br />
During real time management of forest fires, one of<br />
the main objectives is to position the firefight<br />
resources in an optimal way to face the ongoing<br />
fire events, taking into account their dynamics.<br />
The territory is modeled as a graph with three types<br />
of nodes: fires, fire fight stations <strong>and</strong> transit nodes,<br />
introduced to satisfy the hypothesis that the<br />
transfer time on all arcs is unitary.<br />
The notation used in the formulation of this<br />
problem is analogous to the one introduced in the<br />
previous section. Furthermore, let us define Υ⊂V<br />
as the subset of nodes belonging to V <strong>and</strong><br />
representing fires. The (continuous) resources are<br />
represented by the amount of extinguishing power,<br />
<strong>and</strong> variables U j (t), w jl (t) are expressed in kW.<br />
The dem<strong>and</strong> D j (t) - which plays a key role both in<br />
the objective function <strong>and</strong> in the constraints<br />
representing the process dynamics in a real time<br />
resource allocation problem - is supposed to be<br />
simply the overall power which characterizes the<br />
fire corresponding to a given node, namely<br />
( t) p ( t)<br />
D j = j<br />
(8)<br />
In the past years, several approaches aiming at<br />
modeling forest fires dynamics in space <strong>and</strong> time<br />
have been proposed in the literature. Among the<br />
others, the models belonging to the so-called semiphysical<br />
class seem to be the most practical <strong>and</strong><br />
widely used by technicians <strong>and</strong> the scientific<br />
community. Semi-physical models take into<br />
account the physical laws governing the fire spread<br />
phenomenon by means of parametrical<br />
relationships among fuel characteristics,<br />
meteorological variables <strong>and</strong> topography. The<br />
parameters of the model can be estimated on the<br />
basis of empirical observations consisting of infield<br />
measurements, related to real case studies, or<br />
to experimental fires, or collected in a combustion<br />
laboratory (Rothermel, 1972; Van Wagner, 1977;<br />
Albini, 1985).<br />
However, as the purpose of this work is that of<br />
evaluating the effectiveness of the proposed<br />
approach in connection with simple case studies,<br />
<strong>and</strong> since the discussion <strong>and</strong> the selection of<br />
suitable propagation models are far beyond the<br />
scope of the present work, only a very simple fire<br />
propagation model will be used.<br />
Such a model is characterized by a perfect isotropy<br />
(absence of orographic asperities, absence of wind,<br />
etc.). In this case, a rough approximation is to<br />
consider the rate of increase of the burnt area as<br />
linearly dependent on the overall power of the fire,<br />
<strong>and</strong> the rate of increase of the power itself as<br />
constant (at least, as far as no extinguishing action<br />
is taken). The equations representing such quite<br />
simple dynamics are<br />
da<br />
j ( t)<br />
= k<br />
dt<br />
<br />
dp<br />
j ( t)<br />
= k<br />
<br />
dt<br />
j<br />
1<br />
j<br />
2<br />
where<br />
a j (t) [m 2 ]<br />
⋅ p<br />
j<br />
⋅1(<br />
t)<br />
( t)<br />
j ∈ Υ<br />
(9)<br />
is the area burnt by fire j at<br />
instant t;<br />
p j (t) [kW] is the power of fire j at instant t.<br />
In this simple representation, if the fire front is<br />
modeled as a circumference, a linear increase (with<br />
time) of the radius r j (t) gives a quadratic increase<br />
of the area a j (t).<br />
Assuming a linear intensity p lin [ kW/m] of the fire,<br />
constant over the circumference, the overall power<br />
p j (t) of the fire can be expressed as<br />
p<br />
j<br />
( t) p 2π<br />
r ( t)<br />
= (10)<br />
lin<br />
j<br />
Thus, the first equation of (9) can be re-written as<br />
da<br />
j<br />
j<br />
( t) π r ( t) dr = k p 2π<br />
r ( t)dt<br />
= (11)<br />
2 j j 1<br />
that allows obtaining<br />
k<br />
dr 1<br />
= (12)<br />
dt p<br />
j j<br />
1<br />
⋅<br />
where<br />
lin<br />
lin<br />
j<br />
719
dr j<br />
is the spread speed of the fire.<br />
dt<br />
From the second equation of (9)<br />
dp<br />
j<br />
j 2<br />
= k dt<br />
(13)<br />
Then, differentiating equation (10) <strong>and</strong> substituting<br />
in equation (13), one obtains:<br />
p<br />
j<br />
lin j 2<br />
so that<br />
2 π dr = k dt<br />
(14)<br />
j<br />
k 2<br />
is equal to<br />
k<br />
dr<br />
j j<br />
2<br />
= ⋅ 2π<br />
⋅ plin<br />
(15)<br />
dt<br />
The discretization of (9) provides<br />
<br />
a<br />
<br />
<br />
p<br />
j<br />
j<br />
( t + 1) = a j ( t) + k1<br />
p j ( t)<br />
∆t<br />
j<br />
( t + 1) = p ( t) + k ∆t<br />
−U<br />
( t)<br />
j<br />
j<br />
2<br />
j<br />
j ∈Y<br />
(16)<br />
where it has been taken into account the action of<br />
the extinguishing resources U j (t).<br />
Resource location centres<br />
Transit nodes<br />
Fires<br />
Figure 1. Representation of the target area <strong>and</strong> of the directed graph used in order to formalize the optimal<br />
real time resource allocation problem relevant to forest fires risk in the Liguria region. The two fires located<br />
near nodes 2 <strong>and</strong> 6 are the ones relevant to Scenario 2.<br />
As regards the functions needed to describe the<br />
real time resource allocation problem objective,<br />
f ⋅ has been omitted since it is not relevant for<br />
o<br />
j<br />
( )<br />
the considered case, because only one kind of<br />
resources has been considered. Specifically, only<br />
the amount of water that can be carried with truck<br />
engines is considered. Furthermore, such kind of<br />
resources can indifferently be assigned to each one<br />
of the considered location centers. Thus, no<br />
inadequate location for the resources can be found<br />
among the considered nodes (namely, nodes<br />
corresponding to location centers, fires, or transit<br />
nodes).<br />
Besides, it is assumed<br />
P<br />
s<br />
j<br />
jl<br />
2<br />
( t) γ a ( t)<br />
= (17)<br />
j,<br />
t<br />
2<br />
( t) η w ( t)<br />
j,<br />
l,<br />
t<br />
j<br />
= (18)<br />
jl<br />
where γ j, t <strong>and</strong> η j , l,<br />
t are suitable weight parameters.<br />
Then the overall problem can be stated as<br />
min<br />
T −1<br />
∑∑<br />
t= 0 j∈Y<br />
γ<br />
j,<br />
t<br />
a<br />
2<br />
j<br />
T −1<br />
∑∑∑<br />
2<br />
( t) + η w ( t)<br />
t= 0 j∈V<br />
l∈V<br />
l ≠ j<br />
j,<br />
l,<br />
t<br />
jl<br />
(19)<br />
subject to constraints (3) ÷ (7), (16).<br />
Note that the value of quantities a j (0) for each j∈Υ,<br />
<strong>and</strong> U j (0), for each j∈V, are supposed to be known.<br />
3. APPLICATION TO TEST CASES<br />
The model described in the previous section has<br />
been applied to the Liguria region. Among the<br />
Italian northern regions, Liguria is the one that is<br />
more affected by this kind of calamity, with more<br />
than 500 fires a year <strong>and</strong> a very high rate between<br />
the total burnt area (about 3060 ha in 2002, <strong>and</strong><br />
4560 in 2003) <strong>and</strong> the total forest area (about<br />
334000 ha).<br />
Forty-four Forest Service’s stations are spread over<br />
the territory, with an overall nominal value of<br />
extinguishing power of about 270000 kW. This<br />
value has been obtained by considering only the<br />
available l<strong>and</strong>-force resources placed in each<br />
station <strong>and</strong> under the simplifying assumption that<br />
the water flow given by the considered mean<br />
reaches completely the fire front <strong>and</strong><br />
instantaneously evaporates [Fiorucci et al., 2002].<br />
Due to the short distance between some pairs of<br />
stations, <strong>and</strong> in order to simplify the statement <strong>and</strong><br />
the structure of the real time resource allocation<br />
problem, a clustering of the stations in 6 nodes has<br />
720
een performed. The overall graph of the<br />
considered problem is thus composed of 12 nodes<br />
(6 of them are transit nodes) <strong>and</strong> 22 bi-directional<br />
links (see Figure 1). The power present in each<br />
cluster is the result of the sum of all the power<br />
relevant to the stations of the cluster.<br />
Two different scenarios have been taken into<br />
account, <strong>and</strong> are briefly described below.<br />
Scenario 1 is characterized by a single forest fire,<br />
in the eastern part of the region, in correspondence<br />
of node N2; in the following, this fire will be<br />
denoted as S1F1.<br />
Two different fires characterize scenario 2, one in<br />
the neighborhood of node N2 <strong>and</strong> one near N6<br />
(denoted in the following as S2F1 <strong>and</strong> S2F2,<br />
respectively). The two fires are characterized by<br />
different values of parameters k j 1 <strong>and</strong> k j 2; in<br />
particular, S2F2 is characterized by a value of k j 2<br />
greater than S2F1 <strong>and</strong>, therefore it is characterized<br />
by a faster increase of the power over time.<br />
The main characteristics of the three scenarios are<br />
resumed in table 1.<br />
With reference to the two scenarios, two strategies<br />
are described in the following subsections:<br />
heuristic strategy, <strong>and</strong> the strategy defined by the<br />
solution of the real time resource allocation<br />
problem.<br />
Both the scenarios are characterized by an<br />
optimization horizon T of 15 hours, subdivided<br />
into regular time intervals, equals to 1800 seconds.<br />
dr/dt<br />
[m/s]<br />
p lin<br />
S1F1 S2F1 S2F2<br />
0.02 0.01 0.01<br />
[kW/m]<br />
1000 1000 1200<br />
p(0)<br />
[kW]<br />
125600 125600 150720<br />
a(0)<br />
[m 2 ]<br />
1256 1256 1256<br />
k 1<br />
[(m/s)/ 2*10 -5 10 -5 8*10 -6<br />
(kW/m)]<br />
k 2<br />
[(m/s)* 125 62.8 73.36<br />
(kW/m)]<br />
Node N2 N6 N2<br />
Table 1. The main parameters characterizing the<br />
three fires of the two scenarios.<br />
3.1 Heuristic strategies<br />
In this case, the management problem is governed<br />
by a greedy heuristic strategy: the resources are<br />
allocated to the nearest fire until the dem<strong>and</strong> is<br />
satisfied; if two or more fires have to be<br />
contemporarily extinguished, available resources<br />
are assigned proportionally to coefficient k j 2.;<br />
resources can be moved from a fire <strong>and</strong> sent to<br />
another one only when an active fire is completely<br />
extinguished. Recall that the evolution of the<br />
dynamics of the fire is described by equations (9).<br />
The single fire in Scenario 1 (S1F1) required 24<br />
time intervals to be extinguished, affecting an area<br />
greater than 4,000,000 m 2 .<br />
Referring to Scenario 2, the fire affecting node 6<br />
(S2F2) did not give rise to considerable damages<br />
(less than 35,000 m 2 of burned area) as all the<br />
resources were initially concentrated on this fire,<br />
because of its value of parameter k j 2 higher than<br />
S2F1. This choice influenced the trend of fire<br />
S2F1: as a few resources were initially assigned to<br />
this fire, it reached a maximum power similar to<br />
the one of S1F1 (characterized by a higher value of<br />
coefficient k j 2), <strong>and</strong> the burned area was greater<br />
than 1,200,000 m 2 .<br />
3.2 Real time optimal strategies (RTO)<br />
In this case, the management problem is governed<br />
by the strategy defined by means of the solution of<br />
the optimization problem described in subsection<br />
2.2.<br />
S1F1 required 19 time intervals to be extinguished<br />
by the use of these strategies, whereas the burned<br />
area is equal to 250,000 m 2 .<br />
Here again, in scenario 2 resources are first<br />
assigned to S2F2, <strong>and</strong> thus this fire was<br />
extinguished before S2F1: the intervention ended<br />
in time interval 21 on S2F2, <strong>and</strong> in time interval 27<br />
on S2F1, <strong>and</strong> the burned areas were 63,000 m 2 <strong>and</strong><br />
50,000 m 2 , respectively.<br />
3.3 Comparison of the strategies<br />
It is reasonable to expect that the application of<br />
RTO strategies allows an improvement, with<br />
respect to the application of heuristic strategies,<br />
especially as regards the overall burned area. In<br />
fact, this factor is the main term of the cost<br />
function, of the optimization problem.<br />
square meters<br />
Burned Area-S1<br />
450000<br />
400000<br />
350000<br />
300000<br />
250000<br />
200000<br />
150000<br />
100000<br />
50000<br />
0<br />
0 5 10 15 20 25 30<br />
time intervals<br />
Heuristic<br />
RTO<br />
Figure 2. Cumulated burned area [m 2 ] relevant to<br />
Scenario 1 obtained by the use of heuristic<br />
strategies <strong>and</strong> RTO strategies.<br />
721
Referring to Scenario 1 (see figure 2), it can be<br />
noted that this expectation is perfectly confirmed.<br />
In fact, a comparison between heuristic <strong>and</strong> RTO<br />
strategies shows a decrease of the overall burned<br />
area of 37%.<br />
In order to compare the results obtained in<br />
connection to Scenario 2, it is necessary to<br />
consider jointly the fires in each scenario (i.e.,<br />
S2F1 <strong>and</strong> S2F2 for Scenario 2). In fact, the cost<br />
function used for the definition of RTO strategies<br />
penalizes the global burned area, <strong>and</strong> not the<br />
burned area relevant to each fire.<br />
square meters<br />
200000<br />
150000<br />
100000<br />
50000<br />
Burned Area-S2<br />
0<br />
0 5 10 15 20 25 30<br />
time intervals<br />
Heuristic<br />
RTO<br />
Figure 3. Cumulated burnt area [m 2 ] relevant to<br />
Scenario 2 obtained by the use of heuristic<br />
strategies <strong>and</strong> RTO strategies on both S2F1 <strong>and</strong><br />
S2F2 fires.<br />
Referring to Scenario 2, figure 3 shows that the<br />
application of RTO strategies allows a decrease of<br />
burned area of nearly 30% with respect to the<br />
results obtained by the application of the heuristic<br />
procedure.<br />
4. CONCLUSIONS AND FUTURE<br />
DEVELOPMENTS<br />
In the modern view of civil protection comm<strong>and</strong><br />
centers, a continuous coordinated work is<br />
requested to face emergencies. In general,<br />
preventive actions before the event <strong>and</strong><br />
coordination optimal actions either during the<br />
emergence or its very preceding instants are crucial<br />
to minimize the effects of probable catastrophes.<br />
That is why an emerging area of research is the<br />
study of methodologies <strong>and</strong> technologies,<br />
exploiting the application of interdisciplinary<br />
competences (e.g. system sciences, information<br />
technologies, operation research, telematics etc..)<br />
to support decision within this context.<br />
In this work, the formalization of this latter aspect,<br />
specifically the real time optimal resource<br />
allocation in natural hazard management, has been<br />
presented, focusing on a case study relevant to<br />
forest fire hazard. The formalization of the<br />
decision problem, although may be subject to<br />
several improvements (for example, in the forest<br />
fire model), has the capability to support the<br />
decision makers in managing complex situations,<br />
where few resources are available <strong>and</strong> many<br />
dem<strong>and</strong> centers are requesting them. In addition,<br />
the developed decision problem can be used as a<br />
training support in simulated case studies.<br />
A future development of the proposed approach, at<br />
which the authors are already working, is the<br />
coupling of the proposed technique with another<br />
tool which is in charge to move resources in<br />
advance, when probable future requests are<br />
predicted by simulation models relevant to the<br />
specific considered hazard.<br />
5. REFERENCES<br />
Albini F.A., A model for fire spread in wildl<strong>and</strong><br />
fuels by radiation, Combustion Science <strong>and</strong><br />
Technology, 42, 229-258, 1985.<br />
Fiorucci P., Gaetani F., Minciardi R., Dynamic<br />
models for preventive management <strong>and</strong> real<br />
time control of forest fires, proceedings of 15th<br />
IFAC world congress in automatic control,<br />
Barcelona, Spain, 2002.<br />
Friedrich F., Gehbauer F., Rickers U., Optimized<br />
resource allocation for emergency response<br />
after earthquake disasters, Safety Science, 35,<br />
41-57, 2000.<br />
Paggio R., Agre G., Dichev C., Umann G.,<br />
Rozman T., Batachia L., Stocchero M., A costeffective<br />
programmable environment for<br />
developing environmental decision support<br />
systems, <strong>Environmental</strong> <strong>Modelling</strong> & <strong>Software</strong>,<br />
14, 367-382, 1999.<br />
Rothermel R.C., A mathematical model for<br />
predicting fire spread in wildl<strong>and</strong> fuels. USDA,<br />
Forest Service Research Paper. INT-114.<br />
Intermountain Forest <strong>and</strong> Range Experiment<br />
Station, Ogden, UT, 40 p, 1972.<br />
Van Wagner C.E.. Conditions for the start <strong>and</strong><br />
spread of crown fire, Canadian Journal of<br />
Forest Research,. 7, 23-34, 1977.<br />
Wybo J. L.. “EXPERTGRAPH: Knowledge based<br />
analysis <strong>and</strong> real-time monitoring of spatial,<br />
application to forest fire prevention in French<br />
Riviera, proceedings of <strong>International</strong><br />
Emergency Management <strong>and</strong> Engineering<br />
Conference, Managing Risk with Computer<br />
Simulation, Orl<strong>and</strong>o, FL, 1992.<br />
722
Combining Dynamic Economic Analysis <strong>and</strong> <strong>Environmental</strong><br />
Impact <strong>Modelling</strong>: Addressing Uncertainty <strong>and</strong><br />
Complexity of Agricultural Development<br />
Heikki Lehtonen a , Ilona Bärlund b , Sirkka Tattari b <strong>and</strong> Mikael Hilden b<br />
a MTT Economic Research, Agrifood Research Finl<strong>and</strong>, Luutnantintie 13, FIN-00410 Helsinki, Finl<strong>and</strong><br />
e-mail: heikki.lehtonen@mtt.fi<br />
b Finnish Environment Institute P.O.Box 140, FIN-00251 Helsinki, Finl<strong>and</strong><br />
Abstract: In this study the impacts of different agricultural policies on agricultural production <strong>and</strong> nutrient<br />
leaching from agricultural l<strong>and</strong>s are evaluated using the economic DREMFIA agricultural sector model <strong>and</strong> a<br />
field scale nutrient transport model ICECREAM. DREMFIA includes an evolutionary scheme of technology<br />
diffusion which considers farm investments, evolving farm size structure <strong>and</strong> technological change explicitly.<br />
The technology diffusion model allows self-inforcing patterns of technical change driven by the spread of<br />
information <strong>and</strong> farmers’ knowledge related to different technological alternatives. Hence the long-term<br />
changes in agriculture due to policy changes may be essentially larger than those predicted by traditional<br />
static equilibrium models. Larger potential for changes in production provides a larger perspective for<br />
evaluation of environmental impacts. The environmental effects are studied using the field scale nutrient<br />
transport model ICECREAM, based on the l<strong>and</strong> use changes predicted by the DREMFIA model. The modelled<br />
variables are nitrogen <strong>and</strong> phosphorus losses in surface runoff <strong>and</strong> percolation. Eutrophication of surface<br />
waters is the considered environmental effect. In this paper the modelling strategy will be presented <strong>and</strong><br />
highlighted using two case study catchments with varying environmental conditions <strong>and</strong> l<strong>and</strong> use.<br />
Keywords: agricultural policy; economic modelling; technical change; eutrophication; nutrient leaching modelling<br />
1. INTRODUCTION<br />
Water quality, influenced by agricultural activities<br />
<strong>and</strong> policies, has been of great public interest <strong>and</strong><br />
part of agricultural policy debate in Finl<strong>and</strong>. For<br />
this reason, both long-term economic viability of<br />
agriculture, <strong>and</strong> nutrient leaching <strong>and</strong> water quality,<br />
are under focus in this paper.<br />
Dynamic Regional Sector Model of Finnish Agriculture<br />
DREMFIA [Lehtonen 2001, 2004], which<br />
simulates economically rational production decisions,<br />
is used in this study in order to evaluate the<br />
likely impact of agricultural policy change on agricultural<br />
production. The model includes endogenous<br />
sector level investments which are important<br />
when evaluating long-term impacts of agricultural<br />
policy. Endogenous sector level investments, however,<br />
have been relatively rare in agricultural sector<br />
models [Heckelei et. al. 2001].<br />
Investments, while affecting technical change <strong>and</strong><br />
accumulation of knowledge <strong>and</strong> skills of farmers,<br />
have wide ranging consequences in the long run.<br />
Accumulation of knowledge <strong>and</strong> skills of farmers<br />
are important production factors in agriculture.<br />
However, future development may be dependent<br />
on initial conditions or on special dynamics of<br />
learning <strong>and</strong> investments. This increases the complexity<br />
<strong>and</strong> uncertainty of agricultural development.<br />
Under some simplifying assumptions, however,<br />
complex path dependent processes of technical<br />
<strong>and</strong> agricultural change can be modelled without<br />
making the model <strong>and</strong> its results intractable or<br />
too difficult to underst<strong>and</strong>.<br />
The changes in l<strong>and</strong> use, animal production <strong>and</strong> the<br />
use of production inputs, obtained from the<br />
DREMFIA model, are utilised in the nutrient<br />
transport model ICECREAM [Tattari et al. 2001]<br />
723
to evaluate field scale environmental impacts of<br />
different agricultural policies. Two catchments,<br />
which vary in their location <strong>and</strong> characteristics,<br />
have been selected for this study. This paper discusses<br />
the effort <strong>and</strong> the first experiences when<br />
connecting policy scenarios with impact modelling.<br />
2. METHODS<br />
2.1 THE SECTOR MODEL<br />
The DREMFIA model is dynamic recursive <strong>and</strong><br />
includes 17 production regions. The model provides<br />
effects of various agricultural policies on<br />
l<strong>and</strong> use, animal production, farm investments <strong>and</strong><br />
farmers’ income. Endogenous investments in different<br />
production techniques are modelled using<br />
the concept of technology diffusion. Since the<br />
endogenous technical change <strong>and</strong> explicit sector<br />
level investments are rare in st<strong>and</strong>ard economic<br />
models, one may expect the DREMFIA model to<br />
yield impacts of agricultural policy changes different<br />
from those reported by traditional economic<br />
models <strong>and</strong> reasoning. One can compare the<br />
DREMFIA results, for example, with the results of<br />
Jensen & Fr<strong>and</strong>sen [2003].<br />
In the DREMFIA model annual l<strong>and</strong> use <strong>and</strong> production<br />
decisions from 1995 till 2020 are simulated<br />
by an optimisation model which maximises producer<br />
<strong>and</strong> consumer surplus subject to regional<br />
product balance <strong>and</strong> resource (l<strong>and</strong>) constraints.<br />
Products <strong>and</strong> intermediate products may be transported<br />
between the regions. The optimisation<br />
model is a typical spatial price equilibrium model<br />
(see e.g. Cox & Chavas [2001]), except that no<br />
explicit supply functions are specified (i.e. supply<br />
is a primal specification), <strong>and</strong> foreign trade activities<br />
are included in DREMFIA. Armington assumption,<br />
which is a common feature in international<br />
agricultural trade models but less common in<br />
one-country sector models, is used. Imported <strong>and</strong><br />
domestic products are imperfect substitutes, i.e.<br />
endogenous prices of domestic <strong>and</strong> imported products<br />
are dependent. There are 18 different processed<br />
milk products <strong>and</strong> their regional processing<br />
activities in the model.<br />
Technical change <strong>and</strong> investments, which imply<br />
evolution of farm size distribution, are modelled as<br />
a process of technology diffusion. Investments are<br />
dependent on economic conditions such as interest<br />
rates, prices, support, production quotas <strong>and</strong> other<br />
policy measures <strong>and</strong> regulations imposed on farmers.<br />
The model of technology diffusion follows the<br />
main lines of Soete & Turner [1984].<br />
Two crucial aspects about diffusion <strong>and</strong> adaptation<br />
behaviour are included: first, the profitability<br />
of the new technique, <strong>and</strong> second, the risk <strong>and</strong><br />
uncertainty involved in adopting a new technique.<br />
The information about <strong>and</strong> likelihood of adoption<br />
of a new technique will grow as its use becomes<br />
wider spread.<br />
To cover the first point, likelihood of adoption of<br />
a new technique (f βα ) is made proportional to the<br />
fractional rate of profit increase in moving from<br />
technique α to technique β, i.e. f βα is proportional<br />
to (r β -r α )/r α where r α is the rate of return for technique<br />
α <strong>and</strong> r β is the rate of return for technique β.<br />
The second point is modelled by letting f βα be<br />
proportional to the ratio of the capital stock in the<br />
β technique (K β ) to the total capital stock K (in a<br />
certain agricultural production line), i.e. K β /K. The<br />
total investments to α technique, after some simplification,<br />
is<br />
I<br />
α<br />
= σ( Qα<br />
− wLα<br />
) + η(<br />
rα<br />
− r)<br />
Kα, . (1)<br />
where σ is the savings rate (proportion of economic<br />
surplus re-invested in agriculture), η is the<br />
farmers’ propensity to invest in alternative techniques,<br />
Q α is the total production linked revenue<br />
for technique α, w is a vector of input prices, L α is<br />
a vector of variable production factors of technique<br />
α, <strong>and</strong> r is the average rate of return on all<br />
techniques.<br />
The interpretation of this investment function is as<br />
follows. If η were zero then (1) would show that<br />
the investment in the α technique would come<br />
entirely from the investable surplus generated by<br />
the α technique. For η≠0 the investment in the α<br />
technique will be greater or less than the first<br />
term, depending on whether the rate of return on<br />
the α technique is greater than the average rate of<br />
return on all techniques (r). This seems reasonable.<br />
If a technique is highly profitable then it will<br />
tend to attract investment <strong>and</strong> conversely if it is<br />
relatively less profitable investment will decline.<br />
If there are no investments in α technique at some<br />
time period, the capital stock K α decreases at the<br />
depreciation rate. To summarise, the investment<br />
function (1) is an attempt to model the behaviour<br />
of farmers whose motivation to invest is greater<br />
profitability but nevertheless will not adopt the<br />
most profitable technique immediately, because of<br />
uncertainty <strong>and</strong> other retardation factors.<br />
The endogenous investments <strong>and</strong> technical change,<br />
as well as the recursive structure of DREMFIA<br />
model imply that the incentive for changes must<br />
affect production more than one year before significant<br />
changes in production may occur. Hence<br />
the DREMFIA model is designed to be used in<br />
724
evaluation of medium <strong>and</strong> long term effects of<br />
agricultural policy.<br />
Endogenous investments determine animal <strong>and</strong><br />
crop production volume in the long-term, but shortterm<br />
changes in crop production are constrained by<br />
flexibility constraints. The constraints are validated<br />
on the basis of average crop production data from<br />
1990-2002. Consumption trends are given exogenously.<br />
Fertilisation <strong>and</strong> yield levels are dependent<br />
on crop <strong>and</strong> fertiliser prices through crop yield<br />
functions. Feeding of animals may change in the<br />
short-term within certain bounds imposed by fixed<br />
production factors <strong>and</strong> animal biology provided<br />
that nutrition requirements are fulfilled. Specific<br />
production functions are used to model the dependency<br />
between the average milk yield of dairy cows<br />
<strong>and</strong> the amount of the grain based feed stuffs used<br />
in feeding. The yield of dairy cows responds to<br />
price changes of milk <strong>and</strong> feed stuffs. Time series<br />
of the model outputs include number of animals,<br />
areas of different crops <strong>and</strong> feeding of animals. The<br />
detailed presentation of model <strong>and</strong> its parameters<br />
can be found in Lehtonen [2001, 2004].<br />
The technology diffusion model has been validated<br />
to observed evolution of farm size distribution<br />
in 1995-2002. The overall model replicates<br />
very closely ex-post production in 1995-2002.<br />
2.2 THE CATCHMENT MODEL<br />
The ICECREAM model Tattari et al. [2001]; Bärlund<br />
<strong>and</strong> Tattari [2001], used for environmental<br />
impact assessment, is developed to simulate water,<br />
soil loss <strong>and</strong> phosphorus (P) <strong>and</strong> nitrogen (N)<br />
transport in the unsaturated soil of agricultural<br />
l<strong>and</strong>. The model simulates on field scale but the<br />
model results have been aggregated using typical<br />
soil-crop-slope combinations to small catchment<br />
scale to describe transport from agricultural l<strong>and</strong><br />
[Rekolainen et al., 2002]. Special attention in<br />
ICECREAM development has been paid on including<br />
management practices such as various tillage<br />
methods, fertilisation practices <strong>and</strong> l<strong>and</strong> use options<br />
like vegetative strips.<br />
To assess the environmental impacts of the agricultural<br />
policy scenarios, the results of the field scale<br />
simulations with ICECREAM are up-scaled. The<br />
relevant soil-crop-slope combinations form a simulation<br />
matrix of 6 soil types, 11 crop types <strong>and</strong> 9<br />
field slopes, i.e. 594 single simulations. These<br />
results are averages of annual sums of e.g. leached<br />
nitrate-N over the simulation period, here 10 years.<br />
The parameters to characterise soil properties <strong>and</strong><br />
crop development are equal in both simulated areas<br />
but the meteorological conditions are typical for<br />
each region. The response to the results from the<br />
DREMFIA model is gained weighing the ICE-<br />
CREAM matrix by the percentage of each soilcrop-slope<br />
combination in each catchment for each<br />
particular year.<br />
2.3 CATCHMENT AREAS<br />
The two catchments selected for this study vary in<br />
their location <strong>and</strong> characteristics. Yläneenjoki<br />
catchment is situated in the coastal plains of<br />
south-western Finl<strong>and</strong>. Its total area is larger (227<br />
km 2 ) but its field percentage smaller (35%) than<br />
of the Taipaleenjoki catchment (27 km 2 ; 50%),<br />
which is situated in eastern Finl<strong>and</strong>. The main line<br />
of production in Yläneenjoki is spring cereals<br />
whereas in Taipaleenjoki it is dairy production,<br />
which also explains the higher share of grassl<strong>and</strong><br />
in this area. Yläneenjoki region, on the other<br />
h<strong>and</strong>, is strong in pork <strong>and</strong> poultry production.<br />
Yläneenjoki is one of the relatively best grain<br />
production areas in Finl<strong>and</strong>. The yields of wheat<br />
<strong>and</strong> malting barley, in particular, are higher than<br />
average yields in Finl<strong>and</strong>. Farms having dairy <strong>and</strong><br />
beef cattle are of the same size in both Yläneenjoki<br />
<strong>and</strong> Taipaleenjoki areas, but pork <strong>and</strong> poultry<br />
farms in Yläneenjoki area are significantly larger<br />
<strong>and</strong> specialised than in Taipaleenjoki region.<br />
2.4 THE POLICY SCENARIOS<br />
Base-scenario follows Agenda 2000 reform<br />
(agreed in Berlin 1999; CEC [1999]) which is<br />
assumed to stay unchanged until 2020. It is assumed<br />
that producer price of milk would fall by<br />
15% in Finl<strong>and</strong> until 2008 from the average producer<br />
price of 1999-2001 (35,3 c/litre). Hence the<br />
producer price of milk would be 30,01 c/litre in<br />
2008-2015 in Base –scenario. LFA-, environmental<br />
<strong>and</strong> national support, mainly paid per hectare<br />
of different crops, are assumed to stay at 2003<br />
year level in 2004-2015.<br />
Mid Term Review (MTR) –scenario, ranging up<br />
to year 2020, is a combination of EU Commission’s<br />
agricultural policy reform proposal [CEC<br />
2003] presented in January 22 2003, <strong>and</strong> the CAP<br />
reform agreed in June 2003. CAP-support, based<br />
on 2000-2002 historical production levels, is paid<br />
in a single farm payment each year.<br />
Producer price of milk falls by 28% in the EU until<br />
2009. In Finl<strong>and</strong> such a change means that the<br />
average producer price of 1999-2001 (35,3 c/litre)<br />
reduces to 25,4 c/litre in 2009. The milk price cut<br />
is compensated by payments per quota ton. The<br />
payment goes up to 41 euros per ton (prior 5%<br />
modulation) until 2008.<br />
725
An increase in LFA support is assumed. The increase<br />
of LFA support would be directed for milk<br />
<strong>and</strong> cattle farms. The support rate per bovine animal<br />
unit would increase linearly up to 300 euros<br />
per bovine animal until 2009. Overall this would<br />
mean a 50% increase in the total LFA support.<br />
National supports, paid per hectare of certain special<br />
crops <strong>and</strong> per animal, are kept at base scenario<br />
level.<br />
Integrated rural <strong>and</strong> environmental policy<br />
(INT) –scenario is built on MTR-scenario in such a<br />
way that environmental concerns <strong>and</strong> labour in<br />
rural areas are of particular emphasis. This means<br />
that support for grass area is increased, <strong>and</strong> labour<br />
is supported by paying 3 euros per hour of work for<br />
farms which have bovine animals. CAP extensification<br />
premium is not de-coupled from production.<br />
LFA support is kept at the base scenario level. EU<br />
price level of agricultural products would be the<br />
same as in MTR scenario.<br />
Free trade –scenario <strong>and</strong> full scale agricultural<br />
trade liberalisation (LIB) includes the most drastic<br />
changes. All agricultural support is transformed<br />
into an area based flat rate support which is the<br />
same for all crops <strong>and</strong> is de-coupled from production.<br />
This transformation would be complete in<br />
2010. The total sum of agricultural support is decreased<br />
by 10% by year 2014. Prices of agricultural<br />
products in the EU are 5-20 % lower than in MTR<br />
<strong>and</strong> INT –scenarios.<br />
3. RESULTS AND DISCUSSION<br />
Let us briefly discuss the changes in production in<br />
the whole country level because that provides a<br />
major explanation for the changes in production in<br />
Yläneenjoki <strong>and</strong> Taipaleenjoki catchments. Production<br />
in both areas, reported in tables 1 <strong>and</strong> 2, is<br />
influenced by production in other areas because of<br />
balance between total supply <strong>and</strong> dem<strong>and</strong>.<br />
Milk production has a strong effect on l<strong>and</strong> use in<br />
Finl<strong>and</strong>. Dairy capital decreases drastically in INT<strong>and</strong><br />
LIB-scenarios. The rate of return on investment<br />
decreases well below the general interest rate<br />
(assumed 5% until 2020), which implies drastic<br />
downturn in investments. In INT –scenario the<br />
labour support, which should reinforce supply<br />
ceteris paribus, makes investments in large production<br />
units relatively less profitable <strong>and</strong> hence inhibits<br />
the development of competitive farm structures<br />
in the long-term. The decrease in investments <strong>and</strong><br />
dairy capital is less drastic in MTR scenario because<br />
of increased LFA support for bovine animals.<br />
In any case supply of milk will gradually<br />
decrease in MTR scenario due to lower milk prices<br />
<strong>and</strong> de-coupled CAP payments. The decrease of<br />
beef supply is relatively larger than milk supply<br />
due to increasing milk yield <strong>and</strong> reducing dairy<br />
herd.<br />
Agricultural policy changes have little effect on<br />
pig <strong>and</strong> poultry production, since relatively small<br />
changes are expected for pork <strong>and</strong> poultry sectors.<br />
The development of dairy production in both<br />
Yläneenjoki <strong>and</strong> Taipaleenjoki areas are characterised<br />
by the same kind of development <strong>and</strong> drivers<br />
of development as in the whole country: dairy<br />
production decreases significantly in MTR-, INT<strong>and</strong><br />
in LIB-scenarios because of low profitability<br />
of investments. Small dairy farms allocate l<strong>and</strong> to<br />
set-aside instead of investing in dairy production.<br />
Suckler cow numbers, however, increase in base<br />
scenario in both areas, especially in Taipaleenjoki<br />
area. This is due to considerable national support<br />
for grass area while beef prices <strong>and</strong> production<br />
linked supports keep up production. Grass area<br />
increases <strong>and</strong> grain area decreases significantly<br />
(from the 1995 level) in Taipaleenjoki region<br />
already in the base scenario until 2015. This is<br />
because Taipaleenjoki region becomes even more<br />
dominated by dairy production <strong>and</strong> grain production<br />
concentrates to more feasible regions.<br />
In the MTR scenario suckler cow numbers at<br />
whole country level increase only slightly because<br />
increased LFA -support is outweighed by decoupled<br />
CAP support. However, in Taipaleenjoki<br />
region the number of suckler cows increases even<br />
in the MTR scenario because there is a strong<br />
incentive for extensive grass cultivation. The<br />
intensive dairy production is replaced by very<br />
extensive grass cultivation. This is a rational consequence<br />
of low milk price <strong>and</strong> decoupled CAP<br />
payments. In the LIB scenario the dairy herd declines<br />
drastically <strong>and</strong> set aside becomes the relatively<br />
most profitable use of l<strong>and</strong>.<br />
In Yläneenjoki region the milk production reduces<br />
only slightly in the MTR scenario because of lower<br />
feed costs compared to Taipaleenjoki area. Since<br />
Yläneenjoki area is one of the relatively best grain<br />
production areas in Finl<strong>and</strong>, incentive for extensive<br />
grass cultivation is not as strong as in Taipaleenjoki<br />
area. In the MTR scenario, however, grain <strong>and</strong><br />
grass areas decrease slightly in Yläneenjoki region<br />
while set aside areas increase up to 11% of the total<br />
area. In INT <strong>and</strong> LIB scenarios, where LFA support<br />
is lower than in MTR scenario, set aside areas<br />
are high in 2015. It is remarkable that even if the<br />
total grain area in Finl<strong>and</strong> decreases drastically in<br />
the LIB scenario, grain area does not change much<br />
in the Yläneenjoki region.<br />
726
Table 1 Development of the number of animals [1000 heads] according to the 2001 survey <strong>and</strong> estimated by<br />
DREMFIA for the four scenarios BAS (Agenda 2000), MTR (Mid Term Review), INT (Integrated Policy)<br />
<strong>and</strong> LIB (Free Trade) in 2015.<br />
Yläneenjoki<br />
Taipaleenjoki<br />
2001 BAS MTR INT LIB 2001 BAS MTR INT LIB<br />
Dairy cows 1,56 1,30 1,13 0,74 0,58 3,0 2,57 1,41 1,13 0,70<br />
Suckler 0,40 0,46 0,17 0,11 0,09 0,07 0,25 0,14 0,25 0,02<br />
cows<br />
Sows 6,65 3,15 4,26 3,47 1,91 0,18 0,04 0,04 0,04 0,05<br />
Pigs 39,1 21,85 29,55 24,12 13,27 1,08 0,29 0,29 0,29 0,35<br />
Hens 246,8 171,5 271,2 228,9 117,5 0,85 0,22 0,22 0,22 1,07<br />
Other<br />
poultry<br />
519,7 1022,6 704,5 773,3 346,3 0 0 0 0 0<br />
Table 2 Distribution of crops [% of cultivated area] simulated by ICECREAM according to the 1995 survey<br />
<strong>and</strong> estimated by DREMFIA for the four scenarios BAS (Agenda 2000), MTR (Mid Term Review), INT<br />
(Integrated Policy) <strong>and</strong> LIB (Free Trade) in 2015.<br />
Yläneenjoki<br />
Taipaleenjoki<br />
1995 BAS MTR INT LIB 1995 BAS MTR INT LIB<br />
oats 17 22 27 27 29 27 31 13 9.8 1.9<br />
barley 37 57 45 40 39 14 0.68 0.37 0.11 0.20<br />
s_wheat 11 2.4 2.8 2.3 3.6 1.9 0.013 0.013 0.013 0.013<br />
oilseeds 4.1 1.0 1.4 0.97 1.8 0.95 0.0063 0.0063 0.0063 0.0063<br />
w_wheat 4.6 1.1 1.2 1.0 1.6 0 0 0 0 0<br />
rye 4.2 0.97 1.1 0.93 1.5 1.8 0.012 0.012 0.012 0.012<br />
s_beet 2.3 0.54 0.62 0.51 0.80 0 0 0 0 0<br />
potato 1.4 0.31 0.36 0.30 0.47 0.72 0.0048 0.0048 0.0048 0.0048<br />
grass 7.7 6.4 4.8 3.5 4.8 45 64 82 85 39<br />
g_fallow 8.3 4.3 11 19 14 3.9 4.4 4.4 4.4 58<br />
b_fallow 1.0 0.23 0.27 0.22 0.34 3.4 0.023 0.023 0.023 0.27<br />
s_wheat: spring wheat; w_wheat: winter wheat; s_beet: sugar beet; g_fallow: green fallow; b_fallow: bare fallow<br />
Figure 1 Simulated change in average annual sum of soluble (DPr, a) <strong>and</strong> sediment bound (PP, b) P in surface<br />
runoff <strong>and</strong> nitrate-N in percolation from root zone (percNO 3 , c) from arable l<strong>and</strong> in 2015 relative to the<br />
situation in 1995 in Yläneenjoki (YLA) <strong>and</strong> Taipaleenjoki (TAI) catchments.<br />
a) 40<br />
b) 40<br />
c) 40<br />
DPr [%]<br />
20<br />
0<br />
-20<br />
-40<br />
PP [%]<br />
20<br />
0<br />
-20<br />
-40<br />
percNO3 [%]<br />
20<br />
0<br />
-20<br />
-40<br />
-60<br />
BAS MTR INT LIB<br />
-60<br />
BAS MTR INT LIB<br />
-60<br />
BAS MTR INT LIB<br />
YLA<br />
TAI<br />
According to ICECREAM model, the change in<br />
DPr <strong>and</strong> PP due to the base scenario is close to no<br />
change in Yläneenjoki region. All other scenarios<br />
would lead to a small reduction of both variables.<br />
For DPr this is due to reduction of grass <strong>and</strong><br />
increase of green fallow <strong>and</strong> for PP the main<br />
reason is the reduction of bare fallow <strong>and</strong> winter<br />
cereals in the catchment, both l<strong>and</strong> use types<br />
having relatively high PP loss values. Grass is the<br />
only crop receiving surface applied fertilisation<br />
<strong>and</strong> thus a decrease in the grass area reduces DPr<br />
losses effectively The rather high reduction of<br />
percNO 3 can be explained by a smaller area of<br />
oilseeds <strong>and</strong> winter cereals. Both crop types have<br />
rather high N fertilisation compared to simulated<br />
crop uptake, which explains losses in percolated<br />
water.<br />
In Taipaleenjoki region the relative change in P<br />
leaching is higher than in Yläneenjoki <strong>and</strong> for DPr<br />
an increase is indicated for all scenarios except<br />
LIB. For DPr the main reason would be the larger<br />
area under grass in 2015 compared to 1995. The<br />
DPr decrease under the LIB scenario is explained<br />
by the extremely high increase in green fallow<br />
area. The change in grass <strong>and</strong> green fallow area<br />
explains also the reduction of PP for all scenarios.<br />
The results for percNO 3 for MTR <strong>and</strong> INT scenarios<br />
can be interpreted as no change. The reduction<br />
for the other scenarios is a combination of an increase<br />
in the area of oats (BAS) <strong>and</strong> green fallow<br />
with very low nitrate leaching <strong>and</strong> reduced area of<br />
727
oilseeds <strong>and</strong> winter cereals with high nitrate leaching<br />
potential.<br />
The Yläneenjoki area is more susceptible to eutrophication<br />
due to natural conditions <strong>and</strong> loading<br />
history but it has to be investigated what the predicted<br />
change would mean in Taipaleenjoki conditions<br />
over a longer time period. Therefore, future<br />
analysis on the effect of predicted actual nutrient<br />
load change on variables describing eutrophication<br />
(e.g. Secchi depth) is needed.<br />
4. CONCLUSIONS<br />
Large reductions in milk price <strong>and</strong> a simultaneous<br />
de-coupling of CAP payments are likely to cut<br />
dairy investments considerably. This would cease<br />
the ongoing structural change, farm size growth<br />
<strong>and</strong> production specialisation on Finnish dairy<br />
farms 1 . Instead, many dairy farms would refrain<br />
from investment <strong>and</strong> allocate l<strong>and</strong> to set-aside.<br />
Milk <strong>and</strong> especially beef production volumes<br />
would decrease considerably in the long-term. Also<br />
grain area would decrease slightly. De-coupling<br />
CAP-support, however, would have only marginal<br />
effects on pork <strong>and</strong> poultry production.<br />
The coupled use of the economic model DREM-<br />
FIA <strong>and</strong> the environmental model ICECREAM<br />
enabled to test the effect of four different agricultural<br />
policy scenarios on nutrient leaching in two<br />
Finnish catchments with varying characteristics.<br />
The relative change in nutrient leaching was dependent<br />
on the policy scenario applied, the nutrient<br />
leaching variable studied <strong>and</strong> on the catchment<br />
chosen. In the Yläneenjoki catchment in southwestern<br />
Finl<strong>and</strong> a reduction of all variables presented<br />
would be expected, whereas in Taipaleenjoki<br />
in eastern Finl<strong>and</strong> especially soluble P in surface<br />
runoff might be increasing even if product<br />
prices were reduced <strong>and</strong> subsidies were de-coupled<br />
from production. This challenges a common view<br />
that lower prices <strong>and</strong> decoupled subsidies always<br />
imply less environmental harm. In order to utilise<br />
the results in policy dialogue, further refinement of<br />
the method is needed in order to quantify the effect<br />
in each particular area <strong>and</strong> to link the nutrient load<br />
from agriculture to the eutrophication potential.<br />
5. ACKNOWLEDGEMENTS<br />
1 Such effects are not taken into account in st<strong>and</strong>ard<br />
economic tools. For example, Jensen &<br />
Fr<strong>and</strong>sen 2003 report no change in Finnish dairy<br />
production, <strong>and</strong> a remarkable increase in beef<br />
production, due to 2003 CAP reform.<br />
The financial support of the SUSAGFU project<br />
through the Academy of Finl<strong>and</strong> (contract 76724)<br />
is gratefully acknowledged.<br />
6. REFERENCES<br />
Bärlund I. <strong>and</strong> Tattari S., Ranking of parameters<br />
on the basis of their contribution to<br />
model uncertainty. Ecol. Modell., 142,<br />
11-23, 2001.<br />
CEC, http://europa.eu.int/comm/<br />
agenda2000/ index_en.htm, 1999.<br />
CEC, http://europa.eu.int/comm/<br />
agriculture/ mtr/index_en.htm, 2003.<br />
Cox T.L. <strong>and</strong> Chavas J.-P., An interregional<br />
analysis of price discrimination <strong>and</strong> domestic<br />
policy reform in the US dairy sector.<br />
Am. J. Agric. Econ., 83(1), 89-106,<br />
2001.<br />
Heckelei, T., Witzke, P. <strong>and</strong> Henrichsmeyer, W.,<br />
Agricultural Sector <strong>Modelling</strong> <strong>and</strong> Policy<br />
Information Systems. In Proceedings<br />
of the 65 th European Seminar of the<br />
European Association of the Agricultural<br />
Economics (EAAE), March 29-31, Bonn,<br />
Germany 2000, 2001.<br />
Jensen, H.G. <strong>and</strong> Fr<strong>and</strong>sen, S.E., Impacts of the<br />
Eastern European Accession <strong>and</strong> the<br />
2003-reform of the CAP. FOI Working<br />
paper no. 11/2002. Danish Research Institute<br />
of Food Economics, 2003.<br />
Lehtonen H., Principles, structure <strong>and</strong> application<br />
of dynamic regional sector model of<br />
Finnish agriculture. PhD thesis, Systems<br />
Analysis Laboratory, Helsinki University<br />
of Technology. Agrifood Research Finl<strong>and</strong>,<br />
Economic Research (MTTL) Publ.<br />
98. Helsinki, 2001<br />
Lehtonen, H., Impacts of de-coupling agricultural<br />
support on dairy investments <strong>and</strong> milk<br />
production volume in Finl<strong>and</strong>. Forthcoming<br />
in Acta Agriculturae Sc<strong>and</strong>inavica,<br />
Section C: Food Economics,<br />
2004.<br />
Rekolainen S., Salt C.A., Bärlund I., Tattari S.<br />
<strong>and</strong> Culligan-Dunsmore M., Impacts of<br />
the management of radioactively contaminated<br />
l<strong>and</strong> on soil <strong>and</strong> phosphorus<br />
losses in Finl<strong>and</strong> <strong>and</strong> Scotl<strong>and</strong>. Water,<br />
Air, Soil Pollut., 139, 115-136, 2002.<br />
Soete, L. & Turner, R, Technology diffusion <strong>and</strong><br />
the rate of technical change. The Economic<br />
Journal. pp. 612-623. September<br />
1984.<br />
Tattari S., Bärlund I., Rekolainen S., Posch M.,<br />
Siimes K., Tuhkanen H.-R. <strong>and</strong> Yli-<br />
Halla M., <strong>Modelling</strong> sediment yield <strong>and</strong><br />
phosphorus transport in Finnish clayey<br />
728
soils. Trans. ASAE, 44(2), 297-307,<br />
2001.<br />
729
Simulation of Water <strong>and</strong> Carbon Fluxes in Agro- <strong>and</strong><br />
forest Ecosystems at the Regional Scale<br />
J. Post, V. Krysanova & F. Suckow<br />
Potsdam Institute for Climate Impact Research, P.O. Box 601 203, Telegrafenberg, Potsdam, Germany<br />
post@pik-potsdam.de<br />
Abstract: To investigate effects of different l<strong>and</strong> use management practices on carbon fluxes at the regional<br />
scale we developed an integrated model by coupling an ecohydrological river basin model SWIM (Soil <strong>and</strong><br />
Water Integrated Model) <strong>and</strong> a soil organic matter model SCN (Soil-Carbon-Nitrogen model). The latter is a<br />
submodel of the forest growth model 4C. The extended integrated model combines hydrological processes,<br />
crop <strong>and</strong> vegetation growth, carbon, nitrogen, phosphorus cycles <strong>and</strong> soil organic matter turnover. It is based<br />
on a three level spatial disaggregation scheme (basin, subbasin <strong>and</strong> hydrotopes), whereas a hydrotope is a set<br />
of elementary units in the subbasin with a uniform l<strong>and</strong> use <strong>and</strong> soil type. The direct connection to l<strong>and</strong> use,<br />
soil <strong>and</strong> climate data provides a possibility to use the model for analyses of climate change <strong>and</strong> l<strong>and</strong> use<br />
change impacts on hydrology <strong>and</strong> soil organic matter turnover. Aim of this study is to test the model<br />
performance <strong>and</strong> its capability to simulate carbon pools <strong>and</strong> fluxes in right magnitude <strong>and</strong> temporal<br />
behaviour at the regional scale. As a first step, the model was parameterised <strong>and</strong> validated for conditions in<br />
East Germany, incorporating values known from literature <strong>and</strong> regionally available times series of carbon<br />
pools <strong>and</strong> fluxes. This provides verification of carbon pools <strong>and</strong> fluxes in the l<strong>and</strong>scape <strong>and</strong> verifies the<br />
correct representation of the environmental processes therein. Based on this, different l<strong>and</strong> management<br />
strategies (e.g. soil cultivation techniques, crop residue returns) <strong>and</strong> l<strong>and</strong> use change options (e.g. conversion<br />
of agricultural areas to forest or to set-aside areas) can be simulated to assess the behaviour of water <strong>and</strong><br />
carbon fluxes as well as carbon sequestration options.<br />
Keywords: Ecohydrological modelling, soil carbon, soil nitrogen, soil organic matter turnover, global change<br />
1. INTRODUCTION<br />
The major anthropogenic input of CO 2 to the<br />
atmosphere is attributed to fossil fuel combustion,<br />
cement manufacturing <strong>and</strong> l<strong>and</strong> use change. The<br />
latter involves deforestation, biomass burning,<br />
draining of wetl<strong>and</strong>s, plowing, use of fertilisers<br />
<strong>and</strong> manure <strong>and</strong> other agricultural practices. This<br />
source is estimated to be a large global carbon<br />
flux of 1 - 2 * 10 15 g C yr -1 [Houghton, 1996].<br />
L<strong>and</strong> management practises (e.g. cultivation, l<strong>and</strong><br />
conversion) are seen to influence the rate <strong>and</strong><br />
magnitude of this CO 2 flux.<br />
To take these effects into considerations it is<br />
necessary to develop integrated ecological models<br />
which cover the main processes ruling the<br />
turnover of organic matter in the environment <strong>and</strong><br />
hence the release of CO 2 . Hereby soil processes of<br />
organic matter turnover play an important role.<br />
The quantity of soil organic matter (SOM) is<br />
dependent on the balance between litter (dead<br />
plant biomass) production <strong>and</strong> the rate of litter<br />
<strong>and</strong> SOM decomposition. Further on the products<br />
of primary productivity are entering the soil<br />
column containing a mixture of dead plant <strong>and</strong><br />
animal material derived substances with variable<br />
physical <strong>and</strong> chemical properties. These materials<br />
are subject to decomposition by the macro- <strong>and</strong><br />
micro-organisms in the soil. Together with the<br />
decomposition <strong>and</strong> mineralisation of existing<br />
organic materials the heterotrophic soil respiration<br />
produces CO 2 . These processes are influenced by<br />
environmental conditions like soil temperature,<br />
soil moisture <strong>and</strong> soil acidity status.<br />
In the present work an extension of the ecohydrological<br />
river basin model SWIM (Soil <strong>and</strong><br />
Water Integrated Model, Krysanova et al. [1998])<br />
730
y a new module for the turnover of soil organic<br />
matter <strong>and</strong> soil carbon <strong>and</strong> nitrogen dynamics<br />
(SCN- Soil-Carbon-Nitrogen model, a submodel<br />
of the forest growth model 4C, Lasch et al.<br />
[2002], Grote et al. [1999]) is presented. The<br />
advantage of this approach is a possibility to<br />
combine hydrological, soil carbon <strong>and</strong> soil<br />
nitrogen processes in both vertical <strong>and</strong> lateral<br />
dimensions for agro- <strong>and</strong> forest ecosystems in<br />
river basins.<br />
As a prerequisite to perform l<strong>and</strong> use change <strong>and</strong><br />
l<strong>and</strong> management impacts on SOM dynamics at<br />
the regional scale, detailed model verification for<br />
the main ecosystems has to be performed.<br />
2. METHODS AND DATA<br />
2.1. The ecohydrological model SWIM<br />
SWIM (Soil <strong>and</strong> Water Integrated Model,<br />
Krysanova et al. [1998] is a continuous-time,<br />
spatially distributed model. SWIM works on a<br />
daily timestep <strong>and</strong> integrates hydrology,<br />
vegetation, erosion <strong>and</strong> nutrients at the river basin<br />
scale. The spatial aggregation units are subbasins,<br />
which are delineated from digital elevation data.<br />
The subbasins are further disaggregated into so<br />
called hydrotopes, hydrologically homogenous<br />
areas. The hydrotopes are defined by uniform<br />
combinations of subbasin, l<strong>and</strong> use <strong>and</strong> soil type<br />
[Krysanova et al., 2000]. The model is connected<br />
to meteorological, l<strong>and</strong> use, soil <strong>and</strong> agricultural<br />
management data. For detailed process<br />
descriptions, validation studies <strong>and</strong> data<br />
requirements it is referred to publications by<br />
Krysanova et al. [1998, 2000]. Following, the<br />
relevant processes for the presented work are<br />
described briefly.<br />
2.1.1 Hydrological cycle<br />
The hydrology module is based on the water<br />
balance equation, taking into account<br />
precipitation, evapotranspiration, percolation,<br />
surface runoff <strong>and</strong> subsurface runoff for the soil<br />
column which is subdivided into several layers.<br />
The water balance for the shallow aquifer<br />
includes ground water recharge, capillary rise to<br />
the soil profile, lateral flow, <strong>and</strong> percolation to<br />
deep aquifer [Krysanova et al., 1998].<br />
2.1.2 Soil temperature<br />
Soil temperature is calculated on a daily basis at<br />
the center of each soil layer. The calculation is<br />
based on an empirical relationship between daily<br />
average, minimum <strong>and</strong> maximum air temperature<br />
<strong>and</strong> a damping factor for soil depth. The effect of<br />
current weather conditions <strong>and</strong> l<strong>and</strong> cover (snow,<br />
above ground biomass) are considered<br />
[Krysanova et al., 2000].<br />
2.1.3 Vegetation growth<br />
Vegetation growth is simulated separately for<br />
annual (crops) <strong>and</strong> perennial (forest) plant types.<br />
For crop growth a simplified EPIC approach<br />
[Williams et al., 1984] is used for simulating all<br />
crops considered (wheat, barley, corn, potatoes,<br />
alfalfa <strong>and</strong> others) using unique parameter values<br />
for each crop. The simplified EPIC approach is<br />
based on growth dynamics for annual plants. To<br />
describe perennial, <strong>and</strong> especially forest growth<br />
dynamics, a different approach was adopted<br />
recently which is described by Wattenbach et al.<br />
[2004]. The central element of the approach is the<br />
use of an allometric relation for the ratio of leaf<br />
biomass to total biomass that is given by an age<br />
dependent exponential function [Bugmann, 1994].<br />
The forest growth then is based on a robust<br />
computation of the temporal LAI (leaf area index)<br />
dynamics [Bugmann, 1994]. Based on the forest<br />
age <strong>and</strong> forest st<strong>and</strong> density dependent LAI the<br />
biomass energy ratio is used to calculate the daily<br />
biomass increase. The model also considers<br />
phenology, age-dependent mortality <strong>and</strong> simple<br />
forest management practices.<br />
The vegetation growth dynamics <strong>and</strong> the related<br />
dead biomass production at the end of the<br />
growing season deliver the amount of litter (dead<br />
above- <strong>and</strong> below-ground biomass) entering the<br />
litter <strong>and</strong> soil layers.<br />
2.2 The soil Carbon <strong>and</strong> Nitrogen Model SCN<br />
The carbon <strong>and</strong> nitrogen cycle module is based on<br />
the tight relationship between the soil <strong>and</strong> the<br />
vegetation. On the one h<strong>and</strong> an input exists into<br />
the soil by addition of organic material through<br />
accumulating litter, dead fine roots <strong>and</strong> organic<br />
fertilizer, <strong>and</strong> on the other h<strong>and</strong> there is a<br />
withdrawal from the soil of water <strong>and</strong> nitrogen by<br />
the vegetation, release of CO 2 into the atmosphere<br />
<strong>and</strong> export of inorganic nitrogen by soil water<br />
flows (e.g. percolation into the groundwater,<br />
lateral flow processes).<br />
To describe the carbon <strong>and</strong> nitrogen budget<br />
organic matter is differentiated into Active<br />
Organic Matter (AOM) as humus pool <strong>and</strong><br />
Primary Organic Matter (POM) as litter pool. The<br />
731
latter is separated into 5 fractions for each<br />
vegetation <strong>and</strong> crop type. For forest types as<br />
example, the dead plant materials are cut into<br />
stems, twigs <strong>and</strong> branches, foliage, fine roots <strong>and</strong><br />
coarse roots. The fine <strong>and</strong> coarse roots are further<br />
distributed into the soil layers according to<br />
rooting depth <strong>and</strong> a root mass allocation. For all<br />
pools of active <strong>and</strong> primary organic matter the<br />
carbon <strong>and</strong> nitrogen content is considered.<br />
The carbon <strong>and</strong> nitrogen turnover into different<br />
stages (pools) can be pictured as a first order<br />
reaction [Chertov <strong>and</strong> Komarov, 1997; Franko,<br />
1990; Parton et al., 1987, Grote et al., 1999] as<br />
shown in Figure 1. The processes are controlled<br />
by matter specific reaction coefficients.<br />
Soil processes<br />
Soil<br />
Water<br />
Soil C, N<br />
turnover<br />
Soil<br />
Temp.<br />
Vegetation Processes<br />
Vegetation<br />
growth<br />
Litter<br />
production<br />
C<br />
in POM<br />
Qpom<br />
N<br />
in POM<br />
L<strong>and</strong> use<br />
Hydrological Processes<br />
lateral <strong>and</strong> vertical<br />
L<strong>and</strong><br />
management<br />
Uptake of<br />
water &<br />
nutrients<br />
k 2 k 1 C in<br />
AOM<br />
Qaom<br />
k * 1<br />
N in<br />
AOM<br />
k * 2<br />
k aom<br />
k aom<br />
mineral C<br />
percolation<br />
of water<br />
Atmosphere<br />
Meteorology<br />
Deposition<br />
C O 2<br />
k nit<br />
NH 4 NO 3<br />
leached<br />
N<br />
Figure 1. Illustration of main processes of the<br />
SWIM-SCN model.<br />
The dominant process is the carbon<br />
mineralisation, which provides the energy for the<br />
whole turnover of the organic matter. Following<br />
the above concept, the basic turnover in each<br />
layer is described as a reaction of the first order.<br />
The carbon change in the primary organic matter<br />
C POM is controlled by the reaction coefficient k POM<br />
= k 1 +k 2 (see Figure 1). The transformation of<br />
primary organic matter C POM to active organic<br />
matter C AOM is controlled by a synthesis<br />
coefficient k syn , which is specific to the litter type<br />
(plant type <strong>and</strong> litter fraction) whereas k 1 = k syn ⋅<br />
k pom . The turnover of carbon in active organic<br />
matter is made up from the synthesised portion<br />
<strong>and</strong> the carbon used in the process of<br />
mineralisation.<br />
How much nitrogen is absorbed into the active<br />
organic matter <strong>and</strong> what proportion is mineralised<br />
depends on the C/N ratio of both organic fractions<br />
<strong>and</strong> on the carbon used in the synthesis of the<br />
active organic matter. The change in nitrogen in<br />
the active organic matter takes place in a similar<br />
way to the turnover of carbon, whereas the C/N<br />
ratios of both organic fractions Q POM <strong>and</strong> Q AOM<br />
modify the synthesis coefficient k syn to k * syn<br />
[Kartschall et al., 1990].<br />
Heterotrophic (substrate induced) soil respiration<br />
is calculated through the decay of C POM <strong>and</strong> C AOM<br />
pools per day. Root respiration therefore is not<br />
considered.<br />
In addition, changes of nitrogen in the pools of<br />
ammonia N NH4 <strong>and</strong> nitrate N NO3 are considered.<br />
The reduction functions for mineralisation <strong>and</strong><br />
nitrification r min <strong>and</strong> r nit , respectively, show the<br />
effect of soil water content <strong>and</strong> soil temperature<br />
on these processes [Franko, 1990; Kartschall et<br />
al., 1990]. The mineralisation is inhibited, if the<br />
water content decreases below half of the<br />
saturated water content. The reduction of<br />
nitrification by drought is similar to the reduction<br />
of mineralisation, despite the decrease in<br />
nitrification under conditions of a very high water<br />
content, which results from the deficiency of<br />
oxygen.<br />
The influence of soil temperature on the<br />
mineralisation is described by van't Hoffs rule<br />
[Van’t Hoff, 1884]. The temperature depending<br />
reduction function for nitrification is analogous to<br />
that for mineralisation.<br />
2.3 Parameterisation<br />
The model parameterisation was done under the<br />
premise to simulate soil organic matter <strong>and</strong><br />
relevant processes for eastern German conditions<br />
with a special focus on the lowl<strong>and</strong>s. Therefore<br />
related environmental studies in the region <strong>and</strong><br />
literature were used for parameterisation. The<br />
reaction coefficients k pom <strong>and</strong> k syn have to be<br />
determined for each plant species (forest types,<br />
crop types) <strong>and</strong> primary organic matter fractions<br />
(fine roots, foliage, etc.). Determination of these<br />
coefficients is mainly done either by field<br />
experiments (litter bag experiments) or under<br />
laboratory conditions (incubation experiments).<br />
Main source for these parameters for the region<br />
under study are for agricultural plants<br />
investigations by Klimanek [1990 a, b] <strong>and</strong><br />
Franko [1990]. For forest types information can<br />
be found in Bergmann [1999] <strong>and</strong> Berg & Staaf<br />
[1980].<br />
The coefficients k aom <strong>and</strong> k nit are soil <strong>and</strong> l<strong>and</strong><br />
cover specific <strong>and</strong> can be found e.g. in Franko<br />
[1990] <strong>and</strong> Bergmann et al. [1999].<br />
732
2.4 Verification sites description<br />
Two field sites for verification were chosen for<br />
this presentation. For forest sites the Level II<br />
monitoring plot Kienhorst with Scots pine (Pinus<br />
sylvestris l.) investigated in the framework of the<br />
Pan- European Programme for intensive <strong>and</strong><br />
continuous monitoring of forest ecosystems<br />
(Level II, http://www.fimci.nl) was used. For<br />
agricultural sites data from the experimental field<br />
for cultivation of energy crops at the Leibnitz –<br />
Institute of Agricultural Engineering Bornim ATB<br />
[Hellebr<strong>and</strong> et al., 2003] were applied. Both sites<br />
are in the state of Br<strong>and</strong>enburg (east Germany),<br />
on s<strong>and</strong>y to loamy soils <strong>and</strong> sub-continental dry<br />
climate (long term annual precipitation average of<br />
600 mm).<br />
3. RESULTS AND DISCUSSION<br />
The model was verified for the main processes of<br />
soil organic matter turnover described above. For<br />
the verification sites meteorological data, soil<br />
parameterisation <strong>and</strong> management practices of the<br />
sites have been used to ensure that environmental<br />
conditions are represented.<br />
At first simulations of soil moisture conditions<br />
<strong>and</strong> soil temperature were compared with<br />
observed data. These are the deciding<br />
environmental factors influencing decomposition<br />
of organic material <strong>and</strong> humus mineralisation<br />
accounted for by the model. For the forest site soil<br />
moisture in three soil depths <strong>and</strong> soil temperature<br />
in two soil depths were compared with simulation<br />
results for a period of 5 years (1997 – 2001). For<br />
the agricultural site only soil temperature<br />
measurements at 20 cm soil depth for a period of<br />
one year (02/1999 – 02/2000, Hellebr<strong>and</strong> et al.,<br />
[2003]) were available. Both sites showed a good<br />
agreement between observation <strong>and</strong> simulation.<br />
The vegetation growth <strong>and</strong> the formation of litter<br />
determine the input of primary organic matter.<br />
For the forest site, vegetation growth was verified<br />
for LAI development, biomass increase, total<br />
biomass <strong>and</strong> litter, which is described in detail by<br />
Wattenbach et al. [2004]. The measured <strong>and</strong><br />
simulated litter production for a period of 4 years<br />
(1996 – 1999) show an adequate accordance. For<br />
the agricultural site, yearly harvest yields for<br />
Triticale <strong>and</strong> Rye (3 years, 1995 – 1997) were<br />
measured. For Triticale a 3-year average dry<br />
matter yield of 9 t dm ha -1 a -1 on fertilized sites (150<br />
kg N ha -1 a -1 ) was obtained. Rye dry matter yield<br />
was measured with 10.6 t dm ha -1 a -1 on fertilized<br />
sites. Total dry matter values for the agricultural<br />
site represent the total aboveground biomass<br />
harvested. For a comparison to model results<br />
these values have to be multiplied by the harvest<br />
indices for Triticale (0.42) <strong>and</strong> Rye (0.40).<br />
Simulated harvest yields for the respective period<br />
deliver for Triticale an average yield of 3.7 t dm ha -<br />
1<br />
a -1 <strong>and</strong> for Rye 3.8 t dm ha -1 a -1 with<br />
corresponding measured values of 3.8 <strong>and</strong> 4.2 t dm<br />
ha -1 a -1 respectively. Hence, for the agricultural<br />
site vegetation growth <strong>and</strong> litter production is well<br />
represented through the model.<br />
For the decomposition of primary organic matter<br />
no measurements were done at the agricultural<br />
site. For agricultural species decomposition<br />
experiments in the field with litterbags were taken<br />
from literature. In Figure 3 data from Henriksen<br />
<strong>and</strong> Brel<strong>and</strong> [1999] were adopted to show the<br />
ability of the model to represent decomposition of<br />
primary organic matter for barley straw <strong>and</strong> wheat<br />
straw residues. It has to be noted that the field<br />
experiment was carried out in southeast Norway.<br />
Due to different climatic situation the<br />
decomposition behaviour is different than for<br />
eastern German conditions. On average climate is<br />
warmer <strong>and</strong> drier than in southeast Norway.<br />
Warmer temperature may enhance decomposition<br />
due to enhanced microbial activity. In contrary<br />
drier climate weakens decomposition activity.<br />
The soil conditions are comparable for both sites.<br />
So data from that source is seen to be<br />
representative to show the models ability to<br />
simulate decomposition of primary organic matter<br />
for crops.<br />
For the forest site decomposition of Scots pine<br />
needles are shown in the following figure,<br />
adapted from a research site closed to the forest<br />
site [Bergmann et al., 1999]. Figure 2 show that<br />
the model represents the decomposition of<br />
primary organic matter in an appropriate way.<br />
remaining C [% of added]<br />
100<br />
90<br />
80<br />
70<br />
60<br />
50<br />
40<br />
30<br />
20<br />
measured - summer barley straw<br />
simulated - summer barley straw<br />
measured - winter wheat straw<br />
simulated - winter wheat straw<br />
measured - needle litter (Scots pine)<br />
simulated - needle litter (Scots pine)<br />
0 100 200 300 400 500 600 700 800<br />
days<br />
Figure 2. Simulated decomposition of winter<br />
wheat, summer barley straw <strong>and</strong> needle litter<br />
(Scots pine). Measured values from mesh bag<br />
experiments were adopted from Henriksen <strong>and</strong><br />
Brel<strong>and</strong> [1999] <strong>and</strong> Bergmann et al. [1999].<br />
The fact that different plant types (e.g. fine roots,<br />
stems, foliage) have different decomposition rates<br />
733
is accounted for through the separation into litter<br />
compartments by the model.<br />
For soil humus the long-term behaviour has to be<br />
investigated. For agricultural soils for a time span<br />
of 40 years the conditions of the static long term<br />
field experiment in Bad Lauchstädt / Halle<br />
(Saxony-Anhalt / Germany) were simulated. Data<br />
were adopted from Franko [1990] <strong>and</strong> simulations<br />
for the active organic matter pool showed good<br />
accordance in the long run with the measured<br />
data. For forest sites representative values for<br />
humus dynamics were taken from Bergmann et al.<br />
[1999]. In this study humus material was<br />
collected <strong>and</strong> decomposition was measured for<br />
approximately 900 days in the humus layer with<br />
so-called rhizobags. The humus was almost<br />
decomposed after approximately 45 – 54 years<br />
(calculated with a mass loss model using the<br />
measured data, ref. Bergmann et al. [1999]). The<br />
humus dynamic for the forest site could also be<br />
simulated in the right magnitudes <strong>and</strong> temporal<br />
behaviour for a period of 50 years (not shown<br />
here).<br />
The integral quantity determining POM <strong>and</strong> AOM<br />
processes is the formation of substrate induced<br />
soil respiration. For forest <strong>and</strong> agro-ecosystems<br />
measured soil respiration known from literature<br />
<strong>and</strong> from available experimental field data have<br />
been compiled <strong>and</strong> are shown in table 1. It has to<br />
be noted that field measurements of soil<br />
respiration include both, the heterotrophic<br />
(microbes <strong>and</strong> soil fauna) <strong>and</strong> autotrophic (root)<br />
respiration. Model estimates deliver only values<br />
for heterotrophic respiration. To consider that<br />
fact, values proposed by Hanson et al. [2000]<br />
were used to separate root / rhizosphere<br />
contributions to total soil respiration for various<br />
vegetation types in different ecosystems. Table 1<br />
shows that simulation results meet measured<br />
magnitudes for agro- <strong>and</strong> forest ecosystems for<br />
East German conditions.<br />
Table 1. Simulated yearly soil respiration (SR)<br />
values for 3 l<strong>and</strong> cover types compared with<br />
literature cited values <strong>and</strong> measurements.<br />
L<strong>and</strong> SR [gC m -2 a -1 ] SR [gC m -2 a -1 ] Ref.<br />
cover simulated measured<br />
Crop 295 - 840 410 - 660 Beyer<br />
[1991]<br />
Crop - 300 250 ATB*<br />
triticale<br />
Crop - 275 (1999) 211 (1999) ATB*<br />
rye 250 (2001) 236 (2001)<br />
Forest<br />
deciduous<br />
100 – 375 292 - 710 Buchmann<br />
[2000]<br />
Forest<br />
evergreen<br />
300 – 525 475 Beyer<br />
[1991]<br />
* Data provided by the Leibnitz – Institute of Agricultural<br />
Engineering Bornim (ATB) from their experimental field site.<br />
4. CONCLUSIONS AND OUTLOOK<br />
The presented results show that the SWIM-SCN<br />
model is able to represent main processes of soil<br />
organic matter turnover for agro- <strong>and</strong> forest<br />
ecosystems at the regional scale. The model can<br />
be described as robust <strong>and</strong> modest in data<br />
requirement, which can be done using regionally<br />
available data sets <strong>and</strong> literature values. Further<br />
verification for additional crop <strong>and</strong> forest species<br />
still has to be performed using not yet available<br />
data from long-term field experiments in East<br />
Germany. Additionally, detailed sensitivity <strong>and</strong><br />
uncertainty analyses of input data <strong>and</strong> model<br />
parameters will be performed in order to provide<br />
error margins for model results. This delivers<br />
necessary information for the evaluation of model<br />
performance.<br />
Currently two modules for nitrogen cycling are<br />
used within the extended SWIM-SCN model. The<br />
original was already tested at the basin scale, <strong>and</strong><br />
the second one is coupled to the carbon cycle.<br />
Both have to be compared <strong>and</strong> then combined into<br />
one to preserve their advantages. It is further on<br />
relevant to consider coupled C <strong>and</strong> N cycles in<br />
order to properly regard feedbacks <strong>and</strong><br />
interactions between them which highly influence<br />
the process of soil organic matter turnover.<br />
The verification results shown here are a<br />
prerequisite to investigate humus <strong>and</strong> soil<br />
respiration dynamics at the regional scale under<br />
the impact of global change. Especially changed<br />
l<strong>and</strong> use conditions, imposed through socioeconomic<br />
changes in a region, have to be<br />
quantified in respect to soil organic matter<br />
dynamics <strong>and</strong> hence carbon sequestration<br />
possibilities. Here the impact of agricultural <strong>and</strong><br />
forest management practices might play an<br />
important role, too. The proposed model<br />
framework may help here to investigate effects of<br />
l<strong>and</strong> use <strong>and</strong> l<strong>and</strong> management change on water,<br />
carbon <strong>and</strong> nitrogen dynamics. This information<br />
can deliver useful hints for policy <strong>and</strong> decisionmaking.<br />
5. ACKNOWLEDGEMENTS<br />
Special thanks goes to the L<strong>and</strong>esforstanstalt<br />
Eberswalde for the provision of Level II data <strong>and</strong><br />
to the Leibnitz – Institute of Agricultural<br />
Engineering Bornim for the provision of data<br />
from their experimental field site. This work was<br />
supported by the HSP fond of the State of<br />
Br<strong>and</strong>enburg.<br />
734
6. REFERENCES<br />
Berg, B. <strong>and</strong> H. Staaf, Decomposition rate <strong>and</strong><br />
chemical changes of Scots pine needle<br />
litter. I. Influence of st<strong>and</strong> age, Structure<br />
<strong>and</strong> function of northern coniferous<br />
forests - An ecosystem study, T. Persson,<br />
Stockholm, 32, 363-372, 1980.<br />
Bergmann, C., T. Fischer, <strong>and</strong> R.F. Hüttl,<br />
Decomposition of needle-, herb-, rootlitter,<br />
<strong>and</strong> Of-layer-humus in three Scotts<br />
pine st<strong>and</strong>s, Changes of Atmospheric<br />
Chemistry <strong>and</strong> Effects on Forest<br />
Ecosystems. A Roof Experiment Without<br />
Roof, R. F. Hüttl <strong>and</strong> K. Bellmann,<br />
Dordrecht, Kluwer, 3, 151-176, 1999.<br />
Beyer, L., Intersite characterization <strong>and</strong><br />
variability of soil respiration in different<br />
arable <strong>and</strong> forest soils, Biol. Fertil Soils,<br />
12, 122-126, 1991.<br />
Buchmann, N., Biotic <strong>and</strong> abiotic factors<br />
controlling soil respiration rates in Picea<br />
abies st<strong>and</strong>s, Soil Biology <strong>and</strong><br />
Biochemistry, 32(11-12), 1625-1635,<br />
2000.<br />
Bugmann, H.K.M., On the Ecology of<br />
Mountainous Forests in a Changing<br />
Climate: A Simulation Study. Swiss<br />
Federal Institute of Technology Zürich,<br />
Zürich, ETH Zürich, 258, 1994.<br />
Chertov, O.G. <strong>and</strong> A.S. Komarov, SOMM: A<br />
model of soil organic matter dynamics,<br />
Ecoogical <strong>Modelling</strong>, 94(2-3), 177-189,<br />
1997.<br />
Franko, U., C- und N-Dynamik beim Umsatz<br />
organischer Substanz im Boden., Berlin,<br />
Akademie der L<strong>and</strong>wirtschaftswissenschaften<br />
der DDR, 1990.<br />
Grote, R., F. Suckow, <strong>and</strong> K. Bellmann,<br />
<strong>Modelling</strong> of of carbon-, nitrogen-, <strong>and</strong><br />
water balances in pine st<strong>and</strong>s under<br />
changing air pollution <strong>and</strong> deposition,<br />
Changes of Atmospheric Chemistry <strong>and</strong><br />
Effects on Forest Ecosystems. A Roof<br />
Experiment Without Roof, R. F. Hüttl<br />
<strong>and</strong> K. Bellmann, Dordrecht, Kluwer, 3,<br />
251-281, 1999.<br />
Hanson, P.J., N.T. Edwards, C.T. Garten, <strong>and</strong> J.A.<br />
Andrews, Separating root <strong>and</strong> soil<br />
microbial contributions to soil<br />
respiration: A review of methods <strong>and</strong><br />
observations, Biogeochemistry, 48, 115-<br />
146, 2000.<br />
Hellebr<strong>and</strong>, H.J., J. Kern, <strong>and</strong> V. Scholz, Longterm<br />
studies on greenhouse gas fluxes<br />
during cultivation of energy crops on<br />
s<strong>and</strong>y soils, Atmospheric Environment,<br />
37(12), 1635-1644, 2003.<br />
Henriksen, T.M. <strong>and</strong> T.A. Brel<strong>and</strong>,<br />
Decomposition of crop residues in the<br />
field: evaluation of a simulation model<br />
developed from microcosm studies, Soil<br />
Biology <strong>and</strong> Biochemistry, 31(10), 1423-<br />
1434, 1999.<br />
Houghton, R.A., Converting terrestrial<br />
ecosystems from sources to sinks of<br />
carbon, Ambio, 25, 267-272, 1996.<br />
Kartschall, T., P. Döring, <strong>and</strong> F. Suckow,<br />
Simulation of Nitrogen, Water <strong>and</strong><br />
Temperature Dynamics in Soil, Syst.<br />
Anal. Model. Simul., 7(6), 33-40, 1990.<br />
Klimanek, E. M., Umsetzungsverhalten der<br />
Wurzeln l<strong>and</strong>wirtschaftlich genutzter<br />
Pflanzenarten, Arch. Acker- Pfl. Boden,<br />
34(8), 569-577, 1990(a).<br />
Klimanek, E. M., Umsetzungsverhalten von<br />
Ernterückständen, Arch. Acker- Pfl.<br />
Boden, 34(8), 559-567, 1990(b).<br />
Krysanova, V. <strong>and</strong> D.-I. Müller-Wohlfeil,<br />
Development <strong>and</strong> test of a spatially<br />
distributed hydrological/water quality<br />
model for mesoscale watersheds,<br />
Ecological <strong>Modelling</strong>, 106, 263-289,<br />
1998.<br />
Krysanova, V. <strong>and</strong> A. Becker, Integrated<br />
modelling of hydrological processes <strong>and</strong><br />
nutrient dynamics at the river basin<br />
scale, Hydrobiologia, 410, 131-138,<br />
2000.<br />
Lasch, P., F.W. Badeck, M. Lindner, <strong>and</strong> F.<br />
Suckow, Sensitivity of simulated forest<br />
growth to changes in climate <strong>and</strong><br />
atmospheric CO 2 , Forstwiss.<br />
Centralblatt, 121(Supplement 1), 155-<br />
171, 2002<br />
Parton, W. J., D. S. Schimel, C.V. Cole, <strong>and</strong> D.S.<br />
Ojima, Analysis of factors controlling<br />
soil organic matter levels in Great Plain<br />
grassl<strong>and</strong>s, Soil Sci. Soc. Am. J., 51,<br />
1173-1179, 1987.<br />
Van't Hoff, J.H., Etudes de dynamique chimique,<br />
Muller, Amsterdam, 214, 1884.<br />
Wattenbach, M., F. Hattermann, R. Weng, F.<br />
Wechsung, V. Krysanova, <strong>and</strong> F.W.<br />
Badeck, A simplified approach to<br />
implement forest eco-hydrological<br />
properties in regional eco-hydrological<br />
modelling, Ecological <strong>Modelling</strong>,<br />
accepted, 2004.<br />
Williams, J. R., K. G. Renard, <strong>and</strong> P.T. Dyke,<br />
EPIC a new method for assessing<br />
erosion´s effect on soil productivity,<br />
Journal of Soil <strong>and</strong> Water Conservation,<br />
38(5), 381-383, 1984.<br />
735
An integrated geomorphological <strong>and</strong> hydrogeological<br />
MMS modeling framework for a semi-arid mountain<br />
basin in the High Atlas, southern Morocco<br />
Carmen de Jong 1 , Rebecca Machauer 1 , Barbara Reichert 2 , Sebastien Cappy 2 , Rol<strong>and</strong> Viger 3 <strong>and</strong><br />
George Leavesley 3<br />
1<br />
Geography Department, University of Bonn, Germany, dejong@giub.uni-bonn.de 2<br />
2<br />
Institute of Geology, University of Bonn, Germany<br />
3<br />
USGS Denver, USA<br />
Abstract: The aims of this study are to develop <strong>and</strong> implement a precipitation-runoff model in the Modular<br />
Modeling System (MMS) for a small mountain catchment, the Ameskar basin. It is part of the meso-scale,<br />
semi-arid Drâa basin investigated within IMPETUS, an integrated project for the efficient <strong>and</strong> sustainable use<br />
of freshwater in southern Morocco. The River Drâa drains from the High Atlas Mountains to Lac Iriki <strong>and</strong><br />
feeds a large dam for irrigation purposes. Precipitation inputs include rain <strong>and</strong> snow from 3 climate stations.<br />
Snow sublimation plays a significant role in the higher altitudes <strong>and</strong> is integrated accordingly. Vegetation is<br />
scarce outwith the intensively irrigated oases <strong>and</strong> evapotranspiration is limited to small shrubs. Surface runoff<br />
<strong>and</strong> springs are controlled by complex geomorphological <strong>and</strong> geological settings <strong>and</strong> by highly porous,<br />
infiltrating wadi river beds. The MMS model has been developed mainly on the basis of geomorphologically<br />
<strong>and</strong> hydrogeologically defined Hydrological Response Units (HRUs). Discharge is sporadic, extreme <strong>and</strong> varies<br />
according to snowmelt <strong>and</strong> precipitation. Infiltration dominates over surface discharge <strong>and</strong> is important for<br />
groundwater renewal especially in limestone <strong>and</strong> basalt. MMS will be developed for the whole Drâa catchment<br />
for operational discharge forecasting in future.<br />
Keywords: MMS, Hydrological Response Units, semi-arid, mountain<br />
1. INTRODUCTION<br />
Mountains play an important role in the regional<br />
water balance of large, complex, semi-arid basins<br />
[Cunningham et al 1998, Flerchinger <strong>and</strong> Cooley,<br />
2000, Khazaei et al 2003, Pitlick 1994, Salvetti et al<br />
2002, Viviroli et al 2003]. However, the contribution<br />
of snow <strong>and</strong> rainfall to the annual <strong>and</strong> multi-annual<br />
water balance in remote high mountain regions in<br />
north-western Africa is largely unknown [Matthews<br />
1989 et al]. This study aims to investigate the water<br />
balance of the semi-arid Ameskar catchment in<br />
the High Atlas Mountains based on an integrated<br />
geomorphological <strong>and</strong> hydro-geological MMS<br />
(Modular Modeling System) framework.<br />
2. STUDY AREA<br />
The study site is the remote, semi-arid Ameskar<br />
basin located in the northern Drâa basin within<br />
the M’Goun Range (3950 m) of the High Atlas<br />
Mountains (Figure 1). Here, the High Atlas forms a<br />
Mesozoic calcareous massif. The catchment consists<br />
of complex, highly folded geological units, mainly<br />
limestone, basalt, syenite, Quaternary cinder, silt,<br />
s<strong>and</strong>stone <strong>and</strong> gypsum. Geomorphologically, the<br />
basin has rockfaces, extensive scree slopes, debris<br />
flow channels, debris fans, ancient <strong>and</strong> active<br />
l<strong>and</strong>slides on the lower valley flanks <strong>and</strong> coarse,<br />
gravelly river beds. The Assif-n-Ait-Ahmed flows<br />
over a highly porous bed with an average surface<br />
discharge of only 0.5 m 3 s -1 <strong>and</strong> sporadic flow down to<br />
the village of Lower Ameskar. Flow becomes totally<br />
subsurface in the lower reaches of the valley. The<br />
Assif-n-Ait- Ahmed joins the M’Goun which in turn<br />
forms the Drâa below its confluences with the Dades<br />
[Youbi 1990]. At Ifre (1500 m) average discharge is 4<br />
m/s. Vegetation in the valley is scarce <strong>and</strong> dominated<br />
by scattered juniperus trees between 1800-2400 m ,<br />
cushion shrubs above 2400 m <strong>and</strong> accacia wadi<br />
communities in the rivers <strong>and</strong> dry wadi beds. These<br />
are partially irrigated with Mediterranean fruit oases<br />
736
Tabant<br />
JBEL<br />
6°20<br />
WAOUGOULZAT<br />
3763 m<br />
6°10<br />
0 5 km<br />
J B E L<br />
31°30<br />
4068 m<br />
M G<br />
Aflafal<br />
O U N<br />
M Goun<br />
*<br />
*<br />
El Mrabitine<br />
Tizi-n-Tounza<br />
Tichki<br />
Assif-n-Ait-Ahmed<br />
J b e<br />
l<br />
Ta d a r a s<br />
Taria Ifre<br />
Cascade<br />
Ameskar<br />
3266 m<br />
Jbel<br />
Tigounatine<br />
Oued<br />
Mgoun<br />
J b e l<br />
A s s e<br />
l d a<br />
Amajgag<br />
Igherm<br />
Aqdim<br />
Assif-n-Ait Toumert<br />
J b<br />
e l<br />
A k<br />
l<br />
i m<br />
3432 m<br />
Ait Khlifa<br />
*<br />
31°20<br />
climate station<br />
stage recorder<br />
snow pillow<br />
lysimeter<br />
road<br />
Ameskar<br />
basin<br />
Taoujgalt<br />
Figure 1 Map of Ameskar study basin in the<br />
M’Goun range with stream gauges <strong>and</strong> climate<br />
stations<br />
Ifre<br />
G.B.-J.<br />
<strong>and</strong> legumes. In the Upper Drâa, the average annual<br />
available groundwater resource is approximately 80<br />
Mm 3 per annum. At Ifre, the Oued M’Goun has a<br />
catchment size of 1240 km 2 <strong>and</strong> although it covers<br />
only 8% of the total surface area of the Upper Drâa<br />
it provides the main freshwater resource for the<br />
region.<br />
2.1 Climate<br />
The High Atlas mountains play a distinct role<br />
in producing orographic precipitation <strong>and</strong> cloud<br />
formation. Average annual precipitation in the<br />
Ameskar catchment varies around 500 mm <strong>and</strong><br />
is influenced by a strong vertical temperature <strong>and</strong><br />
rainfall gradient. The semi-arid climate can be<br />
subdivided into two dominant seasons: a wetter<br />
winter season with both rain <strong>and</strong> snow at the<br />
beginning <strong>and</strong> towards the end <strong>and</strong> a very dry<br />
summer season with at least 2 months without<br />
precipitation. At the highest peak station (3850 m)<br />
minimum temperatures can reach –25°C during the<br />
winter while average summer temperatures only<br />
reach 15°C <strong>and</strong> wind speeds can attain 25 m/s. On<br />
average there are more than 140 frost days per year.<br />
Snowfall events are erratic throughout the winter<br />
<strong>and</strong> at the higher stations, first modeling results<br />
<strong>and</strong> field experiments indicate a high percentage of<br />
snow lost directly through sublimation [Schulz et al<br />
2003].<br />
2.2 Flood hydrology<br />
Sporadic floods occur as a result of snowmelt <strong>and</strong><br />
/ or catchment wide precipitation in spring <strong>and</strong><br />
autumn (Figure 2). The flood characteristics are a<br />
Figure 2 Rapid surface flow over basalt with flood<br />
in confluence of the main channel, Sept. 2003<br />
[Foto: Cappy]<br />
reflection of extreme rainfall or snowmelt patterns<br />
as well as geomorphologic <strong>and</strong> geologic setting.<br />
Flood peaks are high with steep ascending <strong>and</strong><br />
descending (“shark-tooth”) flood limbs. The periods<br />
between extreme events are marked by low or absent<br />
discharge.<br />
2.3 Hydrogeology <strong>and</strong> Geomorphology<br />
The Ameskar basin can be roughly divided into<br />
two main geological units, the massive limestones<br />
<strong>and</strong> dolomites that shape the higher levels of the<br />
catchment <strong>and</strong> the basalts covered by clays along<br />
the lower slopes (Table 1). The karstic aquifers have<br />
a transmissivity of between 10 -1 to 10 -4 m 2 /s <strong>and</strong><br />
are therefore highly permeable. Since scree slopes<br />
are extensive in the limestones, large amounts of<br />
water can infiltrate easily into the underground.<br />
Table 1 Hydrogeological characteristics of Ameskar<br />
basin [Reichert, Cappy <strong>and</strong> Thein 2003]<br />
Formation Lithology Classification<br />
Quaternary s<strong>and</strong>, gravel,<br />
limestone<br />
clasts<br />
porous aquifer<br />
/ 0 to 30 m / 10 -2 m.s -1<br />
porous aquifer<br />
/ 20 to 50 m / 10 -3 m.s -<br />
Liassic<br />
(carbonatic)<br />
Liassic<br />
(siliciclastic)<br />
Lower Liassic<br />
/Upper Triassic<br />
Triassic<br />
(basalt)<br />
Triassic<br />
(continental)<br />
limestone, dolomite<br />
s<strong>and</strong>stone,<br />
limestone, dolomite<br />
clay, siltstones<br />
doloritic basalt<br />
s<strong>and</strong>-siltstone,<br />
conglomerates<br />
porous <strong>and</strong> fractured<br />
aquifer to a karst<br />
aquifer<br />
/ 100 to 500 m /<br />
10 -6 to 10 -2 m.s -1<br />
fractured aquifer<br />
/ 20 to 50 m /<br />
variable<br />
aquitarde<br />
/ < 5 m / 10 -9 m.s -1<br />
Fractured aquifer<br />
/ 50 to 300 m /<br />
10 -8 to 10 -3 m.s -1<br />
Aquiclude<br />
< 200 m / 10 -6 m.s -1<br />
737
The recession coefficient of the slow discharge<br />
components lies between 120-210 days in limestone<br />
aquifers [Schwarze et al 1999]. The clay sediments of<br />
the Upper Trias form a near impermeable layer with<br />
a clear spring horizon. Triassic basalts are densely<br />
developed with fissures of different transmissivity<br />
ranging between 0.15-1.5 10 -7 m 2 /s according to<br />
Schwarze [1999] <strong>and</strong> have a recession coefficient of<br />
270-310 days. The hydrogeological situation of the<br />
basin can only be roughly estimated at the moment<br />
with the help of tracer <strong>and</strong> dating techniques.<br />
The basin can be subdivided into several<br />
geomorphological units based on the underlying<br />
geology <strong>and</strong> geomorphologic activity. The main<br />
units in the highest regions consist of old terraces<br />
<strong>and</strong> erosional surfaces that are highly permeable<br />
<strong>and</strong> contain loose rock fragments. The most<br />
widespread units in the catchment are the screes<br />
that cover the steeper slopes below the rock faces.<br />
The scree slopes are fed by rock falls originating<br />
are preferentially reworked by a dense network of<br />
debris flows. The scree slope units also have a high<br />
permeability <strong>and</strong> they often consist of moist soils<br />
covered by a protective layer of rock fragments.<br />
Some of the lower slopes are covered by cinder<br />
slopes. This unit is totally impervious due to their<br />
clay content but can be found in close vicinity to<br />
the highly pervious units. The next category are<br />
the river beds themselves <strong>and</strong> the fluvially eroded<br />
lower slopes. The river beds consist of a massive<br />
layer of loose stones <strong>and</strong> rocks reworked by rare<br />
flood events or deposited at the lower valley slopes<br />
as debris fans. They are highly permeable unless in<br />
contact with rock layers near the surface as is typical<br />
for basalt dykes. The last category consists of mass<br />
movements. These are also located along the lower<br />
valley slopes but are of much larger extend <strong>and</strong><br />
volume. They are reworked finer sediments <strong>and</strong><br />
are less permeable than the adjacent river beds or<br />
slopes.<br />
3. METHODOLOGY<br />
3.1 Climate stations<br />
In the Ameskar catchment, climate stations were<br />
installed at three sites (Figure 1): Ameskar (2250<br />
m), Tichki (3260 m) <strong>and</strong> M’Goun (3850 m). Since<br />
2001 common meteorological data include soil <strong>and</strong><br />
air temperature, humidity, radiation, precipitation,<br />
wind direction <strong>and</strong> wind speed is collected at one<br />
level in 15 minute intervals.<br />
3.2 Discharge Stations<br />
Three float gauge stations are operational in the<br />
catchment (Figure 1), one since April 2002 in Taria<br />
(2752 m) with a catchment area of 5.5 km 2 , one since<br />
October 2003 at Cascade (2195 m) with a catchment<br />
area of 53 km 2 <strong>and</strong> one since November 2003 at<br />
Tichki village with a catchment area of 15 km 2 .<br />
Serious problems were encountered with discharge<br />
measurements due to very long periods of low<br />
flow alternating with extreme flood flow. Since no<br />
calibrations were possible during high / flood flow,<br />
discharge is underestimated for floods especially at<br />
Cascade where the flow cross-section is difficult to<br />
define. During the period of 2002, the Taria stage<br />
recorder periodically stopped working over the<br />
summer of 2002 <strong>and</strong> after the flood in April 2003.<br />
This is possibly caused by high suspended sediment<br />
concentrations deposited on the wheel during the<br />
flood <strong>and</strong> causing the float to jam. The observed<br />
stage was corrected manually for this second period<br />
according to the pre-flood reference level.<br />
3.3 Geological <strong>and</strong> hydrogeological mapping<br />
A detailed geological map was produced by a<br />
combination of field work <strong>and</strong> image interpretations<br />
derived from remote sensing. Hydrogeological<br />
measurements distinguished the discharge of<br />
different springs, the age <strong>and</strong> origin of water.<br />
The methods include δ 2 H, δ 18 O <strong>and</strong> Tritium<br />
measurements, discharge <strong>and</strong> water quality<br />
measurements of springs.<br />
3.4 Geomorphological mapping<br />
The geomorphology of the basin was determined<br />
from field work, topographical maps (1: 100 000) as<br />
well as from high resolution remote sensing images<br />
such as IKONOS <strong>and</strong> ASTER (Figure 3).<br />
3.5 Soil water investigations<br />
Soil water investigations included sprinkling <strong>and</strong><br />
infiltration experiments as well as studies on the<br />
structure <strong>and</strong> skeletal content of soils. Diverse units<br />
were studied such as the highly permeable river<br />
beds, impermeable cinder slopes <strong>and</strong> widespread<br />
scree slopes.<br />
Figure 3 Detailed geomorphological basin map<br />
738
4. MMS MODEL<br />
4.1 GIS Weasel<br />
The pre-processor for the Modular Modeling System<br />
(MMS) model consists of the GIS Weasel which<br />
is a GIS-based tool for classifying Hydrological<br />
Response Units (HRUs) based on a Digital Elevation<br />
Model with a grid size of 100 m, vegetation maps<br />
derived from topographical maps <strong>and</strong> remote sensing<br />
images, geomorphological, geological <strong>and</strong> simple<br />
assumptions derived from soil characteristic maps<br />
(Figure 4). All units were digitized as polygons <strong>and</strong><br />
converted into grids as a basis for parameterization<br />
for the MMS model. HRUs were disaggregated to<br />
improve discharge modeling [Carlile 2002].<br />
The catchment was subdivided into two main zones:<br />
the Taria subcatchment at the Taria gauging outlet<br />
<strong>and</strong> the Cascade subcatchment at the Cascade outlet.<br />
No surface discharge is present at the final basin<br />
outlet therefore the model was not run for that part.<br />
4.2 MMS Model<br />
The PRMS (Precipitation Runoff <strong>Modelling</strong><br />
System) developed by Leavesley [2002] was applied<br />
for the Ameskar catchment. It includes a special<br />
function for precipitation that allows precipitation<br />
to be extrapolated between the different stations<br />
for individual HRUs. Precipitation (including snow<br />
<strong>and</strong> snow water equivalent) from the three stations<br />
is computed for the HRUs according to the daily<br />
a)<br />
b)<br />
TARIA<br />
TARIA<br />
Figure 4 Taria <strong>and</strong> Cascade subcatchments with<br />
HRUs for a) vegetation [after de Jong 2003] <strong>and</strong><br />
b) hydrogeology [after Bell 2003, Budewig 2003,<br />
Hofmann 2002, Osterholt 2002 <strong>and</strong> Cappy 2003]<br />
measured minimum <strong>and</strong> maximum temperatures<br />
<strong>and</strong> radiation. This is particularly important for<br />
mountain basins with highly variable precipitation.<br />
Precipitation was corrected for the three stations<br />
according to Sevruk <strong>and</strong> Zahlavova [1994]. The<br />
modeled discharge results for Taria <strong>and</strong> Cascade are<br />
decomposed into several water balance components<br />
(Table 2). Comparison of the modeled results of the<br />
Taria subcatchment shows that a large percentage of<br />
the water is transmitted into storage <strong>and</strong> base flow<br />
(nearly 40%) <strong>and</strong> that there is nearly twice as much<br />
subsurface flow as surface discharge. This is due to<br />
the fact that, geologically, the whole sub-catchment<br />
consists of limestone. In Cascade, on the other h<strong>and</strong>,<br />
a very high percentage of precipitation is lost by<br />
evaporation <strong>and</strong> by sublimation after snow events.<br />
Less than 15% of the flow is subsurface <strong>and</strong> less<br />
than 11% is surface discharge.<br />
Table 2 Modeled components of water cycle for a)<br />
Taria subcatchment <strong>and</strong> b) Cascade subcatchment<br />
[after Machauer 2003].<br />
N ETA Q Q O<br />
Q I<br />
Q B<br />
S<br />
a) mm 569 20 123 6 9 110 101<br />
% 100 38.7 21.6 1 1.6 19.3 17.7<br />
b) mm 384 281 42 6 7 15 33<br />
% 100 73 11 1.6 1.8 3.9 8.8<br />
N = total precipitation, ETA = actual<br />
evapotranspiration, Q = discharge, Q O<br />
= surface flow, Q I<br />
=<br />
interflow, Q B<br />
= base flow, S = storag<br />
Whereas interflow exceeds surface flow for most<br />
of the flood events <strong>and</strong> for average flow at Taria,<br />
surface flow equals or exceeds interflow at Cascade.<br />
This is mainly due to the geomorphological setting<br />
of the basin. The highly porous channels of Taria<br />
<strong>and</strong> Cascade are rapidly filled with water during<br />
heavy <strong>and</strong> intensive precipitation events, such that<br />
surface flow peaks rapidly over short periods after<br />
the interflow storage areas have been saturated.<br />
Flood recession is equally rapid, producing the<br />
characteristic “shark-tooth” flood hydrograph.<br />
Once the precipitation input ceases, which is very<br />
abrupt in these regions, flow retreats rapidly into the<br />
interflow areas which in turn feeds into the deeper<br />
groundwater reservoirs. Thus high flow <strong>and</strong> flood<br />
flow events reflect a high percentage of near surface<br />
<strong>and</strong> surface flow. Since the vegetation cover is sparse<br />
<strong>and</strong> loamy soils are rare, little water is buffered in<br />
the vegetation zone.<br />
These first results show good correlations between<br />
the modelled <strong>and</strong> observed discharge <strong>and</strong> are<br />
promising given the complexity of the catchment<br />
geology <strong>and</strong> the short climate <strong>and</strong> discharge<br />
time series involved (Figure 5). For the Taria<br />
subcatchment the Pearson correlation coefficient<br />
739
etween modeled <strong>and</strong> observed discharge is 0.74<br />
<strong>and</strong> the Index of Agreement is 0.70. The differences<br />
between the observed <strong>and</strong> simulated runoff for Taria<br />
<strong>and</strong> Cascade are the result of different processes in<br />
each basin, for example the timing <strong>and</strong> magnitude of<br />
runoff reflects precipitation versus snowmelt. Thus<br />
the lower flow on the 4th of April at Taria is the result<br />
of snowmelt initiated 4 days after the precipitation<br />
event. At Cascade, the same event causes an<br />
earlier <strong>and</strong> higher modeled flood peak as a result<br />
of immediate rainfall runoff response. The flood<br />
event is considerably overpredicted but considering<br />
that the observed discharge is not totally reliable at<br />
this site, the modelled discharge may, in this case,<br />
be a better approximation of the real situation. The<br />
rapidity of development <strong>and</strong> decay of flood peaks is<br />
a reflectance of both the high porosity of the wadi<br />
river beds <strong>and</strong> the karstic nature of the region.<br />
a)<br />
b)<br />
Figure 5 Modeled versus observed results for a) Taria <strong>and</strong> b) Cascade subcatchment (Oct. – July 2003).<br />
5. CONCLUSIONS<br />
This study shows the benefits of application of the<br />
Modular Modeling System in a small, complex,<br />
mountainous catchment. The modeling procedure<br />
has given insight into the complexity of discharge<br />
patterns in semi-arid mountain basins. A major focus<br />
was put on the identification <strong>and</strong> parameterization<br />
of Hydrological Response Units based specifically<br />
on their geomorphological <strong>and</strong> hydrogeological<br />
characteristics. The model results show that<br />
discharge reacts more sensitively to precipitation<br />
than to the hydrogeological components.<br />
There are two main conclusions that can be<br />
drawn from this study. Firstly, the climatological,<br />
geomorphological <strong>and</strong> geological characteristics<br />
as well as gauging errors of the two basins can<br />
explain the differences between the simulated <strong>and</strong><br />
observed discharge. The dominance of snow over<br />
rainfall, the differences between limestone versus<br />
basalt aquifers <strong>and</strong> the percentage of porous river<br />
beds all influence the timing <strong>and</strong> magnitude of flow.<br />
In future, it is desirable to develop a channel loss<br />
function for the model to analyze the overestimation<br />
in rainfall runoff. Secondly, specific discharge<br />
components should be developed in cooperation<br />
with hydrogeologists. Observations of flow <strong>and</strong><br />
chemistry will improve the knowledge of flow paths<br />
740
<strong>and</strong> residence times of water in the basins <strong>and</strong> this<br />
knowledge will be used to improve the model. It is<br />
also anticipated to validate the high ETA in relation<br />
to discharge.<br />
6. ACKNOWLEDEGMENTS<br />
This project was funded by the BMBF (Federal<br />
Ministry for Education <strong>and</strong> Research) in the frame<br />
of the GLOWA-IMPETUS Morocco project. We are<br />
grateful for the support given by numerous Diploma<br />
<strong>and</strong> PhD students.<br />
7. REFERENCES<br />
Bell S., Geologische Kartierung der IMPETUS-<br />
Testsite Tichki und hydrogeologische<br />
Bewertung des Assif n‘Ait Ahmed<br />
Einzugsgebietes, Region M´Goun, südlicher<br />
Hoher Atlas, Marokko. - Unpublished<br />
diploma thesis, Geological Institute,<br />
University of Bonn, 2003.<br />
Budewig T., Geologische Kartierung der IMPETUS-<br />
Testsite Tichki, Region M´Goun, südlicher<br />
Hoher Atlas, Marokko. Unpublished diploma<br />
thesis, Geological Institute, University of<br />
Bonn, 2003.<br />
Carlile, P. W., A.J. Jakeman, B.F.W. Croke und<br />
B.G. Lees, Use of Catchment Attributes to<br />
Identify the Scale <strong>and</strong> Values of Distributed<br />
Parameters in Surface <strong>and</strong> Sub-surface<br />
Conceptual Hydrology Models. In:<br />
Integrated Assessment <strong>and</strong> Decision Support,<br />
Proceedings of the First Biennial Meeting of<br />
the <strong>International</strong> <strong>Environmental</strong> <strong>Modelling</strong><br />
<strong>and</strong> <strong>Software</strong> Society, <strong>Volume</strong> 1: 346-351,<br />
2002.<br />
Flerchinger, G.N. <strong>and</strong> K.R. Cooley, A ten-year water<br />
balance of a mountainous semi-arid watershed.<br />
Journal of Hydrology 237: 86-99, 2000.<br />
Hofmann H., Geologische Kartierung und<br />
hydrogeologische Bewertung der IMPETUS-<br />
Testsite Ameskar, Region M´Goun, südlicher<br />
Hoher Atlas, Marokko. Unpublished diploma<br />
thesis, Geological Institute, University of<br />
Bonn, 2002.<br />
Khazaei E., Spink A.E.F., <strong>and</strong> J. W. Warner, A<br />
catchment water balance model for estimating<br />
groundwater recharge in arid <strong>and</strong> semi-arid<br />
regions of south-east Iran. Hydrogeology<br />
Journal 11(3): 333-342, 2003.<br />
Leavesley, G.H., Markstrom, S.L., Restrepo, P.<br />
J. <strong>and</strong> R. J. Viger, A modular approach to<br />
adressing model design, scale, <strong>and</strong> parameter<br />
estimation issues in distributed hydrological<br />
modelling. In: Hydrological Processes 16,<br />
173-187, 2002.<br />
Matthews, D.A. et al, Programme Al Ghait – Morocco<br />
winter snowpack augmentation project. Final<br />
report. Department of the Interior, Washington<br />
D.C., 1989.<br />
Machauer, R., Hydrological investigations in the<br />
Assif-n-Ait-Ahmed catchment in the High<br />
Atlas Morocco, Unpublished Diploma Thesis,<br />
University of Bonn, 96, 2003.<br />
Osterholt V., Geologische Kartierung der IMPETUS-<br />
Testsite Ameskar, Region M´Goun, südlicher<br />
Hoher Atlas, Marokko. Unpublished diploma<br />
thesis, Geological Institute, University of<br />
Bonn, 2002.<br />
Pitlick, J., Relation between peak floods, precipitation<br />
<strong>and</strong> physiography for five mountainous regions<br />
in the western USA. Journal of Hydrology,<br />
vol. 158, Issue 3-4, p. 219-240., 1994.<br />
Reichert, B., Cappy, S. <strong>and</strong> J. Thein, Hydrogeology<br />
of the Draa catchment. IMPETUS final<br />
report. Integrated Management Project<br />
for the Efficient <strong>and</strong> Sustainable Use of<br />
Freshwater in West Africa. 2003.<br />
Salvetti, A., Ruf, W., Burl<strong>and</strong>o, P., Juon, U. <strong>and</strong><br />
C. Lehmann, Hydrotope-based river flow<br />
simulation in a Swiss Alpine Catchment<br />
accounting for Topographic, Micro-climatic<br />
<strong>and</strong> L<strong>and</strong>use Controls. Integrated Assessment<br />
<strong>and</strong> Decision Support, Proceedings of the<br />
First Biennial Meeting of the <strong>International</strong><br />
<strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong> <strong>Software</strong><br />
Society, <strong>Volume</strong> 1, 334- 339. June 2002.<br />
Schwarze, R., Dröge, W., <strong>and</strong> Opherden, K.<br />
Regionalisierung von Abfluss-komponenten,<br />
Umsatzräumen und Verweilzeiten für<br />
kleine Mittel-gebirgseinzugsgebiete. In:<br />
Kleeberg, H.-B., Mauser, W., Peschke, G.<br />
<strong>and</strong> Streit, U. (Hgr.): DFG: Hydrologie und<br />
Regionalisierung. Wiley-VCH, Weinheim,<br />
345-370. 1999.<br />
Schulz, O., De Jong, C. <strong>and</strong> M. Winiger, Snow<br />
depletion modelling in the High Atlas<br />
Mountains of Morocco. In: Geophysical<br />
Research Abstracts, <strong>Volume</strong> 5, EGS-AGU-<br />
EUG Joint Assembly. 2003.<br />
Sevruk, B. <strong>and</strong> L. Zahlavova, Classification<br />
system of precipitation gauge site exposure<br />
<strong>and</strong> application. <strong>International</strong> Journal of<br />
Climatology, 14, 681-689, 1994.<br />
Youbi, L., Hydrologie du Bassin du Dades. Office<br />
Regional de mise en valeur agricole de<br />
Ouarzazate, pp. 40, Ouarzazate 1990.<br />
Viviroli, D., Weingartner, R, <strong>and</strong> Messerli, B.,<br />
Assessing the hydrological significance of the<br />
world’s mountains. Mountain Research <strong>and</strong><br />
Development, 23:1, 32-40, 2003.<br />
741
Anticipated Effects of Re-Allocation of Intensive<br />
Livestock in S<strong>and</strong>y Areas in the Netherl<strong>and</strong>s<br />
A.P. van Wezel, R.O.G. Franken, J.D. van Dam <strong>and</strong> P. Cleij<br />
<strong>Environmental</strong> Assessment Agency, National Institute for Public Health <strong>and</strong> the Environment, Bilthoven, The<br />
Netherl<strong>and</strong>s, AP.van.Wezel@rivm.nl.<br />
Abstract: Currently plans are developed for a large-scale 'reconstruction' of the rural area in the<br />
Netherl<strong>and</strong>s. Goals of the reconstruction are to diminish veterinary risks <strong>and</strong> to improve the quality of<br />
environment, nature <strong>and</strong> l<strong>and</strong>scape. Spatial separation of l<strong>and</strong>-use is an important instrument. The plans are<br />
prepared at a regional scale, involving stakeholders. We present an ex-ante analysis of the effects of these<br />
plans, including cost-effectiveness <strong>and</strong> integration with related spatial plans. Scenarios for the autonomous<br />
development of intensive livestock, spatial data-rich analysis <strong>and</strong> research in the field of public administration<br />
are integrated in the ex-ante analysis.<br />
Keywords: intensive livestock – reconstruction – veterinary diseases – Water Framework Directive – Nitrate<br />
Directive<br />
1. INTRODUCTION<br />
After a huge crisis on swine-plague in 1997, the<br />
Dutch parliament in 2002 enacted the ‘Law on<br />
reconstruction of the s<strong>and</strong>y areas with<br />
concentration of intensive livestock’. The purpose<br />
of the law is threefold:<br />
- diminish veterinary risks<br />
- improve quality of nature <strong>and</strong> l<strong>and</strong>scape<br />
- improve environmental quality <strong>and</strong> water<br />
quality<br />
The law applies to half of the Dutch rural areas,<br />
which are divided in 12 areas for reconstruction.<br />
For each of these, a local reconstruction committee<br />
develops a plan. Zoning of functions is an<br />
important instrument, the law discerns:<br />
- areas for agricultural development (AAD)<br />
where enlargement <strong>and</strong> (re-)establishment of<br />
intensive livestock is possible;<br />
- areas for function combination (AFC) of<br />
agriculture, housing <strong>and</strong> nature, where reestablishment<br />
or enlargement of intensive<br />
livestock is possible if in accordance with the<br />
spatial quality <strong>and</strong> functions in the area;<br />
- areas for extensive agriculture (AEX), with<br />
nature or housing as primary function where<br />
enlargement or (re-) establishment of intensive<br />
livestock is impossible.<br />
The reconstruction plans can interfere directly with<br />
spatial planning procedures. The plans are to be<br />
carried out between 2004 <strong>and</strong> 2016. The State,<br />
provinces, water boards <strong>and</strong> the European Union<br />
will invest 220 million Euro yearly.<br />
The reconstruction is region-specific policy, which<br />
is appropriate if there are serious policy shortages<br />
in the area or if there are specific vulnerable<br />
functions such as nature.<br />
This papers describes an ex ante evaluation of the<br />
effects of the joint reconstruction plans. We<br />
considered the effects of the plans in respect to<br />
earlier established (inter) national obligations for<br />
environment, nature <strong>and</strong> water <strong>and</strong> the original<br />
goals of the law; we considered the costeffectiveness<br />
of the intended measures with respect<br />
to the autonomous developments in agriculture <strong>and</strong><br />
intensive livestock in the Netherl<strong>and</strong>s; we<br />
considered juridical <strong>and</strong> governmental aspects of<br />
the reconstruction; <strong>and</strong> we considered the<br />
integration of the reconstruction plans with other<br />
relevant plans.<br />
2. METHODS<br />
The current situation with respect to nature <strong>and</strong><br />
environment is taken as a starting point. Data were<br />
taken from the ‘<strong>Environmental</strong> Balance’ <strong>and</strong><br />
‘Nature Balance’ published yearly by RIVM.<br />
Geospecific data on numbers of animals are taken<br />
from ‘AGRIS’, a database on agriculture. With<br />
GIS, calculations are made on animal densities per<br />
reconstruction area. Data on farm numbers, l<strong>and</strong><br />
use, farmers income were obtained from CBS<br />
742
(www.cbs.nl). With respect to nature, with help of<br />
GIS overlays are made between the reconstruction<br />
areas <strong>and</strong> various categories of nature.<br />
Autonomous development of agriculture was<br />
modelled according to De Bont et al. (2003), based<br />
on numbers of animals per reconstruction area.<br />
Information on the zoning was given by the<br />
provinces involved in the reconstruction, these GIS<br />
files are used for analysis of the relevant<br />
importance of the various zones, <strong>and</strong> overlaps with<br />
current <strong>and</strong> planned l<strong>and</strong>-uses.<br />
Information on juridical <strong>and</strong> governmental aspects<br />
was gained by literature study, <strong>and</strong> by a series of<br />
interviews with stakeholders in the various<br />
reconstruction areas (Driessen <strong>and</strong> De Gier, 2004).<br />
Databases on other relevant plans, such as for<br />
water <strong>and</strong> housing, are used of the consideration of<br />
horizontal integration with other spatial plans.<br />
3. SITUATION AND DEVELOPMENTS<br />
IN THE AREA<br />
Nitrogen deposition, nitrate concentration in<br />
groundwater <strong>and</strong> phosphate saturation of<br />
agricultural soils, are resulting problems. For other<br />
environmental themes -such as biocide pressure,<br />
dryness, sound <strong>and</strong> light disturbance- the situation<br />
in the reconstruction areas is comparable to other<br />
regions in the Netherl<strong>and</strong>s (RIVM, 2002/2003).<br />
The l<strong>and</strong>scapes of international importance are<br />
situated outside the reconstruction area in the lower<br />
western part of the Netherl<strong>and</strong>s.<br />
Areas pointed out after the EU Bird- <strong>and</strong> Habitat<br />
Directives are mainly concentrated outside the<br />
reconstruction areas in the lower parts <strong>and</strong><br />
wetl<strong>and</strong>s.<br />
Half of the existing <strong>and</strong> planned nature lies in the<br />
reconstruction area. The low environmental quality<br />
leads to problems for vulnerable types of nature.<br />
Furthermore a huge part of nature exists of small<br />
areas that are more vulnerable for the adverse<br />
effects of surrounding l<strong>and</strong>-uses (RIVM, 2003).<br />
3.1. Nature <strong>and</strong> Environment<br />
Figure 1. depicts the location of the reconstruction<br />
areas in the s<strong>and</strong>y eastern <strong>and</strong> southern parts of the<br />
Netherl<strong>and</strong>s. The livestock density is also given, in<br />
GVE/ha. In this units all livestock is summarised<br />
based on their phosphate excretion, a cow is one<br />
GVE.<br />
Figure 2. Nitrogen deposition (mol/ha).<br />
3.2. Agricultural developments<br />
Figure 1. Reconstruction areas in the Netherl<strong>and</strong>s,<br />
with livestock density.<br />
High livestock density combined with vulnerable<br />
soils lead to a low environmental quality for<br />
eutrophication <strong>and</strong> acidification (Figure 2.).<br />
There is a strong shrinkage of the number of farms.<br />
Remaining farmers intensify <strong>and</strong> enlarge their<br />
farms, to remain paying in spite of the high prices<br />
for labour <strong>and</strong> l<strong>and</strong>. Thus, the shrinkage in farms<br />
does not self-evidently lead to a shrinkage in<br />
production factors such as cattle <strong>and</strong> l<strong>and</strong>, <strong>and</strong><br />
neither to a strong reduction in environmental<br />
pressure.<br />
743
Shrinkage in livestock was 11% (expressed in<br />
GVE) in the reconstruction area between 1990 <strong>and</strong><br />
2000 (Figure 3.).<br />
Based on a scenario including established policy<br />
<strong>and</strong> CAP reform (De Bont et al., 2003), which was<br />
supposed on the historical figures per region, a<br />
further shrinkage in livestock of 13% until 2010 is<br />
foreseen.<br />
4000000<br />
3500000<br />
GVE<br />
3000000<br />
2500000<br />
2000000<br />
1500000<br />
1000000<br />
500000<br />
0<br />
1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000<br />
year<br />
Figure 3. Development of total livestock in the<br />
reconstruction area<br />
The high prices for l<strong>and</strong> lead to an acceleration of<br />
scale enlargement in l<strong>and</strong>-bounded agriculture, <strong>and</strong><br />
a replacement of badly paying agricultural sectors<br />
by more paying sectors such as horticulture, bulb<br />
farms <strong>and</strong> tree-nurseries.<br />
3.3. Relevant (inter)national laws<br />
Next to existing laws, many water <strong>and</strong><br />
environmental laws are developing. The Manure<br />
Law will be changed in 2006, as a result of the<br />
judgement by the European Court that the Dutch<br />
implementation of the Nitrate Directive is<br />
insufficient. The obliged norm of 170 kg/ha<br />
manure will -without a shrinkage of livestock- lead<br />
to a large rise in costs for livestock farmers.<br />
The Water Framework Directive (WFD) leads to<br />
obligations for nitrogen, phosphate <strong>and</strong> toxicant<br />
concentrations in water which will be equal or<br />
lower than current Dutch ‘Maximum Tolerable<br />
Risk Concentrations’ (MTR). However, the MTRs<br />
are no obligations <strong>and</strong> are not legally implemented.<br />
The Netherl<strong>and</strong>s can hardly meet those obligations<br />
for reasons of the historical pollution, large-scale<br />
exceeding of the current MTRs <strong>and</strong> exceeding of<br />
the less stringent obligations from the Nitrate<br />
Directive. Following the judgement of the<br />
European Court on the Nitrate Directive, the<br />
Netherl<strong>and</strong>s should take additional measures on<br />
reaching the WFD obligations (Van Rijswick et al.,<br />
2004). The reconstruction can be a vehicle for such<br />
measures.<br />
Areas for the implementation of the Bird- <strong>and</strong><br />
Habitat Directive are pointed out, therefore<br />
maintenance goals will be formulated to the end of<br />
2004, which may be more stringent than WFD<br />
obligations.<br />
Finally, emission reduction goals following the<br />
NEC Directive are currently in negotiation for<br />
2020.<br />
All mentioned juridical developments strengthen<br />
the need for a lowering of environmental pressure<br />
by intensive livestock.<br />
3.4. Conclusions<br />
The Netherl<strong>and</strong>s do currently not meet European<br />
Directives, <strong>and</strong> the foreseen shrinkage of livestock<br />
is insufficient to meet targets. The reconstruction is<br />
one of the means that can contribute. The fact that<br />
the agricultural sector is currently dynamic,<br />
ameliorates the possibility to steer developments.<br />
Goals to be reached on the different relevant fields<br />
for reconstruction are, mostly quantitative,<br />
recorded in (inter) national regulations <strong>and</strong> laws.<br />
For an overview see Van Wezel et al. (2004). It is<br />
not set clear beforeh<strong>and</strong> how far the reconstruction<br />
should contribute in meeting these goals. In view<br />
of the large-scale process <strong>and</strong> investments<br />
however, it might be expected that the contribution<br />
of the reconstruction on meeting the goals is<br />
significant.<br />
The major problems in the reconstruction areas are<br />
directly related to the intensive livestock <strong>and</strong> the<br />
high livestock density. Intensive livestock farmers<br />
want to develop <strong>and</strong> enlarge for economical<br />
reasons, but because of other functions in the<br />
region, manure-related environmental targets <strong>and</strong><br />
veterinary vulnerability this is undesirable.<br />
2. RECONSTRUCTION PLANS AND<br />
EFFECTS<br />
3.1. Zoning<br />
All reconstruction plans used the instrument of<br />
zoning; an overview of the situation of the different<br />
zones (AAD, AFC <strong>and</strong> AEX) is given in Figure 4.<br />
Marked are the differences between the various<br />
reconstruction areas in the relative areas for the<br />
different zones. The low scale of plan development<br />
results in an even lower scale of the zoning. A<br />
higher scale of the different types of areas would<br />
result in a lesser mutual adverse influence between<br />
different l<strong>and</strong>-uses.<br />
Farms in the AEX often still have unused<br />
planological rights. These remain respected, which<br />
means that the farmers still can develop which is<br />
not in line with the intentions of the Law on<br />
Reconstruction. However, not respecting these<br />
rights would presumably have lead tot high public<br />
744
costs for settlement of damage (Driessen <strong>and</strong> De<br />
Gier, 2004). On the long term, farmers in AEX will<br />
have no possibilities to further exp<strong>and</strong> than current<br />
rights.<br />
In the AAD there will be public <strong>and</strong> private<br />
investments because farmers with growth potential<br />
will be replaced from AEX to AAD. These<br />
replacements will hardly be effective to diminish<br />
the environmental problems as they do not lower<br />
the environmental pressure, or lower livestock<br />
intensity. The environmental pressure remains <strong>and</strong><br />
as the background concentrations of for example<br />
ammonia remain high, the acidification problems<br />
in nature areas will still exist. If other measures or<br />
general policy can reduce the environmental<br />
pressure, zoning will be more effective as the<br />
background concentration is lower compared to the<br />
point sources around a nature area for example.<br />
Zoning can be viewed as an instrument to<br />
concentrate public investments, for example on<br />
agricultural nature conservation. If this instrument<br />
is put in concentrated, its revenues are higher<br />
(Opdam <strong>and</strong> Geertsema, 2002). Presently, the<br />
available dem<strong>and</strong> for this form of nature<br />
conservation is higher than the supply, especially<br />
in the Dutch provinces with much intensive<br />
livestock (e.g. NB <strong>and</strong> Li, Figure 5).<br />
The Dutch Cabinet ended a broad discussion on<br />
the future of intensive livestock with the statement<br />
that ‘there is a future for intensive livestock in the<br />
Netherl<strong>and</strong>s’. The Dutch Cabinet will give new<br />
room for establishment for farmers in AAD.<br />
However, in view of the preceding, this perspective<br />
can be laid for example in working in the food<br />
production chains, or in the use of state-of-the-art<br />
technology, but the perspective can hardly be laid<br />
in further intensifying livestock intensity in these<br />
areas.<br />
ha<br />
25000<br />
20000<br />
15000<br />
10000<br />
5000<br />
0<br />
Dr Fl Fr Ge Gr Li NB NH Ov Ut Ze ZH<br />
Supply AN<br />
Dem<strong>and</strong> AN<br />
Figure 5. Supply <strong>and</strong> dem<strong>and</strong> of nature<br />
conservation by agriculture in Dutch Provinces<br />
3.2. Investment packages<br />
An analysis of the investment packages shows that<br />
the different reconstruction committees vary in<br />
their spending of means over themes. In general<br />
investments in agriculture (improvement of<br />
agricultural structure), nature <strong>and</strong> water are<br />
important, while investments in recreation <strong>and</strong><br />
tourism, quality of life <strong>and</strong> l<strong>and</strong>scape take only a<br />
small part of the available budget.<br />
Although a veterinary crisis in 1997 was the<br />
immediate cause for the reconstruction, which was<br />
followed by two crises in 2001 <strong>and</strong> 2003, hardly<br />
measures to reduce veterinary risks are taken in the<br />
reconstruction plans. A lower livestock density, a<br />
reduction in contacts, <strong>and</strong> a regionalization of<br />
intensive livestock is inevitable on the long term<br />
(see also RLG/RDA, 2003). Actions to this end<br />
should be taken by the sector; apparently the<br />
reconstruction process gives insufficient impulses<br />
to do this.<br />
The foreseen lowering in environmental pressure is<br />
mainly explained by autonomous development <strong>and</strong><br />
(proposed) general policy, not by the measures<br />
taken in the reconstruction. For example, results<br />
are given for the province of Gelderl<strong>and</strong> (Figure<br />
6.). In view of the present technological<br />
possibilities -for housing <strong>and</strong> feeding of livestock,<br />
fertilisation, manure use <strong>and</strong> –depots-, either<br />
shrinkage of livestock or export of manure is<br />
inevitable to reach environmental targets.<br />
Province Gelderl<strong>and</strong><br />
Integrale zonering<br />
Reconstructie<br />
P-surf.water<br />
Extensiveringsgebied<br />
L<strong>and</strong>bouwontwikkelingsgebied<br />
Niet-reconstructiegebied<br />
Verwevingsgebied<br />
Nader te bepalen<br />
Reconstructiegebieden<br />
N-surf.water<br />
nitrate in groundwater<br />
general policy<br />
reconstruction<br />
Figure 4. Situation of areas for agricultural<br />
development, areas for function combination <strong>and</strong><br />
areas for extensive agriculture.<br />
ammonia protection of nature<br />
stench<br />
0 20 40 60 80 100<br />
% reach of target<br />
Figure 6: Contribution of reconstruction to the<br />
reach of target compared to the contribution of<br />
general policy (data by Gies et al., 2003).<br />
745
3.3. Integration in other spatial plans<br />
All plans are incorporated in the regional planning<br />
system, <strong>and</strong> thus are legally binding. This is not the<br />
case for water plans made on regional level.<br />
3.4. Governmental <strong>and</strong> juridical aspects<br />
Based on literature <strong>and</strong> a series of interviews with<br />
parties involved in the planning process,<br />
conclusions are drawn regarding plan development<br />
(Driessen <strong>and</strong> De Gier, 2004). The study gives the<br />
impression that parties involved subscribe to the<br />
plans, there is support. The emphasis by law on<br />
zoning results in less attention being paid to other<br />
possible solutions for problems related tot<br />
intensive livestock farming. The initial ambitions<br />
in the zones for extensive agriculture were<br />
adjusted, after both the central government <strong>and</strong> the<br />
provinces appeared hesitant to pay for damage<br />
caused by earlier planning. The central government<br />
will assess the plans according to a detailed<br />
framework <strong>and</strong> decide for financing, after the<br />
provinces have drawn up the plans. This will leave<br />
little room for interpretation by provinces<br />
themselves. Due to developments in relevant<br />
environmental laws <strong>and</strong> re-prioritising policy<br />
goals, parties involved had little certainty<br />
beforeh<strong>and</strong>. The emphasis on realising targets will<br />
lead to risk-free planning.<br />
4. CONCLUSIONS<br />
Major problems in the reconstruction regions are<br />
directly related to the intensive livestock <strong>and</strong> its<br />
high density. Intensive livestock farmers want to<br />
develop <strong>and</strong> enlarge, but because of other functions<br />
in the region, manure-related environmental targets<br />
<strong>and</strong> veterinary vulnerability this is undesirable.<br />
The shrinkage in farms does not lead to a<br />
proportional shrinkage in production factors or<br />
reduction in environmental pressure. The foreseen<br />
shrinkage of livestock is insufficient to meet targets<br />
of current European Directives. Water <strong>and</strong><br />
environmental (EU) laws are developing <strong>and</strong><br />
strengthen the need for a lowering of<br />
environmental pressure. The fact that the<br />
agricultural sector is currently dynamic,<br />
ameliorates the possibility to steer developments.<br />
The low scale of plan development, results in an<br />
even lower scale of the zoning. A higher scale of<br />
would result in a dimished mutual adverse<br />
influence between l<strong>and</strong>-uses.<br />
Farms in the AEX often still have unused rights.<br />
These remain respected, so farmers still can<br />
develop on the short term. Farmers with growth<br />
potential will be replaced from AEX to AAD, these<br />
replacements will hardly contribute to diminish the<br />
environmental problems. As the background<br />
concentrations of for example ammonia remain<br />
high, the acidification problems in nature areas will<br />
still exist. Zoning can be more effective if<br />
background concentrations are lowered.<br />
In general planned investments in agriculture,<br />
nature <strong>and</strong> water dominate compared to<br />
investments in recreation <strong>and</strong> tourism, quality of<br />
life <strong>and</strong> l<strong>and</strong>scape.<br />
Although a veterinary crisis was the immediate<br />
cause for the reconstruction, hardly measures to<br />
reduce veterinary risks are taken in the<br />
reconstruction plans. The effectiveness of ‘pigfree’<br />
zones to reduce veterinary risks is broadly<br />
doubted, <strong>and</strong> they scarcely change the current<br />
situation. A lower livestock density, a reduction in<br />
contacts, <strong>and</strong> a regionalization of intensive<br />
livestock is inevitable on the long term.<br />
The foreseen lowering in environmental pressure is<br />
mainly explained by autonomous development <strong>and</strong><br />
(proposed) general policy, not by the measures<br />
taken in the reconstruction. In view of the present<br />
technological possibilities -for housing <strong>and</strong> feeding<br />
of livestock, fertilisation, manure use <strong>and</strong> –depots-,<br />
either shrinkage of livestock or export of manure is<br />
inevitable to reach environmental targets.<br />
5 ACKNOWLEDGEMENTS<br />
The work described has been performed under the<br />
authorization of the Netherl<strong>and</strong>s <strong>Environmental</strong><br />
Assessment Agency. Rienk Kuiper, Wil van<br />
Duijvenbooden <strong>and</strong> Reinier van den Berg are<br />
acknowledged for their contributions in<br />
discussions. Jan Roels <strong>and</strong> Tony Balnikker<br />
(Ministry of Housing, Spatial Planning <strong>and</strong> the<br />
Environment) <strong>and</strong> Herman Wierenga (Ministry of<br />
Agriculture, Nature <strong>and</strong> Food Quality) attributed to<br />
this work by supply of information <strong>and</strong> discussion.<br />
6. REFERENCES<br />
De Bont, C.J.A.M., J.F.M. Helming, <strong>and</strong> J.H. Jager<br />
(2003) Reform of the Common Agricultural<br />
Policy 2003. LEI report 6.03.15, The Hague<br />
(in Dutch)<br />
Driessen, P.P.J. <strong>and</strong> A.A.J. De Gier (2004) Rural<br />
areas on the move? Substantive, executive<br />
<strong>and</strong> legal aspects of progress in<br />
reconstruction of intensive livestock<br />
farming. RIVM report 500025 001 (In<br />
Dutch)<br />
Gies, T.J.A., P. Coenen, A. Bleeker <strong>and</strong> O.F.<br />
Schoumans (2002) <strong>Environmental</strong> analysis<br />
of reconstruction in Gelderl<strong>and</strong> <strong>and</strong> eastern<br />
Utrecht. Alterra report 535.4 (In Dutck)<br />
Opdam, P, W. Geertsema (2002) Agricultural<br />
Nature Conservation on L<strong>and</strong>scape level;<br />
746
more revenues by spatial connection.<br />
L<strong>and</strong>werk 3: 28-32 (In Dutch)<br />
RIVM (2002/2003) <strong>Environmental</strong> Balance. (In<br />
Dutch).<br />
RIVM (2003) Nature Balance. (In Dutch)<br />
RLG/RDA (2003) Policy for veterinary risks with<br />
support. RLG 03/8, RDA 2003/08 (In<br />
Dutch)<br />
Van Rijswick, H.F.M.W., F.C.M.A. Michiels, <strong>and</strong><br />
J. Moe Soe Let (2004) Juridical water<br />
policy monitor. CELP/NILOS, University<br />
Utrecht (In Dutch)<br />
Van Wezel, A.P., R.O.G. Franken, J.D. van Dam<br />
<strong>and</strong> P. Cleij (2004) Ex ante evaluation of<br />
the reconstruction plans. RIVM report<br />
500025 002 (In Dutch)<br />
747
An Integrated System for the Forest Fire Dynamics<br />
Hazard Assessment Over a Wide Area<br />
P. Fiorucci a , F. Gaetani a , R. Minciardi a,b<br />
a CIMA - Centro di Ricerca Interuniv. in Monitoraggio Ambientale via Cadorna, 7- 17100 Savona, Italy.<br />
b DIST – Dipartimento di Informatica, Sistemistica e Telematica – Università di Genova via Opera Pia, 13 -<br />
16145 Genova, Italy. Phone +39010353-2804, Fax -39010353-2154<br />
E-mail: Francesco.Gaetani@unige.it<br />
Abstract: An integrated approach is presented for the assessment of forest fire hazard over a wide<br />
geographical area on the basis of real-time information, meteorological forecasts, <strong>and</strong> territorial data stored in<br />
a geographical information system. The paper describes the architecture of a comprehensive system that can<br />
be designed in order to manage the overall forest fire risk. Specifically, vegetation modelling, in connection<br />
with local meteorological conditions <strong>and</strong> topography, allows obtaining forecast of the dynamics of moisture<br />
contents of the different kinds of dead <strong>and</strong> live fuels over a wide geographical area. Besides, a semi-physical<br />
fire propagation model gives a quantitative evolution of the hazardousness over the whole considered region.<br />
Such hazardousness is related to the spread behaviour of a potential fire after an accidental or deliberate<br />
ignition. A dynamic hazard assessment is carried out, as hazard distribution in time <strong>and</strong> space <strong>and</strong> is<br />
determined over a certain time horizon (24/72 hours). The purpose of dynamic hazard assessment is that of<br />
getting reliable information useful to take a number <strong>and</strong> a variety of pre-operational actions that can reduce<br />
the impact of potentially lighted fire over the considered territory, within the considered time horizon. An<br />
application of the system is described over the whole Italian territory relevant to the Joint Operation Center<br />
of the Italian Civil Protection, in order to demonstrate the effectiveness of the proposed approach.<br />
Keywords: forest fire, dynamic modelling<br />
1. INTRODUCTION AND OBJECTIVES<br />
OF THE PAPER<br />
The forest fires phenomenon is strictly related<br />
with l<strong>and</strong> use <strong>and</strong> vegetational characteristics of<br />
the area where the ignitions have been effected.<br />
As a matter of fact, in Mediterranean basin, forest<br />
fires occurrence is almost in all the cases<br />
attributable to man (either as a voluntary action or<br />
as an involuntary consequence of some activity),<br />
whereas in other geographical areas a great<br />
number of fires are caused by lightning activity.<br />
However, in all the cases, the propagation is<br />
heavily influenced by the characteristics<br />
(topography, vegetation, etc.) of the interested<br />
territory, as well as by the meteorological<br />
conditions <strong>and</strong> the conditions of the fuel (mainly<br />
as regard its moisture content). From the above<br />
discussion, it is clear that, in general, it is<br />
improper to think of forecasting ignition, whereas,<br />
it is sensible to assess <strong>and</strong> forecast the danger that<br />
a (somehow) active fire may find favourable<br />
conditions for its propagation.<br />
In this connection, the purpose of the paper is that<br />
of presenting an integrated approach that has been<br />
developed in order to assess the forest fire hazard<br />
over a wide geographical region, on the basis of<br />
all available information.<br />
The rest of the paper is organized as follow. In the<br />
next section the architecture of the developed<br />
system for dynamic forest risk assessment is<br />
presented. In the third <strong>and</strong> in the forth sections,<br />
the fuel moisture <strong>and</strong> the initial spread model<br />
actually implemented in the system, are<br />
introduced. In the fifth section, the software<br />
implementation of the system is described <strong>and</strong><br />
some results concerning the application of the<br />
748
system to the Italian territory are briefly<br />
presented. Finally, in the last section some<br />
conclusions are drawn <strong>and</strong> some possible<br />
directions for forthcoming research activity are<br />
indicated.<br />
2. THE STRUCTURE OF THE HAZARD<br />
ASSESSMENT MODULE<br />
The paper describes a first implementation of a<br />
“dynamic hazard assessment procedure”, whose<br />
structure (see Fig.1) may be decomposed into two<br />
sub-models, namely the fuel moisture model <strong>and</strong><br />
the initial fire spread model. The function of the<br />
fuel moisture model is to represent the dynamic<br />
behaviour of the distribution throughout the<br />
territory of the variable expressing the water<br />
content of the fuel that is mostly interested by the<br />
ignition process. The second model is used to<br />
quantitatively describe the behaviour of a<br />
(possibly) lighted fire, disregarding any possible<br />
extinguishing action. Such a model is not used to<br />
obtain a forecast of the propagation process of a<br />
given fire, but only to evaluate the risk of<br />
potential spread after a possible ignition.<br />
The information that can be used by the two submodels<br />
represented in Fig. 2 is both static <strong>and</strong><br />
dynamic. The first is related to topographic <strong>and</strong><br />
territorial data (topography, l<strong>and</strong> use, road<br />
networks, etc.) which can be obtained from a<br />
Geographical Information System (GIS), <strong>and</strong> to<br />
the vegetation cover of the areas considered. Note<br />
that vegetation cover (biomass kind <strong>and</strong> density)<br />
may be included within static information; in fact,<br />
seasonal biomass dynamics is much slower than<br />
the dynamics of the two models represented in<br />
Fig. 2, so that one can reasonably think of<br />
considering the average vegetation load for each<br />
season of interest.<br />
On the other h<strong>and</strong>, dynamic information may be<br />
diverse in nature. First of all, there may be a<br />
network of ground sensors (rain gauges,<br />
anemometers, solar radiation sensors, etc.)<br />
capable of providing real-time measurements of<br />
variables whose importance is apparent for the<br />
evaluation of forest fire hazard. Other sensors<br />
may be used to acquire information related to fuel<br />
moisture. For instance, the use of remote sensors<br />
(installed on aircraft or satellites) may provide<br />
real-time information about the state of<br />
vegetation. Such information may refer, for<br />
example, to the tissue moisture content or to some<br />
measure of water deficit index (Burgan et al.,<br />
1996). In addition, remote ground sensors can<br />
provide real-time information about the<br />
meteorological conditions over the territory, or<br />
data concerning the drought level of vegetation.<br />
Finally, valuable dynamic information is also<br />
represented by the outputs of a meteorological<br />
model, assuming that the forecasts over a suitably<br />
long horizon [e.g., 48-72 hours] can be considered<br />
sufficiently reliable. Note that, in principle,<br />
several different meteorological models can yield<br />
a notion about the uncertainty of the<br />
meteorological forecasts. As such forecasts are<br />
provided by the model over a time horizon, even<br />
the forest fire hazard assessment that is obtained<br />
by the overall module in Fig. 2 refers to such a<br />
horizon, <strong>and</strong> thus may be represented as a set of<br />
functions of time <strong>and</strong> space, one for each physical<br />
variable that is assumed to be significant to assess<br />
forest fire hazard.<br />
For the sake of simplicity, the above functions<br />
may be chosen to be discrete both in time <strong>and</strong><br />
space. A suitable choice for the time<br />
discretization interval <strong>and</strong> the space discretization<br />
grid is that of taking the same time-space<br />
discretization that characterizes the outcomes of<br />
the meteorological model. In particular, the<br />
variables that are determined to evaluate <strong>and</strong><br />
represent the forest fire hazard on each cell of the<br />
space grid (<strong>and</strong> for each time interval) are the rate<br />
of spread <strong>and</strong> the linear intensity that a fire could<br />
assume (in case of ignition). Note that the<br />
former’s potential rate of spread is obtained<br />
through the application of a model that is not used<br />
to evaluate the dynamics of a given fire, but to<br />
evaluate the physical characteristics that a fire<br />
could take on, in each cell, on the basis of the<br />
variables that locally condition the possibility of a<br />
successful ignition <strong>and</strong> fire propagation.<br />
The system receives the daily outputs of a<br />
meteorological Limited Area Model (LAM),<br />
namely Lokal Modell (Doms <strong>and</strong> Schättler,<br />
1999), consisting of a set of data discretized in<br />
time steps of three hours over a time horizon of<br />
72 hours, <strong>and</strong> defined over a regular grid<br />
composed of 57.200 cells having a side<br />
corresponding to 0,05 degrees. The available<br />
meteorological (forecast) variables are the<br />
cumulated rainfall [m] in each three-hour time<br />
interval, air temperature [K], dew point<br />
temperature [K], wind speed [m s-1], <strong>and</strong> wind<br />
direction [rad]. In addition, a Digital Elevation<br />
Model (DEM), defined over a regular grid of<br />
5.000 m side, has been utilized to represent the<br />
topography of the target area. This model is used<br />
to define the average value of the aspect angle<br />
[deg], the slope [%], <strong>and</strong> the elevation [m] of each<br />
grid cell.<br />
749
Topography <strong>and</strong><br />
territorial data (GIS)<br />
Vegetation<br />
data<br />
Weather<br />
conditions<br />
Meteorological<br />
forecast<br />
Other<br />
sensor data<br />
Static hazard<br />
distribution<br />
Fuel moisture model<br />
Fuel moisture<br />
behaviour<br />
Potential fire spread<br />
model<br />
Dynamic<br />
hazard<br />
distribution<br />
Figure 1. A schematic representation of the structure of the dynamic forest fire hazard assessment module<br />
The vegetational characteristics have been<br />
introduced in the system by means of a vectorial<br />
map of the whole area (e.g., the CORINE L<strong>and</strong><br />
Cover map). In the proposed procedure, a regular<br />
grid of 0,0125 degrees of side length is<br />
introduced, in order to define the distribution of<br />
fuels in space <strong>and</strong> their characterization in terms<br />
of physical-chemical properties. The available<br />
fuels can be described by specifying their<br />
morphological parameters <strong>and</strong> the behaviour of<br />
their physiological variables. It is assumed that<br />
the values of such morphological parameters<br />
cyclically vary over the year. In this connection, a<br />
quasi-stationary model has been adopted, <strong>and</strong> the<br />
average seasonal value has been used for any<br />
parameter. For each different reported species, the<br />
seasonal fuel loads [kg m -2 ], has been obtained<br />
from the literature (Anderson, 1982; Nunez-<br />
Regueira et al., 1999) <strong>and</strong> then organized in a GIS<br />
database. Moreover, the physiological<br />
characteristics of the fuels, that is the average<br />
seasonal tissue moisture content [%] of live fuel<br />
for each species, have been also introduced in the<br />
GIS database, as well as the average seasonal<br />
Higher Heating Value (HHV) [kJ kg -1 ], both for<br />
dead <strong>and</strong> live fuels.<br />
In the present implementation, the only dynamic<br />
information provided to the models, is represented<br />
by meteorological forecasts, <strong>and</strong> other kinds of<br />
dynamic information are not used. Nevertheless,<br />
as it will be discussed in the next sections, even in<br />
its present version, the module can be considered<br />
as on operative tool, <strong>and</strong> the results of its<br />
applications are quite encouraging.<br />
3. THE FUEL MOISTURE MODEL<br />
In the present implementation of the dynamic fire<br />
hazard assessment system, only the dynamics of<br />
the dead fine fuel is modelled. Instead, the live<br />
fuel moisture is considered practically timeinvariant,<br />
<strong>and</strong> is provided by values<br />
corresponding to the specific vegetation cover <strong>and</strong><br />
to the considered season (Brown et al., 1989).<br />
The dynamics of the dead fine fuel moisture is<br />
represented by using for each cell k over the<br />
considered region a specific model, which does<br />
not interact with the models of the other cells, as<br />
no fire propagation is represented. Then, let<br />
u o k<br />
( t)<br />
represent the dead fine fuel moisture at cell<br />
k at time instant t. It is assumed that the evolution<br />
of the above quantity is governed by the<br />
differential equation<br />
du<br />
o<br />
k<br />
( t)<br />
o<br />
( t) − K u ( t)<br />
= K step<br />
(1)<br />
1<br />
2<br />
k<br />
dt<br />
where step(t) is the unit step function 1 . In fact, the<br />
solution of (1) has an asymptotic behaviour<br />
determined only by the ratio (K 1 /K 2 ), namely<br />
u<br />
( t)<br />
K<br />
=<br />
( 0)<br />
o<br />
o 2 u − K<br />
k 1 2<br />
k<br />
K<br />
2<br />
e<br />
K<br />
K<br />
− K t<br />
1<br />
step( t) + step( t)<br />
(2)<br />
Of course, the asymptotic value (K 1 /K 2 ) is<br />
independent of the initial state u k (0), <strong>and</strong> the<br />
transient behaviour decays (increases) if u k (0) ><br />
(K 1 /K 2 ) (u k (0) < (K 1 /K 2 ). Observe that the “time<br />
constant” characterizing the speed at which the<br />
transient term in the r.h.s. of (2) vanishes is given<br />
by 1/K 2 .<br />
Note that the solution (2) of eq. (1) is correct only<br />
in the assumption of time-invariance of<br />
coefficients K 1 <strong>and</strong> K 2 , which, however, as will be<br />
discussed below, must be considered timevarying,<br />
since their values depend on a set of<br />
meteorological variables. Thus, the use of<br />
solution (2) is correct only whenever the<br />
dynamics of model (1) (which is characterized by<br />
the time constant (1/K 2 )) is considerably slower<br />
than meteorological dynamics (determining the<br />
1 Function step[x] is defined as equal to 1 if x≥0<br />
<strong>and</strong> equal to 0 otherwise.<br />
2<br />
750
variation of K 1 <strong>and</strong> K 2 ). However, discretization<br />
of (2) is in any case allowed, even when K 1 <strong>and</strong><br />
K 2 are significantly time-varying. For this reason,<br />
hereafter the dependence of such coefficients on<br />
time (<strong>and</strong> on cell k) will be explicitly recalled by<br />
the notation.<br />
It is assumed that coefficients K 1,k (t) <strong>and</strong> K 2,k (t)<br />
are functions of the meteorological variables p k (t),<br />
w k (t), ρ k (t), τ k (t), that is the cumulated rain p k (t)<br />
[m], the wind intensity w k (t) [m s -1 , rad], the<br />
relative humidity ρ k (t) [%], <strong>and</strong> the air<br />
temperature τ k [K]. In the proposed model,<br />
instead of trying to model such a dependence<br />
through thermodynamic considerations, a semiphysical<br />
structure is proposed by assuming that<br />
the asymptotic value (K 1,k (t)/K 2,k (t)) can be<br />
expressed as a function of the meteorological<br />
variables as follows<br />
K<br />
K<br />
K<br />
K<br />
2,k<br />
( t)<br />
( t)<br />
ρk<br />
( t)<br />
+α1<br />
α2+α3τ<br />
( t<br />
= e<br />
)<br />
if p k (t) ≤ p* (3)<br />
1 ,k<br />
k<br />
1,k ( t)<br />
β1<br />
2,k ( t) = if p k (t) > p* (4)<br />
where α i (i=1,..,3), β 1 are constants having<br />
suitable dimensions <strong>and</strong> p * [m] is a threshold<br />
value for the cumulated rain. Note that (3) applies<br />
in the absence of significant rainfall (in the last<br />
time interval), whereas (4) holds true whenever<br />
such a rainfall cannot be neglected. Of course, the<br />
constant values must be selected so that<br />
ρk<br />
( t)<br />
+α1<br />
α2<br />
+α3τk<br />
( t<br />
β > e<br />
)<br />
(5)<br />
1<br />
for any possible value of ρ k (t), <strong>and</strong> τ k (t). Note that<br />
the dependence of the r.h.s. of (3) on ρ k (t) can be<br />
justified by observing that the higher the value of<br />
ρ k (t) is the higher the asymptotic value of<br />
u o k<br />
( t)<br />
will be 2 . Besides, the fact that the r.h.s. of<br />
(4) is independent of p k (t) can be justified by the<br />
assumption that the asymptotic values of the fuel<br />
moisture are independent of the rainfall intensity<br />
(whenever such an intensity exceeds a certain<br />
threshold). Finally, the fuel moisture is<br />
uncorrelated with temperature <strong>and</strong> humidity in<br />
case of rain, since rain raises the fuel moisture<br />
condition to the fiber saturation point, which is<br />
greater than 35% (Cheney, 1981).<br />
As regards the dependency of K 2,k (t) from<br />
meteorological variables, recalling that 1/K 2,k (t) is<br />
the time constant which (in time-invariant<br />
2 Note that such an asymptotic values is actually only<br />
“potential”, as it is achieved only when meteorological<br />
conditions are time-invariant.<br />
meteorological conditions) characterizes the<br />
transient behaviour represented in (2), makes so<br />
that, in absence of a significant rainfall, a high<br />
value of temperature <strong>and</strong> wind intensity favours<br />
drying but hampers moistening. Instead, in<br />
presence of significant rainfall, provides a linear<br />
dependence of the moistening speed on the<br />
rainfall intensity.<br />
At this point, after model (1) has been discussed,<br />
also as regard the dependence of the coefficients<br />
appearing in (1) on the meteorological variables,<br />
it is worth explicitly providing the discretized<br />
version of the model that is actually implemented<br />
in the developed system, namely<br />
0<br />
u<br />
k<br />
[ ] + T K ( t)<br />
0<br />
( t 1) = u ( t) 1−<br />
T K ( t)<br />
+ 2,k<br />
1, k (6)<br />
k<br />
where T is the length of the discretization interval<br />
(3 hours), <strong>and</strong> the time variable t is now an integer<br />
number.<br />
The behaviour of the fuel moisture model is<br />
deeply affected by the value of the parameters; an<br />
accurate calibration of such parameters could take<br />
place by means of suitable parameter fitting<br />
techniques <strong>and</strong> on the basis of a wide set of real<br />
data. Such a calibration is beyond the scope of<br />
this paper, also because of the difficulty of<br />
obtaining reliable <strong>and</strong> significantly distributed<br />
data regarding fuel moisture. An estimate of such<br />
parameters derive from empirical evidence over a<br />
wide set of test cases (relevant to detected fires)<br />
that will be described in the following <strong>and</strong> that<br />
refer to the overall performance analysis of the<br />
integrated system consisting of the cascade of the<br />
fuel moisture model <strong>and</strong> the potential spread<br />
model.<br />
4. THE POTENTIAL SPREAD MODEL<br />
The dynamic information that this model uses is<br />
that related to meteorological forecast variables,<br />
<strong>and</strong> that provided by the fuel moisture model. At<br />
the same time, the propagation model makes use<br />
of information related to topography <strong>and</strong><br />
vegetation (kind <strong>and</strong> density per m2), again<br />
referred to the considered cells. The information<br />
concerning density <strong>and</strong> kind of dead <strong>and</strong> live fuel<br />
for each cell is considered as static, since only<br />
seasonal variations are considered simply by<br />
taking into account different values of the relevant<br />
parameters for the various seasons.<br />
The development of the potential fire spread<br />
model follows the same basic lines first proposed<br />
by Drouet (1974) for the definition of a forest fire<br />
propagation model, but introducing some<br />
important novelties as regards the procedures to<br />
evaluate the forest fire hazard.<br />
751
The first information on which the potential<br />
spread model is built is represented by the<br />
nominal rate of spread v 0,k , which is a quantity<br />
referring to st<strong>and</strong>ard conditions as regards the<br />
temperature <strong>and</strong> the average live fuel moisture, in<br />
absence of wind, within a perfectly flat terrain,<br />
<strong>and</strong> with perfectly dry dead fuel. Obviously, v 0,k<br />
depends on cell index k, as it depends on the kind<br />
of fuel (i.e. particle size, bulk density, moisture,<br />
<strong>and</strong> chemical composition of the fuel) <strong>and</strong> on the<br />
vegetation density (biomass per square meter) of<br />
live <strong>and</strong> dead fuel. Besides, such a value has to be<br />
specified in connection to the various seasonal<br />
conditions, as they determine the average<br />
moisture of live fuel. Clearly, the determination of<br />
v 0,k needs a great amount of experimental tests<br />
<strong>and</strong> a deep knowledge on the vegetation covering<br />
over the territory. On this basis, the potential rate<br />
of spread, which takes into account the influence<br />
of the meteorological variables, can be defined<br />
<strong>and</strong> determined as follows<br />
( t)<br />
( t)<br />
Wk<br />
vk ( t) = v0,k<br />
Zk<br />
( t)<br />
Sk<br />
Vk<br />
( t)<br />
(7)<br />
N<br />
where<br />
k<br />
Z k (t) is a (multiplicative) correction<br />
[dimensionless] due to air temperature, at time t<br />
<strong>and</strong> in cell k, with respect to the std. temperature<br />
(0°C) assumed as the reference one;<br />
W k (t) is a (multiplicative) correction<br />
[dimensionless] due to wind speed on flat terrain,<br />
at time t <strong>and</strong> in cell k;<br />
N k (t) is a normalization term [dimensionless]<br />
which takes into account the influence of<br />
topography on coefficient W k (t);<br />
S k is a (multiplicative) correction [dimensionless]<br />
due to the slope of the cell k;<br />
V k (t) is a (multiplicative) correction<br />
[dimensionless] due to the dead fine fuel<br />
moisture, at time t <strong>and</strong> in cell k.<br />
The correction terms introduced above are<br />
defined as parametric functions of the considered<br />
information (slope, aspect, meteo variables),<br />
based on empirical evidence.<br />
Having thus clarified the way to compute the<br />
potential rate of spread v k (t), which provides a<br />
quantification of the swiftness characterizing the<br />
(potential) spread of a fire, it is necessary to also<br />
quantify the intensity of the phenomenon, which<br />
is the ultimate measure of the hazard. To this end,<br />
Byram’s equation (1959) can be used to<br />
determine the (potential) fire linear intensity I k (t)<br />
[kW m -1 ], namely<br />
k<br />
1<br />
( t) = vk<br />
( t)∑<br />
i=<br />
0<br />
i<br />
k<br />
I LHV d<br />
(8)<br />
where<br />
0<br />
k<br />
1<br />
k<br />
d , ( d ) [kg m -2 ] is the density of dead fuel (live<br />
fuel), for the considered season in cell k;<br />
( t)<br />
1<br />
k<br />
i<br />
k<br />
LHVk 0 , ( LHV ) is the Lower Heating Value<br />
[kJ kg -1 ] of the dead fine fuel (live fuel) in cell k<br />
at time t, given by:<br />
0<br />
k<br />
0<br />
k<br />
[ 1−<br />
u (t)] − Q u (t)<br />
LHV (t) = HHV<br />
(9)<br />
1<br />
k<br />
LHV<br />
where<br />
0<br />
k<br />
1<br />
k<br />
0<br />
k<br />
1 1<br />
[ 1−<br />
u k ] − Q u k<br />
= HHV<br />
(10)<br />
1<br />
k<br />
HHV , ( HHV ) is the Higher Heating Value [kJ<br />
kg-1] of the dead fine fuel (live fuel) based on the<br />
prevailing species composition in cell k.<br />
5. SOFTWARE IMPLEMENTATION<br />
Since August 2003 a prototype of the system<br />
provides to Italian Civil Protection a daily fire<br />
hazard assessment for a 72 hours time interval.<br />
The system has been implemented in a MS Visual<br />
C ++ procedure integrate, as it shown in Fig. 2, in a<br />
pre-existent GUI integrated in a dedicated<br />
network <strong>and</strong> used by the Civil Protection for the<br />
data processing <strong>and</strong> the visualization of<br />
information relevant to the other natural hazards.<br />
The system receives daily at 6:00 AM from a<br />
remote station 120 ASCII files (66 MB)<br />
elaborated by the LAM run of 00:00 AM <strong>and</strong><br />
relevant to the national meteorological forecast<br />
for the next 72 hours. As it concerns the static<br />
information, the Italian vegetational cover <strong>and</strong> the<br />
topographic characteristics are stored in 1 file of<br />
1025 KB. The computation environment is a<br />
Windows XP operational system equipped with<br />
AMD Athlon XP 2000 2 GHz CPU, 256 MB<br />
RAM. The time needed for the creation of the<br />
whole set of files is about 50 seconds. The output<br />
files are 7 (10 MB), defined for each time 3-hour<br />
interval belonging to the 72 hours time horizon,<br />
<strong>and</strong> for each cell of 0,05° side covering the whole<br />
target area, i.e. the Italian territory, measuring<br />
302.000 km 2 Each file defines the following<br />
greatness: the air temperature contribution to the<br />
rate of spread [dimensionless], the wind speed<br />
contribution to the rate of spread [dimensionless],<br />
the dead fine fuel moisture [%], the maximum<br />
rate of spread [m h -1 ], the rate of spread [m h -1 ],<br />
the linear intensity for each cell [kW/m], <strong>and</strong> the<br />
linear intensity aggregates for each Italian<br />
regional district [kW/m]. Each output file is in<br />
0<br />
k<br />
752
ASCII format <strong>and</strong> is composed by 26 column <strong>and</strong><br />
13.265 rows. The first <strong>and</strong> second column<br />
represents the coordinates of the cell, whereas the<br />
other 24 columns represent the values of each<br />
variable for each time interval.<br />
the Italian Civil Protection Department, which is<br />
charged to manage <strong>and</strong> dispatch the fleet of<br />
amphibious water bomber (CL 415 Canadair) <strong>and</strong><br />
heavy helitanker (S64F Air Crane) on national<br />
high-risk areas or on signalled active fires.<br />
7. ACKNOWLEDGEMENT<br />
The activities reported in the paper are presently<br />
carried out with reference to the case study<br />
relevant to the Italian Civil Protection <strong>and</strong> have<br />
been funded by the Gruppo Nazionale per la<br />
Difesa dalle catastrofi Idrogeologiche GNDCI,<br />
U.O. n. 3.28, Special project n. 4 Structural <strong>and</strong><br />
operational design of a decision support system<br />
based on a national geographical information<br />
system <strong>and</strong> aiming at forest fire risk management.<br />
Figure 2. The Civil Protection GUI outputs<br />
relevant to forest fires dynamic hazard assessment<br />
over the Italian territory. The (animated) image at<br />
left side of Fig. 2 is the potential rate of spread [m<br />
h -1 ] for the next 24 hours, whereas at right the<br />
linear intensity [kW m -1 ] relevant to the same<br />
time interval is reported.<br />
Figure 3. On the left the dead fine fuel moisture<br />
[%] relevant to a 24 hour forecast; on the right the<br />
air temperature contribution to the rate of spread<br />
[dimensionless] relevant to the same time interval.<br />
6. CONCLUSIONS AND FURTHER<br />
RESEARCH DIRECTIONS<br />
Several practical as well as conceptual problems<br />
remain to be investigated to assess the validity<br />
<strong>and</strong> the practical relevance of the proposed<br />
approach. Experimental evaluation with reference<br />
to a real case study is presently carried out within<br />
8. REFERENCES<br />
Anderson, H.E. Aids to determining fuel models<br />
for estimating fire behavior. USDA, Forest<br />
Service General Technical Report INT-<br />
122. Intermountain Forest <strong>and</strong> Range<br />
Experiment Station, Ogden, UT. 22 p,<br />
1982.<br />
Brown, J.K., G.D. Booth, <strong>and</strong> D.G. Simmerman.<br />
Seasonal change in live fuel moisture of<br />
understory plants in western U.S. Aspen.<br />
In MacIver DC, Auld H, Whitewood R<br />
(editors). Proceedings from 10th<br />
Conference on Fire <strong>and</strong> Forest.<br />
Meteorology Environment Canada,<br />
Forestry Canada, Ottawa, Ontario, p. 406-<br />
412, 1989.<br />
Byram, G.M. Combustion of forest fuels. In<br />
Forest Fire: Control <strong>and</strong> Use. Editor Davis<br />
KP, McGraw-Hill, New York, pp. 113-<br />
126, 1959.<br />
Cheney, N.P. Fire behaviour. In 'Fire <strong>and</strong> the<br />
Australian biota'. Australian Academy of<br />
Science, Canberra, AUS. pp. 151-175,<br />
1981.<br />
Doms, G., U. Schättler. The Non-hydrostatic<br />
Limited-Area Model LM (Lokal-Modell)<br />
of DWD. Deutscher Wetterdienst.<br />
Offenbach, Germany, 1999.<br />
Drouet, J.C. Theorie de la propagation des feux de<br />
forets. Master Thesis Universitè d’Aix-<br />
Marseille, France, 1974.<br />
Nunez-Regueira, L, J. Rodriguez, J. Proupin, <strong>and</strong><br />
B. Mourino. Design of forest biomass<br />
energetic maps as a tool to fight forest<br />
wildfires. Termochimica Acta, 328, pp.<br />
111-120, 1999.<br />
753
Linking Narrative Storylines <strong>and</strong> Quantitative Models<br />
To Combat Desertification in the Guadalentín, Spain<br />
Kasper Kok 1 , Hedwig van Delden 2<br />
1 Laboratory of Soil Science <strong>and</strong> Geology, Wageningen University, Wageningen, the Netherl<strong>and</strong>s.<br />
E-mail: kasper.kok@wur.nl<br />
2 Research Institute for Knowledge Systems, Maastricht, the Netherl<strong>and</strong>s<br />
Abstract: Desertification in Spain is a largely society-driven process, which can be effectively managed only<br />
through an underst<strong>and</strong>ing of ecological, socio-cultural <strong>and</strong> economic driving forces. This calls for a more active<br />
role of decision makers <strong>and</strong> other stakeholders. We present a promising approach, involving stakeholders in the<br />
scenario development process <strong>and</strong> linking these narrative storylines with an integrated quantitative model. Within<br />
the framework of a larger EC-financed project, dealing with desertification in the Mediterranean region, multiscale<br />
scenarios were developed for Europe, the Northern Mediterranean <strong>and</strong> four local areas. In the same project<br />
a Policy Support System (PSS) was developed. The main objective of the present exercise was to establish a link<br />
between the qualitative scenarios <strong>and</strong> the PSS for the watershed of the Guadalentín River in Spain. From the<br />
results of two scenario workshops, three scenarios were selected, all linked to the same Mediterranean scenario.<br />
Our selection aimed at maximising both the variety in the narrative storylines <strong>and</strong> the expected output of the PSS.<br />
The scenarios were subsequently formalised, ensuring that the same information was present for all three<br />
scenarios; semi-quantified ("translated") by linking them to the main entry points of the PSS; <strong>and</strong> quantified by<br />
parameterisation of the model. Although model runs have not yet been carried out, preliminary results indicate<br />
the potential for the constructed quantitative scenarios. The paper illustrates the practical potential <strong>and</strong> pitfalls of<br />
linking qualitative storylines <strong>and</strong> quantitative models. Future research should, however, also focus on the more<br />
fundamental theoretical obstacles that are easily overlooked.<br />
Keywords: scenarios; spatial modelling; Policy Support System; participatory; linking qualitative <strong>and</strong><br />
quantitative methods<br />
1. INTRODUCTION<br />
Desertification in Spain is largely a society-driven<br />
problem, which can be effectively managed only<br />
through a thorough underst<strong>and</strong>ing of the principal<br />
ecological, socio-cultural, <strong>and</strong> economic driving<br />
forces [UNCCD, 1994]. This Integrated Assessment<br />
approach also calls for a much more active role of<br />
decision makers <strong>and</strong> other local stakeholders during<br />
all phases of the process [Rotmans, 1998]. A<br />
particularly pressing issue is establishing the link<br />
between qualitative outputs from employing<br />
participatory methods <strong>and</strong> quantitative, data<br />
dem<strong>and</strong>ing, spatially explicit models. To tackle the<br />
problem, various different methods are being<br />
developed, including e.g. Agent Based Models<br />
[Parker et al., 2002], that can be directly<br />
parameterised by stakeholders [e.g. Barreteau et al.,<br />
2001]. Here we present another approach by linking<br />
qualitative narrative storylines, developed during<br />
scenario workshops, <strong>and</strong> a Policy Support System<br />
(PSS). The work is part of a larger European<br />
project, MedAction.<br />
1.1 MedAction<br />
MedAction (see Appendix) is an EC-financed<br />
project within which an information <strong>and</strong> decisionsupport<br />
base on l<strong>and</strong> degradation is being developed<br />
754
Table 1. Main results <strong>and</strong> methods employed during stakeholder<br />
scenario workshops within Module 1 of MedAction.<br />
Present<br />
(2003)<br />
Short term<br />
(2008)<br />
Long term<br />
(2008-2030)<br />
Long term<br />
(2030)<br />
Workshop #<br />
(date)<br />
1<br />
(Oct/Nov 2002)<br />
2<br />
(Jun/Jul 2003)<br />
2<br />
(Jun/Jul 2003)<br />
1<br />
(Oct/Nov 2002)<br />
Grouping Individual <strong>and</strong> All Groups <strong>and</strong> All Groups Groups<br />
Main method Post-its <strong>and</strong> discussion Discussion Backcasting Collage <strong>and</strong><br />
forecasting<br />
Results Main factors Major current trends Desirable futures 'Real' futures<br />
to assist decision-makers from the local to the<br />
European level in the formal <strong>and</strong> informal decision<br />
<strong>and</strong> policy making process to combat desertification<br />
in the Northern Mediterranean Region. The specific<br />
problems of desertification <strong>and</strong> mitigation measures<br />
are addressed at the European, Mediterranean <strong>and</strong><br />
local scale, with the ultimate goal being to aid local<br />
decision-making with regard to policy formulation<br />
for sustainable l<strong>and</strong> management at the local level.<br />
Work was carried out in four local case studies: the<br />
Guadalentín (Spain); Val d'Agri (Italy); Alentejo<br />
(Portugal); <strong>and</strong> the isl<strong>and</strong> of Lesbos (Greece). A<br />
simplified flow-chart of the main activities within<br />
MedAction is given in Figure 1, highlighting<br />
components important in this paper.<br />
of the future in 2030 that was obtained during a<br />
forecasting [see also Kasemir et al., 2000] session;<br />
an extension of the present representing the<br />
situation in 2008 based on an extrapolation of<br />
current trends; <strong>and</strong> a backcasting exercise [Dreborg,<br />
1996; Robinson, 2003], reasoning back from a<br />
desirable end-point in 2030 to short-term<br />
measurements that are necessary to realise this<br />
future. The diversity of methods has resulted in a<br />
good overview of the perception of stakeholders on<br />
the present situation; short-term fears; <strong>and</strong> longterm<br />
hopes <strong>and</strong> expectations. The methods <strong>and</strong><br />
results are summarised in Table 1; a full description<br />
can be can be downloaded on<br />
http://www.icis.unimaas.nl/medaction/download.ht<br />
ml.<br />
MODULE 3:<br />
Decision<br />
Support<br />
Systems<br />
MODULE 1:<br />
Multi-scale<br />
scenarios<br />
Cost<br />
Benefit<br />
Analysis<br />
Impact of<br />
past<br />
policies<br />
Stakeholders in<br />
four local regions<br />
Desertification Policy Support Framework<br />
Figure 1. Simplified flow-chart of main activities<br />
within MedAction. Grey shades indicate<br />
components important in the paper.<br />
Module 1 of MedAction was coordinated at the<br />
<strong>International</strong> Centre for Integrative Studies (ICIS)<br />
in Maastricht, the Netherl<strong>and</strong>s, <strong>and</strong> dealt with<br />
scenario development at European [Kok et al.,<br />
2003]; Mediterranean [Kok <strong>and</strong> Rothman, 2003],<br />
<strong>and</strong> local [Kok <strong>and</strong> Patel, 2003] levels. Local<br />
scenarios for the various local case studies were<br />
developed during series of workshops with 20-25<br />
local <strong>and</strong> regional stakeholders. Various scenariodevelopment<br />
methods were tested. Four main<br />
products can be distinguished: a story of the present<br />
characterising the perception of the local<br />
stakeholders on the situation in their region; a story<br />
Figure 2. Simplified structure of the<br />
Policy Support System as developed<br />
in Module 3 of MedAction.<br />
Module 3 dealt among other things with the<br />
development of a Policy Support System (see<br />
Figure 2): a software instrument to support policymaking<br />
at the regional level, developed by the<br />
Research Institute for Knowledge Systems. The<br />
MedAction PSS has been developed with the<br />
objective to address a number of policy themes<br />
concerning water resources, sustainable agriculture,<br />
desertification <strong>and</strong> l<strong>and</strong> degradation in<br />
Mediterranean regions. Problems, goals, policy<br />
options <strong>and</strong> policy indicators have been collected<br />
755
<strong>and</strong> structured for each of these themes, <strong>and</strong><br />
translated into a conceptual framework. From this<br />
conceptual framework a policy support system is<br />
being designed <strong>and</strong> developed incorporating socioeconomic<br />
as well as physical models. The PSS<br />
supports policy-makers in underst<strong>and</strong>ing the<br />
impacts of autonomous developments within a<br />
region as well as the impacts of external influences<br />
on the region, such as economic <strong>and</strong> demographic<br />
growth <strong>and</strong> climate change. All impacts can be<br />
measured by means of a number of policy relevant<br />
indicators (e.g. profits in the agricultural sector,<br />
forested area, suitability of the soil for agriculture or<br />
natural vegetation, water use <strong>and</strong> availability, l<strong>and</strong><br />
use), which change dynamically during the run of a<br />
simulation. The PSS is developed as a generic<br />
system <strong>and</strong> is applied in particular to the<br />
Guadalentín river basin in Spain. Previous versions<br />
of this system have also been applied to the Marina<br />
Baixa region in Spain <strong>and</strong> the Argolidas region in<br />
Greece. The MedAction PSS will be described in<br />
more detail in the presentation of Van Delden et al.<br />
(2004) at this conference. More information on the<br />
predecessor of the MedAction PSS, MODULUS,<br />
can be found in Engelen [2003] <strong>and</strong> Oxley et al. [in<br />
press].<br />
1.2 Objectives<br />
The main objective of this paper is to establish the<br />
link between the qualitative scenarios developed in<br />
Module 1 <strong>and</strong> the PSS developed in Module 3 of<br />
MedAction. This paper focuses on the practical<br />
application in the watershed of the Guadalentín in<br />
southeastern Spain.<br />
2. METHODS<br />
2.1 Selecting the Scenarios<br />
The series of workshops in the Guadalentín yielded<br />
a wealth of scenarios. We focused our selection on<br />
the forecasting <strong>and</strong> backcasting exercises, resulting<br />
in seven local scenarios, linked to three<br />
Mediterranean scenarios. From these we selected,<br />
for the exercise described, those two scenarios that<br />
were linked with the European scenario Convulsive<br />
Change, in which climate change is as disruptive as<br />
some are now predicting, triggering a series of<br />
severe droughts <strong>and</strong> desert formation. The<br />
forecasting scenario (Scenario I: Likely future)<br />
provides a most likely future under these<br />
Mediterranean developments; the backcasting<br />
scenario (Scenario II: Desired future) is based on<br />
the desirably future of a strong agricultural sector.<br />
A third scenario (Scenario III: Water shortage) was<br />
added, where one of the key assumptions in the first<br />
two scenarios – the construction of a canal from the<br />
Ebro River– was omitted, thus strongly limiting<br />
water availability <strong>and</strong> the effects thereof. This third<br />
scenario was thus not directly formulated by the<br />
stakeholders, although the possibility was discussed<br />
during the workshops.<br />
The scenarios were chosen to maximise both the<br />
variety present in the narrative stories, as well as the<br />
variety of the expected spatially explicit results. The<br />
link between the narrative storylines <strong>and</strong> the PSS<br />
was not complete. Social changes in the narratives,<br />
for example, could not always be quantified, while<br />
strongly non-linear changes in the stories could not<br />
always be incorporated in the PSS. Similarly,<br />
detailed information on e.g. soil conservation<br />
measurements <strong>and</strong> l<strong>and</strong> management practices in<br />
the PSS could not be extracted from the narrative<br />
storylines <strong>and</strong> were set at default values.<br />
2.2 Formalising the Scenarios<br />
These three narrative scenarios were formalised<br />
using the Factor – Actor – Sector framework that<br />
was also used in the development of the European<br />
<strong>and</strong> Mediterranean scenarios (see Table 1). This<br />
helped in maintaining the links with higher level<br />
scenarios. The developments remain qualitative.<br />
2.3 Translating the Scenarios<br />
These formalised stories were then quantified to the<br />
extent possible by linking them to the main entry<br />
points of the PSS. First a selection was made of<br />
parameters in the PSS that have a link with the<br />
scenarios as formalised. For each of these<br />
parameters, it was then indicated what the expected<br />
change was in each of the three scenarios. Change<br />
was semi-quantitative, ranging from +++ (very<br />
strong increase) to --- (very strong decrease). This<br />
methodology was also applied in earlier work by<br />
White et al. [2004]. Below are the most important<br />
parameters in the PSS that were considered during<br />
this translation. Grouping is by main components in<br />
the PSS as depicted in Figure 2.<br />
756
Table 2. Summary of the formalised scenarios used as input in the PSS,<br />
by the main Factors, Actors <strong>and</strong> Sectors.<br />
FAS Scenario I Scenario II Scenario III<br />
Factors<br />
Water availability Increasingly limited due to Limited, distribution favours Strongly limited, no "Ebro<br />
drought<br />
agriculture<br />
water"<br />
Migration<br />
Rural-urban migration<br />
Fewer permanent tourists Strong rural-urban<br />
European Sunbelt<br />
migration, less immigrants<br />
Morocco<br />
Sectors<br />
Agriculture Increasingly difficult position Multi-functional, favoured for<br />
water<br />
Lack of water, although<br />
still favoured<br />
Tourism Booming Eco-tourism, less in numbers Lack of water stops<br />
expansion<br />
Actors<br />
Businesses Large-scale, mass tourism,<br />
smallholders disappear. Industry<br />
important<br />
Small-scale favoured, industry<br />
under pressure<br />
Lack of water limits<br />
developments<br />
LAND USE MODULE: Total l<strong>and</strong> dem<strong>and</strong> for<br />
Agriculture; Rural residential; Dense residential;<br />
Industry & commercial areas; Tourism; Expats,<br />
Forest reserves.<br />
CLIMATE MODULE: Scenario for future climate<br />
change, based on IPCC scenarios. Main factors are<br />
precipitation, temperature <strong>and</strong> radiation.<br />
WATER MANAGEMENT MODULE: Defined by<br />
resource (aquifer, reservoirs (including Tajo <strong>and</strong><br />
Ebro water), desalinised sea water) <strong>and</strong> by function<br />
for the dem<strong>and</strong>s. Three main parameters were<br />
included: price (also per source); quantity (per<br />
source, per sector or per person/hectare);<br />
distribution (per sector).<br />
An important input in this module is the<br />
presence/absence of irrigated agriculture,<br />
represented by a binary map showing where<br />
irrigation from each water source is possible; with a<br />
choice between drip or spray irrigation.<br />
FARMER’S DECISION MODULE: The choice for<br />
different crop types (including no crop or<br />
ab<strong>and</strong>oned l<strong>and</strong>) depends on, among other things,<br />
the market price of crops, subsidies, taxes, farmers'<br />
resistance to change, water availability <strong>and</strong> the<br />
calculated yield or suitability. Parameters adapted<br />
based on the scenarios were market prices,<br />
subsidies, farmer’s resistance to change, <strong>and</strong> the<br />
introduction of "new" crops which are better<br />
resistant to dry soils.<br />
OTHER: Policy relevant parameters include: zoning<br />
maps for each function, construction of new roads,<br />
canals <strong>and</strong> check dams, dredging of the reservoirs,<br />
terracing, ploughing.<br />
2.4 Quantifying the Scenarios<br />
The last step was the actual parameterisation of the<br />
PSS. We used a baseline scenario for all parameters<br />
that had no relation with the narrative storylines. For<br />
example, there are detailed modules for hydrology,<br />
soil erosion, salinisation <strong>and</strong> plant growth, for which<br />
not much information could be extracted from the<br />
qualitative scenarios. The output of these modules is,<br />
however, influenced by the impacts of the different<br />
scenarios. The possible futures in turn are also<br />
influenced by the output of the modules, since the<br />
core of the PSS is an integrated dynamic model with<br />
strong feedback loops between the processes<br />
represented. In the parameterisation process, we were<br />
as consistent as possible. In general, "+++" translated<br />
into 3% more <strong>and</strong> "---" into 3% less. However, many<br />
small additional assumptions were necessary, given<br />
the amount of parameters in the PSS that were not<br />
explicitly referred to in the narrative stories.<br />
2.5 Running the Model<br />
Unfortunately, the final results of the model runs were<br />
not available in time to be included in the proceedings<br />
of the conference. Preliminary results indicate that the<br />
three scenarios translate into significantly different<br />
l<strong>and</strong> use patterns. However, a full analysis of how the<br />
variability of the input scenario translated into a<br />
variability of the output maps has not been conducted<br />
yet.<br />
In order to get an idea of how the results will look<br />
like, a few of the input <strong>and</strong> dynamic output maps of<br />
757
Figure 3. Examples of maps of the Guadalentín<br />
watershed, used as input in the Policy Support<br />
System.<br />
degradation, but a key input in the crop choice model.<br />
The ESA map is also calculated on a yearly basis, but<br />
is not used as an input in other modules.<br />
3. DISCUSSION<br />
Figure 3a. L<strong>and</strong> use patterns. Black: built-up<br />
areas; dark grey: irrigated agriculture; medium<br />
grey: dryl<strong>and</strong> agriculture; light grey: natural<br />
vegetation; white: water bodies.<br />
Figure 3b. <strong>Environmental</strong>ly Sensitive Areas<br />
(ESAs). Black: = built-up areas <strong>and</strong> water<br />
bodies; dark grey: critical; medium grey: fragile:<br />
light grey: potential.<br />
In this article we have illustrated how narrative<br />
storylines can potentially be linked to a quantitative<br />
model. With the presented detailed methodology we<br />
hope to have emphasised some of the potential<br />
practical pitfalls during translation. We want to<br />
particularly stress the potential difficulties when<br />
linking a highly complex model with many<br />
parameters <strong>and</strong> sub-modules to scenarios that are<br />
partly developed by laymen <strong>and</strong> that therefore<br />
sometimes lack ‘scientific’ argumentation <strong>and</strong> are not<br />
always internally consistent.<br />
We have, however, not touched upon the theoretical<br />
considerations. By successfully solving practical<br />
problems, important theoretical questions might not<br />
receive the attention that is needed. We hope to<br />
further elaborate on:<br />
What is the feasibility of linking qualitative scenarios<br />
that include surprises, radical system changes, <strong>and</strong><br />
non-linearities to a model that may have limited<br />
possibilities to deal with these changes?<br />
How to deal with inconsistencies typically present in<br />
narrative storylines that are developed partly by nonexperts?<br />
To what degree do the worldview of the scientists as<br />
represented in the PSS <strong>and</strong> the worldview of the local<br />
<strong>and</strong> regional stakeholders as formalised in the<br />
scenarios match?<br />
What we want to achieve in the long run is a bottomup,<br />
top-down cycle, in which those <strong>and</strong> other<br />
questions are addressed <strong>and</strong> dealt with by scientists,<br />
policy makers, <strong>and</strong> the general public alike.<br />
Figure 3c. Suitability for agriculture. Light grey<br />
indicates a low suitability; dark grey tones a high<br />
suitability.<br />
the PSS are given in Figure 3. Presented are the<br />
input l<strong>and</strong> use patterns; the suitability for<br />
agriculture; <strong>and</strong> the environmentally sensitive areas<br />
(ESAs), which are measure for the potential for l<strong>and</strong><br />
degradation. The suitability map is calculated yearly<br />
<strong>and</strong> can be used as an indicator for l<strong>and</strong><br />
4. CONCLUSIONS AND<br />
RECOMMENDATIONS<br />
• This paper has demonstrated that it is<br />
practically possible to link qualitative<br />
storylines <strong>and</strong> quantitative models, despite a<br />
number of potential pitfalls.<br />
• A successful practical link is, however, no<br />
guarantee for a successful link from a<br />
theoretical point of view. Inconsistencies in<br />
the participatory stories <strong>and</strong> radical system<br />
changes are but two examples of potential<br />
theoretical obstacles.<br />
758
We therefore need to focus future efforts of linking<br />
qualitative <strong>and</strong> quantitative scenarios on:<br />
• Synchronising the underlying assumptions<br />
of mathematical models <strong>and</strong> the mental<br />
models from which the local stakeholders<br />
reason.<br />
• Involving stakeholders in some phases of<br />
the construction of mathematical models.<br />
5. REFERENCES<br />
Barreteau, O., F. Bousquet, <strong>and</strong> J-M. Attonaty,<br />
Agent-based modelling, game theory <strong>and</strong><br />
natural resource management issues, Journal<br />
of Artificial Societies <strong>and</strong> Social Simulation<br />
4(2). Available online at:<br />
http://www.soc.surrey.ac.uk/JASSS/4/2/5.ht<br />
ml, 2001<br />
Dreborg, K.H., Essence of backcasting. Futures 28,<br />
813-828, 1996.<br />
Engelen, G, Development of a Decision Support<br />
System for the integrated assessment of<br />
policies related to desertification <strong>and</strong> l<strong>and</strong><br />
degradation in the Mediterranean, p. 159-195<br />
in: Giupponi, C. <strong>and</strong> M. Shechter (Eds.),<br />
Climate Change in the Mediterranean: Socioeconomic<br />
Perspectives of Impacts,<br />
Vulnerability <strong>and</strong> Adaptation, Edward Elgar,<br />
Cheltenham.<br />
Kasemir, B., U. Dahinden, Å. G. Swartling, , R.<br />
Schüle, D. Tabara, <strong>and</strong> C.C. Jaeger, Citizens'<br />
perspectives on climate change <strong>and</strong> energy<br />
use, Global <strong>Environmental</strong> Change 10, 169-<br />
184, 2000.<br />
Kok, K., D.S. Rothman, S.C.H. Greeuw, <strong>and</strong> M.<br />
Patel, European scenarios. From VISIONS to<br />
MedAction, MedAction Deliverable 2, ICIS,<br />
Maastricht, Report number I03-E004,<br />
Available online at: http://www.icis.<br />
unimaas.nl/medaction/download.html, 2003.<br />
Kok, K., <strong>and</strong> D.S. Rothman, Mediterranean<br />
scenarios. First Draft, MedAction<br />
Deliverable 3, ICIS, Maastricht, Report<br />
number I03-E001, Available online at:<br />
http://www.icis.unimaas.nl/medaction/downl<br />
oad.html, 2003.<br />
Kok, K., <strong>and</strong> M. Patel (Eds.), Target Area scenarios.<br />
First sketch. MedAction Deliverable 7. ICIS,<br />
Maastricht, Report number I03-E003.<br />
Available online at: http://www.icis.<br />
unimaas.nl/medaction/download.html, 2003.<br />
Oxley, T., B. McIntosh, N. Winder, M. Mulligan,<br />
<strong>and</strong> G. Engelen, Integrated modelling <strong>and</strong><br />
decision-support tools: a Mediterranean<br />
example, <strong>Environmental</strong> <strong>Modelling</strong> &<br />
<strong>Software</strong>, in press.<br />
Parker, D.C., T. Berger, S.M. Manson, <strong>and</strong> W.J.<br />
McConnell, Agent-based models of l<strong>and</strong>-use<br />
<strong>and</strong> l<strong>and</strong>-cover change. Proceedings of an<br />
international workshop, October 4-7, 2001,<br />
Irvine CA, USA, Report no. 6, LUCC Report<br />
Series, 2002.<br />
Robinson, J., Future subjunctive: backcasting as<br />
social learning, Futures 35, 839-856, 2003..<br />
Rotmans, J, Methods for IA: the challenges <strong>and</strong><br />
opportunities ahead. <strong>Environmental</strong> <strong>Modelling</strong><br />
<strong>and</strong> Assessment 3, 155-179, 1998.<br />
UNCCD, United Nations Convention to Combat<br />
Desertification, Convention text as of<br />
September 1994, Available online at:<br />
http://www.unccd.int/convention/menu.php,<br />
1994.<br />
van Delden, H., P Luja, <strong>and</strong> G. Engelen, Integration<br />
of multi-scale dynamic spatial models of socioeconomic<br />
<strong>and</strong> physical processes for river<br />
basin management. Paper presented at this<br />
conference, 2004.<br />
White, R., B. Straatman, <strong>and</strong> G. Engelen,<br />
Planning scenario visualization <strong>and</strong><br />
assessment: a cellular automata based<br />
integrated Spatial Decision Support System, p.<br />
420-442, in: Spatially Integrated Social<br />
Science, M. F. Goodchild, M.F. <strong>and</strong> D. Janelle<br />
(Eds.), Oxford University Press, New York,<br />
2004.<br />
6. APPENDIX<br />
MedAction: Policies to combat desertification in the<br />
Northern Mediterranean region. Research project<br />
supported by the European Commission under the<br />
Fifth Framework Programme <strong>and</strong> contributing to the<br />
implementation of the Key Action 2: "Global Change,<br />
Climate <strong>and</strong> Biodiversity"; Subaction 2.3.3 "Fighting<br />
L<strong>and</strong> Degradation <strong>and</strong> Desertification". Research<br />
period: 2001-2004. Website: www.icis.unimaas.nl/<br />
medaction.<br />
759
Integrated Assessment of Water Stress in Ceará, Brazil,<br />
under Climate Change Forcing<br />
M.S. Krol <strong>and</strong> P. van Oel<br />
Water Engineering <strong>and</strong> Management, Department of Civil Engineering, University of Twente<br />
PO Box 217, 7500 AE Enschede, the Netherl<strong>and</strong>s<br />
Abstract: Surface water is the main source of fresh water supply in Ceará, lying in the semi-arid Northeast of<br />
Brazil.<br />
Keywords: Integrated assessment; Water stress; Climate change; <strong>Modelling</strong><br />
1. INTRODUCTION<br />
Surface water is the main source of fresh water<br />
supply in Ceará, lying in the semi-arid Northeast of<br />
Brazil. The semi-arid climate goes along with a<br />
high intra-annual <strong>and</strong> inter-annual variability in<br />
precipitation, <strong>and</strong> in the same time the state’s<br />
society is heavily dependent on water supply. This<br />
makes the region specifically vulnerable for<br />
unfavourable developments in climate.<br />
Global circulation models (GCMs) are improving<br />
their skill to represent global <strong>and</strong> continental<br />
climate in present <strong>and</strong> recent history (IPCC, 2000)<br />
<strong>and</strong> their results are interpreted in regional impact<br />
studies.<br />
Surface water storage is the main regional strategy<br />
to enhance water availability. The strategy is meant<br />
to serve both the goals of safeguarding water<br />
supply for vital water use, <strong>and</strong> of enabling a growth<br />
of water-dem<strong>and</strong>ing economic activities. The<br />
management of water storage infrastructure <strong>and</strong><br />
water distribution is an important factor in<br />
determining water stress <strong>and</strong> its impacts.<br />
Tendencies towards participatory approaches in<br />
decision-making on water management, that<br />
emerge from Integrated Water Resources<br />
Management (IWRM, XXX) are being in<br />
introduced in Ceará over the last decade.<br />
This paper describes the impact of climate change<br />
on the water balance of Ceará using the Semi-arid<br />
Integrated Model, SIM (Krol et al, 2001), as a case<br />
study for impacts on developing semi-arid regions.<br />
The state of the art in GCM results at the subcontinental<br />
scale for Northeast Brazil is assessed.<br />
Impacts on the water balance focus on water stress<br />
<strong>and</strong> stored water volumes. The representation of<br />
water management in SIM is evaluated, <strong>and</strong> the<br />
options to extend the integrated model are<br />
discussed.<br />
2. CASE STUDY AREA<br />
The study area can be characterised as a semi-arid<br />
environment with a short but intense rainy period<br />
<strong>and</strong> a long dry period (Gaiser et al., 2003). Rainfall<br />
in the rainy period is generally unreliable with<br />
droughts appearing at various temporal <strong>and</strong> spatial<br />
scales. Surface water is stored in small ponds <strong>and</strong><br />
large reservoirs to distribute water availability over<br />
the year <strong>and</strong> into drought years. Groundwater<br />
availability is scattered with often problems of<br />
salinity in the dominantly crystalline area. In<br />
drought years, many groundwater wells lose<br />
capacity <strong>and</strong> show increasing salinity.<br />
Surface water is mostly used for agriculture<br />
(irrigation <strong>and</strong> animal water use) <strong>and</strong> for water<br />
supply for the metropolitan area of Fortaleza<br />
(industrial <strong>and</strong> municipal use), over a canal inlet to<br />
the urban supply-system situated in the<br />
downstream part of the main river Jaguaribe.<br />
760
Figure 1. Location of the study area of state of Ceará <strong>and</strong> the Jaguaribe Basin (marked orange), situated in<br />
the semi-arid ‘drought polygon’ in Northeast Brazil.<br />
three are able to represent in their simulations the<br />
3. CLIMATE SCENARIOS<br />
semi-aridness <strong>and</strong> strong seasonal cycle, that are<br />
characteristic for this region, see Figure 2. One of<br />
Complex physically-based climate models (as<br />
these three models has a serious flaw in<br />
General Circulation Models, GCMs) show an<br />
representing global precipitation, hampering<br />
increasing ability to simulate present day climate<br />
serious interpretation of its results on changes in<br />
as well as historic trends over the last centuries at<br />
precipitation. This lack in skill may be caused by<br />
the global to continental scale (IPCC, 2000). They<br />
the relatively coarse resolution of GCMs, 300 to<br />
project significant global climate warming (1.4 to<br />
900 km, leaving 2 to 12 grid cells only to cover all<br />
5.8 degrees, 1990-2100) <strong>and</strong> precipitation<br />
of North-eastern Brazil. An alternative<br />
increase to take place in the current century, under<br />
explanation may be the imperfect representation<br />
the assumption of a continuous increase in<br />
of regionally important physical processes. Either<br />
atmospheric greenhouse gas concentrations, as<br />
way, the lack in skill seriously affects the<br />
would be caused by a continued intensive use of<br />
applicability of model results for impact<br />
fossil fuels.<br />
assessments.<br />
Still, the skill of these models in representing<br />
The recommended approach to critically review<br />
climate at the scale of North-eastern Brazil (NEB)<br />
regional performance in selecting model results to<br />
is modest. Of seven climate GCMs, whose climate<br />
be used in assessments (IPCC-TGCIA, 1999) is<br />
change experiments were made available for<br />
often ignored, for instance in a specific<br />
climate impact assessments by the IPCC Data<br />
assessment of plausible climate change in Brazil,<br />
Distribution Centre (IPCC-DDC, 1999), only<br />
including a focus on NEB (Hulme <strong>and</strong> Sheard,<br />
761
1999). This can easily lead to inconsistent<br />
interpretations; for example, in one GCM, Northeastern<br />
Brazil turns from very arid into arid<br />
between 2000 <strong>and</strong> 2100, which by only using<br />
climate change output, would be interpreted as a<br />
transformation from semi-arid into sub-humid.<br />
Error for the dryest<br />
months<br />
Simulation of regional precipitation<br />
over NE Brazil<br />
200%<br />
150%<br />
100%<br />
50%<br />
0%<br />
-50%<br />
-100%<br />
ECHAM 4<br />
HADCM 2<br />
other GCMs<br />
-75% -50% -25% 0% 25% 50%<br />
Error in annual precipitation<br />
Figure 2. Simulation of precipitation in Northeast<br />
Brazil for the current climate by GCMs.<br />
The two models involved, still reasonably<br />
allowing a regional interpretation of their results<br />
for North-eastern Brazil are ECHAM4 (Roeckner<br />
et al., 1996) <strong>and</strong> HADCM2 (Johns et al., 1997).<br />
Following the recommendations of IPCC-TGCIA<br />
(1999) we selected results from these models for<br />
our analyses. For an annual increase of<br />
greenhouse gases by 1% per year as of 1990,<br />
projections of precipitation changes over NEB<br />
(2070-2099 compared to 1961-1990) are –50%<br />
for ECHAM <strong>and</strong> +21% for HADCM.<br />
Given the very small number of models meeting<br />
the minimal criteria adopted above for a direct<br />
regional interpretation of its climate change<br />
results, conclusions on likely precipitation<br />
changes in NEB, on median changes or probable<br />
ranges of precipitation change cannot be drawn.<br />
Still, in climate change studies for semi-arid<br />
North-eastern Brazil both the possibilities of a<br />
strong decrease in precipitation (-50%) <strong>and</strong> an<br />
appreciable increase in precipitation (+21%)<br />
should be considered as plausible to take place in<br />
the current century<br />
Assessment of climate change impacts on e.g.<br />
surface hydrology <strong>and</strong> agricultural production for<br />
the states of Ceará <strong>and</strong> Piauí requires, at the<br />
coarsest, a resolution of climate data at the scale<br />
of sub-regions in Ceará <strong>and</strong> Piauí with marked<br />
differences in hydro-meteorological or agrometeorological<br />
conditions, i.e. the scale of 10-100<br />
km. This seriously hampers the direct (grid-cell<br />
based) regional interpretation of GCM-simulated<br />
climate change, whose resolution is much coarser.<br />
Indirect methods, using Local Area Models<br />
(LAMs) of climate or statistical downscaling of<br />
large-scale features to derive regional climate may<br />
overcome this problem. The latter method was<br />
applied in the WAVES project for the generation<br />
of regional climate scenarios (Gerstengarbe <strong>and</strong><br />
Werner, 2001), as no LAMs are presently<br />
operational with sufficient skill for climate studies<br />
in the study region (Böhm, 2001). Future<br />
development of regional LAMs or GCMs with<br />
increased resolution could yield improved<br />
regional simulations of the climate of semi-arid<br />
north-eastern Brazil, as is hinted at by the fact that<br />
the 2 models with reasonable reproduction of<br />
regional <strong>and</strong> global climate exhibit the highest<br />
resolutions in the model set considered.<br />
The downscaling method adopted combines<br />
observed daily historic climate data at the level of<br />
climate stations with long-term climate trends<br />
from GCM projections. Here the tendency in<br />
annual precipitation at the large scale was taken as<br />
the regionally most relevant trend. Simultaneously<br />
observed daily data on precipitation <strong>and</strong><br />
temperature were used to interpret these<br />
tendencies into projections of these variables at<br />
the station level. Other meteorological variables<br />
like relative humidity <strong>and</strong> short-wave radiation<br />
were added using regression relations derived<br />
from the few available daily time series of a more<br />
complete set of meteorological variables. This<br />
resulted in climate scenarios at the level of the<br />
climate stations. Interpolation routines were used<br />
to transform these scenarios into a climate<br />
scenario defined at the level of municipalities.<br />
This additional step was necessary, as the<br />
municipality was taken as the common spatial unit<br />
used in the integrated simulations of hydrological,<br />
agricultural <strong>and</strong> socio-economic processes.<br />
Results for the two selected GCMs show welldefined<br />
spatial patterns of precipitation trends,<br />
arising from station-specific correlations between<br />
local <strong>and</strong> large-scale precipitation amounts. The<br />
difference between the spatial patterns indicates<br />
that this correlation is different for anomalously<br />
dry years than for anomalously wet years. For a<br />
description of the methodology see Gerstengarbe<br />
<strong>and</strong> Werner (2001).<br />
4. CLIMATE CHANGE IMPACTS ON<br />
WATER SUPPLY INDICATORS<br />
For the assessment of the effects of climate<br />
change on Ceará, the semi-arid integrated model<br />
SIM (Krol et al., 2001) was used. This model<br />
represents inter-linkages between climate,<br />
hydrology, water storage, agricultural production,<br />
<strong>and</strong> socio-economic impacts on the long term,<br />
considering regional development <strong>and</strong> external<br />
drivers of global change. The model bases on a<br />
systems analytical approach, where choices on<br />
762
esolutions were made dependent on the spatial<br />
<strong>and</strong> temporal scales of the processes included. A<br />
minimum common resolution (municipality <strong>and</strong><br />
year) was defined for data exchange, but finer<br />
where required.<br />
This study considers simulations for one fixed<br />
scenario of regional <strong>and</strong> global developments, but<br />
with three different assumptions on climate<br />
change, referred to as the ECHAM scenario, the<br />
HADCM scenario <strong>and</strong> the Constant scenario. The<br />
reference scenario ‘Globalisation <strong>and</strong> Cash Crops’<br />
(Döll <strong>and</strong> Krol, 2003) was taken. This scenario<br />
assumes a continuation of historic trends in<br />
demographic <strong>and</strong> economic development, an<br />
increasing on international markets with<br />
development focusing on the coastal region <strong>and</strong><br />
the interior, where water resources are potentially<br />
available (especially the downstream river valleys<br />
<strong>and</strong> mountainous areas). Especially in these area<br />
water supply is enhanced through additional damconstruction.<br />
Agriculturally used area exp<strong>and</strong>s<br />
gradually; the irrigated area more than doubles.<br />
In SIM, the larger reservoirs, with capacity over<br />
50 Mm3, are simulated explicitly. The total water<br />
volume stored in these reservoirs at the beginning<br />
of the dry season (July 1st) shows a strong<br />
increase between 1995 <strong>and</strong> 2015, the period<br />
where a marked increase in storage capacity<br />
occurs in the scenario (by 7000 Mm3). Total<br />
storage capacity in Ceará <strong>and</strong> Piauí then reaches<br />
almost 22000 Mm3, of which 16400 Mm3 is<br />
installed in Ceará. Afterwards, in the HADCM<br />
scenario <strong>and</strong> the scenario with constant climate,<br />
the reservoirs show a variable degree of stored<br />
water, without a significant trend; for the ECHAM<br />
scenario, stored volume in Ceará shows a marked<br />
decline (Fig 3).<br />
Gm3<br />
16<br />
12<br />
8<br />
4<br />
0<br />
Water stored in large reservoirs in Ceará<br />
at the start of the dry season<br />
ECHAM<br />
HADCM<br />
Constant<br />
1990 2000 2010 2020 2030 2040 2050<br />
Figure 3. <strong>Volume</strong> of water stored in reservoirs in<br />
Ceará at the start of the dry season for the base<br />
scenario with 3 assumptions for climate change.<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
Sufficieny of water supply for<br />
irrigation<br />
ECHAM<br />
HADCM<br />
Constant<br />
1990 2010 2030 2050<br />
Figure 4. Relative sufficiency of water supply to<br />
fulfil the dem<strong>and</strong>s from irrigation in Ceará.<br />
5. WATER MANAGEMENT ISSUES<br />
Water management issue play a key role in<br />
determining drought impacts. Ceará has a centuryold<br />
history of water-management oriented towards<br />
drought relief <strong>and</strong> regional development.<br />
Reservoir construction has had a central part in<br />
water management policies, with the object of<br />
guaranteeing water supply both for the dry season<br />
as for possible subsequent droughts (failure of the<br />
rainy season).<br />
Water dem<strong>and</strong> management <strong>and</strong> management<br />
connected to water distribution, especially in<br />
connection with stakeholder involvement, have<br />
emerged only recently. This is reflected by the<br />
small availability of historic data on water use <strong>and</strong><br />
water distribution.<br />
Options in water management that are being<br />
considered in strategic planning still involve the<br />
consideration of reservoir construction, next to<br />
connections of subbasins, combating subregional<br />
drought, strategic reservoir operation that could<br />
possibly gain efficiency by considering long-term<br />
precipitation forecasts, that have gained<br />
substantially in skill over the last decade,<br />
prioritization <strong>and</strong> distribution policies. Especially<br />
the latter two issues are also subject of discussion<br />
in water committees that are operational in Ceará.<br />
In this respect, Ceará is a front-runner in Brazilian<br />
water policy, by even partially implementing<br />
committees before the legal arrangements have<br />
been completed. Here new developments will<br />
emerge <strong>and</strong> experiences are to be gained over the<br />
coming decade.<br />
Very different water management strategies can<br />
be distinguished between, from precautionary or<br />
even conservative to risk-seeking or profitoptimizing.<br />
It is presently unclear how water<br />
management in Ceará could be characterized or<br />
what tendencies in such a characterization could<br />
be expected. It does, however, have a significant<br />
763
influence on the regional vulnerability to drought<br />
<strong>and</strong> to climate change.<br />
At present, water management is only considered<br />
very coarsely in the integrated model SIM. The<br />
model includes responses to water shortage, but<br />
rather to safe-guard internal consistency in the<br />
model, than to realistically represent management;<br />
this was beyond the scope of the project in which<br />
the model was defined.<br />
A consideration of these water management issues<br />
in the model is expected to yield more convincing<br />
simulations, in improving the consistency <strong>and</strong><br />
allowing options to assess not only the sensitivity<br />
of the region to climate change but also the<br />
possible efficiency of water management options.<br />
Agent-based modelling offers an option to define<br />
algorithmic descriptions of water management<br />
strategies that can be directly implemented in the<br />
existing framework of the integrated model.<br />
6. CONCLUSIONS<br />
Global climate change will take effect on the<br />
climate of North-eastern Brazil. The direction of<br />
precipitation changes however cannot be<br />
determined with certainty. Both very significant<br />
precipitation losses <strong>and</strong> moderate precipitation<br />
increases should be considered plausible.<br />
At current climatic conditions, surface water<br />
availability, showing the regional vulnerability.<br />
The impacts of precipitation losses, as projected<br />
by one of the climate models with best regional<br />
performance (ECHAM scenario), would be of big<br />
importance for the region, even enhancing the<br />
vulnerability.<br />
For the state of Ceará, large scale reductions in<br />
the availability of stored surface water leads to an<br />
increasing imbalance between water dem<strong>and</strong> <strong>and</strong><br />
water supply after 2025, under the assumptions of<br />
the reference scenario, where future water<br />
dem<strong>and</strong>s are growing until 2025 but stabilizing<br />
then.<br />
Agricultural production would also show negative<br />
tendencies after 2025 due to insufficiency of<br />
water supply to meet irrigation water dem<strong>and</strong>s.<br />
For the climate scenarios with a constant or<br />
moderately increasing precipitation, no apparent<br />
tendencies in impacts are found.<br />
In the climate scenarios, trends in precipitation are<br />
entangled with the natural variability, leaving long<br />
periods for tendencies to become statistically<br />
significant. This applies to tendencies in the<br />
impacts as well. Even for the climate scenario<br />
with the most marked trends, significance of<br />
impacts is found after 2025 only. Before, impact<br />
levels do not exceed the levels of impacts<br />
emerging from natural variability.<br />
This should not discourage to consider possible<br />
climate change in preparing policies to increase<br />
drought resilience. Measures to increase resilience<br />
will largely rely on long-term policies. In<br />
discussions with responsible regional agencies,<br />
focus was on measures in water infrastructure <strong>and</strong><br />
its management, water use efficiency<br />
improvements, <strong>and</strong> structural changes in the<br />
agricultural sector. All these items refer to long<br />
term changes, where possible climate change<br />
could have a significant influence. The present<br />
analysis suggests, that the efficiency of various<br />
measures under different future climatic<br />
conditions (which can be considered as the<br />
robustness of the measure) might be a more<br />
relevant criteria in selecting measures than<br />
optimising the measure for present climatic<br />
conditions.<br />
Integrated modelling proved an important<br />
instrument in evaluating climate impacts.<br />
Feedbacks between trends in agriculture, water<br />
use, insufficiency of surface water supply have a<br />
relevant influence on model results, especially for<br />
the scenario with diminishing precipitation<br />
volumes. Such feedbacks would not be addressed<br />
by single thematic studies or direct sequential<br />
couplings of models.<br />
Many uncertainties remain, not only in the<br />
possible future climate developments, but also in<br />
the regional responses to water shortage <strong>and</strong><br />
trends in water use; descriptions of specific water<br />
use, water management, tendencies in (irrigated)<br />
agriculture <strong>and</strong> societal processes are based on<br />
scarce data, <strong>and</strong> other relevant themes are still<br />
lacking representation in the scenarios <strong>and</strong><br />
models, e.g. planned adaptation strategies as the<br />
connection of large catchments to reduce impacts<br />
of sub-regional droughts. These uncertainties were<br />
partly studies for the isolated contributing models,<br />
see various contributions in Araújo et al. (2001),<br />
but an ample study of uncertainties in the<br />
integrated model is lacking. Here methods from<br />
agent-based modelling offer promising options,<br />
especially in representing user’s responses to<br />
reduced water availability <strong>and</strong> in representing<br />
decision making on water distribution.<br />
7. ACKNOWLEDGEMENTS<br />
The authors wish to thank the contributors of the<br />
WAVES-programme for their kind collaboration<br />
<strong>and</strong> provision of available data.<br />
8. REFERENCES<br />
Döll, P. <strong>and</strong> M.S. Krol<br />
764
From Narrative to Number: A Role for Quantitative<br />
Models in Scenario Analysis<br />
Eric Kemp-Benedict a<br />
a<br />
KB Creative, 8 Longfellow Road, Cambridge, MA 02138, USA, e-mail: eric@kb-creative.net<br />
Abstract: There is growing concern that the predictive mathematical models conventionally used in policy<br />
analysis are too limiting to serve as tools in futures studies, because they cannot reproduce the sudden<br />
changes seen in real societies. The field of complex systems has successfully produced similar changes in<br />
simplified model systems, but has been less successful in practical futures work. Some recent scenario<br />
exercises (such as the IPCC scenarios, UNEP’s GEO-3 scenarios, the work of the Global Scenario Group <strong>and</strong><br />
the European VISIONS project) have addressed this issue by combining wide-ranging narratives with<br />
quantitative models, demonstrating that a synthesis between qualitative <strong>and</strong> quantitative approaches is<br />
possible. However, there is no consensus on an appropriate methodology. In this paper it is argued that there<br />
are essentially two analytical challenges that scenario models must address in order to achieve the goal of<br />
more robust planning in the face of both gradual <strong>and</strong> sudden change. One is to represent complexity, while the<br />
other is to represent what might be called “complicatedness.” Complex behavior arises from the<br />
interrelatedness of different components of a system, while “complicatedness” as used here means that there<br />
are a lot of factors to keep in mind—constraints, actors, resources, etc. It will further be argued that<br />
complexity is best dealt with in narratives, <strong>and</strong> complicatedness is best dealt with using computers. The<br />
characteristics of appropriate computer models will be presented, <strong>and</strong> extant exemplars of appropriate models<br />
described.<br />
Keywords: Scenarios; Modeling; Futures Studies; Complex Systems.<br />
1. INTRODUCTION<br />
From its earliest inception, there has been a tension<br />
in Futures Studies between the use of qualitative<br />
<strong>and</strong> quantitative techniques. At times this has taken<br />
the form of a contest. Modelers, in particular, have<br />
cast themselves as the guardians of rigor in a field<br />
struggling to gain legitimacy, <strong>and</strong> it can perhaps be<br />
argued that in the past decade, with the increasing<br />
use of Integrated Assessment (IA) models <strong>and</strong><br />
Computable General Equilibrium (CGE) models,<br />
quantitative approaches have dominated. Yet there<br />
has always been an argument for combining<br />
narrative <strong>and</strong> number (see, e.g., deLeon [1984])<br />
<strong>and</strong> recently, as the weaknesses of quantitative<br />
models have once again become apparent [Smil,<br />
2000; DeLeon, 1997; Höjer <strong>and</strong> Mattsson, 2000],<br />
there are increasing calls for balancing qualitative<br />
<strong>and</strong> quantitative approaches in futures work.<br />
In this paper, we join the chorus of authors calling<br />
for change, arguing that a robust scenario emerges<br />
from the interaction between the quantitative <strong>and</strong><br />
qualitative contributions. For evidence of the<br />
usefulness of a synthetic approach, we can turn for<br />
examples to recent scenario exercises, such as the<br />
IPCC scenarios [Nakicenovic <strong>and</strong> Swart, 2000],<br />
UNEP’s GEO-3 scenarios [UNEP, 2002], the<br />
World Water Visions scenarios [Cosgrove <strong>and</strong><br />
Rijsberman, 2000], the work of the Global<br />
Scenario Group [Gallopín et al., 1997] <strong>and</strong> the<br />
European VISIONS project [Rotmans et al., 2000].<br />
However, despite the considerable work that has<br />
been done, there is no consensus on how to go<br />
about synthesizing qualitative <strong>and</strong> quantitative<br />
scenario approaches. As a contribution to this<br />
emerging type of futures work, we offer a set of<br />
765
methodological guidelines for a successful<br />
synthesis. 1<br />
Key to the approach described here is a distinction<br />
between complexity—the subject of complex<br />
systems theory—<strong>and</strong> what we call<br />
complicatedness—merely keeping track of the<br />
numerous factors, such as physical-economicsocial<br />
relationships, that can influence a scenario.<br />
It is argued in this paper that complexity is best<br />
dealt with using traditional qualitative scenario<br />
techniques, while quantitative models—especially<br />
computer models—are best suited to keeping track<br />
of complications. In this view, the narrative drives<br />
the scenario development, while quantitative<br />
models are developed in response to the narrative.<br />
2. MODELS: COMBINING NARRATIVE<br />
AND NUMBER<br />
A model is a representation of a system. A good<br />
model behaves sufficiently like the real system that<br />
conclusions can be drawn from the model’s<br />
behavior to aid in making decisions about the real<br />
system. How “good” a given model is therefore<br />
depends on its purpose. In traditional policy<br />
modeling, comprehensive, predictive mathematical<br />
models have been the norm. However, this sort of<br />
model has a poor record when confronted with the<br />
complex nature of social systems [Rihani, 2002].<br />
In Vinay Lal’s pithy remark, “Since the human<br />
being is the one unpredictable animal, many<br />
planners for the future find Homo sapiens to be a<br />
rather unpleasant reminder of the impossibility of a<br />
perfect blueprint” [Lal, 1999]. In contrast, more<br />
“intuitive” scenario exercises, presented in<br />
narrative form, have captured some of the<br />
surprising features observed in real social systems.<br />
Of necessity, both mathematical studies <strong>and</strong><br />
narrative exercises employ models, although of<br />
very different kinds. In the mathematical approach<br />
the model is explicit, as a set of mathematical<br />
formulae, a computer program, a diagram in Stella,<br />
or some other formal representation that can be<br />
translated into a sequence of numerical<br />
calculations. In the narrative approach the model is<br />
generally implicit in the form of the narrative,<br />
which reflects the shared mental model of its<br />
authors. There are advantages <strong>and</strong> disadvantages to<br />
both the mathematical <strong>and</strong> narrative approaches.<br />
The challenge is to combine narratives with formal<br />
mathematical analysis in a way that builds on the<br />
strengths of the two approaches.<br />
What are those strengths? There are essentially two<br />
analytical challenges that scenario models must<br />
address. One is to represent complexity, while the<br />
other is to represent complicatedness. By<br />
1 For a different approach to a synthesis, see Alcamo [2001].<br />
“complexity,” we mean the behavior of complex<br />
systems, as described by complex systems theory.<br />
In particular, it refers to the behavior arising from<br />
the interrelatedness of different components of a<br />
system, a feature of real systems that helps make<br />
the world so interesting. In contrast, by<br />
“complicatedness” we mean the sort of<br />
bookkeeping that is necessary when there are a lot<br />
of factors to keep in mind—constraints, actors, <strong>and</strong><br />
resources.<br />
People are quite capable of thinking in terms of<br />
complex systems, but they are not in the habit of<br />
doing so. Many futures techniques that result in a<br />
narrative description of the future seek to draw out<br />
this latent ability, mainly by encouraging people to<br />
think “outside the box.” Computers can also<br />
represent complexity. Mathematical models with<br />
very few variables, but with nonlinear interactions<br />
between the variables, or agent-based models that<br />
feature interacting agents following simple<br />
behavioral rules, can exhibit a striking array of<br />
features that parallel those seen in real systems.<br />
They key insight arising from these studies is that<br />
simple rules can lead to rich <strong>and</strong> unexpected<br />
behavior. However, the state of the art in computer<br />
modeling of societies as complex systems is too<br />
crude for applied work. Instead, it is best suited for<br />
academic studies, to learn more about the nature of<br />
complexity <strong>and</strong> to broaden thinking about social<br />
dynamics. 2 Thus, people are good at modeling<br />
complexity in real social systems, while computer<br />
models have a way to go. In contrast, people are<br />
rapidly overwhelmed by mere complication, while<br />
computers are very good at keeping track of<br />
complicated situations. This is one reason why the<br />
spreadsheet <strong>and</strong> the database became the first<br />
“killer apps” of the personal computer revolution. 3<br />
For these reasons, in this essay it is proposed that a<br />
scenario model should consist of two components:<br />
a set of narratives <strong>and</strong> a set of mathematical<br />
models. The dividing line between the two is not<br />
fixed, but generally the narratives should focus on<br />
the complex nature of the system <strong>and</strong> on its<br />
evolution, while the computer-based mathematical<br />
models should h<strong>and</strong>le the complicated features of<br />
the system, to assist the scenario developers in<br />
making a consistent <strong>and</strong> coherent narrative.<br />
2<br />
The view expressed here closely matches Kohler’s<br />
characterization of “Weak Social Simulation” [Kohler, 2002].<br />
3<br />
Rotmans [1999] also draws a distinction between<br />
“complexity” <strong>and</strong> “complication” when describing computer<br />
models for integrated assessment. However, in contrast to the<br />
position argued in this paper, Rotmans believes that complexity<br />
should be incorporated in the computer model. We would argue<br />
that while it may be appropriate for a complex model to<br />
describe the biophysical components of an IA model, it is not<br />
appropriate for the societal components, given the current state<br />
of the art.<br />
766
3. QUANTITATIVE MODELS AS A<br />
RESPONSE TO A NARRATIVE<br />
In the discussion below, the task of building a<br />
combined narrative <strong>and</strong> quantitative scenario is<br />
broken out into two subtasks: narrative writing <strong>and</strong><br />
mathematical analysis. Although the same person<br />
or group of people may do both subtasks, more<br />
often they are carried out by different people with<br />
different sets of skills. In this essay, the two groups<br />
will be called the “narrative team” <strong>and</strong> the<br />
“modeling team.”<br />
In the approach urged in this essay, the narrative<br />
drives scenario development, while the modeling<br />
team follows the narrative team’s lead. However,<br />
the process is not all one-way: the quantitative<br />
analysis also informs the narrative scenario<br />
development. 4 Taking this reciprocal influence into<br />
account, there are four main roles that quantitative<br />
scenario development can play when implemented<br />
in response to a narrative:<br />
1. Force a clarification of terms <strong>and</strong><br />
mechanisms.<br />
2. Expose contradictions in mental models.<br />
3. Provide a feel for the scope of possible<br />
outcomes within a narrative framework.<br />
4. Illustrate a particular scenario narrative.<br />
5. Make a study replicable, extensible <strong>and</strong><br />
transferable.<br />
The first two items provide direct feedback to the<br />
narrative team about the content of the scenarios.<br />
The first is simply the result of constructing a<br />
rigorous statement of what the narrative writers<br />
mean. This is always a good thing to do, <strong>and</strong> the<br />
task of making a formal mathematical model is a<br />
particularly useful way in which to do it. If a<br />
narrative is to be translated into a formal<br />
structure—especially one that is to be coded in a<br />
computer—then many potentially ambiguous<br />
points must be nailed down <strong>and</strong> key decisions must<br />
be made. This process sharpens the narrative<br />
analysis, as the narrative team is forced to address<br />
its ambiguous goals <strong>and</strong> statements. Note that this<br />
salutary outcome is not reached when the<br />
quantitative model drives the analysis, <strong>and</strong> the<br />
narrative follows from it. In this case, the<br />
mathematical model has been built by people (the<br />
modeling team) who have already encountered<br />
ambiguities <strong>and</strong> resolved them in ways that may or<br />
may not be acceptable to the people using the<br />
4 Some recent scenario exercises, such as the IPCC scenarios,<br />
the VISIONS project, the GEO-3 scenarios <strong>and</strong> the scenarios of<br />
the Global Scenario Group have employed this basic approach<br />
of developing quantitative scenarios in response to a narrative,<br />
<strong>and</strong> have mentioned the two-way flow of information.<br />
However, the approach described in this essay differs in some<br />
ways from those exercises.<br />
quantitative outputs [van der Sluijs, 2002]. The<br />
decisions are not made jointly between the<br />
narrative <strong>and</strong> modeling teams, so they do not<br />
provoke discussion.<br />
The second item—exposing contradictions in<br />
mental models—highlights a key role that<br />
scenarios play, that of fostering cognitive<br />
development <strong>and</strong> learning [Chermack <strong>and</strong> van der<br />
Merwe, 2003; Robinson, 2003]. Constructivist<br />
theories of cognition <strong>and</strong> learning posit that people<br />
actively construct mental models through which<br />
they filter their experiences. Those mental models<br />
are remarkably resilient, <strong>and</strong> are relinquished only<br />
when they are shown (repeatedly) to be<br />
inconsistent—either internally inconsistent or<br />
inconsistent with external reality [Kempton et al.,<br />
1997; Yankelovich, 1991]. Narratives reflect the<br />
mental models of their authors, <strong>and</strong> by translating<br />
them into formal terms, contradictions can be<br />
exposed, either through the process of developing<br />
the formal model or through manipulating the<br />
model. This benefit of modeling exercises often<br />
goes unnoticed, because generally when a formal<br />
model does succeed in changing the narrative<br />
team’s mental model, it is not mentioned in the<br />
written report. There are at least two reasons for<br />
this. First, researchers do not report their<br />
conceptual errors—they report the underst<strong>and</strong>ing<br />
they achieve through their research. Second, when<br />
someone’s mental model changes, it is<br />
extraordinarily difficult to capture the original<br />
pattern of thought. Whatever the reason, it is a pity<br />
that the insights are not reported. Incorrect mental<br />
models are widely shared, <strong>and</strong> are likely to be held<br />
by many readers of the report. If they are not<br />
explicitly addressed, they are likely to persist.<br />
The third item, that of providing a feel for the<br />
scope of possibilities within a narrative, offers<br />
indirect but generally very useful feedback to the<br />
narrative team. How responsive is an outcome to<br />
changes in some parameter or condition? Within a<br />
“backcasting” exercise, how constraining are the<br />
long-term goals? What level of action might be<br />
required to achieve them? What is the scope for<br />
alternative approaches? Even with the simplest<br />
formal models, results from this type of<br />
exploratory exercise can be surprising. A perceived<br />
constraint may turn out not to be so constraining,<br />
or not the main factor determining the evolution of<br />
the scenario; an undesired outcome may turn out to<br />
be avoidable only with heroic efforts; <strong>and</strong> a factor<br />
that is initially small may turn out to be<br />
surprisingly large by the end of the scenario period.<br />
While less profound in its implications for the<br />
scenario narrative than the revelation of a<br />
contradiction or an ambiguity, exploring the<br />
boundaries of the model can provide valuable<br />
insight to both the narrative writers <strong>and</strong> the model<br />
builders.<br />
767
The next item—illustrating a particular scenario<br />
narrative—is an opportunity for narrative writers<br />
<strong>and</strong> model builders to share their insights with<br />
others <strong>and</strong> invite external critique. The narrative,<br />
refined by interaction with the model, is finalized<br />
<strong>and</strong> disseminated, along with quantitative figures—<br />
one or more “illustrations” that emerge from the<br />
exploration of the model boundaries.<br />
The final item states that by encoding key<br />
decisions by the narrative team into an agreed set<br />
of quantitative models, the model structure can be<br />
reused, either by the original team or another team.<br />
Potentially, this offers great advantages. By<br />
making the model explicit, it can be subjected to<br />
outside review. However, there is also a danger<br />
that formal models will be reused uncritically. A<br />
central feature of the combined narrative <strong>and</strong><br />
numerical approach proposed in this essay is that<br />
the narrative <strong>and</strong> modeling teams engage in a<br />
mutual critique. When a set of scenarios generated<br />
in this way is adopted by others, or reused, it<br />
should again be subjected to critique. One way to<br />
encourage this is to always start fresh, with a new<br />
set of narratives, but allow the modeling team to<br />
reuse an existing set of models if they seem<br />
appropriate for those narratives. That is, computer<br />
models should be “cannibalized” for parts, not<br />
reused wholesale. Over time, a modeling team<br />
could develop a code base of “parts” to bring into<br />
play for different scenario exercises.<br />
4. APPROPRIATE MODELS<br />
What are the characteristics of an appropriate<br />
quantitative model for scenario development? Bell<br />
[1997] lists four schools of computer modeling:<br />
input-output analysis, econometrics, optimization,<br />
<strong>and</strong> system dynamics. None of these in isolation is<br />
particularly well-suited for the tasks outlined<br />
above. The problem with each, at least as they have<br />
conventionally been used, is that they attempt to<br />
encapsulate too much of the system being studied.<br />
In these approaches, there is little scope for a<br />
narrative team to redirect the analysis. The<br />
narrative team may envision an abrupt shift in<br />
circumstances—e.g., of the same magnitude as the<br />
fall of the Berlin Wall, the events at Tiananmen<br />
Square, the spread of HIV/AIDS, or the<br />
demonstrations against the World Trade<br />
Organization—but in general it will be difficult to<br />
represent it within an existing quantitative model.<br />
This is not to say that such models cannot be<br />
useful. In fact, they can provide very important<br />
insights <strong>and</strong> a well-defined structure to a scenario<br />
exercise, but they are not best suited—when used<br />
in isolation—to the development of wide-ranging<br />
scenarios.<br />
Another type of model is needed. In fact, examples<br />
of appropriate models already exist, but their<br />
common features have not (to the authors’<br />
knowledge) yet been enumerated. Below, we list<br />
the desired characteristics. In addition, we provide<br />
what is essentially a job description for the<br />
modeling team.<br />
Appropriate models for exploratory scenario<br />
analysis should:<br />
1. Represent the narrative.<br />
2. Reflect fundamental constraints (e.g., l<strong>and</strong><br />
<strong>and</strong> energy balances, economic balances).<br />
3. Reflect the spatial <strong>and</strong> temporal scales of<br />
key processes.<br />
4. Offer several “levers” (although not too<br />
many) for the narrative team <strong>and</strong> other<br />
users.<br />
5. Implement likely correlations.<br />
6. Reflect a knowledge of the relevant<br />
literature.<br />
These conditions place considerable dem<strong>and</strong>s on<br />
the modeling team. Not only must it have access to<br />
a variety of modeling techniques but it must also<br />
be cognizant of the literature in various fields. The<br />
modeling team is also required to represent<br />
whatever narratives the narrative team might<br />
produce. The modeling team must try to identify<br />
the model implicit in the narrative, <strong>and</strong> interpret it<br />
in a formal mathematical model. This requires<br />
flexibility <strong>and</strong> creativity. Perhaps even more<br />
dem<strong>and</strong>ingly, the conditions above require the<br />
modeling team to yield up a large measure of<br />
control to the narrative team. That is, what the<br />
modeling team should produce is not a predictive<br />
model, although it may have causal components<br />
(such as a demographic cohort model). Instead, it<br />
should produce a model that allows a narrative<br />
team to explore a numerical “neighborhood” of<br />
possibilities that is consistent with its narrative.<br />
The main role the quantitative model plays is to<br />
take care of complications, by keeping track of<br />
constraints <strong>and</strong> correlations. The complexity of the<br />
system—arising from the mutual interactions<br />
between its constituent parts—is addressed<br />
principally by the narrative team.<br />
Some examples of suitable models will be given in<br />
the next section. However, before proceeding to<br />
them, a comment is in order about the fifth <strong>and</strong><br />
sixth points in the list above. The fifth point states<br />
that “likely correlations” should be implemented.<br />
This is perhaps the most heterodox suggestion in<br />
this paper. A common complaint against<br />
econometric models, as traditionally used, is that<br />
they interpret empirically correlated data as being<br />
causally related, when that might not be the case.<br />
Elaborate analysis <strong>and</strong> relatively large <strong>and</strong> dense<br />
data sets are necessary to demonstrate causality, so<br />
such analyses are only carried out in a few<br />
768
contentious cases. In the approach proposed here,<br />
however, models need not be causal—for many<br />
purposes, correlations are sufficient. This is<br />
because causal connections should be captured in<br />
the narratives (where they should be made quite<br />
explicit), while the quantitative models should<br />
explore the likely consequences of those narratives<br />
to aid the narrative team in making consistent<br />
narratives. One way to do this is by exploiting<br />
likely correlations.<br />
An example can help clarify this point: An<br />
economically liberal narrative may describe rapid<br />
economic growth in a context of liberalized<br />
markets, while saying nothing about transport<br />
choices. But if the environmental implications of<br />
the narrative are of interest, then transport should<br />
be considered. In this case, empirical correlations<br />
between economic output per capita <strong>and</strong> transport<br />
patterns might be introduced by the modeling team.<br />
If they are, then the modeling team should inform<br />
the narrative team, which may respond by either<br />
accepting the empirical pattern or explicitly stating<br />
in the narrative that the historical pattern is broken.<br />
Such an approach is not without its dangers: it is<br />
only too easy to interpret a correlation as a causal<br />
link, <strong>and</strong> to treat correlations as laws of nature. An<br />
open mind equipped with a pragmatic mind-set is<br />
required for this task.<br />
The sixth point is that the model should reflect a<br />
knowledge of the relevant literature. In practice,<br />
this implies that the modeling team should have a<br />
grasp of the literature on a diverse range of<br />
technical fields, such as economics, engineering,<br />
urban studies, ecology, agronomy, etc. But saying<br />
this does not mean that they need to be experts in<br />
those fields. They should not, for example, expect<br />
to be able to do basic research in the fields.<br />
Perhaps a reasonable benchmark is that they should<br />
not be surprised by something that would not<br />
surprise an expert in the field. Even this level of<br />
underst<strong>and</strong>ing is unlikely to be reached by a<br />
modest-sized team over a wide range of topics, but<br />
to the degree it is approached, it should enable the<br />
modeling team to converse meaningfully with<br />
subject experts <strong>and</strong> allow the modeling team to<br />
supply references, provisional parameter values<br />
<strong>and</strong> insights to the narrative team when an expert is<br />
not on one of the teams.<br />
5. EXAMPLES<br />
There already exist models that meet many of the<br />
criteria listed in the previous section. Three<br />
examples are discussed below. The list is intended<br />
to be illustrative, <strong>and</strong> is far from exhaustive. These<br />
examples may function as exemplars for those<br />
wishing to do an exercise of the sort described in<br />
this paper. While none of the examples below is a<br />
causal model, this possibility is not ruled out. For<br />
example, stock-flow models <strong>and</strong> cohort models<br />
could easily satisfy the requirements for an<br />
appropriate model as proposed in this paper, <strong>and</strong> if<br />
a narrative suggests a particular causal, predictive<br />
model then it may be appropriate to introduce it.<br />
One sector-specific example is the PODIUM<br />
model of the <strong>International</strong> Water Management<br />
Institute (IWMI). 5 PODIUM is implemented as a<br />
Microsoft Excel workbook, <strong>and</strong> is intended to be<br />
used by decision makers in an interactive session.<br />
The decision maker moves through a sequence of<br />
pages, making choices about possible future<br />
developments on each page. At the end, the<br />
implications of the decision maker’s choices are<br />
presented in terms of agricultural water use. The<br />
PODIUM model meets several of the criteria of an<br />
appropriate model as envisioned in this paper: 1) it<br />
reflects a narrative (a basic “development”<br />
narrative that matches the framework of the target<br />
audience); 2) it reflects fundamental constraints<br />
(e.g., constraints on food production); 3) it offers<br />
several “levers” for the decision maker to<br />
manipulate; 4) it reflects a knowledge of the<br />
relevant literature.<br />
An example of a model that incorporates several<br />
sectors is the model developed for the Georgia<br />
Basin Futures Project (GBFP). 6 This study intends<br />
ordinary citizens to be enlisted as narrative writers.<br />
The GBFP team developed a wide array of<br />
possible narratives, <strong>and</strong> built structurally simple<br />
(but not simplistic) mathematical models that cover<br />
the range of futures allowed by those narratives.<br />
The user is offered a series of choices, <strong>and</strong> as with<br />
the PODIUM model, once the model is run the<br />
implications of those choices are presented to the<br />
user. The GBFP model satisfies all of the criteria<br />
for an appropriate quantitative model, according to<br />
the framework presented in this essay.<br />
The final example is that of the “convergence<br />
algorithm” of the PoleStar team for the Global<br />
Scenario Group (GSG). 7 While many aspects of<br />
the GSG scenarios fit the conditions for an<br />
appropriate model as outlined in this paper, the<br />
way that the fundamental narrative of convergence<br />
was implemented deserves special mention. To<br />
give coherence to the illustrative quantitative<br />
scenarios, the PoleStar team introduced an<br />
algorithm, called the “convergence algorithm,” for<br />
calculating energy intensities, emission factors <strong>and</strong><br />
activity levels in developing regions [Kemp-<br />
Benedict et al., 2002]. This model meets four of<br />
the criteria listed in the previous section: 1) it<br />
implements the scenario narrative; 2) it reflects the<br />
temporal scale of technological change; 3) it<br />
5 http://www.iwmi.cgiar.org/tools/podium.htm<br />
6 http://www.basinfutures.net/<br />
7 http://www.seib.org/polestar <strong>and</strong> http://www.gsg.org/<br />
769
eflects a knowledge of the relevant literature, in<br />
this case the literature on dematerialization <strong>and</strong><br />
technological leapfrogging; 4) it implements likely<br />
correlations, in that within the scenario narrative,<br />
rising income in developing regions leads to<br />
convergent patterns of consumption <strong>and</strong> resource<br />
use.<br />
6. SUMMARY<br />
The emerging realization that predictive<br />
mathematical models are limiting in futures work is<br />
leading to interesting new approaches in scenario<br />
development. Several recent scenario studies have<br />
attempted a synthesis of narrative <strong>and</strong> quantitative<br />
approaches. However, there is no consensus on<br />
methodology. This paper proposed a set of criteria<br />
for appropriate mathematical models (as well as for<br />
the modelers themselves) <strong>and</strong> discussed how<br />
models can be joined with narratives to make<br />
robust scenarios.<br />
7. ACKNOWLEDGEMENTS<br />
The author would like to thank Annette Huber-Lee,<br />
Winston Yu, Sivan Kartha, Jack Sieber <strong>and</strong><br />
Michael Benedict for comments on drafts of this<br />
paper. Any remaining errors are of course the<br />
responsibility of the author.<br />
8. REFERENCES<br />
Alcamo, J., “Scenarios as Tools for <strong>International</strong><br />
<strong>Environmental</strong> Assessments,” Experts’<br />
Corner Report: Prospects <strong>and</strong> Scenarios<br />
No.5 <strong>and</strong> <strong>Environmental</strong> Issue Report No.<br />
24. European Environment Agency.<br />
Copenhagen: EEA, 2001.<br />
Bell, W., Foundations of Futures Studies: Human<br />
Science for a New Era, <strong>Volume</strong> I: History,<br />
Purposes <strong>and</strong> Knowledge. New Brunswick,<br />
New Jersey: Transaction Publishers, 1997.<br />
Chermack, T.J. <strong>and</strong> L van der Merwe, “The Role<br />
of Constructivist Learning in Scenario<br />
Planning,” Futures 38, pp. 445-460, 2003.<br />
Cosgrove, W.J. <strong>and</strong> F.R. Rijsberman, World Water<br />
Vision: Making Water Everybody’s<br />
Business. London, UK: Earthscan<br />
Publications. 2000.<br />
deLeon, P. “Futures Studies <strong>and</strong> the Policy<br />
Sciences,” R<strong>and</strong> Paper Series P-7000.<br />
Santa Monica, California: R<strong>and</strong><br />
Corporation.<br />
deLeon, P., Democracy <strong>and</strong> the Policy Sciences.<br />
Albany, New York: State University of New<br />
York Press, 1997.<br />
Gallopín G., A. Hammond, P. Raskin, <strong>and</strong> R.<br />
Swart, Branch Points: Global Scenarios<br />
<strong>and</strong> Human Choice. Stockholm: Stockholm<br />
Environment Institute. 1997.<br />
Höjer, M <strong>and</strong> L-G Mattsson, “Determinism <strong>and</strong><br />
Backcasting in Future Studies,” Futures 32,<br />
pp. 613-634, 2000.<br />
Kemp-Benedict, E., C. Heaps <strong>and</strong> P. Raskin,<br />
Global Scenario Group Futures: Technical<br />
Notes. Boston: Stockholm Environment<br />
Institute-Boston. 2002.<br />
Kempton, W., J.S. Boster <strong>and</strong> J.A. Hartley,<br />
<strong>Environmental</strong> Values in American Culture.<br />
Cambridge, Massachusetts: The MIT Press.<br />
1997.<br />
Kohler, T.A. “Putting Social Sciences Back<br />
Together Again: An Introduction to the<br />
<strong>Volume</strong>,” in T.A. Kohler <strong>and</strong> G.J.<br />
Gumerman, Eds., Dynamics in Human <strong>and</strong><br />
Primate Societies: Agent-Based Modeling<br />
of Social <strong>and</strong> Spatial Processes. New York:<br />
Oxford University Press. 2000.<br />
Lal, V., “Futures <strong>and</strong> Knowledge”, in Z. Sardar,<br />
Ed., Rescuing All Our Futures: The Future<br />
of Futures Studies. Westport, Connecticut:<br />
Praeger Publishers, 1999.<br />
Nakicenovic, N. <strong>and</strong> R. Swart (Eds.), Emissions<br />
Scenarios: Special Report of the<br />
Intergovernmental Panel on Climate<br />
Change. Cambridge, UK: Cambridge<br />
University Press. 2000.<br />
Rihani, S., Complex Systems Theory <strong>and</strong><br />
Development Practice. London: Zed Books,<br />
2002.<br />
Robinson, J., “Future Subjunctive: Backcasting As<br />
Social Learning,” Futures 35, pp. 839-856,<br />
2003.<br />
Rotmans, J. “Integrated Assessment Models:<br />
Uncertainty, Quality <strong>and</strong> Use,” Working<br />
Paper 199-E005, <strong>International</strong> Centre for<br />
Integrative Studies. Maastricht: ICIS. 1999.<br />
Rotmans, J., M. van Asselt, C. Anastasi, S.<br />
Greeuw, J. Mellors, S. Peters, D. Rothman,<br />
N. Rijkens, “Visions for a Sustainable<br />
Europe,” Futures 32, pp. 809-831. 2000.<br />
Smil, V., “Perils of Long-Range Energy<br />
Forecasting: Reflections on Looking Far<br />
Ahead,” Technological Forecasting <strong>and</strong><br />
Social Change 65, pp. 251-264. 2000.<br />
United Nations Environment Program (UNEP),<br />
Global Environment Outlook 3: Past,<br />
Present <strong>and</strong> Future Perspectives. London:<br />
Earthscan Publications. 2002.<br />
van der Sluijs, J.P., “A Way Out of the Credibility<br />
Crisis of Models Used in Integrated<br />
<strong>Environmental</strong> Assessment,” Futures 34,<br />
pp. 133-146, 2002.<br />
Yankelovich, D. Coming to Public Judgment:<br />
Making Democracy Work in a Complex<br />
World. Syracuse, New York: Syracuse<br />
University Press. 1991.<br />
770
Scenario Reoptimisation under Data Uncertainty<br />
Antonio Manca a , Giovanni M. Sechi a <strong>and</strong> Paola Zuddas a<br />
a<br />
Research <strong>and</strong> Educational Centre for Network Optimisation, Department of L<strong>and</strong> Engineering<br />
University of Cagliari, Italy. (zuddas@unica.it)<br />
Abstract: Many dynamic planning <strong>and</strong> management problems are typically characterised by a level of<br />
uncertainty regarding the value of data input such as supply <strong>and</strong> dem<strong>and</strong> patterns. Assigning inaccurate<br />
values to them could invalidate the results of the study. Consequently, deterministic models are inadequate<br />
for the representation of these problems where the most crucial parameters are either unknown or are based<br />
on an uncertain future. In these cases, the scenario analysis technique could be an alternative approach.<br />
Scenario analysis can model many real problems in which decisions are based on an uncertain future, whose<br />
uncertainty is described by means of a set of possible future outcomes, called "scenarios". In this paper we<br />
present a scenario analysis approach to dynamic multi-period systems by integrating scenario optimisation<br />
<strong>and</strong> subsequent deterministic reoptimisation. In the scenario optimisation phase we represent data uncertainty<br />
by a robust chance optimisation model obtaining a so-called barycentric value with respect to selected<br />
decision variables. The successive reoptimisation model based on this barycentric solution allows planning a<br />
part of the risk of a wrong decision, reducing the negative consequences deriving from it.<br />
Keywords: Scenario analysis; Optimisation under uncertainty; Dynamic problems; Reoptimisation.<br />
1. INTRODUCTION<br />
A system is dynamic if each component is<br />
associated with a time t <strong>and</strong> represents a decision<br />
in time. A dynamic system can be generated by<br />
replicating a static system over time with interperiod<br />
connections. Multiperiod systems are<br />
defined in a dynamic planning horizon in which<br />
management decisions have to be made<br />
sequentially in time or decided globally as a<br />
decision strategy referring to a predefined set of<br />
data <strong>and</strong> time horizon. Many dynamic planning<br />
<strong>and</strong> management problems are typically<br />
characterised by a level of uncertainty regarding<br />
the value of data input such as supply <strong>and</strong> dem<strong>and</strong><br />
patterns. (Glockner <strong>and</strong> Nemhauser, [2000].<br />
Assigning inaccurate values to them could<br />
invalidate the results of the study. Consequently,<br />
deterministic models are inadequate for the<br />
representation of these problems where the most<br />
crucial parameters are either unknown or are based<br />
on an uncertain future.<br />
The traditional stochastic approach gives a<br />
probabilistic description of the unknown<br />
parameters on the basis of historical data. This is a<br />
very efficient approach when a substantial<br />
statistical base is available <strong>and</strong> reliable<br />
probabilistic laws can adequately describe<br />
parameters’ uncertainty <strong>and</strong> their possible<br />
outcomes (Infanger[1994]; Kall <strong>and</strong> Wallace<br />
[1994]; Ruszczynski[1997]). It is well known that<br />
stochastic optimisation approaches cannot be used<br />
when there is insufficient statistical information on<br />
data estimation to support the model, when<br />
probabilistic rules are not available, <strong>and</strong>/or when it<br />
is necessary to take into account information not<br />
derived from historical data.<br />
In these cases, the scenario analysis technique<br />
could be an alternative approach (Dembo[1991];<br />
Rockafellar <strong>and</strong> Wets[1991]). Scenario analysis<br />
can model many real problems where decisions are<br />
based on an uncertain future, whose uncertainty is<br />
described by means of a set of possible future<br />
outcomes, called "scenarios". Therefore, a scenario<br />
represents a possible realisation of some sets of<br />
uncertain data in the time horizon examined<br />
(Onnis et al., [1999]).<br />
The scenario analysis approach considers a set of<br />
statistically independent scenarios, <strong>and</strong> exploits the<br />
inner structure of their temporal evolution in order<br />
to obtain a "robust" decision policy, in the sense<br />
that the risk of wrong decisions is minimised.<br />
Some examples are given in Pallottino et al. [2003]<br />
for water resources management, in Mulvey <strong>and</strong><br />
Vladimirou[1989] for investment <strong>and</strong> production<br />
planning, in Glockner[1996] for air traffic<br />
771
management <strong>and</strong> in Hoyl<strong>and</strong> <strong>and</strong> Wallace [2001]<br />
for insurance policy <strong>and</strong> production planning.<br />
The aim of this paper is to generalize the<br />
effectiveness of scenario analysis when evaluating<br />
the risk of wrong decisions in order to reduce the<br />
negative consequences.<br />
In Pallottino et al. [2004] the authors analysed the<br />
scenario approach for water resources management<br />
offering some general rules for making a scenario<br />
tree from a predefined set of scenarios <strong>and</strong> for<br />
identifying a complete set of decision variables<br />
relative to all the scenarios under investigation. In<br />
this paper we extend that approach to general<br />
dynamic systems <strong>and</strong> propose a reoptimisation<br />
procedure, which facilitates reaching a robust<br />
solution <strong>and</strong> planning a part of the risk of wrong<br />
decisions caused by wrong assumptions on<br />
adopted parameters.<br />
2. DETERMINISTIC DYNAMIC CHANGE<br />
DYNAMIC MODEL OPTIMISATION<br />
MODEL<br />
In a deterministic dynamic framework we extend<br />
the analysis to a sufficiently wide time horizon <strong>and</strong><br />
assume a time step (time-period), t. The scale <strong>and</strong><br />
number of time-steps must be adequate to reach a<br />
significant representation of the variability the<br />
system components.<br />
A dynamic multi-period system is then generated<br />
by replicating the static basic system over time, for<br />
each time-period t, having previous knowledge of<br />
the time sequence of historical data. We then<br />
connect the corresponding copies for different<br />
consecutive periods by additional components<br />
carrying the information (decision) stored at the<br />
end of each period in such a way that the whole<br />
multi-period system is connected. We call these<br />
components inter-period components.<br />
A dynamic mathematical model is a mathematical<br />
model associated with a dynamic system. The data<br />
<strong>and</strong> the decision variables of the dynamic<br />
mathematical model are associated to each<br />
component of the dynamic system for each timeperiod<br />
t.<br />
In a deterministic approach, the database is derived<br />
from available historical data submitted to<br />
statistical validation on the basis of a forecast <strong>and</strong><br />
adopted as a reference scenario. In the<br />
deterministic optimisation model, we assume that<br />
the manager has previous knowledge of the time<br />
sequence of input data to the system. As a<br />
consequence, the solution obtained is strictly<br />
connected to the adopted scenario. We can<br />
formalize a model (P g ) for a specific scenario g, as<br />
an optimisation model:<br />
(P g ) min f g (x g )<br />
s.t.<br />
x g ∈ X g<br />
Once scenario g is adopted, where x g represents<br />
the vector comprehensive of all management <strong>and</strong><br />
planning variables for all time-periods t, f g (x g )<br />
represents the objective function of the problem<br />
<strong>and</strong>, x g ∈ X g , represents the set of all constraints<br />
(technical, physical, social, etc.) that are peculiar<br />
to the examined problem (st<strong>and</strong>ard constraints).<br />
The solution x g of problem (P g ) represents the set<br />
of decisions that should be adopted if scenario g<br />
takes place.<br />
3. CHANGE DYNAMIC OPTIMISATION<br />
MODEL<br />
Deterministic models are not adequate to describe<br />
the variability of some crucial parameters <strong>and</strong><br />
small differences in data in two different scenarios<br />
can produce significantly different solutions.<br />
Typically, most of the data in model (P g ) can be<br />
affected by a high level of uncertainty. In an<br />
uncertain environment the stochastic optimisation<br />
approach cannot be adopted since it is unreliable to<br />
match a valid occurrence probability to each<br />
scenario.<br />
The simulation approach studies a number of<br />
outcomes obtained by solving a number of<br />
optimisation problems (P g ) for each scenario g.<br />
During the optimisation process, different<br />
scenarios, corresponding to different dynamic<br />
multi-period models, proceed independently<br />
obtaining a different management policy for each<br />
scenario. Simulation verifies the performance of<br />
all policies selecting one for future decisions.<br />
Usually, to reach a viable management policy, a<br />
large number of scenarios must be considered. The<br />
simulation approach can prove very dem<strong>and</strong>ing<br />
from a computational point of view, especially if<br />
continuously replicated when the hydrological<br />
events occurring are very different from those<br />
foreseen in the selected scenario.<br />
The scenario analysis approach attempts to face<br />
the uncertainty factor by taking into account a set,<br />
G, of different supposed scenarios corresponding<br />
to the different possible time evolution of crucial<br />
data. Unlike simulation, the different scenarios are<br />
considered together to obtain a global set of<br />
decision variables on the whole set of scenarios.<br />
More precisely, two scenarios sharing a common<br />
initial portion of data must be considered together<br />
<strong>and</strong> partially aggregated with the same decision<br />
variables for the aggregated part, in order to take<br />
into account the two possible evolutions in the<br />
subsequent diverse parts. In this way, the set of<br />
parallel scenarios is aggregated by producing a tree<br />
structure, called scenario-tree. The aggregation<br />
772
ules guarantee that the solution in any given<br />
period is independent of the information not yet<br />
available. This result can be obtained by inserting<br />
congruity constraints which require that the<br />
subsets of decision variables, corresponding to the<br />
indistinguishable part of different scenarios, must<br />
be equal among themselves In other words, model<br />
evolution is only based on the information<br />
available at the moment. (Rockafellar <strong>and</strong> Wets,<br />
[1991]).<br />
The problem supported by the scenario tree, is<br />
described by a mathematical model that includes<br />
all single-scenario problems (P g ), ∀g∈G, plus<br />
some inter-scenario linking constraints<br />
representing the requirement that if two scenarios<br />
g1 <strong>and</strong> g2 are identical up to time t on the basis of<br />
information available at that time, then the<br />
corresponding set of decision variables, x 1 <strong>and</strong> x 2 ,<br />
must be identical up to time t. These constraints<br />
represent the congruity requirement that the<br />
subsets of decision variables corresponding to the<br />
indistinguishable part of different scenarios must<br />
be equal among themselves. Moreover, a weight<br />
can be assigned to each scenario representing the<br />
“importance” assigned by the manager to the<br />
running configuration. At times the weights can be<br />
viewed as the probability of occurrence of the<br />
examined scenario. More often they are<br />
determined on the basis of background knowledge<br />
about the system.<br />
The resulting mathematical model is named<br />
chance-model to indicate that it is not<br />
stochastically based but, due to the impossibility of<br />
adopting probabilistic rules <strong>and</strong>/or to the necessity<br />
of inserting information that cannot be deduced<br />
from historical data, it attempts to represent the set<br />
scenario g2<br />
scenario s1<br />
resource<br />
scenario g1<br />
0 5 10 τ 15 20 25 30 35 40 45 50<br />
time-periods<br />
Figure 1. Stored resources in scenario <strong>and</strong> deterministic optimisation.<br />
of possible performances of the system, as<br />
uncertain parameters vary.<br />
The chance model (P C ) can have the following<br />
structure:<br />
(P C ) min Σ g w g f g (x g )<br />
s.t.<br />
x g ∈ X g<br />
x* ∈ S<br />
∀g∈ G<br />
Where:<br />
w g represents the weight assigned to a scenario<br />
g∈G; x* represents the vector of variables<br />
submitted to congruity constraints; x g ∈ X g<br />
represents the set of st<strong>and</strong>ard constraints for each<br />
scenario g ; x* ∈ S, represents the set of congruity<br />
constraints.<br />
The objective function is the weighted sum of the<br />
objective functions of problems (P g ) <strong>and</strong> all<br />
st<strong>and</strong>ard constraints are included.<br />
Congruity constraints require that the decision<br />
variables in those scenarios that are<br />
indistinguishable up to a specific time τ<br />
(branching-time) are the same up to τ. Specifically,<br />
the decisions at the end of the time τ , must be the<br />
same of those at the beginning of period τ+1 .<br />
To generate the set G of scenarios, different<br />
approaches such as Monte Carlo generation<br />
scheme, Neural network techniques or ARMA<br />
models can be performed. The aim of this paper is<br />
not to detail these procedures <strong>and</strong> we assume that<br />
the set G is available.<br />
Regarding weight definitions, if the manager were<br />
able to evaluate the weight w g as the probability<br />
773
that scenario g will occur, he could estimate it by<br />
some stochastic technique or statistical test. More<br />
often the manager has few, if any, possibilities to<br />
do this due to the difficulty in deriving a<br />
probabilistic rule from conceptual considerations.<br />
Instead, in scenario analysis, a weight w g assigned<br />
to a scenario g can be interpreted as the "relative<br />
importance" of that scenario in the uncertain<br />
environment. In other words, in scenario analysis,<br />
weights are interpreted as subjective parameters<br />
assigned on the basis of the experience of the<br />
water management board.<br />
3.1 A sample system<br />
To illustrate the scenario analysis approach we<br />
refer to a sample dynamic supply-dem<strong>and</strong> system<br />
with a resource supply <strong>and</strong> a dem<strong>and</strong> centre. The<br />
supply centre can deliver a resource or store it to<br />
deliver in a successive time-period. We assume<br />
that the dimensions of the supply <strong>and</strong> dem<strong>and</strong><br />
centres are known, <strong>and</strong> that the system is<br />
operational. We want to determine the resource<br />
management policy over a time horizon such that<br />
the known resource dem<strong>and</strong> is satisfied (as much<br />
as possible) <strong>and</strong> the total cost is minimized.<br />
Objective function <strong>and</strong> constraints will be<br />
analytically expressed on the basis of the feature of<br />
the examined system. Variables of the optimisation<br />
problem, for each scenario g at time-period t, are<br />
referred to stored resource y t g, delivered resource<br />
from supply centre to dem<strong>and</strong> centre z t g. Resource<br />
dem<strong>and</strong> p is assigned <strong>and</strong> we suppose that<br />
historical data are available. Deficits u t g can be<br />
then calculated as the difference between dem<strong>and</strong><br />
p <strong>and</strong> delivered resources z t g, in each time-period t.<br />
We then generate two scenarios, g1 <strong>and</strong> g2,<br />
assuming that uncertain parameters correspond to<br />
resource supplies in supply centre in period t in<br />
scenario g.<br />
The two scenarios are both identical to the<br />
historical data up to branching timeτ. We suppose<br />
that scenario g2 follows the historical data from<br />
τ +1 to the last time-period, while scenario g1 has<br />
the resource supplies reduced by 50% with respect<br />
to it (“scarce” scenario). This means that two<br />
different possible resource supply configurations<br />
can occur. Finally the two scenarios run until they<br />
dem<strong>and</strong> p<br />
scenario s1<br />
scenario g2<br />
resource<br />
scenario g1<br />
0 5 10 τ 15 20 25 30 35 40 45 50<br />
time-periods<br />
Figure 2. Resources delivered to dem<strong>and</strong> centre in scenario <strong>and</strong> deterministic optimisation.<br />
reach the end of the time-horizon. The<br />
optimisation model requires minimizing a function<br />
representing the total weighted cost Σ g Σ t w g f g ( , y t g,<br />
z t g, u t g ) subject to st<strong>and</strong>ard <strong>and</strong> congruity<br />
constraints.<br />
To illustrate, we show some possible results<br />
concerning stored resources in supply centre, y t g,<br />
<strong>and</strong> resources delivered to dem<strong>and</strong> centre, z t g,<br />
obtained by scenario analysis, solving the above<br />
optimisation chance model.<br />
Figure 1 shows stored resources, y t g, obtained by<br />
scenario analysis <strong>and</strong> those obtained by a<br />
deterministic optimisation model when the<br />
“scarce” scenario g1 is assumed as database. When<br />
scenario g1 is considered independently, it is<br />
referred to as s1. The resulting graph represents the<br />
decisions that would be made for transferring<br />
resources in a deterministic optimisation process.<br />
The zone between the two graphics of the<br />
aggregated scenarios, g1 <strong>and</strong> g2, represents the<br />
possible decisions that can be made for stored<br />
resources. Therefore we can say that any part of s1<br />
not between g1 <strong>and</strong> g2 represents the error that the<br />
774
manager would have made if he had adopted<br />
decision s1.<br />
Figure 2 shows the resources delivered, z t g, from<br />
supply centre to the dem<strong>and</strong> centre. The behaviour<br />
of these flows shows that in the scenario g2<br />
dem<strong>and</strong> is fulfilled while in scenario g1 deficits are<br />
present after branching time τ. But, comparing this<br />
with results in deterministic optimisation under<br />
scenario s1, we can see that as regards the scarcity<br />
of resources conditions, scenario optimisation<br />
gives a smoother distribution, i.e., with a lower<br />
variance of resource distribution in scenario g1<br />
even though the average is almost the same as<br />
scenario s1. Thus, when planning for scarce<br />
resources, scenario analysis provides less dramatic<br />
<strong>and</strong> more easily implementable results then using<br />
deterministic optimisation to determine<br />
management policy.<br />
3.2 A barycentric chance reoptimisation model<br />
In the previous section we showed how scenario<br />
analysis could be more useful than the<br />
deterministic approach in deciding management<br />
policy. This can be crucial if scarce resources<br />
events occur <strong>and</strong> a rationing policy must be<br />
adopted. But, an effective management policy<br />
must be able to establish a target value for<br />
delivering resources to the dem<strong>and</strong> centre. The<br />
community suffers less from resource rationing if<br />
it has been forewarned of a possible shortage. This<br />
target value should take into account the entire<br />
range of possible scenarios of resource availability,<br />
neither too pessimistic in case of abundance, nor<br />
too optimistic in case of scarcity of resources. In<br />
other words, a target value should be sufficiently<br />
barycentric in respect to the different possible<br />
scenarios that could take place. Establishing the<br />
resource dem<strong>and</strong> level at this target value would<br />
permit notifying the resource users (the<br />
community) in a timely fashion. As a consequence,<br />
preventive measures could be adopted in order to<br />
avoid, at least in part, damages derived from an<br />
unexpected drastic cut in resources (water, oil, raw<br />
materials, currency, transportation <strong>and</strong><br />
telecommunications, etc.).<br />
dem<strong>and</strong> p<br />
baricentric<br />
value<br />
programmed deficits<br />
unprogrammed deficits<br />
0 5 10 15 20 25 30 35 40 45<br />
time-periods<br />
Figure 3. Resources delivered to dem<strong>and</strong> centre in deterministic reoptimisation.<br />
If xˆ tg are the decision variables representing the<br />
resources that can be delivered to a dem<strong>and</strong> centre<br />
in time-period t under scenario g, we want to<br />
determine a target dem<strong>and</strong> as the value x b that is<br />
barycentric with respect to all xˆ tg . To obtain this<br />
value we introduce in the objective function of<br />
problem (P C ) a function measuring the weighted<br />
distance from x b to xˆ tg for all g <strong>and</strong> t. If we adopt<br />
the Euclidean norm to measure this distance, the<br />
chance barycentric model (P B ) can be expressed<br />
as:<br />
(P B ) min Σ g w g f g (x g ) + Σ g Σ t<br />
λ g<br />
( xˆ tg– x b ) 2<br />
s.t.<br />
x g ∈ X g<br />
∀g∈ G<br />
x* ∈ S<br />
where λ g<br />
.is the weight associated to the norm.<br />
Once the value x b is determined, a reoptimisation<br />
process can be adopted in order to identify the<br />
sensitivity of the examined system with respect to<br />
deficit programming.<br />
We construct a deterministic dynamic model in<br />
which the predefined dem<strong>and</strong> is settled equal to the<br />
barycentric value x b <strong>and</strong> adopting as data input,<br />
those corresponding to the most crucial scenario<br />
(e.g. what the manager considers the most risky for<br />
the system). The difference between the new<br />
configuration of delivered resources in each time-<br />
775
period t <strong>and</strong> the value x b , identifies the set of<br />
programmed deficits for the system.<br />
In the sample system illustrated in the previous<br />
section we determine a value z b in such a way that<br />
it is barycentric with respect to all z t g. We then<br />
reoptimise the system solving a deterministic<br />
model assigning to the dem<strong>and</strong> centre the obtained<br />
value z b as target value <strong>and</strong> adopt, as data input,<br />
those corresponding to scarce scenario. Figure 3<br />
shows the resources delivered to the dem<strong>and</strong><br />
centre in the reoptimisation phase together with the<br />
programmed deficits (difference between the new<br />
configuration of delivered resources in each timeperiod<br />
t <strong>and</strong> the value x b ) <strong>and</strong> unprogrammed<br />
deficits (difference between the original resource<br />
dem<strong>and</strong> <strong>and</strong> the value x b ). Moreover, comparing<br />
the behaviour of delivered resources with that<br />
showed in figure 2, we observe that management<br />
policy is even better than the policy corresponding<br />
to scenario g2. The programming of deficits makes<br />
it possible to set up adequate preventive measures,<br />
which permit a notable reduction in the event of<br />
resources scarcity.<br />
4 CONCLUSIONS<br />
In this paper we showed how scenario analysis can<br />
be more useful than the deterministic approach in<br />
deciding system management policy when a level<br />
of uncertainty affects data input such as supply <strong>and</strong><br />
dem<strong>and</strong> patterns. Decision policy under<br />
uncertainty condition can be crucial if scarce<br />
resources events occur <strong>and</strong> a rationing programme<br />
must be adopted. The scenario analysis approach<br />
considers a set of statistically independent<br />
scenarios, <strong>and</strong> exploits the inner structure of their<br />
temporal evolution in order to obtain a "robust"<br />
decision policy, in the sense that the risk of wrong<br />
decisions is minimised. This can be done by a<br />
reoptimisation deterministic process using a<br />
barycentric value derived from a previous scenario<br />
optimisation. Finally, this make it possible to<br />
identify programmed deficits to control the<br />
negative consequences deriving from wrong<br />
decisions allowing the system manager to adopt<br />
preventive measures avoiding, at least in part,<br />
damages derived from an unexpected drastic cut in<br />
resources.<br />
Glockner, G. D., Nemhauser G. L., 2000. A<br />
dynamic network flow problem with<br />
uncertain arc capacity: formulation <strong>and</strong><br />
problems structure. Operations Research 48<br />
(2), 233-242.<br />
Hoyl<strong>and</strong>, K., Wallace, S. W., 2001. Generating<br />
scenario trees for multistage decision<br />
problems. Management Science 47 (2), 295-<br />
307.<br />
Infanger, G., 1994. Planning under uncertainty:<br />
Solving large-scale stochastic linear<br />
programming. Boyd & Fraser Publishing<br />
Company, Danvers, MA.<br />
Kall, P., Wallace, S. W., 1994. Stochastic<br />
Programming, John Wiley <strong>and</strong> Sons, New<br />
York.<br />
Mulvey, J. M., Vladimirou, H., 1989. Stochastic<br />
network optimisation models for investment<br />
planning. Annals of Operation Research 20,<br />
187-217.<br />
Onnis, L., Sechi, G. M, Zuddas, P., 1999.<br />
Optimisation processes under uncertainty,<br />
A.I.C.E., Milano, 238-244.<br />
Pallottino, S., Sechi, G.M., Zuddas, P., 2004. A<br />
DSS for Water Resources Management under<br />
Uncertainty by Scenario Analysis, to appear in<br />
<strong>Environmental</strong> <strong>Modelling</strong> & <strong>Software</strong>, Special<br />
Issue.<br />
Rockafellar, R. T., Wets, R. J. B., 1991. Scenarios<br />
<strong>and</strong> policy aggregation in optimisation under<br />
uncertainty. Mathematics of Operations<br />
Research 16, 119-147.<br />
Ruszczynski, A., 1997. Decomposition methods in<br />
stochastic programming. Mathematical<br />
Programming 79, 333-353.<br />
5 REFERENCES<br />
Dembo, R., 1991. Scenario optimisation. Annals of<br />
Operations Research 30, 63-80.<br />
Glockner, G. D., 1996. Effects of air traffic<br />
congestion delays under several flow<br />
management policies. Transportation<br />
Research Record 1517, 29-36.<br />
776
Reliable <strong>and</strong> Valid Identification of a Small Number of<br />
Global Emission Scenarios<br />
Olaf Tietje, SystAim Zurich,<br />
olaf.tietje@systaim.ch<br />
Abstract: Numerous scenarios of global greenhouse gas emissions have been created, that are<br />
difficult to communicate to decision makers. To identify few significantly different <strong>and</strong> consistent scenarios<br />
is time consuming, requires deep underst<strong>and</strong>ing of the underlying driving forces, <strong>and</strong> may depend on the<br />
individual perspective of the scenario analyst. Developed from an expert based scenario technique a new<br />
method was developed, which in step 1 analyzes each given scenario with respect to the relations between its<br />
characteristics (e.g. parameters, state variables). This analysis may include a very large number of qualitative<br />
('nominal'), logical, ordinal <strong>and</strong> metric characteristics. In step 2, a few consistent <strong>and</strong> significantly different<br />
scenarios are reliably determined according to modifications of three recently published scenario<br />
identification methods. The comparison of the different methods for scenario identification shows the<br />
convergent validity of the methodology. The presentation sketches the mathematical background of the<br />
analysis <strong>and</strong> of the identification <strong>and</strong> shows results of an application to the IPCC emission scenarios. It is<br />
concluded that the quantitative scenario selection procedure presented is very helpful for the communication<br />
of scenarios to decision makers. Because the mathematical methodology complies with approaches used in<br />
qualitative scenario techniques, in which experts estimate scenario consistency, a combination of qualitative<br />
<strong>and</strong> quantitative knowledge could be possible, but has not yet been investigated.<br />
Keywords:<br />
Scenario analysis; greenhouse gas emissions; consistency analysis.<br />
1 INTRODUCTION<br />
The IPCC data base on emission scenarios<br />
currently contains 633 scenarios from 197 sources<br />
(Morita, 1999). A part of these scenarios is<br />
published on the web by the Center for<br />
<strong>International</strong> Earth Science Information Network<br />
(CIESIN, 2004). The summary for policy makers<br />
(Nakicenovic, 2000) presents 6 scenarios, which<br />
‘illustrate all scenario groups’. They refer to four<br />
different story lines which were defined in the<br />
open process described in the Special Report on<br />
Emissions Scenarios (SRES) Terms of Reference<br />
(Nakicenovic, 2000). The story lines are output of<br />
a qualitative procedure (confer Alcamo, 2001)<br />
<strong>and</strong> input to the quantitative scenario modeling<br />
efforts. The scenarios cover a wide range of<br />
uncertainty. The uncertainties of each of the<br />
single scenarios overlap largely. Therefore it is<br />
very difficult for policy makers to extract the<br />
information that they need.<br />
The methodology used here is based on three<br />
scenario selection procedures presented in Tietje<br />
(2004), which were shown to be reliable <strong>and</strong><br />
valid. Those selection procedures rely on<br />
consistency ratings by experts. This paper uses a<br />
specific method to estimate scenario consistencies<br />
based on quantitative data <strong>and</strong> then applies one of<br />
the selection procedures, namely the max-min<br />
selection.<br />
The goal of this investigation is to test a<br />
procedure that could be applied to both, the<br />
construction of scenarios, <strong>and</strong> the reliable <strong>and</strong><br />
valid selection of few emission scenarios. This<br />
paper describes a procedure to evaluate <strong>and</strong> select<br />
a small number of scenarios out of the 40 SRES<br />
illustrative / maker scenarios. The procedure can<br />
also be used to evaluate climate model input<br />
parameters, so that scenarios can then be derived<br />
from substantially different sets of parameters.<br />
777
2 MATERIAL<br />
3 METHODS<br />
The basis for the application of the procedure are<br />
the scenarios published on the web by the Center<br />
for <strong>International</strong> Earth Science Information<br />
Network (CIESIN, 2004). These 40 scenarios are<br />
described by 38 quantitative characteristics<br />
calculated for 12 points of time. From these 456<br />
(38 times 12) variables 25 characteristics (see<br />
Table 1) <strong>and</strong> three points of time (2020, 2050,<br />
2100) have been selected so that each scenario is<br />
described by 75 variables. The selection left out<br />
the total of l<strong>and</strong> use (which is constant), a second<br />
GNP/GDP index, six kinds of final energy <strong>and</strong><br />
six kinds of primary energy (but the total is used<br />
in these two cases), <strong>and</strong> carbon sequestration data<br />
(which is only available for the 4 ASF scenarios).<br />
Note that some of the variables are not normally<br />
distributed <strong>and</strong> some are correlated.<br />
Table 1 Selected quantitative characteristics to<br />
describe the IPCC scenarios referred to here<br />
(EJ=Exa Joule, ZJ=Zeta Joule, Gt=Giga tons,<br />
Mt=Mega tons, further explanations see<br />
Nakicenovic et al. 2000, CIESIN 2004)<br />
Characteristic<br />
Population<br />
GNP/GDP (mex)<br />
Final Energy Total<br />
Primary Energy Total<br />
Cumulative Resources Use<br />
Coal<br />
Oil<br />
Gas<br />
Cumulative CO2 Emissions<br />
L<strong>and</strong> Use<br />
Cropl<strong>and</strong><br />
Grassl<strong>and</strong>s<br />
Energy Biomass<br />
Forest<br />
Others<br />
Anthropogenic Emissions<br />
Fossil Fuel CO2<br />
Other CO2<br />
Total CO2<br />
CH4 total<br />
N2O total<br />
SOx total<br />
CFC/HFC/HCFC<br />
PFC<br />
SF6<br />
CO<br />
NMVOC<br />
NOx<br />
Unit<br />
Million<br />
Trillion US$<br />
EJ<br />
EJ<br />
ZJ<br />
ZJ<br />
ZJ<br />
Gt C<br />
Million ha<br />
Million ha<br />
Million ha<br />
Million ha<br />
Million ha<br />
Gt C<br />
Gt C<br />
Gt C<br />
Mt CH4<br />
Mt N2O-N<br />
Mt S<br />
Mt C eq.<br />
Mt C eq.<br />
Mt C eq.<br />
Mt CO<br />
Mt<br />
Mt N<br />
Scenarios are hypothetical visions into the future.<br />
The IPCC SRES scenarios are quantitative<br />
scenarios. For a given story line, a set of model<br />
parameters is constructed, that is used in a<br />
mathematical simulation model for the<br />
quantitative calculation representing the story<br />
line. Note that the story lines have been created<br />
intuitively by a group of experts.<br />
A formative scenario analysis (Scholz & Tietje,<br />
2002; often called scenario technique) does not<br />
use a quantitative simulation model, but a formal<br />
procedure to construct scenarios. This kind of<br />
scenario analysis is often used to construct<br />
economic scenarios. In a formative scenario<br />
analysis the space of possible scenarios is defined<br />
by the set product (here n=75)<br />
S y y y<br />
: = 1× 2× K × n<br />
In formative scenario analysis the scenario space<br />
is finite. Therefore each variable was classified<br />
into 5 equidistant classes between the corresponding<br />
minimum <strong>and</strong> maximum values.<br />
The methodology applied here consists of two<br />
steps. In the first step a consistency matrix is<br />
estimated, with which the consistency of each<br />
scenario can be calculated. In the second step the<br />
max-min scenario selection method is applied to<br />
select few scenarios.<br />
3.1 Consistency estimation<br />
Each combination of variable levels (classes)<br />
occurring in one of the 40 scenarios described<br />
above is summarized into a consistency matrix<br />
mi<br />
m j<br />
( i j ) , = 1,..., ; = 1,..., ; = 1,...,<br />
C: = c y , y<br />
i j n mi ni mj nj<br />
where i,j are the variable numbers <strong>and</strong> the m i , m j<br />
are the class numbers of a scenario (see Tietje,<br />
2004). If for a scenario a level is undefined<br />
because the variable has not been calculated for<br />
that scenario the corresponding combination of<br />
levels does not count. The sum of level<br />
combinations is classified into the four classes -1<br />
to +2.<br />
The resulting consistency matrix C can then be<br />
used to evaluate the consistency of any possible<br />
scenario<br />
778
*<br />
add<br />
n i−1<br />
mi<br />
m j<br />
k = ∑∑ add i j<br />
i= 2 j=<br />
1<br />
c ( S ) c ( y , y )<br />
where S k is the vector of variable classes k. The<br />
resulting consistencies calculated for the 40<br />
scenarios are presented in Table 2.<br />
currently not considered, but could be used in a<br />
model to arrive at that additional scenario. But,<br />
the calculations showed that any scenario from S<br />
that was consistent is already very similar or<br />
equal to one of the 40 scenarios described above.<br />
This result indicates that the 40 IPCC SRES<br />
scenarios seem to cover a large or even the full<br />
range of possible scenarios.<br />
3.2 Scenario selection methods<br />
The max-min-selection is an iterative procedure<br />
that selects a scenario with the maximum<br />
consistency among the set of scenarios that have a<br />
minimum distance to each of the previously<br />
selected scenarios.<br />
The distance measure simply is the number of<br />
differences between the scenarios:<br />
n<br />
⎧1 if yi( Sk) ≠ yi( Sl)<br />
dS ( k , S l)<br />
= ∑⎨ ⎩ 0 otherwise<br />
i=<br />
1<br />
The investigation of sets of scenarios, selected<br />
with (1) the max-min-selection, (2) the distanceto-selected<br />
method, <strong>and</strong> (3) the local efficiency<br />
method, has shown (a) the reliability of the<br />
selection methods, because the repeated selection<br />
with any of the methods lead to the same<br />
scenarios <strong>and</strong> (b) the convergent validity of the<br />
scenario selection methods, because all three<br />
substantially different methods lead to nearly the<br />
same set of selected scenarios (Tietje, 2004).<br />
Hence, the max-min-selection used here proved to<br />
be a reliable <strong>and</strong> valid procedure to select few<br />
consistent scenarios (Tietje, 2004) out of all<br />
possible scenarios S or any subset, such as the set<br />
of IPCC SRES scenarios.<br />
4 RESULTS<br />
4.1 Additional scenarios<br />
The estimation of the consistency matrix was used<br />
to find out, whether there are additional scenarios<br />
that are consistent in the sense that any<br />
combination of levels of such an additional<br />
scenario already occurred in some of the 40<br />
scenarios described above. The rationale behind<br />
this consistency is that if there is an additional<br />
consistent scenario then there might also be an<br />
additional consistent set of parameters that is<br />
Table 2 Scenarios <strong>and</strong> estimated consistencies<br />
according to c * add<br />
Consistency<br />
A1 AIM 342<br />
A1 ASF 215<br />
A1 IMAGE 305<br />
A1 MESSAGE 466<br />
A1 MINICAM 368<br />
A1 MARIA 287<br />
A1C AIM 271<br />
A1C MESSAGE 416<br />
A1C MINICAM 358<br />
A1G AIM 302<br />
A1G MESSAGE 464<br />
A1G MINICAM 334<br />
A1V1 MINICAM 504<br />
A1V2 MINICAM 440<br />
A1T AIM 443<br />
A1T MESSAGE 551<br />
A1T MARIA 329<br />
A2 ASF 93<br />
A2 AIM 83<br />
A2G IMAGE 53<br />
A2 MESSAGE 97<br />
A2 MINICAM 129<br />
A2-A1 MINICAM 123<br />
B1 IMAGE 301<br />
B1 AIM 376<br />
B1 ASF 165<br />
B1 MESSAGE 444<br />
B1 MARIA 180<br />
B1 MINICAM 250<br />
B1T MESSAGE 422<br />
B1HIGH MESSAGE 468<br />
B1HIGH MINICAM 260<br />
B2 MESSAGE 276<br />
B2 AIM 305<br />
B2 ASF 101<br />
B2 IMAGE 78<br />
B2 MARIA 110<br />
B2 MINICAM 222<br />
B2HIGH MINICAM 235<br />
B2C MARIA 119<br />
779
Figure 1 Distance between scenarios measured by the number variables with equal levels<br />
4.2 Scenario selection<br />
To evaluate the scenario selection procedure it<br />
was applied to the 40 IPCC SRES scenarios. The<br />
required minimum distance between the scenarios<br />
was initially set to 75 so that only completely<br />
different scenarios could be obtained. Due to the<br />
classification into only 5 classes there was only<br />
one scenario found. The other scenarios have a<br />
maximum distance of 68 to the first scenario.<br />
Hence for a required minimum distance of 68 or<br />
lower additional scenarios are obtained when<br />
applying the max-min-selection. With a minimum<br />
distance of 53 the 6 scenarios shown in Figure 1<br />
are selected. As presented by the SRES, there are<br />
three scenarios from scenario family A1 <strong>and</strong> one<br />
scenario from each of the other scenario families<br />
(A2, B1, B2, see Nakicenovic, 2000). Further<br />
inspection of the selection results indicates that at<br />
least a considerable number of the IPCC SRES<br />
scenarios are quite well distributed, i.e. the<br />
scenarios are different enough from each other to<br />
be seriously considered each. A distance of 53<br />
between two scenarios means that two thirds of<br />
all variables fall into a different of only 5 classes.<br />
4.3 Differences between scenarios of<br />
each modeling team<br />
differences between the 9 scenarios of the<br />
MESSAGE team. There are a quite large number<br />
of equal levels between scenarios within the A1<br />
<strong>and</strong> within the B1 family. Similar results are<br />
obtained for all teams that provided multiple<br />
scenarios within one family (AIM, MESSAGE,<br />
MINICAM).<br />
The AIM team provided scenarios that are more<br />
similar to each other (on average 25 equal levels<br />
between the four scenarios A1, A2, B1, <strong>and</strong> B2,<br />
see Table 3) <strong>and</strong> the IMAGE team more different<br />
scenarios (15 equal levels between the scenarios<br />
on average, see Table 3).<br />
Table 3 Aggreement of scenarios for different<br />
scenario families<br />
Modeling Team Average number of equal<br />
levels between scenarios<br />
A1, A2, B1, B2<br />
AIM 24.83<br />
ASF 16.67<br />
IMAGE 14.67<br />
MESSAGE 21.17<br />
MINICAM 20.33<br />
MARIA 20<br />
Six modeling teams provided different numbers<br />
of emission scenarios. Figure 2 shows the<br />
780
4.4 Differences between scenarios<br />
from each family<br />
In the scenario families the scenarios of the six<br />
modeling teams are quite different (see Table 4).<br />
The B1 scenarios show the largest average<br />
number of equal levels. The average number of<br />
equal levels within the scenarios from a modeling<br />
team is less than the average number of equal<br />
levels within the scenarios from a scenario family.<br />
Please note that the difference between the<br />
averages is small.<br />
Table 4 Aggreement of scenarios from<br />
different modeling teams<br />
Scenario<br />
Family<br />
A1 25.3<br />
A2 26.4<br />
B1 32.4<br />
B2 27.7<br />
Average number of equal levels<br />
between scenarios provided by AIM,<br />
ASF, IMAGE, MESSAGE, MINICAM,<br />
MARIA<br />
5 Discussion<br />
The test of the procedure to evaluate scenarios<br />
reveals that the 40 IPCC SRES scenarios build a<br />
valuable set of scenarios. The set seems covering,<br />
because no additional scenarios could be found<br />
yet. The set shows considerably different<br />
scenarios.<br />
It has to be admitted, that the approach applied<br />
here is rather coarse, because each variable is<br />
represented by only 5 classes, the consistency<br />
matrix has only 4 classes, <strong>and</strong> the distance<br />
function only counts the number of differences.<br />
The results are credible, because they partly<br />
repeat what was expected: There is more<br />
similarity between the scenarios from each<br />
scenario family than between scenarios from each<br />
modeling team, although the difference in<br />
similarity levels is small. The differences between<br />
scenario variants are small – such as between<br />
variants B1 MESSAGE <strong>and</strong> B1T MESSAGE <strong>and</strong><br />
B1HIGH MESSAGE.<br />
The characteristics of the scenario stories can be<br />
reproduced by the data analysis. For example the<br />
scenario selection procedure results in the six<br />
illustrative marker scenarios emphasized by the<br />
summary for policymakers.<br />
Figure 2 Distance between 9 scenarios provided by the MESSAGE team<br />
781
Therefore the proposed approach seems to have<br />
resulted in a tool that is available to<br />
- construct <strong>and</strong> evaluate scenarios <strong>and</strong> sets<br />
of scenarios<br />
- possibly integrate scenario efforts of<br />
different groups<br />
- contribute to the choice of climate<br />
scenarios which get decision relevant<br />
when being used within further<br />
formative scenario analyses,<br />
- in this way contribute to regarding<br />
climate as decision relevant for<br />
administrative <strong>and</strong> business strategies<br />
<strong>and</strong> plans<br />
The proposed approach supports also to develop a<br />
shift in perspective from investigating single<br />
scenarios to regarding the full scale set of<br />
scenarios for decision purposes.<br />
The approach used a considerably large number<br />
of variables describing the scenarios. The<br />
performance seems to make a more rigid scenario<br />
construction feasible in the sense that scenario<br />
construction already begins formally when the<br />
model parameters are being determined. A rigid<br />
data analysis might lead to construct a scenario as<br />
a set of model parameters (before modeling take<br />
place, story-free scenarios). The aim would be to<br />
derive storylines from constructed scenarios –<br />
rather than deriving scenarios from intuitively<br />
determined storylines..<br />
Another possibility would be to use all 633 SRES<br />
emission scenarios in order to repeat the current<br />
investigation aiming at a further improvement of<br />
the conclusions for policy makers.<br />
Acknowledgements<br />
I am grateful to Prof. Nakicenovic for the<br />
possibility to use the IPCC SRES scenarios. The<br />
investigation was supported by SystAim (Zürich).<br />
References<br />
Alcamo, J., Scenarios as tools for international<br />
environmental assessments, European<br />
Environment Agency, <strong>Environmental</strong> Issue<br />
Report No. 24. Copenhagen, 2001.<br />
CIESIN, SRES illustrative / marker scenarios<br />
(Version 1.1, July 2000), 2004.<br />
Morita, T., Emission Scenario Database prepared<br />
for IPCC Special Report on Emission<br />
Scenarios convened by Dr. Nebosja<br />
Nakicenovic., 1999.<br />
Nakicenovic, N. a., <strong>and</strong> Intergovernmental Panel<br />
on Climate Change Working Group 3<br />
Policy, Special report on emissions<br />
scenarios, pp. 599 S., Cambridge<br />
University Press, Cambridge, 2000.<br />
Scholz, R. W., <strong>and</strong> O. Tietje, Embedded Case<br />
Study Methods: Integrating Quantitative<br />
And Qualitative Knowledge, pp. xvi+392,<br />
Sage, Thous<strong>and</strong> Oaks, 2002.<br />
Tietje, O., Identification of a small reliable <strong>and</strong><br />
efficient set of consistent scenarios,<br />
European Journal of Operational<br />
Research(in press), 2004.<br />
782
SIMULATING GLOBAL FEEDBACKS BETWEEN SEA LEVEL<br />
RISE, WATER FOR AGRICULTURE AND THE COMPLEX<br />
SOCIO-ECONOMIC DEVELOPMENT OF THE IPCC SCENARIOS<br />
Saskia Werners 1 ; Roelof Boumans 2 ; Laurens Bouwer 3<br />
1) Alterra, Wageningen University <strong>and</strong> Research Centre, NL. werners@mungo.nl;<br />
2) Gund Institute for Ecological Economics, The University of Vermont, USA;<br />
3) Institute for <strong>Environmental</strong> Studies, Vrije Universiteit Amsterdam, NL<br />
ABSTRACT<br />
Nature's way of dealing with unhealthy conditions is unfortunately not one that compels us to conduct a<br />
solvent hygiene on a cash basis.<br />
1919, George Bernard Shaw<br />
The Global Unified Meta-model of the BiOsphere (GUMBO) was used to simulate how the socioeconomic<br />
conditions specified in the Special Report on Emission Scenarios (SRES) of the IPCC<br />
(Intergovernmental Panel on Climate Change) influence vulnerability to climate change. Input parameters<br />
are the consumer preferences, investment strategies, natural resources management <strong>and</strong> technological<br />
development associated with the SRES scenarios. From this input the characteristic SRES driving forces<br />
population growth, economic development <strong>and</strong> fossil fuel use were reproduced in GUMBO with the<br />
corresponding climate scenarios (temperature change, sea level rise <strong>and</strong> rainfall patterns). This article<br />
shows alternative pathways of development exist that yield the same SRES driving forces but that differ<br />
significantly in their vulnerability to sea level rise <strong>and</strong> water availability. It concludes that an assessment<br />
of the relative vulnerability of the SRES scenarios that takes into account the socio-economic<br />
characteristics of these scenarios, can challenge assessments based on climate change <strong>and</strong> the driving<br />
forces only. The assessment of alternative complex socio-economic conditions is an important addition to<br />
underst<strong>and</strong> our world’s vulnerability to climate change. GUMBO offers a promising, flexible <strong>and</strong> fast<br />
environment for the assessment. The GUMBO model <strong>and</strong> documentation can be downloaded from<br />
www.uvm.edu/giee/GUMBO.<br />
Keywords: Socio-economic Scenarios, Global Change <strong>Modelling</strong>, Dynamic Feedback<br />
1 INTRODUCTION<br />
This study simulates the dynamic feedback<br />
between different pathways of socio-economic<br />
development <strong>and</strong> climate change. It builds on the<br />
scenarios published by the Intergovernmental<br />
Panel on Climate Change (IPCC) in its Special<br />
report on emission scenarios (SRES) (IPCC,<br />
2000). The SRES scenarios offer a well<br />
documented regime of plausible future for the<br />
world which provides a meaningful basis for<br />
impact assessments (Arnell et al, 2004). The four<br />
SRES scenarios are created in three successive<br />
steps: (1) qualitative storylines represent a<br />
diverse range of different socio-economic<br />
development pathways for the world, (2) the<br />
storylines are translated into quantitative driving<br />
forces, that are harmonised projections of the<br />
indicators population growth, economic<br />
development, technology, energy <strong>and</strong> l<strong>and</strong>-use,<br />
<strong>and</strong> (3) greenhouse gas emission scenarios are<br />
calculated from the driving forces. So far most<br />
analyses of climate change use the driving forces<br />
to characterise the SRES scenarios (e.g., Parry,<br />
2004, Alcamo, 2002; Kabat, 2003). Few use the<br />
socio-economic characteristics of the underlying<br />
storylines (Arnell et al, 2004). The goal of this<br />
study is to underst<strong>and</strong> how complex dynamic<br />
socio-economic conditions influence our world’s<br />
vulnerability to climate change in the coming<br />
century. More specific this paper reports on the<br />
simulation of the dynamic feedback between sea<br />
level rise <strong>and</strong> water availability in agriculture<br />
<strong>and</strong> the world of two of the SRES scenarios.<br />
This paper explicitly takes a systems perspective<br />
in studying global change. The principles of<br />
783
system dynamics allow studying interrelationships<br />
<strong>and</strong> patterns of change in how human<br />
activities are altering the Earth, impacting the life<br />
support system upon which humans depend<br />
(SDU, 2004; Steffen, 2004; Meadows et al,<br />
1992). Emerging insights from system analysis<br />
are how differences flowing from different<br />
pathways of development are often more<br />
important than climate change itself in<br />
influencing the scale of global impacts (Parry,<br />
2004) <strong>and</strong> how different human conditions <strong>and</strong><br />
income level affect vulnerability <strong>and</strong> resilience to<br />
climate change (Turner, 2003; Parry et al, 2004).<br />
To simulate the influence of the socio-economic<br />
characteristics of the SRES storylines on<br />
vulnerability to climate change this study uses<br />
the Global Unified Meta-model of the BiOsphere<br />
(GUMBO) (Boumans et al., 2002). GUMBO is a<br />
meta-model that incorporates a simplified<br />
version of several existing models at an<br />
intermediate level of complexity. GUMBO<br />
simulates the dynamic feedbacks among global<br />
change, human technology, economic<br />
production, welfare <strong>and</strong> ecosystem goods <strong>and</strong><br />
services within the dynamic earth system. Since<br />
GUMBO treats our world as a closed system, the<br />
IPCC driving forces are endogenous variables,<br />
derived dynamically from model characteristics.<br />
Input parameters of GUMBO are changing<br />
socio-economic conditions including consumer<br />
preferences, investment strategies, natural<br />
resources management <strong>and</strong> technological<br />
development.<br />
This study has four main components. First the<br />
SRES scenarios were simulated in GUMBO.<br />
Secondly, alternative interpretations of the<br />
storylines were modelled. Thirdly, two climate<br />
stresses were simulated. Finally the vulnerability<br />
was assessed of the (alternative interpretations<br />
of) SRES storylines to the climate stresses. For<br />
the purpose of the iEMSs 2004 Conference this<br />
article focuses on the modelling aspects of the<br />
study. A more detailed discussion of assessment<br />
will be published separately. The analysis is<br />
limited to two of the four SRES scenarios.<br />
It proved possible to reproduce the SRES driving<br />
forces population growth, economic growth <strong>and</strong><br />
energy use with their corresponding climate<br />
scenarios (temperature change, sea level rise <strong>and</strong><br />
rainfall patterns) in GUMBO. Model parameters<br />
could be chosen to agree essentially with the<br />
different pathways of socio-economic development,<br />
investment strategies <strong>and</strong> technological<br />
development of the SRES storylines.<br />
Alternative pathways of development could be<br />
defined within one SRES storylines that yield the<br />
same SRES driving forces but that differ<br />
significantly in their vulnerability to sea level<br />
rise <strong>and</strong> water availability. This study shows<br />
dynamic combination of environmental <strong>and</strong><br />
social conditions exist that significantly enhance<br />
or reduce vulnerability. Results suggests that,<br />
taking into account the characteristics of the<br />
storylines, an assessment of the relative<br />
vulnerability of the SRES scenarios can<br />
challenge earlier assessments based on climate<br />
change <strong>and</strong> the driving forces only. The<br />
assessment of alternative multidimensional<br />
socio-economic conditions is an important<br />
addition to underst<strong>and</strong> our world’s vulnerability<br />
to climate change.<br />
2 BACKGROUND<br />
2.1 The IPCC scenarios<br />
To translate findings of climate change science<br />
into international politics the IPCC uses<br />
scenarios. These are published in the Special<br />
Report on Emission Scenarios (SRES) (IPCC,<br />
2000). The SRES scenarios build on previous<br />
scenarios published by the IPCC (IPCC, 1992;<br />
IPCC, 1995), but do not include any policies or<br />
intervention; particularly there is no business-asusual<br />
scenario. The IPCC scenarios of climate<br />
change are built in four discrete steps (see also<br />
Figure 1a):<br />
1. Scenario panels created four qualitative<br />
SRES storylines that represent a diverse<br />
range of different development pathways for<br />
the world<br />
2. The storylines are translated into<br />
quantitative SRES driving forces that are<br />
harmonised projections of the indicators<br />
population growth, economic development,<br />
technology, energy <strong>and</strong> l<strong>and</strong>-use<br />
3. Anthropogenic greenhouse gas emission<br />
scenarios are estimated from the driving<br />
forces<br />
4. These emissions are used to drive climate<br />
models (General Circulation Models<br />
(GCMs)) to produce spatially explicit<br />
climate scenarios, including temperature<br />
change, sea level rise <strong>and</strong> precipitation.<br />
The calculations do not include feed backs from<br />
the climate scenarios onto the SRES storylines or<br />
driving forces. Thus effects of climate change<br />
<strong>and</strong> climate variability on e.g. water resources,<br />
the economic system or ecosystem services<br />
remain largely unresolved.<br />
The two storylines used in this study are (IPCC,<br />
2000, 4-5; Mieg, 2002) (see also Table 1):<br />
• A2: This storyline describes a very heterogeneous<br />
world. The underlying theme is<br />
self-reliance <strong>and</strong> preservation of local<br />
identities. Fertility patterns across regions<br />
converge very slowly.<br />
• B1: This storyline describes a convergent<br />
world with rapid changes in economic<br />
784
structures towards a service <strong>and</strong> information<br />
economy, with reduction in material<br />
intensity, <strong>and</strong> the introduction of clean <strong>and</strong><br />
resource-efficient technologies.<br />
2.2 GUMBO<br />
BOX: Model characteristics GUMBO:<br />
• Based on principles of system thinking<br />
(integration, feedbacks, strong sustainability)<br />
• No spatial resolution; accounts for carbon,<br />
nutrient, water fluxes across 11 l<strong>and</strong> covers <strong>and</strong><br />
4 capital stocks (natural, social, human, built<br />
capital). Apart from incoming solar energy all<br />
variables are endogenous.<br />
• Draws concepts <strong>and</strong> data from many<br />
disciplines (Global Climate models,<br />
Atmospheric models, Sociology models,<br />
Economic models, Ecosystem models).<br />
• Almost 1000 variables <strong>and</strong> 2000 parameters<br />
• Programmed in Stella environment (run time<br />
under 30 sec for 200 years on average PC)<br />
• Free available from (www.uvm.edu/giee/GUMBO)<br />
• User can edit all model equations & parameters<br />
GUMBO simulates the integrated earth systems<br />
<strong>and</strong> assesses the dynamics <strong>and</strong> values of<br />
ecosystem services. GUMBO is a meta-model in<br />
that it incorporates the simplified versions of<br />
several existing models at an intermediate level<br />
of complexity (Boumans et al., 2002). GUMBO<br />
is built upon the principles of system thinking<br />
(Meadows et al, 1992; Simonovic, 2002).<br />
GUMBO simulates the dynamics of carbon,<br />
nutrients <strong>and</strong> water within the Atmosphere,<br />
Lithosphere, Hydrosphere, <strong>and</strong> Biosphere<br />
sectors, <strong>and</strong> across eleven l<strong>and</strong> cover types<br />
covering the surface of the planet. GUMBO uses<br />
a one-year time step period, <strong>and</strong> is calibrated for<br />
the period of 1900-2000 for key variables for<br />
which quantitative time-series were available.<br />
In the model, atmospheric processes attenuate<br />
solar radiation energy arriving at the Earth<br />
surface. The atmospheric exchange of carbon<br />
<strong>and</strong> nitrogen with terrestrial systems is regulated<br />
by vegetation growth, decay <strong>and</strong> burning on<br />
terrestrial systems. Producers, consumers <strong>and</strong><br />
decomposers control these processes on ground,<br />
soil <strong>and</strong> water in the different l<strong>and</strong> cover types.<br />
These conditions <strong>and</strong> processes result in the<br />
provision of goods <strong>and</strong> services, which are<br />
referred in the model as natural capital. Hum<strong>and</strong>riven<br />
l<strong>and</strong> cover changes have an effect on the<br />
provision <strong>and</strong> availability of natural capital,<br />
which in turn, is an important determinant of<br />
humans’ economy <strong>and</strong> social welfare. The<br />
dynamics of social interactions, the human<br />
economy <strong>and</strong> welfare are modelled within the<br />
anthroposphere sector of the model. In contrast<br />
to the larger biosphere, only a very small portion<br />
of materials is internally recycled within the<br />
Anthroposphere. Human population, knowledge,<br />
social institutions <strong>and</strong> investment rates drive the<br />
material <strong>and</strong> energy flux.<br />
The atmosphere <strong>and</strong> anthroposphere are<br />
considered to be globally homogenous. The<br />
homogeneous nature of the atmosphere is<br />
justified by the fast exchanges in air masses<br />
between l<strong>and</strong> covers. The homogeneous<br />
character of the anthroposphere, in turn, reflects<br />
the global economy where wealth <strong>and</strong> quality of<br />
life are not registered relative to l<strong>and</strong> cover type.<br />
The other sectors (lithosphere, hydrosphere, <strong>and</strong><br />
biosphere) are divided into 11 l<strong>and</strong> cover types<br />
<strong>and</strong> the structure described is replicated for each<br />
l<strong>and</strong> cover. In addition, there are sectors in the<br />
model for ecosystem services, l<strong>and</strong> use, <strong>and</strong> the<br />
model’s database.<br />
GUMBO is the first global model to explicitly<br />
account for ecosystem goods <strong>and</strong> services <strong>and</strong><br />
factor them directly into the process of global<br />
economic production <strong>and</strong> human welfare<br />
development (Boumans et al., 2002). In<br />
GUMBO, the flow of ecosystem goods <strong>and</strong><br />
services are explicitly combined with<br />
manufactured <strong>and</strong> human capital to produce<br />
human welfare (Costanza et al., 1997a). Such<br />
design is based on the strong sustainability<br />
concept, that is, on the concept that natural<br />
capital is essential for the creation <strong>and</strong><br />
maintenance of the human, physical <strong>and</strong> social<br />
capitals aspects of the anthroposphere.<br />
3 METHOD<br />
There are four main components of the research.<br />
First the SRES scenarios were simulated in<br />
GUMBO. Secondly alternative interpretations of<br />
the storylines were modelled. Thirdly climate<br />
stresses were simulated. Finally the vulnerability<br />
was analysed of the (alternative interpretations<br />
of) SRES storylines to the climate stresses.<br />
1. Simulation of the SRES scenarios in<br />
GUMBO, including the driving forces <strong>and</strong><br />
the associated climate change scenarios<br />
(temperature, sea level rise, precipitation)<br />
This study uses GUMBO to reproduce the SRES<br />
scenarios together with their climate scenarios in<br />
one modelling framework including feedbacks.<br />
This method differs from the IPCC simulations,<br />
that do not include feedbacks from the climate<br />
scenarios onto the SRES storylines or driving<br />
forces (Figure 1b). The socio-economic<br />
conditions described in the SRES storylines were<br />
used as an input to GUMBO (Table 1) <strong>and</strong><br />
introduced by a change in model parameters after<br />
the year 2004. An interpretation of the storylines<br />
was selected that reproduces the driving forces of<br />
785
the IPCC marker scenarios <strong>and</strong> the<br />
corresponding climate change scenarios. This<br />
article concentrates on the A2 & B1 scenario. A2<br />
was chosen because the IPCC Task Group on<br />
Scenarios for Climate Impact Assessment has<br />
asked climate-modelling centres to give priority<br />
to the A2 (<strong>and</strong> B2) scenarios. B1 was chosen to<br />
complement the socio-economic <strong>and</strong> climatic<br />
conditions of A2.<br />
Scenario panels<br />
Storylines<br />
• Socioeconomic<br />
development<br />
<strong>Modelling</strong> groups<br />
Driving forces<br />
• population<br />
growth<br />
• economic<br />
development<br />
• technology<br />
• energy<br />
• l<strong>and</strong>-use<br />
Integrated<br />
assessm. models<br />
Emission<br />
scenarios<br />
• greenhouse<br />
gas emissions<br />
Climate models<br />
Climate<br />
scenarios<br />
• temperature<br />
change<br />
• sea level rise<br />
• precipitation<br />
Vulnerability <strong>and</strong> impact assessments<br />
a: scenarios development <strong>and</strong> assessment in IPCC process<br />
SRES storylines<br />
Model parameter<br />
• Investment<br />
strategies<br />
• Technology<br />
development<br />
• Resources<br />
management<br />
• Labor particip.<br />
• Health&Educat.<br />
Model variables<br />
• population<br />
• gross world prod<br />
• ecosystem<br />
goods&services<br />
• knowledge<br />
• energy<br />
• l<strong>and</strong>-use &cover<br />
Emission<br />
scenarios<br />
• CO 2 emissions<br />
Model equations<br />
feedbacks <strong>and</strong> impacts<br />
Climate<br />
scenarios<br />
• temperature<br />
change<br />
• sea level rise<br />
• precipitation<br />
b: scenario representation <strong>and</strong> feedbacks in GUMBO<br />
Figure 1a&b: Representation of the IPCC scenarios in GUMBO<br />
To represent the IPCC climate change scenarios<br />
GUMBO was modified <strong>and</strong> recalibrated with<br />
recent insights from global change research.<br />
These modifications include recalibration of the<br />
carbon cycle using global estimates of<br />
atmosphere-ocean interaction <strong>and</strong> l<strong>and</strong>atmosphere<br />
interaction (IPCC, 2001).<br />
Characteristic carbon limitation factors were<br />
estimated for each l<strong>and</strong> cover in GUMBO<br />
(CSCDGC, 2002). Since the potential<br />
interactions between CO2, nutrients, water,<br />
weeds, pest insects <strong>and</strong> other stresses are largely<br />
unknown (Parry, et al, 2004) the limitation<br />
factors were calibrated against literature values<br />
of net biome production (Levy, 2004) <strong>and</strong> net<br />
primary production (e.g. Portela, 2004; Malhi,<br />
2002). The water cycle was recalibrated using<br />
Cosgrove <strong>and</strong> Rijsberman (2002), with special<br />
attention to precipitation per GUMBO l<strong>and</strong> cover<br />
type. For this recalibration climate scenarios of<br />
temperature <strong>and</strong> precipitation per GUMBO l<strong>and</strong><br />
cover type were estimated by superimposing a<br />
mask with the 11 GUMBO l<strong>and</strong> covers types that<br />
was derived from a global l<strong>and</strong> cover data set<br />
(DeFries et al 1994a) onto the downscaled model<br />
output of two General Circulation Model (GCM)<br />
(the Hadley Climate Model 3 (HadCM3) <strong>and</strong> the<br />
European Climate Model 4 with the OPYC3<br />
ocean circulation model (ECHAM4/OPYC3))<br />
available from the IPCC Data Distribution<br />
Centre (IPCC-DDC, http://ipcc-ddc.cru.uea.ac.uk/).<br />
Finally the ecosystem service Climate<br />
Regulation was estimated from the net sink for<br />
carbon that the terrestrial ecosystem represents.<br />
2. <strong>Modelling</strong> of alternative interpretations of<br />
storylines that yield the same driving forces.<br />
Two alternatives were simulated for each<br />
scenario. The leading input variable to mark the<br />
difference between the alternatives is agricultural<br />
production. Agricultural production was selected<br />
because recent impact assessments point at<br />
increased stress from climate change (e.g. Aerts,<br />
2003; Parry et al, 2004). To yield the same SRES<br />
driving forces, the shift in agricultural production<br />
was balanced by changing other input parameters<br />
in line with the SRES storylines. Since the share<br />
of alternative energy sources is specified in the<br />
SRES scenarios, this was not used to construct<br />
alternative interpretations of a storyline.<br />
3. Simulation of two stresses from the climate<br />
system<br />
Two climate stresses were selected to target the<br />
economic system <strong>and</strong> food / biome production<br />
respectively: (i) increasing the depreciation value<br />
of built capital with sea level rise <strong>and</strong> (ii)<br />
decreasing crop production with drought stress.<br />
The study aims to assess the relative<br />
vulnerability to these stresses <strong>and</strong> not the<br />
absolute vulnerability. The absolute strength of a<br />
climate stress is therefore less critical in the<br />
simulation. The impact of sea level rise on built<br />
capital was simulated by increasing the<br />
depreciation value of built capital proportional to<br />
sea level rise above a certain limit. This limit was<br />
786
0. 0 0<br />
selected 20 [cm]. The impact of possible drought<br />
conditions was simulated by decreasing the<br />
drought tolerance of crops. In GUMBO this was<br />
realised by changing the groundwater limit<br />
below which crops production starts to decrease.<br />
Assuming no drought stress in 1990, the impacts<br />
of two different groundwater limits were<br />
assessed that were set at 5/3 <strong>and</strong> twice the 1990<br />
groundwater level respectively. It is noted that<br />
drought stress in GUMBO is the combined result<br />
of water supply <strong>and</strong> water use.<br />
4. Assessment of the relative vulnerability of<br />
(alternative interpretations of) SRES<br />
storylines to the climate stresses.<br />
For each scenario two model runs are compared:<br />
one with <strong>and</strong> one without a particular climate<br />
stress. The relative vulnerability of the different<br />
SRES scenarios is assessed, focussing on a<br />
number of key GUMBO variables, including<br />
population, economic growth <strong>and</strong> ecosystem<br />
services.<br />
Storyline A2 B1<br />
Elements of Storylines defining input parameters GUMBO<br />
Pace & direction Slow <strong>and</strong> High, towards<br />
of investment & heterogeneous; efficient resource<br />
technological focus on agricultural use; clean technology;<br />
change production<br />
recycling<br />
<strong>Environmental</strong><br />
concern<br />
Local; Directed<br />
towards easing soil<br />
erosion & water<br />
pollution for agric.<br />
High & global,<br />
including taxation,<br />
regulation <strong>and</strong><br />
reuse<br />
Fertility rates Slowly declining Declining<br />
Education <strong>and</strong><br />
Health programs<br />
- High towards<br />
clean & equitable<br />
development<br />
Dietary patterns -<br />
Much lower meat<br />
consume due to<br />
high food prices<br />
Income gap &<br />
Productivity<br />
Disparity<br />
Maintained or<br />
increasing<br />
Social structures Diversifying<br />
Declining.<br />
Productivity<br />
increases<br />
High social<br />
consciousness<br />
Elements of Storylines reproduced in GUMBO variables<br />
Energy intensity Declining 0.5-0.7 % Declining<br />
of GDP per year<br />
significantly<br />
Capital stock<br />
turn over<br />
Slow<br />
- (focus on quality<br />
& services)<br />
Resource Low; emphasis on -<br />
availability self-reliance<br />
Equity Decreasing Increasing<br />
Global<br />
interaction<br />
Low; cultural<br />
pluralism &<br />
protectionism<br />
High<br />
Driving forces to be reproduced in GUMBO<br />
Population no High; ~ 15 billion Low; ~ 7.2 billion<br />
GDP growth 1)<br />
Medium (<strong>and</strong><br />
differentiated); 243<br />
Medium - High;<br />
328<br />
Storyline A2 B1<br />
GDP per capita Ind.:US$46,200; US$ 46,598<br />
2)<br />
Dev.:$11,000<br />
Energy use High; fossil fuel use Low; fossil fuel<br />
29 [GtC] use 5 [GtC]<br />
Favoured energy Mixed<br />
Alternative energy<br />
L<strong>and</strong> use change Medium-high High<br />
Climate change scenarios to be reproduced as variables<br />
in GUMBO<br />
CO 2<br />
High; 850 parts per Low; 547 ppm<br />
concentrations million (ppm)<br />
Global av.<br />
temperatures<br />
3.8 degrees relative<br />
to 1990<br />
2.0 degrees rel. to<br />
1990<br />
Sea level rise 42 [cm] 31 [cm]<br />
Precipitation High, diversifying Low<br />
1) World GDP (trillion 1990US$) in 2100<br />
2) GDP per capita in 1990US$,market exchange prices<br />
Table 1: SRES scenario qualification (IPCC,<br />
2000) <strong>and</strong> climate change impacts used in this<br />
study; numbers are for 2100<br />
4 RESULTS<br />
Figure 2 illustrates the input variables found to<br />
capture the SRES Story lines <strong>and</strong> reproduce the<br />
SRES scenario A2 <strong>and</strong> B1 in GUMBO.<br />
Variables are shown relative to their maximum<br />
value in the simulations reported in this article.<br />
A2<br />
B1<br />
Waste<br />
Carrying Capacity<br />
Fertility Decrease<br />
with Education<br />
Raw Material needs<br />
Healthcare Development<br />
Agricultural Production<br />
Social Network<br />
Development<br />
Extractable Fossil<br />
Fuel Development<br />
Labour Participation<br />
Figure 2: Relative size of GUMBO Input<br />
Variables in 2100.<br />
Table 2 compares the values of the SRES driving<br />
forces <strong>and</strong> the corresponding climate scenarios<br />
that are reported by the IPCC with those<br />
simulated in GUMBO. The following<br />
observations are made when representing the<br />
SRES driving forces <strong>and</strong> the IPCC climate<br />
scenarios in GUMBO.<br />
Population growth<br />
Two challenges had to be overcome to match the<br />
population growth of the SRES scenarios:<br />
• to decrease population growth in the near<br />
future given substantial economic growth<br />
• to sustain the world’s population at the end<br />
of the century under increasing pressure<br />
from waste, resource shortage <strong>and</strong> stress on<br />
food production.<br />
787
Essential elements of the storylines are: the<br />
effect of education on fertility <strong>and</strong> raising the<br />
waste carrying capacity <strong>and</strong> waste assimilation<br />
capacity, as reported in B1. In addition the effect<br />
of GWP growth <strong>and</strong> food/capita growth on<br />
mortality had to be decreased. This signifies that<br />
GNP growth does only marginally benefit the<br />
poorest <strong>and</strong> inequality will increase as reported<br />
in the A2 storyline.<br />
GNP growth<br />
GNP growth is matched by growing labour<br />
participation <strong>and</strong> labour efficiency with<br />
increasing income <strong>and</strong> technology. The can be<br />
understood from increased appreciation of<br />
services <strong>and</strong> internalisation of the informal<br />
economy, as indicated in storyline B1. Labour<br />
participation was raised strongly in B1 to<br />
simulate high economic growth under falling<br />
population number <strong>and</strong> fossil fuel use. To match<br />
economic growth it had to be assumed that<br />
improved efficiency of resource use stimulates<br />
consumption, rather than decreases (income<br />
from) raw material use. This may be at odds with<br />
the shift from quantity to quality, reported in the<br />
B1 storyline. Finding satisfying assumptions to<br />
sustain economic growth as projected in the<br />
SRES scenarios proved a major challenge in the<br />
GUMBO simulation.<br />
Energy use<br />
For B1 the amount of available fossil fuel had to<br />
be increase by 30% <strong>and</strong> for A2 by over 400%.<br />
Substantial controls had to be installed, to realise<br />
that additional available oil is not consumed<br />
immediately, but gradually over time. This is<br />
realised in GUMBO by decreasing the rate at<br />
which new oil is found when human capital in<br />
the form of knowledge is invested in fossil fuel<br />
exploitation. The share of alternative energy<br />
sources was raised. To match the share of<br />
renewables of the SRES scenarios, it was<br />
assumed that new technology that is developed is<br />
directly used which may not be the case in all<br />
scenarios.<br />
L<strong>and</strong> use changes<br />
Preserving biome productivity is essential to<br />
realise the large terrestrial sink of anthropogenic<br />
carbon, estimated by the GCMs. It is governed<br />
by production limits (of water, carbon, nutrients,<br />
light <strong>and</strong> waste). In GUMBO climate change<br />
affects the production limits within a l<strong>and</strong> cover<br />
type. Presently it does not directly influence l<strong>and</strong><br />
cover change, <strong>and</strong> the rates at which l<strong>and</strong> covers<br />
change from one to another are held constant.<br />
Once more data becomes available on the<br />
influence of climate on l<strong>and</strong> cover change, this<br />
maybe a valuable extension of GUMBO. It was<br />
decided not to reproduce the l<strong>and</strong> cover change<br />
scenarios of the IPCC since these do not exist for<br />
all scenarios <strong>and</strong> are increasing questioned<br />
(Levy, 2004, etc).<br />
CO 2 concentrations<br />
Global atmospheric carbon concentrations are<br />
well represented. Net biome production was<br />
calibrated to yield carbon uptake in line with the<br />
IPCC estimates of atmospheric carbon.<br />
Global average temperatures<br />
Global temperature change relative to 1990 is<br />
well represented. Temperature changes per l<strong>and</strong><br />
cover type are less well represented <strong>and</strong> deserve<br />
future attention.<br />
Global Mean Sea level rise<br />
Sea level rise is well represented in A2 <strong>and</strong><br />
underestimated in the B1 scenario.<br />
Precipitation<br />
The trend in overall change in precipitation is<br />
well represented. Inter yearly variations are not<br />
modelled by GUMBO. Changes in precipitation<br />
per l<strong>and</strong> cover are different from the GCMs<br />
results, especially for those l<strong>and</strong> covers that<br />
change significantly in area. As GUMBO is not<br />
spatially explicit, GUMBO assumes that when<br />
the area of a l<strong>and</strong> cover increases, the new area<br />
receives the average precipitation over l<strong>and</strong>. This<br />
corresponds to the notion that e.g. cropl<strong>and</strong> is<br />
now in the most suitable locations for<br />
agriculture, characterised by high precipitation.<br />
New cropl<strong>and</strong> would be found in areas with less<br />
favourable conditions.<br />
Population number [billions]<br />
Gross National Product (GNP)<br />
[trillion US$]<br />
A2 A2 B1 B1<br />
2050 2100 2050 2100<br />
IPCC 11.3 15.1 8.7 7.0<br />
GUMBO 11.1 15.0 8.8 7.1<br />
IPCC 81.6 242.8 135.6 328.4<br />
GUMBO 115.6 242.6 135.2 328.9<br />
GWP per capita IPCC 7221 16113 15569 46598<br />
GUMBO 10387 16152 15417 46657<br />
Ecosystem Services [trillion US$] GUMBO 19.0 23.0 16.7 15.7<br />
EcoService - Climate regulation GUMBO 9.5 16.1 6.5 6.4<br />
Global Welfare GUMBO 1.3 2.7 4.4 31.8<br />
Fossil Fuel [GtC] IPCC 16.5 28.9 11.7 5.2<br />
GUMBO 19.4 28.9 11.7 5.3<br />
Alternative Energy [EJ] IPCC 175.0 481.8 140.7 103.4<br />
GUMBO 255.1 337.2 177.4 212.1<br />
Atmospheric Carbon [ppm] IPCC 549.0 834.0 492.0 547.0<br />
GUMBO 560.2 825.1 472.9 547.0<br />
Net Ecosystem Prod. (carbon IPCC 6.1 6.8 4.0 3.5<br />
seq) [GtC] 1)<br />
GUMBO 4.7 8.7 3.6 4.4<br />
Sealevel rise relative to 1990 [m] IPCC 0.16 0.42 0.15 0.31<br />
GUMBO 0.22 0.48 0.15 0.25<br />
Temperature change relative to IPCC 1.75 4.13 1.54 2.32<br />
1990 [oC]<br />
GUMBO 2.11 4.09 1.62 2.51<br />
Total precipitation 2) Hadley 866.6 886.0 - -<br />
[mm/yr] 3) ECHAM 862.4 878.8 - -<br />
GUMBO 864.6 891.1 856.1 864.2<br />
Change in Precip. 2) Hadley 23.3 42.7 - -<br />
rel to 1990 [mm/yr] 3) ECHAM 19.8 36.2 - -<br />
GUMBO 22.1 48.7 13.6 21.7<br />
Ocean Atmosphere Exchange GUMBO -0.7 -0.9 -0.3 0.0<br />
1) Values from Levy, 2004<br />
2) results Hadley General Circulation Model<br />
3) results ECHAM4 Model General Circulation Model<br />
Table 2: Comparison of SRES <strong>and</strong> climate<br />
scenarios with GUMBO results<br />
Figure 3 illustrates the input variables that<br />
simulate two alternative interpretations of both<br />
788
0.00<br />
0.00<br />
SRES storylines A2 <strong>and</strong> B1. The Figure shows a<br />
reduction of agricultural production could be<br />
balanced by increased labour participation <strong>and</strong><br />
healthcare development. In terms of economic<br />
growth, labour was substituted for agricultural<br />
production in the alternatives. Population growth<br />
is controlled by food production rather than<br />
healthcare <strong>and</strong> education.<br />
Alternative1-A2<br />
Alternative2-A2<br />
Agricultural Production<br />
Labour Participation<br />
Alternative1-B1<br />
Alternative2-B1<br />
Agricultural Production<br />
Labour Participation<br />
Healthcare Development<br />
Healthcare Development<br />
Social Network<br />
Development<br />
Extractable Fossil<br />
Fuel Development<br />
Social Network<br />
Development<br />
Extractable Fossil<br />
Fuel Development<br />
Figure 3: Input Parameters for the two alternative<br />
interpretations of the SRES storylines<br />
Climate stresses were applied to the two<br />
alternative interpretations of the SRES scenarios<br />
A2 <strong>and</strong> B1 in GUMBO. Table 3 lists the values<br />
of characteristic GUMBO variables for the<br />
alternative interpretations relative to each other<br />
without additional climate stress (first row for<br />
each variable) <strong>and</strong> with an additional climate<br />
stress (row 2-4 for each variable). It illustrates<br />
that the vulnerability to climate stress differs<br />
between the alternatives. Alternative 2, of which<br />
the economy depends stronger on agricultural<br />
production <strong>and</strong> less on service <strong>and</strong> health care, is<br />
more vulnerable to drought stress. This is<br />
particularly true for B1, which does not ease soil<br />
<strong>and</strong> water pollution for agriculture as in A2.<br />
The A2 scenario is found less vulnerable to<br />
drought stress than the B1 scenario, although it is<br />
characterised by high population growth, fossil<br />
use <strong>and</strong> climate change, suggesting growing<br />
stress on food production. In the underlying<br />
storyline this stress is recognised <strong>and</strong> mitigated<br />
through innovation <strong>and</strong> the local management of<br />
soil erosion <strong>and</strong> water pollution. Building the<br />
scenario from its storyline, adaptations to climate<br />
change have been implemented that are not<br />
included in assessments that build on the driving<br />
forces (e.g. Aerts, 2003; Parry et al, 2004). A<br />
more detailed discussion of the assessment will<br />
be published separately.<br />
Population<br />
number<br />
Gross<br />
National<br />
Product<br />
Ecosystem<br />
Services<br />
EcoService<br />
- Nutrient<br />
Controle<br />
Alternative 2<br />
relative to Alt.1<br />
Alternative 2<br />
relative to Alt.1<br />
climate stress A2 B1 A2 B1<br />
no additional stress 1.00 0.99 Atmospheric 0.92 0.97<br />
sea level 1.01 1.06 Carbon 1.00 1.00<br />
sea level+drought 0.91 0.79 1.02 1.02<br />
sea level++drought 0.80 0.69 1.03 1.02<br />
no additional stress 1.01 1.00 Plant growth 1.23 1.08<br />
sea level 1.03 1.03 (Terrestrial 1.00 1.02<br />
sea level+drought 0.97 0.85 GPP) 0.99 0.91<br />
sea level++drought 0.91 0.77 0.95 0.90<br />
no additional stress 1.02 1.18 Temp. 0.99 1.00<br />
sea level 1.00 1.00 change rel. 1.00 1.00<br />
sea level+drought 1.00 0.98 to 1990 1.00 1.00<br />
sea level++drought 0.99 0.98 1.00 1.00<br />
no additional stress 1.76 1.35 Sealevel rise 0.91 0.95<br />
sea level 1.00 1.08 relative to 1.00 0.99<br />
sea level+drought 0.96 0.74 1990 1.02 1.02<br />
sea level++drought 0.87 0.71 1.03 1.02<br />
Legend:<br />
First row for each variable: value of GUMBO variable in<br />
Alternative 2 divided by its value in Alternative 1 for the year<br />
2100; 2 nd to 4 th row for each variable: effect of climate stress<br />
on Alternative 2 divided by effect of climate stress on<br />
Alternative 1, where the effect of the climate stress is<br />
estimated: value of GUMBO variable with the climate stress<br />
divided by its value without the stress (year 2100); sea level:<br />
increasing the depreciation value of built capital with sea<br />
level rise; +drought: decreasing the drought tolerance of crop<br />
production (++: stronger decrease); Numbers in italics:<br />
driving force that is kept equal for the two alternatives.<br />
Table 3: Value of selected GUMBO variables<br />
for alternative interpretations of the SRES<br />
storylines A2 <strong>and</strong> B1 under climate stress<br />
5 DISCUSSION<br />
It proved possible to reproduce the SRES driving<br />
forces population growth, economic growth,<br />
energy use together with their corresponding<br />
climate scenarios (temperature change, sea level<br />
rise <strong>and</strong> rainfall patterns) in GUMBO. Model<br />
input parameters could be chosen to agree with<br />
the different pathways of socio-economic<br />
development, investment strategies <strong>and</strong><br />
technological development of the SRES<br />
storylines. Exceptions are the absolute amount of<br />
accessible fossil fuel that had to be differentiated<br />
between scenarios to meet the scenario specific<br />
fossil fuel use. Improved efficiency of resource<br />
use was assumed to stimulate consumption to<br />
match economic growth. This may be in<br />
contradiction with elements of the storyline that<br />
indicate a shift to quality goods.<br />
Critical relationships that had to be estimated to<br />
harmonise the scenarios in GUMBO with the<br />
SRES scenarios include (i) the effect of carbon,<br />
water, nutrient <strong>and</strong> other limiting factors on net<br />
biome production to yield estimates of the global<br />
carbon sink, (ii) the impact of investment in<br />
knowledge on population growth, technological<br />
change, efficiency of resource use <strong>and</strong> energy<br />
production, (iii) the relation of income <strong>and</strong><br />
labour participation <strong>and</strong> productivity.<br />
Alternative pathways of development can be<br />
defined within one SRES storylines that yield the<br />
same SRES driving forces but that differ<br />
significantly in their vulnerability to sea level<br />
789
ise <strong>and</strong> water availability. This study shows<br />
dynamic combination of environmental <strong>and</strong><br />
social conditions exist that significantly enhance<br />
or reduce vulnerability. It suggests that, taking<br />
into account the characteristics of the storylines,<br />
an assessment of the relative vulnerability of the<br />
SRES scenarios can challenge earlier<br />
assessments based on climate change <strong>and</strong> the<br />
driving forces only. The assessment of<br />
alternative multidimensional socio-economic<br />
conditions is an important addition to underst<strong>and</strong><br />
our world’s vulnerability to climate change. It is<br />
recommended to build on this “inverse”<br />
approach of vulnerability analysis to assess multi<br />
dimensional causes of critical outcomes. It could<br />
extend the merits of vulnerability assessments<br />
that investigate the impacts of multiple scenarios<br />
of one particular global environmental stress.<br />
GUMBO offers a promising, flexible <strong>and</strong> fast<br />
environment for this assessment.<br />
6 ACKNOWLEDGEMENT<br />
The authors would like to thank Prof. Bob<br />
Costanza of the GUND Institute, University of<br />
Vermont <strong>and</strong> Prof. Ekko van Ierl<strong>and</strong>, Prof. Pavel<br />
Kabat <strong>and</strong> Prof. Rik Leemans of Wageningen<br />
University <strong>and</strong> Research Centre for their<br />
comments <strong>and</strong> support throughout this study.<br />
7 REFERENCES<br />
Aerts, J.C.J.H. <strong>and</strong> Droogers P. (Eds) (2003) Climate<br />
Change in Contrasting River Basins. Cabi<br />
Publishers (accepted / in prep.)<br />
Alcamo, J., M. Märker, M. Flörke <strong>and</strong> S. Vassolo<br />
(2002) Preliminary Findings from Assessing Global<br />
Climate Impacts on Water Resources <strong>and</strong> Hot Spot<br />
Regions, First Interim Report of the Project on<br />
Global Analysis <strong>and</strong> Global Information for the<br />
Dialogue on Water <strong>and</strong> Climate, University of<br />
Kassel, Germany<br />
Arnell N.W., M.J.L. Livermore, S. Kovats, P.E. Levy,<br />
R. Nicholls, M.L. Parry, S.R. Gaffin (2004) Climate<br />
<strong>and</strong> socio-economic scenarios for global-scale<br />
climate change impacts assessments: the SRES<br />
storylines, Global <strong>Environmental</strong> Change 14:3<br />
Boumans, R, Costanza R., Farley J., Wilson M.A.,<br />
Portela R., Rotmans J., Villa F., Grasso M. (2002)<br />
Modeling the dynamics of the integrated earth<br />
system <strong>and</strong> the value of global ecosystem services<br />
using the GUMBO model, Ecological Economics<br />
41 529–560<br />
Cosgrove, W. <strong>and</strong> F. Rijsberman (2002) World Water<br />
Vision: Making Water Everybody's Business,<br />
Earthscan, London.<br />
Costanza, R., R. Arge, et al. (1997). "The value of the<br />
world's ecosystems services <strong>and</strong> natural capital."<br />
Nature 387: 253-269<br />
CSCDGC (2002) Plant growth data. Center for the<br />
Study of Carbon Dioxide <strong>and</strong> Global Change in<br />
Tempe, Arizona. http://www.co2science.org<br />
DeFries, R. S. <strong>and</strong> J. R. G. Townshend (1994a) NDVIderived<br />
l<strong>and</strong> cover classification at global scales.<br />
<strong>International</strong> Journal of Remote Sensing, 15:3567-<br />
3586. Special Issue on Global Data Sets<br />
IPCC 1992: Houghton, J. T., Call<strong>and</strong>er, B. A., &<br />
Varney, S. K. (eds.). (1992). Climate Change 1992:<br />
The supplementary report to the IPCC scientific<br />
assessment. Cambridge University Press. UK.<br />
IPCC 1995: Houghton, J. T., Meira Filho, L. G.,<br />
Call<strong>and</strong>er, B. A., Harris, N., Kattenberg, A., &<br />
Maskell, K. (eds.). (1996). Climate Change 1995:<br />
The science of climate change. Cambridge, UK:<br />
Cambridge University Press, UK.<br />
IPCC 2000: Nakicenovic, N. et al. (eds.). (2000).<br />
Special report on emission scenarios. Cambridge,<br />
UK: Cambridge University Press.<br />
IPCC (2001). Climate Change 2001 – The Scientific<br />
Basis. Contribution of Working Group I to the Third<br />
Assessment Report of the Intergovernmental Panel<br />
on Climate Change. Houghton, J.T. et al. (eds.).<br />
Cambridge University Press, UK, 881pp.<br />
Kabat, P, <strong>and</strong> H. van Schaik (co-ordinating lead<br />
authors) (2003) Climate changes the water rules:<br />
How water managers can cope with today’s climate<br />
variability <strong>and</strong> tomorrow’s climate change.<br />
Synthesis Report of the <strong>International</strong> Dialogue on<br />
Water <strong>and</strong> Climate, ISBN 9032703218<br />
Levy, P.E., M.G.R. Cannell <strong>and</strong> A.D. Friend (2004)<br />
<strong>Modelling</strong> the impact of future changes in<br />
climate,CO2 concentration <strong>and</strong> l<strong>and</strong> use on natural<br />
ecosystems <strong>and</strong> the terrestrial carbon sink, Global<br />
<strong>Environmental</strong> Change 14:21<br />
Malhi, Y., P Meir, S. Brown (2002) Forest, carbon <strong>and</strong><br />
global climate. Phil. Trans. R. Soc. Lond. A 360:<br />
1567-1591<br />
Meadows, D. H., D. L. Meadows, <strong>and</strong> J. R<strong>and</strong>ers.<br />
1992. Beyond the limits: confronting global<br />
collapse, envisioning a sustainable future. Chelsea<br />
Green, Post Mills, Vermont<br />
Parry, M.L. (2002) Scenarios for climate impact <strong>and</strong><br />
adaptation assessment. Global <strong>Environmental</strong><br />
Change 12(3), 149-153pp.<br />
Parry, M.L (2004) Global Impacts of climate change<br />
under SRES scenarios, Editorial Special Issue<br />
Global <strong>Environmental</strong> Change, 14:1<br />
Parry, M.L., Rosenzweig C, Iglesias A, Livermore M,<br />
Fischer G (2004) Effects of climate change on<br />
global food production under SRES emissions <strong>and</strong><br />
socio-economic scenarios. Global <strong>Environmental</strong><br />
Change, 14:53-67<br />
Portela, R.G. (2004) Ecosystem services from Brazil’s<br />
Amazon Rainforest: <strong>Modelling</strong> patterns of l<strong>and</strong>-use<br />
change <strong>and</strong> incentives to more sustainable uses;<br />
University of Maryl<strong>and</strong>, USA<br />
SDU (2004) Systems thinking-a mental model for<br />
sustainable futures, Sustainable Development<br />
Update (SDU), Issue 1, <strong>Volume</strong> 4.<br />
www.albaeco.com/sdu<br />
Simonovic, S.P., (2002) "World Water Dynamics:<br />
Global Modeling of Water Resources", Journal of<br />
<strong>Environmental</strong> Management, 66(3):249-267<br />
Steffen, Will (2004) Global Change <strong>and</strong> the Earth<br />
System – a planet under pressure<br />
Turner II, B.L., Kasperson R.E., Matsone P.A.<br />
McCarthy J.J., Corell R.W., Christensene L. et al<br />
(2003) A framework for vulnerability analysis in<br />
sustainability science. PNAS, vol. 100 no. 14<br />
790
Principles of Human-Environment Systems (HES) Research<br />
Rol<strong>and</strong> W. Scholz <strong>and</strong> Claudia R. Binder<br />
Natural <strong>and</strong> Social Science Interface, Institute for Human Environment Systems. Department of<br />
<strong>Environmental</strong> Sciences, Swiss Federal Institute of Technology Zürich (ETH), Switzerl<strong>and</strong><br />
Abstract: This paper presents the basic principles, applications, <strong>and</strong> a methodological discussion of the<br />
approach of Human-Environment Systems (HES). In general, HES includes all environmental <strong>and</strong><br />
technological systems that are relevant for or affected by humans. The basic principles of the HES approach<br />
are: (1) human <strong>and</strong> environmental systems are constructed as complementary systems, (2) a hierarchy of<br />
human systems with related environmental systems are considered, (3) environmental systems are modeled<br />
in their immediate <strong>and</strong> delayed dynamic reactions to human action, (4) the behavior of the human system is<br />
modeled from a decision theoretic perspective differentiating between goal formation, strategy formation,<br />
strategy selection <strong>and</strong> action, (5) a conceptualization of different types of environmental awareness in each of<br />
these three steps can be developed, <strong>and</strong> finally (6) a distinction is made, with corresponding modeling<br />
reflecting this distinction, between primary <strong>and</strong> secondary feedback loops with respect to human action. We<br />
illustrate the principles with an example from bio-waste management. It is shown how the humanenvironment<br />
interaction can be analyzed.<br />
Keywords: Human-environment systems, regulatory mechanisms, feedback mechanisms, interfering<br />
regulatory mechanisms, bio-waste management.<br />
1. INTRODUCTION<br />
The investigation of complex environmental<br />
systems that are affected by human action is<br />
considered a major scientific challenge. This<br />
challenge has to overcome both the gap between<br />
natural <strong>and</strong> social sciences <strong>and</strong> to master modeling<br />
on different scales. Thus, there is a need for<br />
integrating knowledge from natural <strong>and</strong> social<br />
sciences. Human-environment systems (HES) are<br />
defined as the interaction of human systems with<br />
corresponding environmental or technological<br />
systems.<br />
We provide a process <strong>and</strong> structure model, which<br />
is derived from integrative modeling, system<br />
theory, basic cybernetic feedback loop modeling,<br />
cognitive sciences, <strong>and</strong> decision research (Ashby,<br />
1957; Simon, 1957; Scholz, 1987). This process<br />
<strong>and</strong> structure model should allow for investigating<br />
regulatory, feedback, <strong>and</strong> control mechanisms<br />
(RFC-mechanisms) in HES. Our motivation is to<br />
conceptualize environmentally sensitive regulatory,<br />
feedback, <strong>and</strong> control systems of the<br />
anthroposphere. Our goal is to underst<strong>and</strong> the<br />
evaluation, transformation, <strong>and</strong> regulation<br />
processes of human systems with respect to<br />
environmental <strong>and</strong> resource systems. This is done<br />
when considering a multi-hierarchy level of human<br />
systems. The HES approach is considered a<br />
framework for the underst<strong>and</strong>ing of the mechanisms<br />
underlying environmentally sensitive action,<br />
for reflecting on interferences among different<br />
levels (i.e., the individual <strong>and</strong> societal level), <strong>and</strong><br />
for reflecting on different feedback loops or RFC<br />
mechanisms that may lead to sustainable action.<br />
1.1 Relationships between human <strong>and</strong><br />
environmental systems in environmental<br />
sciences<br />
The HES approach conceptualizes a mutualism<br />
between human <strong>and</strong> environmental systems. The<br />
human <strong>and</strong> the environmental system are conceived<br />
as two different systems that exist in essential<br />
dependencies <strong>and</strong> reciprocal endorsement. The term<br />
human systems, meaning social systems ranging<br />
from society to individuals (Apostle, 1952), has<br />
been used since the time of the ancient Greeks.<br />
These systems are supposed to have a memory,<br />
language, foresight, consciousness etc. In contrast<br />
to the concept of human or social systems, the<br />
term environmental systems arose late in the early<br />
19 th century (Simpson & Weiner, 1989 p. 315),<br />
even though Hippocrates had already dealt with<br />
environmental impacts on human health in early<br />
medicine in 420 BC.<br />
In the history of environmental sciences at large,<br />
the relationship between human (H ) <strong>and</strong><br />
environmental (E) systems was dealt with from<br />
different perspectives. The H impact chain was<br />
initially examined from the human perspective. In<br />
the early 18 th century, forest engineers investigated<br />
791
how legal or economic restrictions affect the<br />
texture of forests agricultural, forest. Resource<br />
economics evolved in the early 18 th<br />
century <strong>and</strong><br />
focused on the question of how agricultural <strong>and</strong><br />
forest yields can be sustainably (von Carlowitz,<br />
1732) or most efficiently obtained (Goodwin,<br />
1977). From the environmental research<br />
perspective, the H impact chain has quite a<br />
different focus, namely how human activities affect<br />
the environment or environmental equilibrium <strong>and</strong><br />
how these impacts can be mitigated (Wood, 1995,<br />
Freedmann, 1995).<br />
There are different ways for the relationship of<br />
what we denote as Human <strong>and</strong> <strong>Environmental</strong><br />
Systems to be conceptualized.<br />
The GAIA-approach (Lovelock, 1979) “views the<br />
earth as a single organism, in which the individual<br />
elements coexist in a symbiotic relationship.<br />
Internal homeostatic control mechanisms,<br />
involving positive <strong>and</strong> negative feedbacks,<br />
maintain an appropriate level of stability.” (Kemp,<br />
1998, p.160) GAIA is an example of an<br />
integrative, qualitative approach for studying HES.<br />
In the GAIA approach the equation H = H<br />
holds true.<br />
In integrated modeling (Odum, 1997; Holling,<br />
2001) variables from the social system (such as<br />
resource availability) <strong>and</strong> variables from environmental<br />
systems (such as economic growth) are<br />
considered within one system structure, mutually<br />
<strong>and</strong> functionally related <strong>and</strong> sometimes even<br />
hierarchically related. Integrative modeling starts<br />
from coupled systems <strong>and</strong> provides a quantitative<br />
analysis (Bossel, 1998; Carpenter et al., 1999).<br />
The HES approach presented below separates<br />
human <strong>and</strong> environmental systems <strong>and</strong> studies<br />
their mutualism. Note that the concept of<br />
environment emerged in the early 19 th<br />
century, a<br />
time when the upcoming industrial age<br />
unmistakably revealed the interaction <strong>and</strong> mutual<br />
dependency between these two systems. Mutual<br />
dependency, reciprocity, <strong>and</strong> the H impact<br />
chains can be approached from the environmental<br />
as well as from the human perspective. The former<br />
looks at optimizing environmental quality by<br />
integrating human models into ecosystem analysis<br />
(Naveh & Lieberman, 1994). The latter<br />
investigates the impact of regulatory mechanisms<br />
on the state of the environment when taking an<br />
anthropogenic perspective (Hammond, et al.,<br />
1995).<br />
2. BASIC PRINCIPLES FOR<br />
MODELING HES<br />
2.1 Six basic principles<br />
This paper follows an approach, which begins from<br />
six basic assumptions from the modeling of HES<br />
(Figure 1)<br />
(1) Conceive human <strong>and</strong> environmental systems<br />
as two different, complementary, interrelated<br />
systems with human action <strong>and</strong> “immediate<br />
environmental reaction” being part of both<br />
systems.<br />
(2) Consider a hierarchy of human systems with<br />
related environmental systems.<br />
(3) Construct a ‘state of the art’ model of the<br />
environmental system <strong>and</strong> its long-term<br />
dynamics.<br />
(4) Provide a decision theoretic conceptualization<br />
of the human system with the components<br />
goal formation, strategy formation, strategy<br />
selection <strong>and</strong> action.<br />
(5) Characterize <strong>and</strong> conceptualize different types<br />
of environmental awareness in each<br />
component of (4).<br />
(6) Distinguish <strong>and</strong> model primary <strong>and</strong> secondary<br />
feedback loops with respect to human action.<br />
We will explain these principles <strong>and</strong> illuminate the<br />
specific contribution of the HES approach.<br />
Principle (1) “departs from approaches that try to<br />
underst<strong>and</strong> <strong>and</strong> predict complex dynamics resulting<br />
from endogeneous interactions without any<br />
exogeneous interference (human intervention)”<br />
(quoted from an anonymous review). The Human<br />
species is treated as a separate entity (left part of<br />
Figure 1) with complementary environmental<br />
systems presented at the right side of the figure.<br />
This is done because we have different insight <strong>and</strong><br />
access to natural <strong>and</strong> social systems <strong>and</strong> as we<br />
acknowledge that knowledge about these systems<br />
is organized in disciplines, which emerged from an<br />
underst<strong>and</strong>ing of natural <strong>and</strong> social systems. The<br />
links between these systems are, in a first view,<br />
the immediate physical impacts or perceivable<br />
changes caused by human action, i.e., the felled<br />
tree, which might cause an accident at work.<br />
For human systems, Principle (2) departs from<br />
Miller’s (1978) hierarchical levels <strong>and</strong><br />
distinguishes between the individual, the group,<br />
the organization, <strong>and</strong> society. Of course systems of<br />
a smaller scale such as organ, cell, RNA etc. <strong>and</strong><br />
systems of a higher level such as supranational<br />
systems can also be considered. At each hierarchy<br />
level specific human – environment relationships<br />
<strong>and</strong> regulatory mechanisms are encountered. As<br />
Forman (1995, p. 505) notes, these specific<br />
interactions between human end environmental<br />
systems are of importance as “… control or<br />
regulation mechanisms that produce stability are<br />
usually interpreted in terms of hierarchy, …”<br />
(Forman, 1995, p. 505) Thus it is a specific<br />
challenge for researcher, when considering human<br />
action, to construct the appropriate complementary<br />
environmental systems.<br />
792
Within each hierarchy level, insights from subdisciplines<br />
can be integrated into the process <strong>and</strong><br />
structure modelM. This is particular of interest, if<br />
sustainable action is the object of research: “One<br />
way to generate more robust foundations for<br />
sustainable decision making is to search for<br />
integrative theories that combine disciplinary<br />
strengths while filling disciplinary gaps.”<br />
(Gunderson, Holling, & Ludwig, 2002, p. 8).<br />
Perhaps, to scientists, the ‘state of the art’ request<br />
in Principle (3) seems to be trivial. The message<br />
of this principle, however, is twofold. First it<br />
implies that long term predictions, e.g. on species<br />
biodiversity, resource availability, changes of<br />
resilience etc. are statements on fuzzy <strong>and</strong> context<br />
bound as they include unknown dynamics due to<br />
adaptability or general contextual changes. But,<br />
second, the ‘state of the art’ attribute also indicates<br />
that analysis with the HES approach should refer<br />
to the (robust) current body of knowledge but<br />
cannot go beyond.<br />
Principle (4) is characteristic for the adopted<br />
decision theoretic perspective. We start with<br />
intended action or goals <strong>and</strong> consider human<br />
behavior to be functional <strong>and</strong> purposeful<br />
(Brunswik, 1952; Scholz & Tietje, 2002). We<br />
distinguish goal formation, strategy formation, <strong>and</strong><br />
strategy selection. According to a decision<br />
theoretic framework, preferences <strong>and</strong> strategies are<br />
the basic components of behavior. In this context<br />
we refer to a game theoretic conception of a<br />
strategy in extensive games (Osborne &<br />
Rubinstein, p. 92). We define a strategy as a<br />
complete plan – which the researcher has to<br />
construct – that provides a behavioral directive for<br />
each situation in the course of goal attainment. The<br />
goals establish preference structures that underlie<br />
strategy evaluation. The latter, of course also<br />
depends on the capability, experiences <strong>and</strong><br />
constraints of the HES under consideration. The<br />
goal systems is conceived a rather stable,<br />
situational activated, <strong>and</strong> based on onto- <strong>and</strong><br />
phylogenic history (Scholz, 1987 We assume that<br />
– at least starting from hierarchy level of the<br />
individual upwards – human system can<br />
subjectively evaluate the supposed expected utility<br />
or gain of a strategy. This stage is called foresight<br />
or anticipation.<br />
In Principle (5) we conceptualize environmental<br />
awareness. This can be done, for example, on three<br />
levels with the level of a) completely ignoring the<br />
impacts resulting from action, b) incorporating<br />
environmental sensitivity <strong>and</strong> change, <strong>and</strong> c)<br />
altruistically neglecting oneself <strong>and</strong> only targeting<br />
the benefits for the environment as is partly the<br />
case in deep ecology approaches.<br />
The action <strong>and</strong> the immediate reaction are<br />
conceived of as the changes in the HES system<br />
resulting from a certain strategy, under given<br />
environmental circumstances <strong>and</strong> constraints. This<br />
is the interface, H , between the two systems.<br />
The human <strong>and</strong> the environmental systems are in a<br />
physically different state after a human action is<br />
performed. A critical issue is what is conceived of<br />
as immediate <strong>and</strong> what is conceived of as a delayed<br />
reaction. According to the decision theoretic<br />
perspective, an environmental reaction is defined<br />
by the episode, period or events that are temporally<br />
<strong>and</strong> at least partly causally related to the<br />
consequences of the action. The time period<br />
depends on the memory <strong>and</strong> the environmental<br />
model of the human unit. Note that this statement<br />
holds both for cell conditioning (Brembs et al.,<br />
2002) as well as for governmental environmental<br />
protection programs.<br />
Principle (6) differentiates between two types of<br />
post-decisional evaluation. One takes place<br />
temporally proximally to the environmental<br />
reaction <strong>and</strong> can be conceived of as learning by<br />
primary feedback. Another post decisional<br />
evaluation is considered to take place temporally<br />
<strong>and</strong> spatially proximally to the environmental<br />
reaction. However, human action can result in side<br />
effects, i.e. unintended dynamics <strong>and</strong> dislocated<br />
reaction, which alter the environmental system in a<br />
favorable or unfavorable manner <strong>and</strong> which can<br />
show rebound effects. Side effects are often delayed<br />
(Venix, 1996) or dislocated as, from the human<br />
system perspective, they are not directly related to<br />
the perceived environmental reaction. These<br />
temporal (or spatial) delays (dislocations) in the<br />
environmental system are considered to be second<br />
order feedback to the human system, as the<br />
individual will notice the effects later (or at other<br />
places). A critical question is whether, in which<br />
way (i.e. by which “algorithms”), <strong>and</strong> when a<br />
delayed or dislocated impact is evaluated. If we<br />
follow the principles of bounded rationality,<br />
optimizing primary <strong>and</strong> secondary learning<br />
depends (i) on economically sampling of<br />
appropriate cues or evidences related to action, (ii)<br />
on setting suitable <strong>and</strong> robust time <strong>and</strong> spatial<br />
boundaries, <strong>and</strong> (iii) efficiently changing goal<br />
formation, strategy selection, <strong>and</strong> strategy<br />
evaluation.<br />
Of particular interest is the fact that human action<br />
at one level of the human system, may lead to<br />
environmental impacts, which in turn provide<br />
feedback to the human system at a level different<br />
to the one of action. That is, feedback loops do not<br />
necessarily occur within one scale or level of the<br />
human system, but across levels. In addition, the<br />
human systems might differ in their goals, <strong>and</strong><br />
strategies, generating interfering actions <strong>and</strong><br />
environmental feedbacks. For example, fast<br />
financial success in a market can trigger slow, but<br />
deep changes in structures on another level. “Thus<br />
modern economists are frustrated in their attempts<br />
to underst<strong>and</strong> the interactions between fast- <strong>and</strong><br />
slow moving variables that create emergent dynamics.”<br />
(Gunderson et al. 2002, p. 8)<br />
793
into environmental quality (from<br />
the green side) <strong>and</strong> market<br />
dynamics (including material <strong>and</strong><br />
money flows) from the bankers’<br />
side.<br />
Figure 1 . A structure - process model of HES.<br />
2.3 The concept of regulatory, feedback <strong>and</strong><br />
control (RFC-) mechanisms<br />
One peculiarity of the HES model is that it<br />
challenges environmental research to investigate<br />
RFC-mechanisms, in particular, adaptive cycles<br />
that describe real <strong>and</strong>/or sustainable behavior. The<br />
critical issue in this is the appropriate balance<br />
between change <strong>and</strong> stability. According to<br />
evolutionary approaches homeostasis <strong>and</strong><br />
stationary, dynamic or evolutionary stable<br />
equilibria are of specific interest. From ecosystem<br />
sciences <strong>and</strong> theories of development, however, we<br />
have learnt that adaptive cycles pass stages such as<br />
release, reorganization, exploitation <strong>and</strong><br />
conservation (Holling & Gunderson. 2001). In<br />
principle, these ideas <strong>and</strong> other assumptions (e.g.,<br />
those related to chaos) can be linked to the<br />
development by stages (Piaget, 1953).<br />
Let us illustrate the idea of RFC-mechanism in<br />
HES by considering the case of formation of a<br />
biofuel company. A crucial step of HES research<br />
would be to determine the key actors (e.g. the<br />
technology pioneer or the head of a credit<br />
department of a large bank) <strong>and</strong> their drivers (i.e.<br />
goals). This can be done based on decision<br />
psychology (see Scholz et al 2003). The<br />
environmental changes can be in a first approach<br />
modeled by the physical <strong>and</strong> financial flows with<br />
their environmental impacts. Then, a sophisticated<br />
HES model would describe a primary feedback<br />
loop illustrating both the impacts would be of<br />
certain strategies (i.e. business plans of the<br />
pioneer; credit rating <strong>and</strong> portfolio considerations<br />
of the credit officer) <strong>and</strong> the short-term impacts on<br />
the material fluxes <strong>and</strong> its environmental impacts.<br />
The secondary feedback loop would include insight<br />
3. THE EXAMPLE OF<br />
BIOWASTE MANAGEMENT<br />
In the following, we report the<br />
application of the model presented<br />
above for the case of bio-waste<br />
management in Zurich (see Lang et<br />
al., 2003). We present an ex-post<br />
case analysis. We begin by<br />
studying the history of decisionmaking<br />
in the human system until<br />
the environmental reaction was<br />
obtained. In order to do so we will<br />
study the change in the system<br />
between the years 1995 <strong>and</strong> 2002.<br />
The regulatory level taken is that of<br />
the canton of Zurich, which felt<br />
uneasy with the resource depletion of organic waste<br />
by incineration.<br />
3.1 Methods <strong>and</strong> data<br />
Data were obtained from the environmental agency<br />
in Zurich (AWEL), which has been gathering data<br />
on the quantity <strong>and</strong> quality of bio-waste material<br />
since 1991.<br />
To model the environmental system (i.e., environmental<br />
reaction), we extended the method of Material<br />
Flux Analysis (MFA; Baccini <strong>and</strong> Bader,<br />
1996) to include agents in the system analysis so<br />
that the environmental system can be linked to the<br />
human system (Binder et al., 2004). To investigate<br />
the human system we used literature analysis, oral<br />
history <strong>and</strong> expert interviews.<br />
Utilizing the case study methodology, the impacts<br />
<strong>and</strong> underlying rationale of the development were<br />
reconstructed <strong>and</strong> the key agents in the process of<br />
technology implementation <strong>and</strong> their operations<br />
were identified. The methodology for this<br />
proceeding is described in Scholz & Tietje, 2003;<br />
pp. 84; Laws et al. 2002; Binder et al., 2004).<br />
3.2 System analysis<br />
Figure 2 presents the system analysis for bio-waste<br />
management in the canton of Zurich. The system<br />
border is the canton of Zurich. The system is composed<br />
of 5 processes <strong>and</strong> 10 flows. There are three<br />
main bio-waste delivery processes, i.e., municipal<br />
collection, gardeners, <strong>and</strong> industry. Separately<br />
collected bio-waste can be treated in two ways:<br />
composting or anaerobic digestion, which are<br />
considered as action alternatives that are<br />
components of mixed strategies (i.e., an allocation<br />
of fractions treated with these two modes of waste<br />
processing; for the sake of simplicity, incineration<br />
794
is not considered see Schleiss <strong>and</strong> Scholz, 2002).<br />
The main agents are directly related to the<br />
processes. Additional agents are municipalities, the<br />
canton of Zurich <strong>and</strong> the State (i.e., Switzerl<strong>and</strong>).<br />
Figure 2. System analysis for bio-waste<br />
management in the canton of Zürich<br />
(Adapted from Lang et al., 2003)<br />
3.3 Results<br />
From 1995 to 2002 the total amount of biowaste<br />
separately collected <strong>and</strong> treated (<strong>and</strong> not being<br />
incinerated) increased by 37%, which corresponds<br />
to 36’500 tons (Lang et al., 2003). The largest<br />
relative increase was found for industries followed<br />
by gardeners <strong>and</strong> municipalities. The latter had<br />
already had a high collection <strong>and</strong> delivery rate in<br />
1995 whereas industries had not separately<br />
collected their bio-waste before (Table 1). Nearly<br />
all the newly collected bio-waste was treated by<br />
anaerobic digestion.<br />
Table 1. Amount of bio-waste delivered by the<br />
main agents in 1995 <strong>and</strong> 2001 in t/year (Source:<br />
Lang et al., 2003) a<br />
Year Municipalities Gardeners Industry<br />
1995 51’000 27’500 1’400<br />
2001 70’000 40’000 10’500<br />
a: Rounded values<br />
To explain the development in bio-waste<br />
management we present three major stages <strong>and</strong><br />
related impact factors.<br />
First, the anaerobic digestion technology improved<br />
from in the mid of the 1990ies so that it became<br />
an economic <strong>and</strong> ecologically feasible alternative to<br />
composting for treating bio-waste. This<br />
development was initiated by some pioneers in the<br />
field of green technologies (i.e., KOMPOGAS).<br />
Second, for Swiss industries (including large<br />
retailers such as MIGROS, which has more than<br />
1/3 of the market share), environmental impacts<br />
became increasingly important, with environmental<br />
performance becoming more <strong>and</strong> more of an issue<br />
of prestige where the goal is to demonstrate that<br />
they run an ecologically <strong>and</strong> socially responsible<br />
business. Therefore, the option of delivering biowaste<br />
for anaerobic digestion <strong>and</strong> utilizing the fuel<br />
for driving their business trucks seemed a good<br />
option for showing their commitment towards<br />
utilizing “clean energy”.<br />
Third, for municipalities it was not always easy to<br />
have a well managed composting plant, thus, the<br />
plants provided severe odor emissions, which led<br />
to complaints within the population. Thus, a less<br />
odor intensive treatment seemed to be an appropriate<br />
solution (Lang et al., forthcoming).<br />
Finally, Switzerl<strong>and</strong> has signed the Kyoto protocol<br />
<strong>and</strong> committed itself to reduce the CO2 emissions.<br />
Given the decentralized structure in Switzerl<strong>and</strong>,<br />
the canton Zurich saw a possibility to reduce CO2<br />
emission by fostering biogas production. Thus, it<br />
issued the article 12a of the Zurich energy law,<br />
which became effective in January 1996. This<br />
article states that if it is technically <strong>and</strong><br />
economically viable “all compostable wastes,<br />
which cannot be composted locally, have to be<br />
processed to marketable goods in central facilities<br />
utilizing their energy-potential” 1 .<br />
Thus, the combination of several goals at<br />
completely different hierarchical levels led to the<br />
observed environmental reaction of an increase in<br />
the collection of bio-waste <strong>and</strong> consequent<br />
treatment in anaerobic digestion plants. The<br />
feedback at all levels was visible as the share of<br />
anaerobic digestion increased <strong>and</strong> is likely to grow<br />
further. This example, however, clearly shows, that<br />
processes <strong>and</strong> change within HES have to be<br />
studied by including not only the physical flows<br />
or the social perspective. Rather a combination <strong>and</strong><br />
integration of these analyses within a concise<br />
framework of the process structure model allows<br />
for a better underst<strong>and</strong>ing of the system <strong>and</strong> might<br />
support transition processes in these systems.<br />
4. DISCUSSION<br />
4.1 Current research with HES<br />
The HES approach is currently being used to<br />
organize a new, medium-scale institute with an<br />
identical name at the ETH Zurich. The focus on<br />
environmental decision-making is dominant in this<br />
institute. In HES research, at present, the case<br />
study approach is dominant. Apart from case<br />
studies on technology breakthrough, historical case<br />
studies, e.g. on the mastership of cholera <strong>and</strong> pest<br />
1<br />
Original German wording: “Kompostierbare Abfälle, die<br />
nicht dezentral kompostiert werden können, sind unter<br />
Ausschöpfung des Energiepotentials in zentralen Anlagen<br />
zu marktfähigen Produkten zu verwerten, soweit dies<br />
technisch und wirtschaftlich möglich ist.” (1983, §12a)<br />
795
control were carried out as they ideally allow the<br />
examination of interfering regulatory systems.<br />
However, in principle, psychological experiments<br />
can also be carried out so long as the<br />
environmental dynamics are simulated with<br />
appropriate computer programs.<br />
4.2 Critical issues of HES research<br />
HES research is under construction. The following<br />
four challenges deserve special attention:<br />
• How can reconstructions of human decision<br />
making with the HES framework be validated?<br />
•<br />
2 In which way can we construct appropriate<br />
“secondary feedback loops”? How can we<br />
show that systems become stabilized or<br />
system transformations become optimized if<br />
secondary, long-term dynamics is taken into<br />
account?<br />
• What role can RFC-mechanisms play in<br />
sustainability management?<br />
• Is decision research (including game theory<br />
<strong>and</strong> risk analysis), with its conception of<br />
games against nature, not only a language but<br />
also a tool for integrating knowledge <strong>and</strong><br />
overcoming both the gaps between social <strong>and</strong><br />
natural sciences <strong>and</strong> those between theory <strong>and</strong><br />
practice?<br />
5. CONCLUSIONS<br />
We presented a process structure model that allows<br />
for studying adaptation <strong>and</strong> development processes<br />
within Human-Environment systems. This model<br />
divides the human <strong>and</strong> the environment system,<br />
providing so the basis for integrating disciplinary<br />
knowledge within one framework. Therefore,<br />
complex systems consisting of interactions among<br />
several agents can be understood <strong>and</strong> transition<br />
processes initiated.<br />
6. ACKNOWLEDGEMENTS<br />
The authors wish to thank an anonymous reviewer<br />
for excellent remarks, Daniel Lang for research<br />
cooperation <strong>and</strong> Peter Loukopoulos for assistance<br />
in the editing.<br />
7. REFERENCES<br />
Apostle, H. G., 1952. Aristotle’s philosophy of mathematics.<br />
Chicago: Chicago University Press.<br />
Ashby, W.R., 1956, An Introduction to Cybernetics. John<br />
Wiley, New York.<br />
Baccini, P. & Bader, H.-P., 1996. Regionaler Stoffhaushalt,<br />
Erfassung, Bewertung und Steuerung, Heidelberg: Spektrum.<br />
Brembs et al., 2002. Operant reward learning in aplysia:<br />
neuronal correlates <strong>and</strong> mechanisms, Science 296: 1706-1709<br />
Bossel, H., 1998. Earth at a crossroads: Path to a<br />
sustainable future. Cambridge: Cambridge University Press.<br />
Brunswik, E., 1952. The conceptual framework of<br />
psychology. Chicago: University of Chicago Press.<br />
von Carlowitz H. C., 1732. Sylvicultura oeconomica oder<br />
hausswirthliche Nachricht und naturmässige Anweisung zur<br />
wilden Baum-Zucht. Leipzig: bey Johann Friedrich Braunssel.<br />
Erben.<br />
Binder C.R., Hofer C., Wiek A., <strong>and</strong> Scholz R.W. Transition<br />
towards improved regional wood flow by integrating material<br />
flux analysis with agent analysis: the case of Appenzell<br />
Ausserrhoden, Switzerl<strong>and</strong>, Ecological Economics, Vol 49/1<br />
pp 1-17, 2004.<br />
Carpenter, S., Brock, W., & Hanson, P., 1999. Ecological<br />
<strong>and</strong> social dynamics in simple models of ecosystem<br />
managment. Conservation Ecology, 3(2):pp. 351-360.<br />
Goodwin, J.W., 1977. Agricultural economics. Reston:<br />
Reston.<br />
Gunderson, L.H. & Holling, C.S., & Ludwig, D., 2002. In<br />
quest of a theory of adaptive change. Gunderson, L.H. &<br />
Holling, C.S.: Panarchy. Underst<strong>and</strong>ing transformations in<br />
human <strong>and</strong> natural systems (pp.3-24). Washington: Isl<strong>and</strong><br />
Press.<br />
Hammond, A., Adriaanse, A., Rodenburg, E., Bryant, D., &<br />
Woodward, R., 1995. <strong>Environmental</strong> indicators: A systematic<br />
approach to measuring <strong>and</strong> reporting on environmental policy<br />
performance in the context of sustainable development.<br />
Washington: World Resources Institute.<br />
1983. Energiegesetz [Energy law] (St<strong>and</strong> am 1. Juli 2002 ).<br />
Pages 5 in.<br />
Holling, C.S., 2001. Underst<strong>and</strong>ing the Complexity of<br />
Economic, Ecological, <strong>and</strong> Social Systems. Ecosystems, 4, 390<br />
- 405.<br />
Kemp, D., 1998. The environment dictionary. London:<br />
Routledge.<br />
Lang D.J., Binder C.R., Stäubli B., <strong>and</strong> Scholz R. W., 2003,<br />
Optimization of waste management systems by integrating<br />
material fluxes, agents <strong>and</strong> regulatory mechanisms – The case<br />
of bio-waste. Paper presented at Environment 2010: Situation<br />
<strong>and</strong> Perspectives for the European Union , Porto, Portugal, May<br />
6-10, 2003,<br />
Laws, D., Scholz, R.W., Shiroyama, H., Susskind, L.,<br />
Suzuki, T. & Weber, O., 2002. Expert views on sustainability<br />
<strong>and</strong> technology implementation.<br />
Miller, G. A. (1978). Living systems. New York: McGraw-<br />
Hill.<br />
Naveh, Z., & Liebermann, A. S., 1994. L<strong>and</strong>scape ecology:<br />
Theory <strong>and</strong> application (2nd ed.). New York: Springer.<br />
Odum, E. P., 1997, Ecology: a bridge between science <strong>and</strong><br />
society. Sunderl<strong>and</strong>, MA : Sinauer.<br />
Osborne, M.J. & Rubinstein, A., 1994. A course in game<br />
theory. Cambridge: MIT Press.<br />
Piaget, J., 1953. The origin of intelligence in the child.<br />
Londin: Routledge & Kegan.<br />
Schleiss, K. & Scholz, R. W., 2001. Alternativen der<br />
Kompostbewirtschaftung. Müll und Abfall , 79-82.<br />
Scholz, R. W., 1987. Cognitive strategies in stochastic<br />
thinking. Dordrecht: Reidel.<br />
Scholz, R.W., Mieg, H.,A. & Weber, O., 2003.<br />
Wirtschaftliche und organisationale Entscheidungen, In:<br />
Auhagen, A.E., & Bierhoff, H.W. Wirtschafts- und<br />
Organisationspsychologie (pp. 194-219). Weinheim: Beltz.<br />
Scholz, R. W., & Tietje, O., 2002. Embedded case study<br />
methods. Integrating qualitative <strong>and</strong> quantitative knowledge .<br />
Thous<strong>and</strong> Oaks: Sage Publications.<br />
Simon, H.A., 1957. Models of man. New York: Wiley.<br />
Venix, J.A.M., 1996. Group model building – facilitating<br />
team learning using system dynamics. Wiley: West Sussex.<br />
2 From a theory of science perspective, in many cases a<br />
“gentle verification” by agreements of participants seems<br />
possible.<br />
796
Addressing Sustainability, HIV-AIDS, <strong>and</strong> Water<br />
Resource Questions in Botswana<br />
M. Hellmuth, J. Sendzimir, Yates, David, USA; Strzepek, Kenneth, USA; <strong>and</strong> S<strong>and</strong>erson, Warren,<br />
USA<br />
Abstract: An integrated population, economic, <strong>and</strong> water resource model was developed to address<br />
sustainable development questions for Botswana. Traditionally, water resources planning models have<br />
considered the implications of different assumptions of population <strong>and</strong> economic growth on the<br />
sustainability of existing water resources supply; however, this model extends that capability to consider<br />
feedbacks from one model component to another. The water model uses a physically based hydrologic<br />
rainfall-runoff model, with surface <strong>and</strong> groundwater components, to produce monthly runoff <strong>and</strong><br />
groundwater recharge at the watershed scale. Surface runoff <strong>and</strong> recharge are the inflows into surface <strong>and</strong><br />
groundwater water reservoirs. The demographic sub model is a st<strong>and</strong>ard multi-cohort model that forecasts<br />
the population by age, sex, rural, urban, education <strong>and</strong> hiv/aids status. The economic sub-model is a<br />
computable general equilibrium model with three sectors: agriculture, non-agricultural exports, <strong>and</strong> non<br />
tradables. The model runs an ensemble of scenarios, including climate change, HIV-AIDS, health,<br />
economic, <strong>and</strong> water conservation scenarios, whose output is probabilistic in nature.<br />
1. INTRODUCTION<br />
This paper includes an overview of the Botswana<br />
PDE-IWS model, including a qualitative<br />
description of the water model components, the<br />
population <strong>and</strong> economic models <strong>and</strong> a<br />
description of the model linkages.<br />
2. THE WATER MODEL<br />
The water model is composed of a rainfall-runoff<br />
model, surface <strong>and</strong> groundwater reservoirs, <strong>and</strong> a<br />
water dem<strong>and</strong> model. The rainfall-runoff model<br />
uses simplified but physically based, mathematical<br />
descriptions of hydrologic processes to assess<br />
climate impacts on river basins (Yates 1996,<br />
Nemec <strong>and</strong> Schaake, 1982; Lettenmaier <strong>and</strong> Gan,<br />
1990; Nash <strong>and</strong> Gleick, 1991; Kaczmarek, 1993;<br />
Yates <strong>and</strong> Strzepek, 1996). This model requires 3<br />
parameters for calibration, <strong>and</strong> may be run at time<br />
steps varying from hourly to daily to monthly (see<br />
Yates 1996 for details). It is a lumped conceptual<br />
model, which uses a soil moisture balance to drive<br />
the runoff <strong>and</strong> infiltration processes. Runoff fills<br />
virtual surface reservoirs while infiltration<br />
recharges virtual groundwater systems within<br />
each SER.<br />
For both the surface <strong>and</strong> groundwater models, a<br />
water mass balance is computed at each time step<br />
using the virtual reservoir as the control volume.<br />
These supply reservoirs, balance inflows,<br />
evaporative <strong>and</strong> human dem<strong>and</strong>s, as well as inter<br />
basin transfers, at each time step.<br />
The water dem<strong>and</strong> model computes the following<br />
dem<strong>and</strong>s at each time step: Domestic,<br />
Institutional, Energy, Industrial, Mining,<br />
Livestock <strong>and</strong> Irrigation. Water consumption in<br />
each of these sectors is driven by economic <strong>and</strong>/or<br />
population changes.<br />
3. THE POPULATION SUB-MODEL<br />
The population model is described in detail in<br />
S<strong>and</strong>erson (2001a, 2001b, 2002a, 2002b,<br />
forthcoming 2004). The model divides the<br />
population,<br />
1) by age (100 ages from 0 through 99+);<br />
2) by sex (female <strong>and</strong> male);<br />
3) by education (primary <strong>and</strong> below, secondary,<br />
tertiary);<br />
4) by HIV status (HIV negative; HIV positive,<br />
asymptomatic, <strong>and</strong> not on medication; HIV<br />
positive, asymptomatic, <strong>and</strong> on medication; <strong>and</strong><br />
AIDS, i.e., symptomatic);<br />
797
5) by number of years since HIV infection (15<br />
categories from infected this year to infected 14 or<br />
more years, for people who are HIV positive,<br />
asymptomatic, <strong>and</strong> not on medication);<br />
6) by sexual behavior risk group (not at risk,<br />
sometimes at risk); <strong>and</strong> 7) by onset of sexual<br />
activity (for young women <strong>and</strong> men).<br />
purchased from the other sectors, whose<br />
production functions for these two sectors are as<br />
follows:<br />
Q 1 =<br />
Q 2 =<br />
c 1 *n 1 a1 *m 1 a14 *int 12 a12 *int 13<br />
a13<br />
c 2 *n 2 a2 *m 2 a24 *int 22 a22 *int 23<br />
a23<br />
Eq. 1<br />
Eq. 2<br />
4. Botswana Economic Model<br />
The Botswana economic model is a computable<br />
general equilibrium (CGE) model, based on the<br />
structure of the BMW CGE model that was<br />
produced by Becker et al. (1992) for a study on the<br />
Indian economy. The advantage of using CGE<br />
models for integrated assessments is that they<br />
allow for the impacts of policy decisions to be<br />
distributed across the sectors, as equilibrium in all<br />
markets is sought. The model is based The BMW<br />
model has 10 sectors, while the Botswana model<br />
was aggregated into three, including nonagricultural<br />
exports (NAE), non-tradables (NT),<br />
<strong>and</strong> agriculture (including agricultural exports)<br />
(AG).<br />
5. Linking the Models<br />
The main purpose of combining the three PDE<br />
sub-models is to create an integrated model that<br />
can evaluate the implications of feedback<br />
processes of the different sub-components. This<br />
section will describe the specific model links that<br />
were incorporated, in the following order: 1)<br />
population- water, 2) population-economy, 3)<br />
economy- water, 4) economy- population, 5)<br />
water- population, <strong>and</strong> 6) water- economy.<br />
Population- Water<br />
Starting with the population/water relationship,<br />
the population has first order effects on both the<br />
water resource <strong>and</strong> the economy. Both rural <strong>and</strong><br />
urban population stocks factor into determining<br />
the respective amount of domestic water<br />
consumed.<br />
Population—Economy<br />
The population also has a direct effect on the<br />
economy through the skilled labor size <strong>and</strong> labor<br />
productivity. In particular, the Non Agriculture<br />
Export (NAE) <strong>and</strong> Non-Tradables (NT) sectors<br />
are affected by labor size <strong>and</strong> productivity. These<br />
production sectors are represented at the uppermost<br />
level by Cobb-Douglas production functions<br />
in value-added, imports, <strong>and</strong> intermediate goods<br />
where Q1 <strong>and</strong> Q2 (Pula) are the output of the<br />
NAE (1) <strong>and</strong> NT (2) sectors; c1 <strong>and</strong> c2 are<br />
distribution parameters for each sector; n1 <strong>and</strong> n2<br />
represent value added (Pula); <strong>and</strong> int12 <strong>and</strong> int13<br />
(Pula) represent intermediate goods purchased by<br />
the NAE sector from the NT <strong>and</strong> AG (3) sectors;<br />
int22 <strong>and</strong> int23 (Pula) represent intermediate<br />
goods purchased by the NT sector from the NAE<br />
<strong>and</strong> AG sectors; <strong>and</strong> each a* represents the Cobb<br />
Douglas value share for that sector; <strong>and</strong> m1<strong>and</strong><br />
m2 (Pula) represent imports for each sector.<br />
Skilled <strong>and</strong> unskilled labor size <strong>and</strong> productivity<br />
measured by capital are derived from the valueadded,<br />
nested ces functions, which then impact<br />
economic output. The value added ces functions,<br />
n1 <strong>and</strong> n2, (capital, skilled <strong>and</strong> unskilled labor)<br />
are:<br />
n 1 =(e 1 *(φ 1 σ1h )+(1-e 1 )*((LPM U *LU 1 ) σ1h )) (1/σ1h)<br />
n 2 =(e 2 *(φ 2<br />
e 2 )*((LPM U *LU 2 ) σ2h )) (1/σ2h)<br />
Eq. 4<br />
Eq. 3<br />
σ2h )+(1-<br />
where e1 <strong>and</strong> e2 are distribution parameters; σ1h<br />
= 1-1/δ1h <strong>and</strong> σ2h = 1-1/δ2hl. The parameters,<br />
δ1h <strong>and</strong> δ2h are equal to the elasticity of<br />
substitition of LU1 <strong>and</strong> LU2, the unskilled labor<br />
in both sectors; LPMU represents the unskilled<br />
labor productivity multiplier.<br />
The nested value added skilled labor <strong>and</strong> capital<br />
ces functions, φ1 <strong>and</strong> φ2 (Pula) for the NAE (1)<br />
<strong>and</strong> NT (2) sectors are:<br />
φ 1 = (f 1 *(K 1 σ1l ) + (1-f 1 )*(LPM S *LS 1 σ1l )) (1/σ1l)<br />
Eq. 5<br />
φ 2 = (f 2 *(K 2 σ2l ) + (1-f 2 )*((LPM S *LS 2 σ2l )) (1/σ2l)<br />
Eq. 6<br />
798
where K1 <strong>and</strong> K2 represent capital value in the<br />
NAE <strong>and</strong> NT sectors; LPMS is the Labor<br />
productivity multiplier for skilled labor; LS1 <strong>and</strong><br />
LS2 are skilled labor in both sectors; f1 <strong>and</strong> f2 are<br />
distribution parameters; where σ1l = 1-1/δ1l <strong>and</strong><br />
σ2l= 1-1/ δ2l. The parameters, δ1l <strong>and</strong> δ2l are<br />
equal to the elasticity of substitution.<br />
Finally, the intermediate goods purchased by the<br />
NAE sector from the NT <strong>and</strong> AG sectors <strong>and</strong> the<br />
imports purchased by the NAE sector are defined<br />
by:<br />
int 12 =a 12 *w 1u /(P 2 *a 1 *(1-e 1 )*n 1 (1/δ1h-1) *<br />
(LPM U *LU1) (-1/δ1h) )<br />
int 13 =a 13 *w 1u /(P 1 *a 1 *(1-e 1 )*n 1 (1/δ1h-1) *<br />
Eq. 7<br />
(LPM U *LU 1 ) (-1/δ1h)) Eq. 8<br />
more thoroughly described in S<strong>and</strong>erson (2001a,<br />
2002a, 2002b).<br />
Water—Population<br />
The connection of water to population is through<br />
diarrhea incidence. Research indicates that<br />
HIV/AIDS individuals are vulnerable to more<br />
severe, prolonged <strong>and</strong> recurrent diarrhea episodes<br />
(NIH, 1994), <strong>and</strong> that the progression rate from<br />
HIV to AIDS is influenced by nutrition or stress<br />
(Bogden, et al 2000; Timbo <strong>and</strong> Tollefson, 1994).<br />
In the model, diarrhea incidence affects the<br />
population in two ways, 1) the AIDS death rate;<br />
<strong>and</strong> 2) the HIV/AIDS progression rate from HIV<br />
to AIDS. The number of diarrheal cases per<br />
capita, DC, is a function of precipitation. The<br />
relationships of total annual diarrhea cases per<br />
10,000 capita for the population age group > 5<br />
<strong>and</strong> the population age group < 5 <strong>and</strong><br />
precipitation are:<br />
m 1 =a 14 *w 1u /(P M *(1+τ)*a 1 *(1-<br />
e 1 )*n 1<br />
(1/δ1h- 1) *<br />
(LPM U *LU 1 ) (-1/δ1h)) Eq. 9<br />
where P1, P2 <strong>and</strong> PM are prices of NAE, NT <strong>and</strong><br />
imports(4); τ is the tariff rate; w1u is the unskilled<br />
NAE wage rate; <strong>and</strong> δ1h is the outer layer<br />
substitution elasticity.<br />
Economy—Water<br />
The first order relationship of the economy to the<br />
water model is through the economic outputs of<br />
real GDP per capita, gross output of nontradables,<br />
<strong>and</strong> gross output of exports as drivers of<br />
the domestic, industrial, energy, institutional, <strong>and</strong><br />
mining water consumers. Industrial, energy <strong>and</strong><br />
institutional water dem<strong>and</strong>s are linearly related to<br />
the total industrial <strong>and</strong> commercial output, <strong>and</strong><br />
the growth of water use in mining is driven by<br />
changes in the NAE sector. In addition, direct<br />
investment in water supply <strong>and</strong> sanitation can be<br />
made.<br />
Economy—Population<br />
The economic model is connected to the<br />
population model based on assumptions of<br />
government investment levels in HIV/AIDS<br />
medication <strong>and</strong> education. This connection is<br />
DC>5=-0.0084*P D 2 +3.0292*P D + 108.62<br />
Eq. 10<br />
DC
upon the incidence of diarrhea, where the<br />
subscript s <strong>and</strong> u are the skilled <strong>and</strong> unskilled<br />
labor, respectively.<br />
The measure of “Lost Labor Days” (LLD) creates<br />
a link between diarrhea <strong>and</strong> Labor Productivity:<br />
LLD = (PPD * LF * ASD) Eq. 12<br />
where LF is the labor force size, PPD is the<br />
percentage of population over 5 with diarrhea, <strong>and</strong><br />
ASD is parameter describing the average number<br />
of sick days per diarrheal episode (ASD= 2<br />
days/episode, Pegram et al., 2002). The<br />
percentage of population over 5 with diarrhea,<br />
PPD is a function of DC:<br />
PPD = DC/10,000 Eq. 13<br />
A relationship between diarrhea incidence <strong>and</strong><br />
labor productivity can be established. For the<br />
skilled <strong>and</strong> unskilled labor forces there is a<br />
negative impact on labor productivity when the<br />
predicted diarrhea cases are higher than the mode,<br />
<strong>and</strong> for the skilled laborers there is a positive<br />
impact on labor productivity when the predicted<br />
diarrhea cases are less than the mode. Then, the<br />
Skilled <strong>and</strong> Unskilled Labor Force Multiplier<br />
parameters are equivalent to:<br />
LPM S = 1<br />
LPM U = 1<br />
when DC = mode<br />
Eq.14<br />
= 1 + (LLD M –LLD)/ (LF*WD)<br />
when DC <<br />
mode<br />
= 1 – (LLD- LLD M )/ (LF*WD)<br />
when DC ><br />
mode<br />
when DC <br />
mode<br />
where LLDM is the lost labor days at the mode of<br />
diarrhea incidence.<br />
6. Conclusions<br />
The Botswana PDE-IWS model methodology was<br />
described. The model allows for the examination<br />
of feed-forward <strong>and</strong> feedback processes, the<br />
impact of water-related diarrhea on the population<br />
could be examined.<br />
7. REFERENCES<br />
Becker, C.M.; Williamson, J.G.; Mills, E.S. 1992.<br />
Indian urbanization <strong>and</strong> economic growth<br />
since 1960. The John’s Hopkins University<br />
Press. London.<br />
Bogden, J.D.; Kemp, F.W.; Han, S.; Li, W.;<br />
Bruening, K.; Denny, T.; Oleske, J.M.;<br />
Lloyd, J.; Baker, H.; Perez, G.; Kloser, P.;<br />
Skurnick, J.; Louria, D.B. 2000. Status of<br />
selected nutrients <strong>and</strong> progression of human<br />
immunodeficiency virus type 1 infection.<br />
American Journal of Clinical Nutrition,<br />
72(3): 809-15.<br />
Danish Hydraulics Institute, 2003. MIKE BASIN:<br />
A Versatile Decision Support<br />
Tool for Integrated Water Resources Management<br />
<strong>and</strong> Planning.<br />
FAOSTAT. 1997. FAO Online Database.<br />
http://apps.fao.org/<br />
Fulp, T.J., Vickers, W.B., Williams, B., <strong>and</strong> King,<br />
D.L., 1995, Decision support for water<br />
resources management in the Colorado River<br />
region: IN L. Ahuja, J. Leppert, K. Rojas,<br />
<strong>and</strong> E. Seely (eds.), Workshop on computer<br />
applications in water management, Colorado<br />
Water Resources Research Institute, Fort<br />
Collins, CO., Information Series No. 79, pp.<br />
24-27.<br />
Government of Botswana. Ministry of Finance<br />
<strong>and</strong> Development Planning. 1997. National<br />
Development Plan 8. Government Printer,<br />
Gaborone, Botswana.<br />
Government of Botswana, 1999. Annual<br />
Economic Report. Ministry of Finance <strong>and</strong><br />
Development Planning.<br />
Government of Botswana. Central Statistics<br />
Office. 2002. Preliminary census results.<br />
http://www.cso.gov.bw/cso/<br />
Hellmuth, M.E. 2002. Ph.D. Thesis: A Systems<br />
Modeling Approach to Population,<br />
Economics <strong>and</strong> Water Resource Interactions<br />
in Botswana: A Cost-Benefit Analysis of<br />
Government Investment Options.<br />
Department of Civil Engineering, University<br />
of Colorado, Boulder.<br />
800
Hellmuth, M.E. <strong>and</strong> S<strong>and</strong>erson, W.C. 2001.<br />
Southern Africa – Water Stressed by 2001?<br />
APC-EU Courier, No. 189, November-<br />
December. pp. 40-41.<br />
IIASA. 2001. Botswana's, Namibia's, <strong>and</strong><br />
Mozambique's future: modeling population<br />
<strong>and</strong> sustainable development challenges in<br />
the era of HIV/AIDS. CD-ROM. Laxenburg,<br />
Austria: <strong>International</strong> Institute for Applied<br />
Systems Analysis.<br />
Johnson <strong>and</strong> Verheust. 1998. Database of South<br />
African surface water bodies. ALCOM <strong>and</strong><br />
FAO, Harare: Zimbabwe.<br />
Kaczmarek, Z. 1993. Water balance model for<br />
climate impact analysis, ACTA Geophysica<br />
Polonica, 41(4): 1-16.<br />
Lettenmaier D.P. <strong>and</strong> Gan, T.Y. 1990. Hydrologic<br />
sensitivities of the Sacramento-San Joaquin<br />
river basin, California, to global warming.<br />
Water Resources Research, 26 (1): 69-86.<br />
Nash, L.L. <strong>and</strong> Gleick, P. 1991. The sensitivity of<br />
streamflow in the Colorado basin to climatic<br />
changes. Journal of Hydrology, 125: 221-241<br />
National Institute of Health. 1994. Fact Sheet #7:<br />
HIV/AIDS <strong>and</strong> Alternative Therapies. Office<br />
of Alternative Medicine.<br />
http://www.aegis.com/pubs/cdc_fact_sheets/1<br />
994/cdc94033.html<br />
Nchito, M.; Kelly, P.;Siangongo, S.; Luo, N.P.;<br />
Feldman, R.; Farthing, M; <strong>and</strong> Baboo, K.S.<br />
1998. “Cryptosporidiosis in Urban Zambian<br />
Children: An analysis of the risk Factors.”<br />
American Journal of Tropical Medicine <strong>and</strong><br />
Hygiene. 59(3). Pp. 435-437.<br />
Nemec, J. <strong>and</strong> Schaake, J. 1982. Sensitivity of<br />
water resource systems to climate variation.<br />
Journal of Hydrologic Sciences, 27:327-343.<br />
New, M.; Hulme, M.; <strong>and</strong> Jones, P. 1997.<br />
Representing twentieth-century space-time<br />
climate variability. Part I: development of a<br />
1961-90 mean monthly terrestrial<br />
climatology. Journal of Climate, 12(3): 829-<br />
856.<br />
Priestly, C.H. <strong>and</strong> Taylor, R.J. 1972. On the<br />
assessment of surface heat flux <strong>and</strong><br />
evaporation using large-scale parameters.<br />
Monthly Weather Review, 100: 81-92.<br />
S<strong>and</strong>erson, W. 2001a. The Mixed Blessing of<br />
Antiretroviral Treatment in Botswana. AIDS<br />
Analysis Africa, vol. 11.<br />
S<strong>and</strong>erson, W. 2001b The “How Many Batswana<br />
Will Be Alive in 2021?. APC-EU Courier,<br />
No. 188, September-October. pp. 42-43.<br />
S<strong>and</strong>erson, W. 2002a. The Demographic Impact<br />
of HIV Medication Programs:<br />
With Examples from Botswana. Paper presented<br />
at the Population Association of America<br />
Meetings, May, 9th. Atlanta, Georgia.<br />
S<strong>and</strong>erson W. 2002b. Death, Diamonds, <strong>and</strong> the<br />
Botswana Paradox. Paper presented at the<br />
Population Association of America<br />
Meetings, May 9th 2002, Atlanta, Georgia<br />
S<strong>and</strong>erson, forthcoming 2004. Interactions<br />
between Education <strong>and</strong> HIV: Demographic<br />
examples from Botswana. Chapter 7.<br />
Wolfgang Lutz <strong>and</strong> Warren S<strong>and</strong>erson, eds.<br />
The endof world population ggrowth in the<br />
21st century: New challenges for human<br />
capital formation in sustainable<br />
development. Earthscan Publications.<br />
Schaffner, B.; Chen, Y.Y.; Kelepile, T.;<br />
Kinzelbach W.; Carlsson, L.; Mannathoko.<br />
I.. 2000. Scientific Support for Sustainable<br />
Groundwater Management: A <strong>Modelling</strong><br />
Study in Botswana Using <strong>Environmental</strong><br />
Tracers. Proceedings of the 30th Congress of<br />
IAH, Capetown<br />
SMEC, 1991. The Botswana National Water<br />
Master Plan. Prepared for the Department of<br />
Water Affairs. Gaborone, Botswana. 1-12.<br />
Stockholm <strong>Environmental</strong> Institute. 1994. Water<br />
Evaluation <strong>and</strong> Planning System (WEAP)<br />
model. Tellus Institute. Boston, MA.<br />
Timbo, B.B.; Tollefson, L. 1994. Nutrition: a<br />
cofactor in HIV disease. Journal of American<br />
Dietary Association, 94(9):1018-22.<br />
WHO/UNICEF. 2001. WHO/UNICEF Joint<br />
Monitoring Programme for Water Supply<br />
<strong>and</strong> Sanitation Coverage Estimates 1980-<br />
2000: Access to Improved Sanitation.<br />
Botswana.<br />
801
Yates, D.N. 1996. WatBal: an integrated water<br />
balance model for climate impact assessment<br />
of river basin runoff. Water Resources<br />
Development. 12(2):121-139.<br />
Yates, D. N.; Strzepek, K.M. 1996. Economic<br />
<strong>and</strong> Social Adaptations to Climate Change<br />
Impacts on Water Resources: A Case Study<br />
of Egypt . Journal of Water Resources<br />
Development, Vol. 12, No. 2: 229 – 244.<br />
Yates, D.; Gangopadhyay, S; Rajagopalan, B;<br />
Strzepek, K. . 2003. A technique for<br />
generating regional climate scenarios using a<br />
nearest-neighbor algorithm. Water Resources<br />
Research, Vol. 39, No. 7.<br />
802
<strong>Modelling</strong> Biocomplexity in the Tisza River Basin<br />
within a Participatory Adaptive Framework<br />
J. Sendzimir a , P. Balogh. b , A. Vári c<br />
a<br />
<strong>International</strong> Institute for Applied Systems Analysis, Laxenburg, Austria<br />
b<br />
Village of Nagykörü, Hungary<br />
c<br />
Hungarian Academy of Sciences, Budapest<br />
Abstract: The erosion of biocomplexity in the Tisza River Basin developed slowly <strong>and</strong> incrementally<br />
over the past 130 years since implementation of the original Vásárhelyi river engineering plan. The<br />
Hungarian public view, blinded by flood <strong>and</strong> toxic spill catastrophes, missed the slow <strong>and</strong> subtle<br />
changes to natural, social <strong>and</strong> human capital precipitated by the reshaping of the TRB l<strong>and</strong>scape <strong>and</strong> its<br />
agriculture for flood defence <strong>and</strong> grain production. While conversion of the TRB from a fruit/nut/<br />
fishery polyculture to a wheat monoculture produced a great deal of financial capital for an aristocratic<br />
minority, the gradual drain of alternatives forms of capital left the region less <strong>and</strong> less resilient in the<br />
face of ecological (floods), economic (globalization) <strong>and</strong> political (war) shocks. Domination by central<br />
authorities over the past 50 years reduced local civic capacity to levels of passivity that make most<br />
communities incapable of innovating to find sustainability solutions, <strong>and</strong> this trend is reinforced by ongoing<br />
paternalistic attitudes in the Hungarian national government. Poverty, passivity, apathy <strong>and</strong> the<br />
severe consequences of failure in the event of flooding have severely reduced Adaptive Capacity, the<br />
potential to innovate <strong>and</strong> adapt to uncertainty. Both Nature <strong>and</strong> Society have evolved considerably<br />
since 1870, so simple reverse engineering futilely aims to resurrect a system that no longer exists. Since<br />
the knowledge to un-straighten <strong>and</strong> reflood a river basin is in its infancy, we must learn as we go along,<br />
humble in the knowledge that management interventions often only increase uncertainty <strong>and</strong> can push<br />
the system further into a degraded state. This paper describes an initiative to use conceptual <strong>and</strong> formal<br />
modelling within an Adaptive Management framework to facilitate a regional discussion on how to<br />
manage the TRB while inventing a pathway back to a more resilient socio-ecosystem, linking natural<br />
<strong>and</strong> social processes.<br />
Keywords: Adaptive management, Vulnerability, Resilience, system dynamics models<br />
1. INTRODUCTION<br />
Managing a river basin is less certain than it<br />
was a century ago when flooding was the<br />
prime concern <strong>and</strong> engineering the solution.<br />
Rising damage trends witness the repeated<br />
failure of flood control, but parallel crises with<br />
river valley economic, social <strong>and</strong> cultural<br />
assets reveal a deeper, more entangled<br />
dilemma. Biocomplexity is an attempt to<br />
convey the uncertainty emerging not only from<br />
complex interactions within these sectors, but<br />
also from the tangle of relations across<br />
ecological, economic <strong>and</strong> socio-political<br />
domains. The challenge to underst<strong>and</strong> <strong>and</strong><br />
manage biocomplexity emerges in a history of<br />
surprising reversals of initial policy success,<br />
“policy resistance” (Sterman 2000, 2002).<br />
Attempts to eliminate, at first, <strong>and</strong> then to<br />
merely control disturbances (flood, fire, pests)<br />
have only promoted larger <strong>and</strong> more profound<br />
disturbances. Stubborn resistance to most<br />
policy remedies has earned such problems the<br />
title of “wicked problems” (Rittel <strong>and</strong> Webber<br />
1973), as if evil intention is a metaphor for<br />
how intractable, unknowable <strong>and</strong><br />
uncooperative the world is.<br />
Wicked policy resistance has become<br />
increasingly evident in Tisza River Basin<br />
(TRB) as rising flood crest trends overtop<br />
every effort to raise <strong>and</strong> fortify the dikes, <strong>and</strong><br />
regional agriculture <strong>and</strong> communities struggle<br />
to hold on (Sendzimir et al. 2004). Blame for<br />
rising flood statistics or declining river valley<br />
economies <strong>and</strong> societies cannot simply be<br />
pinned on “the usual suspects”: exogenous<br />
drivers or ignorant human actors or policies.<br />
Analysis of the underlying complexity<br />
continues to improve (Linerooth-Bayer <strong>and</strong><br />
Vári 2003, Molnar 2003, Sendzimir et al.<br />
2004), but underst<strong>and</strong>ing, <strong>and</strong> more<br />
importantly the capacity to adapt, remains<br />
woefully behind the evolving reality. The<br />
move from the “hard” <strong>and</strong> narrow technical<br />
approach to a more adaptive <strong>and</strong><br />
comprehensive “soft” path (Gleich 2001)<br />
803
equires not so much better underst<strong>and</strong>ing or<br />
methods of analysis or management<br />
intervention, but their integration.<br />
Adaptive management offers a framework to<br />
integrate research, policy <strong>and</strong> local practice<br />
into a structured learning cycle (Walters 1985,<br />
Gunderson et al. 1995, 2002). Research,<br />
policy <strong>and</strong> public debate have been meshed<br />
with some success in AM-inspired initiatives<br />
to renovate the Kissimmee (Light <strong>and</strong> Blann<br />
2000) <strong>and</strong> Colorado rivers (Walters et al.<br />
2000). As with the TRB, the historical causes<br />
<strong>and</strong> resultant problems were far better, if<br />
incompletely, understood than the pathway<br />
back to a resilient system. Especially in the<br />
case of the Kissimmee river, the AM approach<br />
allowed managers to invent such a pathway by<br />
integrating stakeholder education <strong>and</strong> feedback<br />
with pilot research projects in the floodplain<br />
with computer modeling simulations of<br />
different policy implementations. This paper<br />
describes an initiative to use modeling within<br />
an AM framework to facilitate a regional<br />
discussion on how to manage the TRB while<br />
inventing a pathway back to a more resilient<br />
socio-ecosystem, linking natural <strong>and</strong> social<br />
processes. The search for new approaches<br />
arises out of frustration with failure of decades<br />
of research to generate concrete means to stem<br />
the rising trends of flooding <strong>and</strong> socioeconomic<br />
decline. The TRB initiative begins<br />
from the practical perspective that ecological<br />
rejuvenation of ecological structure <strong>and</strong><br />
function in the floodplain must also open<br />
opportunities for local employment <strong>and</strong><br />
income. Concrete steps are already evident in<br />
uniquely parallel pilot studies of ecology <strong>and</strong><br />
traditional forms of agriculture <strong>and</strong> fisheries in<br />
a re-flooded floodplain, but the challenge is to<br />
integrate such field research with on-going<br />
efforts to formulate policy, develop commerce<br />
<strong>and</strong> enterprise, <strong>and</strong> improve practices <strong>and</strong><br />
methods at scales ranging from local to<br />
provincial to national to continental. Herein we<br />
describe these challenges <strong>and</strong> our efforts to<br />
model them as a prelude to launching a basinwide<br />
AM effort to increase the TRB’s<br />
resilience in the face of uncertainty.<br />
1.1 Motivation<br />
Parallel crises seem to reinforce one another in<br />
a downward spiral that increases the<br />
vulnerability of the TRB to disturbance from<br />
climate, globalization, <strong>and</strong> centralization of<br />
power in Hungary (Linerooth-Bayer <strong>and</strong> Vári<br />
2003, Molnar 2003, Sendzimir et al. 2004).<br />
Efforts to control variability in river dynamics<br />
through more intensive <strong>and</strong> expensive forms of<br />
management continue to mount in cost as<br />
flooding increases in frequency <strong>and</strong> intensity<br />
(Horvath et al. 2001). Chronic <strong>and</strong> mounting<br />
crises suggest that intense management is<br />
misdirected due to inadequate underst<strong>and</strong>ing<br />
that is not keeping pace with changes from<br />
multiple sources of uncertainty at multiple<br />
scales (Sendzimir et al. 2004). The imperative<br />
to prevent injury, death <strong>and</strong> economic<br />
devastation hampers efforts to explore <strong>and</strong><br />
learn. This raises the challenge to control even<br />
as we explore, to manage as we learn <strong>and</strong> to<br />
counterpose management actions <strong>and</strong> research<br />
in a cycle such that they reinforce one another<br />
in a progressive series that spirals upward to<br />
greater resilience. The challenge requires that<br />
different factors evolve <strong>and</strong> complement one<br />
another across the whole basin. Our ability to<br />
innovate <strong>and</strong> adapt to uncertainty (Adaptive<br />
Capacity sensu Walker et al. 2002, Yohe <strong>and</strong><br />
Tol 2002) has to increase by riding a wave of<br />
trust <strong>and</strong> confidence that comes as our<br />
interventions lower vulnerability <strong>and</strong> increase<br />
resilience to uncertainty. In brief, the<br />
evolutionary challenge is summed by the<br />
question - How can we increase adaptive<br />
capacity as we manage to lower vulnerability<br />
such that our management approaches become<br />
more adaptive? It may mean short-term<br />
excursions into lowered resilience to cross to<br />
another, less vulnerable <strong>and</strong> more resilient,<br />
stability domain.<br />
1.2 Study Area – Hungarian Reach of<br />
the Tisza River Basin<br />
1.2.1 Historical challenges<br />
Starting in the Ukrainian Carpathian<br />
mountains, the Tisza river cuts through<br />
Romania <strong>and</strong> across the great Hungarian plain<br />
(Alföld), eventually issuing into the Danube<br />
river in the Serbian Republic (Figure 1). The<br />
combination of a large mountain catchment<br />
issuing over a short <strong>and</strong> steep outfall onto a<br />
very flat floodplain drives some of the most<br />
sudden (24 hours) <strong>and</strong> extreme water level<br />
fluctuations (12 meters) in Europe (Kovács<br />
2003, Halcrow Group 1999). Such extreme<br />
floods occur on average every 10-12 years in<br />
the Tisza River Basin (Wu 2000), but the last<br />
century has seen rising trends in all facets of<br />
flooding: flood crest or peak height, flood<br />
volume, <strong>and</strong> flooding frequency. Floods have<br />
increased in peak height by an average of 0.35<br />
to 0.73 cm per year in the past fifty years<br />
(Horváth et al. 2001). Since the average<br />
minimal flow has declined, the difference<br />
between flood <strong>and</strong> drought extremes is<br />
increasing. The interval between extreme<br />
floods has declined sharply from once every 18<br />
years (1877 – 1933) to once every 3 to 4 years<br />
804
(1934 – 1964) to almost every other year over<br />
the last decade.<br />
The roots of these increasing flood statistics<br />
may lie in massive river basin engineering that<br />
began with the original Vasarhelyi plan in<br />
1870. In the early phases of the Industrial<br />
Revolution, rising urban populations that<br />
concentrated around factories created an<br />
exploding market for bread in European cities.<br />
The Austrian <strong>and</strong> Hungarian aristocracy seized<br />
this opportunity by modifying the Tisza river<br />
basin morphometry to fit socio-political<br />
dem<strong>and</strong>s for bread production, wheat export,<br />
habitation, <strong>and</strong> flood protection. The river was<br />
deepened to hasten water flow, shortened by<br />
400 km to facilitate export, <strong>and</strong> bracketed with<br />
dykes to prevent flooding of wheat fields <strong>and</strong><br />
habitations. By 1890 Hungary became the first<br />
wheat-exporting nation in Europe. Practically<br />
in step with mounting flood statistics, regional<br />
development has also climbed since the midnineteenth<br />
century, <strong>and</strong> the clash between<br />
these two rising trends has created ever larger<br />
losses. The infrastructure of towns <strong>and</strong> row<br />
crop farms burgeoned <strong>and</strong> spread into the flood<br />
danger zone, the TRB floodplain, reassured by<br />
the apparent security of a dike <strong>and</strong> canal flood<br />
defence system. The security promised by<br />
hydro engineering might hold for a decade or<br />
two, but ever-larger floods breached these<br />
defences, devastating homes, roads <strong>and</strong> crop<br />
fields. Damage to built capital <strong>and</strong> commerce<br />
from one major flood event could reach as high<br />
as approximately 25 percent of the GDP or<br />
riverine basin or 7-9 percent of national GDP<br />
(Halcrow Group 1999). These sudden<br />
catastrophic losses st<strong>and</strong> out against a<br />
backdrop of regional decline in all forms of<br />
capital that contribute to biocomplexity:<br />
natural capital (biodiversity <strong>and</strong> aesthetics lost,<br />
rising flood statistics), economic (previous<br />
industry gone, region no longer prosperous but<br />
empoverished, apathy about farming), <strong>and</strong><br />
socio-political (cities, schools, businesses<br />
disappearing, political apathy as power<br />
concentrates in Budapest) (Sendzimir et al.<br />
2004).<br />
Figure 1. The Tisza river basin with tributaries in catchments in the Carpathian mountain range across<br />
portions of five different national territories (Romania, Ukraine, Slovakia, Federation of Serbia <strong>and</strong><br />
Montenegro, <strong>and</strong> Hungary.<br />
805
1.2.2 Present opportunities<br />
Chronic flooding <strong>and</strong> toxic spill (Kosztolányi<br />
2001) crises have also driven decades of<br />
research (Molnar 2003), but with little concrete<br />
effect on improving any of the facets of<br />
biocomplexity that affect vulnerability or<br />
reslience. Underst<strong>and</strong>ing has increased, but in<br />
sporadic spurts that have not been integrated<br />
<strong>and</strong> have not spread underst<strong>and</strong>ing or<br />
motivated action across disciplines or social<br />
sectors. Recently, however, WWF Hungary<br />
has sponsored a unique research initiative<br />
(Siposs <strong>and</strong> Kiss 2002) that combines analysis<br />
of both ecological functions <strong>and</strong> traditional<br />
agricultural methods in a floodplain with reestablished<br />
hydrological connections.<br />
Underst<strong>and</strong>ing of how ecological <strong>and</strong><br />
agricultural processes could reinforce each<br />
other could take advantage of new<br />
opportunities to trade wheat production to EU<br />
for credits under agri-environmental schemes.<br />
This means that credit gained from ab<strong>and</strong>oning<br />
wheat production could be used to finance the<br />
research, engineering <strong>and</strong> organization to<br />
restore the resilience of the TRB <strong>and</strong> all forms<br />
of capital that compose biocomplexity. This<br />
opportunity raises the issue of how to spread<br />
underst<strong>and</strong>ing <strong>and</strong> trust in these pilot projects<br />
that might motivate wider discussion <strong>and</strong><br />
experimentation that affects the basin as a<br />
whole. We contend that the AM framework<br />
that integrated pilot studies with public<br />
discussion in the Kissimmee river basin of<br />
Florida can serve as a model that we can adapt<br />
here to local conditions.<br />
1.3 Objectives <strong>and</strong> Hypotheses<br />
1.3.1 Objectives<br />
The objectives of this initiative are as follows:<br />
Develop a better underst<strong>and</strong>ing of how key<br />
ecological variables, processes <strong>and</strong><br />
relationships are affected by different flooding<br />
regimes on the Tisza river floodplain; Explore<br />
how the components of biocomplexity<br />
(ecological, economic <strong>and</strong> socio-political<br />
factors) interact to affect the resilience <strong>and</strong><br />
vulnerability of a re-naturalized river<br />
floodplain with greater hydrological<br />
connectivity <strong>and</strong> more frequent flooding;<br />
establish a functional framework, such as<br />
Adaptive Management (AM), that integrates<br />
research <strong>and</strong> policy <strong>and</strong> local practice in a<br />
structured learning cycle; Test various<br />
hypotheses about how a natural flooding<br />
regime affects ecological processes <strong>and</strong><br />
agricultural productivity in pilot projects<br />
prioritized <strong>and</strong> run by participants within an<br />
AM structured learning cycle; Use conceptual<br />
<strong>and</strong> formal modeling as a means to 1. build<br />
trust among collaborators that their separate<br />
experiences are incorporated in a mutually<br />
compelling vision of the key biocomplexity<br />
factors <strong>and</strong> their interactions that affect the<br />
resilience of the TRB; 2. explore the strengths<br />
<strong>and</strong> sensitivities of interactions in order to<br />
prioritize field research as well as the<br />
establishment of economic infrastructure<br />
(marketing <strong>and</strong> sales).<br />
1.3.2 Hypotheses<br />
Confining inquiry within bounds set by a<br />
preliminary set of hypotheses would stifle the<br />
potential of any AM process to incorporate<br />
heretofore-unknown experience <strong>and</strong><br />
knowledge or to derive novel interpretations.<br />
Anticipating that questions, hypotheses <strong>and</strong><br />
predictions will be derived <strong>and</strong>/or shaped by<br />
the AM participatory process (group<br />
assessment to bound the problem <strong>and</strong> derive a<br />
suite of hypotheses that are plausible<br />
alternative views of the key driving factors of<br />
biocomplexity), we pose one overall<br />
hypothesis as starting point for the AM<br />
assessment phase:<br />
Hypothesis: Re-establishment of hydrological<br />
connections across the Tisza river floodplain<br />
will promote nutrient cycling <strong>and</strong> productivity<br />
in a cascade of effects that will build all<br />
component factors of biocomplexity <strong>and</strong> boost<br />
agriculture, biodiversity <strong>and</strong> fisheries <strong>and</strong><br />
lower the region’s vulnerability to extremes of<br />
weather <strong>and</strong> economic variability.<br />
2. METHODS<br />
Two methodological approaches will be<br />
applied to address the need to assess the state<br />
of biocomplexity in the TRB <strong>and</strong> to set<br />
priorities for integrating research with policy<br />
formulation. First, an Adaptive Participatory<br />
Research Framework will be established to<br />
coordinate collaboration between researchers<br />
<strong>and</strong> stakeholders. Second, within the<br />
Framework system dynamics modelling will be<br />
employed to secure a broad underst<strong>and</strong>ing<br />
among all participants of the key variables <strong>and</strong><br />
interactions affecting biocomplexity.<br />
2.1 Adaptive Participatory Research<br />
Framework<br />
Along the TRB increased variability from<br />
climate <strong>and</strong> economic transition only adds to<br />
the uncertainty of a century of biocomplexity<br />
erosion. Coping with uncertainty requires the<br />
sustained capacity to learn <strong>and</strong> to flexibly<br />
manage. For thirty years a decision making<br />
process has been evolving to address the<br />
challenge of learning while managing. This<br />
process, Adaptive <strong>Environmental</strong> Assessment<br />
<strong>and</strong> Management (AEAM), also known as<br />
806
Adaptive Management (AM), offers a<br />
framework to integrate research, policy <strong>and</strong><br />
local practice that has been developed over<br />
three decades of experimental applications to<br />
underst<strong>and</strong> <strong>and</strong> manage crises of collapsed<br />
fisheries, agriculture, forestry <strong>and</strong> rangel<strong>and</strong><br />
grazing (Holling 1978, Walters 1986,<br />
Gunderson et al 1995, Gunderson <strong>and</strong> Holling<br />
2002). AM increases adaptive capacity by<br />
shifting linear decision making processes<br />
(crisis analysis policy) to a cyclic<br />
learning process that iteratively integrates how<br />
we modify conceptualisation, policy<br />
formulation, implementation <strong>and</strong> monitoring in<br />
order to track <strong>and</strong> manage change in the world<br />
(Figure 2).<br />
The TRB initiative attempts to apply<br />
innovations to AM developed in the Oder river<br />
basin (Sendzimir et al. 2003) for communities<br />
with scarce resources of time <strong>and</strong> money. The<br />
innovations aim to lower transaction costs of<br />
determining the composition of the stakeholder<br />
group participating in the AM exercise as well<br />
as the methods <strong>and</strong> ideas best suited to the<br />
question at h<strong>and</strong>. First, the AM process is<br />
conducted on a mini-scale by using only a<br />
h<strong>and</strong>ful of stakeholders (from two to six<br />
people from NGOs <strong>and</strong> government) with<br />
experience broad enough to reasonably convey<br />
the diversity of opinion in the community. The<br />
methods <strong>and</strong> concepts found useful to this<br />
preliminary group can then be applied at the<br />
larger scale of the entire community.<br />
Furthermore, the confidence <strong>and</strong> trust built<br />
within this group can then be extended to<br />
engage a wider segment of the TRB<br />
stakeholders than might have been involved if<br />
the AM framework was naively attempted at<br />
the larger scale to begin with. The second<br />
innovation to sustain <strong>and</strong> intensify stakeholder<br />
involvement throughout the learning cycle is to<br />
engage them in formulating <strong>and</strong> measuring<br />
indicators of progress towards restoration goals<br />
for biocomplexity. The “red thread” that binds<br />
stakeholders in the entire process emerges<br />
from their actions in participating in field<br />
experiments <strong>and</strong> monitoring the very indices<br />
that they themselves proposed as well as from<br />
the progression of ideas <strong>and</strong> model<br />
development within the AM dialogue.<br />
Policy<br />
Formulation<br />
as test of hypothesis<br />
Assessment<br />
Monitoring <strong>and</strong><br />
Evaluation<br />
Management<br />
Action<br />
Policy<br />
Implementation<br />
Figure 2. Adaptive management process as a structured learning cycle that iteratively links four<br />
phases: assessment, formulation, implementation, <strong>and</strong> monitoring.<br />
2.2 <strong>Modelling</strong><br />
The AM learning cycle usually starts with an<br />
Assessment phase where-in stakeholders<br />
explore a range of assumptions <strong>and</strong> ideas in<br />
order to formulate a suite of equally plausible<br />
hypotheses that provide separate predictions of<br />
why the problem in question occurs (Sendzimir<br />
et al. 1999). <strong>Modelling</strong> can serve as a useful<br />
exercise for AM participant stakeholders to<br />
bound the problem <strong>and</strong> examine the key<br />
variables <strong>and</strong> interactions they consider crucial<br />
to the dynamics of resilience <strong>and</strong> vulnerability<br />
in the system. Conceptual models facilitate<br />
discussion <strong>and</strong> comparison of different<br />
interpretations of the system’s structure (which<br />
variables are involved <strong>and</strong> how are they<br />
linked) including identification of reinforcing<br />
<strong>and</strong> balancing feedback loops <strong>and</strong> delays that<br />
affect system dynamics (Sterman 2000).<br />
Formal models involve mathematical<br />
expression of relationships linking key<br />
variables <strong>and</strong> allow participants to explore how<br />
the relative strengths of different interactions<br />
affect system dynamics, particularly with<br />
regard to questions of vulnerability <strong>and</strong><br />
resilience to change. We discuss current<br />
807
progress in application of Conceptual<br />
<strong>Modelling</strong> that is intended eventually to set the<br />
stage for rigorous applications of formal<br />
models<br />
2.2.1 Conceptual <strong>Modelling</strong> – Causal Loop<br />
Diagrams<br />
Following the AM approach used in the Oder<br />
river valley (Sendzimir et al. 2003) NGO<br />
stakeholders <strong>and</strong> systems science researchers<br />
will meet in an initial scoping session to<br />
winnow a list of key variables down to a<br />
practical range (< 25) <strong>and</strong> then use causal loop<br />
diagramming (Sterman 2000) as a discussion<br />
guide in linking variables <strong>and</strong> slowly<br />
developing a graphic image of the system<br />
structure. As the web of relations takes shape,<br />
certain sections become more underst<strong>and</strong>able<br />
as identification of reinforcing <strong>and</strong> balancing<br />
feedback loops reveals the system macrostructure.<br />
The group’s desire to focus on<br />
specific parts of the model often generates submodel<br />
diagrams that clarify some of the causal<br />
details underlying the more aggregate variables<br />
<strong>and</strong> relations in the general model, The TRB<br />
initiative is in the initial stages of mobilizing<br />
the resources to generate a large scale AM<br />
research collaboration that builds on the<br />
research initiative started by WWF in the<br />
Nagykörü region. Thus far conceptual<br />
modelling has helped the organizers to<br />
synthesize an overview vision (Figure 3) of the<br />
key relationships that affect resilience <strong>and</strong><br />
vulnerability of agro-ecosystems in the TRB<br />
floodplain. Preliminary modelling exercises<br />
like these broaden the modellers intuition in<br />
preparation for their facilitating discussion in<br />
group modelling exercises for actors <strong>and</strong><br />
stakeholders in the TRB. The model reveals<br />
the reinforcing feedback loops that trap policy<br />
in flood defence which strangles the<br />
hydrological connectivity that made the region<br />
one of the richest <strong>and</strong> most productive in<br />
Hungary before 1870.<br />
3. CONCLUSIONS AND<br />
RECOMMENDATIONS<br />
Causal loop diagramming has proven a useful<br />
tool to synthesize an initial overview of the<br />
factors <strong>and</strong> relations driving the erosion of<br />
biocomplexity in the TRB <strong>and</strong> will be<br />
improved in a group participatory process that<br />
refines the conceptual models <strong>and</strong> uses them to<br />
build formal models for exploring the relative<br />
strengths with which different interactions<br />
affect system dynamics.<br />
Atmospheric<br />
Temp<br />
Atmospheric<br />
CO2<br />
+<br />
Extreme Drought<br />
Events<br />
+<br />
River<br />
Channel<br />
Engineering<br />
+<br />
River<br />
Velocity<br />
Extreme Rain<br />
Events<br />
+<br />
River bed<br />
erosion<br />
+<br />
+<br />
-<br />
Poltical Pressure for<br />
Flood Control<br />
+<br />
+<br />
Dike Height &<br />
Length<br />
-<br />
+<br />
-<br />
Flooding<br />
+<br />
High water<br />
level +<br />
+ +<br />
-<br />
Local ET Recycle<br />
as Rain<br />
-<br />
Flood Damage<br />
+<br />
Water<br />
contained in<br />
channel<br />
Normal water<br />
level<br />
+<br />
+<br />
-<br />
Silting in<br />
side<br />
channels<br />
Toleration of<br />
Flood<br />
Damage<br />
Human Development<br />
in Floodplain<br />
-<br />
-<br />
Floodplain<br />
elevation +<br />
-<br />
-<br />
Regional +<br />
Income<br />
+<br />
+ +<br />
Wheat<br />
Production<br />
+<br />
'Tree Plantation<br />
Productivity<br />
+<br />
Fish Nursery<br />
Productivity<br />
+<br />
Fishery<br />
Production<br />
+<br />
Orchard<br />
Productivity<br />
+ +<br />
Groundwater<br />
level elevation<br />
+<br />
-<br />
-<br />
-<br />
-<br />
Hydro-connectivity<br />
& Water Storage on<br />
Floodplain<br />
Fruit Tree<br />
Diversity<br />
Figure 3. Conceptual model of key variables <strong>and</strong> causal loops that interact to affect Tisza river<br />
floodplain resilience to climate related hydro-dynamic variability.<br />
+<br />
808
4. ACKNOWLEDGMENTS<br />
This paper benefited from insightful questions<br />
<strong>and</strong> comments from G. Bosch, P.<br />
Magnuszewski, J. Cline.<br />
5. REFERENCES<br />
Gunderson, L.H., Holling, C.S. (2002).<br />
Panarchy: Underst<strong>and</strong>ing Transformations in<br />
Systems of Humans <strong>and</strong> Nature. ed. Isl<strong>and</strong><br />
Press, Washington, D.C.<br />
Gunderson, L.H., Holling, C.S., <strong>and</strong> Light,<br />
S.S., Ed. (1995). Barriers <strong>and</strong> Bridges to the<br />
Renewal of Ecosystems <strong>and</strong> Institutions.<br />
Columbia University Press, New York.<br />
Halcrow Group, Ltd. (1999) Feasibility study<br />
of flood control development in Hungary. No.<br />
Vituki Consult Plc. Budapest. pp. 6<br />
Rittel, H., Webber, M. 1973. "Dilemmas in a<br />
General Theory of Planning." Policy Sciences<br />
4:pp. 155-159<br />
Walters, Carl, Josh Korman, Lawrence E.<br />
Stevens, <strong>and</strong> Barry Gold. 2000. Ecosystem<br />
Modeling for Evaluation of Adaptive<br />
Management Policies in the Gr<strong>and</strong> Canyon.<br />
Conservation Ecology 4(2): 1 [online] URL:<br />
hhttp://www.consecol.org/Journal/vol4/iss2/art<br />
1/index.html<br />
809
Linking Hydrologic Modeling <strong>and</strong> Ecologic Modeling:<br />
An Application of Adaptive Ecosystem Management in<br />
the Everglades Mangrove Zone of Florida Bay<br />
Jon C. Cline a , Jerome J. Lorenz b <strong>and</strong> Eric D. Swain c<br />
a Department of Biology<br />
Case Western Reserve University<br />
10900 Euclid Avenue<br />
Clevel<strong>and</strong>, Ohio 44106-7080, U.S.A.<br />
b National Audubon Society<br />
115 Indian Mound<br />
Tavernier, Florida 33070, U.S.A.<br />
c U.S. Geological Survey<br />
9100 NW 36th Street, STE 107<br />
Miami, Florida 33178, U.S.A.<br />
Abstract: The Across Trophic Levels System Simulator (ATLSS) is a suite of ecological models designed to<br />
assess the impact of changes in hydrology on biotic components of the southern Florida ecosystem. ATLSS<br />
implements a multimodeling approach that utilizes process models for lower trophic levels, structured population<br />
models for middle trophic levels (fish <strong>and</strong> macroinvertebrates), <strong>and</strong> individual-based models for large<br />
consumers. ATLSS requires hydrologic input to assess the effects of alternative proposed restoration scenarios<br />
on trophic structure. An ATLSS model (ALFISH) for functional fish groups in freshwater marshes in the<br />
Everglades of southern Florida has been extended to create a new model (ALFISHES) to evaluate the spatial<br />
<strong>and</strong> temporal patterns of fish density in the resident fish community of the Everglades mangrove zone<br />
of Florida Bay. The ALFISHES model combines field data assessing the impact of salinity on fish biomass<br />
with hydrologic data from the Southern Inl<strong>and</strong> <strong>and</strong> Coastal System (SICS) model. The estuarine l<strong>and</strong>scape<br />
is represented by a grid of 500 × 500-meter cells across the coastal areas of the Florida Bay. Each cell is<br />
divided into two habitat types; flats, which are flooded during the wet season, <strong>and</strong> creeks, which remain wet<br />
<strong>and</strong> serve as refugia during the dry season. Daily predictions of water level <strong>and</strong> salinity are obtained from<br />
the SICS model output, which is resampled at the 500-meter spatial resolution of the ALFISHES model. The<br />
model output may be used to assess the impact of changes in hydrology on fish biomass <strong>and</strong> its availability to<br />
wading bird <strong>and</strong> other consumer populations. With the development of restoration scenario capabilities in the<br />
SICS model, the SICS/ALFISHES coupling should prove an effective tool for evaluating the potential impact<br />
of water management policies on the wading bird population in the Everglades mangrove zone.<br />
Keywords: Everglades; Spatially explicit model; mangrove zone; Fish; Scenario evaluation<br />
1 INTRODUCTION<br />
Wading birds have long been a predominant feature<br />
of the Everglades mangrove zone of Florida<br />
Bay. In particular, the Roseate Spoonbill (Ajaia<br />
ajaja), a key indicator species due to its strong site<br />
fidelity [Lorenz, 2000], has been in decline in recent<br />
years. It has been proposed [Lorenz et al., 2002]<br />
that changes in the natural pattern of water delivery<br />
from the freshwater marshes to the mangrove zone<br />
have played a significant role in the decline of the<br />
local Roseate Spoonbill population, due to reduced<br />
availability of local estuarine fish, its primary food<br />
source.<br />
810
Directly or indirectly, small estuarine fish are an<br />
important food source for many wading birds,<br />
crocodiles, <strong>and</strong> large predatory fish in the southern<br />
Everglades mangrove zone. Changes in hydrology<br />
upstream have increased salinity <strong>and</strong> altered flooding<br />
regimes. A study of the impact of hydrology<br />
on the community of small mangrove fish in Taylor<br />
Slough <strong>and</strong> C-111 basins [Lorenz, 1999] suggests<br />
these changes may have altered the composition of<br />
the resident fish community <strong>and</strong> affected the relative<br />
availability of prey base fish. Thus the ability to link<br />
the predicted hydrology to the ecological response<br />
of fish populations is an important part of evaluating<br />
the effectiveness of water-delivery schemes.<br />
The U. S. Geological Survey has developed two<br />
separate models applicable to the southern Everglades.<br />
The Southern Inl<strong>and</strong> <strong>and</strong> Coastal System<br />
(SICS) model [Swain, 1999; Swain et al., 2004]<br />
is a hydrodynamic surface-water flow model modified<br />
for wetl<strong>and</strong>s application <strong>and</strong> recently coupled<br />
to a ground-water model to account for leakage <strong>and</strong><br />
salinity transfer. The Across Trophic Levels System<br />
Simulator (ATLSS) is a suite of ecological models<br />
designed to assess the impact of changes in hydrology<br />
on biotic components of the southern Florida<br />
l<strong>and</strong>scape [DeAngelis et al., 1998; DeAngelis et al.,<br />
2002]. Both SICS <strong>and</strong> ATLSS are essential parts of<br />
restoration planning in South Florida.<br />
ATLSS implements a multimodeling approach that<br />
utilizes process models for lower trophic levels,<br />
structured population models for functional groups<br />
of fish <strong>and</strong> macroinvertebrates, <strong>and</strong> individual-based<br />
models for large consumers. To simulate the dynamics<br />
of the estuarine fish community, an existing<br />
ATLSS model (ALFISH version 5.0.0) for functional<br />
fish groups in freshwater marshes in the Everglades<br />
(multicolored areas in Figure 1) was extended<br />
to create a new model (ALFISHES) [Cline<br />
<strong>and</strong> Swain, 2002] to evaluate the spatial <strong>and</strong> temporal<br />
patterns of fish density in the Everglades mangrove<br />
zone of Florida Bay.<br />
ALFISHES requires input from a hydrologic model<br />
to assess the effects of alternative proposed restoration<br />
scenarios on trophic structure. The areal distribution<br />
of water depths <strong>and</strong> salinity computed by<br />
SICS is used to drive the various components of AL-<br />
FISHES. This information represents the most complete<br />
application to date of the hydrodynamic <strong>and</strong><br />
transport equations to represent the wetl<strong>and</strong> flow<br />
<strong>and</strong> salinity movement in the coastal area of the<br />
southern Everglades.<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
<br />
Figure 1: Subregions for Fish Model with AL-<br />
FISHES Study Area (ME: Mangrove Estuary;<br />
STS: South Taylor Slough) <strong>and</strong> Field Sites (from<br />
left to right: TR, JB, HC)<br />
2 MODEL DESCRIPTION AND SOLU-<br />
TION METHODS<br />
The restoration of the South Florida Everglades<br />
ecosystems requires linking l<strong>and</strong>scape changes<br />
associated with environmental management with<br />
changes in key biotic components of the l<strong>and</strong>scape<br />
[DeAngelis et al., 2002]. The management process<br />
involves developing scenarios of l<strong>and</strong>scape change,<br />
developing <strong>and</strong> applying a suite of hydrologic <strong>and</strong><br />
ecological models to project the impact of different<br />
scenarios on the Everglades ecosystem, <strong>and</strong> applying<br />
a decision framework analyze model output <strong>and</strong><br />
evaluate management alternatives (see Figure 2).<br />
ALFISHES is designed to utilize the ATLSS modeling<br />
infrastructure to implement individual model<br />
components <strong>and</strong> to integrate these components<br />
into a single framework. The ATLSS modeling<br />
framework combines functionality associated with<br />
traditional Geographic Information System (GIS)<br />
software with an agent-based modeling approach<br />
[Duke-Sylvester <strong>and</strong> Gross, 2002].<br />
The fish model l<strong>and</strong>scape consists of multiple grid<br />
layers including static layers such as vegetation<br />
<br />
811
25°25´<br />
80°50´ 80° 45´<br />
80° 40´<br />
80°35´<br />
80° 30´<br />
80°25´<br />
Main<br />
Park<br />
Road<br />
Taylor<br />
Slough<br />
Bridge<br />
Royal Palm<br />
Ranger<br />
Station<br />
90<br />
Levee 31W<br />
98<br />
Card Sound Road<br />
80<br />
25° 20´<br />
Old Ingraham Highway<br />
70<br />
60<br />
50<br />
Taylor Slough<br />
C-111<br />
West<br />
Highway<br />
Creek<br />
1<br />
East<br />
Highway<br />
Creek<br />
25° 15´<br />
40<br />
Joe Bay<br />
Barnes Sound<br />
30<br />
West Lake<br />
148<br />
25°10´<br />
20<br />
10<br />
Long Lake<br />
FLORIDA BAY<br />
Oregon<br />
140 Creek<br />
Stillwater<br />
Creek<br />
Shell Creek<br />
1<br />
1 10 20 30 40 50 60 70 80 90 100 110 120 130<br />
Alligator Creek<br />
McCormick Creek Taylor River East Creek Mud Creek Trout Creek<br />
Base from U.S. Geological Survey digital data, 1972<br />
Universal Transverse Mercator projection, Zone 17, Datum NAD 27<br />
EXPLANATION<br />
EVERGLADES NATIONAL PARK BUTTONWOOD EMBANKMENT<br />
0 1 2 3 4 5 KILOMETERS<br />
APPROXIMATE AREA OF<br />
LOCATION OF SURFACE CHANNELS<br />
TAYLOR SLOUGH<br />
0 1 2 3 4 5 MILES<br />
BOUNDARY OF SOUTHERN<br />
INLAND AND COASTAL SYSTEMS<br />
(SICS) STUDY AREA<br />
Figure 3: Southern Inl<strong>and</strong> <strong>and</strong> Coastal Systems<br />
(SICS) study area with 305 meter square grid<br />
overlay (USGS, U.S. Geological Survey; NPS,<br />
National Park Service; SFWMD South Florida<br />
Water Management District).<br />
Figure 2: Decision Process for Ecosystem<br />
Restoration (adapted from [DeAngelis et al.,<br />
2002; Pearlstine et al., 2004]).<br />
types <strong>and</strong> topography, combined with dynamic layers<br />
such as hydrology <strong>and</strong> fish biomass. The ATLSS<br />
C++ l<strong>and</strong>scape classes [Duke-Sylvester <strong>and</strong> Gross,<br />
2002], which provide a common interface for manipulating<br />
spatial data, are the primary means of<br />
communicating spatial information between different<br />
model agents. ALFISHES consist of the following<br />
components:<br />
• hydrology component: spatially-explicit time<br />
series of water depth <strong>and</strong> salinity<br />
• l<strong>and</strong>scape component: distribution of marsh<br />
or mangrove habitat within a single l<strong>and</strong>scape<br />
cell<br />
• lower trophic level components: food base for<br />
small fish<br />
• fish component: fish population model<br />
The following subsections describe SICS <strong>and</strong> the<br />
ALFISHES model components in more detail.<br />
2.1 The Hydrology: the SICS Numerical Model<br />
The Southern Inl<strong>and</strong> <strong>and</strong> Coastal Systems (SICS)<br />
model is used to represent the hydrology of the<br />
model area (Swain <strong>and</strong> others, 2003). SICS utilizes<br />
a two-dimensional, dynamic surface-water model,<br />
called SWIFT2D, coupled to a three-dimensional<br />
ground-water model, called SEAWAT. This coupled<br />
model has several features which allow it to produce<br />
an advanced simulation. Both the surface-water<br />
<strong>and</strong> ground-water models simulate salinity transport<br />
<strong>and</strong> the effects on fluid density. The formulations<br />
have been modified to account for wind forcing,<br />
coastal creek flows, evapotranspiration, <strong>and</strong> leakage<br />
between the surface-water <strong>and</strong> ground-water.<br />
The model area is discretized into a 305 meter<br />
square grid as shown in figure 3. The surface-water<br />
model operates on a timestep of 7.5 minutes <strong>and</strong> the<br />
ground-water model has a 1 day timestep. A simulation<br />
period of 7-years, 1996-2002 inclusive, has<br />
been developed <strong>and</strong> verified. For the purpose of<br />
supplying data for ecological models, 1 day averaged<br />
values are output from the simulation.<br />
In order to utilize the SICS model to analyze possible<br />
changes to the system resulting from ecosystem<br />
restoration scenarios, it is necessary to modify<br />
the boundaries of the SICS model to reflect regional<br />
changes to the south Florida hydrologic system.<br />
The boundaries of the SICS model are shown<br />
in figure 4. The boundary modifications are accomplished<br />
by utilizing results from the South Florida<br />
Water Management Model (SFWMM) which represent<br />
the modifications to the hydrologic system proposed<br />
for restoration purposes. The SFWMM is a<br />
much coarser model <strong>and</strong> uses a 2 mile by 2 mile<br />
grid size. Analysis of the SFWMM indicates that<br />
the produced water-levels are more accurate than<br />
the discharges, thus the water-levels produced on<br />
812
80° ´<br />
25°25´<br />
80°50´ 45<br />
80° 40´<br />
80° 35´<br />
80° 30´<br />
80°25´<br />
Main<br />
Park<br />
Road<br />
Taylor<br />
S-175<br />
Slough<br />
Bridge<br />
Taylor Slough<br />
Bridge discharge<br />
27<br />
L-31W<br />
discharge<br />
Levee 31W<br />
25° 20´<br />
Old Ingraham Highway<br />
CY3<br />
CY2<br />
NP46<br />
NP67<br />
C-111<br />
water level<br />
S-18C<br />
1<br />
25° 15´<br />
NMP<br />
(Nine Mile Pond)<br />
Taylor Slough<br />
C-111 discharge<br />
Joe Bay<br />
C-111<br />
S-197<br />
East<br />
Highway<br />
Creek<br />
West<br />
Highway<br />
Creek Long Sound<br />
Barnes Sound<br />
West Lake<br />
Taylor<br />
River<br />
Trout<br />
Creek Culverts beneath<br />
US-1 water level<br />
25°10´<br />
McCormick<br />
Creek<br />
Florida Bay<br />
water level<br />
FLORIDA BAY<br />
Buoy Key<br />
0 1 2 3 4 5 KILOMETERS<br />
2 3 4<br />
0 1 5 MILES<br />
Whipray Basin<br />
Butternut Key<br />
Base from U.S. Geological Survey digital data, 1972<br />
Albers Equal-Area Conic projection, Datum NAD 1983<br />
St<strong>and</strong>ard Parallels 29°30´ <strong>and</strong> 45°30´, central meridian 23°00´<br />
EVERGLADES NATIONAL PARK<br />
APPROXIMATE AREA OF<br />
TAYLOR SLOUGH<br />
NO-FLOW BOUNDARY<br />
BOUNDARY OF SOUTHERN INLAND AND<br />
COASTAL SYSTEMS (SICS) STUDY AREA<br />
EXPLANATION<br />
SPECIFIED WATER-LEVEL MODEL CELL<br />
SPECIFIED FLUX MODEL CELL<br />
OFFSHORE WATER-LEVEL STATION<br />
EVERGLADES NATIONAL PARK WATER LEVEL STATION<br />
COASTAL FLOW AND STAGE STATION<br />
(U.S. GEOLOGICAL SURVEY)<br />
Figure 4: Southern Inl<strong>and</strong> <strong>and</strong> Coastal Systems<br />
(SICS) study area <strong>and</strong> selected data-collection<br />
sites. (USGS, U.S. Geological Survey; NPS is National<br />
Park Service).<br />
Figure 5: Southern Inl<strong>and</strong> <strong>and</strong> Coastal Systems<br />
(SICS) study area with the SFWMM 2 × 2 mile<br />
grid overlay. (USGS, U.S. Geological Survey;<br />
NPS is National Park Service).<br />
the 2 × 2 mile SFWMM grids are interpolated to<br />
develop new boundary data for the SICS model (see<br />
Figure 5).<br />
Replacing the SICS model field-data-produced<br />
boundaries with interpolated water-level values<br />
from the SFWMM model for the base case, produces<br />
very similar results. This indicates that using<br />
the SFWMM boundaries is a valid approach. Several<br />
different restoration scenarios are to be tested<br />
in this manner. A primary scenario is referred to as<br />
D13R, which describes hydrologic conditions that<br />
are expected to exist in the year 2050 if the Comprehensive<br />
Everglades Restoration Plan (CERP) is<br />
implemented. This plan involves removal of canal<br />
sections <strong>and</strong> new hydraulic control structures <strong>and</strong><br />
operating rules.<br />
2.2 The L<strong>and</strong>scape Fish Model: ALFISHES<br />
The model components of ALFISHES are designed<br />
to incorporate the impact of local hydrologic conditions,<br />
including salinity levels, on fish population<br />
dynamics. The basic model architecture <strong>and</strong><br />
behavior is derived from ALFISH. ALFISHES is<br />
designed to mimic the behavior of the ALFISH in<br />
the freshwater marsh habitat in the northern edge of<br />
the SICS/ALFISHES modeling area. Since the AL-<br />
FISHES modeling area straddles a dynamic salinity<br />
gradient that characterizes the estuarine ecotone<br />
between the Everglades freshwater marshes<br />
<strong>and</strong> Florida Bay, salinity plays a significant role in<br />
model dynamics. Along the gradient from freshwa-<br />
Figure 6: SICS vegetation map (adapted from<br />
[Carter et al., 1999] with the SICS <strong>and</strong> AL-<br />
FISHES study areas.<br />
ter marsh to the estuarine mangrove zone, increasing<br />
salinity is associated with changes in the composition<br />
of the habitat structure, the lower trophic<br />
level, <strong>and</strong> the small fish community.<br />
The Hydrology: Linking SICS <strong>and</strong> ALFISHES<br />
to Model Dynamics. ALFISHES requires input<br />
from a hydrologic model, such as SICS, that includes<br />
salinity. In order to process hydrologic output<br />
for ALFISHES, output from SICS simulations<br />
representing different water management scenarios<br />
is archived. A collection of programs utilizing the<br />
ATLSS l<strong>and</strong>scape library was developed to resample<br />
the SICS output at the 500-meter spatial resolution<br />
of the ALFISHES model <strong>and</strong> build spatial data<br />
sets representing the time series of water depths <strong>and</strong><br />
salinity in the model area. These spatial data sets<br />
are used as input for the ALFISHES model.<br />
813
Figure 7: Hydrology of dwarf mangrove creek<br />
habitat. (adapted from Lorenz [2000])<br />
The mangrove zone l<strong>and</strong>scape model is based on<br />
data collected at field sites [Lorenz, 1999] <strong>and</strong> the<br />
static SICS vegetation map [Carter et al., 1999] (see<br />
Figure 6). Each cell in the model l<strong>and</strong>scape represents<br />
a 500 ×500-m cross-section of the Everglades<br />
mangrove zone. The impact of hydrology in the single<br />
cell fish model of DeAngelis et al. [1997] was<br />
captured by dividing the habitat within the cell into<br />
three parts: marsh, pond, <strong>and</strong> solution holes. The<br />
marsh areas reflood periodically, while the ponds<br />
<strong>and</strong> solution holes serve as refugia during periods<br />
of low water.<br />
The field sites are located in dwarf mangrove (Rhizophora<br />
mangle) habitat <strong>and</strong> are characterized by<br />
deep creeks surrounded by flats that are flooded seasonally<br />
[Lorenz, 1999]. When the sites flood, the<br />
fish spread across the flats. The fish either retreat to<br />
refugia (in this case, creeks), or retreat to neighboring<br />
spatial cells, or die, if the cell dries out (Figure<br />
7).<br />
3 MODEL ARCHITECTURE AND IM-<br />
PLEMENTATION<br />
The ATLSS model components are agent-based <strong>and</strong><br />
use a library of C++ l<strong>and</strong>scape classes to model<br />
l<strong>and</strong>scape <strong>and</strong> hydrologic data [Duke-Sylvester <strong>and</strong><br />
Gross, 2002]. The ATLSS l<strong>and</strong>scape class library<br />
provides common interface for manipulating spatial<br />
data, a generic IODevice class for managing data input<br />
<strong>and</strong> output, a generic Metadata class for describing<br />
spatial data <strong>and</strong> other basic data types, <strong>and</strong> support<br />
for manipulating time series of spatial datasets.<br />
More details about the ATLSS approach are available<br />
at http://atlss.org/.<br />
ALFISHES (see Figure 8) incorporates some addi-<br />
Figure 8: Classes of the ALFISHES model<br />
tional features not provided by the original ATLSS<br />
framework: support for XML (eXensible Markup<br />
Language)-base metadata, abstract interfaces for<br />
generic components (e.g. a generic hydrologic<br />
model interface), support for a model repository<br />
allowing dynamic loading of model components<br />
specified by metadata, <strong>and</strong> a CORBA-based clientserver<br />
implementation that combines a Java-based<br />
GUI (Graphical User Interface) for visualizing spatial<br />
data, a C++-based simulation server.<br />
These additional features are provided by SimApp<br />
[Cline et al., 2000], a CORBA-based framework for<br />
spatially-explicit ecological simulations. SimApp is<br />
an object-oriented framework for spatially-explicit<br />
modeling that combines support for implementing<br />
a suite of meta-models in a distributed computing<br />
environment via XML <strong>and</strong> CORBA. This approach<br />
allows for a more modular <strong>and</strong> scalable computing<br />
approach that supports using different applications<br />
in concert for data visualization, data analysis, <strong>and</strong><br />
model computation.<br />
4 CONCLUSIONS AND FUTURE WORK<br />
Prior to the coupling of SICS <strong>and</strong> ALFISHES, the<br />
Everglades mangrove zone had been excluded from<br />
the ATLSS modeling work in support of CERP.<br />
With the incorporation of different water management<br />
scenarios from the SFWMM into the bound-<br />
814
ary conditions for SICS, the combination of SICS<br />
<strong>and</strong> ALFISHES may provide a platform for establishing<br />
the link between water management <strong>and</strong> the<br />
viability of key indicator species such as spoonbills.<br />
The ability to produce reliable projections of both<br />
fish abundances <strong>and</strong> fish availability during the<br />
spoonbill nesting season remains a primary objective<br />
of the modeling effort. The hydrologic <strong>and</strong><br />
l<strong>and</strong>scape components developed for ALFISHES<br />
may be used to facilitate development or refinement<br />
of other models of wildlife populations in the<br />
Everglades mangrove zone such as the American<br />
crocodile (Crocodylus acutus).<br />
ACKNOWLEDGMENTS<br />
Support for this work was provided by the U.S. Geological<br />
Survey, Biological Resources Division <strong>and</strong><br />
Water Resources Division. We particularly appreciate<br />
the advice <strong>and</strong> support of Don DeAngelis.<br />
REFERENCES<br />
Carter, V., N. Rybicki, J. Reel, H. Ruhl, D. Stewart,<br />
<strong>and</strong> J. Jones. Classification of Vegetation<br />
for Surface-Water Flow Models in Taylor Slough,<br />
Everglades National Park. Technical report, U.S.<br />
Geological Survey, 430 National Center, Reston,<br />
VA 20192, December 1999.<br />
Cline, J., F. Guichard, <strong>and</strong> S. A. Levin. SimApp:<br />
an object-oriented framework for spatial simulations<br />
with applications to marine metacommunities.<br />
PISCO/Mellon Symposium, Corvallis, Oregon,<br />
14-20 December, 18 December 2000.<br />
Cline, J. C. <strong>and</strong> E. D. Swain. Linkage of Hydrologic<br />
<strong>and</strong> Ecological Models: SICS <strong>and</strong> AL-<br />
FISHES. on-line, 2002. Second Federal Interagency<br />
Hydrologic Modeling Conference, Las Vegas,<br />
Nevada, July 28 to August 1, 2002.<br />
DeAngelis, D. L., W. F. Loftus, J. C. Trexler, <strong>and</strong><br />
R. E. Ulanowicz. Modeling fish dynamics <strong>and</strong> effects<br />
of stress in a hydrologically pulsed ecosystem.<br />
Journal of Aquatic Ecosystem Stress <strong>and</strong><br />
Recovery, 6:1–13, 1997.<br />
DeAngelis, D. L., S. Bellmund, W. M. Mooij, M. P.<br />
Nott, E. J. Comiskey, L. J. Gross, M. A. Juston,<br />
<strong>and</strong> W. F. Wolff. The Everglades, Florida Bay,<br />
<strong>and</strong> Coral Reefs of the Florida Keys: An ecosystem<br />
sourcebook, chapter Modeling Ecosystem<br />
<strong>and</strong> Population Dynamics on the South Florida<br />
Hydroscape, pages 239–258. CRC Press, Boca<br />
Raton, FL, 2002.<br />
DeAngelis, D., L. Gross, M. Huston, W. Wolff,<br />
D. Fleming, E. Comiskey, <strong>and</strong> S. Sylvester.<br />
L<strong>and</strong>scape modeling for everglades ecosystem<br />
restoration. Ecosystems, 1:64–75, 1998.<br />
Duke-Sylvester, S. M. <strong>and</strong> L. J. Gross. Integrating<br />
spatial data into an agent-based modeling system:<br />
Ideas <strong>and</strong> lessons from the development of the<br />
across-trophic-level system simulation. In Gimblett,<br />
H. R., editor, Integrating Geographic Information<br />
Systems <strong>and</strong> Agent-Based Modeling Techniques<br />
for Simulating Social <strong>and</strong> Ecological Processes,<br />
pages 125–136. Oxford University Press,<br />
Oxford, 2002.<br />
Lorenz, J. J. Impacts of water management on<br />
Roseate Spoonbills <strong>and</strong> their piscine prey in the<br />
coastal wetl<strong>and</strong>s of Florida Bay. PhD thesis, University<br />
of Miami, 2000.<br />
Lorenz, J. J., J. C. Ogden, R. D. Bjork, Powell,<br />
<strong>and</strong> G. V. N. Powell. The Everglades, Florida<br />
Bay, <strong>and</strong> Coral Reefs of the Florida Keys: An<br />
ecosystem sourcebook, chapter Nesting patterns<br />
of Roseate Spoonbills in Florida Bay 1935-1999:<br />
implications of l<strong>and</strong>scape scale anthropogenic<br />
impacts, pages 563–606. CRC Press, Boca Raton,<br />
FL, 2002.<br />
Lorenz, J. J. The response of fishes to physicochemical<br />
changes in the mangrove of northeast<br />
Florida Bay. Estuaries, 22(2B):500–517,<br />
September 1999.<br />
Nutaro, J. Adevs: A discrete event system simulator,<br />
2001. http://www.ece.arizona.edu/˜nutaro/.<br />
OpenGIS Cosortium Inc. Opengis implementation<br />
specification: Grid coverage. www, 2001.<br />
Pearlstine, L., F. Mazzotti, <strong>and</strong> D. DeAngelis. A review:<br />
Spatially explicit decision support sytems<br />
for l<strong>and</strong>scape habitat assessment <strong>and</strong> restoration,<br />
2004. Submitted.<br />
Swain, E. Numerical Representation of Dynamic<br />
Flow <strong>and</strong> Transport at the Everglades/Florida Bay<br />
Interface. Technical report, U.S. Geological Survey,<br />
9100 NW 36th St. #107, Miami, FL 33157,<br />
December 1999.<br />
Swain, E. D., M. A. Wolfert, J. D. Bales, <strong>and</strong><br />
C. R. Goodwin. Two-dimensional hydrodynamic<br />
simulation of surface-water flow <strong>and</strong> transport<br />
to florida bay through the southern inl<strong>and</strong> <strong>and</strong><br />
coastal systems (sics). Water-Resources Investigations<br />
Report 03-4287, U.S. Geological Survey,<br />
Tallahassee, Florida, 2004.<br />
815
On the Local Coexistence of Species in Plant<br />
Communities<br />
J. Yoshimura a,b,c , K. Tainaka a T. Suzuki a <strong>and</strong> M. Shiyomi d<br />
a<br />
Department of Systems Engineering, Shizuoka University, Hamamatsu, 432-8561, Japan<br />
b<br />
Marine Biosystems Research Center, Chiba University, 1 Uchiura, Amatsu-Kominato, 229-5502, Japan<br />
c<br />
Department of <strong>Environmental</strong> <strong>and</strong> Forest Biology, State University of New York College of <strong>Environmental</strong><br />
Science <strong>and</strong> Forestry, Syracuse, New York 13210, USA<br />
d<br />
Faculty of Sciences, Ibaraki University, Mito, 310-8512, Japan<br />
Abstract: Coexistence of many competitive species is very common in natural plant communities. For<br />
example, almost all forests <strong>and</strong> grassl<strong>and</strong>s consist of various species. Extremely high biodiversity is seen in<br />
tropical rain forests. Grassl<strong>and</strong> communities also often consist of many species. In plant communities,<br />
spatially competitive species of plants coexist in a mosaic pattern. Communities with a single species are very<br />
extremely rare in nature. However, mathematical studies show that the local coexistence of spatially<br />
competitive species is rarely achieved even with two competitive species. Many studies have introduced<br />
external factors to promote coexistence, such as immigration of seeds, seed dormancy, spatial heterogeneity<br />
<strong>and</strong> stochastic environments. Certainly coexistence is achieved under some circumstance in these models.<br />
However, we lack the evidence of such external factors in many plant communities. Natural coexistence of<br />
competitive species seems more prevailing than that expected from that with external reasoning. Therefore, it<br />
is reasonable to consider the possibility of internal factors promoting local coexistence of competitive species.<br />
Here we consider a plant community of two spatially competitive species in a lattice environment. We<br />
simulate the competitive interactions between the two species. Unlike the traditional models, we assume that<br />
the competition between the two species induces the replacement/takeover of one species by the other. This<br />
competitive superiority means that the reaction acts like predation in a mathematical context. We show that<br />
such replacement allows the local coexistence of two locally competitive species to some extent. Competitive<br />
interaction may take a various form of mathematical relations in spatially competitive communities. The rarity<br />
of coexistence in previous models may be the artefact of the Lotka-Volterra type competition.<br />
Keywords: Competition; Species diversity; Local coexistence; Lattice modelling; Plant communities<br />
1. INTRODUCTION<br />
Wild communities <strong>and</strong> ecosystems usually consist<br />
of many species [Wilson, 1992; Rosenzweig, 1995;<br />
Peterson et al., 1998]. Communities or ecosystems<br />
with one or few species are very rare in nature.<br />
Coexistence of a large number of species is almost<br />
universal in natural communities <strong>and</strong> ecosystems.<br />
It is well known that food webs can support several<br />
species [May, 1973; Peterson et al 1998].<br />
Species diversity is also high in plant communities,<br />
where plant species are spatially competitive in<br />
nature [Tilman, 1982]. For example, individual<br />
plants in terrestrial plant communities always<br />
compete for light or space to grow, e.g., grassl<strong>and</strong>s<br />
<strong>and</strong> tropical rainforest. Spatial competition is also<br />
seen in animal communities of tidal zones or<br />
aquatic ecosystems. These communities consist of<br />
so many competing species, yet they are coexisting.<br />
Thus communities with many competitive species<br />
are universal in nature [Tilman, 1982; Shiyomi <strong>and</strong><br />
Yoshimura, 2000].<br />
In contrast, almost all theoretical studies imply that<br />
local coexistence of competitive species is a highly<br />
restricted case <strong>and</strong> rarely predicted [MacArthur<br />
<strong>and</strong> Wilson, 1967; Wilson <strong>and</strong> Yoshimura, 1994].<br />
Local coexistence of competitive species may be<br />
achieved in many different mechanisms that allow<br />
animals <strong>and</strong> plants to coexist with high diversity<br />
<strong>and</strong> density. However, local coexistence is rarely<br />
achieved among spatially competitive species.<br />
816
eplacement, local coexistence is impossible in the<br />
current lattice model, as predicted. However, we<br />
show that, by the introduction of replacement<br />
process, local coexistence of competitive plant<br />
species becomes feasible. We discuss the<br />
discrepancy between mathematical theories <strong>and</strong><br />
real ecological interactions. We also discuss the<br />
mechanisms of local coexistence in terms of<br />
ecosystem structure.<br />
Figure 1. A schematic relation of a plant<br />
community of an inferior species X, a superior<br />
species Y <strong>and</strong> vacant site O (b x , b y , m x <strong>and</strong> m y are<br />
birth <strong>and</strong> death rates of X <strong>and</strong> Y, respectively. P is<br />
the replacement/takeover rate of X by Y.).<br />
Local coexistence of plant species seems to be<br />
almost impossible except when some external<br />
maintaining factors such as immigration, spatial<br />
heterogeneity <strong>and</strong> temporal stochasticity<br />
[MacArthur <strong>and</strong> Levins, 1967; Wilson <strong>and</strong><br />
Yoshimura, 1994]. Competitive interactions must<br />
lead to the exclusion of all the inferior species in<br />
plant communities [Harada <strong>and</strong> Iwasa, 1994;<br />
Harada, 1999].<br />
In contrast, many grassl<strong>and</strong> communities show<br />
extremely high diversity of species without strong<br />
external factors [Shiyomi <strong>and</strong> Yoshimura, 2000;<br />
Shiyomi, Takahashi <strong>and</strong> Yoshimura, 2000]. In<br />
some grassl<strong>and</strong> communities, almost no external<br />
factors are detected, but local coexistence of many<br />
species is maintained over many years. Thus we<br />
expect some internal factors promoting local<br />
coexistence of spatially competing species in<br />
grassl<strong>and</strong>s. High species diversity of competitive<br />
species is also found in tropical rainforests <strong>and</strong><br />
aquatic ecosystems [Wilson, 1992].<br />
Recently lattice simulation models have been used<br />
to study the spatial dynamics of communities <strong>and</strong><br />
ecosystems in ecological studies [Tainaka, 1988;<br />
1989; 2003]. Competitive interactions in lattice<br />
models also lead to competitive exclusion of<br />
spatially competing species [Harada <strong>and</strong> Iwasa,<br />
1994; Harada, 1999].<br />
Here we build a lattice model of two plant species<br />
to examine the possibility of local coexistence. The<br />
two plant species compete for space (cell or site in<br />
the lattice) as an exploitive competition. Once the<br />
site is occupied, the other species have no chance<br />
of seed dispersal as in a grassl<strong>and</strong> community.<br />
Unlike the traditional models, we introduce the<br />
third factor: the competitive replacement or<br />
takeover of one species by the other. Without<br />
2. LATTICE MODELL OF COMPETITIVE<br />
COMMUNITIES<br />
2. 1 Model of Competitive Interaction<br />
Here, we consider a simple community composed<br />
of two species X <strong>and</strong> Y in a two-dimensional lattice<br />
habitat. Interactions between species is defined as<br />
follows (Fig. 1):<br />
X + O → 2X<br />
Y + O → 2Y<br />
X + Y → 2Y<br />
mx<br />
X →O<br />
my<br />
bx<br />
by<br />
P<br />
Y →O<br />
(1)<br />
(2)<br />
(3)<br />
(4)<br />
(5)<br />
where X <strong>and</strong> Y means individuals (or covers) of a<br />
plant species with seed reproduction, <strong>and</strong> O, a<br />
vacant site [Tainaka, 1988, 2003]. Each lattice<br />
point is either occupied by a plant (cover) X or Y or<br />
vacant (O). A large individual may cover a few<br />
sites <strong>and</strong> two small plants of a single species may<br />
occupy a site.<br />
The above equations represent respectively,<br />
reproduction of seeds (1) <strong>and</strong> (2), competitive<br />
replacement of X by Y (3), <strong>and</strong> death (4) <strong>and</strong> (5).<br />
The parameters b x <strong>and</strong> b y represent the birth rates<br />
of a X- <strong>and</strong> Y-individual, respectively. The<br />
parameters m x <strong>and</strong> m y represent the death rates of a<br />
X- <strong>and</strong> Y-individual, respectively. In the current<br />
simulation, we kept the death rates identical <strong>and</strong><br />
constant, that is m x = m y = 0.1. We varied the birth<br />
rates of both species to change the competitive<br />
ability of species, since it is determined by the<br />
birth/death ratios [Tainaka, 1988].<br />
The parameter P in Eq. (3) represents the<br />
replacement/invation rate of an X-individual by a<br />
Y-individual. The replacement process [Eq. (3)] is<br />
functionally equivalent to a predator-prey<br />
relationship. When P = 0, the model becomes pure<br />
Lattice Lotka-Volterra competition among species<br />
X <strong>and</strong> Y [Harada <strong>and</strong> Iwasa, 1994; Harada, 1999].<br />
817
(1) Initially, we r<strong>and</strong>omly distribute particles of<br />
two species on a square lattice, where each lattice<br />
site is either vacant (O) or occupied by a single<br />
species X or Y.<br />
(2) Reaction processes are performed in the<br />
following three steps:<br />
A) We perform the reproduction processes<br />
(1) <strong>and</strong> (2). Choose one lattice site<br />
r<strong>and</strong>omly. If the point is occupied by X or Y,<br />
choose one lattice site again r<strong>and</strong>omly. If<br />
this second site is vacant (O), it becomes X<br />
or Y with the probability b x or b y . Here we<br />
employ periodic boundary conditions such<br />
that the edges are connected to the opposite<br />
edges.<br />
B) Next, we perform a one-body reactions<br />
(4) <strong>and</strong> (5). We chose one lattice site<br />
r<strong>and</strong>omly; if the point is occupied by X or Y,<br />
then it becomes O at the death rate m x or m y .<br />
C) We perform a replacement reaction (3).<br />
Chose one lattice site r<strong>and</strong>omly, <strong>and</strong> then<br />
select one of the four nearest neighbour<br />
points (Neumann neighbours). If the two<br />
selected points are one X <strong>and</strong> one Y, X is<br />
replaced by Y with probability P.<br />
(3) Repeat step (2) LL times, where LL is the<br />
total number of square-lattice sites. This step is<br />
called a Monte Carlo step. In this paper, we set L =<br />
100.<br />
(4) Repeat step (3) for 1000-2000 Monte Carlo<br />
steps.<br />
Figure 2. A typical population dynamics of<br />
spatially competitive species with replacement.<br />
The birth rate of Y is varied in A, B <strong>and</strong> C. The<br />
parameters not shown are replacement rate P = 0.4,<br />
<strong>and</strong> mortality rates m x = m y = 0.2.<br />
We carried out a computer simulation. In this<br />
paper, we apply a method of lattice Lotka-Volterra<br />
model (LLVM), similar to a contact process model<br />
[Tainaka, 1988; 1989; 2003]. If replacement<br />
reaction (Equation (3)) has no site specificity, it<br />
becomes a mean-field theory called the Lotka-<br />
Volterra equation. We record the population sizes<br />
of both species X <strong>and</strong> Y.<br />
2.2 Simulation Procedures of Lattice Model<br />
Population dynamics processing of the lattice<br />
model is explained as follows:<br />
3. RESULTS<br />
We carried out simulations for square lattices for<br />
various parameter combinations. We first describe<br />
simulation results where each species has no<br />
interaction (P = 0). Without replacement,<br />
competitive exclusion always takes place. The<br />
species with a higher birth/death ratio always wins<br />
if the initial densities of both species are<br />
sufficiently high. If the ratio of the two species are<br />
identical, the temporal dynamics becomes r<strong>and</strong>om<br />
walk <strong>and</strong> the exclusion takes place in a long run<br />
depending on the total lattice size, <strong>and</strong> the winning<br />
probability is one half. If one species have a<br />
slightly higher ratio, it will exclude the other<br />
species quite rapidly. Thus the competition for<br />
space in the lattice model is quite keen.<br />
When the replacement probability (P > 0) is<br />
introduced, the coexistence of species may be<br />
induced (Fig. 2). In our example, since species Y is<br />
superior, coexistence appears when the birth rate of<br />
Y is smaller than that of X. The steady state<br />
818
ate. This figure shows the distinctive nature of<br />
replacement effects. When the replacement P = 0,<br />
the competition between the two species is<br />
extremely keen <strong>and</strong> the winner of competitive<br />
exclusion is sharply switched at b x = b x = 0.3 (Fig.<br />
3A). However, the replacement is once introduced,<br />
the pattern of steady state densities changes<br />
completely (Fig. 3B-3D).<br />
When P = 0.2, both X <strong>and</strong> Y coexist when the birth<br />
rate of X is sufficiently high (b x > 0.5 in Fig. 3B).<br />
Interestingly, at the same time, the density of Y is<br />
slightly increased with the birth rate of X. This<br />
increase is the effect of replacement reaction.<br />
When the replacement rate is further increased to P<br />
= 0.6, the coexistence is marginally possible when<br />
b x >> 0.95 (Fig. 3C). When it further increased to<br />
P = 1.0, no coexistence becomes possible <strong>and</strong> X is<br />
always eliminated (Fig. 3D). Fig. 3 also shows that<br />
coexistence is possible only in some intermediate<br />
range of P.<br />
Thus, in a lattice environment, stable coexistence<br />
of two species becomes possible under a range of<br />
direct replacement interactions, as long as the two<br />
species are approximately equal in their relative<br />
strengths (Figs. 3B).<br />
The distribution pattern at the steady states shows<br />
the clumping tendencies of X <strong>and</strong> Y (Fig. 4). Both<br />
X <strong>and</strong> Y aggregate compared with the initial<br />
r<strong>and</strong>om distribution. However, the aggregation<br />
tendency is much stronger in Y, probably because Y<br />
chases <strong>and</strong> eats up X individuals.<br />
Figure 3. The steady state densities of X <strong>and</strong> Y<br />
plotted against the X’s birth rate b x with various<br />
replacement rates P. A: P = 0. B: P = 0.2, C: P =<br />
0.6, D: P = 1.0. The Y’s birth rate is constant at b y<br />
= 0.3. The mortality rates m x = m y = 0.2.<br />
densities of both species also depend on the<br />
combinations of the birth rates <strong>and</strong> the replacement<br />
rate (compare Fig. 2A, 2B <strong>and</strong> 2C).<br />
In Figure 3, the steady-state densities of species X<br />
<strong>and</strong> Y are plotted against the birth rate of species X<br />
while keeping the birth rate of Y constant a low<br />
4. DSICUSSIONS<br />
Mathematical <strong>and</strong> simulation studies show that the<br />
coexistence of competing species is not very likely<br />
in general because of competitive exclusions<br />
(MacArthur <strong>and</strong> Wilson 1967; Tilman 1982). For<br />
example, in a Lotka-Volterra competition system,<br />
the coexistence of two species is only possible<br />
when the degree of interspecific density effects is<br />
smaller than that of intraspecific ones. Mutual<br />
exclusion is always a common outcome in many<br />
competitive models.<br />
Competitive exclusion is fierce when the two<br />
species are competing for space, when space is<br />
limited (Fig. 3A). Laboratory experiments on<br />
competition in a closed system also show that<br />
competitive exclusion is a most likely outcome,<br />
e.g., grain beetles, an aquarium, bacterial cultures,<br />
chemostats [Kuwata <strong>and</strong> Miyazaki, 2000]. Our<br />
simulation experiments show that spatial<br />
coexistence is indeed impossible if competition for<br />
space is the Lotka-Volterra type (Fig. 3A).<br />
Therefore, mathematical <strong>and</strong> simulation studies<br />
819
Figure 4. A snapshot of typical stationary patterns<br />
for the lattice competitive communities. Left: the<br />
initial distribution. Right: the distribution pattern at<br />
time step t = 1000. The steady state densities are<br />
about 0.28 for both species. The parameters are b x<br />
= 0.8, b y = 0.1, P = 0.3. The mortality rates m x = m y<br />
= 0.2.<br />
agree well with laboratory experiments. The local<br />
coexistence of competitive species is rarely<br />
possible, as predicted in the previous studies.<br />
In contrast, almost all natural competitive<br />
communities show high species diversity [Wilson<br />
1992; Rosenzweig, 1995; Shiyomi <strong>and</strong> Yoshimura,<br />
2000]. We then face the paradox of local<br />
coexistence of many competitive species. Theory<br />
tells us that most competitive communities should<br />
be pure communities, whilst observation tells us<br />
that almost all plant communities have high species<br />
diversity. There should be some general reasons or<br />
mechanisms for local coexistence of plant species.<br />
Many mathematical <strong>and</strong> simulation models have<br />
been developed <strong>and</strong> explored to show the local<br />
coexistence of plant (or spatially competitive)<br />
species. These models include some forms of<br />
external factors, such as heterogeneity in space <strong>and</strong><br />
time, temporal disturbances or destruction of local<br />
habitats, immigration, dispersal or movements<br />
between isolated patches <strong>and</strong> variations in<br />
microhabitat structure [Tilman, 1988; Rosenzweig,<br />
1995; Peterson et al., 1998]. These models show<br />
some level of local coexistence. However, we find<br />
the evidence of such external factors in plant<br />
communities. For example, isolated grassl<strong>and</strong>s<br />
often consist of many species of grasses <strong>and</strong> herbs.<br />
There is neither indications of a heavy load of<br />
seed immigration nor the variability in habitats<br />
maintaining the stable species components in the<br />
grassl<strong>and</strong> communities [Shiyomi, Takahashi <strong>and</strong><br />
Yoshimura, 2000].<br />
In this paper, we closely examine the nature of<br />
competitive interaction between the plant species.<br />
In grassl<strong>and</strong>s, plant species always compete for<br />
space. Some species is stronger or superior in<br />
competitive ability <strong>and</strong> shading <strong>and</strong> invading the<br />
neighbouring plants. From such a fact, we<br />
introduce the replacement/takeover process<br />
between the two species,. Here one species Y is<br />
superior in competitive ability (replacement<br />
ability). With the replacement process, coexistence<br />
becomes feasible in a relatively wide range of<br />
parameter space (Fig. 2 <strong>and</strong> 3).<br />
The replacement process is functionally equivalent<br />
to prey-predator interaction. In prey-predator<br />
systems, coexistence of several to many species is<br />
generally possible [May, 1973]. This means that<br />
competitive coexistence may be maintained by a<br />
mechanism similar to that of predator-prey<br />
interactions. However, we should note that, in the<br />
current lattice model, the replacement reaction is<br />
weak unlike the predator-prey systems of the most<br />
traditional studies. Interestingly, coexistence tends<br />
to appear when the replacement rates are not in the<br />
extreme values (Fig. 3). Predator-prey<br />
communities is known to have high biodiversity in<br />
oligotrophic environments [Rosenzweig, 1995].<br />
Mathematically such relationships may be seen<br />
among competing plant species in grassl<strong>and</strong><br />
communities.<br />
In grassl<strong>and</strong> communities, the exact distribution of<br />
each plant species are dynamically changing over<br />
the years, but the overall community structure<br />
seems stable with multiple species coexistence<br />
[Tilman <strong>and</strong> Dowing, 1994; Shiyomi <strong>and</strong><br />
Yoshimura, 2000]. Frequent burning may also<br />
820
promote the coexistence by promoting such<br />
replacements [Tilman, 1988]. In the actual<br />
dynamics, temporal invasion of one plant species<br />
by another should be frequent. Here the tradeoff<br />
between reproductive superiority <strong>and</strong> replacement<br />
ability allows the coexistence of species.<br />
5. CONCLUSIONS<br />
We could conclude that local coexistence is<br />
possible even when two plant species are<br />
competing for space. The species diversity in many<br />
plant communities may be functionally different<br />
from that of the strict Lotka-Volterra type. In the<br />
current case, the replacement or takeover<br />
interactions are functionally identical to predatorprey<br />
interactions.<br />
There may be some other types of competitive<br />
interactions that allow the coexistence of plant<br />
species. Our studies imply that the current<br />
mathematical expression of competitive interaction<br />
is not at least adequate for the competitive<br />
interactions in plant communities. Extraordinary<br />
biodiversity in tropical rainforest may be<br />
maintained by different mechanisms. However,<br />
close examinations of species interactions are<br />
necessary to reveal the mechanisms of coexistence<br />
of so many diverse species of trees.<br />
The external factors may be some contributing<br />
factors as in many animal studies [Peterson et al.,<br />
1998]. However, we find no evidence of the<br />
widespread co-occurrence of such factors in most<br />
diverse plant communities. We believe that some<br />
forms of internal mechanisms should be in many<br />
plant communities, as nature is so diverse in life<br />
(Wilson, 1992).<br />
6. ACKNOWLEDGEMENTS<br />
This work was partially supported by a grant-inaids<br />
from the ministry of culture, education <strong>and</strong><br />
sciences in Japan to J. Y. <strong>and</strong> to K. T.<br />
7. REFERENCES<br />
Harada, Y., <strong>and</strong> Y. Iwasa, Lattice population<br />
dynamics for plants with dispersing seeds <strong>and</strong><br />
vegetative propagation. Researches on<br />
Population Ecology 36: 237-249,1994.<br />
Harada, Y., Short vs. long-range disperser: the<br />
evolutionarily stable allocation in a latticestructured<br />
habitat. Journal of Theoretical<br />
Biology 201: 171-187,1999.<br />
Kuwata, A., <strong>and</strong> T. Miyazaki, Effects of<br />
ammonium supply rates on competition<br />
between Microcystis novacekii<br />
(Cyanobacteria) <strong>and</strong> Scenedesmus<br />
quadricauda (Chlorophyta): simulation study,<br />
Ecological <strong>Modelling</strong>, 135, 81-87, 2000.<br />
MacArthur R.H., <strong>and</strong> E.O. Wilson, The Theory of<br />
Isl<strong>and</strong> Biogeography. Princeton University<br />
Press, Princeton, New Jersey,1967.<br />
MacArthur, R.H., <strong>and</strong> R. Levin, The limiting<br />
similarity, convergence, <strong>and</strong> divergence of<br />
coexisting species. American Naturalist 101:<br />
377-385,1967.<br />
May, R.M., Stability <strong>and</strong> Complexity in Model<br />
Ecosystems. Princeton University Press,<br />
Princeton, New Jersey,1973.<br />
Peterson, G., C. R. Allen, <strong>and</strong> C. S. Holling,<br />
Ecological Resilience, Biodiversity, <strong>and</strong> Scale.<br />
Ecosystems, 1: 6-18, 1998.<br />
Rosenzweig, M.L., Species diversity in space <strong>and</strong><br />
time, Cambridge University Press, 1995.<br />
Shiyomi, M., S. Takahashi, <strong>and</strong> J. Yoshimura,<br />
Spatial heterogeneity in grassl<strong>and</strong> community:<br />
analysis by beta-binomial distribution <strong>and</strong><br />
power low. Journal of Vegetation Sciences<br />
11: 627-632,2000.<br />
Shiyomi, M., <strong>and</strong> J. Yoshimura, Measures of<br />
spatial heterogeneity for plant occurrence or<br />
disease incidence with finite-count.<br />
Ecological Research 15: 13-20,2000.<br />
Tainaka, K., Lattice model for the Lotka-Volterra<br />
system. Journal of the Physical Society of<br />
Japan 57: 2588-2590,1988.<br />
Tainaka, K., Stationary pattern of vortices or<br />
strings in biological systems: lattice version of<br />
the Lotka-Volterra model. Physical Review<br />
Letters 63: 2688-2691,1989.<br />
Tainaka, K., Perturbation expansion <strong>and</strong> optimized<br />
death rate in a lattice ecosystem. Ecological<br />
<strong>Modelling</strong> 163: 73-85, 2003.<br />
Tilman, D., Resource Competition <strong>and</strong> Community<br />
Structure. Princeton University Press,<br />
Princeton, New Jersey, 1982.<br />
Tilman, D., Plant Strategies <strong>and</strong> the Dynamics <strong>and</strong><br />
Structure of Plant Communities. Princeton<br />
University Press, Princeton, New Jersey,1988.<br />
Tilman, D., <strong>and</strong> Dowing, J.A., Biodiversity <strong>and</strong><br />
stability in grassl<strong>and</strong>. Nature 367:363-365,<br />
1994.<br />
Wilson D.S., <strong>and</strong> J. Yoshimura, On the coexistence<br />
of specialists <strong>and</strong> generalists. American<br />
Naturalist 144: 692-707,1994.<br />
Wilson, E.O., The Diversity of Life. Harvard<br />
University Press, Cambridge, Massachusetts,<br />
1992.<br />
821
Ecosystems as Evolutionary Complex Systems: A<br />
Synthesis of Two System-Theoretic Approaches Based<br />
on Boolean Networks<br />
Brian D. Fath a <strong>and</strong> W. E. Grant b<br />
a<br />
Biology Department, Towson University, Towson, Maryl<strong>and</strong> 21252, USA<br />
b<br />
Department of Wildlife <strong>and</strong> Fisheries Sciences, Texas A&M University, College Station, Texas, USA<br />
Abstract: Underst<strong>and</strong>ing <strong>and</strong> managing ecosystems as biocomplex wholes is the compelling scientific<br />
challenge of our times. Several different system-theoretic approaches have been proposed to study<br />
biocomplexity <strong>and</strong> two in particular, Kauffman’s NK networks <strong>and</strong> Patten’s ecological network analysis,<br />
have shown promising results. This research investigates the similarities between these two approaches,<br />
which to date have developed separately <strong>and</strong> independently. Kauffman (1993) has demonstrated that<br />
networks of non-equilibrium, open thermodynamic systems can exhibit profound order (subcritical<br />
complexity) or profound chaos (fundamental complexity). He uses Boolean NK networks to describe system<br />
behavior, where N is the number of nodes in the network <strong>and</strong> K the number of connections at each node.<br />
Ecological network analysis uses a different Boolean network approach in that the pair-wise node<br />
interactions in an ecosystem food web are scaled by the throughflow (or storage) to determine the probability<br />
of flow along each pathway in the web. These flow probabilities are used to determine system-wide<br />
properties of ecosystems such as cycling index (Finn 1976), indirect-to-direct effects ratio, <strong>and</strong> synergism.<br />
Here we use a modified version of the NK model to develop a fitness l<strong>and</strong>scape of interacting species <strong>and</strong><br />
calculate how the network analysis properties change as the model’s species coevolve. We find that, of the<br />
parameters considered, network synergism increases modestly during the simulation whereas the other<br />
properties generally decrease. This research is largely a proof of concept test <strong>and</strong> will lay the foundation for<br />
future integration <strong>and</strong> model scenario analysis between two important network techniques.<br />
Keywords: Boolean networks, Coevolution, Ecological Modeling, Fitness l<strong>and</strong>scapes, Network Analysis.<br />
1. INTRODUCTION<br />
One goal of theoretical ecosystem ecology is to<br />
identify <strong>and</strong> quantify system-level concepts <strong>and</strong><br />
find general patterns of ecosystem organization.<br />
One promising method has been to conceptualize<br />
ecosystems as networks connected by their transfer<br />
<strong>and</strong> exchange of energy <strong>and</strong> matter within <strong>and</strong><br />
across system boundaries. Several different<br />
developments of this conceptualization have been<br />
realized. Independently, they have added<br />
significantly to our underst<strong>and</strong>ing of ecosystems<br />
yet there has been a lack of integration with these<br />
methods because of the different terminology,<br />
notation, history, disciplinary genesis, emphasis,<br />
<strong>and</strong> application. The main goal of this project is to<br />
find linkages between two commonly used<br />
Boolean representations of ecological networks.<br />
In particular, we link ecosystem theory based on<br />
network analysis to Kauffman’s theory of selforganized<br />
systems in order to test the hypothesis<br />
that network properties of homogenization,<br />
amplification indirect effects, <strong>and</strong> synergism<br />
increase as an ecosystem co-evolves to higher<br />
fitness levels.<br />
2. BACKGROUND<br />
2. 1 Ecological Network Analysis<br />
Bernard Patten used mathematical systems theory<br />
as a foundation for studying ecosystems (Patten et<br />
al. 1976, Patten 1978, 1981). He stressed the<br />
utility of the inclusion-exclusion principle of set<br />
theory as a way to formalize the transactions that<br />
naturally occur in food webs. A binary interaction<br />
exists in ecological networks, simplified often as a<br />
question of “who eats whom”, but more broadly as<br />
the transfer of conservative energy–matter between<br />
any two entities in the system. Much of the<br />
subsequent work in network ecology builds on this<br />
822
asic premise of direct energy-matter transactions<br />
between coupled binary pairs. These transactions<br />
form the basis of both direct <strong>and</strong> indirect<br />
ecological relations, such as predation (direct),<br />
neutralism (direct), altruism (direct), mutualism<br />
(indirect) <strong>and</strong> competition (indirect) that are of<br />
importance to community ecology. Some of the<br />
primary findings of this research include the<br />
importance of indirect effects as they propagate<br />
through the myriad of network connections<br />
(Higashi <strong>and</strong> Patten 1989) <strong>and</strong> synergism,<br />
individual compartments in an ecosystem gaining<br />
positive value from being embedded in a larger<br />
network (Patten 1992, Fath <strong>and</strong> Patten 1998).<br />
2.2 Ecological Network Properties<br />
Several network properties have been developed<br />
with four in particular: amplification, indirect<br />
effects, homogenization, <strong>and</strong> synergism used most<br />
regularly to investigate ecosystem behavior. Since<br />
they have been described elsewhere, only a brief<br />
description is provided here (see Fath <strong>and</strong> Patten<br />
(1999) for the details). The four properties relate<br />
the distribution <strong>and</strong> contribution of conservative<br />
energy-matter flow through the network’s many<br />
direct <strong>and</strong> indirect pathways. One measure of<br />
resource distribution is given in the direct flow<br />
intensity, or transfer efficiency, matrix G, whose<br />
values, g ij =f ij /T j , represent the likelihood of flow<br />
along a given path, where f ij corresponds to the<br />
flow from compartment j to compartment i, <strong>and</strong><br />
T j =Σ j( i)=0,n f ij is the total sum of flow through<br />
compartment j including input <strong>and</strong> output<br />
boundary flows (f i0 <strong>and</strong> f 0j , respectively). T in out<br />
j =T j<br />
at steady state. In the direct flow intensity matrix,<br />
G, all elements have a non-negative value less<br />
(0¡<br />
than<br />
one g ij 1)<br />
pathways, <strong>and</strong> therefore are always greater than or<br />
equal than the values of G. The G <strong>and</strong> N matrices<br />
are used to define the amplification,<br />
homogenization, <strong>and</strong> indirect effects properties A<br />
specific quantitative test exists to determine each<br />
property (Figure 1).<br />
Amplification occurs whenever an off-diagonal<br />
element of the integral flow matrix is greater than<br />
one (n ij >1). The integral flow from j to i, can<br />
exceed one when cycling drives more than the<br />
equivalent of one unit of input flow over the<br />
pathway. This property was observed in several of<br />
the small-scale models but is rare in large-scale<br />
models (Fath 2004).<br />
Property<br />
Amplification<br />
Homogenization<br />
Synergism<br />
Ratio of direct to<br />
indirect effects<br />
Test<br />
> 1 for i ≠ j<br />
n ij<br />
CV ( G)<br />
> 1<br />
CV ( N)<br />
¢<br />
¢<br />
+ utility<br />
i= 1 j=<br />
1<br />
£¤£<br />
−utility<br />
( n<br />
n<br />
− i<br />
i= 1 j=<br />
1<br />
Figure 1. Four network properties<br />
n<br />
n<br />
£¤£<br />
ij<br />
n<br />
> 1<br />
ij<br />
g<br />
ij<br />
− g<br />
ij<br />
)<br />
> 1<br />
The homogenization property compares the<br />
resource distribution between the direct <strong>and</strong><br />
integral flow intensity matrices. It was observed<br />
that, due to the contribution of indirect pathways,<br />
flow in the integral matrix was more evenly<br />
distributed or more homogenized than that in the<br />
direct matrix, meaning that flow is comprised of<br />
contributions from many parts of the network.<br />
Network homogenization occurs when the<br />
coefficient of variation of N is less than the<br />
coefficient of variation of G because this indicates<br />
that the network flow is more evenly distributed in<br />
the integral matrix.<br />
Indirect effects are calculated as the integral<br />
contributions minus the direct <strong>and</strong> initial boundary<br />
input (Indirect = N–I–G). The indirect to direct<br />
effects ratio is a measure of the relative strength of<br />
these two factors. When the ratio is greater than<br />
one, then indirect effects are greater than direct<br />
effects.<br />
The fourth property, network synergism is based<br />
on a net flow intensity matrix, D, where<br />
d ij =(f ij −f ji )/T i . Unlike the other series in which the<br />
elements are non-negative, entries in D can be<br />
positive or negative (−1¡ d ij
network such as predation, mutualism,<br />
competition, etc. Synergism arises when integral<br />
positive utility exceeds negative utility because of<br />
mutualistic relations in the system <strong>and</strong> is<br />
calculated as the ratio of the magnitude of the<br />
positive <strong>and</strong> negative utilities.<br />
2.3 Kauffman’s NK Model<br />
Stuart Kauffman uses binary Boolean networks to<br />
find general laws of system self-organization<br />
(Kauffman 1993, 1996, 2000). His main thesis is<br />
that biological systems are composed of<br />
autonomous agents, or self-replicating systems that<br />
perform work, which are “co-constructing <strong>and</strong><br />
propagating organization” (Kauffman 2000, p. 5).<br />
An emphasis is placed on co-construction <strong>and</strong><br />
coevolution because of the cybernetic feedback<br />
that makes agents adapt to other agents at the same<br />
time they modify their own environment. There<br />
recently has been renewed interest in the impact<br />
species have on each other <strong>and</strong> on their<br />
environment (e.g., Jones et al. 1997, Odling-Smee<br />
et al. 2003). Coevolution <strong>and</strong> indirect effects are<br />
both manifestations of interacting networks.<br />
In his NK model, Kauffman (2000) addresses<br />
species coevolution by coupling the influence from<br />
genes of one species to genes of another species.<br />
The basic module of the NK model represents an<br />
organism with N genes, each having two alleles, 0<br />
<strong>and</strong> 1. The contribution of each gene to the fitness<br />
of the organism depends on the allele of that gene<br />
<strong>and</strong> the alleles of K other genes in its genome,<br />
called “epistatic” inputs. In this simple model<br />
there are 2 N combinations of alleles that influence<br />
fitness. Each allele combination is r<strong>and</strong>omly<br />
assigned a fitness contribution value. The average<br />
fitness of the N genes is taken as the mean of the<br />
r<strong>and</strong>om values. The result is a fitness l<strong>and</strong>scape,<br />
such that every allele combination has a specific<br />
fitness value (Table 1 shows an example for N=3).<br />
When there is a flip in one allele from 0 to 1 or<br />
vice versa, the fitness contribution of the gene<br />
changes. If the result is higher fitness, then the<br />
allele shift is accepted, if not, then it is not<br />
accepted. Kauffman found that when the number<br />
of connections to other genes, K, is low the system<br />
quickly evolves to a global fitness maximum. As<br />
the number of connections increases there are<br />
more local peaks until the point when the system is<br />
completely interconnected (K=N–1) <strong>and</strong> the<br />
resulting fitness l<strong>and</strong>scape is fully r<strong>and</strong>om. The<br />
more local peaks that occur, the more improbable<br />
it is to “climb” to the global peak, resulting, on<br />
average, in an overall lower fitness. However,<br />
Kauffman maintains that fitness l<strong>and</strong>scapes are not<br />
r<strong>and</strong>om but instead are generated by the<br />
coevolutionary interactions of the various species.<br />
Therefore, the next step is to link NK models of<br />
various species.<br />
Table 1. There are eight possible binary<br />
combinations of 3 genes. Each is assigned a<br />
r<strong>and</strong>om fitness value between 0 <strong>and</strong> 1, <strong>and</strong> the<br />
fitness for each allele combination is the mean of<br />
the three values. This procedure is used to<br />
construct a fitness l<strong>and</strong>scape. For example,<br />
starting with each gene expressing a 0, the fitness<br />
is 0.37. If the allele on the first gene flips to “1”<br />
then fitness increases to 0.43. This simple model<br />
has only one fitness peak at (0,1,0).<br />
1 2 3 fitness<br />
value<br />
fitness<br />
value<br />
fitness<br />
value<br />
w 1<br />
w 1<br />
w 3<br />
Average<br />
fitness<br />
w<br />
0 0 0 0.2 0.5 0.4 0.37<br />
0 0 1 0.7 0.1 0.2 0.33<br />
0 1 0 0.5 0.9 0.8 0.73<br />
0 1 1 0.3 0.3 0.1 0.23<br />
1 0 0 0.5 0.4 0.4 0.43<br />
1 0 1 0.1 0.5 0.3 0.30<br />
1 1 0 0.9 0.2 0.8 0.63<br />
1 1 1 0.6 0.8 0.4 0.60<br />
In the multi-species version of Kauffman’s NK<br />
model, the fitness value of each allele depends not<br />
only on the allele of that gene <strong>and</strong> on the alleles of<br />
K epistatic genes, but also on the alleles of C other<br />
genes in each of S other species. If there are two<br />
species coupled together, then each gene has K+C<br />
inputs, <strong>and</strong> a table of r<strong>and</strong>om fitness contributions<br />
is generated that has 2 (K+C) combinations. A model<br />
in which each species is connected with S other<br />
species has 2 (K+CS) possible states, so the number of<br />
possible states grows combinatorically. The<br />
fitness of the species is calculated as the mean of<br />
the fitness values of the alleles in its current<br />
genotype; each species is assumed to be isogenic.<br />
Now, when one species evolves (a flipping of an<br />
allele on a gene) this likely has ramifications for<br />
the other species by deforming the overall fitness<br />
l<strong>and</strong>scape. Kauffman found that in general<br />
coevolving systems coupled in this manner behave<br />
either in an ordered or chaotic regime, separated<br />
by a phase transition depending on the number of<br />
couplings.<br />
We have recreated Kauffman’s multi-species NK<br />
model here to investigate the fitness of coevolving<br />
species with a particular interest in underst<strong>and</strong>ing<br />
how ecosystem properties may be affected by the<br />
resultant coevolutionary processes. A few<br />
modifications to the original model as presented<br />
above are noted. Each time step during the<br />
simulation, any one of four events, r<strong>and</strong>omly<br />
chosen, may happen. (1) A r<strong>and</strong>omly-chosen<br />
species may evolve to a new genotype via<br />
824
ecombination, if the r<strong>and</strong>omly-chosen new<br />
genotype has a higher fitness value than the current<br />
genotype. A r<strong>and</strong>omly-chosen species may be<br />
replaced by a new species that (2) may have a<br />
different K than the current species, but has the<br />
same C <strong>and</strong> S, (3) may have a different C, but has<br />
the same K <strong>and</strong> S, or (4) may have a different S,<br />
but has the same K <strong>and</strong> C, if the new species has a<br />
higher fitness value than the current species. Thus,<br />
as species evolve or are replaced by invading<br />
species, they change their own fitness l<strong>and</strong>scape<br />
(Kauffman 1996, 2000) as well as the fitness<br />
l<strong>and</strong>scape of the other species. The above<br />
restrictions could be relaxed in future research to<br />
study more general cases, but for now the model<br />
was used to generate a time series of connectance<br />
matrices. We apply ecological network analysis to<br />
each matrix. Eventually, it would be useful to look<br />
at models that have more realistic ecosystem<br />
structures by using the methodologies developed<br />
in Fath (2004) or perhaps to see if over time<br />
species in the models naturally evolve into a<br />
configuration similar to a trophic structure.<br />
However, that is beyond the scope of this paper.<br />
Here we present the initial results from this<br />
research, which uses a five species model<br />
coevolving for 10 time steps under two different<br />
species coupling regimes.<br />
3. INTEGRATED MODEL<br />
In the first simulation, all species were initialized<br />
with S=1 (i.e., each species is connected with one<br />
other species), <strong>and</strong> in the second simulation all<br />
species were initialized with S=4 (connected to all<br />
other species). Every time interval during the<br />
simulation, we generate a connectance matrix<br />
based on the current fitness <strong>and</strong> S values of the set<br />
of species. Elements of the connectance matrix are<br />
equal to 0 if the fitness values of the genes of the<br />
“to” species are not affected by the genes of the<br />
“from” species, <strong>and</strong> diagonal elements are equal to<br />
0, that is, species are not connected to themselves.<br />
Values of the other elements of the connectance<br />
matrix are calculated as the fitness value of the<br />
“to” species divided by its S value, that is, the sum<br />
of all elements “to” a given species is equal to its<br />
fitness value.<br />
The elements of the connectance matrix represent<br />
the fitness contribution among connected species.<br />
In order to apply ecological network analysis to<br />
these matrices, we assume that elements of the<br />
connectance matrix represent relative rates of<br />
energy flow among the set of species. Obviously,<br />
fitness is not flow, but in a more general sense the<br />
fitness represents a measure of influence between<br />
species. The flow probability between two<br />
compartments is the proportion of flow to total<br />
throughflow (g ij = f ij /T j ) where T j is the total<br />
throughflow into compartment j. This could also<br />
be interpreted as the probability of influence<br />
between two compartments (Patten et al. 1976).<br />
Here we assume that the fitness contribution (from<br />
0 to 1) can be used as a measure of the weighted<br />
influence. This allows us to apply ecological<br />
network analysis to each matrix <strong>and</strong> calculate the<br />
cycling index (Finn 1976) as well as the 4<br />
ecological network properties described above.<br />
We then examined the temporal dynamics of these<br />
properties as the set of species co-evolve through<br />
different fitness l<strong>and</strong>scapes to test the hypothesis<br />
that cycling index, homogenization, amplification,<br />
indirect effects, <strong>and</strong> synergism increase as the<br />
ecosystem co-evolves. Note, that ecological<br />
network analysis is a steady-state analysis,<br />
however we treat the model generated from each<br />
time step as a snapshot in time. As the system<br />
changes over time, we can determine the network<br />
properties of the system in that particular state.<br />
One other assumption is needed to run the<br />
analysis, which is that the model ecosystems, as<br />
open systems, receive external input. Energy<br />
enters the system largely through primary producer<br />
<strong>and</strong> lower trophic level species. Usually, for a<br />
model this size (5 compartments) external input<br />
into one compartment is enough, but in some of<br />
these simulations the first compartment is<br />
eliminated after which time there would be no<br />
further input available to higher trophic levels.<br />
Therefore, a unit of input is given to each of the<br />
first two compartments. The other compartments<br />
receive flow from the network of interactions,<br />
which subsequently affects their fitness.<br />
4. NUMERICAL SIMULATION RESULTS<br />
In the first simulation, each species is connected to<br />
one other species. The connectance values can<br />
change at each time step given the occurrence of a<br />
r<strong>and</strong>omly chosen event, as described above (Table<br />
2 shows two matrices generated by the model at<br />
time steps 2 <strong>and</strong> 3). For example, we see that in<br />
the third time step a new species 2 appears which<br />
is also dependent on species 4 <strong>and</strong> the overall<br />
connectance or fitness from species 2 to species 1<br />
increases. Changes such as these continue through<br />
to the end of the simulation after 10 time steps.<br />
When the ecological network properties of these<br />
connectance matrices from each time step are<br />
calculated we find the following: amplification<br />
does not occur at any time step; the cycling index,<br />
homogenization, <strong>and</strong> ratio of indirect to direct<br />
effects all decrease over time; <strong>and</strong> the synergism<br />
parameter rises steadily until a certain point at<br />
which it starts to drop (Figure 2).<br />
825
Table 2. Example of 2 connectance matrices<br />
generated by the first simulation model. Reading<br />
from columns to rows, at time 2, Sp 2 affects Sp 1<br />
(0.52), Sp 3 affects Sp 2 (0.42), Sp 4 affects Sp 3<br />
(0.42), Sp 5 affects Sp 4 (0.66), <strong>and</strong> Sp 1 affects<br />
Sp 5 (0.61). At T=3, Sp 2 is replaced by a new Sp<br />
2 that is affected by Sp 3 <strong>and</strong> Sp 4 (overall fitness<br />
is higher (0.46 versus 0.42). The new Sp 2 also<br />
has caused a change in the fitness value of Sp 1.<br />
T=2 Sp 1 Sp 2 Sp 3 Sp 4 Sp 5<br />
Sp 1 0 0.52 0 0 0<br />
Sp 2 0 0 0.42 0 0<br />
Sp 3 0 0 0 0.42 0<br />
Sp 4 0 0 0 0 0.66<br />
Sp 5 0.61 0 0 0 0<br />
T=3 Sp 1 Sp 2 Sp 3 Sp 4 Sp 5<br />
Sp 1 0 0.66 0 0 0<br />
Sp 2 0 0 0.23 0.23 0<br />
Sp 3 0 0 0 0.42 0<br />
Sp 4 0 0 0 0 0.66<br />
Sp 5 0.61 0 0 0 0<br />
In the second simulation, each species is initially<br />
linked to four other species. Several changes<br />
occur immediately, most notably, the connection<br />
between Sp 4 <strong>and</strong> Sp 5 is lost. During the 10-step<br />
simulation the system becomes more articulated,<br />
meaning there are fewer connections between<br />
species, but these changes would only be accepted<br />
if the overall fitness of the species increases. One<br />
simple measure to consider is the total number of<br />
connections in the system during each time step<br />
(Table 3). We see a similar pattern in the network<br />
parameters in the second simulation as well.<br />
Amplification does not occur at any time step.<br />
Cycling index <strong>and</strong> indirect effects ratio decrease,<br />
while in this simulation homogenization bounces<br />
around but is fairly flat. Synergism also oscillates<br />
reaching a peak in the middle of the simulation <strong>and</strong><br />
dropping again near the end (Figure 3, note in the<br />
figure that synergism is plotted on the alternate y-<br />
axis).<br />
Table 3. Connections in Simulation A (species<br />
initially connected to one species) <strong>and</strong> Simulation<br />
B (species initially connected to four species)<br />
T A: # links B: # links<br />
0 5 20<br />
1 5 19<br />
2 5 19<br />
3 6 19<br />
4 6 19<br />
5 5 19<br />
6 4 19<br />
7 4 18<br />
8 4 18<br />
9 5 18<br />
10 7 14<br />
parameter value<br />
6<br />
5<br />
4<br />
3<br />
2<br />
1<br />
0<br />
0 2 4 6 8 10<br />
time step<br />
Cycling index homogenization synergism i/d ratio<br />
Figure 2. Behavior of network properties over time for first simulation.<br />
826
parameter value<br />
2.5<br />
2<br />
1.5<br />
1<br />
0.5<br />
0<br />
0 2 4 6 8 10<br />
time step<br />
Cycling index homogenization i/d ratio synergism<br />
Figure 3. Behavior of network properties for second simulation. Synergism is plotted on the alternate y-axis.<br />
20<br />
18<br />
16<br />
14<br />
12<br />
10<br />
8<br />
6<br />
4<br />
2<br />
0<br />
5. CONCLUSIONS<br />
In conclusion, we have recreated Kauffman’s<br />
multi-species NK model <strong>and</strong> used it to investigate<br />
the coevolution of a simple model ecosystem.<br />
Furthermore, we have used the fitness values<br />
generated by the model as surrogates for the<br />
probability of influence between the<br />
compartments. This allows the application of<br />
network analysis techniques to determine the<br />
values of specific network properties. In<br />
particular, we found that network synergism<br />
appears to respond positively as fitness increases,<br />
<strong>and</strong> the other properties respond negatively. This<br />
paper represents the first attempt to integrate the<br />
two Boolean techniques; further research is needed<br />
to more deeply underst<strong>and</strong> the interrelation<br />
between them. Future work along these lines is<br />
currently underway, in particular to see how<br />
various network-based ecological goal functions<br />
(Fath et al. 2001) respond to changes in fitness in<br />
these coevolutionary models.<br />
6. REFERENCES<br />
Fath, B.D., Network analysis applied to large-scale<br />
cyber-ecosystems, Ecol. Modell. 171, 329-<br />
337, 2004.<br />
Fath, B.D., <strong>and</strong> B.C. Patten, Network synergism:<br />
emergence of positive relations in ecological<br />
systems. Ecol. Modell. 107, 127-143, 1998.<br />
Fath B.D., <strong>and</strong> B.C. Patten, Review of the<br />
foundations of network environ analysis.<br />
Ecosystems, 2, 167-179, 1999.<br />
Fath B.D., B.C. Patten, <strong>and</strong> J.S. Choi,<br />
Complementarity of ecological goal functions.<br />
J. Theor. Biol., 208(4), 493-506, 2001.<br />
Finn, J.T., Measures of ecosystem structure <strong>and</strong><br />
function derived from flow analysis. J. Theor.<br />
Biol., 56, 363-380, 1976.<br />
Higashi, M., <strong>and</strong> B.C. Patten, Dominance of<br />
indirect causality in ecosystems. Amer. Nat.,<br />
133, 288-302, 1989.<br />
Jones, C.G., J.H. Lawton, <strong>and</strong> M. Shachak,<br />
Positive <strong>and</strong> negative effects of organisms as<br />
physical ecosystem engineers. Ecology 78,<br />
1946-1957, 1997.<br />
Kauffman, S.A., The Origins of Order: Self-<br />
Organization <strong>and</strong> Selection in Evolution.<br />
Oxford University Press, New York, 1993<br />
Kauffman, S.A., At Home in the Universe: The<br />
Search for Laws of Self-Organization <strong>and</strong><br />
Complexity. Oxford University Press, New<br />
York, 1996.<br />
Kauffman, S.A. Investigations. Oxford University<br />
Press, New York, 2000.<br />
Odling-Smee, F.J, K.N, Lal<strong>and</strong>, <strong>and</strong> M.W.<br />
Feldman. Niche Construction: The neglected<br />
process in evoloution. Princeton University<br />
Press, 472 pp. Princeton, 2003.<br />
Patten, B.C., Systems approach to the concept of<br />
environment. Ohio J. Sci., 78, 206-222. 1978.<br />
Patten, B.C., Environs: the superniches of<br />
ecosystems. Amer. Zoo., 21, 845-852, 1981.<br />
Patten, B.C., Energy, emergy <strong>and</strong> environs. Ecol.<br />
Modell. 62, 29-69, 1992.<br />
Patten, B.C., R.W. Bosserman, J.T. Finn, <strong>and</strong><br />
W.G. Cale, Propagation of cause in<br />
ecosystems. In Patten, B.C. (Ed.), Systems<br />
Analysis <strong>and</strong> Simulation in Ecology, Vol. IV.<br />
Academic Press, 593 pp. New York, 457-579,<br />
1976.<br />
827
Benthic Macroinvertebrates <strong>Modelling</strong> Using Artificial<br />
Neural Networks (ANN): Case Study of a Subtropical<br />
Brazilian River<br />
D. Pereira a,c , M. de A. Vitola a , O. C. Pedrollo a , I. C. Junqueira b <strong>and</strong> S. J. De Luca a<br />
a<br />
Hydraulic Research Institut, Federal University of Rio Gr<strong>and</strong>e do Sul, Porto Alegre, Rio Gr<strong>and</strong>e do Sul<br />
State, Brazil.<br />
b<br />
Environment Municipal Department of Porto Alegre.<br />
c<br />
Feevale University Center.<br />
Abstract: Back-propagation Artificial Neural Networks (ANN) were tested with the aim of modelling the<br />
occurrence of benthic macroinvertebrate families in a south Brazilian river. The dataset, consisting of 67 sets<br />
of observations of macroinvertebrate abundance (families Hydrobiidae, Tubificidae, Chironomidae, Baetidae<br />
<strong>and</strong> Leptophlebiidae) <strong>and</strong> water quality variables (pH, temperature, dissolved oxygen, biochemical oxygen<br />
dem<strong>and</strong>, nitrate, phosphate, total solids, turbidity <strong>and</strong> fecal coliforms), was collected at eleven sampling sites<br />
in the Sinos River Basin during 1991-1993. Five different ANN architectures, with one hidden layer <strong>and</strong> 2, 5,<br />
10, 20 <strong>and</strong> 25 neurons were tested. The ANN models were trained using the gradient descendent <strong>and</strong><br />
Levenberg-Marquardt (LM) algorithms, with different combinations of sigmoid transfer functions (log-log,<br />
tan-log, tan-tan). The percentage of success <strong>and</strong> the correlation coefficient were used to choose the best<br />
network architecture for each taxon. The networks with the LM algorithm provided the best predictions of<br />
macroinvertebrate family occurrence, independent of the family’s frequency. The same network architecture<br />
did not always reproduce all the relationships between the taxon occurrence <strong>and</strong> the environmental variables.<br />
The best model, based on a high correlation coefficient among real <strong>and</strong> predicted data <strong>and</strong> a high percentage<br />
of successes, was the ANN for a very common taxon (Hydrobiidae).<br />
Keywords: Macroinvertebrates; Artificial Neural Networks; <strong>Modelling</strong>; Water Quality<br />
1. INTRODUCTION<br />
Artificial Neural Network (ANN) models have<br />
been widely used as a tool for modelling biological<br />
communities in many European countries (Chon,<br />
2000, 2001; Dedecker et al., 2002; Park et al.,<br />
2003). In Brazil, this ecological modelling<br />
approach is less developed due the scarcity of<br />
datasets of the structure <strong>and</strong> function of biological<br />
communities.<br />
The knowledge of taxonomy, structure <strong>and</strong><br />
organization of benthic communities in the south of<br />
Brazil is incipient. Datasets of benthic<br />
macroinvertebrate occurrence in rivers of Rio<br />
Gr<strong>and</strong>e do Sul State (Brazil) were published by<br />
Junqueira (1995), Pereira, (2002) <strong>and</strong> Pereira &<br />
De Luca (2003).<br />
This paper, which uses a back-propagation<br />
algorithmic approach to modelling the family<br />
occurrence of benthic macro invertebrates in a<br />
south Brazilian river based on a dataset obtained<br />
by Junqueira (1995), is, in a Brazilian context, a<br />
pioneering work. The aims of this paper were test<br />
the ability of ANN models to model the<br />
presence/absence of macro invertebrates <strong>and</strong><br />
contribute to the ecological management of Sinos<br />
River Basin.<br />
2. MATERIAL AND METHODS<br />
2.1 Study sites <strong>and</strong> collected data<br />
The Sinos River headwaters is situated at an<br />
altitude of 700 m above the sea level, in Serra do<br />
Mar. The river, from its headwaters to the<br />
confluence with the river Jacuí is 185 km long,<br />
with 65 tributary. The total area of Sinos River<br />
Basin is 4,328 km 2 . In this basin there are 28<br />
municipalities with a total population of 1,185,961<br />
inhabitants. The main forms of water use are for<br />
residential consumption, for industrial processes<br />
828
<strong>and</strong> for agricultural irrigation. The river also serves<br />
as the receiving body for the effluents generated in<br />
the cities.<br />
The data used in this study were collected by<br />
Junqueira (1995) for the purpose of water quality<br />
monitoring at 12 sampling sites in the Sinos River<br />
Basin (period 1991-1993). The dataset consists of<br />
64 sets of observations of benthic<br />
macroinvertebrate (Hydrobiidae, Tubificidae,<br />
Chironomidae, Baetidae <strong>and</strong> Leptophlebiidae)<br />
abundance <strong>and</strong> water quality measurements<br />
(temperature, dissolved oxygen, biochemical<br />
oxygen dem<strong>and</strong>, pH, nitrate, phosphate, turbidity,<br />
total solids <strong>and</strong> fecal coliforms).<br />
25 km<br />
In this study, different neural network architectures<br />
were used based on the supervised training<br />
method, in which the value of the target variables<br />
are know. This training method was based on the<br />
principles of the backpropagation algorithm, which<br />
is the generalization of the learning rule of<br />
Widrow-Hoff for networks with multiple layers<br />
<strong>and</strong> nonlinear differentiable transfer functions<br />
(Rumelhart et al., 1986; Mathworks, 1998).<br />
The backpropagation network normally has three<br />
layers: an input layer, one or more hidden layers,<br />
<strong>and</strong> an output layer, each including one or more<br />
neurons. The figure 2 shows the topology of the<br />
networks used in this study. Each node from one<br />
layer is connected to all nodes in the following<br />
layer, but neither lateral connections within any<br />
layer, nor feed-back connections are used. Nine<br />
input neurons were used, each one represent an<br />
environmental variable, while the output layer<br />
consists of only one neuron indicating the presence<br />
or absence of a macroinvertebrate taxon.<br />
The weights <strong>and</strong> biases should be initialized before<br />
training, normally these values are r<strong>and</strong>om<br />
numbers, the result network should be independent<br />
of the initialization, this indicate the ability<br />
generalization of the network. The network<br />
architectures were train using tow different set of<br />
initial values for weights <strong>and</strong> biases.<br />
Figure 1. Location of monitoring sampling sites in<br />
the Sinos River Basin, Rio Gr<strong>and</strong>e do Sul State,<br />
South Brazil.<br />
2.2. Data processing<br />
All input variables were analyzed for consistency<br />
in order to eliminate any absurd values. The fecal<br />
coliforms data were subjected to a logarithmic<br />
transformation <strong>and</strong> the resulting output data were<br />
converted to represent either presence or absence<br />
(represented by 1 <strong>and</strong> 0 respectively). After this<br />
processing, all input <strong>and</strong> output variables were<br />
rescaled to fall within the interval between –1 <strong>and</strong><br />
1 – although neural networks can deal with input<br />
of different orders of magnitude, this rescaling<br />
leads to more reliable predictions. All variables<br />
were rescaled to be included within the interval<br />
between -1 <strong>and</strong> 1; the MATLAB function was used<br />
for this.<br />
2.3. Artificial Neural Networks<br />
Figure 2. Neural Network topology use in this<br />
study.<br />
Other important parameter to be set before training<br />
is the transfer function, the neural network can use<br />
many different functions; they should be<br />
differentiable only. Two type of sigmoid function<br />
were used in this study: the tangential <strong>and</strong> the<br />
logarithmic sigmoid transfer function.<br />
The training of an ANN consists in adjusting the<br />
weight <strong>and</strong> the biases using an algorithm of back<br />
propagation errors. For each input vector, the<br />
output vector is calculated by the neural network,<br />
829
the error being calculated for the outputs by<br />
comparing the output vector with the “target”.<br />
Using this error, the weights <strong>and</strong> biases are updated<br />
in order to minimize the error. This procedure is<br />
repeated until the errors become small enough or a<br />
predefined maximum number of interactions is<br />
reached.<br />
of success <strong>and</strong> the correlation coefficient among<br />
the real <strong>and</strong> predicted data. These networks were<br />
not sensitive to the initialization. The best results<br />
(Fig. 3) were obtained with a network of five<br />
neurons <strong>and</strong> logsig-tangsig function of activation<br />
with the largest correlation coefficient (r=0.968)<br />
<strong>and</strong> percentage of success (98.51%).<br />
Many algorithms can be used for the training of the<br />
network; most of these are based on optimization<br />
techniques. Two different algorithms were used in<br />
this study: the gradient descendent with variable<br />
learning rate algorithm, which attempts to keep the<br />
learning step size as large as possible while<br />
keeping learning stable, <strong>and</strong> the Levenberg-<br />
Marquardt algorithm, this algorithm appears to be<br />
the fastest method for training moderate-sized<br />
feedforward neural networks.<br />
The validation model was based on the<br />
methodology described by Dececker et al. (2002),<br />
which consists in splitting the data set into<br />
tenfolds. Each tenfold is used for validation while<br />
the others are used for training. This procedure is<br />
repeated until all tenfolds are used for validation.<br />
Figure 3. Comparation of the percentage correctly<br />
classified patterns for Hidrobiidae with the<br />
modified gradient descendent <strong>and</strong> the Levemberg-<br />
Marquardt algorithm with logsig-tansig activation<br />
function.<br />
The neural network models were implemented<br />
using the toolbox of MATLAB 5.3 for MS<br />
Windows TM . Three combinations of transfer<br />
functions were tested: tangential <strong>and</strong> logarithmic<br />
sigmoid transfer functions. For each taxon, two<br />
training algorithms were used: the descendent<br />
gradient algorithm with variable learning rate <strong>and</strong><br />
the Levenberg-Marquardt algorithm. For both<br />
training algorithms, two different initializations<br />
were tested with five different network<br />
architectures with one hidden layer of [2], [5],<br />
[10], [20] e [25] neurons.<br />
3. RESULTS AND DISCUSSION<br />
During the training of the ANNs, the gradient<br />
descendent algorithm didn't present satisfactory<br />
results. Satisfactory results were obtained when the<br />
algorithm of Levenberg-Marquardt was used. This<br />
algorithm resulted in high percentage of success<br />
(86.6-98.5% of the data) <strong>and</strong> correlation<br />
coefficients (r=0.692 <strong>and</strong> 0.968) among real <strong>and</strong><br />
predicted data. The acting of the ANNs<br />
architectures in the modelling of the simulated<br />
families is described below.<br />
Figure 4. Probability of the presence of<br />
Hidrobiidae as a function of dissolved oxygen, by<br />
means of sensitivity analysis.<br />
3.1 Hydrobiidae<br />
This is a very common taxon, occurring in 62% of<br />
the observations. For this taxon the ANNs<br />
presented satisfactory results as for the percentage<br />
Figure 5. Probability of the presence of<br />
Hidrobiidae as a function of dissolved oxygen , by<br />
means of sensitivity analysis<br />
830
The relationship between the occurrence of<br />
Hydrobiidae <strong>and</strong> the nitrate, dissolved oxygen (Fig.<br />
4), DBO 5 (Fig. 5) concentrations, pH <strong>and</strong><br />
temperature, verified through the sensitivity<br />
analysis, demonstrated that this network was<br />
effective in the modelling of this family. As<br />
mentioned in the literature for Brazilian creeks<br />
(Pereira & De Luca, 2003) Hydrobiidae prefer<br />
relatively high levels of dissolved oxygen.<br />
3.2 Tubificidae<br />
This is a moderate taxon, occurring in 32% of the<br />
observations. For this taxon the ANNs was<br />
sensitive to the initialization. A network with two<br />
neurons <strong>and</strong> activation function tangsig-tangsig<br />
presented the largest correlation coefficient<br />
(r=0.825) <strong>and</strong> percentage of success (92.54%).<br />
However, this network didn't reproduce a real<br />
relationship between the tubificidae occurrence <strong>and</strong><br />
the environmental variables. Just a network with<br />
five neurons <strong>and</strong> activation function logsig-tangsig,<br />
with a lower correlation coeficient r=0.692 <strong>and</strong><br />
86.57% of success, has been efficient in modelling<br />
this family (Fig. 6). This network reproduces a real<br />
relationship among the tubificidae occurrence <strong>and</strong><br />
the nitrate, DBO 5 <strong>and</strong> dissolved oxygen<br />
concentrations <strong>and</strong> temperature, as demonstrated in<br />
the sensitivity analysis (Fig. 7 <strong>and</strong> 8). The result<br />
obtained with a network of 20 neurons indicates<br />
unreal relationship among dissolved oxygen, DBO 5<br />
<strong>and</strong> this taxon. It is known that Tubificidae occur<br />
in water with high level of organic material <strong>and</strong><br />
low level of dissolved oxygen.<br />
Figure 6. Comparation of the percentage correctly<br />
classified patterns for Tubificidae with the<br />
modified gradient descent <strong>and</strong> the Levemberg-<br />
Marquardt algorithm with logsig-tansig activation<br />
function.<br />
Figure 7. Probability of the presence of<br />
Turbificidae as a function of dissolved oxygen, by<br />
means of sensitivity analysis.<br />
Figure 8. Probability of the presence of idae<br />
Tubificidae as a function of dissolved oxygen, by<br />
means of sensitivity analysis.<br />
3.3 Chironomidae<br />
This is a very common taxon, occurring in 65% of<br />
the observation. For this taxon the ANN was<br />
insensitive to the initialization. A network with ten<br />
neurons <strong>and</strong> activation function tangsig-tangsig<br />
presented the largest correlation coefficient<br />
(r=0,968) <strong>and</strong> percentage of success (98.51%). The<br />
relationship among the occurrence of<br />
Chironomidae <strong>and</strong> the concentrations of DBO 5 ,<br />
fecal coliforms <strong>and</strong> total solids, was verified<br />
through the sensitivity analysis, indicating that this<br />
network was effective in modelling of this family.<br />
It is known that Chironomidae are resistant to<br />
organic pollution occurring in water with high level<br />
of organic material <strong>and</strong> low level of dissolved<br />
oxygen.<br />
3.4 Leptophlebiidae<br />
This is a rare taxon, occurring in 25% of the<br />
observation. The ANN tested during the modelling<br />
of this family presented satisfactory results when<br />
831
the percentage of success <strong>and</strong> the correlation<br />
coefficient among the real <strong>and</strong> predicted data was<br />
high. The acting of these networks was<br />
independent of the initialization, although it had a<br />
different behavior in the sensitivity analysis. A<br />
network with 20 neurons <strong>and</strong> activation function<br />
logsig-tangsig presented correlation coefficient<br />
(r=0.834) <strong>and</strong> percentage of success (94.03%). The<br />
relation among the occurrence of Leptophlebiidae<br />
<strong>and</strong> the concentrations of DBO 5 , pH <strong>and</strong> turbidity,<br />
was verified by means of sensitivity analysis. The<br />
same network presented the largest correlation<br />
coefficient (r=0.876) <strong>and</strong> number of success<br />
(95.52%), under new initialization. This network<br />
reproduced the relationship among the occurrence<br />
of Leptophlebiidae <strong>and</strong> total solids.<br />
3.5 Baetidae<br />
This is a very common taxon, occurring in 79% of<br />
the observation. For this taxon, the ANN was<br />
sensitive to initialization. A network with 25<br />
neurons <strong>and</strong> activation function tangsig-tangsig<br />
presented the largest correlation coefficient<br />
(r=0.823) <strong>and</strong> percentage of success (94.03%).<br />
This network reproduced only the relationship<br />
between pH <strong>and</strong> this taxon. A network with 20<br />
neurons <strong>and</strong> a function of activation logsig-tangsig,<br />
with a coefficient of correlation r=0.769 <strong>and</strong><br />
91.04% of success, recognized the relationship<br />
between the occurrence of Baetidae <strong>and</strong> the DBO 5 ,<br />
dissolved oxygen concentrations, fecal coliforms<br />
<strong>and</strong> temperature, as demonstrated by the sensitivity<br />
analysis. This family <strong>and</strong> Leptophlebiidae have<br />
been shown to be sensitive to organic pollution in<br />
Brazilian creeks (Pereira, 2002; Pereira & De<br />
Luca, 2003), giving preference to waters with high<br />
level of dissolved oxygen.<br />
4. CONCLUSION<br />
In this paper the modified descendent gradient<br />
algorithm was not appropriate for training or<br />
predicting the presence or absence. The algorithm<br />
of Levemberg-Marquardt was the most effective in<br />
the training <strong>and</strong> predicting the occurrence of<br />
macroinvertebrate families, independent of the<br />
occurrence frequency. According to Dedecker et<br />
al. (2002) the very rare <strong>and</strong> very common taxon<br />
where better modelled by LM <strong>and</strong> moderate taxon<br />
with GDA.<br />
The best model family was a very common taxon<br />
(Hydrobiidae), presenting high correlation<br />
coefficient among real <strong>and</strong> predicted data as well<br />
as percentage of success, reproducing real<br />
relationship with environmental variables.<br />
The ANN that presented larger correlation<br />
coefficient among real <strong>and</strong> predicted data <strong>and</strong> high<br />
percentage of success was not always the best one<br />
to reproduce the relationships between the taxons<br />
occurrence <strong>and</strong> the environmental variables.<br />
Not always the same network architecture<br />
reproduces all the relationships between the taxon<br />
occurrence <strong>and</strong> the environmental variables.<br />
Similar results were obtained by Dedecker et al.<br />
(2002). This could be related to with the size of<br />
dataset, which influences the generalisation ability<br />
of the ANN.<br />
The result could be improved with selection of<br />
more appropriate environmental variables, as<br />
habitat preferences <strong>and</strong> hydrodynamics<br />
characteristics.<br />
The results presented indicate that ANN could be<br />
an appropriate tool to predict the occurrence of<br />
macroinvertebrate families based on environmental<br />
variables. The principal drawback is to build the<br />
best model configuration, which is essential for a<br />
correct ecological application of ANNs for<br />
ecosystem management.<br />
ACKNOWLEDGEMENTS<br />
The first two authors wish to thanks the National<br />
Council of Scientific <strong>and</strong> Technology<br />
Development (CNPq) for the doctoral degree<br />
grants.<br />
REFERENCES<br />
Chon, T.-S.; Park,Y.-S. & Park, J.H. 2000.<br />
Determining temporal pattern of community<br />
dynamics by using unsupervised learning<br />
algorithms. Ecological <strong>Modelling</strong> 132:151–<br />
166.<br />
Chon, T.S.; Kwak, I.S.; Park,Y.S.; Kim, T.H. And<br />
Kim, Y. 2001. Pattern <strong>and</strong> short-term<br />
predictions of macroinvertebrate<br />
community dynamics by using a recurrent<br />
artificial neural network. Ecological<br />
<strong>Modelling</strong> 146:181-193<br />
Dedecker, A.; Goethals, P.; Gabriels, W. & De<br />
Pauw, N. 2002. Optimization of Artificial<br />
Neural Network (ANN) model design for<br />
prediction of macroinvertebrate<br />
communities in the Zwalm river basin<br />
(Fl<strong>and</strong>ers, Belgium). In: Proceedings of<br />
international <strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong><br />
<strong>Software</strong> Society. vol. 2, pag. 142-147,<br />
Lugano, Switzerl<strong>and</strong>, junho de 2002.<br />
832
Junqueira, I. C. 1995. Aplicação de índices<br />
biológicos para a interpretação da qualidade<br />
da água do rio dos Sinos. Porto Alegre,<br />
PUCRS, 115p. (Dissertação).<br />
MATHWORKS. Neural Networks Toolbox User´s<br />
Guide. The MathWorks Inc., Natick, MA,<br />
1998.<br />
Park,Y.S.; P.F.M. Verdonschot; T.S. Chon <strong>and</strong> S.<br />
Lek., Patterning <strong>and</strong> predicting aquatic<br />
macroinvertebrate diversities using articial<br />
neural network, Water Research 37:1749-<br />
1758, 2003.<br />
Pereira, D., Aplicação de índices ambientais para a<br />
avaliação da sub-bacia do arroio Maratá,<br />
bacia do rio Caí (RS, Brasil), Master<br />
Degree Dissertation, Bioscince Institut of<br />
Federal University of Rio Gr<strong>and</strong>e do Sul,<br />
Porto Alegre, 2002.<br />
Pereira, D. & S. J. De Luca, Benthic<br />
macroinvertebrates <strong>and</strong> the quality of the<br />
hydric resources in Maratá Creek basin of<br />
(Rio Gr<strong>and</strong>e do Sul State, Brazil), Acta<br />
Limnologica Brasiliense 15(2): 57-68,<br />
2003.<br />
833
Interspecific Segregation <strong>and</strong> Phase Transition in a<br />
Lattice Ecosystem with Intraspecific Competition<br />
K. Tainaka a , M. Kushida a , Y. Ito a <strong>and</strong> J. Yoshimura a,b,c<br />
a<br />
Department of Systems Engineering, Shizuoka University, Hamamatsu, 432-8561, Japan<br />
b<br />
Marine Biosystems Research Center, Chiba University, 1 Uchiura, Amatsu-Kominato, 229-5502, Japan<br />
c<br />
Department of <strong>Environmental</strong> <strong>and</strong> Forest Biology, State University of New York College of <strong>Environmental</strong><br />
Science <strong>and</strong> Forestry, Syracuse, New York 13210, USA<br />
Abstract: Many empirical studies of ecological community indicate the coexistence of competing species is<br />
extremely common in nature. However, many mathematical studies show that coexistence of competitive<br />
species is not so easy. In the present article, we focus on the segregation of habitat (microhabitat). If habitats<br />
of species are spatially separated, they can coexist easily: under the habitat segregation, net competition does<br />
not work between species. We study a lattice ecosystem composed of two competitive species. The dynamics<br />
of this system is found to be asymptotically stable. In this system both species can coexist, because<br />
intraspecific competition is stronger than interspecific competition. It is found that this system exhibits a<br />
phase transition: if the mortality rate of both species increases, they go extinct. Our main result shows a selforganized<br />
isolation of microhabitat; that is, at the phase transition point, the living regions of both species are<br />
naturally <strong>and</strong> completely separated from each other. In this critical state, each species independently forms<br />
clusters, <strong>and</strong> the shape of each cluster greatly varies with time proceed. Such a phase transition occurs, even<br />
though (i) there is no special condition in space, <strong>and</strong> (ii) the intraspecific competition is stronger than<br />
interspecific competition. We conclude that such segregation comes from an inherent nature of species.<br />
Despite no attraction acts between individuals, each species forms clusters. This conclusion suggests that all<br />
biospecies may have some mechanism that naturally causes the isolation of habitats.<br />
Keywords: Habitat segregation; Lattice model; Competition; Lotka-Volterra model<br />
1. INTRODUCTION<br />
Lattice models are widely applied in the field of<br />
ecology (Nowak, et al., 1994; Harada & Iwasa,<br />
1994; Nakagiri, et al., 2001). Spatial distribution of<br />
individuals usually differs from r<strong>and</strong>omness. In<br />
most cases, individuals of the same species form<br />
clumping patterns; they huddle together. Nonr<strong>and</strong>omness<br />
in spatial distribution influences on<br />
evolutionary argument. In the present paper, we<br />
demonstrate that a habitat segregation naturally<br />
occurs. Although intraspecific competition is<br />
stronger than interspecific one, these species live<br />
separately. Because of segregation, the competition<br />
between species almost disappears.<br />
We deal with a lattice system called “Lotka-<br />
Volterra model (LLVM)” (Tainaka, 1988;<br />
Matsuda, et al., 1992). The simulation method of<br />
this system is similar to that of the so-called Lotka-<br />
Volterra model. The difference between both<br />
simulation methods is very simple. Namely, in the<br />
case of LLVM, interaction is restricted to occur<br />
between adjacent lattice points (local interaction),<br />
whereas in the Lotka-Volterra model interaction<br />
globally occurs between any pair of lattice points<br />
(global interaction). For this reason, the Lotka-<br />
Volterra model is a mean-field theory of LLVM; in<br />
other words, LLVM is a lattice version of Lotka-<br />
Volterra model. The investigation of spatial model<br />
(LLVM) enables us to give useful information for<br />
population dynamics in living systems. Nonr<strong>and</strong>omness<br />
of spatial distribution strongly effects<br />
on the dynamics.<br />
The LLVM model is an extension of the contact<br />
process (CP) which contains a single species. The<br />
CP model, presented by Harris in the field of<br />
mathematics (Harris, 1974), is a lattice version of<br />
the logistic equation. This model has been<br />
834
extensively studied from mathematical (Durret,<br />
1988) <strong>and</strong> physical (Katori & Konno, 1991; Marro<br />
& Dickman, 1999) aspects. The contact process is<br />
defined by birth <strong>and</strong> death processes of a species X<br />
on a lattice space. Each lattice site is either empty<br />
(E) or occupied (X). The site X means an<br />
individual or a sub-population (occupied patch).<br />
Birth <strong>and</strong> death processes are respectively given by<br />
X ⎯ ⎯→ E (death rate: m) (1a)<br />
X + E ⎯ ⎯→ X (reproduction rate: r) (1b)<br />
The processes (1a) <strong>and</strong> (1b) simulate death <strong>and</strong><br />
reproduction, respectively. The reaction (2) occurs<br />
between adjacent lattice sites.<br />
We develop the contact process to deal with<br />
competition between two species. Moreover,<br />
intraspecific competition is also assumed. Hence,<br />
any pair of individuals located in a short distance<br />
compete with each other. It is found that this<br />
system exhibits habitat segregation: living regions<br />
of both species are automatically separated.<br />
Although the intraspecific competition is stronger<br />
than the interspecific one, habitat segregation<br />
occurs.<br />
2. The Model<br />
Consider two competing species X <strong>and</strong> Y. Our<br />
model is defined by<br />
<strong>and</strong><br />
X ⎯ ⎯→ E (rate: m x ) (2a)<br />
X + E ⎯ ⎯→ X (rate: r x ) (2b)<br />
Y ⎯ ⎯→ E (rate: m y ) (2c)<br />
Y + E ⎯ ⎯→ Y (rate: r y ) (2d)<br />
X + X ⎯ ⎯→ X + E (rate: c x ) (2e)<br />
Y + Y ⎯ ⎯→ Y + E (rate: c y ) (2f)<br />
The reactions (2a) - (2d) are the same meaning as<br />
in the contact process. The last two reactions<br />
represent the intraspecific competition. The<br />
parameters c x <strong>and</strong> c y mean competition rates. In<br />
this model, interspecific competition occurs to get<br />
the empty site E.<br />
We describe the simulation method:<br />
1) Initially, we distribute individuals on the square<br />
lattice; the initial distribution is not important,<br />
since the system evolves into a stationary state.<br />
The final equilibrium points are qualitatively the<br />
same, though spatial pattern changes dynamically.<br />
It is an attractor.<br />
2) The reactions (2) are performed in the following<br />
two steps:<br />
(i) we perform two-body reactions (2b), (2d), (2e)<br />
<strong>and</strong> (2f). Choose one lattice site r<strong>and</strong>omly, <strong>and</strong><br />
then r<strong>and</strong>omly specify one of four neighboring<br />
sites. Let the pair react according to two-body<br />
reactions. For example, if the pair of sites are (X,<br />
E) or (E, X), then E is changed into X according to<br />
the reaction (2b).<br />
(ii) we perform one-body reactions (2a) <strong>and</strong> (2c).<br />
Choose one lattice point r<strong>and</strong>omly; if the site is<br />
occupied by X (or Y), the site will become E by<br />
the rate m x (or m y ). In a real simulation, the<br />
maximum mortality max {m} = 2. When m = 2, we<br />
perform mortality reaction twice.<br />
3) Repeat step 2) L x L = 10,000 times, where L x<br />
L is the total number of lattice points. This is the<br />
Monte Carlo step (Tainaka, 1988).<br />
4) Repeat step 3) until the system reaches a<br />
stationary state.<br />
3. Mean-field theory<br />
If the global interaction is allowed between any<br />
pair of lattice sites, the population dynamics of our<br />
system (2) is given by the mean-field theory:<br />
dx<br />
2<br />
= −mx<br />
x + rx<br />
xe − cx<br />
x<br />
(3a)<br />
dt<br />
dy<br />
2<br />
= −my<br />
x + ry<br />
xe − c y x<br />
(3b)<br />
dt<br />
where x, y <strong>and</strong> e are the densities of the sites X, Y,<br />
<strong>and</strong> E, respectively (e=1-x-y). The above equations<br />
can be rewritten by<br />
dx<br />
dt<br />
dy<br />
dt<br />
= R ( K<br />
(4a)<br />
1 x K1<br />
− x − ay)<br />
/<br />
= R ( K<br />
(4b)<br />
2 y K 2 − x − bx)<br />
/<br />
Here the parameters satisfy the following relations:<br />
R<br />
K<br />
= r x − m x<br />
1 , y<br />
r<br />
− m<br />
x x<br />
1 = ,<br />
rx<br />
+ c x<br />
x<br />
x<br />
1<br />
2<br />
R2 = r y − m , (5)<br />
rx<br />
− mx<br />
K 2 = , (6)<br />
r + c<br />
rx<br />
rx<br />
a = , b = (7)<br />
r + c r + c<br />
x<br />
x<br />
x<br />
x<br />
835
The equations (4a) <strong>and</strong> (4b) are called the Lotka-<br />
Volterra model, <strong>and</strong> its result is well known. Final<br />
stationary states are classified into four classes,<br />
depending on the values of parameters: namely, (i)<br />
both X <strong>and</strong> Y coexist, (ii) X only survives, (iii) Y<br />
only survives, <strong>and</strong> (iv) both go extinct. The<br />
condition for the coexistence is given by<br />
K > <strong>and</strong> x K y<br />
x<br />
aK y<br />
This is explicitly expressed by<br />
r ( r − m )<br />
x<br />
y<br />
y<br />
r + c<br />
y<br />
y<br />
< r x − mx<br />
bK < . (8)<br />
( r y −my<br />
)( rx<br />
+ cx)<br />
<<br />
. (9)<br />
r<br />
It is therefore necessary for the coexistence that<br />
intraspecific competition is stronger than<br />
interspecific one; in other words, the competition<br />
rates (c x <strong>and</strong> c y ) should take large values for the<br />
coexistence. If c x =c y =0 , then the condition (9) is not<br />
satisfied.<br />
y<br />
functions mean local densities. Note that they are<br />
scaled by overall densities. In the case of lattice<br />
system, the distance takes discrete values. We can<br />
prove F(r,XY)= F(r,YX). In Fig. 1, we make clear<br />
the meaning of distance r. The shortest distance<br />
(r=1) means the nearest neighbour, <strong>and</strong> r=2 means<br />
the next nearest neighbour, <strong>and</strong> so on. The shortest<br />
distance is most important, since the correlation<br />
function is usually a decreasing function of<br />
distance. Previously, one of authors defined<br />
F(1,XX) as the clumping degree of X (Tainaka <strong>and</strong><br />
Nakagiri, 2000), <strong>and</strong> F(1,XY) as the degree of<br />
symbiosis (coexistence) of both species (Tainaka,<br />
et al. 2003). In the case of present article, it may be<br />
necessary to calculate F not only for the shortest<br />
distance but also for several values of distance.<br />
This is because our model (2) contains the<br />
intraspecific competition; that is reactions (2e) <strong>and</strong><br />
(2f). For example, it is expected that F(1,XX) takes<br />
a smaller value compared to F(2,XX) because of<br />
the competition: If a pair of adjacent sites are<br />
occupied by X, then one site will be changed into E<br />
according to the reaction (2e).<br />
Fig. 1. Schematic illustration of distance. The<br />
numerals in circles denote the distance (r). The<br />
nearest neighbour corresponds to r=1, <strong>and</strong> the next<br />
nearest neighbour is represented by r=2, <strong>and</strong> so on.<br />
4. Correlation Function<br />
It is obvious that the Lotka-Volterra model has no<br />
information on the spatial distribution of<br />
individuals. Main aim of this article is to analyze<br />
spatial distribution. Species in nature usually form<br />
a non-r<strong>and</strong>om pattern. A typical example of such<br />
non-r<strong>and</strong>omness is a clumping pattern. To know<br />
the degree of clumping, it is convenient to define<br />
correlation function on a lattice space. Let F(r,jk)<br />
be the correlation function, where r is the distance<br />
between a pair of individuals <strong>and</strong> j or k represents<br />
a species (j,k=X or Y). For example F(2,XX)<br />
means the probability finding X at the distance r=2<br />
apart from a X individual. Thus correlation<br />
Fig. 2. The steady-state densities of species X <strong>and</strong><br />
Y are depicted against the mortality rate of both<br />
species. Both densities take almost the same value.<br />
5. Result of Lattice Model<br />
The population dynamics for lattice model is<br />
consistent with the prediction of mean-field theory.<br />
The system evolves into a stationary state. Four<br />
types of stationary states are observed: namely, (i)<br />
both X <strong>and</strong> Y coexist, (ii) X only survives, (iii) Y<br />
only survives, <strong>and</strong> (iv) both go extinct. If c x =c y =0 ,<br />
836
(c x <strong>and</strong> c y ), <strong>and</strong> fix the other parameters. In Fig. 2,<br />
steady-state densities of both species are depicted<br />
against the mortality rate. This figure reveals that<br />
the densities decrease with increasing the mortality<br />
rate. Heretofore, the results are qualitatively<br />
predicted by the mean-field theory.<br />
Spatial pattern exhibits specific properties. In<br />
Figs. 3 <strong>and</strong> 4, typical spatial distributions of<br />
species are illustrated, where the mortality rate of<br />
both species is 0.2 for Fig. 3 <strong>and</strong> 0.71 for Fig. 4.<br />
Indeed, the densities decrease with increasing the<br />
mortality rate. We also find from Fig. 4 that a kind<br />
of habitat segregation occurs: the species X <strong>and</strong> Y<br />
live separately.<br />
Fig. 3. A typical stationary pattern in the case of<br />
high densities. The mortality rate of both species<br />
is 0.2 which is relatively a small value.<br />
Fig. 5. The results of correlation functions<br />
F(r,XX), where r=1,2,3. These values mean the<br />
degree of clumping. If distribution of individuals is<br />
r<strong>and</strong>om, the correlation functions take unity. The<br />
top curve denotes the case of r=1. When the<br />
distance r becomes large, the correlation function<br />
tends to become small. The values of F(r,XX)<br />
diverge near the extinction threshold. Namely, the<br />
degree of clumping of each species becomes<br />
extremely large near extinction.<br />
Fig. 3. A Same as Fig. 3, but densities of both<br />
species are low. The mortality rate of both species<br />
is 0.71 which is a large value. This situation is near<br />
the extinction threshold. Interspecific segregation<br />
occurs.<br />
both species cannot coexist. With the increase of<br />
the values of c x <strong>and</strong> c y, both species become to<br />
survive together. In the present paper, we focus on<br />
the coexistence phase. There are six parameters: at<br />
first, we consider a symmetrical case: r x =r y, m x =m y<br />
<strong>and</strong> c x =c y. We change the values of mortality rates<br />
We analyze the segregation by the use of<br />
correlation function. Figures 5 <strong>and</strong> 6 show the<br />
correlation functions F(r,XX) for r=1,2,3. In the<br />
case of Fig. 5, the correlation functions are plotted<br />
against the mortality rate (death rate). In Fig. 6,<br />
they are plotted against the steady-state density<br />
(log-log plot). The functions F(r,XX) represent the<br />
degree of clumping. If F(r,XX) takes a large value,<br />
then the species X is clumped. It is found from Fig.<br />
5 that the degree of clumping increases with the<br />
increase of the mortality rate. Such a profile is also<br />
observed for the species Y, because our system is<br />
unchanged for the exchange of X <strong>and</strong> Y<br />
(symmetrical case). Figure 6 reveals that the<br />
correlation functions satisfies the same power law;<br />
837
when the steady-state densities approach zero, they<br />
diverge; namely, F(r,XX) approaches infinity.<br />
In Fig. 7, F(r,XY) for r=1,2,3 are plotted against<br />
the steady-state densities. This figure implies that<br />
the degree of coexistence decreases with the<br />
increase of the mortality rate. In particular, if the<br />
densities of both species become zero, both species<br />
live separately.<br />
threshold. The results of correlation functions (Figs.<br />
5, 6 <strong>and</strong> 7) demonstrate the phase transition of<br />
habitat segregation. With decreasing the densities<br />
of species X <strong>and</strong> Y, both species live separately.<br />
6. Conclusions <strong>and</strong> <br />
We have develop the spatial explicit model which<br />
is a lattice version of the Lotka-Volterra<br />
competition model. The population dynamics of<br />
our model is well predicted by the Lotka-Volterra<br />
model. It is obvious that the Lotka-Volterra model<br />
has no information on the spatial distribution of<br />
individuals. Our system evolves into a stationary<br />
state. Depending on the values of mortality rates,<br />
the stationary pattern exhibits a kind of phase<br />
transition: when the densities of both species<br />
become zero, the habitat segregation completely<br />
occurs.<br />
Fig. 7. The results of correlation functions<br />
F(r,XY), where r=1,2,3. These values indicate the<br />
degree of coexistence. The values of correlation<br />
functions become zero near the extinction<br />
threshold.<br />
Fig. 6. The relation between the correlation<br />
functions displayed in Fig. 5 <strong>and</strong> the densities<br />
plotted in Fig. 2 (log-log plot). The plots are<br />
almost on lines. This means a kind of power law;<br />
when the steady-state densities approach zero, the<br />
degree of clumping<br />
¢<br />
¡ ¢ <br />
£¤¥¥¦¥§¨©¤¨©¤<br />
of each species diverges.<br />
¡¢ ¢ ¢ ¢<br />
Our system (2) contains the interspecific<br />
competition; namely, both species X <strong>and</strong> Y<br />
compete to get the empty site (E). This type of<br />
competition also exists for the individuals of the<br />
same species (intraspecific competition). Our<br />
system further contains the other type of<br />
intraspecific competition; that is, reactions (2e) <strong>and</strong><br />
(2f). Although intraspecific competition is stronger<br />
than interspecific one, the degree of clumping of<br />
each species infinitely increases near the extinction<br />
Heretofore, we dealt with the symmetrical case:<br />
that is, the system does not change with respect to<br />
the reversal of species X <strong>and</strong> Y. It should be noted<br />
that the result of habitat segregation is almost<br />
unchanged in asymmetrical cases. If a species is<br />
endangered, the degree of clumping becomes large.<br />
Finally, we discuss the origin of habitat<br />
segregation. The enhancement in clumping degree<br />
may be originated in the fact that offspring are<br />
located near their mother. For this reason, many<br />
species potentially have the mechanism of habitat<br />
segregation.<br />
7<br />
REFERENCES<br />
DURRETT, R. (1988). Lecture Notes on Particle<br />
Systems <strong>and</strong> Percolation. California: Wadsworth<br />
<strong>and</strong> Brooka/Cole Advanced Books & <strong>Software</strong>.<br />
HARADA, Y. & IWASA, Y. (1994). Lattice<br />
population dynamics for plant with dispersing<br />
seeds <strong>and</strong> vegetative propagation. Res. Popul.<br />
Ecol. 36, 237-249.<br />
HARRIS, T. E. (1974). Contact interaction on a<br />
lattice. Ann. Prob. 2, 969-988.<br />
KATORI, M. & KONNO, N. (1991). Upper<br />
bounds for survival probability of the contact<br />
process. J. Stat. Phys. 63, 115-130.<br />
MATSUDA, H., OGITA, N., SASAKI, A. &<br />
SATO, K. (1992). Statistical mechanics of<br />
838
population: the lattice Lotka-Volterra model, Prog.<br />
theor. Phys. 88, 1035-1049.<br />
MARRO, J. & DICKMAN, R. (1999).<br />
Nonequilibrium Phase Transition in Lattice<br />
Models. Cambridge: Cambridge Univ. Press.<br />
NAKAGIRI, N., TAINAKA, K. & TAO, T.<br />
(2001). Indirect relation between species extinction<br />
<strong>and</strong> habitat destruction. Ecol. Mod. 137, 109-118.<br />
NOWAK, M. A., BONHOEFFER, S & MAY, R.<br />
M. (1994). More spatial games. Int. J. Bifurcation<br />
<strong>and</strong> Chaos, 4, 33-56.<br />
TAINAKA, K. (1988). Lattice model for the<br />
Lotka-Volterra system. J. Phys. Soc. Japan, 57,<br />
2588-2590.<br />
TAINAKA, K. & NAKAGIRI, N. (2000).<br />
Segregation in an interacting particle system. Phys.<br />
Lett. A 271, 92-99.<br />
TAINAKA, K., et al. (2003). The effect of<br />
mutualism on community stability. J. Phys. Soc.<br />
Japan, 68, 956-961.<br />
839
A Model of the biocomplexity of deforestation in tropical<br />
forest: Caparo case study<br />
Raquel Quintero 1 , Rodrigo Barros 1, 2 , Jacinto Dávila 1 , Ni<strong>and</strong>ry Moreno 1 , Giorgio Tonella 1,3 , Magdiel Ablan 1 .<br />
1<br />
Centro de Simulación y Modelos (CESIMO).<br />
2 Grupo de Investigación BIODESUS.<br />
3 University of Lugano, Switzerl<strong>and</strong>.<br />
Universidad de los Andes. Facultad de Ingeniería.<br />
Av. Don Tulio Febres Cordero. Mérida, Venezuela.5101<br />
raquelq@ula.ve, rbarros@ula.ve, jacinto@ula.ve, morenos@ula.ve, tonella@ieee.org, mablan@ula.ve<br />
Abstract: This paper presents some preliminary results with a multi-agents modeling approach to underst<strong>and</strong><br />
the complexity of deforestation in tropical forests. The approach was applied to the study of the deforestation<br />
of the Caparo Forest Reserve, in the western part of Venezuela. The model includes, among others, the<br />
following types of agents: several instances of settlers, government <strong>and</strong> lumber concessionaries. Settler<br />
agents represent people of limited economical resources that occupied l<strong>and</strong> of the reserve with the aims of<br />
improving their socio-economical status <strong>and</strong> obtaining in the future the property of the occupied l<strong>and</strong>. They<br />
use subsistence agriculture <strong>and</strong> they try to maximize the benefits from the l<strong>and</strong> occupation, without knowing<br />
that they could generate ecological or environmental problems such as soil exhaustion, due to inexistent or<br />
poor management practices. The lumber concessionaires are represented by companies that are constantly<br />
supervised by the State; their work is to exploit the forest using management plans previously approved in<br />
agreement with the Government. In addition to the dynamical interactions of the agents, the used approach<br />
includes also a cellular automata model for the simulation of the dynamic of the natural system. Both aspects<br />
use representational tools developed in house: Galatea [Uzcátegui, 2002] for the multi-agents aspects,<br />
Actilog [Dávila, 2003] a logic language for the description of rules, <strong>and</strong> SpaSim [Moreno, 2001, 2003] for<br />
the Cellular automata aspects.<br />
Keywords: Biocomplexity, Spatio-Temporal models, Multi-agents modeling <strong>and</strong> simulation.<br />
1. INTRODUCTION<br />
This study is a subproject of “Biocomplexity:<br />
Integrating Models of Natural <strong>and</strong> Human<br />
Dynamics in Forest L<strong>and</strong>scapes Across Scales <strong>and</strong><br />
Cultures”<br />
[http://www.geog.unt.edu/biocomplexity]<br />
It is carried out at the Caparo Forest Reserve in<br />
Venezuela wit the aim to model <strong>and</strong> simulate l<strong>and</strong>use<br />
processes <strong>and</strong> changes in vegetation cover as<br />
a consequence of human actions <strong>and</strong> the effects of<br />
the changes in subsequent human decisionmaking.<br />
Human behavior affecting forest sustainability is<br />
simulated using multi-agent models, there are rules<br />
to generate dynamics similar to what is observed at<br />
the forest reserve; meanwhile, forest dynamic is<br />
represented by a Cellular Automata.<br />
Explicit modeling of human actions <strong>and</strong> their<br />
interaction with ecosystems will give policymakers<br />
information about the impact of their decisions on<br />
the future composition, structure, <strong>and</strong> functionality<br />
of local ecosystems. It will also facilitate a more<br />
informed analysis of the long-term consequences<br />
of private choices <strong>and</strong> public policies on the<br />
natural systems in which human systems are<br />
embedded <strong>and</strong> with which they interact [Acevedo<br />
et al. 2003].<br />
The structure of this paper includes: a brief<br />
description of the Caparo Forest Reserve <strong>and</strong> the<br />
agents considered; models’ description;<br />
implementation details; <strong>and</strong> finally, the<br />
conclusions <strong>and</strong> comments about future work.<br />
1.1 Case Study<br />
The Caparo Forest Reserve, CFR, was created in<br />
1961 <strong>and</strong> its original purpose was to support the<br />
development of the logging industry in the zone,<br />
while preserving one of the finest forests of<br />
Venezuela [CESIMO, 1998]. It is located<br />
southeast of the Barinas State, in the Venezuelan<br />
western plains region. Its extension is of 176,434<br />
840
hectares, <strong>and</strong> it has been divided on three units to<br />
facilitate its management (Figure 1).<br />
The study takes place in Unit I, an area of 53,358<br />
hectares, which itself includes a special area called<br />
the Experimental Unit, that is used by the<br />
University of Los Andes for research <strong>and</strong><br />
educational activities.<br />
Currently, only 7,000 ha. survived (all in the<br />
Experimental Unit).Nevertheless, this area is still<br />
not exempted from deforestations due to agrarian<br />
settlement process.<br />
Extensive cattle ranch dominates the l<strong>and</strong>-use.<br />
After some years, the property of the parcels is<br />
transferred to the settlers, by application of the<br />
Agrarian Reform. Then the parcels are sold at very<br />
low prices, to l<strong>and</strong>lords, politicians <strong>and</strong> cattle<br />
dealers who urged <strong>and</strong> supported the original<br />
settlements [Centeno, 1997]. This process,<br />
characterized by the concentration of the property,<br />
forces the initial group of settlers to move towards<br />
primary cycle settlements or to wage-earning work<br />
(as workers for l<strong>and</strong>lords) [Sánchez, 1989].<br />
There is in the model a settlement function that<br />
considers those places that are more attractive for<br />
this agent: l<strong>and</strong>-uses without supervision, such as<br />
plantations, secondary bushes <strong>and</strong> prairies. At the<br />
same time, this function model the movement of<br />
the settlers using weighted by distance buffers<br />
around rivers, borders <strong>and</strong> roads.<br />
1.3 Concessionary Agent<br />
The lumber concessionaires are represented by<br />
private companies that have the function to carry<br />
out the forest exploitation <strong>and</strong> management plans<br />
in the reserve areas under the supervision of the<br />
Government.<br />
Figure 1. Caparo Forest Reserve Units<br />
[Jurgenson, 1994].<br />
Many factors have contributed to forest<br />
disappearing in the CFR: unsuitable forest<br />
management of some lumber concessionaires,<br />
contradictions between different governmental<br />
organisms, poverty <strong>and</strong> the dem<strong>and</strong> of l<strong>and</strong>s for<br />
agricultural activities, <strong>and</strong> the existence of political<br />
interests in favor of settlements, among others<br />
factors [Ablan et al., 2003].<br />
The following is a description of the most<br />
important characteristics of the agents<br />
implemented in the models.<br />
1.2 Settler Agent<br />
According to Rojas [1993], the first settlers took<br />
possession of a certain area at the reserve <strong>and</strong><br />
practiced subsistence (i.e. slash <strong>and</strong> burn)<br />
agriculture. This surface could be an uncultivated<br />
l<strong>and</strong> (previously deforested <strong>and</strong> unoccupied).<br />
Before five years, the soils are exhausted, <strong>and</strong> the<br />
harvests are no longer enough to sustain the settler<br />
<strong>and</strong> his family. Some settlers try to exp<strong>and</strong> their<br />
farms at the expense of new deforestation.<br />
However, sooner or later, they will end facing the<br />
same situation. The alternative is to seed pasture<br />
for cattle (which gives value to the l<strong>and</strong>) so that<br />
later, they will be able to sell its improvements to<br />
l<strong>and</strong>lords or other settlers.<br />
At this stage, pasture retailers <strong>and</strong> l<strong>and</strong>lords<br />
acquire the improvements of primary settlers.<br />
The lumber concessionary agent implemented,<br />
makes a very simplistic <strong>and</strong> hypothetical forest<br />
management within the reserve: the lumber<br />
concessionaire exploits the forest <strong>and</strong> proceeds to<br />
plant commercial valuable species; furthermore,<br />
the concessionaire is in charge of forest plantations<br />
supervision during the first two years [Ablan et al.,<br />
2003].<br />
In case that the concessionaire finds a settler on its<br />
assigned zone, there are two behaviors<br />
implemented: -the first one implies that the<br />
concessionary agent ignores the settler <strong>and</strong><br />
continues the work at another place that is not<br />
occupied by settlers; - the second one implies that<br />
the concessionary agent informs to the<br />
Government about the settlements.<br />
The implemented concessionary agent has a 30<br />
year cycle <strong>and</strong> it is allowed to harvest 1,200 ha of<br />
“Forest” annually (Figure 2); after the<br />
concessionary acts on the site, the use of l<strong>and</strong> is<br />
changed to “Logged Forest”. Once the 30 year<br />
cycle is over, the concessionary could harvest the<br />
first compartment again (the concessionary area is<br />
divided in “compartments”).<br />
1.4 Government Agent<br />
Three different behaviors or scenarios were<br />
implemented for the Government agent. These<br />
behaviors represent different ways of the role of<br />
the government at the CFR. Their specification is<br />
as follows:<br />
841
1. The Government neither interacts nor<br />
interferes with the activities of the others<br />
agents. It does not have any monitoring<br />
activity. This is called the “H<strong>and</strong>s-off”<br />
government model.<br />
2. The Government has a “strong” policy to<br />
keep settlers away from protected forest<br />
areas (called at our models as the “Pro-<br />
Forestry Government”). This agent has a<br />
monitoring process where any settler<br />
founded at the CFR area is evacuated.<br />
Furthermore, if the concessionary agent,<br />
on its exploitation process, finds a settler<br />
in the zone, the government agent<br />
receives the settlement’s information<br />
from the concessionary <strong>and</strong> the indicated<br />
settlers will be removed from the CFR<br />
area in the next government’s monitoring<br />
process<br />
3. The Government has an “agroforestry”<br />
policy, which means that this agent<br />
monitors the forest area trying to protect<br />
it, but when he finds a settler, the settler<br />
will be relocated to a special area for<br />
agricultural activities. At the same time,<br />
the government agent receives<br />
settlements’ information from the<br />
concessionary agent <strong>and</strong> then the<br />
indicated settlers are relocated.<br />
The “Pro-forestry” <strong>and</strong> the “Agroforestry”<br />
governments evaluate the concessionary’s<br />
exploitation <strong>and</strong> plantation quotas. The<br />
concessionary will be punished by the government<br />
in case the concessionary has failed the agreed<br />
quotas.<br />
Monitoring is based on a function that considers<br />
the places that are more attractive for settlements<br />
(buffers around rivers, CFR borders, roads…).<br />
2. THE MODELS<br />
On the above specification, three computational<br />
models have been developed. They differ only in<br />
the implemented behavior of the Government<br />
agent.<br />
Each model counts with a hundred settler agent<br />
instances (identified from 1 to 100), <strong>and</strong> one<br />
concessionary agent.<br />
L<strong>and</strong>-use change is modeled as cellular automata.<br />
State transition rules are simplifications of the<br />
ecological dynamic of forest succession at the<br />
CFR. Other characteristics of the model are:<br />
• Number of layers: 6.<br />
1. L<strong>and</strong>-uses Layer: each cell can be in any<br />
of the fifteen states described on the<br />
Figure 2.<br />
2. Time in Use Layer: used as a time count<br />
layer that indicates the time that a cell has<br />
spent remaining at a determined state.<br />
3. Population Layer: each cell can be in<br />
some of the following states: - 0<br />
represents an unoccupied cell; - 1, there is<br />
a settler occupying the cell; - 2, there is a<br />
l<strong>and</strong>owner occupying the cell.<br />
4. Supervision Layer: each cell can be in<br />
some of the following states: - 0, that<br />
represents a no watched over cell, - <strong>and</strong> 1,<br />
that indicates a watched cell.<br />
Figure 2 L<strong>and</strong>-uses Transition Graph<br />
5. Settler Identification layer: if the cell is<br />
occupied by a settler the cell in this layer<br />
will have the identification number of<br />
the settler.<br />
6. Compartment layer: it indicates the<br />
compartment’s sequence to be followed<br />
by the agent concessionary in his<br />
exploitation process.<br />
842
• Moore Neighborhood (Zeigler et al., 2000)<br />
for every cell (this neighborhood includes the<br />
eight adjacent cells).<br />
• State Transition Rules:<br />
o Each l<strong>and</strong>-use can stay in that state until<br />
the cell remaining time in that states<br />
achieves the transition time indicated at<br />
Figure 2.<br />
o Permanent states are: Flood River Bank,<br />
Seasonal Wetl<strong>and</strong> used for<br />
stockbreeding, Rivers, Roads, Livestock,<br />
<strong>and</strong> Farming.<br />
2.1 Agent’s Interactions with the environment<br />
The interaction between the settler agent <strong>and</strong> the<br />
environment is described as follows:<br />
1. A settler agent can establish a farm in a<br />
cell that is unoccupied <strong>and</strong> without<br />
supervision. Certain l<strong>and</strong>-uses are preferred<br />
for the initial settlements, <strong>and</strong> once the<br />
settler is established, they will change the<br />
l<strong>and</strong> usage to adapt it to its agricultural<br />
activities.<br />
2. A settler agent can exp<strong>and</strong> its funds at<br />
neighboring unoccupied <strong>and</strong> without<br />
surveillance cells.<br />
3. Before five years, the soils are<br />
exhausted, <strong>and</strong> then the settler agent moves<br />
to another place inside the CFR. Once the<br />
place is left by the settler, the l<strong>and</strong> usage is<br />
changed to prairie.<br />
The concessionary agent interacts with the<br />
environment in the following ways:<br />
1. To exploit the forest, the concessionary<br />
agent needs <strong>and</strong> unoccupied cell with a l<strong>and</strong><br />
usage equals to forest. Then, the l<strong>and</strong> usage is<br />
changed to Logged Forest.<br />
2. To reforest, the concessionary agent<br />
needs an unoccupied cell with a l<strong>and</strong> usage<br />
equals to prairie or secondary bushes. Then<br />
the l<strong>and</strong> usage is changed to plantations.<br />
2.2 Sample of Agent’s Rules<br />
To detail settler agent’s rules we use Actilog<br />
Language, which is a language to write<br />
generalized, (condition --> action), activation<br />
rules. The semantics of the language is based on<br />
the assumption that implications (conditional<br />
goals) can be used to state integrity constraints for<br />
an agent. These integrity constraints describe<br />
conditions under which the agent's goals must be<br />
reduced to plans that can be executed. See Dávila,<br />
[2003] for more details.<br />
Here is a simplified example of the implemented<br />
rules:<br />
FARMS EXPANSION: It is carried out whenever<br />
the settler finds a neighboring unoccupied l<strong>and</strong><br />
without supervision The next Actilog language<br />
code line indicates the only way to farm<br />
expansion:<br />
if thinking_on expansion, not (occupied_l<strong>and</strong>),<br />
not (supervision) then funds_expansion.<br />
3. IMPLEMENTATION<br />
The implemented model is a multi-agent spatial<br />
explicit model, where agents are codified using<br />
GALATEA agents’ library, while the space is<br />
modeled as cellular automata representing a<br />
simplified account of the dynamics of the<br />
environmental system. The cellular automata is<br />
implemented by means of the SpaSim-lib library.<br />
Both the libraries <strong>and</strong> the model are encoded in<br />
Java.<br />
The simulation theory that explains the way we<br />
combined the simulator (SpaSim) with the tool<br />
that implements the agents (GALATEA) is<br />
presented in Moreno, [2002].<br />
Galatea [Uzcátegui, 2002] is a multi-agent<br />
simulation platform that nicely fits with SpaSim<br />
[Moreno, 2002] for the sake of an integrated<br />
spatial, agent-based simulation model. Galatea<br />
provides for a collection of classes to model<br />
reactive <strong>and</strong> rational agents, with a scalable,<br />
logic-based, inference engine which will<br />
eventually allow the agents to perform metareasoning,<br />
of the kind required to reason about<br />
other agents’ reasoning. For the time being,<br />
however, the agents are more of a reactive kind,<br />
with behaviors that can be modeled by means of<br />
generalized condition-action rules [Davila, 2003].<br />
The methodological path used here tries to embed<br />
as much behavior as possible with simpler agent<br />
models in such a way that extensions, such as<br />
those required for meta-reasoning, remain<br />
computationally feasible. This is why the research<br />
has developed these simplified agent models,<br />
testing for their expressiveness <strong>and</strong> evaluating<br />
their validity progressively. In this respect, It<br />
coincides with the work done in Monticino et al.<br />
[2004], also reported in this congress volume.<br />
However, these models are not attached to<br />
decision theory. The reason for this is that, even<br />
though, decision-theoretical approaches have the<br />
advantage of their straightforward psychological<br />
interpretation, the same advantage can be<br />
achieved with logic based models, without having<br />
to pre-encode, in numerical values of the potential<br />
consequences, all the qualitative information<br />
about agents’ preferences <strong>and</strong> assumptions for<br />
meta-reasoning.<br />
SpaSim is software that allows the specification,<br />
simulation, visualization <strong>and</strong> analysis of spatial<br />
models in the same environment, using a friendly<br />
user interface while at the same time providing<br />
considerable flexibility. Square cells were used<br />
for the cellular automata to keep compatibility<br />
with most raster GIS systems in use. Also the<br />
843
software integrates simulation techniques (like<br />
cellular automata), spatial analysis, spatio -<br />
temporal analysis, <strong>and</strong> maps visualization [Ablan<br />
et al., 2003].<br />
The implementation includes former processes<br />
that affect the evolution of the l<strong>and</strong>-use cover,<br />
which are carried out by the already implemented<br />
agents. There are some other agents that have not<br />
been implemented yet, like politicians, Los Andes<br />
University, among others…<br />
4. RESULTS<br />
For each one of the government scenarios<br />
implemented, the model was run for 65 years,<br />
because this is the estimated time required to<br />
observe a transformation from a Logged Forest<br />
into a Forest. The initial state of the CFR<br />
corresponds to the l<strong>and</strong>-use reported in Pozzobón<br />
[1996] et al. for the year 1987. Simulation results<br />
are portrayed as maps that show the spatial<br />
distribution of l<strong>and</strong>-use types obtained in each of<br />
the scenarios.<br />
In the H<strong>and</strong>s-off Government Model, at the end of<br />
the simulation the forest have been replaced by<br />
other types of l<strong>and</strong>-uses, the dominant l<strong>and</strong>-use<br />
being cattle <strong>and</strong> ranching activities.<br />
On the other h<strong>and</strong>, in the pro-forestry government<br />
model, the settlers are finally removed from the<br />
CFR area, <strong>and</strong> the forest has the opportunity to<br />
achieve its original state.<br />
The agroforestry government, at the end of the<br />
simulation the forest has the opportunity to<br />
achieve its original state, but the settlers have left<br />
the special place for agricultural activities <strong>and</strong> the<br />
l<strong>and</strong>lords has occupied that zone for cattle <strong>and</strong><br />
ranching activities<br />
5. CONCLUSIONS<br />
Simulation results agree qualitatively well with<br />
what is now known about l<strong>and</strong>-use change,<br />
tropical forest succession <strong>and</strong> forest management<br />
in the area. On the contrary to what was believed<br />
a few decades ago, it takes vegetal succession in<br />
tropical forests relatively long periods to fully<br />
recuperate its original state, both in volumes of<br />
wood <strong>and</strong> in floristic diversity. For example,<br />
Guariguata & Ostertag [2002] <strong>and</strong> Gómez-Pompa<br />
& Vázquez-Yanez [1985] say that the process<br />
leading to the reappearance of the initial forest<br />
species in the way they were found at the moment<br />
of deforestation could take even around a hundred<br />
(100) years. Our results corroborate that the way<br />
in which the forests were managed, with a 30 year<br />
cycle, would end up compromising the<br />
availability of the forest’s resources in the future,<br />
just as Martinez-Angulo (1955), Lamprecht<br />
(1956: cited by Kammescheidt et al. 2001) <strong>and</strong><br />
Veillón (1971) had warned.<br />
Some points to be improved at our future works:<br />
• Population growth of settlers will be<br />
represented at future models representing<br />
the influence of government policies.<br />
• The l<strong>and</strong>lords will be implemented<br />
explicitly as agents. This will enhance<br />
agents’ interactions as they would be<br />
able to exp<strong>and</strong> their properties, acquire<br />
other settler improvements, etc.<br />
• The government agent will implement a<br />
more detailed evaluation of the<br />
concessionary performance; measuring<br />
beyond exploitation <strong>and</strong> reforestation<br />
quotas.<br />
• The ecological realism of the cellular<br />
automata will be improved by estimating<br />
its parameters from detailed gap-model<br />
simulations (as in Acevedo, et al. [2001],<br />
<strong>and</strong> Monticino, et al. [2002]).<br />
Details of the work <strong>and</strong> future developments can<br />
be found at the www page of the project:<br />
http://chue.ing.ula.ve/INVESTIGACION/PROYE<br />
CTOS/BIOCOMPLEXITY/<br />
5. ACKNOWLEDGMENTS<br />
SpaSim software is a product of the research<br />
activities developed by Ni<strong>and</strong>ry Moreno as a part<br />
of a researcher development program (Programa<br />
de Desarrollo del Investigador Novel) at Los<br />
Andes University (ULA-FONACIT). Galatea was<br />
originally developed under project ULA CDCHT<br />
I99-667 <strong>and</strong> it uses the GLIDER simulation<br />
language, also developed at Los Andes<br />
University.<br />
This material is based upon work supported by<br />
the US National Science Foundation under Grant<br />
CNH BCS-0216722. Any opinions, findings, <strong>and</strong><br />
conclusions or recommendations expressed in this<br />
material are those of the author(s) <strong>and</strong> do not<br />
necessarily reflect the views of the National<br />
Science Foundation.<br />
We would like to thank Miguel Acevedo,<br />
Arm<strong>and</strong>o Torres, Hirma Ramirez <strong>and</strong> Emilio<br />
Vilanova for many valuable discussions <strong>and</strong><br />
feedback.<br />
6. REFERENCES<br />
Ablan M., Dávila J., Moreno N., Quintero R.,<br />
Uzcátegui M. Agent <strong>Modelling</strong> of the Caparo<br />
Forest Reserve. The 2003 European<br />
Simulation <strong>and</strong> <strong>Modelling</strong> Conference. ISBN:<br />
90-77381-04-X. Published by EUROSIS-ETI.<br />
Ghent-Belgium. Edited: Beniamino Di<br />
Marino, Laurence Tianruo Yang <strong>and</strong> Carmen<br />
Bobeanu. Section: Simulation <strong>and</strong> Biology.<br />
Subsection: Simulation of Ecosystems. Page:<br />
367-372. (2003)<br />
Acevedo M.F., S. Pamarti, M. Ablan, D.L. Urban<br />
<strong>and</strong> A. Mikler, Modeling forest l<strong>and</strong>scapes:<br />
844
parameter estimation from gap models over<br />
heterogeneous terrain. Simulation 77, 53-68,<br />
2001.<br />
Centeno, J. C. Deforestaciones fuera de control en<br />
Venezuela. Universidad de los Andes, 1997.<br />
http://www.ciencias.ula.ve/~jcenteno/DEFOR<br />
-ES.html]<br />
CESIMO. Gaia. Caso Venezuela: Deforestación y<br />
Políticas de la Propiedad de la Tierra en la<br />
Reserva Forestal de Caparo. Centro de<br />
Simulación y Modelos de la Universidad de<br />
los Andes, Mérida, Venezuela. 1998.<br />
http://chue.ing.ula.ve/GAIA/CASES/VEN/ES<br />
PANOL/datos-especificos.html,<br />
Dávila, J. Actilog: An Agent Activation<br />
Language. Practical Aspects of Declarative<br />
Languages, Dahl, V <strong>and</strong> Wadler, P. Editors.<br />
Lecture Notes in Computer Science. 2562.<br />
Springer. 2003.<br />
Gómez-Pompa, A. & Vázquez-Yánez, C.<br />
Estudios sobre la regeneración de selvas en<br />
regiones cálido – húmedas de México.<br />
Investigaciones sobre la regeneración de<br />
selvas altas en Veracruz, México, Vol. II. 1-<br />
25. Editorial Alambra Mexicana S.A.,<br />
México, 1985.<br />
Guariguata, M. & Ostertag, R. Sucesión<br />
Secundaria en Bosques Neotropicales.<br />
(Guariguata, M. & Kattan, G.) pp. 591-623.<br />
Libro Universitario Regional. Cartago, Costa<br />
Rica (2002).<br />
Jurgenson, O. (1994). Mapa de Vegetación y Uso<br />
Actual del Área Experimental de la Reserva<br />
Forestal de Caparo. Estado Barinas.<br />
Universidad de Los Andes, Facultad de<br />
Ciencias Forestales. Cuaderno Comodato<br />
ULA-MARNR No. 22. 44 p.<br />
Kammescheidt, L., Torres, A., Franco, W. &<br />
Plonczak, M. History of logging <strong>and</strong><br />
silvicultural treatments in the western<br />
Venezuela plain forests <strong>and</strong> the prospect for<br />
sustainable forest management. Forest<br />
Ecology <strong>and</strong> Management, 148: 1-20.<br />
Elsevier Science Publishers, The Netherl<strong>and</strong>s,<br />
2001<br />
Lamprecht, H. Unos apuntes sobre el principio del<br />
rendimiento sostenido en la ley forestal y de<br />
aguas venezolanas. Boletín de la Facultad de<br />
Ingeniería Forestal 10. 9 34, Mérida,<br />
Venezuela, 1956.<br />
Martínez-Angulo. Estudio económico-social de la<br />
empresa maderera en los Estados Portuguesa<br />
y Barinas. Tesis de grado, Facultad de<br />
Ciencias Forestales y Ambientales, Mérida,<br />
Venezuela, 1955.<br />
Monticino M.G., Cogdill T. <strong>and</strong> M.F. Acevedo,<br />
Cell Interaction in Semi-Markov Forest<br />
L<strong>and</strong>scape Models. pp 227-232. In: Rizzoli<br />
A.E. <strong>and</strong> A.J. Jakeman (Eds.). Integrated<br />
Assessment <strong>and</strong> Decision Support,<br />
Proceedings of the First Biennial Meeting of<br />
the IEMSS. Lugano, Switzerl<strong>and</strong>, 2002.<br />
Monticino, M, Acevedo M., Callicott B., Cogdill,<br />
T, Ji M., <strong>and</strong> Lindquist, C. Coupled Human<br />
<strong>and</strong> Natural Systems: A Multi-Agent Based<br />
Approach. Submitted to the Proceedings of<br />
the Second IEMSS Conference, Osnabrück<br />
Germany, 2004.<br />
Moreno, N.; Ablan. M.; Tonella, G. ‘SpaSim: A<br />
software to Simulate Cellular Automata<br />
Models’. In proceedings IEMSs 2002, First<br />
Biennial Meeting of the <strong>International</strong><br />
<strong>Environmental</strong> Modeling <strong>and</strong> <strong>Software</strong><br />
Society, Lugano, Switzerl<strong>and</strong>. 2002. Page:<br />
348-353.<br />
Moreno N.,. Diseño e Implementación de una<br />
estructura para el soporte de simulación<br />
espacial en Glider. Master Theses. Graduate<br />
Computer Program Universidad de los Andes,<br />
Mérida, Venezuela, 2001<br />
North Texas University. Proposal to the National<br />
Science Foundation (NSF): “Biocomplexity:<br />
Integrating Models of Natural <strong>and</strong> Human<br />
Dynamics in Forest L<strong>and</strong>scapes Across<br />
Scales <strong>and</strong> Cultures”, 2003.<br />
http://www.geog.unt.edu/biocomplexity/<br />
Pozzobón, E. Estudio de la Dinámica de las<br />
Deforestaciones en la Reserva Forestal de<br />
Caparo mediante Imágenes HRV SPOT.<br />
Universidad de Los Andes. Facultad de<br />
Ciencias Forestales. Escuela de Ingeniería<br />
Forestal. Departamento de Ingeniería.<br />
Mérida-Venezuela. Octubre, 1996.<br />
Rojas López, J. La Colonización Agraria de las<br />
Reservas Forestales: ¿un Proceso sin<br />
Solución? Universidad de los Andes, Instituto<br />
de Geografía, Mérida, Venezuela, Cuaderno<br />
Geográficos No. 10, Junio de La<br />
Colonización Agraria de las reservas<br />
Forestales: ¿Un proceso sin solución? 1993.<br />
Sánchez, M. Situación Actual del Proceso de<br />
Ocupación de la Reserva Forestal en Caparo.<br />
Universidad de los Andes, Instituto de<br />
Geografía y Conservación de Recursos<br />
Naturales, Mérida, Venezuela. 1989.<br />
Uzcátegui S., M. Diseño de la Plataforma de<br />
Simulación de Sistemas Multi-Agentes<br />
GALATEA. Master Theses. Graduate<br />
Computer Program, Universidad de Los<br />
Andes, Mérida, Venezuela. 2002.<br />
Veillón, J. P. Importancia económico-social de<br />
los bosques del Estado Portuguesa,<br />
Venezuela. Instituto de Silvicultura,<br />
Universidad de Los Andes, Mérida,<br />
Venezuela, 1971.<br />
Zeigler B.P., Praehofer H., Tag Gon Kim. Theory<br />
of Modeling <strong>and</strong> Simulation. Second Edition.<br />
Integrating Discrete Event <strong>and</strong> Continuous<br />
Complex Dynamic Systems. Academic Press,<br />
USA (2000).<br />
845
Developing Tools for Adaptive Integrated Water<br />
Resource Management in a Semi Arid Region:<br />
Possibilities, Probabilities <strong>and</strong> Uncertainties.<br />
Denise Eisenhuth a , Juan Bellot a , , Andreu Bonet a , Juan R. Sánchez a<br />
a Departamento de Ecología, Universidad de Alicante<br />
e-mail: D.M.Eisenhuth@ua.es<br />
Abstract: The objective of the AQUADAPT Project (www.aquadapt.net) is to develop strategic tools<br />
to inform adaptive integrated water resource policy using a co-evolutionary approach. In Spain a<br />
methodology has been developed in the framework of an integrative study to identify evidence of coevolutionary<br />
processes between the hydrological system <strong>and</strong> the water-using communities of the Marina<br />
Baixa over a period of 50 years. The Marina Baixa is comprised of 18 municipalities each with radically<br />
different l<strong>and</strong>-use patterns spread over an area of 671km 2 . The research task is complicated by the fact<br />
that Spain is a country that is currently debating the merits of a move from dem<strong>and</strong> management to<br />
supply augmentation for this water-using region. Models, processes <strong>and</strong> assessments are applied to reflect<br />
a co-evolutionary perspective of the relationships between water resources, ecological quality <strong>and</strong><br />
sustainable development. A variety of mosaics have been assembled with which to identify couplings of<br />
elements that could have spawned a co-evolving process. The paper discusses both the challenges <strong>and</strong> the<br />
merits associated with designing new methodologies in an integrative study framework for dealing with<br />
the possibilities, probabilities <strong>and</strong> uncertainties of adaptive integrated water resource management.<br />
Keywords: adaptive integrated water resource management; co-evolutionary approach; ecological quality;<br />
sustainable community development.<br />
1<br />
. INTRODUCTION<br />
The intellectual orientation of this policy<br />
relevant research is one of how to determine <strong>and</strong><br />
then to identify evidence of the co-evolution 1 of<br />
natural resources <strong>and</strong> human societies. The<br />
application of a co-evolutionary perspective<br />
could further illuminate the rapid <strong>and</strong><br />
unforeseen changes that are inherent to the<br />
environmental, socio-economic <strong>and</strong> governance<br />
contexts within which water supply <strong>and</strong> dem<strong>and</strong><br />
patterns develop. It could also lead to clearer<br />
underst<strong>and</strong>ings of the forces that direct the<br />
structure, processes <strong>and</strong> dynamics of socionatural<br />
systems. (McIntosh & Jeffrey, 2003)<br />
The use of a co-evolutionary approach might be<br />
the means with which to identify the<br />
1 Co-evolution is understood to be reciprocal<br />
evolution in which 2 or more populations evolve<br />
in response to one another.<br />
characteristics of both resilience 2 <strong>and</strong> adaptive<br />
potential 3 , as well as how these concepts link to<br />
the notion of sustainable management of a<br />
socio-natural system. These characteristics<br />
play a critical role in determining the<br />
relationship between water resources, ecological<br />
quality <strong>and</strong> sustainable community development<br />
in a semi-arid region. Particularly, the research<br />
is concerned with developing the types of<br />
models, processes <strong>and</strong> assessments that can be<br />
deployed to uncover: a) the pace <strong>and</strong> tempo of<br />
co-evolutionary processes b) the indicators<br />
(both quantitative <strong>and</strong> qualitative) which can be<br />
used to monitor co-evolutionary processes c)<br />
Best Practice Guidance on planning methods<br />
reflecting a co-evolutionary perspective <strong>and</strong> d)<br />
long term water management strategies for<br />
semi-arid regions.<br />
2 Resilience relates to the potential of a system<br />
to reorganise, restructure or transform<br />
3 Adaptivity relates to the potential of a system<br />
to transform through innovation<br />
846
This paper describes the challenges inherent to<br />
such a task <strong>and</strong> how the methodology is<br />
emerging. It documents attempts thus far at the<br />
Universidad de Alicante to integrate individual<br />
disciplinary underst<strong>and</strong>ings (the research team<br />
is comprised of ecologists, economists,<br />
sociologists <strong>and</strong> institutional theorists) of how<br />
to map <strong>and</strong> interpret the Marina Baixa , in terms<br />
of what is possible <strong>and</strong> probable in modelling in<br />
such a scenario, as well as how the research<br />
team is dealing with the uncertainties that are<br />
inherent to this type of policy relevant research.<br />
1.1 The challenges of the research task<br />
As the focus of this policy relevant research lies<br />
with changes to the state of natural <strong>and</strong> human<br />
systems as they evolve over time, when<br />
designing a methodology the primary challenge<br />
rested with how to frame 4 policy questions.<br />
And then to consider the implications framing<br />
has for the direction of future policy.<br />
Particularly, when it comes to the<br />
operationalisation of the terms ‘ecological<br />
quality’ <strong>and</strong> ‘sustainable community<br />
development’.<br />
The rationale for this approach is as follows.<br />
The terminology adaptive, integrated water<br />
resource management is synonymous with the<br />
terminology sustainable water management.<br />
Yet, there are distinct challenges associated with<br />
exactly how to operationalise these terms. There<br />
is a natural tendency to link the term<br />
sustainability with underst<strong>and</strong>ings of natural<br />
systems dynamics <strong>and</strong> to pay less attention to<br />
underst<strong>and</strong>ing the reciprocal relationships of a<br />
co-evolving socio-natural system (Jeffrey &<br />
McIntosh, 2004) especially when it involves the<br />
impact of water transfers.<br />
Broadly speaking, an increase to water supply<br />
will stimulate l<strong>and</strong>-use changes. Since 1999<br />
there have been water transfers to the Marina<br />
Baixa from the Júcar River. Water transfers can<br />
be described as engineering responses to<br />
maintain resilience of a hydrological system.<br />
However, engineering resilience only seeks to<br />
preserve stability. (McGlade, 2002) While<br />
engineering resilience maintains stability (viz.,<br />
sustainability) in short-term production, natural<br />
<strong>and</strong> social resilience will diminish even though<br />
engineering resilience might be great. The<br />
long-term result is a converse effect on natural<br />
4 For a comprehensive account of the<br />
implications of framing policy refer to Schön &<br />
Rein (1994)<br />
diversity that reduces the resilience <strong>and</strong> adaptive<br />
potential of ecological systems (Gunderson &<br />
Holling, 2002) as well as social systems.<br />
For the purposes of this study the characteristics<br />
of resilience <strong>and</strong> adaptive potential offer far<br />
more useful concepts through which to<br />
underst<strong>and</strong> the implications of a move from<br />
dem<strong>and</strong> management to supply augmentation<br />
<strong>and</strong> exactly how water transfers will determine<br />
future relationships between water resources,<br />
ecological quality <strong>and</strong> sustainable community<br />
development in the Marina Baixa. The focus<br />
then becomes one of how to identify the<br />
characteristics of the type of resilience that<br />
promotes innovation, or the capacity of a socionatural<br />
system to adapt to shocks or surprise.<br />
Therefore, when framing the policy questions<br />
much consideration was devoted to the impact<br />
of past water transfers, as well as the potential<br />
impact of the Júcar-Vinalopó water transfers<br />
that will greatly increase supply to the Marina<br />
Baixa<br />
2.0 Marina Baixa study area<br />
The Marina Baixa catchment (671 km 2 ), with a<br />
complex topography, is characterised by a dense<br />
l<strong>and</strong> use mosaic, with irrigated crops, dryl<strong>and</strong><br />
crops, urbanisation, <strong>and</strong> Mediterranean<br />
shrubl<strong>and</strong>s <strong>and</strong> woodl<strong>and</strong>s. The area has<br />
undergone radical socio-economic change over<br />
recent decades that can be attributed to tourism<br />
development. The main change attractors are<br />
coastal proximity (tourism) <strong>and</strong> water<br />
availability (irrigated crops). We selected three<br />
municipalities with contrasting situations where<br />
we might model <strong>and</strong> illustrate these changes:<br />
Benidorm, on the coast (51873 inhabitants,<br />
3860 has), Callosa d'en Sarriá (7057 inhabitants,<br />
3430 has) <strong>and</strong> Guadalest (180 inhabitants, 1610<br />
has). The former is one of the main tourist<br />
destinations in Europe, receiving more than 3<br />
million tourists each year, <strong>and</strong> registering 20<br />
million over-night stays. Guadalest – also a<br />
tourist destination - is situated inl<strong>and</strong> <strong>and</strong> its<br />
development strategy is very different from<br />
Benidorm. Guadalest has more than 2 million<br />
tourists visit each year, but over-night stays.<br />
Whilst, Callosa d'en Sarriá, also situated inl<strong>and</strong>,<br />
has developed monoculture (medlar <strong>and</strong> citrus<br />
trees).<br />
2.1 Spatial characteristics<br />
Traditional l<strong>and</strong> use activities are at least partly<br />
responsible for maintaining the high levels of<br />
ecological quality found in Mediterranean<br />
l<strong>and</strong>scapes (Blondel & Aronson 1999).<br />
Although the ecological concepts of balance <strong>and</strong><br />
847
stability are contested (e.g. see Perry, 2002) we<br />
refer here to a state of dynamic equilibrium in<br />
which both socio-cultural activities <strong>and</strong><br />
biological diversity <strong>and</strong> function are maintained.<br />
Changes in l<strong>and</strong> cover, water use <strong>and</strong><br />
management of the l<strong>and</strong> have occurred<br />
throughout history in Mediterranean regions <strong>and</strong><br />
other parts of the world. (Dale et al. 2002) The<br />
total l<strong>and</strong> area dedicated to human usage has<br />
grown dramatically, <strong>and</strong> increasing production<br />
of goods <strong>and</strong> services, has intensified both use<br />
<strong>and</strong> control of the l<strong>and</strong>. (Richards, 1990) Water<br />
availability throughout the l<strong>and</strong>scape varies<br />
seasonally <strong>and</strong> from year to year in response to<br />
changing weather conditions <strong>and</strong> water-use<br />
dem<strong>and</strong>s. Water resources are already being<br />
influenced by climate <strong>and</strong> l<strong>and</strong>-use changes.<br />
L<strong>and</strong> use <strong>and</strong> climate changes have a potential<br />
effect on hydrological cycles, on flow damages,<br />
groundwater recharge <strong>and</strong> dem<strong>and</strong> for the<br />
resource (Turner et al. 2001). The<br />
consequences or impacts of such changes on<br />
risk or resource reliability depend not only on<br />
biophysical changes in l<strong>and</strong>scapes, water<br />
recharge <strong>and</strong> water quality but also on the<br />
characteristics of the water management system.<br />
2.2 Temporal characteristics<br />
The history of this water management system<br />
indicates that the region could be typical of a<br />
complex co-evolving system. 5 The Marina<br />
Baixa has a history of water deficiency <strong>and</strong><br />
water-using communities have devised complex<br />
supply <strong>and</strong> dem<strong>and</strong> management arrangements<br />
to accommodate this deficiency. A<br />
distinguishing characteristic is that there has<br />
been no traditional separation of l<strong>and</strong> <strong>and</strong> water<br />
rights as there have been in other parts of the<br />
Alicante province despite the fact that the<br />
history of irrigation in Callosa d'en Sarriá can be<br />
traced to the 15 th century. The autonomy of<br />
water in the Alicante province dates back to the<br />
13 th century. Embedded 6 governance structures<br />
are numerous <strong>and</strong> are engaged in management<br />
activities that are duplicated. Management<br />
responsibilities overlap in a relatively<br />
complicated hierarchical arrangement.<br />
Institutions that we consider to be embedded<br />
function on at least six different layers – from<br />
local to catchment level. Because of their<br />
spatial <strong>and</strong> temporal embeddedness these<br />
5 For a comprehensive account of complex<br />
societies <strong>and</strong> co-evolution see Tainter (1988)<br />
6 Embedded governance structures are defined<br />
as those local cognitive, cultural, structural <strong>and</strong><br />
political institutions that allocate environmental<br />
resources (Zukin & DiMaggio, 1990)<br />
governance structures operate quite differently<br />
from those disembedded governance structures<br />
at the level of the autonomous community of<br />
Valencia, Madrid, Brussels <strong>and</strong> the numerous<br />
international water management forums.<br />
3.0 Methodology<br />
The first exercise was to analyse the<br />
contemporary literature relating to co-evolution,<br />
resilience <strong>and</strong> adaptive potential <strong>and</strong> to make an<br />
assessment of its relevance or non-relevance for<br />
the study. Three key references were selected:<br />
Norgaard’s (1995) approach to co-evolution<br />
where all variables are endogenous; Panarchical<br />
connections (Gunderson & Holling, 2002) that<br />
attempts to reduce complexity to a single theory<br />
of metaphor with which to describe resilience<br />
<strong>and</strong> adaptive potential through the use of a<br />
nested set of adaptive cycles; <strong>and</strong> McGlade’s<br />
(2002) mosaic approach to connect the<br />
hierarchal overlapping of systems, as well as to<br />
accommodate the emergence of new systems<br />
The methodology is designed using a series of<br />
mosaics 7 to provide the required mapping <strong>and</strong><br />
interpretation space. (Eisenhuth, 2003) The<br />
mosaics take the following form. A geographic<br />
mosaic to characterise co-evolutionary<br />
processes between different l<strong>and</strong>-uses <strong>and</strong> their<br />
corresponding hydrological balances <strong>and</strong> a<br />
cultural mosaic with which to overlay social <strong>and</strong><br />
institutional layers to detect differential<br />
response to change (social resilience, or<br />
adaptive capacity) as well as to determine the<br />
relationships between water resources,<br />
ecological quality <strong>and</strong> sustainable community<br />
development.<br />
3.1 Applying mosaics<br />
To study changes in the l<strong>and</strong>scape,<br />
georeferenced l<strong>and</strong> use (LU) /l<strong>and</strong> cover (LC)<br />
maps (1:10.000) from aerial photographs (1956,<br />
1978, 2000) of the study areas are created<br />
(Dunn et al. 1991), <strong>and</strong> 6 LU/LC categories:<br />
urban, dry crops, irrigation crops, shrubl<strong>and</strong>s,<br />
woodl<strong>and</strong>s <strong>and</strong> others (water bodies, dams)<br />
were defined. The analysis of this<br />
chronosequence is executed through the<br />
importation of created maps using GIS. In<br />
addition, the proportional rate change for each<br />
l<strong>and</strong> use can be determined. Finally, the<br />
7 A mosaic provides a cognitive map that<br />
represents units of spatial heterogeneity in<br />
which objects are aggregated to form a structure<br />
that can change over time.<br />
848
proportional historical changes for each l<strong>and</strong> use<br />
are calculated. The proposition is that selected<br />
quantitative indicators can be used to indicate<br />
changes in the l<strong>and</strong>scape structure. (Farina,<br />
1998)<br />
To model <strong>and</strong> evaluate the changes, a simple<br />
non-spatial l<strong>and</strong>scape model analogous to the<br />
method of Markov-chain transition probabilities<br />
for each two-year combination of l<strong>and</strong> use<br />
changes (1956-1978, 1978-2000 <strong>and</strong> 1956-<br />
2000) is constructed (Dale et al., 2002). The<br />
alterations that the l<strong>and</strong> use changes have on the<br />
ecological quality of the zones are evaluated<br />
through qualified transition model matrices,<br />
where ecological complexity, environmental<br />
stability <strong>and</strong> the sustainability of the water<br />
management system are considered.<br />
These criteria of transition qualification take<br />
into account the increase in or decrease of the<br />
ecosystem <strong>and</strong> water indicators such as the<br />
vegetation biomass, successional status, the<br />
potential to reverse a change, variation in water<br />
consumption, the evaporation <strong>and</strong> evapotranspiration<br />
of vegetation cover <strong>and</strong> l<strong>and</strong><br />
fragmentation. Transitions are then grouped in<br />
terms of the processes that the territory has<br />
experienced (succession, degradation, etc.), in<br />
progressive (P), degradative (D) <strong>and</strong> stable (S)<br />
with a possible combination of 5x5=25 types of<br />
transition. Thus, any transformation of pine<br />
forest or natural vegetation would be<br />
degradative, meanwhile transformations of<br />
urban l<strong>and</strong> use would be progressive. In this<br />
context the transition of dryl<strong>and</strong> crops to<br />
irrigated crops could be qualified as<br />
degradative, meanwhile irrigated to dryl<strong>and</strong><br />
crop is progressive as identified in the<br />
Transition Matrix Analysis. (Refer Figure 1)<br />
Woo<br />
dl<strong>and</strong><br />
Degra<br />
dative<br />
Degra<br />
dative<br />
Degra<br />
dative<br />
Degra<br />
dative<br />
Figure 1. Transition matrix analysis.<br />
Stable<br />
Hydrological balances are modelled <strong>and</strong> used to<br />
estimate contributions to the Marina Baixa<br />
hydrological system, during the study period<br />
using water balance models at plot level for<br />
each l<strong>and</strong> use type (Bellot et al, 1999), <strong>and</strong> l<strong>and</strong><br />
surfaces calculated from LU/ LC maps, thematic<br />
maps produced by GIS, <strong>and</strong> climatic data series<br />
provided by meteorological stations. Finally,<br />
the implications of changes in water use in the<br />
study area can be analysed by means of the<br />
comparison with the water flows in each main<br />
l<strong>and</strong> use type <strong>and</strong> the general balance (Refer<br />
Fig. 2) of the study zone during the period 1956<br />
to 2000.<br />
Figure 2 Marina Baixa general balance<br />
model<br />
3.2 Formulation of policy questions using<br />
geographic <strong>and</strong> cultural mosaics<br />
Urba<br />
n<br />
Dry<br />
crops<br />
Irrig<br />
ated<br />
crops<br />
Shru<br />
bl<strong>and</strong><br />
Urba<br />
n<br />
Stable<br />
Degra<br />
dative<br />
Degra<br />
dative<br />
Degra<br />
dative<br />
Dry<br />
crops<br />
Progre<br />
ssive<br />
Stable<br />
Progre<br />
ssive<br />
Degra<br />
dative<br />
Irriga<br />
ted<br />
crops<br />
Progre<br />
ssive<br />
Degra<br />
dative<br />
Stable<br />
Degra<br />
dative<br />
Shrub<br />
l<strong>and</strong><br />
Progre<br />
ssive<br />
Progre<br />
ssive<br />
Progre<br />
ssive<br />
Stable<br />
Wood<br />
l<strong>and</strong><br />
Progr<br />
essive<br />
Progr<br />
essive<br />
Progr<br />
essive<br />
Progr<br />
essive<br />
The Marina Baixa represents an overlapping<br />
hierarchical arrangement of systems that can be<br />
better interpreted using mosaics. In this sense,<br />
each of the 18 municipalities is described as a<br />
territory, or population, that interacts one with<br />
the other, using a geographical grid in a time<br />
sequence <strong>and</strong> modelling transitions. Using a<br />
mosaic approach it is possible to analyse the<br />
socio-economic behaviour <strong>and</strong> the<br />
appropriateness of governance structures in the<br />
three water using municipalities by contrasting<br />
the structures, processes <strong>and</strong> dynamics that have<br />
driven l<strong>and</strong>-use patterns. Some examples of the<br />
policy questions that have been formulated are<br />
as follows: What drives competition in dem<strong>and</strong><br />
for water for contrasting uses:, eg., tourism<br />
versus irrigated <strong>and</strong> dryl<strong>and</strong> crops? How price<br />
is influenced by dem<strong>and</strong>? Why alliances were<br />
849
formed to manage the resource more efficiently,<br />
<strong>and</strong> why some municipalities have elected not to<br />
be part of such alliances? What drives changes<br />
in human behaviour e.g., in times of water<br />
shortages <strong>and</strong> climatic limitations? How to<br />
interpret the water budget of uniform<br />
hydrological response units (Bellot et al. 1999)<br />
using different types of models? These units can<br />
be LU/LC categories.<br />
Using this simplified process-based approach it<br />
is possible to select inputs <strong>and</strong> outputs, e.g.,<br />
rainfall, aquifer pumping, evapo-transpiration,<br />
run-off <strong>and</strong> then to calculate the budget for the<br />
Marina Baixa using the surfaces of each l<strong>and</strong><br />
use calculated by using GIS. The results of the<br />
models form a kind of metric to inform endusers<br />
regarding the severity of water deficit or<br />
the lowering of aquifer levels (Bellot et al.<br />
2001.) calculated for each study period: 1956,<br />
1978, 2000. In some cases the hydrological<br />
budget in each unit of l<strong>and</strong> changes over time.<br />
There may be an increase in urban supply<br />
because of increases to per capita water use.<br />
Conversely, there may be a decrease in water<br />
use for irrigated crops if technology is<br />
improved. In other L/Us such as shrubl<strong>and</strong> or<br />
woodl<strong>and</strong> it is possible to isolate smaller<br />
changes in water usage over the same time<br />
scale. The rates of change from one category to<br />
another can be modelled according to social or<br />
economic trends observed by other researchers<br />
from the study group.<br />
Other policy questions relate to the effect of<br />
water transfers on ecological quality <strong>and</strong> social<br />
resilience <strong>and</strong> if this could lead to the<br />
emergence of a socio-natural system that can<br />
respond to surprise or shock. This can be<br />
determined by analysing the relationship<br />
8<br />
6.0 References<br />
Bellot, J., Sanchez, J.R., Chirino, E., Hern<strong>and</strong>ez,<br />
N., Abdelli, F., & Martinez, J.M. (1999) Effect<br />
of different vegetation type cover effects on the<br />
soil-water balance in semi-arid areas of southeastern<br />
Spain, Physics <strong>and</strong> Chemistry of the<br />
Earth (B) Vol. 24, 353-357<br />
Bellot, J., Bonet, A., Sánchez, J.R., & Chirino<br />
E., (2001) Likely effects of l<strong>and</strong> use changes on<br />
the runoff <strong>and</strong> aquifer recharge in a semiarid<br />
l<strong>and</strong>scape using a hydrological model,<br />
L<strong>and</strong>scape <strong>and</strong> Urban Planning, 778, 1-13<br />
between l<strong>and</strong>-use change <strong>and</strong> water usage, as<br />
well as determining what types of institutions,<br />
(viz., embedded or disembedded) nurture social<br />
resilience, or receptivity to change. Finally, how<br />
knowledge about this socio-natural system is<br />
transferred. This list of policy questions is not a<br />
conclusive list as integrative research has to be<br />
an iterative process. (Winder, 2002)<br />
4.0 Conclusions<br />
The strengths associated with the use of a coevolutionary<br />
approach can be best described as<br />
follows. A better opportunity to pool individual<br />
disciplinary knowledge relating to adaptive,<br />
integrated water resource management. Greater<br />
potential to track the pace <strong>and</strong> tempo of<br />
reciprocal evolutionary processes. The means<br />
with which to assess the impact of the politics of<br />
water transfers as they are applied to a situation<br />
that is currently unsustainable. An opportunity<br />
to analyse institutional transformation <strong>and</strong><br />
transaction costs <strong>and</strong> the implications these have<br />
for water management efficiency. Finally, a<br />
clearer underst<strong>and</strong>ing of the terms resilience <strong>and</strong><br />
adaptive potential <strong>and</strong> how these concepts relate<br />
to adaptive, integrated water resource<br />
management of a semi-arid region.<br />
The major challenges lie with selection of a<br />
research methodology for adapting the analogy<br />
of co-evolution for socio-natural policy relevant<br />
research. And then to identify the populations<br />
that could be co-evolving. Another challenge is<br />
the framing of policy questions <strong>and</strong> the impact<br />
this has for the future direction of governance<br />
arrangements, the formulation of announced<br />
policy such that it can be translated into policy<br />
in practice <strong>and</strong> technology choice.<br />
Bellot, J., Peña, J., Tejada, J., Bonet, A., &<br />
Sánchez, J.R., (2003), Implicaciones De Los<br />
Cambios De Uso Del Territorio Sobre Los<br />
Balances Hídricos En La Marina Baixa<br />
(Alicante), II Jornadas Ibéricas de Ecología<br />
Paisaje, ‘Presente y Futuro de la Ecología del<br />
Paisaje en la Península Ibérica’, IALE-España,<br />
Alcalá de Henares, Madrid, 24,25 y 26<br />
Septiembre de 2003<br />
Blondel, J., Aronson, J., (1999) Biology <strong>and</strong><br />
Wildlife of the Mediterranean Region, Oxford,<br />
Oxford University Press<br />
8 Acknowledgements: The authors gratefully acknowledge the comments made by Roger Seaton <strong>and</strong><br />
Paul Jeffrey. This paper was written as a contribution to the ‘AQUADAPT’ project (‘Strategic tools to<br />
support adaptive, integrated water resource management under changing conditions at catchment scale: A<br />
co-evolutionary approach’) supported by the EC under contract EVK1-CT-2001-00104<br />
850
Dale, M., Dale, P., & Edgoose, T., (2002) Using<br />
Markov models to incorporate serial<br />
dependence in studies of vegetation change,<br />
Acta Oecologica, 23, 261-269<br />
Eisenhuth, D., (2003) Intellectual orientations<br />
for underst<strong>and</strong>ing the reality of water<br />
governance structures, The AQUADAPT<br />
Papers, www.aquadapt.net<br />
Farina, A., (1998) Principles <strong>and</strong> Methods in<br />
L<strong>and</strong>scape Ecology, Chapman & Hall,<br />
Cambridge<br />
Gunderson, L.H., L.H. & Holling C.S., C.S.,<br />
(2002), Resilience <strong>and</strong> Adaptive Cycles, in<br />
Gunderson, L.H., & Holling, C.S., (eds.)<br />
Panarchy Underst<strong>and</strong>ing Transformations in<br />
Human <strong>and</strong> Natural Systems, Washington,<br />
Isl<strong>and</strong> Press, pp.25-62<br />
Hunsaker, C.T., O’Neill, R.V., Jackson, B.L.,<br />
Timmins, S.P., Levine, D.A. & Norton, D.J.,<br />
(1994) Sampling to characterize l<strong>and</strong>scape<br />
pattern, L<strong>and</strong>scape Ecology 9, 207-226<br />
Schön, D.A., & Rein, M., (1994) Frame<br />
Reflection Towards the Resolution of<br />
Intractable Policy Controversies, Basic Books,<br />
New York<br />
Tainter, J.A., (1988) The Collapse of Complex<br />
Societies, Cambridge University Press,<br />
Cambridge<br />
Turner, M, Gardner, R., & O’Neill, R.V.,<br />
(2001) L<strong>and</strong>scape Ecology in theory <strong>and</strong><br />
practice. Pattern process, Springer Verlag. 401,<br />
New York<br />
Winder, N. (2002) Why Integrative Research is<br />
like Herding Weasels, Discussion Paper<br />
prepared for the AQUADAPT Workshop,<br />
Montpelier, October 25-27, 2002<br />
Zukin, S. & DiMaggio, P., (1990) Structures of<br />
Capital, the Social Organisation of the<br />
Economy, Cambridge University Press,<br />
Cambridge<br />
Jeffrey, P. & McIntosh, B., (2004) Coevolutionary<br />
theory as a conceptual model in the<br />
search for sustainable modes of water<br />
management, The AQUADAPT Brief, The<br />
AQUADAPT Papers, www.aquadapt.net<br />
McGlade, J., (2002) L<strong>and</strong>scape Sensitivity,<br />
Resilience <strong>and</strong> Sustainable Watershed<br />
Management: A Co-evolutionary perspective,<br />
Discussion paper prepared for the AQUADAPT<br />
Workshop, Montpelier, October 25-27, 2002<br />
McIntosh, B. & Jeffrey, P., (2003) Transferring,<br />
interpreting <strong>and</strong> using theories of biological<br />
(co)evolution in socio-natural science: A reply<br />
to Rammel <strong>and</strong> Staudinger (in print)<br />
<strong>International</strong> Journal of Sustainable<br />
Development <strong>and</strong> Global Change<br />
Norgaard, R., (1995) Development Betrayed the<br />
end of progress <strong>and</strong> a co-evolutionary<br />
revisioning of the future, Routledge, London<br />
Perry, G.L.W. (2002) L<strong>and</strong>scapes, space <strong>and</strong><br />
equilibrium: some recent shifts, Progressive<br />
Physical Geography 26, 339-359<br />
Richards (1990) L<strong>and</strong> Transformations. in<br />
Turner, B.L., II et al (eds)The Earth as<br />
Transformed by Human Action, Global <strong>and</strong><br />
Regional Changes in the Biosphere Over The<br />
Past 300 Years, Cambridge University Press,<br />
London<br />
851
as<br />
Stability Analyses of the 50/50 Sex Ratio<br />
Using Lattice Simulation<br />
Y. Itoh, K. Tainaka <strong>and</strong> J. Yoshimura<br />
Department of Systems Engineering, Shizuoka University,<br />
3-5-1 Johoku, Hamamatsu 432-8561 Japan<br />
Abstract: The observed sex ratio is nearly one half in many animals including humans. Fisher explained that<br />
the 50/50 sex ratio is optimal. However, the 50/50 sex ratio seems highly unstable because a slight deviation<br />
from 50/50 changes the optimal ratio to the opposite extremes; zero or unity. Thus sex ratio should be<br />
fluctuating wildly around 50/50. In contrast, the observed 50/50 sex ratios in wild populations seem to be<br />
very stable. There should be some unknown mechanisms to stabilize the sex ratio 50/50. We build the lattice<br />
model of mating populations. Each cell represents a male, a female, or an empty site. We perform lattice<br />
simulation by two different methods: local interaction (lattice model) <strong>and</strong> global interaction. In the case of<br />
lattice model, chance of reproduction is determined based on the numbers of males <strong>and</strong> females adjacent to<br />
the vacant site. In the global interaction method, we select four sites r<strong>and</strong>omly instead of the adjacent sites.<br />
The highest density is achieved in the 50/50 sex ratio. This density peak st<strong>and</strong>s out sharply in the lattice<br />
simulation, but it is rather flat in the global interaction. With a high mortality, the 50/50 sex ratio becomes the<br />
sole survivor; all other ratios becomes extinct. The stability <strong>and</strong> persistence of the 50/50 sex ratio becomes<br />
evident especially in a harsh environment. The superiority of the 50/50 sex ratio in the lattice model is due to<br />
the decreased chance of mating in a local site, known as the Allee effect. Our approach can extend to show<br />
that any value of sex ratio is evolutionary stable.<br />
Keywords: Sex ratio; Lattice model; Allee effect; Evolutionary maintainable strategy<br />
1. INTRODUCTION<br />
The observed sex ratio is almost 50/50 in many<br />
animals [Charnov, 1982]. The 50/50 sex ratio is<br />
first explained by Fisher [1930]. Using the concept<br />
of evolutionary stability, it is described as follows.<br />
If there are many more males in the population, a<br />
female parent producing more females becomes<br />
adaptive because, being a parent of many<br />
daughters, she produces more gr<strong>and</strong>children than a<br />
female with less daughters. In the opposite case,<br />
producing more males is adaptive. Only when the<br />
sex ratio is 50/50, a parent with 50/50 males or<br />
females becomes balanced. This Fisher’s 50/50 sex<br />
ratio is an evolutionarily stable strategy (ESS).<br />
However, the stability of sex ratio is highly<br />
questionable. A slight deviation from 50/50 sex<br />
ratio in either directions results immediately in the<br />
optimality of producing only one sex opposite to<br />
the deviation. If the sex ratio is slightly femalebiased,<br />
then the strategy producing just only males<br />
(sex ratio<br />
˺=1) is the most adaptive. In contrast,<br />
with a slight male-bias, the optimal sex ratio<br />
becomes 0. Furthermore, when the sex ratio of the<br />
population is exactly 50/50, a parent with 1 sex<br />
ratio (all sons) or sex ratio 0 (all daughters) leaves<br />
nearly equal numbers of gr<strong>and</strong>children compared<br />
with a parent with 50/50 sex ratio. Thus, even<br />
though 50/50 sex ratio is an ESS, it seems highly<br />
unstable like a ropewalking. We need an<br />
explanation for the observed stability of 50/50 sex<br />
ratio in wild animal populations.[Bulmer, 1994]<br />
Our study applys not ESS but evolutionary<br />
maintainable strategy (EMS). The strategy of EMS<br />
takes the maximum value of the population size in<br />
stationary state. EMS is the most sustainable<br />
strategy. Throughout this paper, we represent the<br />
sex ratio value the male ratio. The sex ratio<br />
˺=0.7 means that, out of 10 individuals, 7 are<br />
males <strong>and</strong> 3 females.<br />
˺<br />
We build a two-dimensional lattice model. In the<br />
lattice model individuals are spatially distributed.<br />
The reproduction of an individual depends on<br />
adjacent individuals. In each simulation we fix the<br />
sex ratios of newborn offsprings <strong>and</strong> measure the<br />
852
steady-state densities. In the simulation<br />
experiments, we also employ global interaction,<br />
where no spatial structure is assumed. In the global<br />
interaction, the reproduction depends on the<br />
individuals r<strong>and</strong>omly chosen from the entire<br />
population. The steady-state densities of global<br />
interaction is analyzed mathematically.<br />
In both lattice model <strong>and</strong> global interaction, the<br />
boundary of extinction is determined by the Allee<br />
effect in the chance of sexual reproduction. We<br />
first discuss the Allee effect. Then we describe the<br />
lattice <strong>and</strong> global interaction models. We next<br />
analyse the steady-state densities of global<br />
interaction. The simulation results of both lattice<br />
model <strong>and</strong> global interaction is explained next.<br />
Finally we discuss the stability of 50/50 sex ratio.<br />
2. THE ALLEE EFFECT<br />
Our work is closely related to the Allee effect. If a<br />
population is not so small, it is generally persistent<br />
<strong>and</strong> resistant against extinction. The risk of<br />
extinction drastically increases, when the<br />
population size becomes below a critical number of<br />
individuals. The Allee effect denotes such a<br />
threshold termed the minimum viable population<br />
(MVP).<br />
dx<br />
dt<br />
( x − a)( b − x)<br />
= Rx<br />
, (1)<br />
where x(t) is the population size, <strong>and</strong> R, a <strong>and</strong> b are<br />
positive constants. Equation (1) has three<br />
equilibriums; both states x=0 <strong>and</strong> x=b are stable,<br />
whereas the state x=a is unstable. In Fig. 1, typical<br />
population dynamics of (1) is shown. When an<br />
initial value of x exceeds the constant a, then x<br />
recovers the stationary value b. In contrast, when it<br />
is below the value a, then x becomes zero. Hence,<br />
the parameter a corresponds to the MVP size. To<br />
avoid the extinction, it is necessary that the<br />
population size exceeds the value a.<br />
3. MODEL<br />
We consider population composed of a single<br />
species on a two-dimensional lattice. Each lattice<br />
site is occupied by a male (M), a female (F) or<br />
empty (O). Birth <strong>and</strong> death processes are given by<br />
(2a)<br />
(2b)<br />
(2c)<br />
The processes (2) simulate reproduction <strong>and</strong> death.<br />
Figure 1. The conceptual population dynamics of<br />
Allee effect. If the initial size is larger than<br />
threshold value a, the population goes to stationary<br />
state b. If the initial size is smaller than a, the<br />
population goes extinct.<br />
Possible mechanisms for Allee effect may be the<br />
necessity of finding a mate for reproduction,<br />
satiation of a generalist predator, group defense<br />
against predators, or preservation of genetic<br />
diversity. A simple mathematical model of Allee<br />
effect is given by<br />
Figure 2. Schematic illustration of the model. The<br />
central site (empty O) is surrounded by two males<br />
<strong>and</strong> one female. Therefore, the reproduction rate at<br />
the site is B = r × 2 × 1 = 2r<br />
. Therefore the<br />
reproductive probability is 2rα<br />
for male <strong>and</strong><br />
2r (1 − α ) for female. Note that if either sex is<br />
absent, B becomes zero.<br />
853
e<br />
The parameter m represents the mortality rate of an<br />
individual. The parameter B is the reproduction<br />
rate that is determined by equation<br />
B = rP M<br />
P F<br />
(3)<br />
where P M <strong>and</strong> P F represents the number of males<br />
<strong>and</strong> females, respectively, adjacent to the vacant<br />
lattice site (the cental site in Fig. 2). The<br />
reproductive parameter r is set to a constant (one).<br />
The birth process (2a) is carried out in two ways:<br />
lattice model <strong>and</strong> global interaction. The following<br />
is the method of lattice model simulation.<br />
i) Distribute species r<strong>and</strong>omly over some squarelattice<br />
points in such a way that each point is<br />
occupied by only one individual.<br />
ii) Choose one lattice site r<strong>and</strong>omly. If the site is<br />
empty (O), perform birth process defined in<br />
reaction (2a) where the reproduction rate B is<br />
determined by the states of adjacent four lattice<br />
sites. Here we employ periodic boundary<br />
conditions.<br />
iii) Choose one lattice site r<strong>and</strong>omly. If the site is<br />
occupied by an individual (male or female),<br />
perform death process defined in reaction (2b) or<br />
(2c).<br />
iv) Repeat step (ii) <strong>and</strong> (iii) by L x L times, where<br />
L x L is the total number of the square-lattice sites.<br />
This step is called a Monte Carlo step [Tainaka,<br />
1988].<br />
v) Repeat the step (iv) until the system reaches<br />
stationary state.<br />
Throughout the simulation, each individual on a<br />
lattice site is assumed not to move: this assumption<br />
is applicable for plant, <strong>and</strong> may be approximately<br />
valid even for animals, provided that the radius of<br />
action of an individual is much shorter than the<br />
size of the entire system [Satulovsky <strong>and</strong> Tome,<br />
1994]. Empirical data suggest interactions occur<br />
locally; long-ranged interaction is rather<br />
exceptional [Price <strong>and</strong> Waser, 1979]. In the case of<br />
lattice model, reproduction rate B is determined by<br />
adjacent four lattice sites.<br />
In contrast, in the case of global interaction<br />
simulation, the long-ranged interaction is allowed.<br />
Here the reproduction rate B is determined by any<br />
four r<strong>and</strong>omly chosen lattice sites.<br />
4. THEORY FOR GLOBAL INTERACTION<br />
In the case of global interaction, we can obtain the<br />
evolution equation for the population dynamics.<br />
Let the sex ratio of male, <strong>and</strong> x, y be the<br />
population sizes (densities) of male <strong>and</strong> female,<br />
˺<br />
respectively. Hence, the density of empty site is<br />
expressed by ( 1−<br />
x − y)<br />
. The time dependences of<br />
both densities are given by<br />
dx<br />
dt<br />
dy<br />
dt<br />
−mx<br />
+ crxy( 1−<br />
x − y)α<br />
= , (4a)<br />
= −my<br />
+ crxy( 1−<br />
x − y)(1<br />
−α)<br />
. (4b)<br />
The first term in the right h<strong>and</strong> side of equation (4)<br />
represents the death process <strong>and</strong> the second term<br />
corresponds to the birth process. Note that the<br />
parameter c in (4) is the reproductive constant<br />
(c=16), because P M<br />
= 4x<br />
<strong>and</strong> P F<br />
= 4 y .<br />
Equation (4) is similar as reported in the previous<br />
papers (Boukal & Berec, 2002; Hopper & Roush,<br />
1993). However, equation (4) has a distinct<br />
advantage: it has fast <strong>and</strong> slow dynamics, so that<br />
stationary densities are easily obtained. The fast<br />
dynamics comes from the fact that the ratio of<br />
female to the male in the birth process is constant.<br />
Namely, the system rapidly (exponentially)<br />
approaches to the following condition:<br />
/ x = (1 −α ) /α<br />
y (5)<br />
The stationary state of basic equation (4) should<br />
satisfy the condition (5).<br />
On the other h<strong>and</strong>, the slow dynamics is given from<br />
equations (5) <strong>and</strong> (4a) [or (5) <strong>and</strong> (4b)]. It follows<br />
that<br />
dx<br />
dt<br />
= (6)<br />
2<br />
−mx<br />
+ r(1<br />
−α ) x ( α − x)<br />
/ α<br />
Equation (6) is just the same as (1). The steadystate<br />
densities a in equation (1) are given by<br />
a = ( α − D) / 2 , (7a)<br />
b = ( α + D) / 2 , (7b)<br />
where D is the discriminant:<br />
2 4m<br />
α<br />
D = α − • . (8)<br />
r 1−<br />
α<br />
For the survival, it is necessary for D to be positive.<br />
In contrast, when D ≤ 0 or when<br />
α ( 1−α)<br />
≤ 4m / r , (9)<br />
extinction always occurs; there is only one stable<br />
equilibrium ( x = y = 0 ). When<br />
˺(1-˺) takes a<br />
small value, then the target population goes extinct.<br />
Thus a kind of phase transition is predicted<br />
according to either equation (9) holds or not. The<br />
854
population size of male in stable equilibrium is<br />
given by b. Similarly, that of female is b(1-˺)/˺.<br />
Hence, the total population size in stable state is<br />
given by<br />
x y = b + b( 1−α ) / α = b / α<br />
+ . (10)<br />
Concrete form of this equation is expressed by<br />
1<br />
+ y = +<br />
2<br />
1 m<br />
−<br />
4 rα(1<br />
−α)<br />
x . (11)<br />
rather flat. In contrast, the density peak become<br />
pointed at the 50/50 sex ratio in the lattice<br />
simulation (Fig. 4).<br />
In summary, for the lattice model, a pointed peak is<br />
observed near the sex ratio<br />
˺=0.5. In contrast, for<br />
the global interaction, a peak is not pointed but<br />
rather plateau (Figs. 2-4).<br />
The total population size for survival phase<br />
becomes maximum at 50/50 sex ratio, <strong>and</strong> it<br />
always exceeds 1/2.<br />
5. RESULTS<br />
Simulations are performed for different values of<br />
mortality rate m <strong>and</strong> sex ratio<br />
˺. In Figs. 2-4, the<br />
steady-state density of a population is shown for<br />
both lattice model (open squares) <strong>and</strong> global<br />
interaction (filled points). Here we set the mortality<br />
level of the lattice model 10 times lower than that<br />
of global interaction, because a lattice population<br />
is much easier to go extinct.<br />
In the case of global interaction, the phase<br />
transition occurs:<br />
i) the population goes extinct, when inequality (9)<br />
holds.<br />
ii) it survives, when the condition (9) does not hold.<br />
The phase transition discontinuously occurs. In the<br />
case of lattice model, the phase transition is also<br />
observed; however, it continuously occurs.<br />
Under a low mortality condition (Fig. 2), the range<br />
of sustainable sex ratios are between 0.3 <strong>and</strong> 0.7 in<br />
the lattice simulation (squares). The population<br />
goes extinct if the sex ratio is out of this<br />
sustainable range. The sustainable range exp<strong>and</strong>s<br />
between 0.05 <strong>and</strong> 0.95 with global interactions<br />
(filled points in Fig. 2). The phase transition to the<br />
extinction for global interactions are described in<br />
the section 2. The spatial reproductive structure<br />
strongly enhances the stability of the 50/50<br />
adaptive sex ratio. Under a moderate-level<br />
mortality condition (Fig. 3), the range of<br />
sustainable sex ratios narrows to that between 0.4<br />
<strong>and</strong> 0.6 in the lattice model. In contrast, the range<br />
is still wide (between 0.1 <strong>and</strong> 0.9) in the global<br />
interaction version. In a high mortality condition<br />
(Fig. 4), the population survives at a low steadystate<br />
density in a very narrow range of sex ratios,<br />
or goes extinct. Here the sustainable range become<br />
narrower even under global interaction. However,<br />
within this range the steady-state densities are<br />
Figure 2. The steady-state densities for sex ratio<br />
˺.<br />
The open squares are for lattice model (m = 0.01),<br />
the filled points are for global interaction<br />
(mortality rate m = 0.1),<br />
Figure 3. Same as Fig. 2 , but the mortality rate is<br />
0.03 for lattice model, 0.3 for global interaction.<br />
855
Figure 4. Same as Fig .2 , but the mortality rate is<br />
0.07 for lattice simulation, 0.7 for global<br />
interaction.<br />
6. DISCUSSIONS<br />
Simulations are carried out for two methods: the<br />
lattice model <strong>and</strong> the global interaction. There are<br />
a sustainable range of sex ratios for the persistence<br />
of populations in both lattice model <strong>and</strong> global<br />
interaction. Outside of this range, the populations<br />
are destined to extinction. In this sense, the results<br />
of both approaches are qualitatively same.<br />
However, a markedly distinctive quantitative<br />
difference is observed between the two methods.<br />
The chance of mating extremely limited in the<br />
lattice model. We set mortality of the lattice model<br />
10 times lower than that of the global interaction.<br />
Sharp peaks are observed for lattice model but<br />
plateaus for global interaction (Figs. 2-4).<br />
Allee effects appears in the global interaction. The<br />
steady-state densities goes to zero (extinction)<br />
abruptly when the deviation from 50/50 sex ratio<br />
increased gradually. Here the boundary is a step<br />
function from a finite number to zero (Equations 6-<br />
10, see also Figs. 2-4). Due to the sex ratio<br />
deviation, the reproduction rate cannot cope with<br />
the mortality rate, resulting in the negative<br />
population growth. Within this boundary, the<br />
average density changes slightly with the<br />
difference in sex ratio.<br />
Allee effects also play an important role in the<br />
lattice model. In the lattice model, the chance to<br />
reproduce is dependent on the existence of both<br />
sexes in each confined locality rather than the<br />
entire population. With little deviation of sex ratio<br />
from 50/50 range, monosexual regions are easily<br />
formed, <strong>and</strong> the opportunity of reproduction goes<br />
to zero in these regions. Because of the locality<br />
problem, the thresholds of the sex ratio deviation<br />
appears much narrower in the lattice model than<br />
those in the global interaction. However, due to the<br />
averaging of the Allee effects over the entire lattice<br />
space, the density decline profile along the sex<br />
ratio becomes gradual unlike the step function in<br />
the global interaction (Figs. 2-4). Thus the lattice<br />
model relates the localized Allee effect.<br />
The stability of 50/50 sex ratio may be strongly<br />
enhanced by the Allee effect. This effect is severe<br />
when the population is “philopatric” (lattice<br />
model) rather than “panmictic” (global interaction).<br />
In both cases, the deviation from 50/50 ratio is<br />
extremely costly when the population is under<br />
severe environmental conditions.<br />
The current model may be applicable to the<br />
stability analyses of non-50/50 sex ratios. If we<br />
apply the sexual difference in mortality or<br />
reproductive values, then we expect that the<br />
optimal sex ratio takes a value of non-50/50.<br />
Our approach may account for the superiority of<br />
sexual reproduction. It is well known that the<br />
reproduction rate of asexual reproduction is much<br />
larger than that of sexual reproduction (twice). If<br />
we apply the concept of ESS without trade-off,<br />
then the asexual reproduction beats sexual one.<br />
However, our approach insists the importance of<br />
sustainability (steady-state density). When we<br />
apply the concept of EMS, the superiority of sexual<br />
reproduction can be explained. This is because the<br />
sexual reproduction is more maintainable; an<br />
example of such sustainability is the Red-Queen<br />
effect [Hamilton, 1980]. Moreover, it is known that<br />
the increase of reproduction rate does not mean the<br />
increase of steady-state density (paradox of<br />
enrichment) [Rosenzweig, 1971]. Interaction with<br />
other species may play an important role in the<br />
advantage of sex.<br />
In this century, many environmental problems will<br />
be serious. Among all, mass extinction of<br />
biospecies is one of the most serious problems. In<br />
order for a biospecies to avoid extinction, it is<br />
necessary that its population size is not so small.<br />
The risk of extinction drastically increases, when<br />
the population size becomes below a critical<br />
number of individuals. Such a threshold has been<br />
termed the minimum viable population (MVP).<br />
Several authors have empirically tried to estimate<br />
the MVP for a number of different organisms.<br />
Belovsky [1987] investigated a variety of mammal<br />
species, <strong>and</strong> determined that populations of several<br />
thous<strong>and</strong> individuals were necessary to achieve a<br />
95% chance of persistence for 100 years. Thomas<br />
[1990] made MVP recommendations for birds <strong>and</strong><br />
mammals that ranged from 1000 to 10,000.<br />
Similarly, Wilcove, et al. [1993] analysed the<br />
endangered species in the United States <strong>and</strong><br />
concluded that a minimum of 1000 individuals<br />
856
were necessary for persistence. Hence, empirical<br />
data suggest that MVP takes large values. Our<br />
approach of lattice model may explain such large<br />
values of MVP.<br />
Wilcove, D. S., McMillan, M. <strong>and</strong> Winston, K. C.,<br />
What exactly is an endangered species? An<br />
analysis of the U.S. endangered species list:<br />
1985-1991, Conservation Biology, 7, 87-93,<br />
1993.<br />
7. ACKNOWLEDGEMENTS<br />
This work was partially supported by grant-in-aids<br />
from the ministry of Education, Culture <strong>and</strong><br />
Science of Japan to K. T. <strong>and</strong> to J. Y.<br />
8. REFERENCES<br />
Allee, W. C., Animal Aggregations: a Study in<br />
General Sociology, University of Chicago<br />
Press, Chicago, 1931.<br />
Belovsky, G. E., Extinction models <strong>and</strong><br />
mammalian persistence, Viable Populations<br />
for Conservation. Cambridge University Press,<br />
Cambridge, 1987.<br />
Boukal, D.S. <strong>and</strong> L. Berec, Single-species models<br />
of the Allee effect: extinction boundaries, sex<br />
ratios <strong>and</strong> mate encounters, Journal of<br />
Theoretical Biology, 218, 375-394, 2002.<br />
Bulmer, M., Theoretical Evolutionary Ecology.<br />
Sinauer Associates, Sunderl<strong>and</strong>, 1994.<br />
Charnov, E. L., The Theory of Sex Allocation,<br />
Prinston University Press, Princeton, 1982.<br />
Fisher, R. A., The Genetical Theory of Natural<br />
Selection, Oxford University Press, Oxford,<br />
1930.<br />
Hamilton, W. D., Sex versus non-sex versus<br />
parasite, Oikos, 35, 282-290, 1980.<br />
Hopper, K. R. <strong>and</strong> R. T. Roush, Ecological<br />
Entomology, 18, 321-331, 1993.<br />
Price, M. V. <strong>and</strong> N. M. Waser, Pollen dispersal<br />
<strong>and</strong> optimal outcrossing in Delphinium<br />
nelsoni, Nature, 277, 294-297, 1979.<br />
Rosenzweig, M. L., The paradox of enrichment:<br />
destabilization of exploitation ecosystems in<br />
ecological time, Science, 171, 385-387, 1971.<br />
Satulovsky, J. E. <strong>and</strong> T. Tome, Physics Review,<br />
Stochastic lattice gas model for a predatorprey<br />
system, E49, 5073-9, 1994.<br />
Tainaka, K., Lattice model for the Lotka-Volterra<br />
system, Journal of the Physical Society of<br />
Japan, 57(88), 2588-2590, 1988.<br />
Tainaka, K. <strong>and</strong> N. Araki, Press Perturbation in<br />
Lattice Ecosystems: Parity Law <strong>and</strong> Optimum<br />
Strategy, Japanese Theoretical Biology, 197,<br />
1-13, 1999.<br />
Thieme, H. R., Mathematics in Population Biology,<br />
Princeton University Press, Princeton, 2003.<br />
Thomas, C. D., What do real population dynamics<br />
tell us about minimum viable population<br />
sizes?, Conservation Biology, 4, 324-327,<br />
1990.<br />
857
Reproductive Strategies of Marine Green Algae: the<br />
Evolution of Slight Anisogamy <strong>and</strong> the <strong>Environmental</strong><br />
Conditions of Habitats<br />
Tatsuya Togashi a, c , Tatsuo Miyazaki a , Jin Yoshimura a, b , John L. Bartelt c <strong>and</strong> Paul Alan Cox c<br />
a Marine Biosystems Research Center, Chiba University, Amatsu-Kominato, 299-5502, Japan.<br />
e-mail: togashi@faculty.chiba-u.jp<br />
b Department of Systems Engineering, Shizuoka University, Hamamatsu, 432-8561, Japan.<br />
c National Tropical Botanical Garden, 3530 Papalina Road, Kalaheo, HI 96741, USA.<br />
Abstract: In marine green algae, isogamous or slightly anisogamous species are taxonomically widespread.<br />
They produce positively phototactic gametes with phototactic devices including an eye-spot in both sexes.<br />
We developed a new numerical simulator of gamete behavior using C++ <strong>and</strong> pseudo-parallelization methods<br />
to elucidate potential advantages of phototaxis. Input parameters were set based on experimental data. Each<br />
gamete swimming in a virtual rectangular test tank was tracked <strong>and</strong> the distances between the centers of<br />
nearby male <strong>and</strong> female were measured at each step to detect collisions. Our results shed light on the roles of<br />
gamete behavior <strong>and</strong> the mechanisms of the evolution of anisogamy <strong>and</strong> more derived forms of sexual<br />
dimorphism. We demonstrated that not only gametes with positive phototaxis were favored over those<br />
without particularly in shallow water because they could search for potential mates on the two-dimensional<br />
water surface rather than r<strong>and</strong>omly in three dimensions, but also phototactic behavior clarified the difference<br />
between isogamy <strong>and</strong> slight anisogamy. Isogamous species produced significantly more zygotes than slightly<br />
anisogamous ones only under the phototactic conditions. Our results suggested that “sperm limitation” might<br />
be resolved in the slightly anisogamous species. In marine green algae, some more markedly anisogamous<br />
species produce the smaller male gametes that have no eye-spot <strong>and</strong> swim r<strong>and</strong>omly. In contrast, the larger<br />
female gametes have an eye-spot <strong>and</strong> show positive phototaxis. As a result of careful experimental<br />
observations, we discovered the first pheromonal attraction system in marine green algae. This pheromonal<br />
attraction system might have played a key role in the evolution of anisogamy in marine green algae, because<br />
it may enable markedly anisogamous species achieve 2D search efficiencies on the water surface. The mating<br />
systems appear to be tightly tuned o the environmental conditions of their habitats.<br />
Keywords: Anisogamy; Gamete behavior; Marine green algae; Pheromonal attraction; Phototaxis<br />
1. INTRODUCTION<br />
Anisogamy with gametes of two different sizes is<br />
common to many organisms <strong>and</strong> only one<br />
universal difference between males <strong>and</strong> females<br />
[R<strong>and</strong>erson <strong>and</strong> Hurst, 2001]. This anisogamy<br />
underlies the evolution of sex differences in<br />
behavior <strong>and</strong> morphology, because it generates<br />
sexual selection whenever the number of small<br />
gametes produced by males exceeds the number<br />
necessary to fertilize the ova of a single female<br />
[Schuster <strong>and</strong> Wade, 2003]. Thus, sexual<br />
selection resulting from the variance in mate<br />
numbers of the sex producing small gametes does<br />
not exist in asexual populations.<br />
Two main theories have been proposed to account<br />
for the evolution of anisogamy. The one is a<br />
theory in which there is disruptive selection<br />
acting on gamete size based on the two<br />
conflicting selection forces of search efficiency<br />
<strong>and</strong> postzygotic survival [e.g. Parker et al., 1972],<br />
<strong>and</strong> the other is sperm limitation theory that<br />
considers an escape from sperm limitation as a<br />
mechanism driving anisogamy [e.g. Levitan,<br />
1996].<br />
In oogamous sea urchins, it has been reported that<br />
females are often sperm limited [Levitan, 1996].<br />
On the other h<strong>and</strong>, in marine green algae,<br />
isogamous or slightly anisogamous are<br />
taxonomically widespread. Their gametes not<br />
only have specific mating types, but also have a<br />
phototactic system with an eye-spot. It has been<br />
suggested that the eye-spot evolved in the most<br />
primitive green flagellate taxa [Melkonian, 1982].<br />
Such gametes initially show positive phototaxis<br />
prior to mating, swimming upward in the water<br />
column towards the light at the sea surface.<br />
Positively phototactic gametes may gain<br />
858
significant advantages, especially in shallow<br />
water, by being able to search for potential mates<br />
in a two-dimensional surface rather than in threedimensional<br />
space [Cox <strong>and</strong> Sethian, 1985].<br />
There are some experiments that support this idea<br />
[Togashi et al., 1999]. So, sperm limited<br />
conditions might not be ubiquitous in these<br />
species.<br />
In this paper, considering the effect of diffusion<br />
of gametes (inherent in 3-D r<strong>and</strong>om walks that are<br />
non-recurent) through time, we sought to<br />
elucidate potential advantages of gamete<br />
phototaxis <strong>and</strong> to study the mechanism of the<br />
evolution of slight anisogamy <strong>and</strong> the<br />
environmental conditions of their habitats.<br />
In the study of fertilization kinetics of gametes,<br />
numerical simulations using computer<br />
programming languages are an alternative to<br />
laboratory or field experiments, <strong>and</strong> can gain<br />
realism if specific sizes, swimming velocities, <strong>and</strong><br />
trajectories of real gametes are used as input<br />
parameters. [Although mathematical models are<br />
often the best encapsulation of ecological <strong>and</strong><br />
evolutionary mechanisms [Wilson, 2000], they<br />
may be unsuitable to analyze isogamous or near<br />
isogamous species, because it is difficult to<br />
remove fused gametes from the mating<br />
populations through time with mathematical<br />
methods.] Such gamete behaviors can be<br />
determined from video recordings of individual<br />
gamete swimming paths. Cox <strong>and</strong> Sethian [1985]<br />
used such inputs from Pommerville's films of the<br />
swimming behaviour of gametes of the fungal<br />
genus Allomyces to simulate gamete motion, but<br />
were limited to two-dimensional analysis given<br />
the 2-D plane of the film. Analytical solutions of<br />
three-dimensional r<strong>and</strong>om gamete motion are<br />
difficult to obtain, because unlike twodimensional<br />
r<strong>and</strong>om walks, three-dimensional<br />
motions are non-recurrent. Subsequent<br />
supercomputer simulations of three-dimensional<br />
search [Cox et al., 1991] resulted in the prediction<br />
that elliptically deformed, rather than spherical<br />
objects of equivalent biomass would result in<br />
greater encounter rates in 3-D r<strong>and</strong>om searches,<br />
but no attempts were made to compare isogamy to<br />
anisogamy. In this paper, our simulation code is<br />
compiled from C++.<br />
Figure 2. Experimental conditions <strong>and</strong> regimes.<br />
Figure 1. Gamete motion <strong>and</strong> sexual fusion.<br />
2. NUMERICAL MODELLING OF<br />
GAMETE BEHAVIOR<br />
2.1. Introduction<br />
2.2. Model description <strong>and</strong> input parameters<br />
Gametes are idealized as spheres. Body width of<br />
gamete is used as the diameter. Our idealization<br />
of gametes as spheres might be slightly unrealistic<br />
(since gametes of marine algal species are pearshaped)<br />
but would be of little mathematical<br />
consequence since drag forces are determined by<br />
the cross-sectional radius orthogonal to the<br />
direction of travel at low Reynold’s numbers<br />
according to Stoke’s law [Le Méhauté, 1976].<br />
Comparative data concerning gamete traits<br />
collected by a literature survey have demonstrated<br />
that, in isogamous or slightly anisogamous<br />
species, the range of gamete size is relatively<br />
narrow, <strong>and</strong> that it is intermediate between male<br />
<strong>and</strong> female gamete size in species with marked<br />
anisogamy [Togashi et al., 2002]. Therefore,<br />
using the data of a slightly anisogamous species<br />
Monostroma angicava Kjellman [Togashi et al.,<br />
1997] as representatives, the radii of male <strong>and</strong><br />
female gametes in a slightly anisogamous species<br />
are set at 1.48 µm <strong>and</strong> 1.85 µm, respectively. The<br />
radius of the isogametes is assumed to be the<br />
average of the slight anisogametes:<br />
(1.48+1.85)/2=1.67 µm.<br />
Each gamete swims at a given speed in water,<br />
starting from a r<strong>and</strong>omly distributed position on<br />
the bottom of a virtual rectangular test tank of 10<br />
mm (length), 10 mm (width) <strong>and</strong> 25 mm (depth).<br />
Thus, the distance traveled by each gamete during<br />
each time interval is the same. Experimental<br />
859
studies on gamete size <strong>and</strong> swimming speed exist<br />
in some species of marine green algae [Togashi et<br />
al., 1997; Togashi, 1998; Togashi et al., 1998]. At<br />
low Reybold’s numbers relevant here, movement<br />
is governed by viscous forces [F = 6 π ε c r;<br />
ε: viscosity of the liquid, c: swimming speed of<br />
gamete, r: radius of gamete] (see R<strong>and</strong>erson <strong>and</strong><br />
Hurst, 2001). Experimental data suggest that these<br />
forces provided by flagellar propulsion are<br />
equivalent for male <strong>and</strong> female gametes across<br />
species, supporting our assumption that gamete<br />
size is inversely related to swimming speed.<br />
At the beginning of each time interval (step),<br />
every gamete changes swimming direction threedimensionally<br />
since small motile objects at low<br />
Reynold’s numbers maintain straight paths for<br />
only limited time due to the impact of Brownian<br />
forces [Dusenbery, 1992]. Based on the analysis<br />
of gamete swimming paths [Togashi <strong>and</strong> Cox, in<br />
press], the step interval is set at 0.3 second, then,<br />
two angles are independently chosen separately<br />
for each gamete from a r<strong>and</strong>om sequence of<br />
integers between -30 <strong>and</strong> +30 to determine the<br />
changes of direction in the X-Y (horizontal) <strong>and</strong><br />
the Y-Z (vertical) planes, respectively. When a<br />
gamete collides with the tank or water surface,<br />
angles of incidence equal those of reflection. In<br />
gametes exhibiting positive phototaxis, a vector<br />
sum of the current unit velocity vector <strong>and</strong> the<br />
normal unit vector (light direction) is taken <strong>and</strong><br />
renormalized. Then the same r<strong>and</strong>om tilt <strong>and</strong><br />
rotation matrices that the non-phototactic method<br />
uses are applied thus ensuring steady upward<br />
motion of the gamete.<br />
The biomass allocated to produce gametes is<br />
assumed to be equal between mating types (i.e.<br />
1.0X10 5 µm 3 ) as experimentally confirmed in<br />
some organisms including marine green algae [e.g.<br />
Togashi et al., 1997). There are few reports on<br />
biased sex ratio of gametophytes in natural<br />
populations of marine green algae so far. So, sex<br />
ratios of gametophytes are assumed to be 1:1<br />
(=male:female).<br />
Each gamete is tracked <strong>and</strong> the distances between<br />
the centers of nearby male <strong>and</strong> female gametes<br />
are measured at each step to detect collisions. All<br />
encounters of sexually different gametes are<br />
deemed to result in sexual fusion. We divide the<br />
test tank into equally sized subrooms to increase<br />
the speed of calculation, but, our simulator can<br />
detect collisions even if a male <strong>and</strong> female gamete<br />
are in two different subrooms, but within “limit”<br />
distance of each other across the shared subroom<br />
face, in such a case, we should count it as a<br />
mating. Fused gametes are then removed from the<br />
mating population.<br />
3. NUMERICAL EXPERIMENTS<br />
3.1. Preliminary tests<br />
As an example of the speed <strong>and</strong> precision of our<br />
simulations, we made thirty runs with populations<br />
of 10,000 male <strong>and</strong> female gametes each for 2000<br />
time steps per run in less than 100 minutes.<br />
Because in our method we start by r<strong>and</strong>omly<br />
placing the population over the bottom of the test<br />
tank, <strong>and</strong> also because the swimming motion of<br />
the gametes has r<strong>and</strong>om elements at each time<br />
step, we were curious about the possible run to<br />
run variations that might occur in our results.<br />
Figure 3. Mating experiments in the normal<br />
species.<br />
3.2. Mating experiments<br />
Our experimental conditions (i.e. gamete size,<br />
number, swimming speed) <strong>and</strong> explored<br />
experimental regimes were shown in Figure 1.<br />
First, we performed mating experiments in the<br />
normal isogamous <strong>and</strong> slightly anisogamous<br />
species under both non-phototactic <strong>and</strong><br />
phototactic conditions (Figure 1a <strong>and</strong> b). The<br />
numbers of gametes at the surface of water were<br />
also monitored in the isogamous species.<br />
Secondly, in the slightly anisogamous species<br />
under the phototactic condition, we slightly<br />
increased only the size of male gametes to that of<br />
the isogametes (Figure 1c).<br />
Thirdly, in the slightly anisogamous species under<br />
the phototactic condition, we slightly decreased<br />
only the size of female gametes to that of the<br />
isogametes (Figure 1d).<br />
Lastly, we performed a mating experiment in a<br />
hypothetical isogamous species, in which the size<br />
of gametes of both mating types was the same as<br />
“male” gametes of the slightly anisogamous<br />
860
species (Figure 1e). Therefore, in this experiment,<br />
gametes of both mating types were smaller than<br />
those of the normal isogamous species.<br />
4. RESULTS<br />
In our preliminary tests, using a particular set of<br />
thirty replicas with identical conditions except for<br />
different r<strong>and</strong>om starting points, we found that the<br />
mean number of fertilizations that had occurred<br />
by the 2000 th step was 5844 with a st<strong>and</strong>ard error<br />
of 6.19. These results mean that we can predict<br />
the mean value of 5844 fertilizations within a<br />
95% confidence interval of 0.2%. Because we use<br />
large numbers of gametes in our simulations, we<br />
expect similar precision in other runs.<br />
In the normal isogamous <strong>and</strong> slightly<br />
anisogamous species (Figure 2), under the nonphototactic<br />
conditions, the numbers of formed<br />
zygotes were nearly identical <strong>and</strong> observed at a<br />
low level. So, many gametes remained<br />
unfertilized in both species. In contrast, under the<br />
phototactic conditions, they remarkably increased<br />
<strong>and</strong> the difference of the numbers of formed<br />
zygotes between the two species was clarified. As<br />
a result, species with isogamy was significantly<br />
more successful in producing zygotes than species<br />
with slight anisogamy. In the isogamous species,<br />
some unfertilized gametes remained in both<br />
mating types. However, in the slightly<br />
anisogamous species, most female gametes were<br />
soon fertilized. The numbers of gametes at the<br />
surface of water during the experiments in the<br />
isogamous species were shown in Figure 3. It<br />
appears that, only under the phototactic<br />
conditions, gametes of both sexes actually<br />
gathered just under the surface of water.<br />
Figure 4. The number of gametes at the surface.<br />
The numbers of formed zygotes in the<br />
hypothetical species were shown in Figure 4. In<br />
the hypothetically established slightly<br />
anisogamous species with “larger male gametes”,<br />
some female gametes remained unfertilized. Thus,<br />
the number of zygotes formed in this species was<br />
smaller than that in the normal slightly<br />
anisogamous species.<br />
However, in the hypothetically established<br />
slightly anisogamous species with “smaller<br />
female gametes”, most female gametes were soon<br />
fertilized. Thus, the number of formed zygotes<br />
was larger than that in the normal slightly<br />
anisogamous species.<br />
In the hypothetically established isogamous<br />
species with “smaller gametes” of both mating<br />
types than the normal isogamous species, the<br />
number of formed zygotes was the largest of all<br />
mating experiments in this study. However, some<br />
gametes of the both mating types remained<br />
unfertilized.<br />
Figure 5. Mating experiments in the hypothetical<br />
species.<br />
5. DISCUSSION<br />
The superiority of phototactic gametes (Figure 2<br />
<strong>and</strong> 4) may be widely expected in nature,<br />
especially for species that release gametes in<br />
shallow waters, because gametes continue to<br />
swim showing positive phototaxis for more than<br />
12 hours in slightly anisogamous <strong>and</strong> isogamous<br />
species [e.g. Togashi et al., 1997]. These species<br />
often inhabit upper or middle intertidal zones [e.g.<br />
Dawes, 1998]. Their zygotes become negatively<br />
phototactic just after they are formed <strong>and</strong> swim<br />
back down to the water substrate [e.g. Togashi et<br />
al., 1997]. This should facilitate settlement on the<br />
intertidal substratum in photosynthetically<br />
advantageous areas, preventing the zygotes from<br />
drifting out to deep waters as they might if<br />
phototaxis remained positive. Behaviors of such<br />
swarmers (i.e. gametes <strong>and</strong> swimming zygotes)<br />
may not be overwhelmed by sea conditions,<br />
861
ecause they often possess mechanisms for<br />
synchronous gamete release during extremely low<br />
daytime tides under calm conditions when<br />
swarmers could make the best use of their<br />
phototaxis avoiding turbulent water movement<br />
[e.g. Togashi <strong>and</strong> Cox, 2001].<br />
Phototaxis may actually function to introduce<br />
gametes to a two-dimensional realm (the water<br />
surface) where search efficiencies <strong>and</strong> target<br />
encounter probabilities are much higher than in<br />
three-dimensional r<strong>and</strong>om searches in the water<br />
column, which characterize those of nonphototactic<br />
gametes (Figure 3).<br />
Our numerical experiments in the normal<br />
isogamous <strong>and</strong> slightly anisogamous species<br />
suggest that, although the isogamous species may<br />
be under sperm limiting conditions where the<br />
number of formed zygotes increases as more<br />
sperm is released, such sperm limitation appears<br />
to be resolved in the slightly anisogamous species,<br />
where most female gametes are fertilized <strong>and</strong> the<br />
number of zygotes depends on the number of<br />
released female gametes (Figure 2). Thus, slightly<br />
anisogamous species may be often under spermabundant<br />
(competitive) conditions.<br />
We hypothetically increased only the size of male<br />
gametes in the slightly anisogamous species. It is<br />
because mating efficiency of such a species might<br />
be as high as the normal species, if slight<br />
anisogamy always resolves sperm limitation. The<br />
volume of zygotes formed in this species is larger<br />
than that in the normal species (by 15 %). Thus, if<br />
this hypothesis is true, assuming a positive<br />
relationship between zygote volume <strong>and</strong> fitness,<br />
such a species might be more advantageous than<br />
the normal species. However, our results have<br />
demonstrated that the number of zygotes formed<br />
in this hypothetical species is smaller than that in<br />
the normal species (Figure 4). This suggests that<br />
the size of male gametes has a large impact on<br />
mating efficiency. Some advantages of smaller<br />
gametes (e.g. higher speed, larger number) appear<br />
to work. This may be one reason why male<br />
gametes do not increase their size in anisogamous<br />
species in nature. Such a size should be<br />
evolutionary stable for males, because it has been<br />
suggested that it is nearly a minimal size to<br />
maintain a phototactic system [Togashi et al.,<br />
2002].<br />
Potential advantages of larger female gametes<br />
(e.g. larger target size) may have weaker effects<br />
on mating than those of smaller male gametes,<br />
because, even if the size of female gametes is<br />
slightly decreased, most female gametes are still<br />
easily fertilized (Figure 4). In this case, the<br />
number of formed zygotes is larger than that in<br />
the normal species. However, the volume of<br />
zygotes formed in this species is smaller than that<br />
in the normal species (by 17 %).<br />
Advantages of small gametes do not always give<br />
satisfactory results because some gametes<br />
remained unfertilized in the hypothetical<br />
isogamous species with the smaller gametes<br />
(Figure 4). Anisogamy may be necessary to<br />
fertilize most (female) gametes. However, this<br />
hypothetical species with smaller gametes<br />
produced the largest number of zygotes in this<br />
study.<br />
Comparing mating efficiency between the two<br />
normal species, it is greater for isogamous than<br />
for slightly anisogamous species (Figure 2). Thus,<br />
we should also note that the evolution of<br />
anisogamy from primitive isogamy may not be<br />
explained solely by high encounter rates of<br />
anisogamous male <strong>and</strong> female gametes <strong>and</strong><br />
resultant high mating efficiency [e.g. Levitan,<br />
1996]. Two conflicting selection forces of search<br />
efficiency <strong>and</strong> zygote fitness may be needed to<br />
explain the evolution of anisogamy in marine<br />
green algae [e.g. Parker et al., 1972].<br />
Such a stronger anisogamy as male gametes are<br />
too small to maintain a phototactic system may<br />
not be predictable through this study without<br />
some other mechanisms to compensate the loss.<br />
Some species of marine green algae (e.g. the<br />
genus Bryopsis) have got over the fence by a<br />
pheromonal attraction from female gametes<br />
which retain a phototactic system [Togashi et al.,<br />
1998]. It has been considered that female<br />
characteristics to increase probability of<br />
fertilization would not have evolved without<br />
sperm limitation with the exception of egg size<br />
[Levitan, 1996]. However, this pheromonal<br />
attraction system could have developed to connect<br />
discrete gamete behaviors between sexes, <strong>and</strong><br />
realized such a marked anisogamy even under<br />
sperm-abundant (competitive) conditions if there<br />
is strong selection for large zygote size.<br />
Figure 6. Mating systems <strong>and</strong> habitats in marine<br />
green algae.<br />
862
Advantages of positive phototaxis may be lost in<br />
deep water <strong>and</strong> larger zygotes should be needed to<br />
develop safely in such a photosynthetically<br />
disadvantageous area. In fact, some species of the<br />
genus Derbesia are usually observed in deep<br />
water [Chapman et al., 1964] <strong>and</strong> produce<br />
strongly anisogamous non-phototactic male <strong>and</strong><br />
female gametes <strong>and</strong> resultant large zygotes. Our<br />
simulations <strong>and</strong> other observations of real mating<br />
systems [Togashi et al., 2002] in marine green<br />
algae suggest that smaller zygotes might be<br />
occasionally disadvantageous, even if they had<br />
larger numbers of zygotes. In marine green algae,<br />
the mating systems appear to be tightly tuned to<br />
the environmental conditions of their habitats (see<br />
Figure 5).<br />
6. REFERENCES<br />
Chapman, V.J., A.S. Edomonds <strong>and</strong> F.I.<br />
Dromgoole, Halicystis in New Zeal<strong>and</strong>,<br />
Nature, 202, 414, 1964.<br />
Cox, P.A., S. Cromar <strong>and</strong> T. Jarvis, Underwater<br />
pollination, three-dimensional search, <strong>and</strong><br />
pollenmorphology: predictions from a<br />
supercomputer analysis, In: Blackmore, S.<br />
<strong>and</strong> S.H. Barnes, eds. Pollen <strong>and</strong> Spores,<br />
Systematics Association Special <strong>Volume</strong><br />
no 44, Clarendon Press, 363-375, Oxford,<br />
1991.<br />
Cox, P.A. <strong>and</strong> J.A. Sethian, Gamete motion,<br />
search, <strong>and</strong> the evolution of anisogamy,<br />
oogamy, <strong>and</strong> chemotaxis, American<br />
Naturalist, 125, 74-101, 1985.<br />
Dawes, C.J., Marine botany. 2nd ed., John Wiley,<br />
New York, 1998.<br />
Dusenbury, D.B., Sensory ecology, Freeman,<br />
New York, 1992.<br />
Le Méhauté, B., An introduction to<br />
hydrodynamics <strong>and</strong> water waves, Springer,<br />
Berlin , 1976.<br />
Levitan, D.R., Effects of gamete traits on<br />
fertilization in the sea <strong>and</strong> the evolution of<br />
sexual dimorphism, Nature, 382, 153-155,<br />
1996.<br />
Melkonian, M., Structural <strong>and</strong> evolutionary<br />
aspects of the flagellar apparatus in green<br />
algae <strong>and</strong> l<strong>and</strong> plants, Taxon, 31, 255-265,<br />
1982.<br />
Parker, G.A., R.R. Baker <strong>and</strong> V.G.F. Smith, The<br />
origin <strong>and</strong> evolution of gamete<br />
dimorphism <strong>and</strong> the male-female<br />
phenomenon, Journal of Theoretical<br />
Biology, 36, 529-553, 1972.<br />
R<strong>and</strong>erson, J.P. <strong>and</strong> L.D. Hurst, The uncertain<br />
evolution of the sexes, Trends in Ecology<br />
<strong>and</strong> Evolution, 16, 571-579, 2001.<br />
Shuster, S.M. <strong>and</strong> M.J. Wade, Mating Systems<br />
<strong>and</strong> Strategies. Princeton University Press,<br />
Princeton, 2003.<br />
Togashi, T., Reproductive strategies, mating<br />
behaviors <strong>and</strong> the evolution of anisogamy<br />
in marine green algae, Ph.D. thesis,<br />
Hokkaido University, Sapporo, 1998.<br />
Togashi, T. <strong>and</strong> P.A. Cox, Tidal-linked synchrony<br />
of gamete release in the marine green alga,<br />
Monostroma angicava Kjellman, Journal<br />
of Experimental Marine Biology <strong>and</strong><br />
Ecology, 264, 117-131, 2001.<br />
Togashi, T. <strong>and</strong> P.A. Cox, Phototaxis <strong>and</strong> the<br />
evolution of isogamy <strong>and</strong> “slight<br />
anisogamy” in marine green algae: insights<br />
from laboratory observations <strong>and</strong><br />
numerical experiments, Botanical Journal<br />
of the Linnean Society, in press.<br />
Togashi, T., T. Miyazaki <strong>and</strong> P.A. Cox, Sexual<br />
reproduction in marine green algae:<br />
gametic behavior <strong>and</strong> the evolution of<br />
anisogamy, Proceedings of Two Symposia<br />
on Ecology ad Evolution in VIII<br />
INTECOL, Seoul, Korea, Aug. 2002,<br />
Sangaku Publisher, 70-79, Ohtsu, 2002.<br />
Togashi, T., T. Motomura, T. Ichimura,<br />
Production of anisogametes <strong>and</strong> gamete<br />
motility dimorphism in Monostroma<br />
angicava, Sexual Plant Reproduction, 10,<br />
261-268, 1997.<br />
Togashi, T., T. Motomura <strong>and</strong> T. Ichimura,<br />
Gamete dimorphism in Bryopsis plumosa:<br />
phototaxis, gamete motility <strong>and</strong><br />
pheromonal attraction, Botanica Marina,<br />
41, 257-264, 1998.<br />
Togashi, T., T. Motomura, T. Ichimura <strong>and</strong> P.A.<br />
Cox, Gametic behavior in a marine green<br />
alga, Monostroma angicava: an effect of<br />
phototaxis on mating efficiency, Sexual<br />
Plant Reproduction, 12, 158-163, 1999.<br />
Wilson, W. Simulating Ecological <strong>and</strong><br />
Evolutionary Syatems in C. Cambridge<br />
University Press, Cambridge, 2000.<br />
863
Predicting predation efficiency of biocontrol agents:<br />
linking behavior of individuals <strong>and</strong> population dynamics<br />
Brigitte Tenhumberg<br />
School of Natural Resources, University of Nebraska-Lincoln, Nebraska, USA, btenhumberg2@unl.edu<br />
Abstract: Behavioral ecology <strong>and</strong> population ecology are two separate branches of ecology; studies linking<br />
the effect of individual behavior <strong>and</strong> population dynamics are rare. This paper connects a stochastic optimal<br />
foraging model of insect predators with an age structured population model of its prey. I modeled syrphid<br />
larvae feeding on cereal aphids, an interaction critical to cereal crops in Germany. The key stochastic element<br />
in this model is the foraging success of predators, which influences survival <strong>and</strong> developmental time of<br />
predators <strong>and</strong> mortality of the prey population. The model predicts that the level of control incurred by<br />
predators is highest if predators arrive when prey numbers are still small, the growth rate of prey population is<br />
small, <strong>and</strong> predator density is moderately high. If the number of predators per prey was high or prey<br />
distribution was much aggregated, predators were less successful in finding prey. As a result predation<br />
efficacy was reduced.<br />
Keywords: Behavior; Population Dynamics; Biocontrol; Escalator Boxcar Train<br />
1. INTRODUCTION<br />
Mortality caused by insect predators <strong>and</strong> parasitic<br />
wasps is a major biotic factor shaping the<br />
population dynamics of any insect prey (host)<br />
species [Symondson et al., 2002] <strong>and</strong> can be<br />
exploited for biocontrol. The impact of predators<br />
(parasitic wasps) on their prey (host) population<br />
likely depends on their foraging behaviour. There<br />
is a large body of literature documenting different<br />
factors influencing foraging behavior (“optimal<br />
foraging theory”), but individual level responses<br />
do not necessarily affect population level<br />
processes. For example, Tenhumberg et al [2001]<br />
demo nstrated that the behavioral response of<br />
individual female parasitic wasps, Cotesia<br />
rubecula, can compensate for the effect of small<br />
scale variation in host distribution. This results in<br />
equal reproductive success over a range of small<br />
scale distributio n pattern s. In this paper I explicitly<br />
link individual behavior with population processes<br />
by simulating the impact of “optimally” behaving<br />
insect predators on their prey population, <strong>and</strong><br />
examine the conditions under which predators can<br />
prevent pest outbreaks.<br />
I used the economically important aphid species,<br />
Sitobion avenae (prey) <strong>and</strong> its syrphid predator,<br />
Episyrphus balteatus as a model system. In<br />
general, the composition of aphid species in<br />
German winter wheat fields includes S. avenae,<br />
Metopolophium dirhodum, <strong>and</strong> Rhopalosiphum<br />
padi [Tenhumberg, 1992]. Only the first two<br />
species occur in high numbers, but they generally<br />
feed on separate plant parts: M. dirhodum feeds on<br />
leafs, while S. avenae feeds mainly on the ear <strong>and</strong><br />
has the highest impact on the yield. In western<br />
Germany syrphids are by far the most important<br />
predators of cereal aphids (~80% of all<br />
stenophagous predators) <strong>and</strong> E. balteatus<br />
constitutes >90 % of the composition of syrphid<br />
species [Groeger, 1992; Tenhumberg, 1992]. Other<br />
insects contributing to the control of cereal aphid<br />
populations include lady beetles, parasitic wasps,<br />
<strong>and</strong> spiders.<br />
2. MODEL DESCRIPTION<br />
2.1 Aphid Model (S. avenae)<br />
To simulate the population dynamics of aphids I<br />
used the “escalator boxcar train” (EBT) technique<br />
[Leffelaar, 1999], which can be used to model<br />
continuous time populations with mixed age<br />
distributions. Before a simulation starts, the<br />
developmental axis of one stage is broken up into a<br />
number of classes or boxcars, each with identical<br />
developmental width. Here, we constructed two<br />
chained EBTs, one for larval aphids <strong>and</strong> one for<br />
adult aphids. Note that aphid eggs do not occur<br />
864
during the growing season of winter wheat. Each<br />
EBT consisted of 10 boxcars representing different<br />
age classes. All individuals of the aphid population<br />
were distributed among the boxcars. Individuals of<br />
a particular boxcar had unique vital rates, so the<br />
model could account for stage <strong>and</strong> age specific<br />
mortality <strong>and</strong> reproduction rates. The<br />
developmental process was simulated by shifting<br />
individuals continuously to a higher stage of<br />
development at the same rate. Newborn aphids<br />
entered the first boxcar of the larvae-EBT; unless<br />
dying they successively moved through all boxcars<br />
of the larvae-EBT <strong>and</strong> the adult-EBT <strong>and</strong> were<br />
removed from the population after reaching the end<br />
of the last boxcar, which is their maximum life span.<br />
The EBT technique is described in detail in<br />
Leffelaar [1999].<br />
Model parameters were estimated based on<br />
laboratory studies on S. avenae at 20 o C [Dean,<br />
1974; Simon et al., 1991] <strong>and</strong> listed in Table 1.<br />
According to Dean [1974] 97% of aphid larvae<br />
survive to adult phase <strong>and</strong> the average adult<br />
lifespan is 22 days. We assume that juvenile<br />
survival rate is constant <strong>and</strong> adult survival follows<br />
a Weibull function. In general, with increasing<br />
temperatures larval development increases <strong>and</strong><br />
survival of adult aphids decreases; reproduction<br />
<strong>and</strong> the intrinsic growth rate increase up to 20 o C<br />
<strong>and</strong> decrease at higher temperatures [Dean, 1974].<br />
The model does not include the effect of<br />
temperature directly; however the sensitivity<br />
analysis revealed the effect of changes in the<br />
developmental time <strong>and</strong> reproduction.<br />
The simulation model predicts exponential growth<br />
of aphids. Real aphid populations are regulated by<br />
density dependent mechanisms, such as an<br />
increasing proportion of migrating aphids<br />
(alatifome = aphids with wings) [Watt <strong>and</strong> Dixon,<br />
1981], presumably limiting aphid numbers to < 1000<br />
aphids per shoot. As this paper is concerned with<br />
predator-prey interactions at much lower aphid<br />
densities we ignore density dependent<br />
mechanisms.<br />
Table 1: Parameters used in aphid model. Daily rates were normalized through division by λ. (a = 1.05, b=011.5,<br />
c=0.040976, x is time in days, κ = 3.5, <strong>and</strong> ρ = 0.034)<br />
Larvae-EBT (L) Adult-EBT (A) References<br />
Stage length, D D L = 8 days D A = 45 days [Dean, 1974]<br />
Number of boxcars, n 10 10<br />
Developmental width, γ γ L = D L /n = 0.8 γ A = D A /n = 4.5<br />
Mortality per day, µ 0.003<br />
L<br />
1<br />
x κ −<br />
µ = µ A = κρ( ρ )<br />
Age dependent reproduction , φ 0 ln ( )<br />
φ x a bx e<br />
modified from Dean [1974]<br />
−cx<br />
= modified from Simon et al.<br />
[1991]<br />
2.2 Syrphid model<br />
The syrphid model has been published elsewhere<br />
[Tenhumberg et al., 2000], so I present only an<br />
overview here. The model uses stochastic dynamic<br />
programming to calculate the optimal statedependent<br />
behavior that maximizes lifetime<br />
reproduction. At any point in time syrphid larvae<br />
have three behavioural options: foraging for<br />
aphids, resting, or pupating. Syrphid larvae may<br />
find food while foraging; the probability of<br />
catching aphids is a function of aphid density <strong>and</strong><br />
distribution. Syrphid larvae need food for<br />
maintenance <strong>and</strong> growth; but foraging uses up<br />
energy <strong>and</strong> increase the risk of being preyed upon.<br />
Syrphid reproduction is a function of size,<br />
consequently the higher the accumulated weight of<br />
a syrphid when pupating, the higher is her<br />
expected future reproductive success. Conversely ,<br />
the longer a syrphid postpones pupating to<br />
accumulate a higher weight, the more likely she is<br />
to die as a result of starvation or predation. What<br />
behavior is best at any point in time depends on<br />
the states: gut content, weight, age, <strong>and</strong> food<br />
availability (mean <strong>and</strong> variance). Syrphid larvae<br />
estimate their chances to find food based on the<br />
distribution of past prey encounters [weighted<br />
maximum likelihood estimate, Mangel, 1990].<br />
Foragers catch A prey units, where A is a negative<br />
binomial r<strong>and</strong>om variable with some mean m <strong>and</strong> an<br />
aggregation index k:<br />
( )<br />
! ( )<br />
a<br />
⎡Γ k+ a ⎤⎛ m ⎞ ⎛ k ⎞<br />
p = P<br />
A { A= a}<br />
= ⎢ ⎥⎜ ⎟ ⎜ ⎟<br />
⎣ a Γ k ⎦⎝m+ k⎠ ⎝m+<br />
k ⎠<br />
k<br />
865
where Γ(k ) is a gamma function [Krebs, 1989], <strong>and</strong><br />
m is syrphids expectation of average food<br />
availability. Based on field observations on cereal<br />
aphids [Ohnesorge <strong>and</strong> Viereck, 1983], I set k=2,<br />
indicating a slightly aggregated distribution.<br />
2.3 Linking predator <strong>and</strong> prey model<br />
The aphid <strong>and</strong> syrphid models were connected<br />
through syrphid feeding activity, imposing<br />
additional mortality on the aphid population (see<br />
Figure 1). In turn, aphid density influenced syrphid<br />
foraging success, <strong>and</strong> consequently syrphid<br />
performance (rate of weight increase, starvation).<br />
To facilitate comparison with empirical data I will<br />
present syrphid density per m 2 <strong>and</strong> aphid density<br />
per shoot (assuming there are 550 shoots per m 2 ).<br />
Shift<br />
µ L Larvae-EBT<br />
µ P<br />
φ<br />
Predators<br />
Adult-EBT<br />
OF<br />
E<br />
µ Α<br />
µ Α<br />
Figure 1. Flow Chart. µ indicates mortality, φ age<br />
dependent reproduction, “shift” individuals<br />
shifting from larvae-EBT to adult-EBT, E small<br />
syrphid larvae enter the model, P syrphid larvae<br />
pupate, <strong>and</strong> OF optimal consumption rate of<br />
predators.<br />
Egglaying behaviour of syrphid females is<br />
influenced by aphid abundance such that females<br />
only oviposit if aphid populations are above some<br />
threshold density, which varies between years<br />
[Tenhumberg <strong>and</strong> Poehling, 1991], <strong>and</strong> can be as<br />
low as 0.2 aphids per shoot [Chambers, 1991].<br />
Syrphid larvae hatch after three days [Tenhumberg,<br />
1992]. Analogous to the egg distribution, I modeled<br />
the distribution of new syrphid larvae entering the<br />
model (freshly hatched) as a normal distribution,<br />
with the first larvae entering the model after aphid<br />
density reached some threshold density.<br />
Each time step the interactions between aphid <strong>and</strong><br />
syrphid populations were modeled sequentially.<br />
- The change in aphid population for one time step<br />
(=10 hours) was calculated based on the EBT<br />
model.<br />
- At the beginning of each time step, the model<br />
determined optimal decisions of predators, which<br />
follow from the tradeoff between the likelihood of<br />
accumulating more weight <strong>and</strong> of dying.<br />
P<br />
- The per capita aphid consumption was simulated<br />
based on the probability distribution determined<br />
by aphid density <strong>and</strong> distribution.<br />
- Then the model calculated the changes in<br />
individual states: age increases; gut content<br />
increased according to the number of prey<br />
consumed; some of the gut content was used for<br />
maintenance <strong>and</strong> weight increase. If gut content<br />
decreased below a threshold predators died of<br />
starvation.<br />
- The model removed pupating <strong>and</strong> dying syrphid<br />
larvae from the population <strong>and</strong> new arriving<br />
larvae entered the population.<br />
- The total number of predated aphids were<br />
removed according to their relative frequency in<br />
the boxcars of larvae-EBT <strong>and</strong> adult-EBT. This<br />
assumes that prey encounter is r<strong>and</strong>om <strong>and</strong><br />
syrphids do not have any preferences for prey<br />
size .<br />
2.4 Sensitivity Analysis<br />
For the sensitivity analysis I employed Latin<br />
Hypercube Sampling [LHS Blower <strong>and</strong><br />
Dowlatabadi, 1994], which is a type of stratified<br />
Monte Carlo sampling. This technique has been<br />
used in the analysis of complex ecological models<br />
elsewhere [Rushton et al., 2000a; Rushton et al.,<br />
2000b; Tenhumberg et al., in press]. LHS is an<br />
extremely efficient sampling design because each<br />
value of a parameter is only used once in the<br />
analysis. The estimation of uncertainty for each<br />
parameter is modeled by treating each parameter as<br />
a r<strong>and</strong>om variable. Probability distribution<br />
functions (pdfs) are defined for each parameter. I<br />
used uniform distributions, but other distributions<br />
are possible. I broke each of these distributions<br />
into N intervals, each of equal probability. I then<br />
chose the midpoint of each interval <strong>and</strong> generated<br />
an LHS table as an N * K matrix, where N is the<br />
number of simulations <strong>and</strong> K is the number of<br />
sampled input parameters. I chose N=100 <strong>and</strong><br />
K=10. 12 parameter combinations were excluded<br />
from the analysis because they either resulted in an<br />
exponential decline of the aphid population without<br />
syrphids present or aphid populations increased<br />
too rapidly for syrphid larvae to have any effect. I<br />
repeated each run 20 times because the syrphid<br />
model is stochastic; therefore the whole sensitivity<br />
analysis is based on 1760 simulations (88*20). All<br />
simulations are stopped after 33 days or 80 time<br />
steps.<br />
I used partial rank correlation coefficients (PRCC)<br />
to evaluate statistical relationships between each<br />
input parameter <strong>and</strong> each output parameters while<br />
keeping all other input parameters constant at their<br />
866
expected value [Conover, 1980]. This partial rank<br />
correlation is based on ranks of the results <strong>and</strong> of<br />
the parameter values within their columns, rather<br />
than on the raw values. This analysis determines<br />
the independent effect of each parameter, even if<br />
the parameters are correlated. The sign of the<br />
PRCC indicates the qualitative relationship<br />
between input <strong>and</strong> output variable, <strong>and</strong> the relative<br />
importance of the input variables can be directly<br />
evaluated by comparing the PRCC values. The<br />
calculation of PRCC is described in Blower [1994].<br />
3. RESULTS AND DISCUSSION<br />
Figure 2 illustrates a typical simulation run using<br />
the parameters listed in Table 1. Overall 70 syrphid<br />
larvae hatched, but as a result of pupation <strong>and</strong><br />
larval mortality the maximum syrphid density was<br />
only 39 individuals per m 2 . When the last syrphid<br />
larvae disappeared (32 days) aphid density reached<br />
30 individuals per shoot. For comparison, in the<br />
absence of predators aphid population was 475<br />
individuals per shoot. In the real world the ears of<br />
winter wheat plants usually start drying up around<br />
20-30 days after syrphid larvae appear<br />
[Tenhumberg, 1992] <strong>and</strong> the resulting rapid<br />
decrease in plant quality causes the break down of<br />
aphid populations through elevated aphid<br />
mortality <strong>and</strong> development of a large proportion of<br />
migrating aphids [Watt <strong>and</strong> Dixon, 1981]. Thus,<br />
aphid populations are unlike ly to increase<br />
considerably after all syrphids have pupated.<br />
Number of Individuals<br />
40<br />
30<br />
20<br />
10<br />
0<br />
0 10 20 30<br />
Time [Days]<br />
Figure 2. Simulated population dynamics of aphids<br />
(solid line) <strong>and</strong> syrphid larvae (dotted line), using<br />
parameter values from Table 1. 1 st syrphid larva<br />
appeared when aphid density > 0.05.<br />
If syrphid predators have such high potential to<br />
control aphid populations, why do aphid<br />
populations regularly outbreak in Northern<br />
Germany? The sensitivity analysis (see Table 2)<br />
sheds some light on this question. First, I will<br />
discuss the results of aphid parameters, then the<br />
parameters specifying the interactions of predator<br />
<strong>and</strong> prey populations.<br />
A. Aphid specific parameters:<br />
Most prominent factors influencing maximum aphid<br />
density (A max ) are the parameters of the age<br />
dependent reproduction curve (φ, Table 1) <strong>and</strong><br />
larval developmental time which determines how<br />
quickly aphids start reproducing (Table 2). In<br />
general, the larger the values of a <strong>and</strong> b the higher<br />
is the maximum reproductive output (φ max ). c is<br />
inversely correlated to aphid reproductive output:<br />
the smaller c the larger φ max <strong>and</strong> the slower the<br />
decrease in the age dependent reproduction.<br />
Within the parameter range tested the effect of<br />
larval <strong>and</strong> adult survival is small (small PRCC’s <strong>and</strong><br />
only κ is significant).<br />
Reproduction <strong>and</strong> developmental time are<br />
influenced by the temperature in the field. If the<br />
weather is warm, aphid development is short <strong>and</strong><br />
the peak reproduction is reached earlier [Dean,<br />
1974]. According to the results of the sensitivity<br />
analysis these conditions greatly promote high<br />
aphid densities. Conversely, aphid populations<br />
usually reach much higher densities in northern<br />
Germany (cooler climate) compared to southern<br />
Germany (warmer climate) [Tenhumberg <strong>and</strong><br />
Poehling, 1995].<br />
B. Predator specific parameters:<br />
The input parameters influencing predator-prey<br />
interactions are aphid density when 1 st syrphid<br />
larvae appear (synchronization), the total number<br />
of predators <strong>and</strong> aphid distribution, which<br />
influences predator foraging success. The impact<br />
of syrphid predators on aphid population is not<br />
only influenced by input parameters, but also by<br />
mortality (i.e. starving) <strong>and</strong> behavioral response<br />
<strong>and</strong> of syrphid larvae (functional response, timing<br />
of pupation). As an indication of syrphid<br />
responses I included in the sensitivity analysis the<br />
maximum number of syrphids (S max ), the time period<br />
over which syrphid larvae were present (syrphid<br />
days, S d ), <strong>and</strong> the average per capita consumption<br />
(C). In the following, I will refer to the PRCC’s in<br />
colum S max as PRCC -S max , <strong>and</strong> so on.<br />
Synchronization: By far most important in keeping<br />
aphid numbers low is the synchronization between<br />
aphids <strong>and</strong> syrphid predators (PRCC-A max =0.89). A<br />
high aphid density when the 1 st predators arrive<br />
results in high food availability <strong>and</strong> syrphid<br />
predators increase their consumption rate (large<br />
positive PRCC-C). This functional response is<br />
consistent with empirical findings [Tenhumberg,<br />
1995]. As a response to high food availability<br />
syrphid larvae accumulate weight quicker <strong>and</strong><br />
867
pupate at an earlier age [Tenhumberg et al., 2000].<br />
As a consequence, S d <strong>and</strong> S max are shorter<br />
(negative PRCC-S d <strong>and</strong> S max ), which means the<br />
growth rate of aphid populations is slowed down<br />
for a shorter period of time <strong>and</strong> the maximum<br />
number of predators is smaller. So, the reduced<br />
larval period of syrphids counteracts somewhat the<br />
increased feeding rate of syrphid predators.<br />
Predator abundance: Interestingly the effect of the<br />
cumulative number of syrphid larvae appearing on<br />
maximum aphid density is much smaller than the<br />
effect of synchronization. The reason for this is<br />
interspecific competition resulting in decreasing<br />
per capita consumption with increasing predator<br />
density (negative PRCC-C), <strong>and</strong> syrphid larvae<br />
need a longer time to accumulate a sufficiently<br />
large weig ht to pupate (positive PRCC-S d ).<br />
Aphid distribution: The degree of aggregation of<br />
aphid distributions also influences maximum aphid<br />
densities (negative PRCC-A max ) through syrphid<br />
mortality <strong>and</strong> foraging efficiency. A high degree of<br />
aggregation (small k-value) translates to large<br />
variation in foraging success between capturing<br />
bouts, which in turn increases the probability of<br />
starvation because of the high frequency of<br />
successively finding no or not enough food. The<br />
increased mortality rate results in overall reduced<br />
syrphid densities (positive PRCC-S max ). If aphid<br />
distributions are highly aggregated the per capita<br />
consumption of predators decreases (positive<br />
PRCC-C). As a result of the slow rate of weight<br />
accumulation syrphid larvae need a longer time to<br />
pupate, which increases the length of the period<br />
where syrphid predators are present (negative<br />
PRCC-S d ).<br />
Table 2: Partial rank correlation coefficients (PRCC) of maximum aphid density, A max , syrphid maximum<br />
density, S max , number of days syrphid larvae are present, S d , <strong>and</strong> the average per capita consumption per day<br />
of present larvae, C. Absolute values >0.235 (>0.19) are significant at p=0.01 (p=0.05) <strong>and</strong> are indicated by **<br />
(*). Range specifies the rage over which input parameters were v aried in the sensitivity analysis. The analysis<br />
is based on 88 different parameter combinations.<br />
Input variables Parameter Range A max S max S d C<br />
Reproduction a 4-7 0.651 ** 0.210 * - 0.279 ** 0.206 *<br />
b 1-3 0.857 ** 0.120 - 0.441 ** 0.472 **<br />
c 0.3-0.7 - 0.704 ** - 0.190 * 0.251 ** - 0.274 **<br />
Adult mortality κ 2-5 0.292 ** 0.034 - 0.200 * 0.182<br />
ρ 0.025-0.05 0.058 - 0.161 - 0.113 0.044<br />
Larval mortality 0.02-0.1 - 0. 137 - 0.118 0.162 - 0.062<br />
Larvae DT 6-9 - 0.767 ** - 0.136 0.489 ** - 0.509 **<br />
Aphid distribution k 0.01-2 - 0.232 * 0.583 ** - 0.564 ** 0.640 **<br />
Threshold Density 0.01-1 0.891 ** - 0.273 ** - 0.596 ** 0.812 **<br />
Total predator number 50-100 - 0.380 ** 0.960 ** 0.265 ** -0.195 *<br />
4. CONCLUSIONS<br />
This model suggests that syrphid larvae are most<br />
likely to suppress aphid outbreaks if syrphid<br />
larvae arrive when aphid density is still is small.<br />
Differences in the synchronization between<br />
syrphid <strong>and</strong> aphids populations are hypothesized<br />
to be the main reason why in northern Germany<br />
aphid populations regularly reach outbreak<br />
densities in winter wheat fields (if no insecticides<br />
are applied) <strong>and</strong> in southern Germany not<br />
[Tenhumberg <strong>and</strong> Poehling, 1995].<br />
The potential of syrphid larvae to prevent<br />
outbreak densities of aphid populations is also<br />
influenced by intraspecific competition <strong>and</strong><br />
syrphid responses to aphid population, such as<br />
timing of pupation, starvation <strong>and</strong> foraging<br />
success. The latter is not only dependent on<br />
aphid density but also aphid distribution. Ignoring<br />
these responses in models forecasting the risk of<br />
pest outbreaks [e.g., Gosselke et al., 2001] might<br />
result in overestimating predation efficiency <strong>and</strong><br />
consequently erroneous risk assessment.<br />
5. ACKNOWLEDGEMENTS<br />
I thank A. J. Tyre for providing R- functions to<br />
calculate LHC matrix <strong>and</strong> PRCC’s. His editorial<br />
comments also greatly improved this paper.<br />
868
6. REFERENCES<br />
Blower, S.M., <strong>and</strong> H. Dowlatabadi, Sensitiv ity <strong>and</strong><br />
uncertainty analysis of complex models of<br />
disease transmission: an HIV model, as an<br />
example, <strong>International</strong> Statistical Review, 62,<br />
229-243, 1994.<br />
Chambers, R.J., Oviposition by aphidopha-gous<br />
hoverflies (Diptera: Syrphidae) in relation to<br />
aphid density <strong>and</strong>distribution in winter wheat,<br />
p. 115-121, In L. Polgar, et al., eds. Behaviour<br />
<strong>and</strong> impact of Aphidophaga. SPB Academic<br />
Publish-ing, Den Haag, Netherl<strong>and</strong>s, 1991.<br />
Conover, W.J., Practical nonparametric statistics.<br />
2 ed. John Wiley <strong>and</strong> Sons Inc, New<br />
York.1980.<br />
Dean, G.J., Effect of temperature on cereal aphids<br />
Metopolophium dirhodum (Wlk),<br />
Rhopalosiphum padi (L.) Macrosiphum<br />
avenae (F.) (Hem., Aphididae), Bulletin of<br />
Entomological Research, 63, 401-409, 1974.<br />
Gosselke, U., H. Triltsch, D. Rossberg, <strong>and</strong> B.<br />
Freier, GETLAUS01 - the latest version of a<br />
model for simulating aphid population dynamics<br />
in dependence on antagonists in<br />
wheat, Ecological <strong>Modelling</strong>, 145, 143-157,<br />
2001.<br />
Groeger, U., Untersuchungen zur Regulation von<br />
Getreideblattläusen unter Einfluß der<br />
L<strong>and</strong>schaftsstruktur, Agrarökologie, 6, 1-169,<br />
1992.<br />
Krebs, C.J., Ecological Methodology Har-per <strong>and</strong><br />
Row, New York, USA.1989.<br />
Leffelaar, P.A., On systems analysis <strong>and</strong> simu -<br />
lation of ecological processes with examples<br />
in CSMP, FST <strong>and</strong> FORTRAN. 2nd ed. Kluwer<br />
Academic Publishers, Dordrecht, The Netherl<strong>and</strong>s.1999.<br />
Mangel, M., Dynamic information in uncertain <strong>and</strong><br />
changing Worlds, Journal of Theoretical<br />
Biology, 146, 317-332, 1990.<br />
Ohnesorge, B., <strong>and</strong> A. Viereck, Zur Befalls -<br />
dichteabschätzung bei Getreideblattläu-sen,<br />
Zeitschrift für Pflanzenkrankhei-ten und<br />
Pflanzenschutz, 90, 213-219, 1983.<br />
Rushton, S.P., P.W.W. Lurz, J. Gurnell, <strong>and</strong> R.<br />
Fuller, <strong>Modelling</strong> the spatial dynamics of<br />
parapoxvirus disease in red <strong>and</strong> grey squirrels:<br />
a possible cause of the decline in the red<br />
squirrel in the UK, Journal of Applied<br />
Ecology, 37, 997-1012, 2000a.<br />
Rushton, S.P., G.W. Barreto, R.M. Cormack, D.W.<br />
Macdonald, <strong>and</strong> R. Fuller, Model-ing the<br />
effects of mink <strong>and</strong> habitat fragmentation on<br />
the water vole, Journal of Applied Ecology,<br />
37, 475-490, 2000b.<br />
Simon, J.C., C.A. Dedryver, J.S. Pierre, S. Tanguy,<br />
<strong>and</strong> P. Wegorek, The influence of clone <strong>and</strong><br />
morph on parameters of intrinsic rate of<br />
increase in the cereal aphids Sitobion avenae<br />
<strong>and</strong> Rhopalo-siphum padi, Entomologia<br />
Experimen-talis et Applicata, 58, 211-220,<br />
1991.<br />
Symondson, W.O.C., K.D. Sunderl<strong>and</strong>, <strong>and</strong> M.H.<br />
Greenstone, Can generalist predators be<br />
effective biocontrol agents, Annual Review of<br />
Entomology, 47, 561-594, 2002.<br />
Tenhumberg, B., Untersuchungen zur Popula -<br />
tionsdynamik von Syrphiden in Winterweizenbeständen<br />
und Quantifi-zierung ihrer<br />
Bedeutung als Antago-nisten von<br />
Getreideblattläusen. PhD, University of<br />
Göttingen, Gòttingen (FRG), 1992.<br />
Tenhumberg, B., Estimating predatory efficiency<br />
of Episyrphus balteatus (Diptera: Syrphidae)<br />
in cereal fields, <strong>Environmental</strong> Entomology,<br />
24, 687-691, 1995.<br />
Tenhumberg, B., <strong>and</strong> H.M. Poehling, Studies on<br />
the efficiency of syrphid larvae, as predators<br />
of aphids in winter wheat, p. 281-288, In A.<br />
Dixon <strong>and</strong> I. Hodek, eds. Behaviour <strong>and</strong><br />
impact of Aphidophaga, Proceedings of the<br />
4th meeting of the IOBC working group. SPB<br />
Academic Publishing,, Den Haag, Netherl<strong>and</strong>s,<br />
1991.<br />
Tenhumberg, B., <strong>and</strong> H.M. Poehling, Syrphids as<br />
natural enemies of cereal aphids in Germany:<br />
Aspects of their biology <strong>and</strong> efficacy in<br />
different years <strong>and</strong> regions, Agriculture Ecosystems<br />
& Environment, 52, 39-43, 1995.<br />
Tenhumberg, B., A.J. Tyre, <strong>and</strong> B.D. Roitberg,<br />
Stochastic variation in food availability in -<br />
fluences weight <strong>and</strong> age at maturity, Journal<br />
of Theoretical Biology, 202, 257-272, 2000.<br />
Tenhumberg, B., M. Keller, H.P. Possingham, <strong>and</strong><br />
A.J. Tyre, The effect of resource aggre gation<br />
at different scales: optimal foraging behaviour<br />
of Cotesia rubecula, American Naturalist,<br />
158, 505-518, 2001.<br />
Tenhumberg, B., A.J. Tyre, A.R. Pople, <strong>and</strong> H.P.<br />
Possingham, Do harvest refuges buffer kangaroos<br />
against evolutionary responses to<br />
selective harvesting, Ecology, in press.<br />
Watt, A.D., <strong>and</strong> F.G. Dixon, The role of cereal<br />
growth stages <strong>and</strong>crowding in the induction<br />
of alatae in Sitobion avenae <strong>and</strong> its consequences<br />
for population growth., Ecological<br />
Entomology, 6, 441-447, 1981.<br />
869
The Coexistence of Plankton Species with Various<br />
Nutrient Conditions: A Lattice Simulation Model<br />
T. Miyazaki a , T. Togashi a , T. Suzuki b , T. Hashimoto b , K. Tainaka b <strong>and</strong> J. Yoshimura a,b,c<br />
a<br />
Marine Biosystems Research Center, Chiba University, 1 Uchiura, Amatsu-Kominato, 229-5502, Japan<br />
b<br />
Department of Systems Engineering, Shizuoka University, Hamamatsu, 432-8561, Japan<br />
c<br />
Department of <strong>Environmental</strong> <strong>and</strong> Forest Biology, State University of New York College of <strong>Environmental</strong><br />
Science <strong>and</strong> Forestry, Syracuse, New York 13210, USA<br />
Abstract: In aquatic ecosystems, species diversity is known to be higher in poor nutrient conditions. The<br />
enrichment of nutrition often induces the loss of biodiversity. This phenomenon is called the paradox of<br />
enrichment, since higher nutrient levels can support more species. Furthermore, the species diversity is<br />
usually high in most natural communities of phytoplankton. However, the niches of planktonic algae seem<br />
almost identical in apparently homogeneous, aquatic environments. Therefore, the high species diversity of<br />
phytoplankton is incomprehensible <strong>and</strong> called the paradox of plankton. Mathematical studies show that local<br />
coexistence of competitive species is rare. In a competitive community, the most superior species eliminates<br />
all the inferior species in the long run. Experimental results using chemostats also support this theoretical<br />
prediction. Thus we have no sound explanation for the local coexistence of many planktonic species in low<br />
nutrient conditions. Here we build a lattice model of ten planktonic species. All ten species are under<br />
competition for space in a relatively large lattice space. We report a few cases of simulation run. Simulation<br />
shows that, in an ecological time scale, coexistence of many species is observed when all species have low<br />
identical birth rates. We also show that, when the average birth rates are high, the most superior species<br />
exclude all the inferior species immediately. Our results suggest that competition for space does not function<br />
among species, when the densities of species are extremely low. The results of current simulation<br />
experiments may be related to the paradox of enrichment as well as that of plankton.<br />
Keywords: Paradox of plankton; Species diversity; Coexistence; Lattice model; Paradox of enrichment<br />
1. INTRODUCTION<br />
Enrichment is empirically known to reduce the level<br />
of species diversity of animal <strong>and</strong> plant<br />
communities. However, a community should be<br />
able to support more species with enrichment<br />
because of increased productivity. Therefore, the<br />
loss of biodiversity with enrichment is<br />
counterintuitive <strong>and</strong> called the paradox of<br />
enrichment [Rosenzweig, 1975, 1995, Tilman,<br />
1982]. Here we limit our argument in the aquatic<br />
ecosystems.<br />
In the aquatic systems, the loss of biodiversity is<br />
often correlated with enrichment of water<br />
conditions [Ogawa <strong>and</strong> Ichimura, 1984a, 1984b,<br />
Ogawa, 1988]. High biodiversity is observed in still<br />
waters with low nutrients. Recent pollution due to<br />
domestic <strong>and</strong> factory wastewaters increases the<br />
nutrient levels of almost all aquatic systems,<br />
invoking the serious enrichment problem.<br />
The nutrient concentrations are low in most wellpreserved<br />
aquatic ecosystems. The species<br />
diversities of phytoplankton are usually high in<br />
these ecosystems. Because water environment is<br />
homogeneous <strong>and</strong> the niches of phytoplankton are<br />
almost identical, the most superior species should<br />
exclude all the rest of inferior species. However, it<br />
seems that many species of phytoplankton usually<br />
coexist in a single natural aquatic ecosystem<br />
without apparent competitive exclusions. This<br />
unexplainable phenomenon is called the paradox of<br />
plankton after Hutchinson [1961].<br />
In contrast with the observed high diversity in<br />
natural aquatic ecosystems, theoretical studies<br />
predict that local coexistence of species is highly<br />
limited. Many mathematical analyses <strong>and</strong><br />
simulations show that local coexistence of<br />
competitive species is usually impossible unless<br />
interspecific competition is weaker than<br />
intraspecific competition. Simulation experiments<br />
870
usually show that the outcomes are the dominance<br />
of a single species resulting in the exclusion of all<br />
the rest (inferior) species.<br />
To explain the extreme diversity in some<br />
communities, external factors are suggested, such as<br />
climatic changes, immigration from other habitats.<br />
Many mathematical models <strong>and</strong> theories try to<br />
achieve coexistence of many species by means of<br />
external factors, such as environmental changes<br />
(stochasticity), immigration of adjacent individuals.<br />
However, such external factors do not necessarily<br />
seem to be applicable to the diversity of plankton.<br />
Many empirical studies of small pond <strong>and</strong> lake<br />
ecosystems with low nutrient still waters show high<br />
species diversity. There seems no indication of<br />
external factors in these ecosystems in general.<br />
Thus we have three-fold mysteries in the planktonic<br />
communities with low nutrient conditions: (1)<br />
paradox of enrichment, (2) paradox of plankton <strong>and</strong><br />
(3) competitive exclusion of species with identical<br />
niches.<br />
In this paper, we built a simulation model of ten<br />
planktonic species in a large lattice habitat. We<br />
assumed that the competition between planktonic<br />
species (or individuals) is achieved through the<br />
growth difference of species. We carried out quite a<br />
few simulation runs with various birth rates,<br />
keeping the constant death rate. We show a typical<br />
dynamics of low <strong>and</strong> high nutrition conditions. In<br />
low nutrient conditions, we show that many species<br />
persist <strong>and</strong> coexist in ecological time. In high<br />
nutrient conditions, we show the case of instant<br />
elimination of all the inferior species by the most<br />
superior species. We discuss the implication of the<br />
current simulation trials in relation to the paradoxes<br />
of enrichment <strong>and</strong> plankton.<br />
2. LATTICE MODEL OF MULTIPLE<br />
COMPETITIVE SPECIES<br />
2. 1 Lattice Model<br />
We consider a competitive ecosystem of ten<br />
planktonic species (S i ; i = 1,.., 10) on a large square<br />
lattice (500×500 cells). Birth <strong>and</strong> death processes<br />
are given by<br />
X<br />
X<br />
i<br />
i<br />
+<br />
+<br />
X<br />
O<br />
O<br />
i<br />
d<br />
m<br />
i<br />
→<br />
b<br />
i<br />
i<br />
→ 2 X<br />
→<br />
O<br />
O<br />
+<br />
i<br />
X<br />
i<br />
(1)<br />
( 2 )<br />
(3)<br />
where each lattice site is either occupied by species<br />
S i (X i ) or empty (O). The reactions (1), (2) <strong>and</strong> (3)<br />
simulate reproduction (birth), death, <strong>and</strong> dispersal<br />
(movement), respectively. The parameters b i <strong>and</strong> m i<br />
represent the birth <strong>and</strong> death rates of an individual,<br />
respectively. All parameters are kept constant<br />
during a simulation run. The death rate m i is kept at<br />
m i = 0.3 for all simulations. The parameter d i<br />
represents the accidental dispersal (movement) rate<br />
of an individual, where an individual move to one<br />
cell to another, r<strong>and</strong>omly. The dispersal is<br />
implemented to prevent clumping or extreme<br />
aggregation, simulating an aquatic system. The<br />
reaction is carried out in two ways: the contact<br />
process (CP) where interaction occurs between<br />
adjoining lattices [Harris, 1974] <strong>and</strong> the mean-field<br />
simulation (MFS) where interaction globally occurs<br />
between any pair of lattices.<br />
We study two distinct growth conditions assuming<br />
low <strong>and</strong> high productivities. In the high productivity,<br />
we assume that all species have species-specific<br />
birth rates, while in the low productivity, all species<br />
have the identical low birth rate due to the critical<br />
threshold for growth rates. We set b i = 0.5 (i =<br />
1,..,10) for the low productivity. At this birth rate,<br />
the net growth (reproductive) rate is positive, but<br />
very close to zero. For the high productivity, we set<br />
b i = 1.01 - 0.01i (i = 1,..,10). Here max b i = b 1 =<br />
1.00, <strong>and</strong> min b i = b 10 = 0.91.<br />
2.2 Simulation Procedure<br />
The simulation procedures for the contact process<br />
(CP) are as follows:<br />
(I) Algal cells are distributed r<strong>and</strong>omly over some<br />
square-lattice points in such a way that each point is<br />
occupied by only one individual cell, if the point is<br />
occupied. The initial density of X i is set to 0.0001<br />
for all simulations.<br />
(II) Each reaction process is performed in the<br />
following three steps.<br />
(i) We perform the single body reaction (2).<br />
Choose one square-lattice point r<strong>and</strong>omly. Let<br />
change the point to O with probability m i , if it is<br />
occupied by a X i individual.<br />
(ii) Next, we perform the two-body reaction (1).<br />
Select one point r<strong>and</strong>omly <strong>and</strong> specify one of<br />
adjacent points. Here the adjacent site is set as the<br />
Neumann neighbors (4 sites: up, down, left <strong>and</strong><br />
right). If the selected pair is X i <strong>and</strong> O, then the latter<br />
point will become X i with probability b i . Here we<br />
employ periodic boundary conditions.<br />
(iii) At last, we perform the two-body reaction (3).<br />
Select one point r<strong>and</strong>omly. If the selected point is X i ,<br />
then we choose another point r<strong>and</strong>omly. If the<br />
second point is not occupied (O), then we move X i<br />
to the second site (interchange X i <strong>and</strong> O).<br />
871
Figure 1. A typical result of population dynamics for the lattice ecosystem of ten competitive species S i (i =<br />
1,..,10). A: the contact process. An identical low birth rate b i = 0.5 is assumed for all species, implying a poor<br />
nutrient condition. At this birth rate, the net growth (reproductive) rate is positive, but very close to zero. B:<br />
the contact process. High different birth rates are assumed for ten species, such that b i = 1.01 - 0.01i (i =<br />
1,..,10). Here max b i = b 1 = 1.00, <strong>and</strong> min b i = b 10 = 0.91. C: the mean-field simulation (MFS) for A. D: the<br />
mean-field simulation (MFS) for B. The death rate m i = 0.3 <strong>and</strong> the dispersal rate d i = 0.01. The time is<br />
measured by the Monte Carlo step. The total number of square-lattice sites is 500 × 500.<br />
872
Figure 2. Temporal dynamics of ten competitive species in the spatial ecosystems in ecological time with an<br />
identical low birth rate b i = 0.5. Top: the contact process model (exerted from Fig. 1A). Bottom: the meanfield<br />
simulation (MFS). The density of each species (left cells) <strong>and</strong> the remaining number of species (right<br />
cells) is plotted against time evolution. Up to 20,000 Monte Carlo steps are shown.<br />
(III) Repeat the step (II) by L × L times, where L ×<br />
L is the total number of the square-lattice sites.<br />
Here we set L = 500. This step is called a Monte<br />
Carlo step [Tainaka, 1988]<br />
(IV) Repeat the step (III) for a specific length, that<br />
is 100,000 Monte Carlo steps.<br />
In the case of mean-field simulation (MFS), the<br />
above procedure is slightly different. In the contact<br />
process, the interaction (1) occurs between adjacent<br />
lattice sites. However, in the MFS, the long-ranged<br />
(global) interaction is allowed: the reaction (1)<br />
takes place between any pair of lattice sites. The<br />
second sentence in Step (ii) is changed as follows:<br />
(ii’) … Two lattice sites are r<strong>and</strong>omly <strong>and</strong><br />
independently selected.<br />
Note also that the reaction (3) has no meaning<br />
(effect) on the dynamics in the MFS.<br />
3. RESULTS<br />
We run a long-term simulation for various birth rate<br />
conditions, while keeping the death rate constant at<br />
m i = 0.3. A typical example of long-term dynamics<br />
is shown in Fig. 1 for both low <strong>and</strong> high birth rates.<br />
There is a threshold value for birth rates to achieve<br />
positive or net reproductive rates, resulting in zero<br />
net growth where the birth <strong>and</strong> death rates are<br />
balanced. When the birth rates are slightly lower<br />
than this threshold value (for example, b i = 0.49),<br />
all species go extinct quite rapidly. In an ecological<br />
time scale of about 10,000 time steps (Monte Carlo<br />
873
Figure 3. Snapshots of a temporal pattern in the<br />
lattice model (CP) of Fig. 1 at a time point<br />
(top: 20,000, bottom: 40,001). The birth rate b i<br />
= 0.5. The density of S i are listed above. The<br />
100x100 sites are cut from 500x500 sites.<br />
steps, almost all species still coexist in the<br />
ecosystem, when the birth rate is positive, but close<br />
to zero growth rates (b i = 0.50; Fig. 1A). In much<br />
longer time scales, most species are eliminated by<br />
chance, as a r<strong>and</strong>om walk.<br />
In contrast, when the birth rates are significantly<br />
higher, all inferior species are immediately<br />
excluded by the most superior species in a very<br />
short time much shorter than 5,000 time steps (Fig.<br />
1B). When the birth rates are high <strong>and</strong> different<br />
among species, only one dominant species with the<br />
highest growth rate eliminate all the rest species<br />
immediately in almost any simulation. This happens<br />
irrespective of the simulation methods (either<br />
contact process models or mean-field simulations).<br />
Fig. 2 shows the results of the contact process <strong>and</strong><br />
the mean field simulation in which the birth rate is<br />
close to zero growth rate value. Note that the<br />
dynamics of up to 20,000 time steps is long enough<br />
to cover ecological time scales. In both the contact<br />
process <strong>and</strong> the mean-field simulation, the<br />
coexistence of most species is maintained in these<br />
time steps (Fig. 2). Between the two simulations,<br />
there are only slight differences in the average<br />
density <strong>and</strong> extinction dynamics. In the contact<br />
process, the average density is slightly higher (Fig.<br />
2, top-left) than that of the mean-field simulation<br />
(Fig. 2, bottom-left). The remaining number of<br />
species is also higher in the contact process (Fig. 2,<br />
top-right) in contrast with that in the mean-field<br />
simulation (Fig. 2, bottom-right).<br />
These slight differences should be due to the spatial<br />
structure of lattice model in the reaction (e.g. step<br />
(II)). Fig. 3 shows the temporal pattern dynamics of<br />
Fig. 1 at a time point of 20,000 <strong>and</strong> 40,001 time<br />
steps. Fig. 3 clearly shows clumping tendency. It<br />
indicates the effects of lattice spatial structure on<br />
the coexistence trends in Fig. 2.<br />
We also tested various conditions in birth rates. For<br />
example, we run the simulation with low variable<br />
birth rates (b i = 0.49 + 0.01i). In the low density,<br />
the effects of the 0.01 differences in birth rate on<br />
the dynamics are extraordinary. All the inferior<br />
species are instantly eliminated from the ecosystem.<br />
The elimination rate is a few times faster than that<br />
in the high birth rates. The implication of<br />
variability (differences) in low <strong>and</strong> high birth rates<br />
will be discussed in detail later in the discussion.<br />
4. DISCUSSION<br />
In the current simulations, we vary birth rates of ten<br />
species to see the persistence <strong>and</strong> coexistence of<br />
species in ecological time scales. When the birth<br />
rates of ten species are identical, most species<br />
coexist. However, a slight difference are introduced,<br />
the species with the highest growth rate eliminate all<br />
other species. In natural ecosystems of poor nutrient<br />
conditions, growth rates are closely zero <strong>and</strong><br />
virtually no species variability in growth rate is<br />
expected [see e.g., Tilman, 1982]. Thus the<br />
coexistence in ecological time scale in our<br />
simulation is underst<strong>and</strong>able in nutrient-limited<br />
aquatic systems. In contrast, in nutrient-rich<br />
conditions, the species-specific growth rates should<br />
be extremely variable [Kuwata <strong>and</strong> Miyazaki, 2000].<br />
Thus, the elimination of all the inferior species<br />
should take place due to the competitive interaction<br />
between species.<br />
874
The lattice size (500 × 500) in our simulation is<br />
larger than usual lattice models, but it is still<br />
extremely small in comparison with the real sizes of<br />
natural aquatic ecosystems. The total densities of<br />
plankton in natural ecosystems are lower in several<br />
magnitudes than those in our low-density simulation.<br />
Due to the computational limitation of lattice size<br />
(500×500), it is impossible to get the stable steady<br />
state with lower birth rates (closer to the threshold<br />
value. The general trends we observed in the lattice<br />
simulation could be much more significant in the<br />
natural ecosystems.<br />
Our simulation shows that the local coexistence of<br />
phytoplanktonic species in ecological time may be<br />
achieved by the internal factors alone. The<br />
coexistence in the ecosystem is virtually not<br />
coexistence at the same site in the lattice; rather<br />
almost all individual planktonic species survive <strong>and</strong><br />
reproduce independently from other species due to<br />
the vast space between them. Low nutrient<br />
conditions of natural ecosystems may prohibit the<br />
reproduction to reach the high density that incurred<br />
competitive interaction.<br />
Even though the current simulations are limited <strong>and</strong><br />
only trial runs with limited combinations of<br />
parameters are carried out, these results indicate<br />
that local coexistence of many species in very low<br />
birth rates is possible, while the instant elimination<br />
of all inferior species by a single dominant species<br />
is also possible. Thus the mechanisms underlining<br />
the current lattice model may relate to the paradox<br />
of enrichment, as well as that of plankton.<br />
Microcystis novacekii (Cyanobacteria) <strong>and</strong><br />
Scenedesmus quadricauda (Chlorophyta):<br />
simulation study, Ecological <strong>Modelling</strong>, 135,<br />
81-87, 2000.<br />
Levins, R., Coexistence in a variable environment,<br />
American Naturalist, 114, 765-783, 1978.<br />
Ogawa, Y., Net increase rates <strong>and</strong> dynamics of<br />
phytoplankton populations under<br />
hypereutrophic <strong>and</strong> eutrophic conditions, Jpn.<br />
J. Limnol., 49(4), 261-268, 1988.<br />
Ogawa, Y. <strong>and</strong> S. Ichimura, Phytoplankton diversity<br />
in isl<strong>and</strong> waters of different trophic status, Jpn.<br />
J. Limnol., 45(3), 173-177, 1984.<br />
Ogawa, Y. <strong>and</strong> S. Ichimura, The relationship<br />
between phytoplankton diversity <strong>and</strong> trophic<br />
status of isl<strong>and</strong> waters, Jap. J. Ecol., 34, 27-33,<br />
1984.<br />
Rosenzweig, M.L., Paradox of enrichment:<br />
destabilization of exploitation ecosystems in<br />
ecological time, Science, 171, 385-387, 1971.<br />
Rosenzweig, M.L., Species diversity in space <strong>and</strong><br />
time, Cambridge University Press, 1995.<br />
Rosenzweig, M.L., Win-win ecology: how earth's<br />
species can survive in the midst of human<br />
enterprise, Oxford University Press, 2003.<br />
Tilman, D., Resource competition <strong>and</strong> community<br />
structure, Princeton University Press, 296pp,<br />
Princeton, 1982.<br />
5. ACKNOWLEDGEMENTS<br />
This work was partially supported by grants-in-aids<br />
from the Ministry of Cultures, Education <strong>and</strong><br />
Sciences in Japan to T. M., T. T., K.T. <strong>and</strong> J. Y.<br />
6. REFERENCES<br />
Alam, K., <strong>and</strong> A.C. Falkl<strong>and</strong>, Home Isl<strong>and</strong><br />
groundwater modelling Stage 2, ECOWISE<br />
<strong>Environmental</strong>, Canberra, 1998.<br />
Grover, J.P., Resource competition, Chapman &<br />
Hall, 342pp, London, 1977.<br />
Harris, T.E., Contact interaction on a lattice, Ann.<br />
Prob., 2, 969-988, 1974.<br />
Hulsman, J. <strong>and</strong> F.J. Weissing, Biodiversity of<br />
plankton by species oscillations <strong>and</strong> chaos,<br />
Nature, 402, 407-410, 1999.<br />
Hutchinson, G.E., The paradox of plankton,<br />
American Naturalist, 95,137-145, 1961.<br />
Kuwata, A. <strong>and</strong> T. Miyazaki, Effects of ammonium<br />
supply rates on competition between<br />
875
ÅØÑØÐ ÑÓÐÐÒ Ó ÖÑÙÐ ÐÐ ÐÓÓÑ×<br />
ʺʺ ËÖÖ<br />
ÑÖÝÓÐÓÝ Ê×Ö ÍÒظ ÁÒÒ ËØØ×ØÐ ÁÒ×ØØÙØ<br />
¾¼¿¸ º ̺ ÊÓ¸ ÃÓÐØ ¼¼½¼¸ ÁÒ<br />
¹ÑÐ ÖÑ Ö×кºÒ»ÖÑÖÙÔ¾ÝÓÓºÓÑ<br />
×ØÖØ<br />
ÈÐÒØÓÒ × Ø ×× Ó ÐÐ ÕÙØ ÓÓ Òº ÈÝØÓÔÐÒ¹<br />
ØÓÒ¸ Ò ÔÖØÙÐÖ¸ ÓÙÔ× Ø ¬Ö×Ø ØÖÓÔ ÐÚк ÈÐÒØÓÒ<br />
Ó Ù ×ÖÚ× ÓÖ ÓÙÖ Öظ ÓÓ ÓÖ ÑÖÒ Ð¸ ÓÜÝÒ<br />
ÓÖ ÙÑÒ Ð Ò Ð×Ó ×ÓÖ× Ð Ó Ø ÖÓÒ ¹¹<br />
ÓÜ ÖÓÑ Ø ÖØ ØÑÓ×ÔÖº Ì ÝÒÑ× Ó ÖÔ<br />
´ÓÖ Ñ××Úµ ÒÖ× ÓÖ Ö× Ó ÔÐÒØÓÒ ÔÓÔÙÐØÓÒ× ×<br />
Ò ÑÔÓÖØÒØ ×ÙØ Ò ÑÖÒ ÔÐÒØÓÒ ÓÐÓÝ Ò Ò¹<br />
ÖÐÐÝ ØÖÑ × ³ÐÓÓѳº ÀÖÑÙÐ ÐÐ ÐÓÓÑ× ´À×µ<br />
Ú ÚÖ× «Ø× ÓÒ ÙÑÒ Ðظ ¬×Öݸ ØÓÙÖ×Ѹ Ò<br />
ÒÚÖÓÒÑÒغ ÁÒ ÖÒØ ÝÖ× ÓÒ×ÖÐ ×Òج ØØÒ¹<br />
ØÓÒ × Ò Ô ÓÒ À׺ ÌÓÜ ×Ù×ØÒ× ÖÐ×<br />
Ý ÖÑÙÐ ÔÐÒØÓÒ ÔÐÝ Ò ÑÔÓÖØÒØ ÖÓÐ Ò Ø× ÓÒ¹<br />
ØÜغ ÁÒ Ø× ÔÔÖ¸ ÑØÑØÐ ÑÓÐ ÓÒ××ØÒ ØÛÓ<br />
ÖÑÙÐ ÔÝØÓÔÐÒØÓÒ Ò ÞÓÓÔÐÒØÓÒ ×Ý×ØÑ ÛÐÐ ×¹<br />
Ù×׺ Ì ÒÐÝØÐ × ÛÐÐ × ÒÙÑÖÐ ¬ÒÒ× ÛÐÐ <br />
ÚÖ¬ ØÖÓÙ ÓÙÖ ÜÔÖÑÒØÐ Ó×ÖÚØÓÒ× Û Û×<br />
ÖÖ ÓÒ Ø ×ØÖÒ ÔÖØ Ó Ý Ó ÒÐ ÓÖ Ø Ð×Ø<br />
ØÖ ÝÖ׺<br />
ÃÝ ÛÓÖ×<br />
ÌÛÓ ÖÑÙÐ ÔÝØÓÔÐÒØÓÒ ÞÓÓÔÐÒØÓÒ<br />
ÐÓÓÑ ÓÑÔØØÓÒ ÓÐÓÐ ÓÒØÖÓÐ<br />
½º ÁÒØÖÓÙØÓÒ<br />
ÈÝØÓÔÐÒØÓÒ Ö Ø ×ÓÙÖ Ó ÐÑÓ×Ø ÐÐ ÒÖÝ<br />
Ô××Ò ØÖÓÙ ÕÙØ ÓÓ Û× Ò ÓÑÔÖ×<br />
×ÓÑ ¼± Ó Ø ØÓØÐ ¬Ü ÐÓÐ ÔÖÑÖÝ ÔÖÓÙ¹<br />
ØÚØÝ´ÐÓÛ׸½µº ÐÖ ÓÑÔÓÒÒØ Ó Ø×<br />
ÔÖÓÙØÚØÝ Ò ØØÖÙØ ØÓ Ø ÓÙÖÖÒ<br />
Ó ÓØ ××ÓÒÐ Ò ×ÔÓÖ Ð ÐÓÓÑ× ØØ<br />
ÓÖÑ × ÔØ× ÓÚÖ Ø ÓÒ³× ×ÙÖ Ò Ö×<br />
ÐÓÐÞ ÒÙØÖÒØ ÒÖÑÒغ ÆÓÛ Ø Ö×ÖÖ×<br />
Ö ÑÒÐÝ ÓÒÒØÖØÒ ÓÒ ÐÓÓÑ ÝÒÑ× ÛØ<br />
×ÔÐ ÑÔ×× ÓÒ ÖÑÙÐ ÐÐ ÐÓÓÑ× ´À×µº<br />
Ì ÔÐÒØÓÒ ÐÓÓÑ× ÑÝ ØÓÖÞ ÒØÓ<br />
×ÔÖÒ ÐÓÓÑ× Ò Ö Ø׺ ËÔÖÒ ÐÓÓÑ× ÓÙÖ<br />
××ÓÒÐÐÝ Ù ØÓ Ò× Ò ØÑÔÖØÙÖ ÓÖ ÒÙØÖ¹<br />
ÒØ ÚÐÐØݺ Ê Ø× Ö Ø Ö×ÙÐØ Ó ÐÓй<br />
Þ ÓÙØÖ× ××ÓØ ÛØ ÛØÖ ØÑÔÖØÙÖ<br />
´ÌÖÙ×ÓØ Ò ÖÒÐݸ½µº<br />
ÈÝØÓÔÐÒØÓÒ ×Ô× Û Ù× ¬× ÔÓ×ÓÒ¹<br />
Ò ´ÙØÖµ Ò ×ÐЬ׹ÚØÓÖ ÔÓ×ÓÒÒ Ó<br />
ÙÑÒ× Ö ÐÖÐÝ ÖÑÙÐ ´ÀÐÐÖ«¸ ½¿µº<br />
ÀÖÑÙÐ ×Ô× Ò ÖØÖÞ Ý ØÖ ×¹<br />
ØÒØ ÑÓÐ× Ó ÒØÚ ÑÔØ ×Ù × ³ØÓܳ¸<br />
ÒÓÜÓÙ× Ò ÒÙ×Ò Ù× ØÖ ×ÝÒÓÒÝÑÓÙ×ÐÝ ÓÖ<br />
ØÓ ÖØÖÞ ÖÐØÚ Ö Ó ÑÔغ ÊÒØ<br />
×ØÙ× ÖÚÐ ØØ ×ÓÑ ØÑ× ÐÓÓÑ Ó ÖØÒ<br />
ÖÑÙÐ ×Ô× Ð× ØÓ ÖÐ× Ó ÓØ ØÓÜÒ× Ò<br />
ÐÐÐÓÔØ ×Ù×ØÒ׺ ÐÐÐÓÔØ ×Ù×ØÒ×<br />
ØÒ ØÓ ÖØÐÝ ØÖØ Ò ÑÝ ÔÝ×ÓÐÓ¹<br />
ÐÐÝ ÑÔÖ¸ ×ØÙÒ¸ ÖÔи ÒÙ ÚÓÒ Ö¹<br />
ØÓÒ×¸Ò ÐÐ ÖÞÖ׺ ÜÑÔÐ× Ó Ø× ÚÖ×<br />
ÑÓÐ× Ó ÑÔØ ÛÖ ÚÒ Ý ËÑÝ ´½¾µº<br />
ÁÒÓÖÑØÓÒ ÓÒ Ø ÖÒØÐÝ ×ÓÚÖ¸ ÖÑÖ¹<br />
Ð ÑÙ× ÔÖØÓÖÝ ÚÓÖ Ó Ø ÖÑÙÐ<br />
ÒÓÐÐظ Ȭ×ØÖ Ô× Û× ÔÖ×ÒØ<br />
Ý ÙÖÓÐÖ Ò Ð×ÓÛ ´½µ Ò ÄÛØÙ×<br />
Ø Ðº ´½µº ÌÖ Ö ØÐ×Ø Ø «ÖÒØ<br />
ÑÓ× Ò ÑÒ×Ñ× Ý Û ÖÑÙÐ ÔÝØÓ¹<br />
ÔÐÒØÓÒ ×Ô× Ò Ù× ÑÓÖØÐØݸ ÔÝ×ÓÐÓ¹<br />
Ð ÑÔÖÑÒظ ÓÖ ÓØÖ ÒØÚ Ò ×ØÙ «Ø׺<br />
Ì× ÐÐ ÒØÓ ØÛÓ ÒÖÐ ØÝÔ× ÒÓÒ ÑÐ ¹<br />
Ø× Û Ð ØÓ ×ØÖÚØÓÒ ÓÖ Ù× ÖÑÙРѹ<br />
ÒÐ Ò ÔÝ×Ð Ñ¸Ò ÑÐ «Ø×<br />
ØØÖÙØÐ ØÓ ÔÝ×йÑÐ ÖØÓÒ׸ÔÝÓ¹<br />
ØÓÜÒ׸ ÓÖ ÓØÖ ÑØÓÐØ׺ ÑÓÒ Ø Ñ¹<br />
ÐÐÝ ÖÑÙÐ ÑÒ×Ñ׸¹Ó«× Ù ØÓ ÒÓÜ<br />
ÓÖ ÝÔÓÜ ÓÐÐÓÛÒ ÐÓÓÑ× Ó ÐÖ¸ÖÐØÚÐÝ ÙÒ¹<br />
ÖÞ ×Ô׸ ×Ù ×ÖØÙѸ Ö ÛÐÐ ÒÓÛÒ<br />
´ÖÒÐ Ø Ðº¸ ½µº ÆÓØÐÙ ÔÖÓÙ× Ö¹<br />
ÑÖÐÝ ×ÑÔÐ ØÝÓØÓÜÒ ÆÀ ´Ç Ò<br />
Æ×Ó¸½µº ÅÓÖØÐØÝ Ò ÓÙÖ ØÖÓÙ ÖØ<br />
Ò×ØÓÒ ´ÒÓØÓÜÒµ Ó Ø ÖÑÙÐ ×Ô׸ÙÔÓÒ<br />
ÜÔÓ×ÙÖ ØÓ ×ÖØ ØÓÜÒ× ´ÜÓØÓÜÒ×µ¸ ÓÖ ÖÓÑ<br />
ØÓÜÒ ÚØÓÖÒ ØÖÓÙ Ø ÓÓ Û¸ ÓÑÔ¹<br />
Ò Ý ÓÒÓÖÑØÓÒÐ Ò× Ò Ø ØÓÜ ÔÖÒ¹<br />
ÔÐ× Ò ØÖ ÔÓØÒ× ´ËÑÞÙ ½ ×ÙÑÓØÓ<br />
Ò ÅÙÖظ ½¿µº Ì ÑÔØ ÑÝ ÓÒ¹<br />
876
877
878
879
880
881
882
Implications of processing spatial data from a forested<br />
catchment for a hillslope hydrological model<br />
T. Kokkonen a , H. Koivusalo a , A. Laurén b , S. Penttinen c , S. Piirainen b , M. Starr d , <strong>and</strong> L. Finér e<br />
a Laboratory of Water Resources, Helsinki University of Technology; tkokko@cc.hut.fi<br />
b Joensuu <strong>and</strong> d Vantaa Research Centres, Finnish Forest Research Institute<br />
c Kuopio Office, Geological Survey of Finl<strong>and</strong><br />
e Faculty of Forestry, University of Joensuu<br />
Abstract: Finl<strong>and</strong> has committed to both increasing timber production <strong>and</strong> decreasing the nutrient loading<br />
caused by forestry, which calls for development of methods to assess environmental impacts of forest<br />
management. A simulation model based on the concept of a typical hillslope is applied to describe water <strong>and</strong><br />
nitrogen processes in a forested catchment. Application of the model requires that spatially distributed<br />
catchment data are processed to create parameterisation for a vertical two-dimensional profile. In such a twodimensional<br />
catchment description, behaviour of the system at different distances to a stream can be<br />
considered. This study explores 1) how changing the location of a clear-cut area is reflected in model results,<br />
<strong>and</strong> 2) how the inevitable simplifications when representing a catchment as a single hillslope may affect the<br />
model outcome. The results suggest that description of the catchment with a single two-dimensional profile is<br />
a reasonable approximation as long as areas having a high fraction of subsurface runoff (> 60-70%) are not<br />
combined with areas where the surface runoff component is dominant. At low hydraulic conductivities the<br />
nitrate load was strongly controlled by the distance from the cut area to the stream, <strong>and</strong> the load increased<br />
almost linearly with the inverse of the distance. But when the conductivity value became sufficiently large,<br />
the effect of the cutting location became smaller, <strong>and</strong> the relationship to the inverse of the distance was<br />
obscured by snowmelt timing differences in open <strong>and</strong> forested environments.<br />
Keywords: Catchment; Forest harvesting; Hydrology; Nitrate; Mathematical modelling; Spatial description<br />
1. INTRODUCTION<br />
<strong>Environmental</strong> impacts of forest management<br />
practises are particularly important to countries<br />
like Finl<strong>and</strong>, where 75 % of the l<strong>and</strong> area is<br />
covered with forests <strong>and</strong> 84 % of those forests are<br />
managed for timber production (Finnish Forest<br />
Research Institute, 2002). At the national level,<br />
forest management contributes ca. 9% of the<br />
nitrogen load to surface waters (Kenttämies, 2003).<br />
Locally forest management can be the most<br />
significant human activity responsible for nutrient<br />
leaching to lakes <strong>and</strong> rivers. Finl<strong>and</strong> has<br />
committed to increase timber production <strong>and</strong> to<br />
decrease the nutrient loading caused by forestry<br />
(Ministry of Agriculture <strong>and</strong> Forestry, 1999).<br />
These controversial aims will require development<br />
of computation methods to support the forest<br />
manager in planning treatments in such a manner<br />
that the environmental loading is minimised.<br />
The aim of this paper is first to explore how a<br />
hydrological model based on the concept of a<br />
typical hillslope can exploit distributed data<br />
depicting topography <strong>and</strong> l<strong>and</strong>-use of a first-order<br />
headwater catchment. The inevitable<br />
simplifications when representing a catchment as a<br />
single hillslope are discussed <strong>and</strong> assessed. The<br />
model is finally used to assess how the location of<br />
a clear-cut area is reflected in nitrate export.<br />
2. METHODS<br />
2.1. Two-dimensional description of a<br />
catchment<br />
In the present study implications of simplifying the<br />
three-dimensional catchment domain into two<br />
dimensions are explored. The simplification relies<br />
on the concept of a characteristic profile, which is<br />
defined to represent a typical flowpath from a<br />
water divide into the nearest stream (Kokkonen et<br />
al., 2001).<br />
Determination of the surface geometry of a<br />
characteristic profile, i.e. length <strong>and</strong> slope, is based<br />
on an analysis of the digital elevation model of a<br />
catchment. The elevation difference between each<br />
pixel <strong>and</strong> its receiving stream pixel is computed,<br />
883
<strong>and</strong> these differences are categorised according to<br />
the distance from the stream along the flowpath.<br />
Average value of the elevation differences at a<br />
given distance determines the elevation of the<br />
profile at that distance. Convergent or divergent<br />
topography within the catchment is accounted for<br />
with the aid of a width function. The width<br />
distribution along the profile is identified from the<br />
number of pixels located at a given distance. More<br />
information on how the characteristic profile is<br />
determined can be found in Kokkonen et al. (2001)<br />
<strong>and</strong> Koivusalo et al. (2003).<br />
2.2. Hydrological modelling<br />
The characteristic profile model (CPM) applied in<br />
this study comprises separate routines for<br />
estimating meteorological variables beneath the<br />
canopy (Koivusalo <strong>and</strong> Kokkonen, 2002),<br />
calculating accumulation <strong>and</strong> melt of a snowpack<br />
(Koivusalo et al., 2001), <strong>and</strong> describing soil water<br />
movement <strong>and</strong> runoff generation processes along a<br />
typical hillslope (Karvonen et al., 1999). The<br />
canopy <strong>and</strong> snow routines operate at an hourly<br />
time-step, <strong>and</strong> the runoff generation procedure<br />
operates at a daily time scale. The model has been<br />
developed for areas where shallow soils (1 – 3 m)<br />
are underlain by an impermeable bedrock, <strong>and</strong><br />
where the infiltration capacity of the soil exceeds<br />
the intensity of rainfall or snowmelt.<br />
The characteristic profile is divided into vertical<br />
soil columns, which are further divided into soil<br />
layers. Vertical fluxes in all columns are computed<br />
by the Skaggs (1980) approximation to the<br />
Richards equation. Water that cannot infiltrate is<br />
transported downslope the profile as surface runoff<br />
<strong>and</strong> it either reaches the stream, or infiltrates if the<br />
air volume further down the profile allows it. After<br />
the vertical fluxes <strong>and</strong> the resulting groundwater<br />
levels have been resolved, lateral groundwater<br />
flow between soil columns is computed from<br />
Darcy’s law. The groundwater flow from the<br />
column adjoining the stream forms one of the<br />
runoff components <strong>and</strong> is later called groundwater<br />
flow. When the groundwater level in any column<br />
rises above the soil surface, the model generates<br />
exfiltration runoff which is transported downslope<br />
in a similar manner as surface runoff. Groundwater<br />
flow <strong>and</strong> exfiltration runoff together form that part<br />
of runoff that has infiltrated into the soil before<br />
getting discharged into the stream. The sum of<br />
these two runoff components is later referred to as<br />
subsurface runoff.<br />
Hydrological effects of a forest cutting are<br />
described by bypassing the canopy model.<br />
Koivusalo et al. (2003) have demonstrated the<br />
performance of the model against field data<br />
measured in Kangasvaara.<br />
2.3. Nitrogen modelling<br />
Nitrogen dem<strong>and</strong> of the tree st<strong>and</strong> is linearly<br />
related to the rate of photosynthesis, which is<br />
estimated with the FINNFOR forest ecological<br />
model (Kellomäki <strong>and</strong> Väisänen, 1997).<br />
FINNFOR also simulates the canopy litter-fall to<br />
the ground, which forms the input to a litter<br />
decomposition routine modified from the ROMUL<br />
model (Chertov et al., 2001). Nitrogen released<br />
from decomposing litter together with the<br />
atmospheric deposition of nitrogen form the<br />
nitrogen input to the CPM. Nitrogen is transported<br />
along a characteristic profile according to the<br />
water fluxes computed in the CPM. Process<br />
descriptions for nitrification, denitrification <strong>and</strong><br />
retention have been adopted from Jansson <strong>and</strong><br />
Karlberg (2001). While the current model accounts<br />
for nitrate, ammonium, <strong>and</strong> dissolved organic<br />
nitrogen processes, only the nitrate results are<br />
addressed in the present study.<br />
The effect of clear-cutting on nitrogen processes is<br />
described as decreased plant dem<strong>and</strong> <strong>and</strong> as an<br />
instantaneous litter input in the form of logging<br />
residues.<br />
3. SITE AND DATA DESCRIPTION<br />
The study utilises meteorological data at the<br />
Kangasvaara (56 ha) catchment located in eastern<br />
Finl<strong>and</strong> (63º 51’ N, 28º 58’ E). This catchment was<br />
instrumented as part of the VALU project<br />
commencing in 1992 (Finér et al. 1997).<br />
Meteorological data covering the period from 1992<br />
to 2001 were compiled from both on-site<br />
measurements <strong>and</strong> records obtained from the<br />
nearest (ca. 20 km) weather station operated by the<br />
Finnish Meteorological Institute. The forests in<br />
Kangasvaara were dominated by mature<br />
coniferous trees. In late 1996 a total of 35% of the<br />
catchment area was clear-cut. In Kangasvaara 92%<br />
of the area is covered by glacial till, while the<br />
remaining l<strong>and</strong> area is covered by peat.<br />
Long-term mean annual precipitation <strong>and</strong> air<br />
temperature in the area are 700 mm <strong>and</strong> 1.5 ºC,<br />
respectively. The bulk nitrate deposition was 101<br />
kg-N/km 2 /a in the period from 1993 to 1996<br />
(Piirainen et al., 1998), <strong>and</strong> the mean annual nitrate<br />
export was in the order of 2 kg-N/km 2 /a before the<br />
harvest. A few years after the clear-cut nitrate<br />
concentrations have significantly increased<br />
(Ahtiainen et al., 2003).<br />
For the model simulations presented in Section 4 a<br />
reference profile, which reflects the average<br />
topography <strong>and</strong> soil depths in the Kangasvaara<br />
catchment, was created. The reference profile,<br />
discretised into 20 columns, has a length of 500 m,<br />
a constant width, a slope of 6º, a depth of 1.5 m,<br />
<strong>and</strong> a 100% forest cover.<br />
884
4. RESULTS AND DISCUSSION<br />
4.1. Possibilities <strong>and</strong> compromises of a twodimensional<br />
catchment description<br />
The presented vertical two-dimensional<br />
representation of a catchment allows consideration<br />
of horizontally distributed catchment properties<br />
(e.g. topography <strong>and</strong> clear-cut locations) as a<br />
function of distance from the stream. Also, the<br />
travel distance of water <strong>and</strong> solutes to the stream<br />
<strong>and</strong> different moisture conditions along the<br />
flowpath can be taken into account in such a<br />
simplified catchment representation. This contrasts<br />
with models that have a lumped catchment<br />
representation.<br />
However, the simplification of the three<br />
dimensional catchment domain into two<br />
dimensions inevitably results in a loss of<br />
information, which may have consequences that<br />
are not obvious. When values of catchment<br />
properties depend only on the distance to a stream,<br />
no information is lost even though one dimension<br />
is omitted. In such an ideal setting the spatial<br />
distribution of catchment properties can be fully<br />
described in a single characteristic profile. In any<br />
real catchment such symmetry does not exist, <strong>and</strong><br />
information is lost through averaging variables that<br />
define the profile shape, assigning values for class<br />
variables along the profile, <strong>and</strong> changing the<br />
connectivity structure of catchment sub-areas<br />
having different properties.<br />
Two profiles with different properties, such as<br />
slope, can justifiably be aggregated together to<br />
form a single characteristic profile, when the<br />
model response is linear with respect to that<br />
property. However, when this linearity<br />
requirement is violated, the model output from the<br />
single profile differs from the sum of outputs from<br />
the two profiles. Class variables, such as<br />
vegetation type <strong>and</strong> soil type, can only have one<br />
value in the characteristic profile at a given<br />
distance from the stream, even though several<br />
classes may exist at the same distance. Assigning a<br />
single value for a class variable at any distance<br />
along the characteristic profile may distort the<br />
structure how areas with different properties drain<br />
through each other. The connectivity structure may<br />
also be distorted by aggregating short hillslopes<br />
with the near-stream parts of longer hillslopes,<br />
which in essence increases the downslope width of<br />
the single profile.<br />
In the following simulations response of the model<br />
output to changes in location of a clear-cut area<br />
(35% of the total area), slope of the profile, <strong>and</strong><br />
length of the profile is studied by varying one of<br />
those parameters at a time. Previous applications<br />
of the CPM have shown that the lateral hydraulic<br />
conductivity value of soil has an important role in<br />
determining the fraction of subsurface runoff<br />
(Koivusalo <strong>and</strong> Kokkonen, 2003). Therefore, the<br />
model response to changes in the above parameters<br />
is studied across conductivity values ranging from<br />
0.1 to 50 cm/h. Otherwise, the soil hydraulic<br />
parameters were adopted from Möttönen (2000).<br />
As the runoff generation mechanism <strong>and</strong> leaching<br />
of nitrate are likely to be related, the model<br />
response is assessed in terms of the fraction of<br />
subsurface runoff <strong>and</strong> the average annual nitrate<br />
load.<br />
4.2. Location of a clear-cut area<br />
Figure 1a shows how the subsurface runoff<br />
fraction changes when the distance between the<br />
centre of the clear-cut area <strong>and</strong> the stream<br />
decreases from 413 to 88 m. When the distance is<br />
at its minimum the cut area resides next to the<br />
stream. At low lateral conductivity values the<br />
location of the clear-cut had no effect on the<br />
subsurface runoff percentage, but when the<br />
conductivity value was increased a profile with a<br />
clear-cut area close to the water divide produced<br />
less subsurface runoff than a profile where a clearcut<br />
area was located next to the stream. This model<br />
result is explained by the earlier melt of snow in<br />
the cut than in the forested parts of the profile. In<br />
the model only subsurface flow is delayed along<br />
the profile, while surface runoff reaches the stream<br />
within one computation time-step independent of<br />
the distance. When the subsurface runoff<br />
component dominates the transport of water along<br />
the profile, a certain combination of hydraulic<br />
conductivity <strong>and</strong> distance to a stream can lead to a<br />
situation where melting waters from the clear-cut<br />
areas located further upslope on the profile reach<br />
the stream at the same time as snow melts in forest<br />
areas near the stream. This leads to an increased<br />
saturation in the near stream areas, which causes<br />
more surface runoff to be generated.<br />
Changes in nitrate export resulting from varying<br />
location of the clear-cut area are presented in<br />
Figure 1b. When the fraction of surface runoff is<br />
high, <strong>and</strong> the conductivity is low, nitrate loads<br />
increase almost linearly with the inverse of<br />
distance to the stream until the minimum value is<br />
reached. The nitrate load is much more sensitive to<br />
the cutting location at low than at high<br />
conductivities. At the highest tested conductivity<br />
value the nitrate load did not reach the minimum<br />
value when the clear-cut area is furthest away from<br />
the stream. This perhaps counterintuitive finding is<br />
explained by the behaviour of the subsurface<br />
runoff fraction as explained earlier.<br />
885
2a)<br />
a)<br />
subsurface flow. This can be seen as a decrease in<br />
nitrate export with the inverse of profile length.<br />
a)<br />
b)<br />
b)<br />
Figure 1. Fraction of subsurface runoff as a<br />
function of the mean distance between the stream<br />
<strong>and</strong> the clear-cut area (a) <strong>and</strong> the mean nitrate load<br />
as a function of the inverse of distance (b).<br />
4.3. Profile slope <strong>and</strong> profile length<br />
Figure 2a shows how the fraction of subsurface<br />
runoff is related to the slope of the profile. With<br />
increasing slope the share of subsurface runoff<br />
increases, particularly for soils with high<br />
conductivities <strong>and</strong> gentle slopes. This dependency<br />
is nearly linear until a subsurface runoff fraction of<br />
60-70% is reached. After this the fraction<br />
gradually approaches the maximum value of 100%<br />
as the profile becomes steeper. On the contrary to<br />
the fraction of subsurface runoff, the nitrate export<br />
decreased with the increase of the profile slope<br />
(Figure 2b). In fact, the fraction of subsurface<br />
runoff <strong>and</strong> the annual nitrate load showed a strong<br />
negative correlation (R = -0.99).<br />
Increasing the profile length led to a decrease in<br />
the fraction of subsurface runoff (Figure 3). This is<br />
a reflection of an increasing upslope area that<br />
drains into a stream through an equal width at the<br />
downslope end of the profile. The decrease of the<br />
subsurface flow fraction is almost linearly related<br />
to the inverse of the profile length until the fraction<br />
becomes sufficiently large (60-70 %). The nitrate<br />
load has again almost a perfect negative<br />
correlation (R = -0.99) with the fraction of<br />
Figure 2. Fraction of subsurface runoff (a) <strong>and</strong><br />
mean nitrate load (b) as a function of slope.<br />
Figure 3. Fraction of subsurface runoff as a<br />
function of profile length.<br />
The effect of width distribution of equally long<br />
profiles on the fraction of subsurface runoff<br />
depends on the runoff generation mechanism.<br />
When groundwater flow is the dominant<br />
mechanism producing subsurface runoff, the width<br />
distribution has a similar effect as the profile<br />
length, i.e. expansion of the upslope area by<br />
increasing the profile convergence decreases the<br />
fraction of subsurface runoff. However,<br />
886
exfiltration runoff, which is initiated when the<br />
profile convergence becomes sufficiently large,<br />
has a compensating effect on the amount of<br />
subsurface runoff. Increasing the profile<br />
convergence, as opposed to groundwater flow, led<br />
to a greater amount of exfiltration runoff.<br />
Therefore, after exfiltration runoff is activated, the<br />
fraction of subsurface flow is insensitive to a<br />
change in the profile convergence.<br />
Fraction of Subsurface Runoff [%]<br />
100<br />
90<br />
80<br />
70<br />
60<br />
50<br />
40<br />
30<br />
20<br />
10<br />
0<br />
Long<br />
Area-weighted sum of the short <strong>and</strong> long profiles<br />
Combined divergent profile<br />
Short (uniform width)<br />
Long (uniform width)<br />
Combined<br />
Short<br />
0.1 1 10 100<br />
Hydraulic Conductivity [cm/h]<br />
Figure 4. Fraction of subsurface runoff as a<br />
function of lateral hydraulic conductivity for four<br />
different cases. The short (250 m) <strong>and</strong> long (500<br />
m) profiles represent 1/3 <strong>and</strong> 2/3 of the catchment<br />
area, respectively.<br />
The effect of breaking the connectivity structure of<br />
different areas within a catchment was studied<br />
with the following simulation example. A<br />
catchment with one side of the stream having short<br />
hillslopes <strong>and</strong> the other side having long hillslopes<br />
was modelled first with two constant width<br />
profiles. Subsequently, the same catchment was<br />
represented with one divergent profile. According<br />
to the results (Figure 4), the two model set-ups<br />
gave nearly identical values for the fraction of<br />
subsurface runoff across the range of tested<br />
conductivities. Because the simulated nitrate load<br />
correlated strongly with the subsurface runoff<br />
fraction, there was also little difference in nitrate<br />
export values for the two cases.<br />
5. CONCLUSIONS<br />
Results indicated that the nitrate load showed a<br />
strong linear relationship with the fraction of<br />
subsurface runoff. The share of subsurface runoff,<br />
however, showed a non-linear behaviour with the<br />
profile length <strong>and</strong> slope.<br />
Simulation results depicting the influence of the<br />
profile slope on the runoff generation mechanism<br />
revealed that the slope angle had a nearly linear<br />
control on the subsurface runoff fraction when the<br />
fraction remained below 60-70%. The averaging of<br />
profiles having different slopes into a single profile<br />
can be justified under such conditions. But when<br />
profiles producing large fractions of subsurface<br />
flow are aggregated with profiles generating only<br />
little subsurface runoff, the assumption of linearity<br />
is violated <strong>and</strong> the averaging procedure leads to an<br />
overestimation of the subsurface fraction.<br />
Aggregating profiles with different lengths distorts<br />
the connectivity structure of the catchment as the<br />
short profiles are combined with the near-stream<br />
parts of longer hillslopes. With uniform soils <strong>and</strong><br />
l<strong>and</strong>-use (forest), however, it did not matter<br />
whether the catchment was described with two<br />
parallel profiles of different length <strong>and</strong> uniform<br />
width or with a single divergent profile having the<br />
same length as the longer of the two profiles.<br />
Breaking the structure how catchment sub-areas<br />
connect with each other is likely to have a more<br />
pronounced effect when some areas within the<br />
catchment have significantly different<br />
characteristics. This is the case, for example, when<br />
the catchment has been subject to clear-cutting or<br />
has soils with varying hydraulic characteristics <strong>and</strong><br />
vegetation cover.<br />
Despite the inevitable simplifications that arise<br />
when representing a catchment as a single<br />
hillslope, the two-dimensional framework for<br />
assessing environmental impacts of forest<br />
management practises offers advantages over a<br />
lumped, one-dimensional catchment description.<br />
The model structure based on a characteristic<br />
profile can accommodate the effects of slope angle<br />
<strong>and</strong> length of the profile, <strong>and</strong> also the effects of<br />
location of a cut area. Simulation of the effects of<br />
clear-cutting indicated that at low soil<br />
conductivities, nitrate loads increased linearly with<br />
the inverse of distance between the cut area <strong>and</strong><br />
the stream. But when the conductivity value<br />
became sufficiently large, effects arising from<br />
differences in the timing of snowmelt between<br />
open <strong>and</strong> forested environments obscured this<br />
relationship.<br />
Although not all assumptions incorporated in the<br />
current version of the model may be accurate, the<br />
two-dimensional catchment description provides a<br />
framework where hypotheses about clear-cutting<br />
effects can be evaluated <strong>and</strong> tested. In a complex<br />
forest ecosystem, where the controls are not<br />
necessarily straightforward <strong>and</strong> intuitive, a<br />
simulation model can be a valuable tool in<br />
analysing <strong>and</strong> interpreting field measurements.<br />
6. ACKNOWLEDGEMENTS<br />
This work was funded by the Academy of Finl<strong>and</strong><br />
project FEMMA (No 52740), which develops tools<br />
for assessing environmental impacts of forest<br />
management practises. On-site hydrometeorological<br />
data were collected in the VALU<br />
project. Additional meteorological data were<br />
887
kindly provided by the Finnish Meteorological<br />
Institute. Support from prof. Pertti Vakkilainen,<br />
prof. Tuomo Karvonen, prof. Hannu Mannerkoski,<br />
Dr. Keijo Nenonen, <strong>and</strong> Dr. Pekka Hänninen is<br />
greatly acknowledged.<br />
7. REFERENCES<br />
Ahtiainen, M., L. Finér, M. Haapanen, K.<br />
Kenttämies, T. Mattsson, <strong>and</strong> A. Rämö,<br />
Näkyvätkö hakkuun ja maanmuokkauksen<br />
vaikutukset valumaveden laadussa –<br />
tehoavatko ympäristönsuojeluohjeet? (in<br />
Finnish) In: L. Finér, A. Laurén, <strong>and</strong> L.<br />
Karvinen (eds.), Proceedings of the Koli<br />
Seminar, Research Papers 886, Finnish<br />
Forest Research Institute, 25-33, Joensuu,<br />
2003.<br />
Chertov, O.G., A.S. Komarov, M.<br />
Nadporozhskaya, S.S. Bykhovets, <strong>and</strong> S.L.<br />
Zudin, ROMUL – a model of forest soil<br />
organic matter dynamics as a substantial<br />
tool for forest ecosystem modelling,<br />
Ecological <strong>Modelling</strong>, 138, 289-308, 2001.<br />
Finér, L., M. Ahtiainen, H. Mannerkoski, V.<br />
Möttönen, S. Piirainen, P. Seuna, <strong>and</strong> M.<br />
Starr, Effects of harvesting <strong>and</strong> scarification<br />
on water <strong>and</strong> nutrient fluxes, Research<br />
Papers 648, Finnish Forest Research<br />
Institute, 38 p., Joensuu, 1997.<br />
Finnish Forest Research Institute, Statistical<br />
Yearbook of Forestry 2002, Finnish Forest<br />
Research Institute, 2002.<br />
Jansson, P.-E. <strong>and</strong> L. Karlberg, Coupled heat <strong>and</strong><br />
mass transfer model for soil-plantatmosphere<br />
systems, Royal Institute of<br />
Technology, Dept. of Civil <strong>and</strong><br />
<strong>Environmental</strong> Engineering, 321 pp.,<br />
Stockholm, 2001.<br />
Karvonen, T., H. Koivusalo, M. Jauhiainen, J.<br />
Palko, K. Weppling, A hydrological model<br />
for predicting runoff from different l<strong>and</strong> use<br />
areas, Journal of Hydrology, 217, 253-265,<br />
1999.<br />
Kellomäki, S., <strong>and</strong> H. Väisänen, <strong>Modelling</strong> the<br />
dynamics of the forest ecosystem for<br />
climate change studies in the boreal<br />
conditions, Ecological <strong>Modelling</strong>, 97, 121-<br />
140, 1997.<br />
Kenttämies, K., Tilanne ja tavoitteet metsätalouden<br />
vesistökurmituksen vähentämiseksi (in<br />
Finnish), In: L. Finér, A. Laurén, <strong>and</strong> L.<br />
Karvinen (eds.), Research Papers 886,<br />
Finnish Forest Research Institute, 2003.<br />
Koivusalo, H., <strong>and</strong> T. Kokkonen, Snow processes<br />
in a forest clearing <strong>and</strong> in a coniferous<br />
forest, Journal of Hydrology, 262, 145-164,<br />
2002.<br />
Koivusalo, H., <strong>and</strong> T. Kokkonen, Snow processes<br />
in a forest clearing <strong>and</strong> in a coniferous<br />
forest, Journal of Hydrology, 262, 145-164,<br />
2002.<br />
Koivusalo, H., <strong>and</strong> T. Kokkonen, <strong>Modelling</strong> runoff<br />
generation in a forested catchment in<br />
southern Finl<strong>and</strong>, Hydrological Processes,<br />
17, 313-328, 2003.<br />
Koivusalo, H., M. Heikinheimo, <strong>and</strong> T. Karvonen,<br />
Test of a simple two-layer parameterisation<br />
to simulate the energy balance <strong>and</strong><br />
temperature of a snowpack, Theoretical <strong>and</strong><br />
Applied Climatology, 70, 65-79, 2001.<br />
Koivusalo, H., T. Kokkonen, A. Laurén, M.<br />
Ahtiainen, T. Karvonen, H. Mannerkoski,<br />
S. Penttinen, P. Seuna, M. Starr, P.<br />
Vakkilainen, <strong>and</strong> L. Finér, Parameterisation<br />
<strong>and</strong> application of a hillslope model to<br />
assess hydrological impacts of forest<br />
harvesting, In: D. Post (ed.), MODSIM<br />
2003. The <strong>Modelling</strong> <strong>and</strong> Simulation<br />
Society of Australia <strong>and</strong> New Zeal<strong>and</strong>,<br />
Townsville, Australia, 855, 2003.<br />
Kokkonen, T., H. Koivusalo, T. Karvonen, A<br />
semi-distributed approach to rainfall-runoff<br />
modelling - a case study in a snow affected<br />
catchment, <strong>Environmental</strong> <strong>Modelling</strong> &<br />
<strong>Software</strong>, 16, 481-493, 2001.<br />
Ministry of Agriculture <strong>and</strong> Forestry, Finl<strong>and</strong>’s<br />
National Forest Programme 2010,<br />
Publications 2/1999, Ministry of<br />
Agriculture <strong>and</strong> Forestry, 1999.<br />
Möttönen, V., Moreenin vesitaloustunnusten<br />
vaihtelu tuoreessa kangasmetsässä ennen<br />
hakkuuta ja hakkuun jälkeen (in Finnish),<br />
Lic. For. Thesis, University of Joensuu,<br />
Faculty of Forestry, 56 p., Joensuu, 2000.<br />
Piirainen, S., L. Finér, <strong>and</strong> M. Starr, Canopy <strong>and</strong><br />
soil retention of nitrogen deposition in a<br />
mixed boreal forest in eastern Finl<strong>and</strong>,<br />
Water, Air, <strong>and</strong> Soil Pollution, 105, 165-<br />
174, 1998.<br />
Skaggs, R.W., A water management model for<br />
artificially drained soils, North Carolina<br />
Agricultural Research Service, 54 p.,<br />
Raleigh, 1980.<br />
888
Generic process-based plant models for the analysis of<br />
l<strong>and</strong>scape change<br />
B. Reineking ab , A. Huth a <strong>and</strong> C. Wissel a<br />
a Department of Ecological <strong>Modelling</strong>, UFZ Centre for Ecological Research, P.O. Box 500135 Leipzig,<br />
Germany; bjoern.reineking@ufz.de<br />
b Natural <strong>and</strong> Social Science Interface (ETH-UNS), Swiss Federal Institute of Technology, Zurich, Switzerl<strong>and</strong><br />
Abstract: The analysis of l<strong>and</strong>scape change impacts on community composition <strong>and</strong> dynamics is difficult for<br />
species rich plant communities, because of their high complexity. One approach to deal with this challenge<br />
are generic process-based models. In these models, the species are described by a common set of parameters<br />
<strong>and</strong> functional responses. Thus, they allow both the integration of knowledge on key processes, <strong>and</strong> a common<br />
description for several ecological patterns. An important aspect of these models are trade-offs in the species’<br />
physiological <strong>and</strong> life-history traits, which prevent ‘super-species’ that dominate under all environmental conditions.<br />
We compare process-based models with two other model types that have been applied to similar ends – statistical<br />
habitat models, <strong>and</strong> phenomenological population models. These process-based models come at the price<br />
of an increased number of parameters for an individual species. However, a description of the interactions<br />
between species, which has proven difficult to incorporate in statistical habitat models, or requiring excessively<br />
many parameters in phenomenological population models, can be included easily. Finally, processed-based<br />
models produce a rich set of patterns on several organizational levels that can be compared to empirical observations,<br />
<strong>and</strong> thus be used for model calibration <strong>and</strong> validation.<br />
The approach is illustrated with a case study of Southern African plant communities. The investigated semi-arid<br />
l<strong>and</strong>scapes are characterized by high stochastic fluctuations in population sizes. These fluctuations may in the<br />
short term mask the effects of environmental or l<strong>and</strong> use change, <strong>and</strong> models allow to assess likely long-term<br />
consequences. Questions pertinent to the management of these l<strong>and</strong>scapes include the effect of grazing on the<br />
diversity of the plant communities <strong>and</strong> the impact of climate change.<br />
Keywords: L<strong>and</strong>scape change; modelling; generic; process-based models; species richness<br />
1 INTRODUCTION<br />
The analysis of the effects of l<strong>and</strong>scape change<br />
on species rich plant communities has received increased<br />
interests over the last years. Drivers of l<strong>and</strong>scape<br />
change are climate change or changes in management<br />
practices.<br />
In pasture l<strong>and</strong>scapes, for example, some traditional<br />
management systems have become economically<br />
unsustainable [Kleyer et al., 2002]. Consequently,<br />
management alternatives are being sought that are<br />
both economically feasible <strong>and</strong> acceptable in their<br />
effect on the plant <strong>and</strong> animal communities (see<br />
for example, the MOSAIK project, [Kleyer et al.,<br />
2002]).<br />
Common challenges to the assessment of the effect<br />
of l<strong>and</strong>scape change are long time scales <strong>and</strong> transient<br />
dynamics, the need to assess a multitude of<br />
management options, <strong>and</strong> high species diversity.<br />
In the following, we will first compare three model<br />
approaches for the assessment of l<strong>and</strong>scape change<br />
on plant communities. We then exemplify the approach<br />
of generic, process-based models in more<br />
detail for a Southern African succulent plant community,<br />
where the impact of spatial <strong>and</strong> temporal<br />
variation in water availability as well as that of different<br />
grazing regimes on species richness is investigated.<br />
In the last two sections we discuss the opportunities<br />
<strong>and</strong> challenges of the process-based modelling<br />
approach, <strong>and</strong> draw conclusions.<br />
889
2 MODEL APPROACHES<br />
In this section we present three approaches to assess<br />
the likely impact of l<strong>and</strong>scape change, in the order<br />
of increased structural complexity: statistical habitat<br />
models, phenomenological population models,<br />
<strong>and</strong> generic, process-based (‘mechanistic’) models.<br />
2.1 Statistical habitat models<br />
Statistical habitat models quantify the habitat requirements<br />
of species based on presence/absence<br />
records or density estimates of the species, <strong>and</strong> information<br />
on the environmental conditions at the investigated<br />
sites. Frequently used statistical methods<br />
are generalized linear models, generalized additive<br />
models or classification trees [Austin, 2002].<br />
When l<strong>and</strong>scape change can be related to changes in<br />
the environmental variables used in the construction<br />
of the habitat models, changes in the spatial distribution<br />
of suitable habitat as a consequence of l<strong>and</strong>scape<br />
change can be predicted. Usually, separate<br />
models are developed for different species, <strong>and</strong> the<br />
community response is assessed as the sum of the<br />
individual species’ responses.<br />
Advantages. An important advantage of statistical<br />
habitat models is that, given available empirical<br />
data, they are quickly to develop, <strong>and</strong> that there<br />
are tools to quantify uncertainty in the predictions<br />
(though they are based on certain model assumptions<br />
that need not be fulfilled).<br />
Where the interaction of species plays a key role in<br />
determining the presence or relative abundance of<br />
the species, the predictions from models neglecting<br />
competitive effects may be misleading.<br />
Finally, it would often be useful to have a prospective<br />
assessment of management alternatives. However,<br />
extrapolating from correlational models to new<br />
situations is problematic.<br />
2.2 Phenomenological population models<br />
Dynamic population models address the issue of<br />
transient dynamics initiated by l<strong>and</strong>scape change.<br />
This is true for phenomenological as well as<br />
process-based models. Phenomenological models<br />
here refers to those models that do not attempt to<br />
incorporate the mechanism underlying the observed<br />
phenomena, but focus on capturing key aspects of<br />
the observed dynamics. The value of model parameters<br />
are usually assigned by fitting the model to observed<br />
data. Matrix models or models of the Lotka-<br />
Volterra type belong to this class.<br />
Advantages. There exist a lot of experience with<br />
phenomenological models, <strong>and</strong> they tend to be<br />
structurally fairly simple. Therefore, they can<br />
be implemented <strong>and</strong> analyzed reasonably quickly.<br />
These models do not need many parameters for an<br />
individual species.<br />
Problems. The models do not explicitly incorporate<br />
the dynamics of the system <strong>and</strong> assume usually<br />
that the observed patterns of species occurrences reflect<br />
an (quasi-)equilibrium state, given the values<br />
of the explanatory variables. Temporal dynamics<br />
can only be captured in a phenomenological way by<br />
explicitly incorporating a time variable such as time<br />
since last disturbance.<br />
It is usually difficult to develop models for large<br />
sets of species, because many species are rare, such<br />
that there are few presence records to construct<br />
the statistical models from. As a rule of thumb,<br />
there should be a minimum of ten occurrences per<br />
explanatory variable used in a logistic regression<br />
model. Otherwise, there will be high uncertainty<br />
in model parameters <strong>and</strong> predictions.<br />
In addition, models assume that the occurrences of<br />
different species do not interfere with each other.<br />
Problems. Parameterization of the models poses a<br />
key problem. Estimation of competition parameters<br />
is difficult. In addition, the number of required competition<br />
parameters quickly grows as the number of<br />
modelled species increases. If only pairwise interactions<br />
are included, the number increases quadratically<br />
in the number of species. Usually, it is not<br />
feasible to collect data for many species, so a few<br />
species representing different functional groups are<br />
selected. Also, it is necessary to have information<br />
on changes of the parameter values under the different<br />
l<strong>and</strong>scape change scenarios. One option is<br />
to model how the values of the parameters change<br />
with altered l<strong>and</strong> use, i.e. to develop a model of the<br />
relationships of species model parameters with the<br />
l<strong>and</strong> use characteristics. One example of such an<br />
approach, where the parameters in a matrix population<br />
model of a soil mite species are related to different<br />
levels of temperature <strong>and</strong> soil contamination,<br />
is given by Stamou et al. [2004].<br />
890
2.3 Process-based models<br />
In process-based models, species are described<br />
by morphological, physiological <strong>and</strong>/or life-history<br />
traits, <strong>and</strong> the model explicitly describes how resource<br />
uptake (e.g. water, nutrients, light) translates<br />
into population growth. An example of this<br />
approach is Tilman’s ALLOCATE model of grassl<strong>and</strong><br />
plant communities, where plants compete for<br />
nutrients <strong>and</strong> light, <strong>and</strong> depending on their allocation<br />
strategy for photosynthates (roots, stem, leaves)<br />
face trade-offs that lead to different relative competitiveness<br />
under varying resource levels [Tilman,<br />
1988].<br />
Advantages. Process-based models in general<br />
need more parameters to characterize a single<br />
species than phenomenological models. However,<br />
because the interactions between species are the<br />
outcome of the modelled processes (water <strong>and</strong> nutrient<br />
uptake, light interception), additional parameters<br />
that describe the interactions are not necessary.<br />
The number of model parameters therefore<br />
increases only linearly with the number of model<br />
species.<br />
Physiological <strong>and</strong> life-history traits determine how<br />
plant species respond to l<strong>and</strong>scape change. By modelling<br />
the link between species traits <strong>and</strong> population<br />
dynamics, process-based models allow to investigate<br />
the effect of l<strong>and</strong>scape change on a range<br />
of species for which the relevant traits are known.<br />
This way, they tie in the database projects on species<br />
traits with the underst<strong>and</strong>ing of l<strong>and</strong>scape change<br />
effects. <strong>Modelling</strong> of the processes helps to identify<br />
key parameters that have to be estimated. In addition,<br />
this can help to identify traits that can easily be<br />
measured (with low time <strong>and</strong> money investment),<br />
or that can be reliably related to the traits that are<br />
directly relevant on the process level, but that are<br />
difficult to observe or quantify. This approach has<br />
been successfully applied in modelling the dispersal<br />
of plant species. Based on mechanistic models, a<br />
minimum set of plant <strong>and</strong> seed characteristics could<br />
be established, that together with information on the<br />
wind distribution allow to predict the distribution of<br />
primary dispersal distances [Tackenberg, 2003].<br />
Process-based models produce patterns at several<br />
hierarchical levels. This can be used in model parameterization<br />
<strong>and</strong> validation [Grimm et al., 1996].<br />
Problems. Process-based models aim at a controlled<br />
increase in complexity, i.e. to strike a balance<br />
between generality <strong>and</strong> specificity. However, the increased<br />
complexity in the model structure comes at<br />
a cost.<br />
Often, these models put a high dem<strong>and</strong> on computing<br />
resources. Therefore, it may not be possible to<br />
explicitly model large stretches of the l<strong>and</strong>scape, but<br />
rather only smaller patches. In order to scale to the<br />
whole l<strong>and</strong>scape, model simplifications have to be<br />
carried out. Yet, such aggregated descriptions also<br />
increase clarity <strong>and</strong> underst<strong>and</strong>ing of the model behavior.<br />
Although many species traits that are represented in<br />
the model can be measured in principle in the field,<br />
they may not be available for the majority of the<br />
species. In addition, complex model structure allows<br />
for a rich set of dynamics, leading to substantial<br />
uncertainty in model predictions.<br />
Finally, process-based models pose a greater challenge<br />
to the software development than the other<br />
approaches discussed. Dissemination <strong>and</strong> reuse<br />
of models or model components between research<br />
projects is difficult.<br />
In the following section we present an extended example<br />
of a process-based plant model.<br />
3 EXAMPLE: MODELLING A SEMI-<br />
ARID SUCCULENT PLANT COMMU-<br />
NITY<br />
The arid winter-rainfall region of the western<br />
Richtersveld (South Africa) harbors an unusually<br />
high plant species richness, with species densities<br />
approaching 40 perennials per 100 m 2 [Jurgens<br />
et al., 1999]. Although a wealth of processes have<br />
been invoked to explain biodiversity in plant communities,<br />
the relative importance of different factors<br />
remains poorly understood.<br />
3.1 Model description<br />
The model calculates plant water uptake <strong>and</strong> transpiration,<br />
carbon assimilation <strong>and</strong> respiration on a<br />
daily basis. The water <strong>and</strong> carbon cycles are coupled<br />
via the plant’s water use efficiency. Immature<br />
plants allocate carbon to the compartments roots,<br />
succulent tissue, <strong>and</strong> leaves. Once plants have<br />
reached their size at maturity, all net carbon gain<br />
is invested in seeds. In times of drought, plants<br />
rely on water stored in succulent tissue for transpiration.<br />
If the carbon balance is negative, the plant<br />
suffers from increased mortality. At the level of the<br />
population, the key processes are germination, sur-<br />
891
vival of individuals <strong>and</strong> seeds in the seed bank, <strong>and</strong><br />
seed production. They are calculated on an annual<br />
basis. The germination rate <strong>and</strong> the seed survival<br />
rate are constant in the model. However, plant survival<br />
<strong>and</strong> seed production depend on environmental<br />
conditions, in particular rainfall <strong>and</strong> potential evapotranspiration.<br />
The strategy types, i.e. ‘species’, differ only in the<br />
values of five parameters that define (a) biomass<br />
allocation to roots, leaves, <strong>and</strong> storage (effectively<br />
two parameters, as the sum of the fractions has to<br />
sum to 1), (b) size at maturity, <strong>and</strong> (c) germination<br />
rate <strong>and</strong> date. Allocation to the three compartments<br />
roots, leaves, <strong>and</strong> storage is assumed to be independent<br />
of total plant biomass. The key environmental<br />
state variable is soil water content. Soil water<br />
content increases through rainfall, <strong>and</strong> decreases<br />
through plant water uptake, drainage <strong>and</strong> evaporation.<br />
The soil is characterized by soil depth, saturation<br />
water content, <strong>and</strong> water content at permanent<br />
wilting point. An overview of the main model processes<br />
is shown in Figure 1.<br />
The model takes as environmental input sequences<br />
of daily rainfall <strong>and</strong> potential evapotranspiration.<br />
With few exceptions, parameters in the plant model<br />
can be measured in the field. Parameter values were<br />
based on the literature <strong>and</strong> represent typical values<br />
for plants of semi-arid regions, or values chosen in<br />
similar process-based models of plant communities.<br />
The values of three parameters relating to drought<br />
mortality <strong>and</strong> water storage were selected such that<br />
some viable plant strategy types were possible under<br />
the most arid scenarios investigated.<br />
The process-based model of plant growth <strong>and</strong><br />
survival is combined with a spatially explicit<br />
individual-based population model. The simulated<br />
area is subdivided into square sites, with a side<br />
length of 25 cm. The maximum number of plants<br />
per cell is six, however, the roots <strong>and</strong> leaves of<br />
plants can extend over several cells.<br />
The population processes (seed production, dispersal,<br />
germination) operate on annual time steps.<br />
Seeds are dispersed according to a log-normal dispersal<br />
kernel. Competition during germination is<br />
modelled as lottery competition.<br />
Plant growth <strong>and</strong> survival are calculated in daily<br />
time steps. Established plants compete for water in<br />
areas where roots overlap. Shading effects are not<br />
taken into account. The state of individual plants is<br />
given by their age (i.e. cohort assignment), mass,<br />
amount of water in succulent tissue, <strong>and</strong> the time<br />
Figure 1: Overview of the key model processes.<br />
Light grey arrows indicate water fluxes, black arrows<br />
represent carbon fluxes.<br />
period over which the growth rate has been negative.<br />
3.2 Simulated environmental change scenarios<br />
An area of 10 times 10 m, corresponding to the dimensions<br />
of long-term observation sites established<br />
in the Richtersveld, was simulated. Starting from a<br />
situation with no established plants <strong>and</strong> a seed bank<br />
containing equal seed densities of all model species,<br />
the population dynamics were simulated for a period<br />
of 200 years. Within this time frame, the community<br />
dynamics reached an equilibrium state. The<br />
model species pool consisted of 36 species, comprised<br />
of 12 different allocation strategies <strong>and</strong> 3<br />
sizes at maturity.<br />
Rainfall. With respect to water availability, we<br />
present results on the relevance of the following two<br />
factors: (a) Spatial heterogeneity of water availability<br />
through redistribution of precipitation, evaluated<br />
at three levels (no redistribution, moderate redistribution,<br />
strong redistribution). The total amount of<br />
precipitation was held constant (see Figure 2).<br />
(b) Temporal heterogeneity of water availability<br />
through fluctuations in precipitation, evaluated at<br />
three levels (a st<strong>and</strong>ard scenario that corresponded<br />
to the model parameters of the Interactive South<br />
Africa Rain Atlas for the study region, an increased<br />
level of seasonality as well as a reduced level of seasonality).<br />
The mean annual rainfall was held constant<br />
at 70 mm by adjusting the mean daily probability<br />
of rainfall. In all scenarios, dew fall was simulated<br />
as a precipitation event with low magnitude<br />
(0.2 mm) <strong>and</strong> constant probability (see Figure 3).<br />
892
Vertical extent (m)<br />
0 2 4 6 8 10<br />
Expected daily precipitation (mm)<br />
0.0 0.1 0.2 0.3 0.4 0.5 0.6<br />
Reduced<br />
St<strong>and</strong>ard<br />
Enhanced<br />
Dew<br />
0 2 4 6 8 10<br />
Horizontal extent (m)<br />
Figure 2: Water redistribution map for the most heterogeneous<br />
scenario. Darker colors correspond to<br />
increased water availability. The total amount of<br />
available water is identical in all redistribution scenarios.<br />
Grazing. The expected proportion of a given<br />
plant to be grazed was held constant across species,<br />
i.e. no preferences of livestock for certain species<br />
were modelled. Grazing was applied spatially homogeneously<br />
in the model. The grazing intensity,<br />
i.e. the total amount of biomass removed annually,<br />
was constant over time, <strong>and</strong> two levels of intensity<br />
were simulated. A second aspect of the investigated<br />
grazing regimes are the frequencies, i.e. the number<br />
of times grazing occurred during a year. Three<br />
levels of grazing frequencies were simulated.<br />
3.3 Results<br />
Rainfall. The influence of rainfall variability <strong>and</strong><br />
spatial water redistribution on species diversity is<br />
shown in Figure 4. Shannon diversity H was calculated<br />
as H = − ∑ N<br />
i=1 p i ln p i , where p i is the<br />
proportional abundance of species i, <strong>and</strong> the sum is<br />
over all N species. It is evident that water redistribution<br />
exhibits a strong positive effect on community<br />
diversity at the studied scale. The maximum<br />
diversity at a given level of temporal variability in<br />
rainfall is reached at the maximum level of heterogeneity.<br />
There is a positive effect of redistribution<br />
on diversity. The effect of spatial heterogeneity is to<br />
provide spots with increased water supply <strong>and</strong> thus<br />
improved growing conditions, allowing for a larger<br />
set of species to coexist. Temporal heterogeneity, on<br />
0 100 200 300<br />
Day of year<br />
Figure 3: Expected daily precipitation for three precipitation<br />
scenarios for the Richtersveld site, including<br />
dew.<br />
the other h<strong>and</strong>, does not have a positive influence on<br />
diversity in the studied form. The more aggregated<br />
the rainfall is in time, the longer the periods of unfavorable<br />
growing conditions. Since the plants in the<br />
system are ‘living on the edge’, the increased variability<br />
appears to increase the overall extinction risk<br />
to an extent that it outweighs the potentially positive<br />
effect of temporal niche differentiation.<br />
Grazing. Grazing reduced the number of surviving<br />
species. Only the dominant species in the scenario<br />
without grazing were viable under grazing<br />
pressure. As expected, grazing intensity was overall<br />
more important than grazing frequency. However,<br />
the effect of grazing frequency changed under<br />
low <strong>and</strong> high grazing pressure. While a higher<br />
frequency had a marginally positive influence under<br />
low grazing pressure (Figure 5), it exerted a<br />
strong negative effect under high grazing pressure,<br />
where species only survived if grazing occurred infrequently.<br />
4 CONCLUSIONS<br />
We argue that generic, process-based plant simulation<br />
models, though no panacea, can be expected<br />
to become a key tool in the assessment of l<strong>and</strong>scape<br />
change. These models are able to meet the<br />
challenges posed by the assessment of future l<strong>and</strong>scape<br />
change – long time scales <strong>and</strong> transient dynamics,<br />
the need to assess a multitude of manage-<br />
893
Shannon diversity index<br />
2.2 2.4 2.6 2.8<br />
none<br />
medium<br />
high<br />
●<br />
Water redistribution<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
Shannon Diversity Index<br />
0.0 0.5 1.0 1.5<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
●<br />
low<br />
high<br />
●<br />
reduced normal enhanced<br />
Rainfall variability<br />
Figure 4: Notched boxplots showing the relative effect<br />
of rainfall variability <strong>and</strong> water redistribution<br />
on community diversity. Each boxplot represents 5<br />
replications.<br />
ment options, <strong>and</strong> high species diversity. They can<br />
be geared to specific environmental situations, thus<br />
allowing model results to be directly compared to<br />
specific patterns observed in the field. Additionally,<br />
key findings are likely to generalize to other ecosystems<br />
of similar environmental conditions, because<br />
of the models’ generic structure.<br />
ACKNOWLEDGMENTS<br />
We are grateful to Tamara Münkemüller, Frank<br />
Schurr <strong>and</strong> two anonymous reviewers for their valuable<br />
comments on the manuscript. B. R. has been<br />
financially supported by the German Federal Ministry<br />
of Education <strong>and</strong> Research (BIOTA project,<br />
grant 01-LC-0024).<br />
REFERENCES<br />
Austin, M. P. Spatial prediction of species distribution:<br />
an interface between ecological theory <strong>and</strong><br />
statistical modelling. Ecological <strong>Modelling</strong>, 157:<br />
101–118, 2002.<br />
0 10 20 30 40 50<br />
Grazing frequency (events per year)<br />
Figure 5: Effects of grazing frequency on community<br />
diversity at two levels of grazing intensity.<br />
succulent mesembryanthemaceae shrubs in the<br />
winter-rainfall desert of northwestern namaqual<strong>and</strong>,<br />
south africa. Plant Ecology, 142:87–96,<br />
1999.<br />
Kleyer, M., R. Biedermann, K. Henle, H. J. Poethke,<br />
P. Poschlod, <strong>and</strong> J. Settele. MOSAIK: Semi-Open<br />
Pasture <strong>and</strong> Ley - a Research Project on Keeping<br />
the Cultural L<strong>and</strong>scape Open, pages 399–412.<br />
Springer, Heidelberg, 2002.<br />
Stamou, G. P., G. V. Stamou, E. M. Papatheodorou,<br />
M. D. Argyropoulou, <strong>and</strong> S. G. Tzafestas. Population<br />
dynamics <strong>and</strong> life history tactics of<br />
arthropods from mediterranean-type ecosystems.<br />
Oikos, 104:98–108, 2004.<br />
Tackenberg, O. Modeling long-distance dispersal<br />
of plant diaspores by wind. Ecol. Monogr., 73:<br />
173–189, 2003.<br />
Tilman, D. Plant Strategies <strong>and</strong> the Dynamics <strong>and</strong><br />
Structure of Plant Communities. Princeton University<br />
Press, Princeton, New Jersey, 1988.<br />
Grimm, V., K. Frank, F. Jeltsch, R. Br<strong>and</strong>l, J. Uchmanski,<br />
<strong>and</strong> C. Wissel. Pattern-oriented modelling<br />
in population ecology. Sci. Total Environ.,<br />
183:151–166, 1996.<br />
Jurgens, N., I. H. Gotzmann, <strong>and</strong> R. M. Cowling.<br />
Remarkable medium-term dynamics of leaf<br />
894
The role of local spatial heterogeneity in the maintenance<br />
of parapatric boundaries: agent based models of<br />
competition between two parasitic ticks<br />
A. J. Tyre a , B. Tenhumberg a , <strong>and</strong> C. M. Bull b<br />
a<br />
School of Natural Resources, University of Nebraska-Lincoln, Lincoln, Nebraska, USA;<br />
atyre2@unl.edu<br />
b<br />
School of Biological Science, Flinders University of South Australia, Adelaide, South Australia,<br />
Australia<br />
Abstract: Recent models of ecological parapatry, where the geographical distributions of two similar<br />
species abut without overlapping, have shown that spatial gradients in intrinsic growth rates can lead to sharp<br />
boundaries when dispersal is density dependent. However, a well documented parapatric boundary in<br />
southern Australia between two tick species that parasitise a large lizard lacks one or both of these features;<br />
dispersal of ticks is r<strong>and</strong>om <strong>and</strong> there may not be a gradient of population growth rates for one of the<br />
species. There is local variation in population growth rates arising from variation in the number of host<br />
lizards with overlapping host ranges. When more hosts are available there is a shorter waiting time for a<br />
host to arrive, <strong>and</strong> consequently higher survival rates. We construct a spatially explicit agent based model of<br />
the interaction between the two ticks <strong>and</strong> their lizard host <strong>and</strong> explore the role that this fine scale spatial<br />
heterogeneity plays in maintaining the parapatric boundary between the two tick species geographic<br />
distributions.<br />
Keywords: Ecological parapatry; Tiliqua rugosa; Aponomma hydrosauri; Amblyomma limbatum<br />
1. INTRODUCTION<br />
Parapatric boundaries occur where the<br />
biogeographic distribution of two species abut but<br />
do not overlap [Bull 1991]. When there is no<br />
hybridization between the two species, the<br />
situation is described as ecological parapatry. A<br />
number of processes have been proposed to<br />
explain the maintenance of ecological parapatry<br />
including interspecific competition [MacArthur<br />
1972], reproductive interference [Ribeiro <strong>and</strong><br />
Speilman 1986], <strong>and</strong> habitat patchiness [Bull <strong>and</strong><br />
Possingham 1995].<br />
A recent 1-dimensional diffusion model found<br />
that density dependent dispersal could sharpen a<br />
boundary by narrowing the overlap zone [García-<br />
Ramos et al. 2000]; density independent dispersal<br />
lead to complete overlap. However, the biology of<br />
a well documented boundary between two species<br />
of reptile tick in Australia [Bull <strong>and</strong> Burzacott<br />
2001] seems unlikely to have density dependent<br />
dispersal by the two participants. Ticks only move<br />
when attached to hosts, <strong>and</strong> hosts reduce<br />
movement in response to tick infestation [Main<br />
<strong>and</strong> Bull 2000]. If anything, this would lead to<br />
inverse density dependent dispersal by ticks. Host<br />
abundance also varies across the boundary [Bull<br />
1995], <strong>and</strong> when more hosts are available there is<br />
a shorter waiting time for a host to arrive, <strong>and</strong><br />
consequently higher survival rates. This variation<br />
in tick survival could create habitat patchiness<br />
capable of maintaining the boundary [Bull <strong>and</strong><br />
Possingham 1995]. We used an agent based model<br />
of the system to examine the effect of varying<br />
host abundance <strong>and</strong> dispersal rates on the<br />
maintenance of the parapatric boundary.<br />
2. THE PARAPATRIC BOUNDARY<br />
The study area is mixed chenopod shrubl<strong>and</strong> <strong>and</strong><br />
mallee woodl<strong>and</strong> near Mt. Mary in the mid-north<br />
of South Australia. The region has an annual<br />
rainfall of about 250 mm. Aponomma hydrosauri<br />
895
<strong>and</strong> Amblyomma limbatum are ectoparasites of<br />
large reptiles in southern Australia. The<br />
predominant host is the sleepy lizard, Tiliqua<br />
rugosa. The study area straddles a parapatric<br />
boundary between the distribution of both species;<br />
north of the boundary there are generally no A.<br />
hydrosauri ticks except for occasional outbreaks<br />
Adult males take in small meals, <strong>and</strong> wait for long<br />
periods on the host where they mate with attaching<br />
females. Tick activity <strong>and</strong> development is<br />
confined to the spring <strong>and</strong> summer months when<br />
temperatures are warm <strong>and</strong> lizards are active [Bull<br />
<strong>and</strong> Burzacott 2001].<br />
A<br />
0.8<br />
2. AN AGENT BASED MODEL OF TICK<br />
POPULATION DYNAMICS<br />
Incidence, A. hydrosauri<br />
0.6<br />
0.4<br />
B<br />
Incidence<br />
0.2<br />
0.0<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0.0<br />
1995<br />
Year<br />
-10<br />
1990<br />
1985<br />
0<br />
Position [km]<br />
-10<br />
10<br />
0<br />
Position [km]<br />
1985<br />
10<br />
1990<br />
Year<br />
1995<br />
Figure 1. Space-time perspective plots of the<br />
incidence of Ap. hydrosauri (A) <strong>and</strong> Amb.<br />
limbatum (B) along Transect 1 between 1982 <strong>and</strong><br />
1997. South is in the direction of decreasing<br />
position.<br />
(Figure 1A). South of the boundary there are no<br />
Amb. limbatum (Figure 1B). The life cycle of<br />
both ticks has four stages: egg, larva, nymph <strong>and</strong><br />
adult [Bull <strong>and</strong> Burzacott 2001]. They require<br />
three hosts to complete their life cycle. Larvae,<br />
nymphs <strong>and</strong> adult females each attach to a host,<br />
engorge, <strong>and</strong> then detach (usually when the host is<br />
in an overnight refuge). Engorged larvae <strong>and</strong><br />
nymphs moult to the next stage. Engorged females<br />
lay eggs that hatch into larvae. Unfed larvae,<br />
nymphs <strong>and</strong> adults then wait in the refuge for a<br />
new host individual (or the same host) to attach to.<br />
From a tick’s point of view, the l<strong>and</strong>scape consists<br />
of the lizard hosts <strong>and</strong> their nocturnal refuge sites.<br />
There are R refuges in a 1 km x 20 km rectangle<br />
oriented perpendicular to the boundary zone;<br />
refuges are distributed with complete spatial<br />
r<strong>and</strong>omness. These refuges are used by L lizards<br />
whose home ranges are centred on a r<strong>and</strong>omly<br />
chosen refuge. All refuges within some distance h<br />
of the centre refuge are included in the home<br />
range. The l<strong>and</strong>scape "wraps" in the short<br />
direction, so the model l<strong>and</strong>scape is a long<br />
cylinder; the short boundaries are reflecting. The<br />
l<strong>and</strong>scape is initialised with 10,000 ticks of each<br />
species. Each species is confined to ½ of the<br />
l<strong>and</strong>scape at initialisation.<br />
There are two time scales in the model. On the<br />
short time scale, movement of lizards, birth,<br />
development, <strong>and</strong> death of ticks is modelled each<br />
day. A series of days is then aggregated into a<br />
season, which is 210 days (1st September to 31st<br />
March) long. Development is frozen between<br />
seasons, under the assumption that autumn/winter<br />
temperatures are too low for tick activity. Ticks<br />
experience overwintering mortality, <strong>and</strong> host<br />
population dynamics also occurs between<br />
seasons.<br />
The choice of a single day as the basic time step is<br />
logical given the assumption that all significant<br />
movement of ticks on <strong>and</strong> off lizards occurs only<br />
in refuges entered at night. Ticks are adapted to<br />
detach in the nocturnal refuges of their hosts,<br />
where desiccation risks are lower, <strong>and</strong> the chances<br />
of finding another host are higher [Bull <strong>and</strong><br />
Burzacott 2001]. Within a single day, the model<br />
goes through several steps in the following order:<br />
ticks board lizards, lizards move between refuges,<br />
engorged ticks disembark from lizards, <strong>and</strong> ticks<br />
develop (Figure 2).<br />
2.1 Tick embarkation, Lizard movement, Tick<br />
disembarkation<br />
896
At the beginning of a model day, all lizards are in<br />
the overnight refuges in which they spent the<br />
previous night. The model checks all ticks in<br />
lizard occupied refuges, <strong>and</strong> any ticks that are<br />
found to be in a suitable state (ie. unfed larvae,<br />
nymphs, or adults) are moved onto the lizard.<br />
Assuming that all suitable ticks board lizards is<br />
almost certainly<br />
Ticks board<br />
lizards<br />
Lizards<br />
redistributed over<br />
refuges<br />
Within season processes<br />
Between season processes<br />
Overwintering mortality of<br />
ticks<br />
Daily Loop<br />
(210 days / season)<br />
Development,<br />
Engorgement,<br />
Survival of ticks on<br />
all lizards <strong>and</strong> in all<br />
refuges<br />
Engorged ticks<br />
disembark<br />
Lizard mortality,<br />
birth, <strong>and</strong> dispersal<br />
Figure 2 Flowchart of main model processes.<br />
Processes that are attributes of lizard population<br />
dynamics <strong>and</strong> behaviour which indirectly affect<br />
the ticks are placed in ovals, while processes<br />
directly affecting ticks are in rectangles.<br />
an overestimate. If there is more than one lizard in<br />
a particular refuge, the number of ticks boarding<br />
each lizard is multi-nomially distributed with<br />
equal probability of boarding each lizard.<br />
In the next step of the daily cycle, lizards move to<br />
new refuges. Each day, lizards move from one<br />
overnight refuge to another overnight refuge<br />
chosen r<strong>and</strong>omly with equal probability from<br />
among those in their home range.<br />
The third step within the daily cycle is to drop off<br />
successfully engorged ticks into their new<br />
refuges. Essentially, ticks which completed<br />
engorgement on the previous development step<br />
(ie. the previous night), are dropped off in the new<br />
refuge chosen by their host lizard.<br />
The final step of the daily cycle h<strong>and</strong>les<br />
development <strong>and</strong> mortality of all ticks, regardless<br />
of their current location. During this step, each<br />
tick is checked to see whether it ages, survi ves, or<br />
lays eggs, depending on its current stage <strong>and</strong><br />
whether it is on a lizard or not.<br />
2.2 Growth <strong>and</strong> Feeding<br />
Stages that are engaged in growth or feeding<br />
(eggs, engorged stages in refuges, <strong>and</strong> unfed<br />
stages on lizards) follow a threshold process,<br />
where each stage lasts for a fixed number of days<br />
for each individual. Each individual is assigned a<br />
normally distributed r<strong>and</strong>om number as a<br />
development or engorgement time on entry to a<br />
new life history stage; values less than zero were<br />
truncated to zero. Both the mean <strong>and</strong> the variance<br />
are stage specific (Table 1), <strong>and</strong> refer to the nontruncated<br />
distributions. Although there are<br />
differences between the two species, at present<br />
we assume that all life history rates are equal.<br />
During the daily development step, each individual<br />
tick has its development or engorgement index<br />
decreased by one day. On the day the index<br />
reaches 0, the individual moves to the next stage<br />
(eg. an egg hatches to an unfed larvae, or a feeding<br />
nymph detaches). This means that the duration of<br />
all growing <strong>and</strong> feeding stages are normally<br />
distributed. This method is similar to those used<br />
for physiologically structured population models<br />
[Gurney et al. 1986], but includes variability<br />
between individuals.<br />
Not all individuals succeed in attaching, engorging<br />
<strong>and</strong> detaching, <strong>and</strong> this is where density<br />
dependence (<strong>and</strong> hence competition) is known to<br />
occur in the system [Tyre et al. 2003]. The<br />
probability of successfully engorging is<br />
e<br />
p{ successfulengorgement} =<br />
1 + e<br />
β =−0.243−0.002(# ofticks)<br />
β<br />
β<br />
. (1)<br />
This depends on the total number of ticks of all<br />
stages at the time engorgement is complete. The<br />
mechanism underlying the relationship between<br />
tick density <strong>and</strong> engorgement success is presently<br />
unknown, <strong>and</strong> this empirical relationship is the<br />
simplest to implement in the model. We currently<br />
have no evidence of density dependence in<br />
engorgement success for nymphs or adults. We<br />
assumed nymphs had a 50% chance of success,<br />
<strong>and</strong> adults 100%, regardless of the number of<br />
ticks present on the lizard.<br />
2.3 Survival<br />
897
Predation on ticks within refuges by other both spatially <strong>and</strong> temporally unpredictable, <strong>and</strong><br />
invertebrates does occur [Bull et al. 1988], but is<br />
Table 1 Developmental, feeding, <strong>and</strong> mortality parameters used in the baseline model. All values are<br />
estimated from data in [Chilton 1989], assuming temperatures of 21 0 C <strong>and</strong> 50-55% RH. All means have<br />
units of days. Feeding times for adult females includes the time required to be mated. Values in italics were<br />
extrapolated from estimates for larvae.<br />
Stage durations<br />
Stage Location Process Mean [days] SD<br />
Egg Refuge Hatching 53 1.32<br />
Unfed Larvae Refuge Mortality 13.8 4.9<br />
Unfed Larvae Lizard Feeding 30.6 11.7<br />
Engorged Larvae Refuge Moulting 21.9 4.07<br />
Unfed Nymphs Refuge Mortality 37.3 5.5<br />
Unfed Nymphs Lizard Feeding 22.7 16.7<br />
Engorged Nymphs Refuge Moulting 28 7.34<br />
Unfed Adults Refuge Mortality 100 7.3<br />
Unfed Females Lizard Feeding 39 17.6<br />
Engorged Females Refuge Pre-oviposition 55.2 8.44<br />
Mature Females Refuge egg-laying 40 --<br />
we do not include it in the current model. When a<br />
host lizard dies from predation (primarily<br />
automobiles near Mt. Mary) or old age all ticks on<br />
the lizard also die. We included this mortality in a<br />
single, between season event (see below). We<br />
assume that the primary source of daily mortality<br />
is desiccation. The habitat has low rainfall (150-<br />
250 mm annually), <strong>and</strong> most development occurs<br />
during the hot, dry summer. The only moisture<br />
source available to ticks is a blood meal, <strong>and</strong><br />
newly moulted, unfed ticks must wait until another<br />
host arrives before they can replenish their<br />
moisture supply. Eggs, engorged ticks in refuges,<br />
<strong>and</strong> ticks feeding on lizards are assumed to be<br />
unaffected by desiccation. We modelled mortality<br />
similarly to development, by providing each<br />
individual with a normally distributed time to<br />
death. This is the number of days that each<br />
individual is expected to survive without feeding.<br />
The time is decreased by one day in each<br />
developmental step, <strong>and</strong> individuals that reach zero<br />
are killed. Death is presumed to have occurred as<br />
a result of higher temperatures during the day, <strong>and</strong><br />
so mortality in a refuge precedes ticks boarding<br />
lizards that enter that refuge on the next day.<br />
2.4 Mating <strong>and</strong> Oviposition<br />
When an unfed adult tick boards a lizard, it is<br />
r<strong>and</strong>omly assigned to be a male or female with a<br />
sex ratio of 1:1. Adult male ticks remain on<br />
lizards for the remainder of their life, assumed to<br />
be a fixed 180 days. The only contribution they<br />
have beyond this point is to mate with unfed<br />
female ticks. After boarding a lizard there is a<br />
fixed five day period before an adult male is<br />
mature <strong>and</strong> capable of mating. When an unfed<br />
female boards a lizard, if there is one or more<br />
mature males aboard she is mated immediately.<br />
Otherwise, she waits on that lizard until a mature<br />
male appears, or 180 days passes. There is no<br />
negative impact of waiting to mate on a females<br />
subsequent reproductive output, although a<br />
negative impact has been observed in laboratory<br />
experiments [Chilton 1989]. Once a female is<br />
mated, she begins to engorge as described above<br />
for all other stages. This does introduce a slight<br />
Allee effect through delaying reproduction by<br />
females that board lizards without males.<br />
Adult female ticks that have mated, successfully<br />
engorged, <strong>and</strong> dropped off in a refuge enter a preoviposition<br />
stage, the duration of which is<br />
normally distributed <strong>and</strong> h<strong>and</strong>led identically to<br />
aging, feeding, <strong>and</strong> moulting. Once the preoviposition<br />
period is complete, each day for 40<br />
days they add a number of new eggs to that refuge.<br />
2.5 Between season processes<br />
There are two processes that occur between the<br />
end of one season <strong>and</strong> the beginning of the next:<br />
overwintering tick mortality <strong>and</strong> lizard population<br />
dynamics. All ticks, regardless of location, have a<br />
stage specific chance of mortality over winter.<br />
This reflects exposure, desiccation, disease,<br />
predation, <strong>and</strong> fungal infection. We set this to<br />
10% for all stages other than eggs. It is set low<br />
898
elative to mortality during the active season<br />
because the risk of desiccation is reduced in the<br />
cooler, wetter climate of winter, <strong>and</strong> invertebrate<br />
predators are less active. However, laboratory<br />
observations indicate that no eggs hatch when held<br />
at temperatures of less than 15 o C. Therefore, egg<br />
mortality is 95% over the winter in the model,<br />
which allows for a small margin of error in the<br />
laboratory estimate of 100%.<br />
Lizard population dynamics is also simplified. A<br />
flat 10% of all lizards are chosen at r<strong>and</strong>om <strong>and</strong><br />
killed at the end of each season. Empirical<br />
observations place annual adult survival at around<br />
90% [Bull 1995]. Any ticks on the killed lizards<br />
are also killed. The killed lizards are replaced<br />
from newborns whose mothers are chosen at<br />
r<strong>and</strong>om from the surviving lizards. These newborn<br />
lizards spend one season in their mother’s home<br />
range, before r<strong>and</strong>omly choosing a home range of<br />
their own (natal dispersal). This results in no net<br />
change in the number of lizards available, but<br />
tends to redistribute 10% of the population to new<br />
locations each season after the first two. New<br />
home range sites are selected in one of two ways:<br />
exponentially distributed dispersal distance with a<br />
mean of 500 m (limited dispersal scenario), or<br />
effectively unlimited dispersal with a mean of<br />
> 6000 m (high dispersal scenario).<br />
species is shown in Figure 4. The distributional<br />
boundaries of both species move slowly through<br />
time, leading to an increase in the breadth of the<br />
zone where both species can be found. Across 4<br />
replicate runs the width of the boundary zone<br />
increases at a rate of 20 m / year (SE=4 m / year;<br />
Figure 5). Although the rate of increase is small,<br />
the boundary zone is clearly not stable.<br />
Joint Incidence<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0<br />
5<br />
Position [km]<br />
10<br />
Figure 3 Joint incidence of both ticks in a high<br />
dispersal run. Position is in km, with zero the midpoint<br />
between the two species initial distributions.<br />
15<br />
0<br />
50<br />
100<br />
Year<br />
150<br />
200<br />
3. RESULTS<br />
All results are presented as smooth fits to data<br />
sampled along a transect positioned down the<br />
centre of the simulated l<strong>and</strong>scape. Simulated<br />
samples are collected once per week. Each lizard<br />
whose home range overlaps the transect has a 10%<br />
probability of being captured <strong>and</strong> having its<br />
current load of ticks enumerated. The incidence is<br />
worked out for all captures in a year within a 1 km<br />
segment of the transect. This mimics the kind of<br />
sampling carried out in reality (Figure 1). In the<br />
following, the boundary zone is defined as the area<br />
where joint incidence of both species is greater<br />
than 1%.<br />
High Dispersal: A representative space-time<br />
perspective plot of the "joint" incidence<br />
(probability that a host has both species of tick) is<br />
shown in Figure 3. Although both species are<br />
separated by 4 km at initialisation, their<br />
distributions broadly overlap in less than 50 years.<br />
In addition, the simulated data show none of the<br />
stable spatial heterogeneity evident in Figure 1.<br />
Joint Incidence<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0.0<br />
0<br />
5<br />
Position [km]<br />
10<br />
Figure 4 Joint incidence of both ticks in a low<br />
dispersal run.<br />
15<br />
0<br />
50<br />
100<br />
Year<br />
150<br />
200<br />
Low Dispersal: A representative space-time<br />
perspective plot of the joint incidence of both tick<br />
899
Relative width of overlap zone [km]<br />
0 2 4 6 8 10 12<br />
50 100 150 200<br />
zone can move. In addition, heterogeneity in host<br />
abundance arising simply through r<strong>and</strong>om birth,<br />
death, <strong>and</strong> movement of the hosts does influence<br />
tick abundance, but does not lead to a stationary<br />
boundary. However, despite the absence of a<br />
gradient in life history performance or density<br />
dependent dispersal the boundary zone remains<br />
quite static for relatively long periods of time.<br />
Our future work will concentrate on comparisons<br />
of the agent based model outlined here with<br />
models based on diffusion approximations to<br />
movement, <strong>and</strong> on comparing the output of the<br />
model with the empirical data.<br />
Time [Years]<br />
Figure 5 Change in width of the boundary zone<br />
with time. Each symbol represents a replicate<br />
simulation.<br />
Table 2 Results of GAM fits to incidence of Ap.<br />
hydrosauri for one replicate south of the initial<br />
boundary. s( ) indicates the smooth term, Y is<br />
year, P is Position, <strong>and</strong> L is relative lizard<br />
abundance. Lizard effect is the linear coefficient<br />
(st<strong>and</strong>ard error).<br />
Model Deviance explained Lizard<br />
Effect<br />
s(Y, P) 77% -<br />
s(Y,P)+ 78% 0.05 (0.003)<br />
L<br />
s(Y) + L 33% 0.15 (0.002)<br />
Effect of heterogeneity in host abundance: Table<br />
2 compares three Generalised Additive Model<br />
(GAM) fits to incidence of Ap. hydrosauri from<br />
south of the initial boundary. A model with a<br />
smooth term in both space <strong>and</strong> time explains 77%<br />
of the deviance in incidence. Adding a linear<br />
effect for the abundance of hosts at each point<br />
only increases the explanatory power to 78%.<br />
Deleting spatial position from the smooth term<br />
dramatically reduces the explanatory power of the<br />
model, although the effect of lizard number<br />
increases. It appears that heterogeneity in host<br />
abundance does influence tick abundance (positive<br />
coefficient), but that this effect is largely<br />
overridden by spatial autocorrelation in the<br />
abundance of ticks themselves.<br />
5. CONCLUSIONS<br />
Our preliminary results for this model clearly<br />
indicate that dispersal of juvenile hosts has a<br />
dramatic effect on the rate at which the boundary<br />
6. ACKNOWLEDGEMENTS<br />
This work has been funded over many years by<br />
Australian Research Council grants to CMB.<br />
7. REFERENCES<br />
Bull, C.M., Ecology of parapatric distributions,<br />
Annual review of ecology <strong>and</strong> systematics ,<br />
22, 19-36, 1991.<br />
Bull, C.M., Population ecology of the sleepy<br />
lizard, tiliqua rugosa, at mt. Mary, south<br />
australia, Australian Journal of Ecology, 20,<br />
393-402, 1995.<br />
Bull, C.M., <strong>and</strong> D. Burzacott, Temporal <strong>and</strong> spatial<br />
dynamics of a parapatric boundary between<br />
two australian reptile ticks, Molecular<br />
Ecology, 10, 639-648, 2001.<br />
Bull, C.M., N.B. Chilton, <strong>and</strong> R.D. Sharrad, Risk<br />
of predation for two reptile tick species,<br />
Experimental & Applied Acarology, 5, 93-<br />
100, 1988.<br />
Bull, C.M., <strong>and</strong> H. Possingham, A model to<br />
explain ecological parapatry, American<br />
Naturalist, 145, 935-947, 1995.<br />
Chilton, N.B. 1989. Life cycle adaptations <strong>and</strong><br />
their implications in the distribution of two<br />
parapatric species of tick. in C.M. Bull,<br />
editor., Flinders University, South Australia.<br />
García-Ramos, G., F. Sanchez-Gárduño, <strong>and</strong> P.K.<br />
Maini, Dispersal can sharpen parapatric<br />
boundaries on a spatially varying environment,<br />
Ecology, 81, 749-760, 2000.<br />
Gurney, W.S.C., R.M. Nisbet, <strong>and</strong> S.P. Blythe, The<br />
systematic formulation of models of stagestructured<br />
populations, in J.A.J. Metz <strong>and</strong> O.<br />
Diekmann, editors. The dynamics of<br />
physiologically structured populations.<br />
Springer-Verlag, Heidelberg.<br />
900
MacArthur, R.H., Geographical ecology. Patterns<br />
in the distribution of species, Harper & Row,<br />
New York, 1972.<br />
Main, A.R., <strong>and</strong> C.M. Bull, The impact of tick<br />
parasites on the behaviour of the lizard tiliqua<br />
rugosa, Oecologia, 122, 574-581, 2000.<br />
Ribeiro, J.M.C., <strong>and</strong> A. Speilman, The satyr effect:<br />
A model predicting parapatry <strong>and</strong> species<br />
extinction, American Naturalist, 128, 513-<br />
528, 1986.<br />
Tyre, A.J., C.M. Bull, B. Tenhumberg, <strong>and</strong> N.B.<br />
Chilton, Indirect evidence of densitydependent<br />
population regulation in<br />
aponomma hydrosauri(acari: Ixodidae), an<br />
ectoparasite of reptiles, Austral Ecology, 28,<br />
196-203, 2003.<br />
901
How to Compare Different Conceptual Approaches to<br />
Metapopulation <strong>Modelling</strong><br />
Frank M. Hilker a , Martin Hinsch b <strong>and</strong> Hans Joachim Poethke b<br />
a Institute of <strong>Environmental</strong> Systems Research, Department of Mathematics <strong>and</strong> Computer Science,<br />
University of Osnabrück, Germany, e-mail: fhilker@uos.de<br />
b Ecological Research Station, Bavarian Julius-Maximilians-University of Würzburg, Germany<br />
Abstract: Models are essential tools in underst<strong>and</strong>ing population dynamics <strong>and</strong> deriving management measures<br />
in the context of population viability analysis. However, very often the question arises which type of<br />
model architecture is appropriate for a given situation. Mostly this situation is characterized by a shortage of<br />
data for model parameterization. In this study, an approach is presented to overcome this lack of real-world<br />
data by using the output of long-term simulation runs of specific individual-based models. Thus, it is possible<br />
to evaluate the quality of macroscopic model predictions. Furthermore, this setting allows to compare totally<br />
different types of metapopulation models. As an exemplary case study, this approach is applied to generic<br />
grasshopper species in highly fragmented habitat l<strong>and</strong>scapes, assessing on the one h<strong>and</strong> the well-known incidence<br />
function model <strong>and</strong> on the other h<strong>and</strong> a grid-based approach. The results show that predictions of both<br />
models have substantial biases. Nonetheless, recommendations can be derived how to obtain more accurate<br />
model estimates. Finally, the patch-matrix model proves to be more adequate than the grid-based approach.<br />
Keywords: Metapopulation models; incidence function model; patch-matrix model; grid-based model;<br />
individual-based model<br />
1. INTRODUCTION<br />
In the last decades, a variety of modelling approaches<br />
has been developed in order to underst<strong>and</strong><br />
population dynamics as well as to be able to derive<br />
appropriate management measures in the context of<br />
population viability analysis. These models differ in<br />
the degree in which they take space, time <strong>and</strong> state<br />
variables into account. For example, there are ordinary<br />
<strong>and</strong> partial differential equations, (integro-)<br />
difference <strong>and</strong> integrodifferential equations, cellular<br />
automata, coupled map lattices, interacting particle<br />
systems or individual-based models (e.g., see<br />
Czárán [1998] <strong>and</strong> references therein).<br />
Hence, modelleres are often confronted with the<br />
question, which conceptual approach seems to be<br />
most appropriate for a certain problem. This paper<br />
is concerned with a comparison between the wellknown<br />
”practical model of metapopulation dynamics”<br />
of Hanski [1994] <strong>and</strong> a grid-based approach<br />
proposed by Settele [1998]. Both are incidence<br />
function models (IFMs), which can relatively easily<br />
be parameterized with the species’ occupancy data.<br />
They differ in the spatial representation of habitats<br />
(cf. Figure 1). In the Hanski model, which shall<br />
be referred to as patch-matrix model (PMM), the<br />
species is assumed to inhabit circular patches of different<br />
sizes within a hostile matrix. In distinction to<br />
the spatial implicit model of Levins [1969] <strong>and</strong> to<br />
spatially explicit models, the PMM is called a spatially<br />
realistic model [Hanski, 1999]. The PMM has<br />
frequently been used in population viability analyses<br />
of endangered species (see Hanski [2001] <strong>and</strong><br />
references therein). In Settele’s approach, which<br />
shall be referred to as grid-based model (GBM),<br />
space is sub-divided into equally sized cells with<br />
different carrying capacities. Each cell is assumed<br />
to be a possible habitat which may be occupied by<br />
the species. Please note the differentiation between<br />
the general habitat, patch (PMM) <strong>and</strong> cell (GBM)<br />
throughout this paper.<br />
There is an increasing dem<strong>and</strong> for grid-based models,<br />
since data on the distribution of various species<br />
are often available in a grid-based format, as this can<br />
easily be h<strong>and</strong>led (for example with Geographic Information<br />
Systems). Regarding Settele’s approach,<br />
however, there is some severe scepticism about the<br />
902
a)<br />
IBM<br />
simulations<br />
Snapshot occupancies<br />
Long-term data<br />
b) c)<br />
K i<br />
IFM-parameters estimators<br />
”Real” parameter values<br />
A j<br />
d i j<br />
A i<br />
r i j<br />
K j<br />
ϕ i j<br />
Figure 1. Representation of habitat configurations<br />
(a) in the patch matrix model (b) <strong>and</strong> in the gridbased<br />
approach (c). Habitats are coloured with grey<br />
values corresponding to their average occupancies.<br />
biological realism in the underlying assumptions,<br />
as will be pointed out in the model description in<br />
the following section. The aim of this paper is<br />
to demonstrate the application of a general method<br />
for the comparison of conceptually different modelling<br />
approaches. Especially in conservation biology,<br />
this is of increasing interest, as generally little<br />
is known about the species under focus <strong>and</strong> often<br />
few quantitative data are available. This study<br />
makes use of an approach which has recently been<br />
presented by Hilker [2002], cf. Figure 2. As there is<br />
a lack of real-world data of sufficient resolution, an<br />
individual-based model (IBM) is used to simulate<br />
complex population dynamics of ”virtual”, generic<br />
species. The simulation runs generate extensive<br />
long-term data sets. On the one h<strong>and</strong>, the PMM <strong>and</strong><br />
the GBM, which are both highly aggregated models,<br />
can now be parameterized with various shortterm<br />
data samples. In this study, snapshot data of<br />
two or five consecutive years are used as it is typical<br />
for field campagains. On the other h<strong>and</strong>, the<br />
”real” parameter values describing the metapopulations<br />
dynamics can be extracted directly from the<br />
full amount of IBM-data (which consist in this study<br />
of 400 years). Thus, one obtains parameter estimators<br />
from the highly aggregated models (based on<br />
short-term snapshot data) <strong>and</strong> ”real” values from<br />
the specific model (based on long-term data), which<br />
can be compared with each other, especially with respect<br />
to the (dis-)advantages of different space representation.<br />
This paper is organized as follows. Firstly, all three<br />
models (PMM, GBM, IBM) are introduced. As<br />
a result of the PMM <strong>and</strong> GBM, the estimators of<br />
Figure 2. An IBM simulates a species’ dynamics in<br />
fragmented habitats. From the available long-term<br />
data, the ”real” parameter values can be extracted.<br />
With several snapshot data both the patch-matrix<br />
<strong>and</strong> the grid-based IFM are parameterized. Their<br />
parameter estimators can then be compared with the<br />
”real” values.<br />
the metapopulation dynamic parameters are yielded<br />
(Subsection ”Parameter estimators”). The ”real”<br />
values are determined each from the full amount<br />
of IBM-data including dispersal events (Subsection<br />
”Extraction of ’real’ values”). Please note, that all<br />
settings of the simulated species, habitat configuration<br />
<strong>and</strong> snapshot sampling are the same as in Hilker<br />
[2002]. Next, the results are given with a focus on<br />
the accuracy of the parameter estimators. Finally,<br />
potential reasons for the resulting deviations in the<br />
estimators are discussed.<br />
2. INCIDENCE FUNCTION MODELS (IFMS)<br />
IFMs rely on presence-absence data of a species in a<br />
set of habitats. In typical field campaigns, these occupancy<br />
data are collected over a single or (better) a<br />
few generations, which have not to be consecutive.<br />
Because of that, these data are often referred to as<br />
patch occupancy pattern or snapshot data. Henceforth,<br />
the latter notation will be used throughout this<br />
paper. The observed snapshot data are assumed to<br />
represent the quasi-equilibrium of metapopulation<br />
dynamics. The modelling objective is to fit the incidence<br />
function to the observed snapshot data, thus<br />
obtaining metapopulation dynamic parameter estimators.<br />
Once these parameters are estimated, the<br />
IFMs can be used to predict habitat-specific colonization<br />
<strong>and</strong> extinction probabilites for a particular<br />
habitat configuration. Thus, occupancies, transient<br />
dynamics, <strong>and</strong> regional population persistence may<br />
be predicted.<br />
If habitat i is extinct (respectively occupied), it has<br />
the colonization probability C i (respectively extinction<br />
probability E i ) of becoming occupied (respectively<br />
extinct) at the next time step. These transi-<br />
903
¥<br />
¥<br />
¡<br />
¥<br />
tions are assumed to occur at r<strong>and</strong>om for each habitat.<br />
The probability that habitat i will be occupied<br />
tends toward the stationary probability<br />
J i<br />
C i<br />
(1)<br />
¡ ¢<br />
C i E i 1 C i¤¦¥ £<br />
which is called the incidence <strong>and</strong> assumes a quasisteady<br />
state of metapopulation dynamics conditional<br />
on non-metapopulation extinction. In (1), the<br />
rescue effect is included. Mathematically, IFMs are<br />
time homogeneous, discrete time first order finite<br />
state Markov chains.<br />
2.1 Patch-matrix model (PMM)<br />
The PMM is described in detail by Hanski<br />
[1994, 1999], or see references therein. The extinction<br />
probability E i is assumed to vary with the<br />
x<br />
patch area A i (in ha): E i min§ e 0 i , A¨ where 1© e 0<br />
<strong>and</strong> x are extinction parameters. Next, the colonization<br />
probability C i is approximated by the number<br />
Mi<br />
of immigrants M i arriving at patch i: C 2 i<br />
Mi 2 y , 2 <br />
where y is a colonization parameter. M i itself depends<br />
on the connectivity S i through S i βM i<br />
β∑ j i p j A j exp £ αd i ¢ j¤ , with d i j being the distance<br />
between patches i <strong>and</strong> j (in km), p j the relative frequency<br />
of patch occupancy. β is assumed to equal<br />
unity <strong>and</strong> α is a migration parameter.<br />
Finally, one can combine y yβ¨<br />
the parameters<br />
<strong>and</strong> e 2 e 0 <strong>and</strong> then incorporate y<br />
C i <strong>and</strong> E i in (1).<br />
Note that only patches with A i A 0 : e x 1 0 are<br />
considered, due to the minimum-operator in the extinction<br />
probability. A 0 is the critical patch area, below<br />
which the extinction probability E i equals unity.<br />
2.2 Grid-based model (GBM)<br />
The GBM has been suggested by Settele [1998].<br />
Space is represented by a grid, whose cells may either<br />
be occupied by local populations or not. Since<br />
all cells are equally sized, the carrying capacity cannot<br />
be approximated by the area as in the PMM. Instead,<br />
the extinction probability is described by<br />
E i exp £ κK ¢ i¤<br />
(2)<br />
where κ is an extinction parameter <strong>and</strong> K i a measure<br />
for the carrying capacity. K i is set to the relative<br />
frequency p i with which the cell is occupied in<br />
the snapshot data. In the case, that a cell is always<br />
unoccupied <strong>and</strong> one can exclude that it is hostile to<br />
the species, one assigns the minimum capacity of all<br />
cells which have been occupied at least once.<br />
The colonization probability is along the line of the<br />
PMM<br />
1<br />
M 2 i<br />
C i<br />
Mi<br />
2 µ 2 (3)<br />
¥<br />
with µ being a colonization parameter. The mean<br />
number of immigrants is approximated by M i<br />
∑ j i M i j with<br />
M i j p j K j exp ¢ £ ρr i j¤ ϕ i j (4)<br />
The term p j K j is a measure for the population abundance<br />
in cell j. The fraction of individuals dispersing<br />
the Euclidean distance r i j (in km) between the<br />
source cell j <strong>and</strong> the target cell i is determined by<br />
1<br />
π<br />
arctan ¢ D<br />
2ri j<br />
¤ is<br />
the migration parameter ρ. ϕ i j<br />
the maximum angle of a circle-segment from the<br />
midpoint of the source patch to the ends of the target<br />
patch, cf. Figure 1. D is the cell length.<br />
Principally, the GBM resembles the PMM in being<br />
a stochastic patch occupancy model based on a<br />
regression model. However, by dividing the l<strong>and</strong>scape<br />
in a grid, local populations inhabiting an<br />
area greater than a single cell are also subdivided.<br />
Hence, the assumption of panmixia for local populations<br />
(patches) is relaxed. Or, contrariwise, two or<br />
more small habitats might be subsumed in one cell.<br />
2.3 Parameter estimators<br />
The PMM as well as the GBM are characterized<br />
by an initially unknown<br />
¢<br />
set of species-specific<br />
metapopulation dynamic e x¤<br />
¢<br />
parameters θ α<br />
or θ µ¤ ρ κ , respectively. These are ¥ obtained<br />
by fitting ¥ ¥ (1) to the snapshot data. Using<br />
maximum pseudo-likelihood regression, the difference<br />
between the snapshot data p i (approximating<br />
the quasi-steady state of the metapopulation)<br />
<strong>and</strong> the model-predicted incidences J i is minimized.<br />
In the pseudo-likelihood function, a binomial<br />
distribution of the species’ occurences is assumed.<br />
Dealing with an optimization problem, the<br />
permutation term can be neglected <strong>and</strong> the likelihood<br />
be log-transformed, θ¤<br />
thus yielding l ¢ ¢ ¢ ¡<br />
∑ i p i log £ J ¢ i¤ 1 p £ i¤ log 1 J ¢ i¤¤ . For maximization<br />
of this function, the simulated annealing algorithm<br />
is used, because it is able to escape from local<br />
optima in the search space <strong>and</strong> find global solutions.<br />
Note, that the PMM-parameters e 0 <strong>and</strong><br />
can e be separated from y by defining A 0 as the<br />
area of the smallest occupied habitat patch (e 0 A x 0 ,<br />
y e e 0 ).<br />
3. SIMULATION OF LONG-TERM DATA<br />
3.1 Individual-based model (IBM)<br />
The IBM simulates stochastically the metapopulation<br />
dynamics of generic bush crickets (or any similar<br />
invertebrate species) with a one-year life-cycle<br />
904
¢<br />
(egg – larva – adult) <strong>and</strong> non-overlapping generations<br />
in a highly fragmented, realistic l<strong>and</strong>scape<br />
with a binary habitat distinction (habitat vs. nonhabitat).<br />
Larvae <strong>and</strong> adults move with certain distances<br />
<strong>and</strong> turning angles, in the matrix much longer<br />
<strong>and</strong> more straight-forward than within the habitat.<br />
Adults have a detection radius for finding mating<br />
partners in their vicinity (Allee effect). The<br />
number of eggs per female is Poisson-distributed<br />
<strong>and</strong> the number of propagules additionally depends<br />
on available resources (density dependence). The<br />
number of available resources fluctuates because<br />
of overlapping local catastrophes (locally correlated<br />
environmental fluctations, but global stochasticity).<br />
For more details, please see Hilker [2002], where<br />
the emergence of metapopulation dynamics from<br />
the individual behaviour <strong>and</strong> the patchy distribution<br />
of habitats has been demonstrated.<br />
3.2 Extraction of ”real” parameter values<br />
A method to extract the IFMs-parameters from the<br />
long-term IBM data has been developed in Hilker<br />
[2002]. Here, it shall be focused on the method regarding<br />
the GBM (which is principally analogous to<br />
the PMM). Contrary to the maximum-likelihood approach<br />
for yielding the parameters estimators (Subsection<br />
2.3), each of the real values can be extracted<br />
by fitting the mechanistic functions of the GBM, i.e.<br />
Eq.s (2), (3), (4), to the long-term IBM-data. The<br />
IBM is run 200 years to let the metapopulation dynamics<br />
reach its quasi-equilibrium. Then, further<br />
400 years are simulated, in which the occupancies<br />
of each cell <strong>and</strong> thus the transitions between being<br />
occupied or empty are recorded.<br />
Let Ni<br />
kl denote the number of transitions of cell i<br />
from state k to l (k l 1: cell occupied, k l 0:<br />
empty). Then one obtains ¥ as likelihood function ¥ for<br />
¢<br />
the recorded transitions: P i 1 C<br />
N 00<br />
i¤<br />
i N<br />
C 01<br />
£ i i E i £<br />
E i C<br />
N 10 ¡ ¢<br />
£ i¤<br />
i 1 E i E i C<br />
N 11<br />
i¤<br />
i . Now, C i <strong>and</strong> E i can be<br />
approximated by maximizing P i (which has been<br />
done with the Fletcher-Reeves conjugate gradient<br />
algorithm [Ueberhuber, 1997]).<br />
Once the extinction <strong>and</strong> the colonization probabilities<br />
of each cell are known, the model equations<br />
can be fitted to them in nonlinear least-square fits,<br />
thus yielding the unknown parameter set θ. Firstly,<br />
the extinction parameter κ can be extracted from<br />
the relationship K i – E i , cf. (2). Next, consider<br />
the migration parameter ρ. Transforming (4) yields<br />
M i j<br />
p i K i ϕ i j<br />
exp £ ρr i ¢ j¤ . Since the values of the exponential<br />
function for negative arguments are always<br />
in the unit interval, the left-h<strong>and</strong> side is scaled by<br />
Table 1. Mean ”real” parameter values (st<strong>and</strong>ard<br />
deviations) of the GBM.<br />
Species ρ µ κ<br />
28 ¢ 8¤ ¢ 7 20 19 0 ¢ 0¤<br />
0 09¤ 8 0 07¤ 2 ¢ 0 ¢ 25¤<br />
9¤ 25 0<br />
20 0 ¢ 0 05¤ 9 1 ¢ 7¤ 0 07¤<br />
1 3 08 9<br />
2 8 10 1<br />
3 4 10 4<br />
dividing through the maximum number of recorded<br />
M<br />
immigrants: i j<br />
p i K i ϕ i j maxM i j<br />
exp £ ρr i ¢ j¤ . Now, this<br />
equation can be fitted as well. Having determined<br />
ρ, the cell connectivities can be computed, which<br />
allows to fit (3), thus finally obtaining the colonization<br />
parameter µ.<br />
4. SIMULATIONS<br />
With the IBM, three different species have been<br />
simulated in varying habitat configurations. Table 1<br />
shows the mean values of the ”real” GBM parameter<br />
values extracted from the long-term IBM data.<br />
Since the parameters are assumed to be speciesspecific,<br />
they are averaged over all habitat configurations<br />
as well as replications. In all simulations, the<br />
cell length of the GBM has been set to D 100 m.<br />
How accurate <strong>and</strong> precise are the parameter estimators<br />
of the GBM parameterized with snapshot data<br />
of two <strong>and</strong> five consecutive years? In Table 2, the<br />
relative errors <strong>and</strong> variation coefficients are given,<br />
which are measures for the accuracy <strong>and</strong> the precision.<br />
If the relative error equals zero, this means a<br />
perfect match. If it is positive/negative, the parameter<br />
is over-/underestimated, respectively. As one<br />
can easily see, there are enormous deviations in the<br />
colonization parameter µ. By using more extensive<br />
snapshot data with five years, these deviations are<br />
reduced, but they are still huge.<br />
In many studies which make use of the PMM, the<br />
migration parameter is estimated by independent<br />
data (cf. the survey in Hanski [1999]). Analogously,<br />
consider the situation in which the ”real” value of ρ<br />
is known. Then the dimension of the search space<br />
in the parameter estimation process is lowered from<br />
three to two. The results are listed in Table 3. For<br />
snapshot data consisting of two years, the colonization<br />
parameter µ is still heavily overestimated. But<br />
with five years, the extreme deviations vanish. The<br />
same tendency holds for the extinction parameter κ.<br />
With two years, it is obviously overestimated for<br />
two of the three species. Using five years, κ can<br />
be determined more accurately.<br />
In the lower rows of Table 2 <strong>and</strong> 3, the PMM-<br />
905
Table 2. Relative errors [variation coefficients] of the parameter estimators. Upper row: GBM, lower row: PMM.<br />
Sp. 2 years 5 years 2 years 5 years 2 years 5 years 2 years 5 years<br />
ρ µ κ<br />
0¡ ¢ 35£¤ ¢ 0¡ 403¡ 130¡ 0¡ ¢ 3£ 0¡ 0¡ 0¡ 20£ 07£<br />
159¡ 94¡ 0¡ ¢ 17£¤ ¢ 0¡ ¢ 1£<br />
4£ 196¡ ¢ 02£¤ 33¡ 7£ 4¡ ¢ 0¡ 0¡ 39£ 2¡ ¢ 0¡ 25£<br />
0¡ 0¡ 01£ 390¡ 0¡ 55¡ 0¡ 21£ 369¡ ¢ 0¡ 29¡ 0£ 148¡ ¢ 01£¤ 55¡ 7£ 0¡ 6¡ ¢ 0¡ 42£ 3¡ 0¡ ¢ 0¡ 74£<br />
y¥<br />
1 55 72 0 3 23 59<br />
2 85 91 3 2 37 24<br />
3 85 83 4 8 29 26<br />
α x e 0<br />
0¡ ¢ 17£¤ ¢ 0¡ 9¡ 8¡ 0¡ ¢ 9£ 0¡ 0¡ 0¡ 19£ 38£ ¢ ¢ 0¡ 0¡ 1£ 1¡ ¢ 44£ ¢ 0¡ 0¡ 0¡ 20£¦ ¢ 28£<br />
12¡ 12¡<br />
30£ 0¡ 1¡ ¢ 0¡ 37£ 1¡ 0¡ ¢ 0¡ 40£ 0¡ ¢ 0¡ 0¡ 17£¦ 0¡ ¢ 0¡ 09£ 12£<br />
3¡ 0¡ 5¡ 11£¤ 0£ 3¡ 1¡<br />
8¡ 0¡ ¢ 6¡ 8£ 8¡ ¢ 13£¤ 3¡ 61£ 1¡ ¢ 0¡ 0¡ 35£ 0¡ 1¡ ¢ 0¡ 41£ 0¡ 0¡ ¢ 0¡ 05£¦ 0¡ ¢ 11£ 0¡ 06£<br />
1 37 42 7 6 17 89 68 50<br />
2 26 24 7 0 20 26 81 83<br />
3 33 33 9 7 71 69 93 93<br />
Table 3. Relative errors [variation coefficients] of the parameter estimators resulting from a 2-dimensional estimation<br />
process with given ”real” migration parameter values. Upper row: GBM, lower row: PMM.<br />
Species 2 years 5 years 2 years 5 years 2 years 5 years<br />
µ κ<br />
77¡ 1 7 96¡ 17£ 2¡ 91 ¢ 4¡ 15£ 0¡ 21 ¢ 0¡ 25£§ 0¡ 67 ¢ 0¡ 09£<br />
2 ¢ 0 ¢ 146¡ 1£¤ 0¡ 11 ¢ 0¡ 72£ 5¡ 11 ¢ 3¡ 17£ 1¡ 12 ¢ 0¡ 40£<br />
77¡<br />
43¡ 3 0 114¡ 2£ 1¡ 12 ¢ 1¡ 41£ 6¡ 17 ¢ 3¡ 90£ 1¡ 91 ¢ 0¡ 75£<br />
¢<br />
y¥<br />
0¡ ¢ 1¡ 0¡ ¢ 79£ ¢ 0¡ 1¡ 0¡ 35£ ¢ 22£¤ 0¡ 0¡ 04£ 0¡ 0¡ 1¡ ¢ 22£<br />
50£ ¢ 0¡<br />
51£ 1¡ ¢ 0¡ 29£ 0¡ 0¡ ¢ 0¡ 08£¤ 0¡ ¢ 0¡ 0¡ 06£<br />
0¡ 46£ 1¡ 0¡ 0¡ 31£<br />
75£ 0¡ 1¡ ¢ 0¡ 30£ 1¡ 2¡ ¢ 0¡ 22£ 0¡ 91£ ¢ 0¡ 2¡ 04£¤ 0¡ ¢ 0¡ 03£<br />
0¡<br />
x e 0<br />
1 39 46 08 00 61 60<br />
2 68 92 34 43 87 89<br />
3 48 43 59 60 92 93<br />
estimators are considered as well. Note, that there<br />
is one more extinction parameter in the PMM. Only<br />
in the case of using five years <strong>and</strong> with a predetermined<br />
migration parameter, the GBM yields deviation<br />
ranges similar to those of the PMM. In all other<br />
settings, the PMM is more accurate.<br />
5. DISCUSSION<br />
There are enormous deviations in the estimators of<br />
the GBM. They decrease, if the migration parameter<br />
is predetermined from independent data. That is<br />
not surprising, because the dimension of the search<br />
space is reduced. However, the usage of the GBM<br />
seems to be applicable only if the migration parameter<br />
is known. Moreover, the GBM proves to be<br />
relatively accurate only in the case when five snapshot<br />
years are available. This can be explained as<br />
follows. With two snasphot years, the relative frequency<br />
of occupancy p i may either be 0.0, 0.5 or<br />
1.0. Remember the usage of a minimum carrying<br />
capacity, which will be in this case at least ¨ 0 5.<br />
Hence, there is an implicit tendency to homogenization<br />
of space, because nearly all cells are possible<br />
habitats. Using five snapshot years, instead, the<br />
minimum carrying capacity can be as low as 0.2.<br />
So far the accuracy of the estimators has been considered.<br />
What happens, if the estimators are used<br />
in the IFM simulation process? Incorporating the<br />
parameters into the model equations, i.e. in the<br />
case of GBM (2) <strong>and</strong> (3), the predicted incidences<br />
M i<br />
M i<br />
M i<br />
1.4<br />
1.2<br />
1<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
Species1<br />
0<br />
0 0.2 0.4 0.6 0.8 1<br />
K i<br />
Species2<br />
1.4<br />
1.2<br />
1<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0<br />
0 0.2 0.4 0.6 0.8 1<br />
K i<br />
Species3<br />
1.4<br />
1.2<br />
1<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0<br />
0 0.2 0.4 0.6 0.8 1<br />
K i<br />
M i<br />
M i<br />
M i<br />
4<br />
3<br />
2<br />
1<br />
Species1<br />
0<br />
0 1 2 3 4 5 6 7<br />
A i<br />
Species2<br />
4<br />
3<br />
2<br />
1<br />
0<br />
0 1 2 3 4 5 6 7<br />
A i<br />
Species3<br />
4<br />
3<br />
2<br />
1<br />
0<br />
0 1 2 3 4 5 6 7<br />
A i<br />
Figure 3. Contour plots of the residuals in J i (left: GBM,<br />
right: PMM). White means a perfect match, <strong>and</strong> each contour<br />
line / darker shading corresponds to an increase in the<br />
residuals of 0.1. Details are explained in the text. Note the<br />
different scaling of the ordinate axes, i.e. the connectivities<br />
in the sense of the PMM <strong>and</strong> GBM.<br />
906
J i can be calculated as a function of the parameter<br />
estimators, M i <strong>and</strong> K i . If this process is repeated<br />
with the ”real” values, the residual differences between<br />
estimated incidences can be calculated (Figure<br />
3). These may be taken as a good measure for<br />
the relevance of errors in parameter estimators. In<br />
the PMM the patch areas have nearly no influence<br />
on the residuals. They are determined by the connectivity.<br />
In contrast, the residuals in the GBM are<br />
not only influenced by the connectivity, but much<br />
more by the carrying capacities.<br />
Settele [1998] originally proposed only to consider<br />
cells which may be potential habitat. Moreover,<br />
he suggested to approximate K i by the mean number<br />
of observed individuals. In this study, the relative<br />
occupancy frequency has been used, in order<br />
to ensure the comparability of the IFM-approaches.<br />
When extracting the ”real” value of the GBM extinction<br />
parameters, we scaled the number of immigrants<br />
in order to obtain a first approximation. This<br />
might be a problem, since the maximum number of<br />
recorded immigrants depends on the l<strong>and</strong>scape. Alternatively,<br />
one could use for the fit a second extinction<br />
parameter (note that then the number of parameters<br />
would be equal to the PMM). However, these<br />
modifications could resolve the essential deviations<br />
(but much more field work would be necessary).<br />
The cell length has been chosen in the size of the<br />
smallest habitat. A systematic investigation of the<br />
influence of the cell length would be of interest, of<br />
course. Nonetheless, the great deviations in the estimators<br />
elucidate severe disagreements in the underlying<br />
assumptions.<br />
6. CONCLUSIONS<br />
Testing the quality of metapopulation models is<br />
generally a difficult issue, because little or even no<br />
data are available. Highly specific models can be<br />
used to substitute missing ”real-world” data. This<br />
allows not only to parameterize single models. Additionally,<br />
different model architectures can be compared.<br />
This has exemplarily been demonstrated to<br />
a grid-based approach vs. the well-known patchmatrix<br />
model.<br />
It shows, that the GBM leads to worrying misestimations.<br />
However, conditions have been derived,<br />
under which the accuracy is in the same range as<br />
for the PMM. Apart from the possible dissection of<br />
natural habitats, the results of this study indicate<br />
another shortcoming of the grid-based approach,<br />
namely that there seems to be the need of a profound<br />
number of snapshot years to determine the carrying<br />
capacity of a cell. Regarding the PMM, the amount<br />
of snapshot years has been considered in the context<br />
whether the metapopulation has reached its quasiequilibrium<br />
[Moilanen, 2000].<br />
Concerning more general aspects, in many studies<br />
has been stated a gap between simple, highly aggregated<br />
models on the one h<strong>and</strong> <strong>and</strong> specific models<br />
on the other h<strong>and</strong>. The former are often analytically<br />
tractable due to their rather general assumptions<br />
about population dynamics (which are often<br />
simply ignored). Thus being parameter-sparse, they<br />
allow to give insight into elementary relationships<br />
of state variables. On the other h<strong>and</strong>, specific models<br />
need a lot of information about the species’ life<br />
cycle. This paper is situated at the edge of these<br />
model types, utilizing the different conceptual approaches<br />
<strong>and</strong> trying to make them more comparable.<br />
REFERENCES<br />
Czárán, T. Spatiotemporal models of population<br />
<strong>and</strong> community dynamics. Chapman & Hall,<br />
London, 1998.<br />
Hanski, I. A practical model of metapopulation dynamics.<br />
Journal of Animal Ecology, 63:151–162,<br />
1994.<br />
Hanski, I. Metapopulation Ecoloy. Oxford University<br />
Press, New York, 1999.<br />
Hanski, I. Spatially realistic theory of metapopulation<br />
ecology. Naturwissenschaften, 88:372–381,<br />
2001.<br />
Hilker, F. M. Parametrisierung von Metapopulationsmodellen.<br />
Diploma thesis, Department of<br />
Mathematics <strong>and</strong> Computer Science, University<br />
of Osnabrück, 2002.<br />
Levins, R. Some demographic <strong>and</strong> genetic consequences<br />
of environmental heterogeneity for biological<br />
control. Bulletin of the Entomological Society<br />
of America, 15:237–240, 1969.<br />
Moilanen, A. The equilibrium assumption in estimating<br />
the parameters of metapopulation models.<br />
Journal of Animal Ecology, 69:143–153, 2000.<br />
Settele, J. Metapopulationsanalyse auf Rasterdatenbasis.<br />
Teubner Verlag, Leipzig, Stuttgart,<br />
1998.<br />
Ueberhuber, C. W. Numerical Computation 2.<br />
Springer, Berlin, 1997.<br />
907
Simulation of Dynamic Tree Species Patterns in the Alpine<br />
Region of Valais (Switzerl<strong>and</strong>) during the Holocene<br />
Heike Lischke<br />
Spatio-temporal L<strong>and</strong>scape <strong>and</strong> Forest <strong>Modelling</strong>; Swiss Federal Research Institute (WSL); Zürcherstrasse<br />
111; CH-8903 Birmensdorf, Switzerl<strong>and</strong>; lischke@wsl.ch<br />
Abstract: The spatio temporal forest l<strong>and</strong>scape model TreeMig is presented. It is based on a forest dynamics<br />
model that incorporates spatial variability by frequency distributions for tree densities <strong>and</strong> light intensities. It<br />
also takes into account seed production, intra-specific density regulation <strong>and</strong> seed dispersal. As a case study,<br />
the change of the tree species patterns in the Central-Alpine region of Valais was simulated with the model,<br />
using climate anomaly <strong>and</strong> immigration scenarios. The simulations were run on a grid with 1km x 1km<br />
resolution <strong>and</strong> with a yearly time step during the Holocene in the highly structured <strong>and</strong> heterogeneous<br />
environment of the valley of Valais in the Alps. As input, a scenario of temperature anomalies in the<br />
Holocene, spatially interpolated climate data, <strong>and</strong> times of species immigration into the simulation area were<br />
used. The results show a vivid pattern of species spread, changes of dominance, <strong>and</strong> up <strong>and</strong> down shifts of<br />
the timberline, which are triggered by the variability of the external factors but exhibit endogenous dynamics,<br />
i.e. migration <strong>and</strong> succession, after drastic changes of the boundary conditions, such as immigration of<br />
species into the simulation area or strong climate changes. However, on the observation scale no purely<br />
endogenous effects such as pattern formation can be observed.<br />
Keywords: Spatially explicit model; spatially linked model; process- model; tree species migration; seed<br />
dispersal; TreeMig; l<strong>and</strong>scape pattern<br />
1. INTRODUCTION<br />
Spatio-temporal patterns in a l<strong>and</strong>scape can arise from<br />
purely endogenous processes, like feedbacks <strong>and</strong><br />
nonlinear interactions, such as shown in many studies<br />
with simple abstract models. In nature, however,<br />
ecosystems are influenced by external factors <strong>and</strong> their<br />
heterogeneity, <strong>and</strong> it is not clear how in such natural<br />
systems external forcing <strong>and</strong> intrinsic dynamics<br />
interact, i.e. in which situations patterns are<br />
predominantly formed endogenously, <strong>and</strong> when<br />
patterns are determined mainly by the environment.<br />
One example for an ecological process exhibiting<br />
complex l<strong>and</strong>scape patterns, is the spatio-temporal<br />
vegetation development since the last glacial, such as<br />
preserved in pollen records. Such vegetation changes<br />
are interesting, since in the context of present <strong>and</strong><br />
anticipated future rapid climate changes, the question<br />
of whether ecosystems are able to follow fast enough<br />
to shifts in the local site conditions is crucial [see e.g.,<br />
Kirschbaum <strong>and</strong> Fischlin, 1996]. How will for<br />
instance tree species spread in a structured<br />
environment, such as the European Alps under<br />
changing environmental conditions? Particularly the<br />
potentially delayed immigration of species to new<br />
habitats is important. Such a migrational lag can cause<br />
a considerable delay of natural aforestation beyond the<br />
current timberlines <strong>and</strong> thus of carbon sequestration,<br />
as compared to the assumption that the species are<br />
already present <strong>and</strong> only limited by low temperatures.<br />
Tree species migration is influenced by a variety of<br />
factors <strong>and</strong> interacting processes: by the spatial<br />
interaction through seed dispersal limiting the<br />
maximum speed of spread, by the nonlinear dynamics<br />
of the local forest communities with processes such<br />
as growth, birth, <strong>and</strong> death, interactions like<br />
(hierarchical) competition or self regulation through<br />
shading, by environmental factors influencing all<br />
these processes, <strong>and</strong> by the heterogeneity of this<br />
environment. Due to these complicated interactions,<br />
the processes <strong>and</strong> patterns cannot be studied by<br />
analyzing the pollen data alone. Models are required<br />
which incorporate the essential processes,<br />
interactions <strong>and</strong> dependencies on environmental<br />
factors of tree migrations to underst<strong>and</strong> the past <strong>and</strong><br />
to assess the future response of vegetation to climate<br />
change.<br />
In this study, we present the model TreeMig, which is<br />
suitable to simulate the migration of tree species in a<br />
structured environment <strong>and</strong> under changing<br />
environmental conditions. In a regional case study in<br />
the European Alps during the Holocene, the patterns<br />
resulting from simulations with this model are<br />
evaluated with respect to internal processes <strong>and</strong><br />
external forcing.<br />
2. MATERIAL AND METHODS<br />
908
2.1. The spatio-temporal tree migration<br />
model TreeMig<br />
The model is originally based on the forest gap model<br />
ForClim [Bugmann, 1994, Bugmann, 1996], in which<br />
birth, growth <strong>and</strong> death of individual trees of many<br />
different species are followed on a set of small<br />
patches. The process functions depend on light,<br />
climate <strong>and</strong> other site conditions. Since birth <strong>and</strong><br />
mortality are stochastic in these models <strong>and</strong> the<br />
simulated areas <strong>and</strong> subpopulations are small the<br />
dynamics on the different patches differ. This results<br />
in a horizontal structure. The differential growth of the<br />
individuals results in a certain vertical structure. Since<br />
the fate of single trees is followed <strong>and</strong> many replicates<br />
of the stochastic dynamics have to be calculated to<br />
obtain reliable mean values, gap models are very<br />
computing time consuming, <strong>and</strong> thus not suitable for<br />
large-scale applications. Therefore, ForClim was<br />
aggregated to the distribution based, height structured<br />
population model DisCForM [Lischke, Löffler et al.,<br />
1998b, Löffler <strong>and</strong> Lischke, 2001]. DisCForM<br />
describes the dynamics of the population densities of<br />
trees in each tree height class. The variability from<br />
patch to patch is described by assuming the trees in<br />
each height class to be r<strong>and</strong>omly distributed over the<br />
patches, which results in a Poisson distribution of the<br />
tree population densities. From this distribution,<br />
frequency distributions of the light intensity <strong>and</strong> of the<br />
light dependent establishment, growth <strong>and</strong> death rates<br />
are calculated. By this, competition through shading<br />
<strong>and</strong> its spatial variability is included. This approach<br />
produces a purely deterministic description of the<br />
dynamics, which still reflects the variability in a<br />
forest, but is much faster than the stochastic gap<br />
model.<br />
To obtain the migration model TreeMig [Lischke <strong>and</strong><br />
Löffler, 2004], DisCForM has been implemented on a<br />
grid of square grid cells of 1km x 1km. TreeMig<br />
simulates explicitly seed production, seed dispersal<br />
<strong>and</strong> the development of seedlings/saplings. The<br />
number of seeds produced per year by each tree<br />
depends on its height, species <strong>and</strong> mast seeding<br />
period. The seed inflow into a cell is then defined by<br />
the seeds produced in all other cells multiplied with<br />
the dispersal kernel, which is a distance dependent<br />
probability density function, given by the combination<br />
of two negative exponentials (exp(-x/α)) for short<strong>and</strong><br />
long-distance transport. The values of α range<br />
from 25 m <strong>and</strong> 200 m. The seeds arriving at a cell<br />
build up the local seed bank, which is decreased by<br />
loss of germinability, predation <strong>and</strong> germination. Tests<br />
with several formulations of the reproduction<br />
submodel revealed that it was necessary to limit the<br />
seed number of each species separately to get a<br />
realistic biodiversity [Lischke <strong>and</strong> Löffler, 2004]. This<br />
was achieved by introducing an intra-specific<br />
competition term or species-specific antagonists (e.g.<br />
seed predators or pathogens). The seedlings<br />
germinating from the seed bank add to the saplings,<br />
which grow <strong>and</strong> die similarly to the adult trees. The<br />
local behavior of TreeMig has been tested against<br />
Swiss National Forest Inventory data representing<br />
different climatic conditions <strong>and</strong> st<strong>and</strong> ages [Bolliger<br />
<strong>and</strong> Lischke, 2004]. Parameters for the reproduction<br />
model have been compiled from various sources<br />
[Lischke <strong>and</strong> Löffler, 2004]. The parameter limiting<br />
species-specific seedling numbers was fitted by<br />
eyeball to a subset of the data. The overall<br />
correspondence between model <strong>and</strong> the entire data set<br />
was satisfying for most species <strong>and</strong> conditions.<br />
2.2. Input data<br />
The study area (Figure 1) encompasses the central<br />
Alpine region of Valais, which spans a large range of<br />
environmental conditions, altitudes from 400 m to<br />
4000 m, yearly mean temperatures between –1 °C<br />
<strong>and</strong> 11 °C, <strong>and</strong> yearly precipitation sums between<br />
350 mm in the eastern parts of the valley <strong>and</strong> 2000<br />
mm in the high altitudes. The central Alps separate<br />
the main valley from the glacial refuges of many<br />
species in the south <strong>and</strong> east. This region has defined<br />
paths where species could immigrate, namely the<br />
northern opening of the valley <strong>and</strong> several lower<br />
mountain passes in the southeast. An area of 50 km*<br />
100 km was chosen, simulation were carried through<br />
on 1kmx1km grid cells, from the end of the last<br />
glacial about 14000 before present (BP) to present.<br />
Four test sites were chosen to evaluate the temporal<br />
simulation pattern, characterized by their westerneastern<br />
position <strong>and</strong> their altitude. Site 2 is situated in<br />
the central part of the valley bottom (670 m), site 1<br />
at the timberline (2390 m), site 3 below timberline<br />
(1990 m) close to the Simplon pass, an site 4 in<br />
medium altitudes (1475 m).<br />
Figure 1: Simulation area in the Valais, Switzerl<strong>and</strong>. The<br />
numbers refer to selected sites, where temporal courses of<br />
the simulation are evaluated.<br />
The model uses as input the bioclimatic variables day<br />
degree sum (above 5.5 °C), lowest monthly<br />
temperature mean <strong>and</strong> a drought stress index<br />
(between 0 <strong>and</strong> 1). They were derived according to<br />
the model ForClim-E [Bugmann <strong>and</strong> Cramer, 1998]<br />
for each cell in the simulation area <strong>and</strong> each year in<br />
the simulation period. The bioclimatic variables were<br />
based on monthly temperatures <strong>and</strong> precipitations<br />
interpolated from climate station values <strong>and</strong> on a<br />
temperature anomaly scenario, reconstructed from<br />
chironomids in an Alpine lake [cf fig. bottom, Heiri,<br />
909
Lotter et al., 2003]. The species were assumed to<br />
immigrate from the Northwest, i.e. from the lake<br />
Geneva <strong>and</strong> from the Southeast over the lowest pass,<br />
i.e. the Simplon (2000 m altitude). In the simulation,<br />
at 14000 BP no trees were present. The approximate<br />
immigration years for the species arriving from the<br />
North were assessed from pollen records from the<br />
pollen database for the European Alps<br />
13900 BP<br />
8600 BP<br />
13800 BP<br />
7800 BP<br />
12000 BP<br />
7000 BP<br />
11000 BP<br />
6000 BP<br />
² T (°C)<br />
Tilia platyphyllos<br />
Tilia cordata<br />
Salix alba<br />
Quercus robur<br />
Quercus pubescens<br />
Quercus petraea<br />
Populus tremula<br />
Populus nigra<br />
Fagus silvatica<br />
Betula pendula<br />
Acer pseudoplatanus<br />
Acer platanoides<br />
Acer campestre<br />
Pinus silvestris<br />
Pinus cembra<br />
Picea excelsa<br />
Larix decidua<br />
Abies alba<br />
14000 12000 10000 8000 6000 4000 2000 0<br />
Simulation time (years BP)<br />
Figure. 2: TreeMig simulation of tree species spread on a 1 km * 1 km grid over 100 km * 50 km in the region of<br />
Valais, Switzerl<strong>and</strong>. In each cell the species biomasses (t/ha) are drawn as stacked columns. A completely filled<br />
cell corresponds to 450 t/ha total biomass. The graph at the bottom indicates the assumed temperature anomaly<br />
along with the times corresponding to the maps shown above (vertical lines).<br />
910
[van der Knaap <strong>and</strong> Ammann, 1997] from the<br />
southern part of the Swiss Central Plateau; those of the<br />
species immigrating over the Simplon were assessed<br />
from a pollen record from Simplon.<br />
Simulation experiments<br />
The simulations were run with the described time<br />
series of bioclimate <strong>and</strong> species immigration. Speciesspecific<br />
biomasses were stored for every cell <strong>and</strong><br />
every 200 years <strong>and</strong> displayed as a movie <strong>and</strong> as a<br />
series of maps. To separate the influence of<br />
succession <strong>and</strong> migration (intrinsic dynamics) from<br />
that of the environment, at selected timepoints the<br />
model was run into equilibrium keeping the bioclimate<br />
constant at the values of these times. All species<br />
which had immigrated before into the simulation area<br />
were allowed to be present everywhere. The<br />
simulated equilibrium species biomasses eq x,y,s were<br />
compared to the species biomasses y x,y,s with transient<br />
climate <strong>and</strong> migration with a similarity index<br />
S =1− ∑ y x,y,s<br />
−eq x,y,s ∑ (y x,y,s<br />
+ eq x,y,s<br />
).<br />
x,y,s<br />
3. RESULTS<br />
x,y,s<br />
The spatial patterns resulting from the simulations are<br />
shown as maps of species composition at selected<br />
time points (Figure. 2), the similarity to the<br />
equilibrium composition in Table 1. During the initial<br />
colonization (14000 to 13800 BP), fast migrating<br />
species such as birch (Betula pendula) <strong>and</strong> aspen<br />
(Populus tremula) spread rapidly, poplar (Populus<br />
nigra), pine (Pinus sylvestris), Swiss stone pine<br />
(Pinus cembra) <strong>and</strong> larch (Larix decidua) (from the<br />
East) follow slower. The vegetation is far from<br />
equilibrium (Table 1, 13900, 13800). At 12000 BP,<br />
the low <strong>and</strong> medium elevations of the central valleys<br />
are completely populated, in the valley by poplar <strong>and</strong><br />
pine, at the slopes by larch <strong>and</strong> at the timberline by<br />
Swiss stone pine, which drives back larch from the<br />
east <strong>and</strong> from the west. At 11000 BP, the timberline<br />
has shifted upwards, Swiss stone pine has mostly<br />
displaced larch, a small population of fir has<br />
developed south of the Simplon-pass, <strong>and</strong> in the<br />
eastern part of the valley, which is characterized by<br />
low precipitation sums, pine dominates. The<br />
vegetation resembles more to the equilibrium<br />
composition but still migration goes on. Until 8000<br />
BP, maple has spread far through of the valley, oak<br />
follows slower (9600 BP, 8000 BP). They push pine<br />
back to the very dry areas. A few firs have passed the<br />
Simplon-pass, <strong>and</strong> spread in the valley <strong>and</strong> at the<br />
Years BP Similarity index<br />
13900 0.16<br />
13800 0.24<br />
12000 0.48<br />
11000 0.54<br />
9600 0.56<br />
8000 0.60<br />
6600 0.61<br />
5200 0.69<br />
Table 1 : Similarity between simulated biomasses<br />
in transient simulation (Figure. 2) <strong>and</strong> equilibrium<br />
simulation .<br />
slopes from the east. Spruce (Picea abies) <strong>and</strong> beech<br />
(Fagus sylvatica) enter the valley in the north. At<br />
6600 BP, spruce <strong>and</strong> beech have spread from the<br />
west. At their eastern limit, still many species<br />
coexist. At 5200 BP, spruce dominates in the region.<br />
It has outcompeted fir, oak, <strong>and</strong> beech at the medium<br />
altitudes, <strong>and</strong> pushed Swiss stone pine back to high<br />
elevations. At this time the simulated vegetation is<br />
close to equilibrium which itself is determined by the<br />
climate.<br />
Figure 3 shows the temporal pattern of the species<br />
composition at the four selected sites. The long-term<br />
climate pattern is manifested most strongly at the<br />
high timberline site 1, by the slow initial increase of<br />
biomass at 11000 BP <strong>and</strong> by the gaps at about 8000<br />
BP <strong>and</strong> between 4000 <strong>and</strong> 1500 BP. Species<br />
biomasses fluctuate at all sites, which reflects the<br />
temperature fluctuations given by the climate<br />
anomaly but also by the stochastic climate generator.<br />
However, the fluctuations are strongest at the borders<br />
of the distribution ranges of the species, such as at<br />
the high timberline site (1), or at the higher border of<br />
the distribution range of pine (site 2) or spruce (site 3,<br />
after 7000 BP). At the lower timberline site (3), the<br />
fluctuations are small until spruce appears, which<br />
competes with Swiss stone pine in warm periods,<br />
resulting in strong fluctuations also in this species.<br />
Such switches between two competing species (Swiss<br />
stone pine <strong>and</strong> pine) appear also at site 4, where<br />
Swiss stone pine is driven by the fluctuations of pine.<br />
The immigration of species is reflected mainly by the<br />
appearance of spruce. Larch <strong>and</strong> Swiss stone pine<br />
seem to appear later in the west (site 1) than in the<br />
east (site 3). However, the spatio-temporal<br />
simulations, particularly the movie, reveal that both<br />
species are present in the region of site 1 already<br />
around 12600 BP. Thus, this apparent migration lag<br />
is due to too cold temperatures.<br />
911
4. DISCUSSION<br />
The large-scale spatio-temporal pattern is dominated<br />
by three aspects:<br />
1) the initial colonization of the empty habitat, 2) the<br />
immigration waves of the various species with<br />
intermediary species intermingling <strong>and</strong> outcompeting<br />
of residents, if the immigrants such as spruce or beech<br />
are dominant, <strong>and</strong> 3) the spatial separation of the<br />
species according to the environmental conditions.<br />
The relative importance of these factors differs<br />
between times <strong>and</strong> locations. The influence of the<br />
spatio-temporal pattern of the environmental factors is<br />
especially strong at the borders of species ranges. The<br />
environment forms the stage for the endogenous<br />
dynamics, i.e. migration <strong>and</strong> competition, which play<br />
particularly in transient phases after drastic changes of<br />
the boundary conditions, i.e. immigrations.<br />
At the studied scale (grain = 1 km * 1 km) no<br />
endogenous pattern formation can be observed. This is<br />
partly because the resolution is too coarse with respect<br />
to the interaction ranges <strong>and</strong> the gradients in the<br />
environmental variables. Furthermore, the<br />
simulations are rarely in equilibrium due to the<br />
constantly changing climate <strong>and</strong> the pulses of<br />
immigration.<br />
The simulations demonstrate that single local data<br />
sets, such as pollen records are difficult to interpret<br />
with respect to spatio-temporal patterns: The clear<br />
pattern of spread in the spatial simulations is hardly<br />
detectable in the single site trajectories. Taking<br />
further into account the uncertainties in pollen data,<br />
climate scenario <strong>and</strong> immigration scenarios related to<br />
dating <strong>and</strong> interpretation [Lischke, Guisan et al.,<br />
1998a] it becomes even more evident that an analysis<br />
of the processes in the past leading to such data sets<br />
is impossible without models – <strong>and</strong> that for a<br />
thorough model testing many pollen data sets are<br />
required which additionally stem from a less complex<br />
area than the Alps.<br />
The presented simulations are a first step towards<br />
analyzing past <strong>and</strong> assessing future tree species<br />
migrations under a changing climate. The presented<br />
simulations serve to demonstrate the generic behavior<br />
of the model in quasi-realistic situations. However,<br />
some traits of the simulations are rather unrealistic,<br />
1<br />
450<br />
3<br />
350<br />
450<br />
350<br />
250<br />
250<br />
150<br />
150<br />
50<br />
50<br />
Simulated biomass (t/ha)<br />
14000 12000 10000 8000 6000 4000 2000 0 -5014000<br />
2 450 4<br />
350<br />
250<br />
150<br />
12000<br />
10000<br />
8000<br />
6000<br />
4000<br />
2000<br />
0 -50<br />
450<br />
350<br />
250<br />
150<br />
50<br />
50<br />
14000<br />
12000<br />
10000<br />
8000<br />
6000<br />
4000<br />
14000 12000 10000 8000<br />
2000 0<br />
-50<br />
Simulation time (years BP)<br />
6000<br />
4000<br />
2000<br />
0 -50<br />
Abies alba Larix decidua Picea excelsa Pinus cembra<br />
350<br />
Pinus silvestris Acer campestre Acer plat./pseudoplat. Betula pendula<br />
150<br />
Fagus silvatica Populus Quercus Salix alba<br />
-50<br />
-12000 Tilia cordata -10000 -8000 Tilia -6000 platyphyllos -4000 -2000 0<br />
Figure 3: Simulated species biomasses at selected sites with different altitudes (1 <strong>and</strong> 3 close to timberline, 2<br />
low, 4 medium) <strong>and</strong> different longitudes (1,2 west, 3,4 east).<br />
912
e.g. the overwhelming dominance of spruce. The<br />
simulated species composition for current conditions<br />
resembles that at 5200 BP, i.e. is dominated by spruce,<br />
oak, <strong>and</strong> beech in the low altitudes, <strong>and</strong> pine at the<br />
very dry sites. The current forests in Valais such as<br />
recorded in the Swiss National Forest Inventory<br />
[EAFV, 1988], have much less spruce <strong>and</strong> more fir<br />
<strong>and</strong> larch. This deviation is probably to a large extent<br />
due to the oversimplified climate change scenario we<br />
used, which assumes that precipitation was the same<br />
as today, whereas it was probably considerably drier<br />
in the late glacial <strong>and</strong> early Holocene [Guiot, Harrison<br />
et al., 1993]. In future simulations, various<br />
combinations of climate anomaly scenarios shall be<br />
tested.<br />
5. REFERENCES<br />
Bolliger, J. <strong>and</strong> Lischke, H., Sensitivity analysis <strong>and</strong><br />
evaluation of the spatio-temporal forest l<strong>and</strong>scape<br />
model TREEMIG, Ecological <strong>Modelling</strong> , in<br />
prep., 2004.<br />
Bugmann, H., On the ecology of mountainous forests<br />
in a changing climate: A simulation study,<br />
Dissertation, Swiss Federal Institute of Technology<br />
Zurich, Zurich, 1994.<br />
Bugmann, H. <strong>and</strong> Cramer, W., Improving the<br />
behaviour of forest gap models along drought<br />
gradients, Forest Ecology <strong>and</strong> Management 103<br />
(2-3), 247-263, 1998.<br />
Bugmann, H. K. M., A simplified forest model to<br />
study species composition along climate gradients,<br />
Ecology 77 (7), 2055-2074, 1996.<br />
EAFV, Schweizerisches L<strong>and</strong>esforstinventar.<br />
Ergebnisse der Erstaufnahme 1982-1986. ,<br />
Berichte Eidgenössische. Forschungsanstalt für<br />
Wald, Schnee und L<strong>and</strong>schaft, Vol. 305, pp. 375.<br />
Eidgenössische Anstalt für das forstliche<br />
Versuchswesen in Zusammenarbeit mit dem<br />
Bundesamt für Forstwesen und L<strong>and</strong>schaftsschutz,<br />
Birmensdorf, 1988.<br />
Guiot, J., Harrison, S. P. <strong>and</strong> Prentice, I. C.,<br />
Reconstruction of Holocene precipitation patterns<br />
in Europe using pollen <strong>and</strong> lake-level data,<br />
Quaternary Research 40 , 139-149, 1993.<br />
Heiri, O., Lotter, A. F., Hausmann, S. <strong>and</strong> Kienast, F.,<br />
A chironomid-based Holocene summer air<br />
temperature reconstruction from the Swiss Alps,<br />
The Holocene 13 (4), 477-484, 2003.<br />
Kirschbaum, M. <strong>and</strong> Fischlin, A., Climate change<br />
impacts on forests. In: R. T. Watson, M. C.<br />
Zinyowera<strong>and</strong> R. H. Moss (Eds.), Climate change<br />
1995 - Impacts, adaptations <strong>and</strong> mitigation of<br />
climate change: Scientific-technical analyses:<br />
Contribution of working group II to the second<br />
assessment report of the intergovernmental panel<br />
on climate change, Cambridge University Press ,<br />
pp. 95-131, 1996.<br />
Lischke, H., Guisan, A., Fischlin, A., Williams, J.<br />
<strong>and</strong> Bugmann, H., Vegetation responses to<br />
climate change in the Alps - Modeling studies. In:<br />
P. Cebon, U. Dahinden, H. Davies, D.<br />
Imboden<strong>and</strong> C. Jaeger (Eds.), A view from the<br />
Alps: Regional perspectives on climate change,<br />
MIT Press , pp. 309-350, 1998a.<br />
Lischke, H. <strong>and</strong> Löffler, T., Intra-specific density<br />
dependence required to maintain diversity in a<br />
spatio-temporal forest model with reproduction,<br />
OIKOS , (submitted), 2004.<br />
Lischke, H., Löffler, T. J. <strong>and</strong> Fischlin, A.,<br />
Aggregation of individual trees <strong>and</strong> patches in<br />
forest succession models - Capturing variability<br />
with height structured r<strong>and</strong>om dispersions,<br />
Theoretical Population Biology 54 (3), 213-226,<br />
1998b.<br />
Löffler, T. J. <strong>and</strong> Lischke, H., Incorporation <strong>and</strong><br />
influence of variability in an aggregated forest<br />
model, Natural Resource Modeling 14 (1), 103-<br />
137, 2001.<br />
van der Knaap, W. O. <strong>and</strong> Ammann, B., Depth-age<br />
relationships of 25 well-dated Swiss Holocene<br />
pollen sequences archived in the Alpine<br />
Palynological Data Base, Revue de Paléobiologie<br />
16 , 433-480, 1997.<br />
913
Aphid Population Dynamics in Agricultural L<strong>and</strong>scapes:<br />
An Agent-based Simulation Model<br />
Hazel Parry 1, 2 , Andrew J Evans 1 , Derek Morgan 2<br />
1 School of Geography, University of Leeds, LS2 9JT, Engl<strong>and</strong> (h.parry@geog.leeds.ac.uk)<br />
2 Central Science Laboratory, S<strong>and</strong> Hutton, York, YO41 1LZ, Engl<strong>and</strong><br />
Abstract: Presently, there are many population models in existence, but these are often case specific,<br />
function at a single spatial scale <strong>and</strong> fail to tackle the complexity arising from individual actions <strong>and</strong><br />
interactions that exist in the real-world. A spatially explicit agent-based simulation model has been<br />
developed to represent aphid population dynamics in agricultural l<strong>and</strong>scapes. Over time, the aphid agents<br />
interact with the l<strong>and</strong>scape <strong>and</strong> with one another. The construction of the model is detailed, including<br />
parameterisation <strong>and</strong> coupling to a geographical information system (GIS). The results show that a spatial<br />
modelling approach that considers both l<strong>and</strong>scape properties <strong>and</strong> factors such as wind speed <strong>and</strong> direction<br />
provides greater insight into aphid population dynamics both spatially <strong>and</strong> temporally. This forms the basis<br />
for the development of further simulation models that can be used to analyse how changes in l<strong>and</strong>scape<br />
structure impact upon important species distributions <strong>and</strong> population dynamics.<br />
Keywords: Aphids; Agriculture; Agent-based <strong>Modelling</strong>; L<strong>and</strong>scape Ecology.<br />
1. INTRODUCTION<br />
Sixty percent of the British l<strong>and</strong>scape is farml<strong>and</strong>.<br />
Most of this has been intensively farmed, which<br />
has resulted in wildlife populations being highly<br />
fragmented <strong>and</strong> pest species controlled primarily<br />
by pesticides. However, it is quite possible to<br />
transform the l<strong>and</strong>scape so that it would be more<br />
beneficial to wildlife, <strong>and</strong> to find alternatives to<br />
high levels of chemical usage. The difficulty is to<br />
determine what would be the optimal way that<br />
would maximize desirable populations but<br />
minimize disruption to existing l<strong>and</strong> management<br />
practices.<br />
The creation of a generic agent-based insect<br />
simulation model for agricultural l<strong>and</strong>scapes will<br />
facilitate concurrent examination of the potential<br />
impacts of l<strong>and</strong>scape change upon populations of<br />
species of both agricultural <strong>and</strong> ecological<br />
interest. In this way more sensitive l<strong>and</strong>scape<br />
management can be achieved, through an<br />
underst<strong>and</strong>ing of the differing implications for a<br />
wide range of species of the introduction or<br />
removal of l<strong>and</strong>scape features or management<br />
regimes [Hunter, 2002].<br />
The final model is still in the development stage,<br />
but a single species simulation will be presented<br />
that illustrates the usage of the model to study the<br />
population dynamics of the bird cherry-oat aphid,<br />
Rhopalosiphum padi (L.), in a 5 × 5 km region of<br />
North Yorkshire. The model is termed ‘agentbased’,<br />
as the extent to which the individuals in<br />
the model react to their environment <strong>and</strong><br />
‘remember’ (physiologically) past events defines<br />
them as ‘agents’ [Topping et al., 2003].<br />
2. SIMULATION MODELLING OF<br />
SPECIES POPULATION<br />
DYNAMICS<br />
2.1 Introduction<br />
There is a tradition in ecology for models that are<br />
based upon mathematical ‘top-down’<br />
relationships between variables [Parrott et al.,<br />
2001]. This has meant many models prior to the<br />
1990s have focused on populations or species<br />
groups, rather than individual animals. However,<br />
such models do not take into account the<br />
complexity of the multiple concurrent interactions<br />
in ecosystems [Laval, 1996]. By ignoring<br />
individual behaviour, important factors are not<br />
914
taken into account, including reproduction <strong>and</strong><br />
competition between individuals, which may<br />
greatly influence general population trends.<br />
A significant need exists for ecological models to<br />
address real-world management problems, but the<br />
lack of transferability, scalability, complexity <strong>and</strong><br />
realism in traditional models <strong>and</strong> their uncertainty<br />
is a key issue [Conroy et al., 1995]. In order to<br />
produce models that are capable of furthering<br />
underst<strong>and</strong>ing of the processes that influence<br />
population dynamics spatially <strong>and</strong> temporally, as<br />
well as forecasting the effects of management or<br />
other human activity on population distributions,<br />
it has been necessary to change the way<br />
ecological systems are modelled over the last<br />
decade or so. Models have become more spatially<br />
explicit <strong>and</strong> attempts have been made to link these<br />
to real l<strong>and</strong>scapes via geographical information<br />
systems [DeAngelis et al., 1998].<br />
2.2 Agent-based <strong>Modelling</strong> in L<strong>and</strong>scape<br />
Ecology<br />
In agent-based models, individual insects are<br />
modelled as individuals (agents), with a unique<br />
history <strong>and</strong> the ability to interact both with the<br />
environment <strong>and</strong> with other agents. The inherent<br />
flexibility of an agent-based, object-oriented<br />
approach enables modellers to attempt to create<br />
more generic models [Ziv, 1998]. Multi-agent<br />
simulation also provides a framework that allows<br />
for interactions at different scales <strong>and</strong> the<br />
simulation of emergent ecosystem properties<br />
[Ferber, 1999]. The agents, their behaviour <strong>and</strong><br />
interactions, allow for realistic representation of a<br />
phenomenon as the result of the interactions of a<br />
group of autonomous agents. Multi-agent<br />
systems are also able to consider both quantitative<br />
<strong>and</strong> qualitative parameters, <strong>and</strong> have the capacity<br />
to integrate quantitative variables, differential<br />
equations <strong>and</strong> rule based behaviour into the same<br />
model. Modifications to the model are also quite<br />
straightforward (such as adding another species).<br />
The approach therefore helps in the search for a<br />
model, rather than simply model implementation<br />
<strong>and</strong> response analysis.<br />
However, despite the advantages, the use of<br />
agent-based modelling techniques in l<strong>and</strong>scape<br />
ecology is still a growing trend, with few<br />
examples of existing models to date [Mathevet et<br />
al., 2003; Parrott et al., 2001; Topping et al.,<br />
2003].<br />
3. AGENT-BASED SIMULATION OF<br />
BIRD-CHERRY OAT APHID<br />
(Rhopalosiphum padi (L.))<br />
POPULATION DYNAMICS IN AN<br />
AGRICULTURAL LANDSCAPE<br />
3.1 Model Description<br />
The model is written using the object-oriented<br />
programming language Java (http://java.sun.com)<br />
<strong>and</strong> the Repast agent-based modelling toolkit<br />
(http://repast.sourceforge.net). It is run in daily<br />
time steps. The key inputs are habitat data<br />
(derived from raster data of a chosen region,<br />
where size <strong>and</strong> extent are defined by the user),<br />
daily minimum, maximum <strong>and</strong> mean temperature,<br />
wind speed <strong>and</strong> wind direction (the latter are<br />
currently single values for prevailing wind).<br />
Classes that represent different species of insect<br />
structure the model, each hierarchically derived<br />
from an ‘Insect’ superclass. This paper focuses<br />
on the use of the model to simulate the spatial<br />
population dynamics of the bird cherry-oat aphid<br />
(Rhopalosiphum padi (L.)) during the autumn <strong>and</strong><br />
winter. Key information about any Insect agent<br />
includes a unique ID tag for the agent, the agent's<br />
‘age’ (0.00-2.00, becoming adult at 1.00) <strong>and</strong> the<br />
agent's position in three-dimensional space. In<br />
addition, for Aphid agents, information on<br />
whether or not the agent has undergone migration<br />
<strong>and</strong> the agent's morphology (alate or apterous) is<br />
also important.<br />
At each daily time step in a model run for the<br />
region the following events take place:<br />
• Adult alate aphid agents may immigrate into<br />
the region.<br />
• Alate aphids may move according to the<br />
wind speed, wind direction, habitat, <strong>and</strong> their<br />
development stage. This movement may be<br />
local foraging, or long distance migration.<br />
• Aphid agents age.<br />
• Aphid agents may die.<br />
• Adult aphid agents may reproduce <strong>and</strong> new<br />
agents may be born.<br />
3.2 Initial Immigration<br />
Before the simulation is started, initial<br />
immigration is input as a number of immigrants,<br />
which are then r<strong>and</strong>omly distributed across the<br />
region. For aphids, the immigrants are assumed<br />
to be reproductive alate adults, of uniform age.<br />
They are also assumed to have undergone<br />
'migration', thus will probably not have a desire to<br />
migrate long distances again [Kennedy et al.,<br />
1963].<br />
915
3.3 Reproduction<br />
Aphid agents become reproductive once the agent<br />
achieves the appropriate age for reproduction, for<br />
alate aphids this is 0.9522, for apterous this is<br />
0.9463. The birthrate depends on the morphology<br />
of the reproductive aphid, <strong>and</strong> the daily minimum,<br />
maximum <strong>and</strong> mean temperatures (for equations<br />
see [Morgan, 2000]).<br />
Nymphs are then located at the same location as<br />
their parent. The stimulus to produce alates<br />
capable of dispersal is related to crowding <strong>and</strong>/or<br />
tactile responses to the nutrient quality of the host<br />
[Loxdale et al., 1999]. The aphid density per m 2<br />
at the location nymphs are born therefore<br />
determines the morphology of the nymphs created<br />
(for equation see [Morgan, 2000]).<br />
3.4 Ageing <strong>and</strong> Mortality<br />
Aphid agents at any life-stage may die depending<br />
on a survival rate affected by the number of daydegrees<br />
below 2.8 0 C for the day. The survival<br />
rates of the aphid agents are calculated from the<br />
daily minimum, maximum <strong>and</strong> mean<br />
temperatures (for equation see [Morgan, 2000]).<br />
Other abiotic factors such as rainfall may be<br />
relevant [Morgan, 2000] as well as the effects of<br />
predation <strong>and</strong> parasites or fungi, but these are not<br />
included in the model as yet. Mortality also<br />
occurs when the aphid agents reach maximum age<br />
2.00 (the number of days that this will take<br />
depends again on temperatures, see below), <strong>and</strong><br />
when they remain on unfavourable habitat for<br />
more than three days (at present the absence of<br />
research in this area makes this an estimate of the<br />
agent's ability to survive poor conditions). The<br />
age of the aphid agent increases each day, at a rate<br />
determined by the daily temperatures (see<br />
[Morgan, 2000]).<br />
3.5 Movement<br />
The flight of alate aphids can be separated into<br />
two phases. The first is a migratory phase<br />
followed by a foraging phase [Kennedy et al.,<br />
1963; Moericke, 1955; Ward et al., 1998].<br />
The rules of migratory flight used in this model<br />
(Figure 1) follow four principles: firstly, alate<br />
aphids will all attempt to migrate voluntarily if<br />
wind speed is not above 8km/hr [Haine, 1955;<br />
Johnson, 1962; Kennedy et al., 1963]. Second,<br />
aphid migration will take place within a day <strong>and</strong><br />
during daylight hours (thus a migration event will<br />
complete within a single run of the model, as this<br />
functions on a daily basis) [Loxdale et al., 1993].<br />
Thirdly, an individual can only migrate a distance<br />
of several kilometres once (if at all) during its<br />
lifetime [Ward et al., 1998]. Finally, migration<br />
will last for a r<strong>and</strong>om duration of between 2.5 <strong>and</strong><br />
6.5 hours [Loxdale et al., 1993] during which time<br />
the aphid will be carried by the wind a distance<br />
determined by the flight duration multiplied by<br />
the wind speed, in the direction of the wind's<br />
movement [Haine, 1955; Loxdale et al., 1993]. It<br />
is also assumed that a ‘boundary layer’ at a height<br />
of 1m exists, below which the aphid is unaffected<br />
by the wind <strong>and</strong> free to move at will <strong>and</strong> above<br />
which the aphid’s movement is controlled by the<br />
wind [Taylor, 1974].<br />
Density <strong>and</strong><br />
resources<br />
No<br />
Is Aphid adult<br />
Alate?<br />
Does Aphid choose to takeoff?<br />
Yes<br />
Is aphid flight speed greater<br />
than wind speed?<br />
No<br />
Yes<br />
Yes<br />
Is wind too strong for takeoff?<br />
Does Aphid fly above<br />
the boundary layer?<br />
Aphid carried by wind in direction of wind at<br />
wind speed, for r<strong>and</strong>om distance relating to<br />
wind speed <strong>and</strong> average flight duration (2.5-<br />
6.5 hours). Then aphid descends rapidly<br />
below boundary layer<br />
No<br />
Yes<br />
No<br />
No<br />
Yes<br />
Figure 1: Flow diagram of movement rules<br />
Remains on plant<br />
Aphid flies r<strong>and</strong>omly<br />
according to perception<br />
of local resources (if<br />
good then remains in<br />
locality, if bad then flies<br />
further away) unaffected<br />
by wind<br />
Aphids loose control of their flight at wind speeds<br />
of around 2km/hr [Haine, 1955; Loxdale et al.,<br />
1993]. Thus it can be inferred that foraging flight<br />
may occur at low wind speeds (2km/hr or less),<br />
taking the form of increasingly 'r<strong>and</strong>om<br />
movement' as wind speeds lower, <strong>and</strong> short flights<br />
tend to be concentrated around host plants<br />
[Kennedy et al., 1959]. The speed of these<br />
movements is set to be the aphid maximum flight<br />
speed of 0.9m/s (3.24km/hr) [Compton, 2002].<br />
To obtain the distance flown this is then<br />
multiplied by the average flight time of an aphid,<br />
which is about 100-200 minutes [Lewis et al.,<br />
1965].<br />
916
4. SIMULATION RESULTS<br />
A simulation was run for the autumn <strong>and</strong> winter<br />
of 1985/86. An initial population of 10,000 alate<br />
aphids were distributed across a grid of 25m cells,<br />
in a region 5 km × 5 km. This grid was derived<br />
from an ASCII raster taken from a LCM2000<br />
dataset of Hertfordshire, Engl<strong>and</strong> (origin<br />
51°51'12"N, 0°19'37"W), with data on l<strong>and</strong> cover<br />
used in a GIS to identify areas of favourable <strong>and</strong><br />
unfavourable habitat. The population levels over<br />
time for the region are shown in Figure 2, <strong>and</strong> the<br />
spatial pattern of dispersal was observed (Figure<br />
3, <strong>and</strong> to be presented at the conference). There<br />
are two major population peaks, at day 313 <strong>and</strong><br />
day 357. Numbers reach their peak in early<br />
autumn due to the influx of alate immigrants. The<br />
second peak is lower due to lower temperatures as<br />
well as a lack of immigrants.<br />
b<br />
Figure 2: Mean density of aphids per occupied<br />
25m grid cell.<br />
c<br />
Figure 3: Spatial distribution of R. padi at a)<br />
julian day 0, b) julian day 10 <strong>and</strong> c) julian day 50<br />
showing the population dynamics as alates first<br />
move into favourable habitat, populations<br />
increase <strong>and</strong> then alates diffuse across the<br />
l<strong>and</strong>scape.<br />
4.1 Validation <strong>and</strong> Sensitivity Analysis<br />
a<br />
The model is validated against independent field<br />
data collected at plant scale (scaled to 1m 2 ,<br />
assuming 300 plants per m 2 ), Figure 4. Aphid<br />
densities are slightly over-predicted by the model,<br />
but follow a very similar trend; populations<br />
increased rapidly from very low numbers <strong>and</strong><br />
917
peaked around 40 days later. Thereafter numbers<br />
declined gradually, although aphids were present<br />
throughout the winter, albeit at low density. As<br />
the model presented here is developed further,<br />
more comprehensive, l<strong>and</strong>scape scale validation<br />
shall also be used.<br />
which could include hedgerow removal, l<strong>and</strong> use<br />
change or climate change amongst others. The<br />
use of an underlying cellular automata model or<br />
Monte Carlo simulation to represent this change<br />
may be necessary to model gradual spatial<br />
changes over time.<br />
Total number of Aphids per m 2<br />
10000<br />
1000<br />
100<br />
10<br />
Model<br />
Field<br />
It can be concluded from this study that<br />
significant progress has been made to establish an<br />
extendable <strong>and</strong> powerful l<strong>and</strong>scape model of<br />
insect population dynamics using agent-based<br />
simulation. Much work is still required to provide<br />
a tool that examines the effects of l<strong>and</strong>scape<br />
change on more than one species, but this study<br />
shows that useful insights into spatial <strong>and</strong><br />
temporal dynamics across spatial scales can be<br />
gained by the use of this model. It may<br />
eventually be possible to adapt this flexible model<br />
to simulate broader ecosystems including, for<br />
example, mammals or birds.<br />
1<br />
266 286 306 326 346 366 386<br />
Julian Day<br />
Figure 4: Simulation for single 1m 2 crop cell<br />
(solid line, with StDev) <strong>and</strong> observed (■) R. padi<br />
populations in 1985 at Rothampsted (data from<br />
Morgan, pers. comm.)<br />
Simple tests of the sensitivity of the model to<br />
several population processes were carried out.<br />
These were found to be similar to the sensitivity<br />
of the model developed by Morgan [2000], where<br />
mortality rates are a key influence on the<br />
population density <strong>and</strong> structure. For example, an<br />
increase in mortality of only 5% suppresses peak<br />
densities by at least five-fold.<br />
5. CONCLUSIONS<br />
Three major challenges for the model now exist.<br />
The model will need to h<strong>and</strong>le realistic aphid<br />
densities across larger regions, which will<br />
increase run-time <strong>and</strong> computational power<br />
required. Millions of aphids may come from<br />
heavily infested crops [Johnson, 1962]. One<br />
solution is to parallelise the model, or to<br />
implement scaling solutions such as 'superindividuals'<br />
[Scheffer et al., 1995].<br />
The second challenge is to add more insect<br />
species. This includes the addition of predators or<br />
parasites to control aphid populations, as well as<br />
the introduction of insects of conservation value.<br />
The third is to more tightly couple the model to<br />
the GIS in order to examine the impacts of<br />
l<strong>and</strong>scape change upon the insect populations,<br />
6. ACKNOWLEDGEMENTS<br />
This work is part of a PhD funded by the Central<br />
Science Laboratory. Thanks to Daniel Parry for<br />
his advice on Java programming.<br />
7. REFERENCES<br />
Compton, S., Sailing with the wind: dispersal by<br />
small flying insects, Chapter 6 in<br />
Dispersal Ecology, Bullock, J.M., R.E.<br />
Kenward, R.S. Hails, Blackwell<br />
Science, Oxford, 2002.<br />
Conroy, M.J., Y. Cohen, F.C. James, Y.G.<br />
Matsinos, <strong>and</strong> B.A. Maurer, Parameter<br />
estimation, reliability, <strong>and</strong> model<br />
improvement for spatially explicit<br />
models of animal populations,<br />
Ecological Applications, 5(1), 17-19,<br />
1995.<br />
DeAngelis, D.L., L.J. Gross, M.A. Huston, W.F.<br />
Wolff, D.M. Flemming, E.J. Comiskey,<br />
<strong>and</strong> S.M. Sylvester, L<strong>and</strong>scape<br />
modeling for everglades ecosystem<br />
restoration, Ecosystems, 1, 164-45,<br />
1998.<br />
Ferber, J., Multi-agent systems: an introduction to<br />
distributed artificial intelligence,<br />
Pearson Education, Harlow, 1999.<br />
Haine, E., Aphid take-off in controlled wind<br />
speeds, Nature, 175, 474-475, 1955.<br />
918
Hunter, M.D., L<strong>and</strong>scape structure, habitat<br />
fragmentation <strong>and</strong> the ecology of<br />
insects, Agricultural <strong>and</strong> Forest<br />
Entomology, 41, 59-166, 2002.<br />
Johnson, C.G., Aphid migration, New Scientist,<br />
305, 622-625, 1962.<br />
Kennedy, J.S. <strong>and</strong> C.O. Booth, Free flight of<br />
aphids in the laboratory, Journal of<br />
Experimental Biology, 40, 67-85, 1963.<br />
Kennedy, J.S., C.O. Booth, <strong>and</strong> W.J.S. Kershaw,<br />
Host finding by aphids in the field II.<br />
Aphis fabae Scop. (gynoparae) <strong>and</strong><br />
Brevicoryne brassicae L.; with a reappraisal<br />
of the role of host-finding<br />
behaviour in virus spread, Annals of<br />
Applied Biology, 47(3), 424-444, 1959.<br />
Laval, Ph., The representation of space in an<br />
object-oriented computational pelagic<br />
ecosystem, Ecological <strong>Modelling</strong>, 88<br />
113-124, 1996.<br />
Lewis, T. <strong>and</strong> L.R. Taylor, Diurnal perodicity of<br />
flight by insects, Transactions of the<br />
Royal Entomological Society London,<br />
116(15), 393-479, 1965.<br />
Loxdale, H.D., J. Hardie, S. Halbert, R. Foottit,<br />
N.A.C. Kidd, <strong>and</strong> C.I. Carter, The<br />
relative importance of short <strong>and</strong> longrange<br />
movement of flying aphids,<br />
Biological Review, 68, 291-311, 1993.<br />
Loxdale, H.D. <strong>and</strong> G. Lushai, Slaves of the<br />
environment: the movement of<br />
herbivorous insects in relation to their<br />
ecology <strong>and</strong> genotype, Philosophical<br />
Transactions of the Royal Society<br />
London B, 354, 1479-1498, 1999.<br />
Mathevet, R., F. Bousquet, C. Le Page, <strong>and</strong> M.<br />
Antona, Agent-based simulations of<br />
interactions between duck population,<br />
farming decisions <strong>and</strong> leasing of<br />
hunting rights in the Carmargue<br />
(southern France), Ecological<br />
<strong>Modelling</strong>, 165, 107-126, 2003.<br />
Morgan, D., Population dynamics of the bird<br />
cherry-oat aphid, Rhopalosiphum padi<br />
(l.), during the autumn <strong>and</strong> winter: a<br />
modelling approach, Agricultural <strong>and</strong><br />
Forest Entomology, 2, 297-304, 2000.<br />
Parrott, L. <strong>and</strong> R. Kok, Use of an object-based<br />
model to represent complex features of<br />
ecosystems, InterJournal of Complex<br />
Systems, Manuscript 37, 2001.<br />
Scheffer, M., J.M.Baveco, DeAngelis, D.L., Rose,<br />
K.A. <strong>and</strong> van Nes, H., Superindividuals<br />
a simple solution for<br />
modeling large populations on an<br />
individual basis, Ecological <strong>Modelling</strong>,<br />
80, 161-170.<br />
Taylor, L.R., Insect migration, flight perodicity<br />
<strong>and</strong> the boundary layer, Journal of<br />
Animal Ecology, 1974.<br />
Taylor, L.R. <strong>and</strong> J.M.P. Palmer, Aerial sampling,<br />
Chapter 5 in Aphid Technology, van<br />
Emden, H.F., Academic Press, London,<br />
189-234, 1972.<br />
Topping, C.J., T.S. Hansen, T.S. Jensen, J.U<br />
Jepsen, F. Nikolajsen, <strong>and</strong> P.<br />
Odderskær, ALMASS, an agent-based<br />
model for animals in temperate<br />
european l<strong>and</strong>scapes, Ecological<br />
<strong>Modelling</strong>, 167, 65-82, 2003.<br />
Ward, S.A., S.R. Leather, J. Pickup, <strong>and</strong> R.<br />
Harrington, Mortality during dispersal<br />
<strong>and</strong> the cost of host specificity in<br />
parasites: how many aphids find hosts?,<br />
Journal of Animal Ecology, 67, 763-<br />
773, 1998.<br />
Ziv, Y., Effects of environmental heterogeneity<br />
on species diversity: a new processbased,<br />
multi-species, l<strong>and</strong>scape<br />
simulation model (SHALOM), The<br />
University of Arizona, 1998.<br />
Moericke, B.V., Über die lebensgewohnheiten der<br />
geflügelten blattläuse (aphidina) unter<br />
besonderer berücksichtigung des<br />
verhaltens beim l<strong>and</strong>en, Zeitschrift für<br />
Angew<strong>and</strong>te Entomologie, 37, 29-91,<br />
1955.<br />
919
Integrating Wetl<strong>and</strong>s <strong>and</strong> Riparian Zones in Regional<br />
Hydrological Modeling<br />
F.F. Hattermann, V. Krysanova, A. Habeck<br />
Potsdam Institute for Climate Impact Research (PIK), hattermann@pik-potsdam.de<br />
Abstract: Wetl<strong>and</strong>s, <strong>and</strong> in particular riparian wetl<strong>and</strong>s, are at the interface between well drained l<strong>and</strong> <strong>and</strong> the<br />
aquatic environment, where they control the exchange of water <strong>and</strong> related chemical fluxes from catchment areas<br />
to surface waters like lakes <strong>and</strong> streams. Integrating wetl<strong>and</strong>s <strong>and</strong> riparian zones in regional hydrological<br />
modeling is challenging because of the complex interactions between soil water, groundwater <strong>and</strong> surface water.<br />
The model must be able to reproduce the special hydrologic processes in wetl<strong>and</strong>s like groundwater dynamics,<br />
plant water <strong>and</strong> nutrient uptake, nutrient degradation <strong>and</strong> leaching to surface waters. An additional problem at<br />
the regional scale is the identification of riparian zones based on regionally available data.<br />
The model used in this study is the eco-hydrological model SWIM (Soil <strong>and</strong> Water Integrated Model), in which<br />
a riparian zone <strong>and</strong> wetl<strong>and</strong> module was incorporated. SWIM was chosen because it integrates the hydrological<br />
processes, vegetation, erosion <strong>and</strong> nutrient dynamics which are relevant at the watershed scale. The study shows<br />
simulation results of river discharge, groundwater dynamics <strong>and</strong> plant groundwater uptake <strong>and</strong> first results of<br />
simulated nutrient fluxes in wetl<strong>and</strong>s.<br />
Keywords: Riparian zones; wetl<strong>and</strong>s; water quality; groundwater dynamics; nutrient retention<br />
1 INTRODUCTION<br />
The water framework directive of the European<br />
Commission dem<strong>and</strong>s to bring water bodies in<br />
Europe into “a good ecological status” (EC 2000).<br />
Many efforts <strong>and</strong> improvements have been done,<br />
mainly in the implementation of waste water<br />
treatment plants. But these measures only help to<br />
improve the water quality of point sources, whereas<br />
the main origin of some important contaminants are<br />
diffuse sources like atmospheric decomposition <strong>and</strong><br />
fertilisation of crop l<strong>and</strong>. Here, riparian zones <strong>and</strong><br />
wetl<strong>and</strong>s play an important role in the control of the<br />
water quality of surface water systems (Dall’O’ et<br />
al., 2001).<br />
The paper presents an integrated catchment model<br />
with which it is possible to analyse the processes in<br />
wetl<strong>and</strong>s <strong>and</strong> riparian zones in meso- to macroscale<br />
river basins, the scale relevant for water<br />
management planning <strong>and</strong> for the implementation<br />
of the water framework directive. A simple but<br />
comprehensive mechanistic wetl<strong>and</strong> module was<br />
developed <strong>and</strong> coupled with the eco-hydrologocal<br />
model SWIM (Soil <strong>and</strong> Water Integrated Model,<br />
Krysanova et al., 1998), which integrates<br />
hydrological processes, vegetation, erosion <strong>and</strong><br />
nutrient dynamics at the watershed scale. The<br />
reliability of the model results was tested under<br />
well defined boundary conditions by comparing the<br />
results with those from a two dimensional numeric<br />
groundwater model under steady-state <strong>and</strong> transient<br />
conditions (Hattermann et al., 2004b) as well as<br />
with observed data of a meso-scale basin, using<br />
contour maps of the long-term mean water table,<br />
observed groundwater level data <strong>and</strong> observed river<br />
discharge <strong>and</strong> nutrient concentrations.<br />
The study area is located in the lowl<strong>and</strong> part of the<br />
Elbe river basin, which is representative for semihumid<br />
l<strong>and</strong>scapes in Europe, where water<br />
availability during the summer season is the main<br />
limiting factor for plant growth <strong>and</strong> crop yields.<br />
The water <strong>and</strong> nutrient balance of the catchments is<br />
influenced by water <strong>and</strong> l<strong>and</strong> use management like<br />
implementation of drainage systems, lowering of<br />
the drainage base <strong>and</strong> increased groundwater<br />
extraction. Large parts of the area have very<br />
shallow groundwater, <strong>and</strong> in particular here the<br />
water cycle is strongly influenced by water<br />
management practices like the installation of<br />
drainage systems for groundwater control (Freude,<br />
2001, L<strong>and</strong>graf, 2001, Bork et al., 1995).<br />
The results of the study show that riparian zones<br />
<strong>and</strong> wetl<strong>and</strong>s have a high potential to reduce the<br />
nutrient transport into surface water systems. Their<br />
impact is so large because they are at the interface<br />
between catchment <strong>and</strong> river systems, where the<br />
greater part of the nutrients in the catchment<br />
originally applied as fertilizers or minerlized from<br />
plant residues is already degraded. Restoration <strong>and</strong><br />
management of wetl<strong>and</strong>s is therefore of high<br />
priority for the control of non point source<br />
contamination of surface waters.<br />
2 MATERIAL AND METHODS<br />
2.1 THE MODEL<br />
2.1.1 SWIM<br />
The eco-hydrological watershed model SWIM<br />
integrates hydrological processes, vegetation,<br />
erosion <strong>and</strong> nutrient dynamics at the basin scale. A<br />
920
is<br />
three-level scheme of spatial disaggregation from<br />
basin to subbasins <strong>and</strong> to hydrotopes is used.<br />
A hydrotope is a set of elementary units in the<br />
subbasin, which have the same geographical<br />
features like l<strong>and</strong> use, soil type, <strong>and</strong> average water<br />
table depth. Therefore it can be assumed that they<br />
behave in a hydrologically uniform way<br />
(Krysanova et al., 2000). Water fluxes, plant<br />
growth <strong>and</strong> nitrogen dynamics are calculated for<br />
every hydrotope, where up to 60 vertical soil layers<br />
can be considered. The outputs from the hydrotopes<br />
are aggregated at the subbasin scale. Mean<br />
resistance time <strong>and</strong> potential retention of water <strong>and</strong><br />
nutrient fluxes are calculated using spatial features<br />
of the hydrotopes like distance to next river,<br />
gradient of the groundwater table <strong>and</strong> permeability<br />
of the aquifer. The approach allows to consider <strong>and</strong><br />
investigate the spatial pattern of l<strong>and</strong> use <strong>and</strong> l<strong>and</strong><br />
use changes. The lateral fluxes are routed over the<br />
river network, taking transmission losses into<br />
account. Plant dynamics are simulated using a<br />
simplified EPIC approach (Williams et al., 1984). A<br />
full description of the model can be found in<br />
Krysanova et al. (1998, 2000). An extensive<br />
hydrological validation of the model in the Elbe<br />
basin including sensitivity <strong>and</strong> uncertainty analyses<br />
is described in Hattermann et al. (2004a).<br />
2. 2 THE WETLAND MODULE<br />
Important for the investigation of meso- to<br />
macroscale river basins is to apply methods which<br />
are physically sound but simple enough to save<br />
computation time <strong>and</strong> data dem<strong>and</strong>. The wetl<strong>and</strong><br />
module described here consists of two parts: one<br />
part describes the groundwater fluxes <strong>and</strong> water<br />
table dynamics, where the time scale is of days or<br />
weeks. The second part describes the nutrient fluxes<br />
<strong>and</strong> degradation, where the time scale is much<br />
larger (years <strong>and</strong> decades, sometimes centuries,<br />
because of the mean residence time of the<br />
groundwater).<br />
Two cases have to be taken into account when<br />
calculating groundwater recharge: The first<br />
describes areas <strong>and</strong> time periods, where the<br />
groundwater table is relatively deep. SWIM uses an<br />
exponential delay function to calculate the effective<br />
groundwater recharge after drainage through the<br />
unsaturated geologic horizons from the last soil<br />
layer to the groundwater table (Arnold et al., 1993).<br />
The second case describes time periods with high<br />
recharge <strong>and</strong> areas with shallow groundwater,<br />
where the water table may rise <strong>and</strong> affect the lower<br />
soil zones. The soil is discretized in SWIM<br />
vertically into 5 cm layers. Layers (i, i+1, …) that<br />
are affected by groundwater are deactivated <strong>and</strong> the<br />
percolate from the layer i-1 is defined as<br />
groundwater recharge. The layer is reactivated<br />
when the water table sinks.<br />
Important for the hydrological processes <strong>and</strong><br />
nutrient fluxes in wetl<strong>and</strong>s is a good reproduction<br />
of the groundwater dynamics. Smedema & Rycroft<br />
(1983) derived a linear storage equation following<br />
the Dupuit-Forchheimer assumptions to predict the<br />
non-steady-state response of groundwater flow to<br />
periodic recharge from Hooghoudt’s (1940) steadystate<br />
formula, assuming that the variation in return<br />
flow q in mm d -1 at time step t is linearly related to<br />
the rate of change in water table height h in m (only<br />
headlosses in horizontal direction are considered):<br />
dq 8* T dh<br />
= *<br />
2<br />
dt L dt<br />
(1),<br />
where T is the transmissivity in m 2 d -1 <strong>and</strong> L the<br />
slope length in m. If the groundwater body is<br />
recharged by deep soil percolation or another<br />
source (Rc in mm d -1 ) <strong>and</strong> is depleted by drain<br />
discharge (q), it follows that the water table will<br />
rise when Rc-q > 0 <strong>and</strong> fall when Rc-q < 0. The<br />
water table fluctuations may be described as<br />
(Smedema & Rycroft 1983):<br />
dh ( Rc − q)<br />
=<br />
dt C * S<br />
(2).<br />
S is again the specific yield. It follows that by<br />
assuming that the integration constant C = 0.8:<br />
dq 10* T<br />
= *( Rc − q)<br />
= α *( Rc − q)<br />
2<br />
dt S * L<br />
(3),<br />
so that the change in drain discharge dq/dt is<br />
proportional to the excess recharge Rc-q, with<br />
being the proportionality factor (reaction factor).<br />
Equation 2 can be transformed to gain the equation<br />
for return flow:<br />
˺<br />
q<br />
t<br />
= q −<br />
* exp( −α<br />
* ∆ )<br />
t 1<br />
t<br />
+ Rc∆ t<br />
* (1 − exp( −α * ∆t)<br />
(4).<br />
Using the linear relationship between q <strong>and</strong> h<br />
(equation 1), we get:<br />
h<br />
t<br />
= h −<br />
*exp( −α<br />
* ∆ ))<br />
t 1<br />
t<br />
Rc +<br />
∆ t<br />
*(1 − exp( − α * ∆ t)<br />
0.8* S *<br />
α (5).<br />
The equations are scale independent <strong>and</strong> the spatial<br />
unit for which h <strong>and</strong> q are calculated can be either<br />
the hydrotope or the subbasin. In this study, the<br />
mean groundwater dynamics were calculated on the<br />
subbasin scale <strong>and</strong> the changes in height (dh/dt)<br />
where then added to the mean water table h of the<br />
hydrotopes U in the subbasins:<br />
dh(<br />
U ) dh<br />
= h ( U ) +<br />
dt<br />
dt<br />
(6),<br />
taking into accound the distance of the hydrotopes<br />
to the river, the slope length L.<br />
The factor a function of the transmissivity T<br />
<strong>and</strong> the slope length L:<br />
˺<br />
921
is<br />
can<br />
has<br />
is<br />
α =<br />
10 * T<br />
S * L<br />
2<br />
(7).<br />
Therefore, the reaction factor has a physical<br />
meaning, as illustrated by the comparison with the<br />
results of the numerically solved Bousinesq<br />
Equation (Hattermann et al., 2004b), where the<br />
same geo-hydrological parameters (T, L, S) were<br />
used. However, for meso- to macro-scale basins the<br />
basic geo-hydrological parameters, namely<br />
transmissivity <strong>and</strong> specific yield, are usually not<br />
available. Especially the specific yield is difficult to<br />
determine. Hattermann et al. (2004b) suggested<br />
another method to estimate the reaction factor ˺<br />
from field observations: From Equation 5, it<br />
follows that in periods without recharge (Rc = 0):<br />
ln ht<br />
−1<br />
− ln ht<br />
=<br />
∆t<br />
α (8).<br />
Therefore, be estimated directly by using<br />
observations of the groundwater head h. This was<br />
done using an automatic calibration algorithm by<br />
adjusting T <strong>and</strong> S in physically sound limits. The<br />
˺<br />
inverse value of the dimension of time <strong>and</strong><br />
can be interpreted as the reaction time of the<br />
groundwater table <strong>and</strong> discharge to changes in<br />
recharge. It has a time scale of days to weeks.<br />
˺<br />
While it is possible to describe water table<br />
dynamics using the mean reaction time, the time<br />
scales which have to be considered for the<br />
simulation of nutrient retention are much larger<br />
(years <strong>and</strong> decades), because the actual residence<br />
time is the crucial value which determines the<br />
intensity of degradation. According to Wendl<strong>and</strong> et<br />
al. (1993), the degradation of nitrate N can be<br />
approximated by a linear decay equation, where<br />
a function of temperature <strong>and</strong> available oxygen.<br />
The full retention of a l<strong>and</strong>scape is then a function<br />
of mean residence time <strong>and</strong> degradation:<br />
̄<br />
conductivity of layer z, J the hydraulic gradient, dz<br />
in m the distance <strong>and</strong> n the number of layers:<br />
− k * J ( z)<br />
v s<br />
( z)<br />
= (12),<br />
S<br />
n<br />
dzi<br />
χ = ∑<br />
(13).<br />
i=<br />
1 vs<br />
( zi<br />
)<br />
Plant uptake of water <strong>and</strong> nutrients from<br />
groundwater is only possible in times when the<br />
plant roots have excess to it <strong>and</strong> if the plant dem<strong>and</strong><br />
cannot be satisfied by soil water <strong>and</strong> nutrient<br />
recourses. A resistance function controls the ability<br />
of plant roots for water <strong>and</strong> nutrient uptake from<br />
groundwater.<br />
2.3 THE BASIN<br />
The northern lowl<strong>and</strong> part of the German Elbe<br />
basin, where the model was tested in the Nuthe<br />
catchment (1998 km 2 , see Figure 1), is climatically<br />
one of the driest regions in Germany, with mean<br />
annual precipitation of about 600 mm per year.<br />
Hence, water availability during the summer season<br />
is the limiting factor for plant growth. The lowl<strong>and</strong><br />
is formed by mostly s<strong>and</strong>y glacial sediments <strong>and</strong><br />
drained by slowly flowing streams with broad river<br />
valleys. The upper sites with deep water tables are<br />
covered by s<strong>and</strong>y, highly permeable soils <strong>and</strong><br />
mostly pine forests or by arable l<strong>and</strong> on ground<br />
moraine with till soils that tend to have layers with<br />
lower water permeability. Valleys are covered by<br />
loamy alluvial soils with grassl<strong>and</strong> <strong>and</strong> riparian<br />
forests, where the groundwater is very shallow, <strong>and</strong><br />
arable l<strong>and</strong> elsewhere. During the last two decades,<br />
decreasing water levels in rivers <strong>and</strong> groundwater<br />
have been observed (L<strong>and</strong>esumweltamt<br />
Br<strong>and</strong>enburg 2000a & 2002b). The main mean<br />
climatic <strong>and</strong> hydrologic characteristics of the study<br />
area are listed in table 1.<br />
N<br />
out<br />
−∆t<br />
/ β1<br />
in −∆t<br />
/ β1<br />
s, t s,<br />
t−1<br />
s,<br />
t<br />
(1 )<br />
− N<br />
= N<br />
s,<br />
t−1<br />
e<br />
−λ1∆t<br />
e<br />
+ N<br />
− e<br />
N<br />
out<br />
−∆t<br />
/ β 2 in −∆t<br />
/ β 2<br />
i, t i,<br />
t−1<br />
i,<br />
t<br />
(1 )<br />
− N<br />
= N<br />
i,<br />
t−1<br />
e<br />
−λ2∆t<br />
e<br />
+ N<br />
− e<br />
N<br />
out<br />
−∆t<br />
/ β3<br />
in −∆t<br />
/ β3<br />
b, t b,<br />
t−1<br />
b,<br />
t<br />
(1 )<br />
− N<br />
= N<br />
e<br />
−λ3∆t<br />
e<br />
+ N<br />
− e<br />
b,<br />
t−1<br />
(9, 10, 11),<br />
where the mean residence time of water in a<br />
subbasin. Since SWIM distinguishes between<br />
surface flow s, interflow i <strong>and</strong> base flow b, each<br />
having different retention characteristics (residence<br />
time, oxygen content), there has to be one equation<br />
for each of the fluxes. The mean residence time of<br />
the water in the subbasin from hydrotope to river<br />
(̐) in s -1 is calculated using the seepage velocity v s<br />
[m s -1 ], where k in m s -1 is the hydraulic<br />
˻<br />
Gauge station<br />
observation well<br />
precipitation station<br />
climate station<br />
Elbe river system<br />
Berlin<br />
Nuthe basin<br />
Stepenitz<br />
Figure 1: The location of the Nuthe basin <strong>and</strong> the<br />
observation points.<br />
922
50.5<br />
Jun- Jun- Jun- Jun- Jun- Jun- Jun- Jun-<br />
46.5<br />
45.5<br />
4<br />
water table [m]<br />
[m NN]<br />
All necessary spatial information to derive the<br />
subbasin <strong>and</strong> hydrotope structure of the basins, the<br />
digital elevation model (DEM), the soil map of the<br />
State Br<strong>and</strong>enburg, the geo-hydrological map, the<br />
l<strong>and</strong> use map <strong>and</strong> water table contour maps were<br />
stored on a grid format with 50 m resolution. The<br />
Nuthe basin was subdivided into 122 subbasins<br />
based on the DEM <strong>and</strong> the drainage network.<br />
Table 1: Long term mean annual precipitation (P),<br />
mean annual temperature (T) <strong>and</strong> river discharge<br />
(Q) of the basin under study.<br />
basin<br />
area<br />
[km 2 ]<br />
P<br />
[mm a -1 ]<br />
T<br />
[°C]<br />
Q<br />
[m 3 s -1 ]<br />
Nuthe 1938.0 590.5 8.8 9.06<br />
3. RESULTS AND DISCUSSION<br />
3.1 GROUNDWATER AND RIVER<br />
FLOW DYNAMICS<br />
First, the simulated mean annual water table depth<br />
of all subbasins in the Nuthe basins were calibrated<br />
automatically using the transmissivity in a<br />
physically sound range. The mean simulated<br />
amplitude was too high <strong>and</strong> had to be smoothed by<br />
a moderate increase in the value of specific yield<br />
(as taken from the geo-hydrological map). The<br />
Mean Absolute Error of the long term mean<br />
observed against the mean simulated water table in<br />
all subbasins was 0.026 m. The groundwater<br />
reaction factors of the subbasins had values<br />
between 0.1 (loamy sediments) <strong>and</strong> 0.3 (s<strong>and</strong>y /<br />
loamy sediments). The time dynamics of the<br />
simulated water tables in terms of rising <strong>and</strong><br />
retention periods were not calibrated.<br />
53.5<br />
52.5<br />
51.5<br />
44.5<br />
43.5<br />
42.5<br />
41.5<br />
40.5<br />
39.5<br />
38.5<br />
37.5<br />
32.5<br />
31.5<br />
30.5<br />
29.5<br />
Jun-<br />
1981<br />
Jun-<br />
1982<br />
Jun-<br />
1983<br />
water table observed<br />
water table simulated<br />
Jun-<br />
1984<br />
5<br />
3<br />
2<br />
1<br />
Jun-<br />
1985<br />
Jun-<br />
1986<br />
Jun-<br />
1987<br />
Jun-<br />
1988<br />
[m] [mm]<br />
Figure 2: Comparison of observed <strong>and</strong> simulated<br />
groundwater table for five locations in the Nuthe<br />
basin (Hattermann et al., 2004b).<br />
Figure 2 shows a comparison of five observed<br />
groundwater table hydrographs from the Nuthe<br />
basin with those simulated. The observation wells<br />
were selected in order to represent a cross section<br />
through the basin from the lowl<strong>and</strong>s in the north to<br />
the hilly area in the south. Well 1 is located next to<br />
the outlet of the Nuthe river catchment. The curves<br />
show a good fit, especially for the early 1980s. The<br />
rise of the groundwater level in 1987 <strong>and</strong> 1988 is<br />
slightly overestimated by the model in subbasins 2,<br />
4 <strong>and</strong> 5. As explained in section 1, the natural flow<br />
regime in the Nuthe basin is influenced by stream<br />
flow control (weir <strong>and</strong> reservoir management), <strong>and</strong><br />
especially in the lowl<strong>and</strong> areas the water level is<br />
controlled by l<strong>and</strong> drainage. The simulated<br />
groundwater hydrographs are very similar, whereas<br />
the observations show more differences. The higher<br />
variability in the observed water levels is the result<br />
of small-scale heterogeneities in the aquifer <strong>and</strong> of<br />
local precipitation events which are missing in the<br />
observed records. An even better fit would be<br />
possible by implementing additional management<br />
information. However, this was not the objective of<br />
the study. On the contrary, the study aimed at<br />
showing that a simplified model approach yields<br />
satisfactory results using commonly available data.<br />
Figure 3 illustrates the impact of plant water uptake<br />
on the simulated water table. While the<br />
groundwater tables simulated with <strong>and</strong> without<br />
plant water uptake converge during the winter term,<br />
they separate during the vegetation period, where<br />
the plant uptake leads to a decline of the<br />
groundwater table.<br />
8<br />
6<br />
4<br />
2<br />
0<br />
-2<br />
evapotranspiration<br />
GW-recharge<br />
GW-vegetation<br />
GW+vegetation<br />
-4<br />
1.6.81 1.6.82 1.6.83 1.6.84 1.6.85<br />
Figure 3: Comparison of simulated <strong>and</strong> observed<br />
groundwater table with <strong>and</strong> without plant water<br />
uptake from groundwater.<br />
The mean long term difference between the<br />
observed <strong>and</strong> simulated river discharge at the basin<br />
outlet is 3.0% for the calibration period 1981 -<br />
1988, indicating that the water balance is correctly<br />
calculated by SWIM. The daily Nash & Sutcliffe<br />
efficiency is 0.7 (only 0.54 for the validation period<br />
1989-2000). The hydraulic regime of the Nuthe<br />
923
Q [m 3 /s]<br />
basin is strongly influenced by water management<br />
regulations like drainage systems <strong>and</strong> weir plants,<br />
so that it is difficult to reproduce the hydrograph<br />
with higher accuracy. The summer discharge is in<br />
some years overestimated by the model (see Figure<br />
4). This can be explained by water abstraction <strong>and</strong><br />
25<br />
20<br />
15<br />
10<br />
5<br />
0<br />
Jan. 98<br />
Mar. 98<br />
May. 98<br />
Jul. 98<br />
Sep. 98<br />
regulation measures, when a minimum river flow is<br />
provided by reservoir management in dry summer<br />
periods. It is worth mentioning that the efficiency<br />
was notably higher for other meso- <strong>and</strong> macro-scale<br />
subbasins of the Elbe located in hilly <strong>and</strong><br />
mountainous areas (Hattermann et al., 2004a).<br />
Figure 4: Comparison of daily river flow observed<br />
<strong>and</strong> simulated (gauge Babelsberg).<br />
Without additional plant water uptake from<br />
groundwater, the total evapotranspiration would be<br />
24% lower, leading to an increase in river discharge<br />
of about 77%.<br />
3.2 NITRATE CONCENTRATIONS<br />
Nov. 98<br />
The nitrate concentration in the Nuthe river during<br />
the eighties was strongly influenced by point<br />
sources (irrigation of waste waters in very small<br />
areas, municipal waste waters, even direct<br />
Jan. 99<br />
Mar. 99<br />
May. 99<br />
Q observed<br />
Q simulated<br />
Jul. 99<br />
Sep. 99<br />
Nov. 99<br />
Nitrate [mg/l]<br />
atmospheric decompositions (about 40 kg/ha), <strong>and</strong><br />
plant decompositions after harvest <strong>and</strong> fall. The<br />
comparison shows that the periodicity <strong>and</strong><br />
amplitude of the observed values is mostly well<br />
reproduced by SWIM, although the difference<br />
between observed <strong>and</strong> simulated values is large<br />
especially at the end of the year. The reason is that<br />
the diffuse sources for nitrate contamination (in<br />
particular fertilization) are not very well known,<br />
because information about crop rotation schemes<br />
<strong>and</strong> fertilization regimes are not available at the<br />
regional scale. In addition, the flow regulation by<br />
dams <strong>and</strong> weirs <strong>and</strong> the drainage systems influence<br />
of course not only river discharge but also nutrient<br />
4.0<br />
3.5<br />
3.0<br />
2.5<br />
2.0<br />
1.5<br />
1.0<br />
0.5<br />
0.0<br />
1.1.90<br />
1.1.91<br />
conc -veg<br />
conc +veg<br />
difference<br />
1.1.92<br />
1.1.93<br />
fluxes. The mean residence time of groundwater is<br />
41 years, with a maximum of approximately 400<br />
years. The values are in good agreement with<br />
L<strong>and</strong>esumwelamt (2002b), who estimated the<br />
nutrient loads <strong>and</strong> retention in the lowl<strong>and</strong><br />
catchments of the Elbe basin.<br />
Figure 6: Comparison of simulated <strong>and</strong> observed<br />
nitrate concentration with <strong>and</strong> without plant water<br />
uptake from groundwater.<br />
Figure 6 illustrates the impact of plant uptake of<br />
nitrate in riparian zones <strong>and</strong> wetl<strong>and</strong>s. As shown<br />
also for the impacts of plants on the water level in<br />
Figure 3, the differences are the highest during the<br />
1.1.94<br />
1.1.95<br />
0.0<br />
0.5<br />
1.0<br />
1.5<br />
2.0<br />
2.5<br />
3.0<br />
3.5<br />
4.0<br />
difference [mg/l]<br />
5.0<br />
4.0<br />
nitrate sim<br />
nitrate obs<br />
Nitrate [mg/l]<br />
3.0<br />
2.0<br />
1.0<br />
0.0<br />
01.01.90 01.01.91 02.01.92 01.01.93 02.01.94 02.01.95<br />
discharge of liquid manure into surface waters),<br />
where the records are vague <strong>and</strong> incomplete, so that<br />
the comparison in Figure 5 is done for a time period<br />
in the ninetieth, where impact of point sources is<br />
very limited because of the implementation of<br />
waste water treatment plants in the basin.<br />
Figure 5: Simulated <strong>and</strong> observed nitrate<br />
concentrations in the Nuthe river.<br />
Diffuse sources in this study are fertilizer<br />
applications (about 180 kg/ha for winter wheat),<br />
summer season, when plant dem<strong>and</strong> is high <strong>and</strong> can<br />
therefore not be satisfied by the soil water<br />
concentrations. The difference becomes smaller<br />
during the late summer, because the total amount of<br />
available nutrients in soils <strong>and</strong> hence also the<br />
leaching of nutrients has its minimum.<br />
Figure 7: Additional nitrate uptake by plants in<br />
riparian zones <strong>and</strong> wetl<strong>and</strong>s.<br />
924
Figure 7 shows a map of the additional plant nitrate<br />
uptake from groundwater in kg/ha. The values are<br />
not so large in comparison with the total plant<br />
uptake (up to 180 kg/ha). The additional uptake is<br />
only about 6% of the total uptake, but this leads to a<br />
retention of about 35.5% of the total river load. The<br />
reason is that the additional uptake happens in an<br />
area next to the surface water bodies, where the<br />
largest part of the nitrate which was originally<br />
applied by fertilizers, mineralised from plant<br />
residues <strong>and</strong> decomposed from the atmosphere is<br />
already degraded.<br />
4 CONCLUSIONS<br />
The simulation results indicate that relatively small<br />
parts of the total catchment area have a relatively<br />
high impact on the water <strong>and</strong> nutrient balance in the<br />
catchment (additional evapotranspiration of about<br />
24%, additional nitrate uptake of about 6%, leading<br />
to a decrease in river discharge of about 77% <strong>and</strong> to<br />
an decrease in annual river nitrate load of about<br />
35%). Riparian zones <strong>and</strong> wetl<strong>and</strong>s are buffer<br />
systems able to reduce contamination of surface<br />
waters, as long as the vegetation has access to<br />
groundwater. On the other h<strong>and</strong>, restoration of<br />
wetl<strong>and</strong>s will lead to increased water losses by<br />
evapotranspiration, crucial in a region where river<br />
discharge during the summer season is only<br />
possible by water regulation through dams <strong>and</strong><br />
weirs, <strong>and</strong> where a trend to lower annual<br />
precipitation has been observed during the last<br />
decades. It follows that water managers have to find<br />
a sensitive balance between water quality <strong>and</strong> water<br />
quantity aspects in the planning process.<br />
5 ACKNOWLEDGEMENTS<br />
The authors would like to thank all their colleagues<br />
at PIK who contributed to this paper with technical<br />
help, particularly Daniel Doktor. Part of this work<br />
was supported by the German BMBF programme<br />
GLOWA (GLObal WAter) Elbe <strong>and</strong> the<br />
Br<strong>and</strong>enburg State <strong>Environmental</strong> Agency (LUA<br />
Br<strong>and</strong>enburg).<br />
6 REFERENCES<br />
Arnold, J.G., Allen, P.M., Bernhardt, A., 1993. A<br />
comprehensive surface-groundwater flow<br />
model. Journal of Hydrology 142, pp. 47-<br />
69.<br />
Borg, H.-R., Dalchow, C., Kächele, H., Piorr, H.-P.,<br />
Wenkel, K.-O., 1995.<br />
Agrarl<strong>and</strong>schaftsw<strong>and</strong>el in Nordost-<br />
Deutschl<strong>and</strong>. Ernst & Sohn.<br />
Dall’O’, M., Kluge, W., Bartels, F., 2001.<br />
FEUWAnet: a multi-box water level <strong>and</strong><br />
lateral exchange model for riparian<br />
wetl<strong>and</strong>s. Journal of Hydrology 250, pp.<br />
40-62.<br />
EC, 2000. Establishing a framework for community<br />
action in the field of water policy.<br />
Directive 2000/60/EC of the European<br />
Parliament <strong>and</strong> of the Council of 23<br />
October 2000, Official Journal of the<br />
European Communities, Brussels.<br />
Freude, M., 2001. L<strong>and</strong>schaftswasserhaushalt in<br />
Br<strong>and</strong>enburg: Situationsanalyse und<br />
Ausblick. In: Korth, B. & Prinzensing, G.<br />
(Eds.), Dokumentation der SPD<br />
Veranstaltung<br />
‚L<strong>and</strong>schaftswasserhaushalt’.<br />
Hattermann, F.F., Krysanova, V., Wechsung, F.,<br />
Wattenbach, M., 2004a. Macroscale<br />
validation of the eco-hydrological model<br />
SWIM for hydrological processes in the<br />
Elbe basin with uncertainty analysis. In:<br />
Fohrer, N. <strong>and</strong> Arnold, J., 2004. Regional<br />
Assessment of Climate <strong>and</strong> Management<br />
Impacts Using the SWAT Hydrological<br />
Model. Hydrological Processes (in press).<br />
Hattermann, F.F., Krysanova, V, Wechsung, F,<br />
Wattenbach, M. 2004b. Integrating<br />
groundwater dynamics in regional<br />
hydrological modelling. <strong>Environmental</strong><br />
<strong>Modelling</strong> <strong>and</strong> <strong>Software</strong> (in press).<br />
Krysanova, V., Becker, A., Müller-Wohlfeil, D.-I.,<br />
1998. Development <strong>and</strong> test of a spatially<br />
distributed hydrological / water quality<br />
model for mesoscale watersheds.<br />
Ecological <strong>Modelling</strong> 106, pp. 261-289.<br />
Krysanova, V., Wechsung, F., Arnold, J.,<br />
Srinivasan, R., Williams, J., 2000. PIK<br />
Report Nr. 69 "SWIM (Soil <strong>and</strong> Water<br />
Integrated Model), User Manual".<br />
L<strong>and</strong>esumweltamt Br<strong>and</strong>enburg, 2000.<br />
Flächendeckende Modellierung von<br />
Wasserhaushaltsgrößen für das L<strong>and</strong><br />
Br<strong>and</strong>enburg. Studien und<br />
Tagungsberichte, B<strong>and</strong> 27.<br />
L<strong>and</strong>esumweltamt Br<strong>and</strong>enburg, 2002a.<br />
Umweltdaten aus Br<strong>and</strong>enburg - Bericht<br />
2002 des L<strong>and</strong>esumweltamtes.<br />
L<strong>and</strong>esumweltamt Br<strong>and</strong>enburg, 2002b.<br />
Stoffeinträge in die Gewässer des L<strong>and</strong>es<br />
Br<strong>and</strong>enburg. B<strong>and</strong> 68.<br />
L<strong>and</strong>graf, L. 2001. Tätigkeitsbericht der<br />
Projektgruppe<br />
“L<strong>and</strong>schaftswasserhaushalt”. In: Korth,<br />
B. & Prinzensing (Eds.), G.Dokumentation<br />
der SPD Veranstaltung<br />
‚L<strong>and</strong>schaftswasserhaushalt’.<br />
Smedema, L.K., Rycroft, D.W., 1983. L<strong>and</strong><br />
Drainage – Planning <strong>and</strong> Design of<br />
Agricultural Drainage Systems. Cornell<br />
University Press, Ithaca, NY.<br />
Wendtl<strong>and</strong>, F., Albert, H., Bach, M., Schmidt, R.<br />
1993. Atlas zum Nitratstrom in der<br />
Bundesrepublik Deutschl<strong>and</strong>. Springer<br />
Verlag, Berlin.<br />
925
Williams, J.R., Renard, K.G., Dyke, P.T., 1984.<br />
EPIC – a new model for assessing<br />
erosion’s effect on soil productivity.<br />
Journal of Soil <strong>and</strong> Water Conservation<br />
38(5), pp. 381-383.<br />
926
Ecoregion Classification Using a Bayesian Approach <strong>and</strong><br />
Model-based Clustering<br />
D. Pullar a S. Low Choy b , W. Rochester a<br />
a<br />
Geography Planning <strong>and</strong> Architecture, The University of Queensl<strong>and</strong>, Brisbane QLD 4072, Australia.<br />
(D.Pullar@uq.edu.au)<br />
b <strong>Environmental</strong> Information Systems, <strong>Environmental</strong> Protection Agency, PO Box 155, Albert Street,<br />
Brisbane QLD 400, Australia.<br />
Abstract: Ecological regions are increasingly used as a spatial unit for planning <strong>and</strong> environmental<br />
management. It is important to define these regions in a scientifically defensible way to justify any decisions<br />
made on the basis that they are representative of broad environmental assets. The paper describes a<br />
methodology <strong>and</strong> tool to identify cohesive bioregions. The methodology applies an elicitation process to<br />
obtain geographical descriptions for bioregions, each of these is transformed into a Normal density estimate<br />
on environmental variables within that region. This prior information is balanced with data classification of<br />
environmental datasets using a Bayesian statistical modelling approach to objectively map ecological regions.<br />
The method is called model-based clustering as it fits a Normal mixture model to the clusters associated with<br />
regions, <strong>and</strong> it addresses issues of uncertainty in environmental datasets due to overlapping clusters.<br />
Keywords: Biogeography; Bayesian statistical modelling; GIS; Elicitation; Mixture models; Clustering<br />
1. INTRODUCTION<br />
Ecoregions define recognizable areas which<br />
embody broad environmental <strong>and</strong> l<strong>and</strong>scape<br />
structures. Ecoregion classification <strong>and</strong> subsequent<br />
boundary definition have a significant impact on<br />
natural resource management. The need for<br />
bioregionalisations was initially driven by<br />
conservation planning, but they have taken on<br />
extended roles as spatial units for tabulating<br />
environmental information (as opposed to socioeconomic<br />
administrative units) <strong>and</strong> for the<br />
allocation of funding for the environment. In<br />
Australia a bioregional planning framework, called<br />
the Interim Biogeographic Regionalisation of<br />
Australia (IBRA) has been established [EA, 2000].<br />
The biogeographical regions in IBRA are l<strong>and</strong><br />
areas comprised of interacting ecosystems that are<br />
repeated in similar form across the l<strong>and</strong>scape.<br />
Typically the IBRA regions are based upon factors<br />
such as climate, lithology, geology, l<strong>and</strong>forms <strong>and</strong><br />
vegetation as surrogate indicators of the ecological<br />
processes that occur on l<strong>and</strong>, particularly as<br />
relevant to conservation strategies <strong>and</strong> natural<br />
resource capability. The ecoregions are mapped at<br />
different scales within a hierarchy ranging from<br />
broad l<strong>and</strong> types to local regional ecosystems (See<br />
Figure 1).<br />
Integrated<br />
combination of<br />
climate, geology,<br />
l<strong>and</strong>form, soils<br />
<strong>and</strong> vegetation.<br />
Fauna<br />
Flora<br />
Bioregions<br />
L<strong>and</strong> Zones<br />
Sub-bioregion<br />
L<strong>and</strong>scapes<br />
Catchments<br />
Regional Ecosystems<br />
L<strong>and</strong> Types<br />
Bioclimatic<br />
- Geology<br />
- Geomorphology<br />
- Climate<br />
L<strong>and</strong>scapes<br />
- Patterns<br />
- L<strong>and</strong>forms<br />
- Vegetation<br />
Processes:<br />
- Spatial<br />
- Temporal<br />
- Functional<br />
Regional Ecosystems<br />
- Vegetation groups<br />
- Biodiversity<br />
Species<br />
Figure 1. Conceptual hierarchy of bioregional<br />
classification at four levels. Adopted from Sattler<br />
<strong>and</strong> Williams [1999]<br />
The focus of this paper is on sub-bioregions as<br />
areas of l<strong>and</strong> that have a distinctive pattern of<br />
l<strong>and</strong>form <strong>and</strong> vegetation which indicates major<br />
differences in l<strong>and</strong> processes <strong>and</strong> biological<br />
communities [Sattler <strong>and</strong> Williams, 1999]. Subbioregions<br />
are mapped at a scale of 1:100,000. In<br />
Queensl<strong>and</strong>, the delineation of sub-bioregions is<br />
largely overseen by an expert scientific panel who<br />
interpret available mapped information sources<br />
using their knowledge of the region. The regions<br />
927
are mapped as areas that have distinctive l<strong>and</strong>scape<br />
patterns with permeable boundaries. With growing<br />
use of these regions within natural resource<br />
decision-making there is pressure to shift from<br />
subjective expert-based methods for defining<br />
bioregions to a more repeatable, scientifically<br />
defensible <strong>and</strong> objective system of classification.<br />
In response to this need a project was undertaken<br />
to make the expert input more explicit <strong>and</strong> to<br />
incorporate classification based on statistical<br />
analysis of geographic information. The guiding<br />
principle in the classification is to determine the<br />
key drivers amongst a range of abiotic<br />
environmental factors using cluster analysis to<br />
identify cohesive <strong>and</strong> separable classes from<br />
geophysical datasets.<br />
The outline for the paper is as follows. The next<br />
section explains the location for the study area.<br />
Section 3 describes the Bayesian approach to<br />
classification. Section 4 describes the spatial <strong>and</strong><br />
graphical tool used to elicit knowledge from<br />
experts that is used as prior information to guide<br />
the classifier. Section 5 illustrates the results for a<br />
classification. Section 6 summarises <strong>and</strong> discusses<br />
the significance of the work.<br />
2. STUDY AREA<br />
The results of the research are to be applied to<br />
eastern bioregions within the state of Queensl<strong>and</strong><br />
in Australia, however the paper will focus on one<br />
bioregion in the south-eastern corner of the state<br />
(Figure 2).<br />
Figure 2. Locality map showing bioregion <strong>and</strong><br />
sub-bioregions for the study region in South-East<br />
Queensl<strong>and</strong>.<br />
The bioregion covers 66,000 km 2 <strong>and</strong> comprises<br />
coastal plains, a major drainage basin for the<br />
Brisbane river catchment, <strong>and</strong> mountain ranges.<br />
The area is sub-tropical <strong>and</strong> is considered one of<br />
the most species-rich <strong>and</strong> diverse parts of Australia<br />
for flora <strong>and</strong> fauna [Sattler <strong>and</strong> Williams, 1999].<br />
There is significant settlement of the region with a<br />
population of approx. 2 million people, <strong>and</strong> the<br />
expectation this population will double in the next<br />
40 years. Despite a number of national parks <strong>and</strong><br />
smaller reserves the area has several vulnerable<br />
species that are endangered <strong>and</strong> bioregional<br />
planning plays an important part in decisionmaking<br />
for future development.<br />
3. METHODOLOGY<br />
Previous approaches to bioregionalisation have<br />
tended to be either expert-driven or data-driven [eg<br />
Bunce et al 2002, Hargrove <strong>and</strong> Hoffman 1999).<br />
For example the most recent set of Queensl<strong>and</strong>’s<br />
sub-bioregions [Sattler <strong>and</strong> Williams 1999] is<br />
based on expert opinion on sub-bioregional<br />
boundaries [see Morgan <strong>and</strong> Terry 1990]. As is<br />
common in these situations a Delphic approach<br />
was used, where a panel of several experts were<br />
consulted together about the location of<br />
boundaries, based on mapped <strong>and</strong> well-defined<br />
topographic features such as regional ecosystem<br />
boundaries (derived from aerial photography),<br />
ridgelines, etc [Neldner 2002]. In their<br />
assessments, experts also referred to other spatial<br />
information such as soils <strong>and</strong> climate. In contrast<br />
the most recent sub-bioregions for Tasmania<br />
[Peters <strong>and</strong> Thackway 1998] take a data-driven<br />
approach <strong>and</strong> make use of spatially extensive fine<br />
scale information both biotic <strong>and</strong> abiotic. This data<br />
was input to multivariate clustering techniques<br />
[Everitt <strong>and</strong> H<strong>and</strong>, 1981], <strong>and</strong> then use this as<br />
input, post-hoc, to an expert panel process to<br />
address inconsistencies <strong>and</strong> other model<br />
inadequacies.<br />
Here we propose a regionalisation approach that<br />
aims to balance inputs from both experts <strong>and</strong> data,<br />
integrated within a Bayesian statistical modelling<br />
framework. The basic premise [Congdon 2001] is<br />
that updating prior information on parameters<br />
using information provided by data (likelihood)<br />
provides posterior information on these<br />
parameters:<br />
Posterior ∝ Prior × Likelihood (1)<br />
This provides a natural framework for continually<br />
updating old models (priors) with new data<br />
(likelihood) as it arises to produce new improved<br />
models (posteriors). Readers are referred to<br />
Gelman et al [2004] for further information on<br />
Bayesian statistical modelling.<br />
928
3. 1 Models<br />
In many situations statistical distributions (eg<br />
Normal, Poisson, exponential, etc) do not fit the<br />
observed data. This is particularly true of<br />
environmental data where a mixture of<br />
environmental conditions could lead to different<br />
patterns in the data. Mixture models address this<br />
issue by explicitly allowing for a mixture of<br />
components, each described by a separate<br />
distribution, to combine together into an overall<br />
mixture distribution.<br />
More precisely, we define a mixture distribution<br />
for K clusters or mixture components indexed k =<br />
1…K. Let w k denote the weight or proportion of<br />
observations in each cluster. Denote by x the<br />
dataset with one row per observation <strong>and</strong> one<br />
column per variable (eg environmental attribute).<br />
Let θ represent the set of mixture model<br />
parameters. Then the overall mixture likelihood<br />
p(·) is defined as the weighted sum of mixture<br />
components f(·):<br />
K<br />
p( x | ) = k =<br />
wk<br />
f<br />
k<br />
( x | θ<br />
1<br />
k<br />
)<br />
θ (2)<br />
A common choice for the model for each cluster is<br />
a multivariate Normal, giving rise to a Gaussian<br />
mixture model. In the k th cluster, for the i th<br />
observation on all variables x i :<br />
f<br />
(<br />
k<br />
d k k<br />
x i<br />
| θ ) ≡ MVN ( µ , ) (3)<br />
This indicates that each observation in the cluster<br />
is drawn from a multivariate Normal distribution of<br />
d dimensions with d×1 vector of means µ <strong>and</strong> d×d<br />
variance-covariance matrix Σ. This could mean for<br />
bioregionalisation that a particular region is<br />
defined by a 3D Normal distribution with mean<br />
rainfall 50mm pa, soil moisture 0.10, <strong>and</strong> elevation<br />
50m. The st<strong>and</strong>ard errors could be, say,<br />
respectively 10mm pa, 0.04, <strong>and</strong> 20m, with the<br />
only non-negligible covariance being 42% between<br />
soil moisture <strong>and</strong> rainfall. If the variability of a<br />
variable is narrow (small st<strong>and</strong>ard error) then the<br />
cluster/region is closely linked to that<br />
environmental attribute. Similarly wide variance<br />
leads to little relationship between that<br />
geographical region <strong>and</strong> the environmental<br />
attribute in question. See Figure 3.<br />
A method proposed by Dempster et al [1977] relies<br />
on introduction of extra (auxiliary) variables to<br />
facilitate the computations. These keep track of<br />
cluster membership for each observation. Let z i = k<br />
if the i th observation falls into the k th cluster. See<br />
figure 4. Then the weight of each cluster is just the<br />
same as the probability of cluster membership:<br />
w<br />
= p( z k)<br />
(4)<br />
k i<br />
=<br />
Min. temperature of bc06 coldest month<br />
0 50 100 150<br />
300 350 400 450<br />
Annual Precipitation<br />
bc12<br />
Figure 3. A mixture model with two variables <strong>and</strong><br />
two clusters. The ellipsoids mark the st<strong>and</strong>ard<br />
deviation of the bivariate normal distribution that<br />
defines each cluster. The size, shape <strong>and</strong><br />
orientation of a cluster's ellipsoid indicates the<br />
means, variances <strong>and</strong> correlations of the two<br />
variables for sites in the cluster.<br />
Figure 4. A mixture model in which the<br />
distribution of one variable is modelled as a<br />
mixture of the distributions of the variable at sites<br />
in each of two clusters. Site 1 is assigned to cluster<br />
1 because the probability density at x 1 is greater for<br />
cluster 1 than that for cluster 2.<br />
The computation of the mixture model is applied<br />
iteratively to explore the posterior distribution of<br />
each parameter repeatedly until most important<br />
parts of the distribution have been explored. This is<br />
the general idea behind Markov Chain Monte<br />
Carlo (MCMC) [Gelman et al , 2004]. Through<br />
MCMC we obtain dependent simulations that, once<br />
they’ve reached equilibrium, model the target<br />
posterior distributions (1) for each parameter. The<br />
challenge is to design an MCMC sampler that<br />
converges to equilibrium efficiently.<br />
929
A general approach for implementing either<br />
Bayesian approach comprises three steps:<br />
1. Designing priors<br />
2. Designing MCMC samplers<br />
3. Implementing MCMC<br />
Designing the MCMC samplers <strong>and</strong> implementing<br />
the MCMC are not the focus of this paper. We<br />
focus on the first stage of designing appropriate<br />
priors in the next section.<br />
3. 2 Priors <strong>and</strong> eliciting expert knowledge<br />
In equation (1) the priors <strong>and</strong> likelihood have equal<br />
impact. The theoretical mixture model likelihood is<br />
defined through equations (2)-(4). Designing<br />
appropriate priors is somewhat of an “art” <strong>and</strong><br />
requires two main stages:<br />
1. Select appropriate priors to enable dialogue<br />
with expert(s) so that they can describe their<br />
prior knowledge in a form suitable for input to<br />
the model.<br />
2. Design <strong>and</strong> implement elicitation processes<br />
(experiments) to quantify the prior knowledge<br />
held by experts.<br />
These two stages are closely linked. Without<br />
knowing the form of the prior the elicitation<br />
process at worst can provide irrelevant<br />
information. On the other h<strong>and</strong>, without a rigorous<br />
elicitation experiment, it is difficult to ensure the<br />
validity, repeatability <strong>and</strong> transparency of priors<br />
obtained.<br />
For Gaussian mixture models there are four main<br />
types of priors we can consider, depending on the<br />
type of expert knowledge available.<br />
Expert knowledge<br />
Experts know nothing<br />
(objective ~ Frequentist)<br />
Experts know something<br />
about model coefficients<br />
(means <strong>and</strong> variances on each<br />
variable in each cluster)<br />
Appropriate prior<br />
Non-informative<br />
(improper) priors<br />
Informative<br />
conjugate<br />
Informative semiconjugate<br />
Experts know something else Data Augmentation<br />
priors<br />
We focus on the second more usual choice, the<br />
informative conjugate prior: informative since<br />
prior knowledge on means <strong>and</strong> variances in each<br />
cluster informs the model (has impact on results),<br />
<strong>and</strong> conjugate since the choice of prior distribution<br />
factors out “nicely” mathematically. For the<br />
Gaussian mixture model, this prior comprises a<br />
Normal distribution for cluster means conditional<br />
on known cluster variance (µ k | Σ k in Equation 5),<br />
with an inverse Wishart distribution for the inverse<br />
covariance matrix (Σ k -1 in Equation 5) [Diebolt <strong>and</strong><br />
Robert 1994]. <br />
−1<br />
−1<br />
µ | Σ ~ N(<br />
m , s ) Σ ~ W ( ν , ϕ ) (5)<br />
k<br />
k<br />
k<br />
k<br />
Each prior has a number of hyperparameters m k , s k ,<br />
µ k , ν k describing respectively the best guess of the<br />
value <strong>and</strong> precision of the cluster means, <strong>and</strong> best<br />
guesses for cluster covariance matrix <strong>and</strong> the<br />
“effective” amount of prior information used to<br />
derive these.<br />
These priors match expert knowledge about<br />
average <strong>and</strong> st<strong>and</strong>ard deviation of each<br />
environmental attribute within each cluster, where<br />
the mean depends on the st<strong>and</strong>ard deviation.<br />
4 SPATIAL ELICITATION TOOL<br />
Eliciting information from experts for input into<br />
Bayesian models requires a blend of psychological<br />
survey design skills, designing questions for<br />
interview, determining who is interviewed <strong>and</strong> how<br />
many times. The challenge is that instead of factual<br />
information, we require knowledge as synthesized<br />
by the expert to be deconstructed <strong>and</strong> quantified in<br />
a form like (5) suitable for input into modelling.<br />
These issues are addressed in the expert elicitation<br />
literature [O’Hagan 1988].<br />
4.1 Design<br />
To this end we have designed a computer assisted<br />
elicitation process that uses a spatial <strong>and</strong> graphical<br />
tool to help the user visualize <strong>and</strong> explore the data.<br />
Essentially the user can interact with data from<br />
various “viewpoints” each with a different activity:<br />
Cartographic: select an existing sub-bioregion<br />
or select attributes to spatially define a “new”<br />
sub-bioregion,<br />
Data exploration: inspect <strong>and</strong> adjust<br />
histograms of each environmental attribute,<br />
Spatial analysis: map several environmental<br />
attributes within the geographic region.<br />
Thus a user can choose to define a sub-bioregion:<br />
in geographic space as a cartographic view or in<br />
variable space as an environmental “domain”. The<br />
aim is to elicit the priors (µ k , Σ k ) for each cluster,<br />
where a cluster corresponds to a sub-bioregion.<br />
The two step process in eliciting these priors is<br />
explained below.<br />
Use the cartographic view to geographically select<br />
areas that characterise each sub-bioregion. This is<br />
typically specified in terms of l<strong>and</strong> classes for<br />
vegetation types, l<strong>and</strong>forms <strong>and</strong> species<br />
distributions. For example, an ecologist may select<br />
areas that form a bio-region made from coastal<br />
lowl<strong>and</strong>s with Banksia open forest. This is carried<br />
k<br />
k<br />
k<br />
930
out in a GIS with custom tools to assist in making<br />
attribute selections. The geographical selections<br />
are used to analyse environmental datasets <strong>and</strong><br />
extract variables within the selected regions.<br />
The data exploration view shows histograms for<br />
the set of environmental variables within the above<br />
geographical selection. These variables are the key<br />
abiotic factors used to classify <strong>and</strong> differentiate<br />
between sub-bioregions. For example, continuous<br />
variables for climate, topography <strong>and</strong> soil<br />
characteristics. An important facet of the data<br />
exploration view is the ability to define thresholds<br />
for variables. For instance the user may clearly<br />
want to eliminate a certain range of values for a<br />
variable (eg particular soil qualities or low rainfall<br />
values). These thresholds can be defined in a<br />
univariate or bivariate fashion. A graphical tool<br />
invoked from the GIS shows adjustable histograms<br />
of several environmental attributes within that<br />
region. This provides hyper-prior parameter<br />
estimates m k ,s k for the mean. Instead of eliciting<br />
the covariance matrix from the user, we use a<br />
sample covariance matrix ϕ k estimated from a subsample<br />
for that region. The user can adjust another<br />
control to set the degrees of freedom (or effective<br />
prior sample size)ν k to reflect certainty in this<br />
matrix. The graphical tool has functions to store<br />
the prior estimates of mean (best guess <strong>and</strong><br />
certainty) as well as the degrees of freedom of the<br />
covariance matrix for each sub-bioregion to be<br />
classified. This “experimental” data along with<br />
other basic metadata is used as priors in the<br />
Bayesian cluster classification.<br />
4.2 Implementation <strong>and</strong> Visualisation Interface<br />
A map-based user interface has been developed<br />
using GIS technology to display parameters as<br />
maps <strong>and</strong> charts. See Figure 5. In the data<br />
exploration view, means are visualized by splitting<br />
the x-axis on histogram <strong>and</strong> slider bar into three<br />
colours. Data symbolization is based upon a<br />
variation of a boxplot to show where credible<br />
intervals are which are then displayed on the map<br />
view [Car et al, 1999]. The slider control allows a<br />
user to adjust class breaks interactively, <strong>and</strong> these<br />
changes are automatically reflected in the colours<br />
displayed on the chart <strong>and</strong> the map cartographic<br />
view. This provides an effective means for a user<br />
to interactively set an estimated credible interval<br />
for single variables within the region.<br />
This information gathered from experts is feed into<br />
the informative priors <strong>and</strong> is recorded as part of an<br />
experimental workbook. Elicitation information<br />
includes the name of the expert, date, remarks,<br />
centre value <strong>and</strong> bounds for each variable<br />
analysed. This information may be used to weight<br />
(e.g. based upon certainty or expert knowledge)<br />
informative priors <strong>and</strong> to document the results of a<br />
classification.<br />
Figure 5. Univariate data visualization.<br />
The graphical interfaces includes a map-based<br />
(cartographic) view <strong>and</strong> a graph-based (data<br />
exploration) view. The user may add up to three<br />
environmental variables as maps. These two views<br />
are linked so that changes in one are<br />
simultaneously reflected in the other. Up to three<br />
maps may be added in this way, <strong>and</strong> then a<br />
combined map or overlay may be created to see the<br />
mapped overlap distributions. The overlap<br />
distribution is representative of the confidence<br />
intervals of the means of the environmental<br />
attributes of a sub-bioregion. The expert can then<br />
interact with the map to add or remove areas from<br />
the sub-bioregion. Hence, the cartographic view<br />
<strong>and</strong> exploration view are dynamically linked. The<br />
expert may view another graphical interface for<br />
exploration of combinations of the variables. A<br />
map <strong>and</strong> a scatter diagram with a background<br />
density frequency chart are displayed for<br />
combinations of the two intervals attributes in two<br />
dimensions. The co-occurrence of related variables<br />
show up as clusters which the expert can refine by<br />
selecting the representative center or mean µ in two<br />
dimension, <strong>and</strong> an area around this center<br />
representing the credible interval. This information<br />
is also saved as prior information for the classifier.<br />
5. RESULTS<br />
The approach may be validated visually against the<br />
existing sub-bioregions by fitting mixture models<br />
to the most significant environmental variables <strong>and</strong><br />
a comparison made to see what adjustments are<br />
suggested by the resulting clusters. Figure 6 shows<br />
the results of this computation for south-east<br />
Queensl<strong>and</strong> <strong>and</strong> it is seen that the adjustments are<br />
minor. The most significant variables used in the<br />
analysis were selected statistically with a<br />
dimension reduction technique. In our south-east<br />
Queensl<strong>and</strong> case study we were able to adequately<br />
931
fit mixture models to the existing sub-bioregions<br />
with a manageable number of topographic, climate<br />
<strong>and</strong> soil variables. Conformance with the existing<br />
sub-bioregions could be effectively controlled by<br />
manipulation of the relative weights placed on the<br />
priors <strong>and</strong> data variables (Figure 6).<br />
(a) (b) (c)<br />
Figure 6. The existing sub-bioregions for southeast<br />
Queensl<strong>and</strong> shown by solid lines on Bayesian<br />
mixture model classifications with: (a) no prior, (b)<br />
moderate weighting on priors, <strong>and</strong> (c) strong<br />
weighting on priors. The priors were calculated<br />
from the existing sub-bioregions.<br />
6. CONCLUSION<br />
The significance of the research is that a Bayesian<br />
approach allows us to combine qualitative<br />
information <strong>and</strong> quantitative data in classification.<br />
Hence combining - the previously competing -<br />
approaches of expert panel <strong>and</strong> data classification.<br />
Bayesian mixture models provide a method for<br />
classifying ecoregions with a formal statistical<br />
procedure that fits overlapping clusters. When<br />
mapped spatially the cluster components relate<br />
well to coherent bio-regions. This is illustrated in<br />
Figure 6, which also shows the results of adjusting<br />
the relative weightings on expert knowledge <strong>and</strong><br />
data between bio-regions. Our elicitation tool<br />
enables experts to interactively specify quantitative<br />
model parameters (e.g. means <strong>and</strong> covariance<br />
matrices) by viewing <strong>and</strong> manipulating familiar<br />
entities such as maps <strong>and</strong> histograms. The results<br />
are presented visually in this paper; future work<br />
will provide details on model diagnostics, model<br />
performance, <strong>and</strong> model comparisons.<br />
7. REFERENCES<br />
Environment Australia, Revision of the Interim<br />
Biogeographic Regionalisation for Australia<br />
(IBRA) <strong>and</strong> Development of Version 5.1.<br />
Summary Report, Canberra, Environment<br />
Australia, 2000.<br />
Bunce, R., P. Carey, R. Elena-Rossello, J. Orr, J.<br />
Watkins <strong>and</strong> R. Fuller, A comparison of<br />
different biogeographical classifications of<br />
Europe, Great Britain <strong>and</strong> Spain, Journal of<br />
<strong>Environmental</strong> Management 65, 121-134,<br />
2002<br />
Carr, D., A. Olsen, S. Pierson <strong>and</strong> J-Y. Courbois,<br />
Boxplot Variations in a Spatial Context.<br />
Statistical Computing & Statistical Graphics<br />
Newsletter, American Statistical Association,<br />
1999.<br />
Congdon, P. Bayesian Statistical <strong>Modelling</strong>,<br />
Wiley, New York, 2001.<br />
Dempster, AP., N. Laird <strong>and</strong> D. Rubin, Maximum<br />
likelihood from incomplete data via the EM<br />
algorithm (with discussion), J Roy Statist Soc<br />
Ser B, 39: 1-38, 1977.<br />
Diebolt, J., <strong>and</strong> C. Robert, Estimation of finite<br />
mixture distributions through Bayesian<br />
sampling, J. Roy. Statist. Soc. Ser. B, 56(2),<br />
363-375, 1994.<br />
Everitt, B., <strong>and</strong> D. H<strong>and</strong>, Finite mixture<br />
distributions, London: Chapman <strong>and</strong> Hall,<br />
1981.<br />
Gelman, A., J. Carlin, H. Stern, <strong>and</strong> D. Rubin,<br />
Bayesian Data Analysis, 2 nd edition.<br />
Chapman <strong>and</strong> Hall/CRC, Florida., 2004<br />
Hargrove, W., <strong>and</strong> F. Hoffman, Using multivariate<br />
clustering to characterize ecoregion borders,<br />
Computing in Science <strong>and</strong> Engineering 1(4),<br />
18-25, 1999.<br />
Morgan, G., <strong>and</strong> J. Terrey, Natural regions of<br />
western New South Wales <strong>and</strong> their use for<br />
environmentl management, Proc Ecol Soc<br />
Aust 16, 467-473, 1990.<br />
Neldner, VJ. Summary of procedure for creating<br />
regional ecosystem maps as defined under<br />
the Vegetation Management Act 1999.<br />
Technical Report. Brisbane: <strong>Environmental</strong><br />
Protection Agency, 2002.<br />
O’Hagan, A., Elicitation of expert beliefs in<br />
substantial practical applications. The<br />
Statistician 47(1), 21–35, 1998.<br />
Peters, D., <strong>and</strong> R. Thackway, A New<br />
Biogeographic Regionalisation for Tasmania.<br />
Hobart, Tasmanian Parks <strong>and</strong> Wildlife, 1998<br />
Sattler, P. <strong>and</strong> R. Williams, The Conservation<br />
Status of Queensl<strong>and</strong>’s Bioregional<br />
Ecosystems, <strong>Environmental</strong> protection<br />
Agency, Queensl<strong>and</strong> Government, 1999.<br />
8. ACKNOWLEDGEMENTS<br />
We thank Petra Kuhnert <strong>and</strong> Robert Denham for<br />
helpful early discussions, ideas <strong>and</strong> suggestions.<br />
This work was supported by ARC-SPIRT Grant<br />
C00107484 <strong>and</strong> by industry partner Queensl<strong>and</strong><br />
<strong>Environmental</strong> Protection Agency, Australia.<br />
932
Assessing management systems for the conservation of<br />
open l<strong>and</strong>scapes using an integrated l<strong>and</strong>scape model<br />
approach<br />
M. Rudner a , R. Biedermann a , B. Schröder b , <strong>and</strong> M. Kleyer a<br />
a<br />
L<strong>and</strong>scape Ecology Group, Institute of Biology <strong>and</strong> <strong>Environmental</strong> Sciences, University of Oldenburg,<br />
26111 Oldenburg, Germany, e-mail: michael.rudner@uni-oldenburg.de<br />
b<br />
Institute of Geoecology, University of Potsdam,14415 Potsdam, Germany<br />
Abstract: The aim of the MOSAIK-project is to test alternative management systems regarding their<br />
efficiency in maintaining the characteristic species composition of dry grassl<strong>and</strong>s. We present an integrated<br />
l<strong>and</strong>scape model approach to test an alternative management system for applicability in preserving dry<br />
grassl<strong>and</strong>s. By rototilling, i.e. cyclic, massive disturbance in the vegetation cover, we established a controlled<br />
mosaic cycle comprising a successional series from heavily disturbed areas to grassl<strong>and</strong> <strong>and</strong> shrubs. The<br />
disturbance regime affects the l<strong>and</strong>scape on different temporal <strong>and</strong> spatial scales. The resulting shifting<br />
mosaics determine the habitat qualities for plant <strong>and</strong> animal species. Changes in habitat quality may reduce<br />
the survival of local or regional populations. To predict the local <strong>and</strong> regional risk of extinction of specific<br />
plant <strong>and</strong> animal functional types, we apply modelling approaches on different scales <strong>and</strong> levels of hierarchy.<br />
We achieve to integrate different modules regarding abiotic <strong>and</strong> biotic state variables, processes <strong>and</strong> complex<br />
interactions in a spatially explicit way into the MOSAIK l<strong>and</strong>scape model, implementing static as well as<br />
dynamic model approaches. The parameters <strong>and</strong> data necessary for reliable modelling were determined<br />
empirically in two study sites in Germany. Subsystems of the overall model are empirically parameterized <strong>and</strong><br />
validated by means of extensive field surveys. The MOSAIK l<strong>and</strong>scape model is still in development. In this<br />
paper we give an overview on the proposed l<strong>and</strong>scape model approach <strong>and</strong> show the general structure of the<br />
MOSAIK l<strong>and</strong>scape model. Preliminary results are exemplified in respect to habitat modelling <strong>and</strong> economic<br />
modelling of two simple management scenarios.<br />
Keywords: L<strong>and</strong>scape modelling; Cyclic disturbance; Management costs; Shifting mosaic of habitat quality;<br />
Integrated modelling<br />
1. INTRODUCTION<br />
The structural change in Central Europe’s<br />
agriculture causes a loss of species rich ecosystems<br />
that depend on traditional l<strong>and</strong> use [Poschlod <strong>and</strong><br />
Schumacher, 1998; Waldhardt et al., 2003]. In<br />
most regions the agricultural practise has been<br />
intensified. Instead traditional (extensive) practise<br />
to preserve open l<strong>and</strong>scapes, expensive l<strong>and</strong>scape<br />
conservation measures like mowing are currently<br />
applied. Consequently, it would be generally<br />
desirable to shift from static costly conservation to<br />
dynamic, more cost-effective management regimes.<br />
To minimise these costs, we examine free grazing<br />
as well as infrequent rototilling as alternative<br />
management systems characterised by an artificial<br />
disturbance regime.<br />
Both systems are characterised by secondary<br />
successions which are periodically reset by small<br />
scale disturbance events. Therefore, the alternative<br />
regimes proposed results in a mosaic of habitat<br />
qualities for plant <strong>and</strong> animal species shifting in<br />
space <strong>and</strong> time. The species’ habitats in this mosaic<br />
cycle become dynamic with respect to location <strong>and</strong><br />
time frame affecting colonisation rate <strong>and</strong><br />
persistence probability. The alternative systems<br />
contrast the classical conservation by cutting that<br />
conserves low <strong>and</strong> closed vegetation cover <strong>and</strong><br />
does not allow periodical succession.<br />
Before recommending the proposed cyclic<br />
disturbance regimes as an alternative to traditional<br />
conservation measures, a number of questions<br />
concerning regional species persistence <strong>and</strong><br />
(inter-)relationships between management, abiotic<br />
933
conditions <strong>and</strong> biotic response have to be<br />
answered. Only if the species’ requirements <strong>and</strong><br />
attributes meet the long-term spatio-temporal<br />
pattern of habitat quality in this mosaic cycle, the<br />
dynamic management regime proposed may serve<br />
as a cost-efficient alternative.<br />
We empirically studied rototilled <strong>and</strong> traditionally<br />
managed plots on the l<strong>and</strong>scape scale to analyse<br />
these management regimes regarding to their<br />
conservational <strong>and</strong> economical efficiency in preserving<br />
the species richness of dry grassl<strong>and</strong>s<br />
[Kleyer at al., 2002]. We regionalised our findings<br />
by applying modelling approaches on different<br />
scales <strong>and</strong> levels of hierarchy to assess the risk of<br />
extinction of plant <strong>and</strong> animal species. Therefore,<br />
we integrate static <strong>and</strong> dynamic modules regarding<br />
abiotic <strong>and</strong> biotic state variables, processes <strong>and</strong><br />
interactions into a spatially explicit l<strong>and</strong>scape<br />
model. There are several examples of successful<br />
applications of l<strong>and</strong>scape models for equivalent<br />
tasks, especially in forest ecology <strong>and</strong> management<br />
[e.g. Kurz et al., 2000; Li et al. 2000; Liu <strong>and</strong><br />
Ashton, 1998]. Other l<strong>and</strong>scape models explicitly<br />
evaluate the effect of management scenarios on<br />
habitat quality [Gaff et al., 2000; Li et al., 2000]<br />
<strong>and</strong> population persistence [Cousins et al., 2003] of<br />
species.<br />
2. MOSAIK LANDSCAPE MODEL<br />
2. 1 Introduction<br />
The MOSAIK l<strong>and</strong>scape model was implemented<br />
in Borl<strong>and</strong> Delphi <strong>and</strong> integrates several abiotic<br />
<strong>and</strong> biotic modules (see below <strong>and</strong> Fig. 1), based<br />
on a simple grid-based Geographic Information<br />
System (GIS). Hence, the different modules are<br />
coupled by a GIS. An interface to ESRI ArcView®<br />
enables the import <strong>and</strong> export of digital maps.<br />
Each module was empirically parameterised <strong>and</strong><br />
validated by means of extensive field surveys.<br />
Combining the modules the l<strong>and</strong>scape model<br />
allows:<br />
i) scaling <strong>and</strong> regionalisation, i.e. extrapolating<br />
surveys <strong>and</strong> predicted probabilities of occurrence<br />
from plot scale to l<strong>and</strong>scape scale,<br />
ii) spatially explicit modelling of processes,<br />
interactions <strong>and</strong> interdependencies between different<br />
abiotic <strong>and</strong> biotic features, <strong>and</strong><br />
iii) assessing the ecological consequences as well<br />
as the socio-economic costs of management scenarios<br />
regarding rototilling <strong>and</strong> traditional mowing.<br />
The management regimes consider the frequency,<br />
spatial extent <strong>and</strong> temporal sequence of rototilling<br />
measures. It depends on the regime if rototilling<br />
can be considered a cost-effective alternative for<br />
the conservation of open dry grassl<strong>and</strong>s that helps<br />
to preserve the specific species composition.<br />
2.2 Model structure<br />
The MOSAIK l<strong>and</strong>scape model comprises the<br />
following modules (see also Fig. 1):<br />
i) Maps<br />
Maps of e.g. elevation, slope, aspect, etc. derived<br />
by means of digital terrain analysis on the basis of<br />
a digital elevation model.<br />
i) Abiotic models<br />
A soil-l<strong>and</strong>scape model providing information on<br />
soil properties that determine soil-water conditions,<br />
i.e. statistical analysis of the spatial distribution of<br />
soil properties with respect to soil samples <strong>and</strong><br />
their position in the terrain.<br />
ii) Habitat models<br />
Statistical habitat models predicting the shifting<br />
mosaic of habitat qualities for plant <strong>and</strong> animal<br />
species as well as the spatial distribution of these<br />
species.<br />
iv) Economic models<br />
Financial models calculating the costs of the<br />
management scenarios regarding the time schedule<br />
<strong>and</strong> spatial regime of rototilling <strong>and</strong> traditional<br />
mowing.<br />
Figure 1. Internal structure of the MOSAIK<br />
l<strong>and</strong>scape model.<br />
934
3. CASE STUDY: LEY LANDSCAPE IN<br />
THE NATURE RESERVE “HOHE<br />
WANN”, SOUTHERN GERMANY<br />
3.1 Study area <strong>and</strong> data sources<br />
The empirical studies in order to parameterise the<br />
MOSAIK l<strong>and</strong>scape model have been carried out<br />
from 2000 to 2003 in the nature reserve “Hohe<br />
Wann”. It is located in the Hassberge area in<br />
Lower Franconia, Germany (50°03‘ N, 10°35‘ O,<br />
see Fig. 2) that belongs to the “Fränkische<br />
Schichtstufenl<strong>and</strong>schaft / franconian escarpment<br />
l<strong>and</strong>scape”.<br />
Figure 3. Map of habitat types within the nature<br />
reserve “Hohe Wann” in the Hassberge area.<br />
3.2 Scenarios<br />
In order to test the habitat modelling module <strong>and</strong><br />
the socio-economic module of the proposed l<strong>and</strong>scape<br />
model approach we developed two rototilling<br />
scenarios (cf. Fig. 6) as examples for more<br />
complex scenarios:<br />
i) Scenario SiLa (single large): rototilling of a<br />
single contiguous patch with an area of ca 7 ha.<br />
ii) Scenario SeSma (several small): rototilling of<br />
16 scattered patches with an area of ca 4 ha.<br />
Figure 2. Map of Germany with Hassberge<br />
study area.<br />
The area of investigation with an extent of about<br />
7 x 3 km² is characterised by heterogeneous geological<br />
substrates, i.e. triassic s<strong>and</strong> <strong>and</strong> gypsum<br />
keuper as well as the traditional system of<br />
inheritance by equal division. South-facing slopes<br />
that receive higher-than-average insolation are<br />
either used as vineyards or they are fallow l<strong>and</strong><br />
after ab<strong>and</strong>onment. They can be characterised as a<br />
mosaic of dry grassl<strong>and</strong>s <strong>and</strong> shrubs within a<br />
matrix of arable l<strong>and</strong> <strong>and</strong> forestry (see Fig. 3).<br />
The surveys of habitat types, l<strong>and</strong> use <strong>and</strong> soil<br />
characteristics were carried out between 2000 <strong>and</strong><br />
2002. Data sets regarding the incidence of plant<br />
<strong>and</strong> animal species as well as habitat features were<br />
carried out on 120 plots following a stratified<br />
r<strong>and</strong>om sampling design. [Hein et al., submitted].<br />
3.3 Preliminary results<br />
3.3.1 Habitat modelling<br />
The l<strong>and</strong>scape model enables the application of<br />
habitat models to different disturbance scenarios.<br />
Habitat models quantify habitat quality in respect<br />
to environment. We used logistic regression<br />
[Hosmer & Lemeshow, 2000] to formulate the<br />
habitat models [e.g. Hein et al., 2003; Kühner &<br />
Kleyer, 2003]. Based on maps of environmental<br />
variables (like habitat type, soil properties, l<strong>and</strong><br />
use, slope, aspect, insolation, wetness index,<br />
amount of plant available soil water between April<br />
to June etc.) we use the habitat models to calculate<br />
the probability of occurrence for the entire study<br />
area, i.e. we perform a spatial extrapolation from<br />
our 120 sample plots. These maps of the<br />
probability of occurrence can be calculated for<br />
single species, species groups or functional types<br />
[Bonn <strong>and</strong> Schröder, 2001; Kleyer, 2002]. Further,<br />
935
these maps may be transformed to maps showing<br />
matrix versus suitable habitat using classification<br />
thresholds. These patch maps may be used for the<br />
analysis of the effects of spatial configuration (e.g.<br />
area, connectivity) on the incidence [e.g. Keitt et<br />
al., 1997; Schröder, 2000] or (meta-)population<br />
dynamics of species [Biedermann, 2000; Söndgerath<br />
<strong>and</strong> Schröder, 2002].<br />
Although, habitat models assume equilibrium<br />
conditions, there are some issues that allow their<br />
application in a dynamic context: applying spacefor-time<br />
substitution [Pickett, 1989] we use timedependent<br />
predictor variables. Predictors directly<br />
describing the disturbance regime in terms of<br />
frequency as well as depth of disturbance integrate<br />
over longer time periods but they directly affect the<br />
soil water balance according to their dynamics.<br />
Bare soil after rototilling differs in evaporation rate<br />
compared to vegetated soil. This aspect is taken<br />
into account when calculating time-dependent<br />
predictors (e.g. amount of plant available soil<br />
water between April <strong>and</strong> June).<br />
sigmoidal. Since the regression coefficient is<br />
negative, the species was found to avoid soils with<br />
high air capacity, that dry fast.<br />
Figure 5. Application of an exemplary habitat<br />
model for Thlaspi perfoliatum: maps of predictor<br />
variables (left), regression equation (top right),<br />
response surface <strong>and</strong> derived map of predicted<br />
occurrence probabilities (bottom right).<br />
12<br />
Frequency<br />
10<br />
8<br />
6<br />
4<br />
2<br />
0<br />
| | | | || | ||| ||| | ||| | | || ||||| | |||| | || || | | ||| || | | | |<br />
0.70 0.75 0.80 0.85 0.90 0.95 1.00<br />
AUC<br />
Figure 4. Performance of the habitat models of 52<br />
plants species: frequency distribution of levels of<br />
AUC-values.<br />
We modelled the probability of occurrence of 52<br />
plant species <strong>and</strong> a considerable number of habitat<br />
models with good performance (Fig. 4). As a case<br />
species the annual plant Thlaspi perfoliatum was<br />
chosen (Fig. 5 depicts the steps in applying the<br />
habitat model). The species’ spatial distribution<br />
was found to depend on the frequency of<br />
disturbance <strong>and</strong> air capacity of the top soil (maps<br />
in Fig. 5) [Kühner <strong>and</strong> Kleyer, 2003]. The model<br />
showed a comparatively good performance<br />
(Nagelkerke-R² = 0.305 <strong>and</strong> AUC = 0.780). The<br />
species showed an unimodal response regarding<br />
the disturbance frequency, meaning that the<br />
probability of occurrence reached its maximum for<br />
intermediate frequencies (around once per year,<br />
what is expected for an annual plant). The response<br />
with respect to the second predictor variables is<br />
Table 1. Maps of predicted occurrence probabilities<br />
for Thlaspi perfoliatum regarding the rototilling<br />
scenarios given in Fig. 6 compared to the<br />
recent situation (regular mowing).<br />
Probability P<br />
(occurrence Thlaspi<br />
perfoliatum)<br />
Proportion rototilled<br />
(total area =<br />
108 ha)<br />
Habitat units<br />
(P × area)<br />
Status quo<br />
Scenario<br />
SiLa<br />
Scenario<br />
SeSma<br />
0% 6.5 % 3.7 %<br />
269247<br />
(100 %)<br />
295425<br />
(110 %)<br />
276582<br />
(103 %)<br />
To include the dynamic aspects related to the<br />
management applied we used results of frequency<br />
analyses conducted on experimental plots (Fritzsch<br />
et al., in prep.): if a species revealed significant<br />
increase or decrease in the first two years after<br />
management, we increased or decreased the<br />
probabilties of occurrence estimated by means of<br />
the habitat models.<br />
The application of the habitat model onto the two<br />
scenarios shown in Fig. 6, the spatial distribution<br />
of habitat quality changes. Overall, Thlaspi per-<br />
936
foliatum would benefit from rototilling (Table 1<br />
compares some summary measures <strong>and</strong> the derived<br />
maps of occurrence probabilities). Both scenarios<br />
yield higher habitat units.<br />
3.3.2 <strong>Modelling</strong> of management costs<br />
In two scenarios SiLa <strong>and</strong> SeSma the costs of<br />
rototilling were modelled in a spatially explicit<br />
way. The calculation of the costs of the rototilling<br />
is based on parameters like frequency (e.g. each<br />
year or every 5 years), effective working time, time<br />
for preparation of machines, labour costs, capital<br />
costs, costs for farm machines [after Kuratorium<br />
für Technik und Bauwesen in der L<strong>and</strong>wirtschaft,<br />
1998]. The effective working time depends on site<br />
parameters like area, slope, soil type, reachability<br />
<strong>and</strong> distance to the next site or farm. As at steep<br />
slopes rototilling has to process upwards<br />
additionally the orientation of the sites with respect<br />
to slope <strong>and</strong> thus the length of the possible<br />
rototilling tracks is relevant. Short tracks require<br />
frequent turning of the machines.<br />
Both scenarios imply costs of almost 7000 €/a (see<br />
Fig. 6), however, the area-dependent costs are<br />
almost twice as large in the second scenario due to<br />
higher time budgets (frequent transposing, higher<br />
relative amount of fixed costs).<br />
systems like rototilling. The application of the<br />
l<strong>and</strong>scape model seems especially relevant in<br />
situations where the development of sites should be<br />
confronted with the costs of the development, as a<br />
large number of scenarios can be evaluated in a<br />
short time period. Further, the l<strong>and</strong>scape model<br />
may be useful for the prediction of future<br />
development within environmental planning<br />
processes (e.g. impact assessment). However,<br />
further developments of the MOSAIK l<strong>and</strong>scape<br />
model, like integration of population dynamic<br />
models or economic models for pasture<br />
management, are necessary in order to achieve<br />
valid predictions of the biodiversity of plants <strong>and</strong><br />
animals as well as management costs.<br />
5. ACKNOWLEDGEMENTS<br />
The authors are grateful to all partners cooperating<br />
in the MOSAIK-project (www.unioldenburg.de/mosaik/mosaik.htm).<br />
The project was<br />
funded by the Federal Ministry of Education <strong>and</strong><br />
Research (FKZ 01 LN 0007). Thanks to M.<br />
Weisensee <strong>and</strong> W. Tecklenburg, University of<br />
Applied Science Oldenburg for their help in<br />
generating the digital elevation model. The aerial<br />
photographs were provided by the Bavarian Survey<br />
Authority. Thanks to the Bavarian State Ministry<br />
for Agriculture <strong>and</strong> Forestry <strong>and</strong> to the department<br />
of agriculture of the city of Würzburg for agrometeorological<br />
<strong>and</strong> soil data provided online.<br />
Figure 6. Comparison of two rototilling scenarios:<br />
SiLa - one contiguous single patch versus<br />
SeSma - several scattered small patches.<br />
4. CONCLUSION<br />
Based on comprehensive field surveys, the<br />
MOSAIK l<strong>and</strong>scape model aims to integrate<br />
abiotic models, habitat models <strong>and</strong> financial<br />
models. Using the l<strong>and</strong>scape model, in the study<br />
area a number of different management scenarios<br />
can be evaluated in respect to nature conservation<br />
value <strong>and</strong> management costs. The results may build<br />
the basis of decisions concerning the management<br />
of concrete sites, using alternative management<br />
6. REFERENCES<br />
Biedermann, R., Metapopulation dynamics of the<br />
froghopper Neophilaenus albipennis (F.,<br />
1798) (Homoptera, Cercopidae) - what is the<br />
minimum viable metapopulation size?<br />
Journal of Insect Conservation, 4, 99-107,<br />
2000.<br />
Bonn, A. <strong>and</strong> B. Schröder, Habitat models <strong>and</strong><br />
their transfer for single- <strong>and</strong> multi-species<br />
groups: a case study of carabids in an alluvial<br />
forest, Ecography, 24, 483-496, 2001.<br />
Cousins, S.A.O., S. Lavorel, <strong>and</strong> I. Davies,<br />
<strong>Modelling</strong> the effects of l<strong>and</strong>scape pattern<br />
<strong>and</strong> grazing regimes on the persistence of<br />
plant species with high conservation value in<br />
grassl<strong>and</strong>s in south-eastern Sweden, L<strong>and</strong>scape<br />
Ecology, 18, 315-332, 2003.<br />
Gaff, H., D.L. DeAngelis, L.J. Gross, R. Salinas,<br />
<strong>and</strong> M. Shorrosh, A dynamic l<strong>and</strong>scape<br />
model for fish in the Everglades <strong>and</strong> its<br />
application to restoration, Ecological <strong>Modelling</strong>,<br />
127, 33-52, 2000.<br />
Hein, S., Gombert, J., Stanke, C., Voss, J. <strong>and</strong> H.J.<br />
Poethke, Survival of crickets in habitats<br />
937
variable in space <strong>and</strong> time, Verh<strong>and</strong>lungen<br />
der Gesellschaft für Ökologie, 33, 54, 2003.<br />
Hein, S., Voss, J., Schröder, B., <strong>and</strong> H.-J. Poethke,<br />
Habitat suitability models for the conservation<br />
of thermophilic grasshoppers <strong>and</strong> bush<br />
crickets, submitted.<br />
Hosmer, D.W. <strong>and</strong> S. Lemeshow, Applied logistic<br />
regression, Wiley, 2000.<br />
Keitt, T.H., D.L. Urban, <strong>and</strong> B.T. Milne, Detecting<br />
critical scales in fragmented l<strong>and</strong>scapes,<br />
Conservation Ecology [online] 1,<br />
www.consecol.org/vol1/iss1/art4, 1997.<br />
Kleyer, M., Validation of plant functional types<br />
across two contrasting l<strong>and</strong>scapes, Journal of<br />
Vegetation Science 13, 167-178, 2002.<br />
Kleyer, M., R. Biedermann, K. Henle, H.J.<br />
Poethke, P. Poschlod, & J. Settele, MOSAIK:<br />
Semi-open pasture <strong>and</strong> ley - a research<br />
project on keeping the cultural l<strong>and</strong>scape<br />
open. In: Redecker, B., P. Fink, W. Härdtle,<br />
U. Riecken & E. Schröder (eds.), Pasture<br />
L<strong>and</strong>scape <strong>and</strong> Nature Conservation,<br />
Springer, Heidelberg, 399-412.<br />
Kühner, A. <strong>and</strong> M. Kleyer, Habitat models for<br />
plant functional types in relation to grazing,<br />
soil factors <strong>and</strong> fertility, Verh<strong>and</strong>lungen der<br />
Gesellschaft für Ökologie, 33, 248, 2003.<br />
Kuratorium für Technik und Bauwesen in der<br />
L<strong>and</strong>wirtschaft (ed.), L<strong>and</strong>schaftspflege :<br />
Daten zur Kalkulation von Arbeitszeiten und<br />
Maschinenkosten, KTBL-Schriften-Vertrieb<br />
im L<strong>and</strong>wirtschaftsverlag, 1998.<br />
Kurz, W.A., S.J. Beukema, W. Klenner, J.A.<br />
Greenough, D.C.E. Robinson, A.D. Sharpe,<br />
<strong>and</strong> T.M. Webb, TELSA: the tool for<br />
exploratory l<strong>and</strong>scape scenario analyses I,<br />
Computers <strong>and</strong> Electronics in Agriculture,<br />
27, 227-242, 2000.<br />
Li, H., D.I. Gartner, P. Mou, <strong>and</strong> C.C. Trettin, A<br />
l<strong>and</strong>scape model (LEEMATH) to evaluate<br />
effects of management impacts on timber <strong>and</strong><br />
wildlife habitat. Computers <strong>and</strong> Electronics<br />
in Agriculture, 27, 263-292, 2000.<br />
Liu, J. <strong>and</strong> P.S. Ashton, Formosaic: An individualbased<br />
spatially explicit model for simulating<br />
forest dynamics in l<strong>and</strong>scape mosaics,<br />
Ecological <strong>Modelling</strong>, 106, 177-200, 1998.<br />
Pickett, S.T.A., Space-for-time substitution as an<br />
alternative to long-term studies. In: Likens,<br />
G.E. (ed.), Long-term studies in ecology.<br />
Springer, Heidelberg, 110–135, 1989.<br />
Poschlod, P. <strong>and</strong> W. Schumacher, Rückgang von<br />
Pflanzen und Pflanzengesellschaften des<br />
Grünl<strong>and</strong>s - Gefährdungsursachen und H<strong>and</strong>lungsbedarf,<br />
Schriftenreihe für Vegetationskunde<br />
29, 83-99, 1998.<br />
Schröder, B., Zwischen Naturschutz und Theoretischer<br />
Ökologie: Modelle zur Habitateignung<br />
und räumlichen Populationsdynamik für Heuschrecken<br />
im Niedermoor. Dissertation,<br />
Institut für Geographie & Geoökologie, TU<br />
Braunschweig, 228 pp, 2000.<br />
Söndgerath, D. <strong>and</strong> B. Schröder, Population<br />
dynamics <strong>and</strong> habitat connectivity affecting<br />
spatial spread of populations - a simulation<br />
study. L<strong>and</strong>sacpe Ecology, 17, 57-70, 2002.<br />
Waldhardt, R., D. Simmering <strong>and</strong> H. Albrecht,<br />
Floristic diversity at the habitat scale in<br />
agricultural l<strong>and</strong>scapes of Central Europe -<br />
summary, conclusions <strong>and</strong> perspectives.<br />
Agriculture, Ecosystems <strong>and</strong> Environment,<br />
98, 79-85, 2003.<br />
938
Forecasting UV Index by NEOPLANTA Model:<br />
Methodology <strong>and</strong> Validation<br />
S. Malinoviç a , D.T. Mihailoviç a,b , Z. Mijatoviç a,c , D. Kapor a,c <strong>and</strong> I.D. Arseniç a,b<br />
a University Center for Meteorology <strong>and</strong> <strong>Environmental</strong> <strong>Modelling</strong>, University of Novi Sad,<br />
Novi Sad, Serbia <strong>and</strong> Montenegro, slavicans@polj.ns.ac.yu<br />
b Faculty of Agriculture, Research Institute of Field <strong>and</strong> Vegetable Crops, University of Novi Sad,<br />
Novi Sad, Serbia <strong>and</strong> Montenegro<br />
c Department of Physics, Faculty of Natural Sciences <strong>and</strong> Mathematics, University of Novi Sad,<br />
Novi Sad, Serbia <strong>and</strong> Montenegro<br />
Abstract: A major consequence of decreasing stratospheric ozone is the increase of solar ultraviolet radiation<br />
(UV) passing through the atmosphere. Ultraviolet radiation is very harmful to the entire ecosystem, including<br />
health of the human population <strong>and</strong> for that reason, in the last few decades, scientists have placed a large<br />
emphasis on monitoring UV radiation <strong>and</strong> development <strong>and</strong> use of estimation procedures. Model<br />
NEOPLANTA estimates UV irradiance under cloudless conditions on a horizontal surface <strong>and</strong> computes the<br />
UV index. Model includes effects of the absorption of UV radiation by ozone, SO 2 <strong>and</strong> NO 2 <strong>and</strong> absorption<br />
<strong>and</strong> scattering by aerosol <strong>and</strong> air molecules in the atmosphere. Aerosols are incorporated in model using the<br />
model OPAC that provides optical properties for ten different aerosol types. Surface influence on UV<br />
irradiance was taken into account using spectral albedo values for nine different surface types. Model can use<br />
st<strong>and</strong>ard atmosphere meteorological profiles but it is possible to include real time meteorological data<br />
coming from the high-level resolution mesoscale models. The capability of the model to reproduce correctly<br />
processes in atmosphere is tested by changing input parameters. The performance of the model has been<br />
tested in relation to its predictive capability of global solar irradiance in the UV range (290-400 nm). For this<br />
test we have used data recorded by the radiometer YANKEE UVB-1 biometer located on the Novi Sad<br />
University campus (45.33 o N, 19.85 o E, 84 m a.s.l).<br />
Keywords: Modeling; Irradiance; UV index; Ozone; Aerosol<br />
1. INTRODUCTION<br />
Ultraviolet (UV) radiation is a small part of solar<br />
spectrum, but it has a large impact on human<br />
health. It is defined as radiation between 100 <strong>and</strong><br />
400 nm, <strong>and</strong> can be divided into three categories:<br />
UV-C (100-280 nm) that is not reach the ground,<br />
UV-B (280-320 nm) which is very small but very<br />
harmful part of terrestrial UV spectrum <strong>and</strong> UV-A<br />
(320-400 nm) which is major part of terrestrial UV<br />
spectrum but much less erythmogenic than UV-B<br />
radiation [Diffey, 1991]. A major consequence of<br />
decreasing stratospheric ozone in last several<br />
decades is the increase of solar UV radiation<br />
passing through the atmosphere, especially the<br />
most harmful UV-B part. Although ozone is one of<br />
the major factors in determining the UV-B<br />
irradiance, the UV flux reaching the ground<br />
depends on many other factors such as aerosols<br />
<strong>and</strong> other UV absorbing gases. Underst<strong>and</strong>ing the<br />
processes affecting UV radiation allows scientists<br />
to estimate UV levels when measurements are not<br />
available. UV exposure of the skin depends very<br />
strongly on the behavior of the human being. It can<br />
be reduced to quite a small extent in many cases by<br />
providing daily information about current values<br />
<strong>and</strong> expected values for the next day(s). The easily<br />
understood parameter that describes potential<br />
detrimental effects on health from UV radiation is<br />
UV index. For that reason UV index is<br />
recommended by several international institutions<br />
<strong>and</strong> widely used to inform public.<br />
This paper describes the NEOPLANTA model,<br />
which computes the UV index under cloudless<br />
conditions on a horizontal surface. Model<br />
simulates the effects of the absorption of UV<br />
radiation by ozone, SO 2 <strong>and</strong> NO 2 <strong>and</strong> absorption<br />
<strong>and</strong> scattering by aerosol <strong>and</strong> air molecules in the<br />
atmosphere. In order to investigate the performance<br />
of the model, we have tested model by<br />
changing input parameters <strong>and</strong> compared the<br />
computed results with clear sky solar UV index<br />
measured in Novi Sad.<br />
939
͌<br />
single<br />
2. MODEL DESCRIPTION<br />
The numerical model NEOPLANTA computes the<br />
solar direct <strong>and</strong> diffuse UV irradiances under<br />
cloud-free conditions for the wavelength range<br />
280-400 nm, <strong>and</strong> computes UV index. The UV<br />
irradiances can be computed at any location at<br />
different altitudes <strong>and</strong> times of day.<br />
The global irradiance is the sum of the direct <strong>and</strong><br />
the diffuse components. The values of the direct<br />
<strong>and</strong> diffuse irradiances can be calculated separately<br />
<strong>and</strong> can be integrated over any wavelength range<br />
with resolution of 1 nm. Model divides atmosphere<br />
into maximum 40 layers <strong>and</strong> takes into account its<br />
curvature. The vertical resolution of the model is<br />
one kilometre for altitudes less than 25 kilometres,<br />
<strong>and</strong> 5 kilometres above this height. Direct <strong>and</strong><br />
diffuse intensity are computed at lower limit of<br />
each layer. Calculation of the direct part of<br />
radiation is carried out by the Beer-Lambert law.<br />
The starting point for calculation of diffuse part of<br />
radiation is the set of equations from Bird <strong>and</strong><br />
Riordan spectral model [1986], which represents<br />
equations from previous parametric models<br />
[Leckner, 1978; Brine <strong>and</strong> Iqbal, 1983; Justus <strong>and</strong><br />
Paris, 1985] improved after comparisons with<br />
rigorous radiative transfer model <strong>and</strong> with<br />
measured spectra. In contrast to mentioned models<br />
our model ignores the absorption by O 2 , CO 2 <strong>and</strong><br />
H 2 O since they not absorb in the relevant<br />
wavelength range. However the absorption by SO 2 ,<br />
NO 2 was added for both the direct <strong>and</strong> diffuse<br />
components.<br />
The required input parameters are the local<br />
geographic coordinates <strong>and</strong> time or solar zenith<br />
angle, altitude <strong>and</strong> the amount of gases. Aerosols<br />
are incorporated in model using the model OPAC<br />
that provides optical properties for ten different<br />
aerosol types [Hess at al., 1998]. Surface influence<br />
on UV irradiance was taken into account using<br />
spectral albedo values for nine different surface<br />
types. The model includes its own vertical gas<br />
profiles <strong>and</strong> extinction cross-sections,<br />
extraterrestrial solar irradiance, aerosol optical<br />
properties <strong>and</strong> spectral surface reflectivity. The<br />
model uses st<strong>and</strong>ard atmosphere meteorological<br />
profiles but it has possibility of including real time<br />
meteorological data coming from the high-level<br />
resolution mesoscale models.<br />
Output data are UV-A <strong>and</strong> UV-B intensity,<br />
biologically active UV irradiance at the surface,<br />
UV index, spectral direct, diffuse <strong>and</strong> global<br />
irradiance, total spectral optical depth <strong>and</strong> values<br />
for each component <strong>and</strong> spectral transmitivity. All<br />
values can be obtained at the surface <strong>and</strong> at the<br />
lower limit of each layer.<br />
2.1 Extraterrestrial Source Spectra<br />
The model uses the extraterrestrial spectral<br />
irradiance from the solar flux atlas of Kurucz et al.<br />
[1984]. The ellipticity of the Earth’s orbit about<br />
the sun is considered, as a correction to the<br />
extraterrestrial solar spectrum.<br />
2.2 Sun’s Position <strong>and</strong> Optical Masses<br />
The model NEOPLANTA calculates instantaneous<br />
spectral irradiance for a given solar zenith angle.<br />
There is also possibility of calculation of the UV<br />
index for the whole day at half-hour intervals from<br />
sunrise to sunset. Solar zenith angle depends on<br />
latitude, longitude, day of the year <strong>and</strong> time of day.<br />
The zenith angle is derived using spherical<br />
trigonometry [Spencer, 1971]. Model has<br />
possibility taking into account daylight saving<br />
time.<br />
optical mass, the optical mass for air<br />
molecules or “air mass”, is used to estimate the<br />
total slant path for all the extinction processes in<br />
the atmosphere, except ozone. Optical mass that is<br />
taking into account Earth curvature <strong>and</strong> refraction<br />
is calculated using formula proposed by Hardie in<br />
Hiltner [1962]. Different expressions for ozone<br />
optical mass is considered here because ozone<br />
extinction process corresponds to a different<br />
vertical concentration profile [Komhyr, 1980].<br />
2.3 Gas Absorption<br />
Extinction of UV radiation by gases is calculated<br />
by the product of the cross-sectional area <strong>and</strong> the<br />
particle concentration for each atmospheric layer:<br />
Ozone is the most important gas in the atmosphere<br />
that absorbs UV radiation. Ozone extinction crosssection<br />
values as a function of wavelength <strong>and</strong><br />
temperature were obtained from Burrows at al.<br />
[1999] for the wavelength range 280-400 nm.<br />
Values are given for five temperatures, from 202 K<br />
to 293 K, <strong>and</strong> for particular layer estimated by a<br />
linear interpolation. Particle concentration is<br />
calculated by combining vertical profiles <strong>and</strong> total<br />
gases amount. Model uses four ozone profiles that<br />
are representative for seasons in mid-latitudes.<br />
Forecasted total ozone amounts for 24, 48 <strong>and</strong> 72<br />
hours are provided by German National Service.<br />
Extinction cross section SO 2 <strong>and</strong> NO 2 as a function<br />
of wavelength <strong>and</strong> temperature were obtained from<br />
940
̄<br />
̄<br />
is<br />
SCIAMACHY spectrometer measurements for the<br />
280-400 nm wavelength range [Bogumil at al.,<br />
2000]. SO 2 <strong>and</strong> NO 2 profiles are used from<br />
Nakajima <strong>and</strong> Tanaka [1986]. The model uses the<br />
total dioxides content of the day before under the<br />
assumption of persistency. In the case of<br />
unavailability of actual measurements long-term<br />
mean values are used instead.<br />
2.4 Aerosols Extinction<br />
As mentioned before ten different aerosol mixtures<br />
that are representative of a boundary layer of<br />
certain origin from OPAC aerosol model are<br />
available in the NEOPLANTA model. These types<br />
differ from one another in the way their scattering<br />
efficiency, single scattering albedo <strong>and</strong> asymmetry<br />
factors vary with wavelength. The software<br />
package OPAC also gives optical properties of<br />
upper atmosphere aerosol, which are representative<br />
of free troposphere (boundary layer-12 km) <strong>and</strong><br />
stratospheric aerosol properties (12-36 km). OPAC<br />
also describes the vertical distribution of aerosol<br />
particles by exponential profile [Hess at al., 1998].<br />
To estimate amount of aerosols in layer at the<br />
ground model NEOPLANTA permits different<br />
possibilities. The user can choose the averaged<br />
conditions provided by OPAC aerosol model,<br />
Angstrom’s turbidity coefficient [Angstrom,<br />
1961], visibility [Koschmieder, 1924; Gueymard,<br />
1995] or aerosol optical depth on 550 nm.<br />
2.5 UV index<br />
To provide the public with easily understood<br />
information about UV radiation <strong>and</strong> its harmful<br />
effects, scientists are defined UV index (UVI). UV<br />
index is related with erythemal effects of UV<br />
radiation on human skin <strong>and</strong> it is st<strong>and</strong>ardised<br />
under the umbrella of several international<br />
institutions [Vanicek, 1999]. A unit of UV index<br />
corresponds to 25 mWm -2 biologically active UV<br />
radiation (UV bio ) <strong>and</strong> it is defined as:<br />
UVI = UVbio × 40 .<br />
To estimate biologically active UV radiation,<br />
spectral irradiance at the surface must be weighted<br />
with an action spectrum. An action spectrum<br />
describes the relative efficiency of UV radiation at<br />
a particular wavelength in producing a particular<br />
biological response.<br />
Figure 1 shows normalized erythemal action<br />
spectrum by McKinley <strong>and</strong> Diffey [1987], spectral<br />
UV irradiance, <strong>and</strong> its spectral overlap. The solid<br />
lines are for a total ozone column of 348 DU <strong>and</strong><br />
the thin lines for 250 DU. It can be seen that the<br />
overlap is greatest in the 300-320 nm range <strong>and</strong> is<br />
very sensitive to ozone amounts [Madronich at al.,<br />
1998]. The potential biologically active UV<br />
irradiance (UV bio ) at the surface is found by a<br />
multiplication of the UV spectrum <strong>and</strong> the action<br />
spectrum, <strong>and</strong> integration between 290 <strong>and</strong> 400<br />
nm:<br />
UVbio<br />
= B I d̄<br />
where<br />
∫<br />
B is ̄<br />
normalized erythemal action<br />
spectrum, I is spectral UV irradiance <strong>and</strong> ̄ ̄<br />
wavelength.<br />
Figure 1. Biologically active UV radiation. The<br />
overlap between the spectral irradiance I(̄) <strong>and</strong><br />
the erythemal action spectrum B(̄) given by<br />
McKinlay <strong>and</strong> Diffey shows the spectrum of<br />
biologically active radiation I(̄) B(̄) [Madronich<br />
at al., 1998].<br />
3. RESULTS AND DISCUSSION<br />
3.1 Sensitivity Studies<br />
The capability of the model to reproduce correctly<br />
processes in atmosphere is tested by changing<br />
input parameters such as ozone <strong>and</strong> dioxides<br />
content, solar zenith angle, amount <strong>and</strong> type of<br />
aerosols <strong>and</strong> altitude.<br />
Under cloud-free skies UV index depends largely<br />
on solar zenith angle <strong>and</strong> ozone content. Model<br />
results show decreasing UV index with increasing<br />
solar zenith angle <strong>and</strong> increasing of diffuse<br />
component of radiation. The response of UV-B<br />
radiation to ozone changes is strongly dependent<br />
on wavelength because of the rapid increase of the<br />
ozone absorption cross section toward shorter<br />
wavelengths, with greater sensitivity at short<br />
941
wavelengths <strong>and</strong> low sun, where the slant ozone<br />
optical depth is greater [Madronich at al., 1998].<br />
As can be seen on Figure 2, biologically weighted<br />
radiation calculated by model NEOPLANTA<br />
shows the theoretically expected dependence on<br />
ozone <strong>and</strong> it is in accordance with measured values<br />
showed in Madronich at al. [1998].<br />
Biologically weighted UV<br />
radiation change (%)<br />
200<br />
150<br />
100<br />
50<br />
0<br />
-60 -50 -40 -30 -20 -10 0<br />
Ozon change (%)<br />
Figure 2. Dependence of erythemal ultraviolet<br />
radiation calculated by NEOPLANTA model at<br />
fixed solar zenith angles on atmospheric ozone<br />
SO 2 absorbs predominantly in the UV-B region<br />
while NO 2 except UV-B region has also noticeable<br />
absorption in UV-A part of spectrum. Model is<br />
estimating only a minor effect of SO 2 on<br />
decreasing global UV irradiance (about 2% for<br />
wavelengths below 300nm for 1 DU gas<br />
increasing). At UV-A part of spectrum model<br />
estimates less than 1% influence NO 2 on global<br />
UV irradiance.<br />
Atmospheric aerosols are characterized by their<br />
amounts <strong>and</strong> chemical composition. Model results<br />
showed decreasing direct <strong>and</strong> increasing diffuse<br />
UV component with aerosol amounts. Diffuse<br />
component increases with relative humidity, that is<br />
result of increasing aerosol radius. Aerosol<br />
influence on UV irradiance is equalized in whole<br />
UV spectrum.<br />
scattered broadb<strong>and</strong> ultraviolet irradiance from the<br />
hemisphere of the sky <strong>and</strong> calculates UV index.<br />
The spectral sensitivity of the device is similar to<br />
the human skin. The measurement technique<br />
employs colored glass filters in combination with<br />
fluorescing ultraviolet-sensitive phosphorus.<br />
Measurement errors introduced by changes in<br />
ambient temperature are significantly reduced<br />
[Dichter at al., 1992]. The used device is located<br />
on the Novi Sad University campus (45.33 o N,<br />
19.85 o E, 84 m a.s.l). The measurements are<br />
collected with a temporal resolution of 10 minute.<br />
Nearly clear sky conditions were observed on<br />
several days during the summer of year 2003. UV<br />
index is calculated every half hour from sunrise to<br />
sunset by means of variation in solar zenith angle.<br />
St<strong>and</strong>ard atmosphere meteorological profiles are<br />
employed with summer humidity profile. All input<br />
parameters were assumed to be constant over the<br />
day. On Figure 3 measured values are presented by<br />
solid line <strong>and</strong> forecasted by rhombus.<br />
In the presented comparison, total ozone content of<br />
the atmosphere is the input parameter that has to<br />
be known, so therefore we consider that ozone<br />
does not represent a great source of uncertainty.<br />
Total column ozone over the Novi Sad coordinates<br />
for these days was taken from the on-line database<br />
of TOMS Meteor-3 observations. There exist no<br />
aerosol chemical composition <strong>and</strong> amount<br />
measurements in Novi Sad so we consider aerosol<br />
to be main source of difference between measured<br />
<strong>and</strong> calculated values. Due to a large portion of<br />
soil particles <strong>and</strong> soot presence in the air of the<br />
town, continental averaged aerosol type is<br />
assumed. Aerosol extinction is calculated using<br />
visibility data for previous day by Koschmieder<br />
equation [1924]. Overestimation partly results<br />
from extinction of UV radiation by clouds because<br />
the sky was not completely cloud free <strong>and</strong>, as<br />
mentioned above, cloudiness is not an input<br />
parameter of the model.<br />
Elevation of the surface above sea level has a<br />
small effect, but it becomes considerable in<br />
mountainous areas. The enhancement of the UV<br />
index according to model results can be about 3 %<br />
to 10 % per kilometer.<br />
3.2 Comparison with Measurements<br />
The accuracy of the model was tested by<br />
comparing model output to clear sky measured<br />
data recorded by the radiometer YANKEE UVB-1<br />
biometer. The biometer measures direct <strong>and</strong><br />
942
10<br />
8<br />
YANKEE UVB-1<br />
NEOPLANTA<br />
in Figure 3, model calculations are only slightly<br />
higher than the measurements. This gives<br />
confidence that this model provides a satisfactory<br />
representation of the UV intensity at the surface.<br />
UVI<br />
UVI<br />
6<br />
4<br />
2<br />
0<br />
10<br />
8<br />
6<br />
4<br />
2<br />
25.06.2003. 26.06.2003.<br />
3. CONCLUSIONS<br />
A numerical model for computing clear sky UV<br />
index at surface was constructed. The model<br />
results were satisfactorily tested on input<br />
parameters change. The performance of the model<br />
has been tested in relation to its predictive<br />
capability of global solar irradiance in the UV<br />
range using data recorded by the radiometer<br />
YANKEE UVB-1 biometer. It was found that<br />
model calculations are slightly higher than the<br />
measurements. The main source of difference is<br />
lack of necessary measurement that can provide<br />
better input atmospheric conditions.<br />
UVI<br />
UVI<br />
0<br />
8<br />
6<br />
4<br />
2<br />
0<br />
6<br />
4<br />
2<br />
0<br />
15.07.2003. 16.07.2003.<br />
27.08.2003. 28.08.2003.<br />
20.09.2003. 21.09.2003.<br />
Figure 3. Comparison model calculations <strong>and</strong><br />
clear sky measurements<br />
UV Index indicates the highest possible UV<br />
irradiance reaching down to the earth’s surface.<br />
Therefore forecasted UV index should be higher<br />
than the measured one, so results from model<br />
calculation are called unsatisfactory if the forecast<br />
value is lower than the observation. As can be seen<br />
REFERENCES<br />
Angstrom, A., Technique of Determining the<br />
Turbidity of the Atmosphere, Tellus, Vol.<br />
13, 214-231, 1961.<br />
Bird, R. E. <strong>and</strong> C. Riordan, Simple solar spectral<br />
model for direct <strong>and</strong> diffuse irradiance on<br />
horizontal <strong>and</strong> titled planes at the Earth's<br />
surface for cloudless atmosphere, J. Appl.<br />
Meteor. 25, 87-97, 1986.<br />
Bogumil, K., J. Orphal <strong>and</strong> J.P. Burrows,<br />
Temperature dependent absorption cross<br />
section of O 3 , NO 2 <strong>and</strong> other atmospheric<br />
trace gases measured whit SCIAMACHY<br />
spectrometer, Proc. ERS- Envistat<br />
Symposium Gothenburg, 2000.<br />
Brine, D. T., <strong>and</strong> M. Iqbal, Solar Spectral Diffuse<br />
Irradiance Under Cloudless Skies, Solar<br />
Energy, Vol. 30, 447-453, 1983.<br />
Burrows J. P., A. Richter, A. Dehn, B. Deters, S.<br />
Himmelmann, S. Voigt, <strong>and</strong> J. Orphal,<br />
Atmospheric Remote–sensing Reference<br />
Data from GOME—2. Temperature–<br />
dependent Absorption Cross Section of O3<br />
in the 231–794 nm Range, J. Quant.<br />
Spectrosc. Radiat. Transfer, 61, 509–517,<br />
1999.<br />
Dichter, B.K., A.F.Beaubien, <strong>and</strong> D.J.Beaubien,<br />
Development <strong>and</strong> characterization of new<br />
solar ultraviolet-B detector, J. Atmos. Sci.<br />
<strong>and</strong> Oc. Technology, 10, 337-344, 1992.<br />
Diffey, B.L., Solar ultraviolet radiation effects on<br />
biological systems, Review in Physics in<br />
Medicine <strong>and</strong> Biology, 36(3): 299-328,<br />
1991.<br />
Gueymard, C., SMARTS2, A simple model of the<br />
atmospheric transfer of sunshine. Florida<br />
943
Solar Energy Center, Rep. FSEC-PF-270-<br />
95, 1995.<br />
Hiltner, W.A., Stars <strong>and</strong> stellar systems,<br />
Compendium of Astronomy <strong>and</strong><br />
Astrophysics, Vol.II, Ch. 8 (R. H. Hardie,<br />
Photo electric reductions), The University<br />
of Chicago Press, 180, 1962.<br />
Hess, M., P. Koepke, <strong>and</strong> I. Schult, Optical<br />
Properties of Aerosols <strong>and</strong> Clouds: The<br />
<strong>Software</strong> Package OPAC. Bulletin of the<br />
American Meteorological Society, 79, No.5,<br />
831-844, 1998.<br />
Justus, C.G. <strong>and</strong> M.V.Paris, A model for solar<br />
spectral irradiance at the bottom <strong>and</strong> top of<br />
a cloudless atmosphere, J. Climat. Appl.<br />
Meteorol., 24, 193-205, 1985.<br />
Kurucz, R. L., I. Furenlid, J. Brault, <strong>and</strong> L.<br />
Testerman, Solar Flux Atlas from 296 to<br />
1300 nm, National Solar Observatory Atlas<br />
No. 1, 1984.<br />
Komhyr, W.D., Operations h<strong>and</strong>book – ozone<br />
observations with a Dobson<br />
spectrophotometer, Prepared for the World<br />
Meteorological Organization Global Ozone<br />
Research <strong>and</strong> Monitoring Project, WMO<br />
report No.6, 1980.<br />
Koschmieder, H., Theorie der horizontalen<br />
Sichtweite, Beitr. Phys. Atmos., 12, 33-53,<br />
1924.<br />
Leckner, B., The Spectral Distribution of Solar<br />
Radiation at the Earth's Surface—Elements<br />
of a Model, Solar Energy, 20, 143-150,<br />
1978.<br />
Madronich, S., L. O. Björn, M. Ilyas <strong>and</strong> M. M.<br />
Caldwell, Changes in biologically active<br />
ultraviolet radiation reaching the Earth's<br />
surface. Journal of Photochemistry <strong>and</strong><br />
Photobiology B: Biology, 46, 5–19, 1998.<br />
McKinley, A.F. <strong>and</strong> B.L. Diffey, A reference<br />
action spectrum for ultraviolet induced<br />
erythema in human skin. CIE Journal, 6,<br />
17-22, 1987.<br />
Nakajima, T., <strong>and</strong> M. Tanaka, Matrix formulations<br />
for the Transfer of Solar Radiation in a<br />
Olane-Parallel Scattering Atmosphere, J.<br />
Quant. Spectrosc. Radiat. Transfer, 35, 13-<br />
21, 1986.<br />
Spencer, J. W., Fourier series representation of the<br />
position of the sun. Search, 2, 272, 1971.<br />
Vanicek, K., T. Frei, Z. Litynska <strong>and</strong> A.<br />
Schmalwieser, UV-Index for the Public, A<br />
guide for publication <strong>and</strong> interpretation of<br />
the solar UV Index forcasts for the public<br />
prepared by the Working Group 4 of the<br />
COST-713 Action “UVB Forecasting,<br />
Brussels, 1999.<br />
944
Mathematical Models for Gene Flow from GM Crops in<br />
the Environment<br />
O. Richter, K. Foit <strong>and</strong> R. Seppelt<br />
Department of <strong>Environmental</strong> Systems Analysis, Institute of Geoecology, Technical University of<br />
Braunschweig, Germany<br />
Abstract: Risk assessment of gene flow from GM crops into the environment requires both the development<br />
of physical transport models <strong>and</strong> biological models for the assessment of outcrossing probabilities. Our<br />
starting point is a Lagrangian approach for pollen dispersal, which describes the concentration statistics in<br />
terms of the stochastic properties of the paths of ensembles of particles. Transport of a particle from a<br />
location (x’,y’,z’) to a location (x,y,z) is mediated by a probability density or transfer function<br />
Q(x,y,z|x’,y’,z’). The transfer function depends on the statistics of the wind field during pollination. The total<br />
amount of pollen, which reaches a single plant, is then derived by the integral over all donators. In the context<br />
of gene flow, particle transport is but one aspect. The target variable is not primarily pollen density but the<br />
amount of outcrossing. The transfer function Q thus has to take into account both transport <strong>and</strong> biological<br />
processes <strong>and</strong> is devised to combine a transport submodel capable of integrating the statistics of wind<br />
velocities, a pollen viability submodel, a phenological submodel, a submodel for pollen redistribution by<br />
insects <strong>and</strong> a pollen competition submodel. Model parameters are estimated from data of outcrossing studies<br />
of maize <strong>and</strong> oil seed rape. The model is then applied to study the effect of field geometries on outcrossing<br />
rates.<br />
Keywords: gene flow modelling, spatial spread of genetic information<br />
1. INTRODUCTION<br />
Since October 2002, the European <strong>Environmental</strong><br />
directive 2001/18/EG is brought into force. It<br />
regulates the permit for commercial <strong>and</strong><br />
experimental release of genetically modified (GM)<br />
plants. An important feature of the directive is the<br />
strict risk assessment with regard to direct <strong>and</strong><br />
indirect environmental effects. Furthermore,<br />
threshold values of tolerated contaminations by<br />
GM material in food <strong>and</strong> food ingredients are<br />
established (Commission Regulations No<br />
49/2000). The unwanted dispersal of genetic<br />
information from GM crops to neighbouring<br />
conventional crops or related weeds is possible <strong>and</strong><br />
negative consequences are not excluded by now.<br />
Fundamental aim of all efforts is to minimize the<br />
risk <strong>and</strong> to enable coexistence between GMO <strong>and</strong><br />
non-GMO agriculture.<br />
Prerequisite of risk assessments <strong>and</strong> management<br />
programs is the estimation of pollen transport.<br />
With the model presented here, simulations of<br />
pollen transport <strong>and</strong> outcrossing rates are possible.<br />
The model takes into account physical transport<br />
processes as well as biological influences.<br />
Biological influences are for instance the degree of<br />
cross-pollination, the influence of insects or the<br />
overlap of fertile periods of donor <strong>and</strong> target<br />
population. After some simulation studies to<br />
demonstrate the model characteristics, the<br />
simulation results of outcrossing experiments with<br />
maize <strong>and</strong> oil seed rape are shown below.<br />
2. THE MODEL<br />
2.1 List of Symbols<br />
x, y, z coordinates of the observation point<br />
x’, y’, z’<br />
φ<br />
coordinates of the source point<br />
wind direction<br />
a x, a y, a z dispersion coefficient in x-, y- <strong>and</strong> z-<br />
direction<br />
q<br />
pollen release rate [kg m-3 s-1]<br />
945
¥<br />
¢<br />
£<br />
¤<br />
s<br />
, ¡ parameter of the wind distribution [°]<br />
st<strong>and</strong>ard deviation of the wind distribution<br />
[°]<br />
w mean wind direction [°]<br />
u<br />
wind speed [m/s]<br />
Q ( x,<br />
y,<br />
z x',<br />
y',<br />
z')<br />
probability density function for transport<br />
from location (x’,y,’z’) to (x,y,z)<br />
G<br />
S(x,y,z)<br />
Source region<br />
source density [kg m-3 s-1]<br />
f ( u,<br />
φ)<br />
Probability density function for wind speed<br />
<strong>and</strong> direction<br />
C wind(x,y)<br />
GM, REC<br />
INSECT<br />
b(x,y)<br />
particle concentration after averaging over<br />
the wind distribution [kg m-3]<br />
indices for transgenic <strong>and</strong> not-transgenic<br />
donor populations<br />
index for influence of insects<br />
insect activity<br />
P(x,y) outcrossing probability [%]<br />
c b(x,y) pollen density of background sources [kg m-<br />
3]<br />
efficiency factor<br />
portion of fertile target plants<br />
efficiency factor for fertilization by<br />
background sources<br />
efficiency factor for outcrossing<br />
2.2 General concept<br />
The Lagrangian theory is used to describe the<br />
dispersal of particles in the atmosphere (generally<br />
in fluids). It is based on the stochastic description<br />
of distances of a group of particles by way of using<br />
a probability density function Q ( x,<br />
y,<br />
z x',<br />
y',<br />
z'<br />
)<br />
for the transport from location (x’,y,’z’) to (x,y,z).<br />
The concentration at location (x,y,z) is given by the<br />
integral over the product of the source S(x’,y’,z’)<br />
<strong>and</strong> the density function.<br />
c( x,<br />
y,<br />
z)<br />
=<br />
Q( x,<br />
y,<br />
z x'<br />
, y'<br />
, z'<br />
) S( x'<br />
, y'<br />
, z'<br />
)<br />
G<br />
dx'<br />
dy'<br />
dz'<br />
(1)<br />
The intention, however, is not to calculate pollen<br />
density, but to calculate the outcrossing rate which<br />
not only depends on a single transport event, but<br />
integrates the entire previous history, i.e. weather<br />
<strong>and</strong> competition during the entire fertile period. In<br />
addition, pollen is distributed by insects. An<br />
outcrossing event depends on<br />
• The pollen concentration at the location<br />
(x,y,z) of donor <strong>and</strong> target population.<br />
• The degree of self-fertilization of the<br />
target plants<br />
• The overlap of the fertile period of the<br />
donor population <strong>and</strong> the sensitive phase<br />
of the target population.<br />
The goal is to find a formulation within the scope<br />
of the Lagrangian approach which provides the<br />
probability of gene transfer at location (x,y,z),<br />
p g<br />
x, y,<br />
z M , B,<br />
G , in dependence on<br />
( )<br />
meteorological variables M, biological variables B<br />
<strong>and</strong> geometrical parameters G.<br />
2.3 The Transfer function<br />
For the probability density function in the<br />
Lagrangian approach, a Gauss dispersal model is<br />
used. It provides a stationary particle transport<br />
from a point source at location (x’,y’,z’) in a<br />
stationary wind field. The resulting concentration<br />
field is given by<br />
( , y,<br />
z x',<br />
y',<br />
z',<br />
u( φ)<br />
) =<br />
u<br />
( φ)<br />
C x<br />
2π<br />
q<br />
a<br />
⋅a<br />
r<br />
y<br />
z<br />
exp <br />
a<br />
x<br />
r − ( x − x')cosφ<br />
−<br />
<br />
( y − y'<br />
) sinφ <br />
with (2)<br />
r<br />
2<br />
=<br />
( ( x − x'<br />
) cos φ + ( y − y'<br />
) sin φ )<br />
a<br />
+ <br />
a<br />
a<br />
+<br />
<br />
a<br />
x<br />
y<br />
x<br />
z<br />
<br />
<br />
<br />
<br />
( − ( x − x'<br />
) sin φ + ( y − y'<br />
) cos φ )<br />
z<br />
2<br />
Figures 1A <strong>and</strong> 1B show projections of the plume<br />
for two different wind directions <strong>and</strong> wind<br />
velocities u.<br />
Figure 1. Projections of plume concentrations<br />
(arbitrary units). Spatial units are m.<br />
A: In main wind direction with u = 4 m/s.<br />
B: In opposite wind direction with u = 0.5 m/s.<br />
2.4 Influence of wind velocity <strong>and</strong> wind<br />
direction<br />
Equation (2) is valid for a point source in a wind<br />
field with velocity u <strong>and</strong> directionφ . However, the<br />
entire previous history of weather during the fertile<br />
period has to be taken into account. This is<br />
summarized in a bivariate probability density<br />
2<br />
2<br />
946
function f ( u, φ) fϕ ( φ)<br />
f ( u,<br />
φ)<br />
= , which is derived<br />
u<br />
from the histograms of wind direction <strong>and</strong> wind<br />
velocity. The single plume equation (2) is weighted<br />
according to wind direction <strong>and</strong> strength via the<br />
integral<br />
( x,<br />
y x',<br />
y'<br />
)<br />
c wind<br />
2π<br />
,<br />
∞ (3)<br />
c<br />
<br />
0<br />
0<br />
=<br />
( x,<br />
y x',<br />
y',<br />
φ) ⋅ f ( u φ) du dφ<br />
For the evaluation of the density function f ( u,φ)<br />
only those readings are taken into account, which<br />
fall into periods when pollen emission is possible<br />
(daytime, favourable moisture conditions).<br />
u,φ<br />
Technical remark: the density functions ( )<br />
were modelled by truncated normal distributions<br />
<strong>and</strong> the density function f (ϕ ) was approximated<br />
by a periodic interpolation polynomial. Figure 3<br />
shows an example.<br />
2.5 Influence of field geometry<br />
The outcrossing probability is determined by the<br />
geometry of the fields of the donor <strong>and</strong> target<br />
populations. The GMO-pollen concentration at<br />
location (x,y) is obtained by integrating over the<br />
GMO plot with the source density S(x’,y’).<br />
c<br />
<br />
G<br />
GM<br />
c<br />
( x,<br />
y)<br />
wind<br />
=<br />
( x,<br />
y x',<br />
y'<br />
) S ( x',<br />
y'<br />
) dx'<br />
dy'<br />
GM<br />
In the case of a constant source density<br />
1<br />
( x',<br />
y'<br />
) ∈G<br />
S ( x',<br />
y'<br />
) = <br />
.<br />
0<br />
else<br />
ϕ<br />
f u<br />
(4)<br />
2.6 Competition between donor <strong>and</strong> target<br />
pollen<br />
The outcrossing probability is determined by the<br />
competition between the pollen of donor <strong>and</strong> target<br />
populations<br />
( , y)<br />
=<br />
cGM<br />
( x,<br />
y)<br />
c ( x,<br />
y) + ηc<br />
( x,<br />
y) + γ c ( x,<br />
y)<br />
P x<br />
κ<br />
GM<br />
REC<br />
b<br />
(5)<br />
while c REC (x,y) is to be computed by an integral<br />
analogously to equation (4). The efficiency factor<br />
κ reflects possible advantages of pollen of the<br />
recipient population over the donor population.<br />
If<br />
c<br />
GM<br />
A<br />
N<br />
( φ<br />
)<br />
f φ<br />
N<br />
0.15<br />
0.1<br />
0.05<br />
A<br />
N<br />
f φ<br />
N<br />
0.2<br />
0.1<br />
( φ<br />
)<br />
-0.4 -0.3 -0.2 -0.1 -0.05<br />
0.1 0.2<br />
-0.1<br />
-0.15<br />
-0.6 -0.4 -0.2<br />
-0.1<br />
-0.2<br />
B<br />
frequ f u<br />
( u,φ<br />
.<br />
)<br />
frequ f u<br />
( u,φ<br />
)<br />
0.3<br />
2.5 3<br />
.<br />
0.25<br />
0.2<br />
0.15<br />
1.5 2<br />
0.1<br />
1<br />
0.05<br />
west<br />
0.5<br />
1 2 3 4 5 6 u east<br />
u<br />
2 4 6 8 u<br />
C<br />
300<br />
[m]<br />
200<br />
100<br />
0<br />
-100<br />
-200<br />
-300<br />
-200 -100 0 100 200 300 400<br />
[m]<br />
Figure 2. Wind field <strong>and</strong> corresponding pollen<br />
dispersal. A: Wind rose with local maxima in west<br />
<strong>and</strong> east direction (polar plot). B: Distributions of<br />
the wind velocities; gentle shape in the west <strong>and</strong><br />
strong wind forces in the easterly wind direction.<br />
C: Resulting dispersal pattern of pollen<br />
concentration, projection.<br />
Figure 3 shows the same kind of plots based on<br />
real wind data. The wind distribution function<br />
(Figure 3B) is based on 24 empirical velocity<br />
distributions (cf. 2.4). The corresponding dispersal<br />
pattern (Figure 3C) is considerably distorted but is<br />
still showing the predominance of south westerly<br />
winds.<br />
B<br />
f u<br />
1<br />
0.75<br />
0.5<br />
0.25<br />
0<br />
0<br />
C<br />
300<br />
[m]<br />
( u,φ<br />
)<br />
200<br />
100<br />
0<br />
-100<br />
-200<br />
2<br />
u<br />
4<br />
2<br />
6 0<br />
4<br />
φ<br />
-300<br />
-200 -100 0 100 200 300 400<br />
[m]<br />
Figure 3. Wind field <strong>and</strong> pollen dispersal based on<br />
a real data set (data courtesy of P. Zwerger <strong>and</strong> A.<br />
Dietz-Pfeilstetter, Biological Research Centre for<br />
Agriculture <strong>and</strong> Forestry, Braunschweig,<br />
Germany).<br />
A: Wind rose (polar plot). B: Distributions of wind<br />
velocities. C: Resulting dispersal pattern of pollen<br />
concentration.<br />
6<br />
3.2 Pollination by insects<br />
A fictive pollen distribution is shown in Figure 4A.<br />
The redistribution by insects (Figure 4B) is carried<br />
out by an increasing insect activity from f(x,y) =0<br />
to f(x,y)=1 in east direction. The original strong<br />
decline of concentrations is smoothed with<br />
increasing insect activity<br />
948
c wind<br />
A<br />
1<br />
0.75<br />
0.5<br />
0.25<br />
0<br />
0 20 40<br />
100<br />
80<br />
60<br />
40<br />
y [m]<br />
the experiment was to establish methods for<br />
quantifying transgenic contamination in the crop<br />
by outcrossing [Schiemann et al., 2002]. In the<br />
year 2000, the field trial was carried out with an<br />
herbicide-resistant line of maize (Zea mays). The<br />
transgenic square with a size of 100- by 100 m was<br />
surrounded by crops of ordinary maize of the same<br />
variety. Altogether 96 sampling points were<br />
monitored. The results of the simulations are<br />
shown in Figure 5.<br />
x [m]<br />
60<br />
80<br />
20<br />
100 0<br />
A<br />
( x<br />
y<br />
)<br />
P ,<br />
0.2<br />
B<br />
0.15<br />
0.1<br />
c INSECT<br />
1<br />
0.75<br />
0.5<br />
0.25<br />
0<br />
0 20 40<br />
x [m]<br />
60<br />
80<br />
20<br />
100 0<br />
100<br />
80<br />
60<br />
40<br />
y [m]<br />
Figure 4. Redistribution of pollen by insects.<br />
A: Original pollen distribution by wind transport.<br />
B: Redistribution with increasing insect activity in<br />
easterly direction.<br />
0.05<br />
150 0<br />
100<br />
-50<br />
50<br />
0<br />
-50<br />
[m]<br />
simulated<br />
observed<br />
B<br />
simulated<br />
0.2<br />
0<br />
50<br />
100<br />
[m]<br />
150<br />
4. COMPARISON WITH EXPERIMENTAL<br />
DATA<br />
In the following examples, the parameter<br />
identification proved to be cumbersome because of<br />
large computer time. The reason is that the<br />
calculating the objective function<br />
n<br />
( ) [ ( ) ( )] 2<br />
a a q = c x , y a a q − P x y<br />
<br />
SQ (9)<br />
x, y,<br />
i i x,<br />
y,<br />
i,<br />
i=<br />
1<br />
necessitates the evaluation of multiple integrals for<br />
each measurement point. For the parameter<br />
identification the FindMinimum procedure in<br />
Mathematica was employed, which is based on the<br />
Levenberg-Marquardt algorithm. The multiple<br />
integrals were evaluated by the adaptive Genz-<br />
Malik algorithm, which is also implemented in<br />
Mathematica.<br />
4.1 Maize<br />
The model was applied to cross-pollination data<br />
from a monitoring farm scale experiment with<br />
maize carried out at the federal Biological<br />
Research Centre for Agriculture <strong>and</strong> Forestry<br />
(BBA) in Braunschweig, Germany. Objective of<br />
i<br />
0.15<br />
0.1<br />
0.05<br />
0.05 0.1 0.15 0.2<br />
observed<br />
Figure 5. Gene flow from transgenic maize.<br />
A: Direct comparsion between simulation <strong>and</strong> data<br />
(percent outcrosssing). B: Correlation between data<br />
<strong>and</strong> model predictions.<br />
4.2 Oil seed rape<br />
Furthermore the model was applied to the data of<br />
an experiment with oil seed rape (Brassica napus)<br />
in the year 1999/2000, carried out by the BBA<br />
Braunschweig as well. Two different herbicideresistant<br />
lines were used (Glufosinat- <strong>and</strong><br />
Glyphosat-resistance) <strong>and</strong> the outcrossing in the<br />
neighbouring transgenic field <strong>and</strong> in the<br />
surrounding non-transgenic seed was examined<br />
[Dietz-Pfeilstetter et al. (2004)]. The data shown<br />
here refer to the transference of Glufosinat-<br />
Resistance (Liberty Link, LL).<br />
949
A<br />
160<br />
180<br />
200<br />
220<br />
0.02<br />
0.015<br />
0.01<br />
0.005<br />
[m]<br />
150<br />
simulated<br />
observed<br />
B<br />
simulated<br />
100<br />
50<br />
0.01<br />
0.005<br />
0<br />
[m]<br />
0<br />
( x<br />
y<br />
)<br />
P ,<br />
0.02<br />
0.015<br />
0.005 0.01 0.015 0.02 observed<br />
Figure 6: Gene flow from transgenic oil seed rape.<br />
A: Direct comparisonn between simulation <strong>and</strong><br />
data (percent outcrossing). B: Correlation between<br />
data <strong>and</strong> model predictions.<br />
5. DISCUSSION<br />
The modelling approach presented here allows to<br />
explicitly incorporate the statistics of wind velocity<br />
<strong>and</strong> wind speed into Lagrangian based transfer<br />
models. Although the general dispersal patterns are<br />
fitted quite well, systematic deviations between<br />
observed <strong>and</strong> simulated values occur. At large<br />
distances from the source the model tends to<br />
overestimate outcrossing rates, whereas<br />
outcrossing rates are underestimated at very short<br />
distances. This discrepancy can be removed, as<br />
was shown by Loos et al. (2003) for a simple<br />
Lagrangian approach by a superposition of two<br />
Gaussian plumes distinguishing between far <strong>and</strong><br />
near transport processes. Further model<br />
developments concern the incorporation of<br />
l<strong>and</strong>scape structures <strong>and</strong> an improvement of the<br />
fitting algorithms.<br />
6. REFERENCES<br />
Loos, C., R. Seppelt, S. Meier-Bethke, J.<br />
Schiemann, O. Richter (2003): Spatially<br />
explicit modelling of transgenic maize pollen<br />
dispersal <strong>and</strong> cross pollination. Journal for<br />
Theoretical Biology. 225(2). 241-255<br />
Richter, O., R. Seppelt (2004): Flow of genetic<br />
information through agricultural ecosystems:<br />
a generic modelling framework with<br />
application to pesticide-resistance weeds <strong>and</strong><br />
genetically modified crops. Ecological<br />
<strong>Modelling</strong> (in Print)<br />
Schiemann, J. <strong>and</strong> S. Meier-Bethke, (2002): Subproject:<br />
"Einkreuzung transgener<br />
Eigenschaften aus Mais in benachbarte nicht<br />
transgene Felder und Etablierung von<br />
Methoden zur Quantifizierung transgener<br />
Kontamination im Erntegut". Of BMBFproject:<br />
"Methodenentwicklung für ein<br />
anbaubegleitendes Monitoring von<br />
gentechnisch veränderten Pflanzen (GVP) im<br />
Agrarökosystem." BBA, Braunschweig<br />
Walklate, P.J., J.C.R. Hunt, H. L. Higson & J. B.<br />
Sweet (2003): A model of pollen mediated<br />
gene flow for oilseed rape. Transactions of<br />
the Royal Society, in press for 2004.<br />
Dietz-Pfeilstetter, A. <strong>and</strong> P. Zwerger, (2004):<br />
Verbreitung von Herbizidresistenzgenen beim<br />
Anbau von gentechnisch verändertem Raps<br />
mit unterschiedlichen Herbizid-resistenzen.<br />
Zeitschrift für Pflanzenkrankheiten und<br />
Pflanzenschutz, Sonderheft XIX, 831-838.<br />
ACKNOWLEDGEMENTS<br />
We gratefully thank the team of J. Schiemann for providinng<br />
the data on cross-pollination of maize <strong>and</strong> the team of A. Dietz-<br />
Pfeilsticker for providing the oild seed rape data, both Federal<br />
Biological Research Centre for Agriculture <strong>and</strong> Forestry,<br />
Braunschweig, Germany.<br />
950
Simulation of Herbicide Transport in an Alluvial Plain<br />
K. Meiwirth 1 <strong>and</strong> A. Mermoud<br />
Institute of <strong>Environmental</strong> Science <strong>and</strong> Technology, Swiss Federal Institute of Technology, ISTE/HYDRAM,<br />
ENAC, EPFL, 1015 Lausanne, Switzerl<strong>and</strong><br />
Abstract: Herbicide transport through the vadose zone was studied in the Upper Rhône River Valley (South-<br />
West Switzerl<strong>and</strong>). The herbicides atrazine <strong>and</strong> isoproturon were applied to instrumented field plots <strong>and</strong> the<br />
concentrations reaching the groundwater were measured. The solute transport is closely linked to precipitation.<br />
Following the first heavy rainfall after the application, the chemicals are quickly transported through the vadose<br />
zone <strong>and</strong> reach the groundwater in a short time. The transport experiments were simulated with the mechanistic<br />
deterministic model HYDRUS-1D. The mobile-immobile water concept was used to account for the rapid<br />
transport. In the study area, the shallow groundwater influences considerably the water conditions in the unsaturated<br />
zone; in such cases the use of a one-dimensional model to simulate the water flow <strong>and</strong> the chemical<br />
transport in the vadose zone is difficult because of problems in defining the lower boundary condition. Groundwater<br />
flow is typically three-dimensional <strong>and</strong> therefore, a saturated - unsaturated 3-D model or the coupling of<br />
an unsaturated 1-D model to a 3-D saturated model would be more appropriate. Nevertheless, HYDRUS-1D<br />
allowed to describe qualitatively some observations <strong>and</strong> to confirm the assumption that accelerated flow occurs<br />
on the experimental plots.<br />
Keywords: herbicide transport; groundwater contamination; water <strong>and</strong> transport modeling<br />
1. meiwirth@gmx.de<br />
1. INTRODUCTION<br />
The contamination of groundwater (GW) by pesticides<br />
has become an increasing problem throughout<br />
the last decades. Pesticides that are ineffectively retained<br />
or rapidly transported through the unsaturated<br />
zone may reach the GW. The transport of<br />
pesticides in the unsaturated zone depends on the<br />
physico-chemical characteristics of the substance,<br />
on the intensity <strong>and</strong> frequency of its application, on<br />
the soil properties, <strong>and</strong> on regional characteristics<br />
such as climate <strong>and</strong> hydrogeology. Heavy rainfall<br />
may transport chemicals deep into the vadose zone,<br />
especially in highly porous or fractured soils [Flury<br />
et al., 1998]. Shallow GW tables are especially vulnerable<br />
for such pesticide contamination [Flury,<br />
1996].<br />
Transport experiments may be carried out using different<br />
techniques at several scales. In order to analyse<br />
the experiments, numerical models are often<br />
used. Within the last decades several models have<br />
been developed to simulate the water flow <strong>and</strong> solute<br />
transport in agricultural environments [Wauchope<br />
et al., 2003], some of them account for<br />
physical non-equilibrium processes [Simunek et al.,<br />
2003].<br />
Mechanistic models that account for physical nonequilibrium<br />
have recently been divided into two<br />
groups: dual-porosity <strong>and</strong> dual-permeability models<br />
[Simunek et al., 2003]. Both groups divide the soil<br />
into two separate pore domains. While dual-porosity<br />
models assume that water in the matrix domain is<br />
stagnant, dual-permeability models allow for water<br />
flow in both, the macropores <strong>and</strong> the micro (matrix)<br />
pores [Simunek et al., 2003]. Dual-permeability<br />
models are frequently used to describe flow <strong>and</strong><br />
transport in fractured or structured media displaying<br />
shrinkage cracks, earthworm channels, root cracks,<br />
or heterogeneous soil textures [e.g., Larsson <strong>and</strong><br />
Jarvis, 1999]. In dual-porosity models the water<br />
flow is restricted to one flow domain (inter-aggregate<br />
pores), while the matrix domain (intra-aggregate<br />
pores) retains <strong>and</strong> stores water, but does not<br />
permit convective flow. An exchange between the<br />
pore regions is described as a first-order process.<br />
This mobile-immobile water concept [MIM; Van<br />
951
Genuchten <strong>and</strong> Wierenga, 1976] is often used to describe<br />
solute transport processes in aggregated porous<br />
media [V<strong>and</strong>erborght et al., 1997].<br />
This paper describes the simulation of transport experiments<br />
carried out in the Rhone River Valley between<br />
Martigny <strong>and</strong> Charrat (Switzerl<strong>and</strong>). Two<br />
herbicides <strong>and</strong> a tracer were applied to instrumented<br />
field plots <strong>and</strong> the concentrations reaching the<br />
groundwater were measured. The model HYDRUS-<br />
1D was chosen for the simulations <strong>and</strong> a dual-porosity<br />
approach was used to simulate the observed<br />
transport. The simulations aimed at explaining the<br />
dominant processes involved in pesticide transport<br />
towards the shallow GW table.<br />
2. MATERIALS AND METHODS<br />
Experimental plots (2,50 m x 1,60 m) were instrumented<br />
in April 2001 with TDR probes <strong>and</strong> tensiometers<br />
between the soil surface <strong>and</strong> 1 m depth.<br />
Additionally, 2.5 m deep stainless steel piezometers<br />
<strong>and</strong> a rain gauge were installed. Two herbicides <strong>and</strong><br />
Iodide (tracer) were applied to the bare soil surface<br />
on May 24, 2002. During the following months,<br />
groundwater samples were collected using a peristaltic<br />
pump. The samples were analysed for their<br />
solute concentration using HPLC.<br />
The soil at the experimental site is a rather homogeneous<br />
silt loam with low organic carbon <strong>and</strong> clay<br />
contents. Because the solutes reached the shallow<br />
GW table (1.4-2 m depth) surprisingly quick, a<br />
dual-porosity model was used to simulate the accelerated<br />
transport. Moreover, evaporation in the valley<br />
is strong <strong>and</strong> varies considerably within a day;<br />
therefore, an hourly time step had to be considered<br />
for the simulation of the strong capillary rise <strong>and</strong><br />
quick flux changes derived from the experiments.<br />
Water flow <strong>and</strong> solute transport were simulated<br />
with HYDRUS-1D, a mechanistic deterministic<br />
model. It uses Richard's equation for water flow <strong>and</strong><br />
the Convection-Dispersion Equation for solute<br />
transport. The model allows for dual-porosity calculations<br />
using the MIM-concept. Boundary conditions<br />
like evaporation <strong>and</strong> rainfall can be introduced<br />
at an hourly time step.<br />
3. OBSERVATIONS<br />
Figure 1 shows the herbicide concentration in the<br />
GW during the summer 2002 <strong>and</strong> the water table<br />
depth. Hardly any chemicals were found in the GW<br />
during the first two weeks without rainfall after the<br />
application (Fig. 1). After the first heavy precipitation<br />
on June 5, a sudden peak was observed. The<br />
concentrations decreased during the following dry<br />
period <strong>and</strong> a second concentration peak appeared as<br />
a consequence of rainfall at the end of June (49<br />
mm). A third attenuated concentration peak was observed<br />
in mid July. The concentration development<br />
of the two herbicides is very similar. Moreover, the<br />
tracer iodide was transported as quickly as the herbicides<br />
(Fig. 2) indicating that adsorption does not<br />
play a predominant role in this soil.<br />
Concentration [µg l -1 ]<br />
GW Depth [m]<br />
Concentration [mg l -1 ]<br />
160<br />
140<br />
120<br />
100<br />
30<br />
80<br />
40<br />
Atrazine<br />
60<br />
Isoproturon 50<br />
40<br />
60<br />
20<br />
70<br />
0<br />
80<br />
21.05.02 31.05.02 10.06.02 20.06.02 30.06.02 10.07.02 20.07.02 30.07.02<br />
1.50<br />
1.75<br />
2.00<br />
Figure 1: Herbicide concentration in GW, daily<br />
rainfall, <strong>and</strong> depth of the GW table<br />
3<br />
2<br />
1<br />
0<br />
24-Mai-02 13-Jun-02 03-Jul-02 23-Jul-02<br />
Figure 2: Iodide concentration in the GW <strong>and</strong> daily<br />
rainfall<br />
4. NUMERICAL MODELING<br />
The experiment has been simulated with the mechanistic<br />
deterministic model HYDRUS-1D [Simunek<br />
et al., 1998]. The aim of the simulations was to<br />
define the processes involved in pesticide transport<br />
<strong>and</strong>, if possible, to predict the fate of chemicals applied<br />
at the soil surface.<br />
4.1 Water Flow Simulations<br />
The model domain is one-dimensional, extending<br />
from the soil surface to a depth of 2.5 m. At the beginning<br />
of a simulation, the lower part of the profile<br />
is water saturated <strong>and</strong> the pressure head at the bottom<br />
node is specified as the height of the water col-<br />
0<br />
10<br />
20<br />
30<br />
40<br />
50<br />
60<br />
70<br />
0<br />
10<br />
20<br />
Daily Rainfall [mm]<br />
Daily Rainfall [mm]<br />
952
umn. From the water table to the soil surface the<br />
profile is supposed to be at hydraulic equilibrium.<br />
At the upper boundary, atmospheric conditions with<br />
possible surface ponding were imposed. Hourly potential<br />
evapotranspiration rates (Penman-Monteith)<br />
were calculated based on data from nearby weather<br />
stations <strong>and</strong> specified together with hourly measured<br />
rainfall as a time variable boundary condition.<br />
At the lower boundary, two basically different conditions<br />
were considered: i) a variable pressure head<br />
boundary condition (approach 1), where the GW<br />
level was specified as pressure head on the bottom<br />
node at an hourly time step; thus, water is allowed<br />
to enter <strong>and</strong> leave the profile through the lower<br />
boundary, ii) a zero flux boundary condition (approach<br />
2), where no water can enter or leave the profile<br />
at the lower boundary; the GW level is<br />
calculated by the model.<br />
The soil profile was assumed to consist of one uniform<br />
soil layer; the parameters of the soil water retention<br />
function [van Genuchten, 1980] were<br />
estimated from the field measurements using the<br />
code RETC [van Genuchten et al., 1994]. The parameters<br />
of the hydraulic conductivity function (Ks,<br />
l) were determined by inverse modeling on pressure<br />
heads <strong>and</strong>/or on GW levels. In order to implement<br />
the MIM concept, a value of the immobile water<br />
fraction had to be defined. Because the model considers<br />
that immobile water cannot evaporate, the<br />
lowest measured value of the water content (0.28<br />
cm 3 cm -3 ) was taken as the immobile water content.<br />
The simulation started at the end of March 2002,<br />
roughly two month before the chemical application<br />
<strong>and</strong> ended in late August 2002. Figure 3 shows the<br />
measured <strong>and</strong> simulated pressure heads at 10 <strong>and</strong> 90<br />
cm depth in the unsaturated zone for approach 1<br />
(A1) <strong>and</strong> approach 2 (A2). The observed pressure<br />
heads at both depths are well reproduced when using<br />
approach 1. Approach 2 matches the data less<br />
well, especially at 10 cm where the observed pressure<br />
heads are significantly underestimated. The<br />
GW level changes, however, are relatively well predicted<br />
in approach 2 (Fig. 4).<br />
The cumulative evaporation was calculated for both<br />
approaches <strong>and</strong> additionally, for approach 1 the cumulative<br />
water flow at the lower boundary was considered<br />
(Fig. 5). In approach 1, the simulated actual<br />
evaporation (Act. E, A1; Fig. 5) is close to the potential<br />
evapotranspiration (Pot. ET). At the bottom<br />
water continuously enters the profile (Bot. In, A1).<br />
When using approach 2, the actual evaporation<br />
(Act. E, A2) accounts for only 50 % of the potential<br />
evapotranspiration.<br />
Analysis of the instantaneous fluxes shows only little<br />
drainage of rainwater to the GW with approach<br />
1; the observed GW level rise is not a consequence<br />
of percolating water, but of the inflow of water<br />
through the lower boundary. This is not consistent<br />
with the observations that show rapid herbicide<br />
transport towards the GW during rainfall events.<br />
Therefore, approach 2 was used for the transport<br />
simulations.<br />
Pressure Head [mm]<br />
Pressure Head [mm]<br />
-5000<br />
-4000<br />
-3000<br />
-2000<br />
-1000<br />
-5000<br />
-4000<br />
-3000<br />
-2000<br />
-1000<br />
10 cm<br />
90 cm<br />
A1<br />
0<br />
25.03.02 24.04.02 24.05.02 23.06.02 23.07.02<br />
10 cm<br />
90 cm<br />
A2<br />
0<br />
25.03.02 24.04.02 24.05.02 23.06.02 23.07.02<br />
Figure 3: Measured (points) <strong>and</strong> simulated (lines)<br />
pressure heads at 10 <strong>and</strong> 90 cm depth for<br />
approaches A1 <strong>and</strong> A2<br />
GW table depth [m]<br />
25.03.02 24.04.02 24.05.02 23.06.02 23.07.02 22.08.02<br />
1<br />
1.5<br />
2<br />
2.5<br />
Figure 4: Measured (points) <strong>and</strong> simulated (line)<br />
GW table depth in approach 2<br />
953
Cum. Flux [mm]<br />
600<br />
400<br />
200<br />
Pot. ET<br />
Act. E, A1<br />
Act. E, A2<br />
Bot. In, A1<br />
Observed Concentration [mg mm -3 ]<br />
1.6E-07<br />
8.0E-08<br />
2.E-08<br />
1.E-08<br />
Simulated Concentration [mg mm -3 ]<br />
0<br />
27.03.02 26.04.02 26.05.02 25.06.02 25.07.02<br />
Figure 5: Cumulative boundary fluxes : Pot.<br />
ET=potential evapotranspiration; Act. E,<br />
A1=actual evaporation, approach 1; Act. E,<br />
A2=actual evaporation, approach 2; Bot. In,<br />
A1=inflow at lower boundary, approach 1.<br />
4.2 Transport Simulations<br />
The transport parameters were either measured directly<br />
in the laboratory (distribution coefficient,<br />
degradation constant) or assessed by roughly adjusting<br />
the simulations to the available concentration<br />
data (dispersivity, mass transfer coefficient,<br />
fraction of adsorption sites in contact with the mobile<br />
liquid). Initially, the soil profile was free of<br />
chemicals. At the upper boundary, a concentration<br />
flux condition was imposed <strong>and</strong> the herbicides were<br />
introduced with the first rainfall after the application.<br />
At the lower boundary, a zero gradient condition<br />
was chosen. The GW concentrations were<br />
calculated as an average over the water column.<br />
Figure 6 shows observed <strong>and</strong> simulated atrazine<br />
concentrations in the GW. A first small peak is predicted<br />
in early June, but its relative intensity is significantly<br />
underestimated compared to the observed<br />
value. The shape of the second <strong>and</strong> third concentration<br />
peaks are rather well reproduced, although the<br />
third simulated peak occurs slightly earlier than observed.<br />
A sensitivity analysis shows that both, the<br />
immobile water content <strong>and</strong> the dispersivity, have a<br />
considerable influence on the herbicide transport<br />
<strong>and</strong> consequently, on the concentrations in the GW.<br />
For the simulation presented in Fig. 6 a dispersivity<br />
value as high as 1000 mm was assumed. This value<br />
is clearly exaggerated <strong>and</strong> must be considered as a<br />
lumped parameter accounting for the quick transfer<br />
of the solutes through the vadose zone. The immobile<br />
water content was set to 0.28 cm 3 cm -3 ,the<br />
maximum possible value, as it represents the lowest<br />
measured water content. In spite of this, the model<br />
underestimates the concentrations in the GW <strong>and</strong><br />
simulates a very small first concentration peak.<br />
0.0E+00<br />
0.E+00<br />
24.05.02 23.06.02 23.07.02 22.08.02<br />
Figure 6: Observed (points) <strong>and</strong> simulated (line)<br />
atrazine concentration in the GW<br />
5. DISCUSSION<br />
The experimental results have shown that the herbicide<br />
transport is closely linked to precipitation. The<br />
applied solutes are rapidly transported towards the<br />
GW after the first heavy rainfall subsequent to the<br />
application. The rapidity of the transport is surprising<br />
bearing in mind that the soil is not visibly structured.<br />
A dual-porosity model was therefore chosen<br />
to simulate the observed GW concentrations.<br />
For the hydraulic simulations, two different lower<br />
boundary conditions were considered.<br />
When assuming a variable pressure head lower<br />
boundary condition (A1), the strong evaporation is<br />
well reproduced. The GW level rise subsequent to<br />
precipitation, however, is caused entirely by an inflow<br />
of water through the lower boundary <strong>and</strong> not<br />
by rainwater percolation; this is not consistent with<br />
the rapid transport of the herbicides observed after<br />
rainfall events.<br />
On the other h<strong>and</strong>, when no inflow of water is assumed<br />
at the bottom boundary (A 2), the evapotranspiration<br />
is significantly underestimated.<br />
Therefore, it is difficult to define precisely the lower<br />
boundary condition. Either the model is too unrestricted<br />
resulting in unreasonable water fluxes at the<br />
bottom node or the soil column is considered isolated<br />
<strong>and</strong> excluded from the regional water flow.<br />
The transport of the chemicals as observed in the<br />
field experiment was extremely quick. In order to<br />
account for this rapidity, a high dispersivity value<br />
together with a great immobile water content were<br />
assumed. Still, the simulations were not satisfactory<br />
<strong>and</strong> the concentration peak observed in the GW<br />
were only roughly reproduced.<br />
954
6. CONCLUSIONS<br />
In conclusion, the model HYDRUS-1D is badly<br />
adapted for predicting herbicide transport in the<br />
specific context of the study area because of two<br />
principal problems :<br />
The shallow GW <strong>and</strong> the high evaporation rate influence<br />
considerably the water conditions in the unsaturated<br />
zone; therefore, a correct simulation of the<br />
water flow depends to a great extend on a realistic<br />
definition of the lower boundary condition. When<br />
using a one-dimensional model, however, the<br />
boundary conditions cannot be correctly defined.<br />
The problem could probably be solved by using a<br />
three-dimensional saturated-unsaturated model or<br />
an unsaturated 1-D model coupled to a 3-D saturated<br />
model.<br />
Even with the MIM concept, it is difficult to reproduce<br />
the extremely rapid transport observed after<br />
rainfall events. Dual-permeability models might be<br />
more appropriate to reproduce the rapid transport.<br />
This conclusion is somewhat surprising as the soil<br />
appears to be homogeneous without cracks or earthworm<br />
burrows. The simulation results indicate that<br />
even in a rather homogeneous soil significantly accelerated<br />
flow may occur.<br />
7. ACKNOWLEDGEMENTS<br />
This research was undertaken within the scope of<br />
the European project PEGASE (Pesticides in European<br />
Groundwaters: detailed study of representative<br />
aquifers <strong>and</strong> simulation of possible evolution scenarios;<br />
EU contract number: EVK1-CT1999-<br />
00028). The funding of the research by the Swiss<br />
Government under the 5th Framework Programme<br />
is gratefully acknowledged.<br />
Simunek, J.; Sejna, M. <strong>and</strong> M.Th. Van Genuchten,<br />
The HYDRUS-1D software package for simulating<br />
the one-dimensional movement of water,<br />
heat, <strong>and</strong> multiple solutes in variablysaturated<br />
media, version 2.0, US Salinity Laboratory,<br />
USDA/ARS, Riverside, CA, 1998.<br />
Simunek, J.; Jarvis, N.; van Genuchten, M. Th. <strong>and</strong><br />
A. Gärdenäs, Review <strong>and</strong> comparison of models<br />
for describing non-equilibrium <strong>and</strong> preferential<br />
flow <strong>and</strong> transport in the vadose zone,<br />
Journal of Hydrology, 272, 14-35, 2003.<br />
V<strong>and</strong>erborght, J.; Mallants, D.; Vanclooster, M. <strong>and</strong><br />
J. Feyen, Parameter uncertainty in the mobileimmobile<br />
solute transport model, Journal of<br />
Hydrology, 190(1), 75-101, 1997.<br />
Van Genuchten, M. T. <strong>and</strong> P. J. Wierenga, Mass<br />
transfer studies in sorbing porous media. I.<br />
Analytical solutions, Soil Science Society of<br />
America Journal, 40, 473-480, 1976.<br />
Van Genuchten, M.Th., A closed-form equation for<br />
predicting the hydraulic conductivity of unsaturated<br />
soils, Soil Science Society of America<br />
Journal, 44, 892-898, 1980.<br />
Van Genuchten, M.Th; Leij, F.J.; Yates, S.R. <strong>and</strong><br />
W. B. Williams, RETC, Code for quantifying<br />
the hydraulic functions of unsaturated soils.<br />
US Salinity Laboratory, USDA/ARS, Riverside,<br />
CA, 1994.<br />
Wauchope, R.D.; Lajpat, R.A.; Jeffrey, G. A.; Bingner,<br />
R.; Lowrance, R.; Van Genuchten, M.T.<br />
<strong>and</strong> L.D. Adams, <strong>Software</strong> for pest-management<br />
science: Computer models <strong>and</strong> databases<br />
from the United States Department of Agriculture,<br />
Agricultural Reserach Service, Pest Management<br />
Science, 59, 691-698, 2003.<br />
8. REFERENCES<br />
Flury, M., Experimental evidence of transport of<br />
pesticides through field soils - a review, Journal<br />
of <strong>Environmental</strong> Quality, 25 (1), 25-45,<br />
1996.<br />
Flury, M; Jury, W.A. <strong>and</strong> E.J. Kladivko, Field-scale<br />
solute transport in the vadose zone: experimental<br />
observations <strong>and</strong> interpretation, in:<br />
Magdi Selim, H. (ed.), Physical nonequilibrium<br />
in soils: modeling <strong>and</strong> application, Ann Arbor<br />
Press, 349-369, 1998.<br />
Larsson, M.H. <strong>and</strong> N. J. Jarvis, Evaluation of a dualporosity<br />
model to predict field-scale solute<br />
transport in a macroporous soil, Journal of<br />
Hydrology, 215, 153-171, 1999.<br />
955
¢<br />
¡<br />
The influence of the averaging period on calculation of<br />
air pollution using a puff model<br />
Rajković Borivoj , Gršić Zoran ¡ , Popov Zlatica ¢ , Djurdjević Vladimir<br />
Department for Meteorology, College of Physics, Belgrade University,<br />
Serbia <strong>and</strong> Montenegro<br />
Radiation <strong>and</strong> <strong>Environmental</strong> Protection Laboratory, Institute of Nuclear Sciences Vinca P.O. Box.522,<br />
11001 Belgrade, Serbia <strong>and</strong> Montenegro<br />
Center for <strong>Environmental</strong> Modeling, University of Novi Sad, Republican Meteorological Institute of Serbia,<br />
Meteorological Observatory, Novi Sad, Serbia <strong>and</strong> Montenegro<br />
Abstract: The main goal of this paper is to assess differences in calculations of air pollution using st<strong>and</strong>ard,<br />
one hour wind averages <strong>and</strong> shorter time averages of ten minutes. A puff model has been used to estimate<br />
concentrations of a passive substance for four days in January, March, June <strong>and</strong> September as representatives<br />
of variations of wind <strong>and</strong> stability during a year. Meteorological inputs were ten meters winds <strong>and</strong> two meters<br />
temperature, measured at a weather station. The additional data of temperature gradients, measured at the same<br />
station two times a day, were also available.<br />
The st<strong>and</strong>ard practice is to form hourly averages of wind speed <strong>and</strong> hourly prevailing direction based on the<br />
data tape records. To infer the importance of the shorter averaging period, tapes were re-analyzed forming ten<br />
minutes averages. After extrapolation to fifty meters heights we forced a puff model with these, two differently<br />
averaged, wind data.<br />
Keywords: Wind extrapolation; Monin-Obukhov theory ; Stability of the PBL; Calculation of pollution dispersion<br />
; Puff model<br />
1 INTRODUCTION<br />
If we want to estimate possible influence of a future<br />
pollution source, we should perform calculations<br />
of concentrations of a passive pollutant for an<br />
extended period of time such as one year or even up<br />
too five years. On a given location we might be in<br />
a situation where only the st<strong>and</strong>ard measurements<br />
of wind <strong>and</strong> temperature are available. That means<br />
that we have data of wind at ten meters <strong>and</strong> temperature<br />
at two meters with time resolution of one<br />
hour. The wind direction is the so-called prevailing<br />
wind direction, which is defined as the most frequent<br />
wind direction in an hour. Since the source<br />
of pollution is usually at greater heights than those<br />
of ten meters we have to perform vertical extrapolation<br />
of the wind speed. The st<strong>and</strong>ard procedure<br />
would be to use Monin-Obukhov theory (MO in the<br />
further text)y. Unfortunately in that case we need<br />
temperature gradients as well. If we have temperature<br />
gradients MO theory enables us to calculate the<br />
sensible heat flux <strong>and</strong> the friction velocity, which<br />
finally leads the extrapolation of the wind speed.<br />
Holstag <strong>and</strong> Van Ulden [1] <strong>and</strong> Holstag [2] have<br />
proposed an alternative procedure for calculation of<br />
the sensitive heat flux using only st<strong>and</strong>ard wind,<br />
temperature measurements <strong>and</strong> cloud cover. Once<br />
we have sensible heat flux we can, using the MO<br />
theory do the extrapolation. Once we do the wind<br />
extrapolation we can then use some simple model<br />
to estimate possible influence of a pollution source.<br />
That can be done for instance with a Gaussian<br />
plume model as less computer dem<strong>and</strong>ing method<br />
or a puff model, of the Gaussian type, but with considerable<br />
more dem<strong>and</strong> for computer time.<br />
956
2 THE WIND EXTRAPOLATION<br />
Following Holstag <strong>and</strong> Van Ulden [1] <strong>and</strong> Holstag<br />
[2] we can estimate sensible heat flux using only<br />
routine measurements, wind at ten meters, temperature<br />
at two meters <strong>and</strong> cloud cover. Basically the<br />
method relies upon energy balance of the ground<br />
surface. To reassess the quality of this approach we<br />
have compared this extrapolation method with the<br />
st<strong>and</strong>ard MO theory for one site where concurrently<br />
with the routine measurements, measurements of<br />
temperature gradients were done. The station is<br />
Rimski Šancevi near Novi Sad lat( N) <strong>and</strong><br />
£¥¤§¦©¨¥<br />
¦¤<br />
E). We have also addressed the question<br />
lon(<br />
of time resolution of the wind data by doing the re<br />
analysis of the anemometer tapes <strong>and</strong> thus forming<br />
ten minutes winds with corresponding changes in<br />
directions.<br />
In course of a year we meet high range of stability,<br />
from very unstable stratification during summer<br />
days to very stable stratification during winter<br />
nights. In the case of very stable stratification<br />
straightforward use of MO theory will give excessively<br />
high values of wind. In that case Holsatg [2]<br />
had proposed an ad hock procedure, which seems to<br />
perform well in such extreme conditions. In order<br />
to reexamine that method we did extrapolation using<br />
the same wind data but now without the use of<br />
the temperature gradients. Comparison of the two<br />
methods is presented in one variable, the diurnal<br />
variation of wind averaged over a year, figure 1. The<br />
Wind speed (m/s)<br />
5<br />
4.5<br />
4<br />
3.5<br />
3<br />
2.5<br />
Wind at 10 met.<br />
Wind at 50 met M-O<br />
Wind at 50 met. heta fl.<br />
2<br />
0 5 10 15 20<br />
locat time (h)<br />
Figure 1: Diurnal variation of wind averaged over a<br />
year<br />
lowest curve (black) shows the measured data. The<br />
next curve is the extrapolated wind using MO theory<br />
with the proposed modification for high stability<br />
when needed(red) <strong>and</strong> the third curve (green) is<br />
the extrapolated wind from st<strong>and</strong>ard measurements<br />
only. Inspection of that figure shows that there is<br />
quite good agreement between the two methods except<br />
between sixteen hours <strong>and</strong> nineteen hours when<br />
Holstag <strong>and</strong> Van Ulden method produces slightly<br />
higher values for the wind speed. We should note<br />
that the proposed method has several parameters<br />
which vary for different locations. They are related<br />
to the state of the ground <strong>and</strong> to its radiation properties<br />
as well. If one has more accurate local values<br />
concerning these processes that will improve the<br />
quality of the results.<br />
3 CALCULATION OF POLLUTION DISPERSION<br />
For the purpose of calculation of the atmospheric<br />
dispersion of airborne material we can use the st<strong>and</strong>ard<br />
Gaussian plume model. It is a simple concept<br />
<strong>and</strong> is extremely computationally efficient. Its<br />
shortcomings are pronounced if we have large temporal<br />
variability of wind <strong>and</strong>/or if we want to estimate<br />
concentrations for larger areas, say beyond<br />
ten kilometers. In that case it is better to use a<br />
model from the puff category wherein one has series<br />
of consecutively released puffs. Details of a such<br />
model design are given in [3] <strong>and</strong> [4]. Local dispersion,<br />
of a individual puff, is still Gaussian like.<br />
This means that we still have dispersion parameters<br />
in horizontal <strong>and</strong> vertical whose values are parameterized<br />
using the Pasquill-Gilfford scheme with the<br />
use of the vertical temperature gradients.<br />
When we want to give an estimate of the possible influence<br />
of a source of pollution, we should perform<br />
calculations covering a longer period say one year<br />
or, if possible, up to five years. Here, at the beginning<br />
of our work, we did just a few, pilot runs, covering<br />
all four seasons <strong>and</strong> with runs three hours long<br />
which were performed twice a day, at midnight <strong>and</strong><br />
at noon. Puff releases were done every ten minutes<br />
in both cases. In the case of hourly averaged winds<br />
the releases were done but with the same wind, inside<br />
each hour. To quantify, in some extent, the results<br />
we have presented, in table 1, values of the<br />
maximum concentrations for each run. We see that<br />
ratios of maximums, for two types of wind averaging,<br />
are quite different from one month to another.<br />
Values of these ratios are 22.6, 4.6, 7.1 <strong>and</strong> 1.1 for<br />
night cases for January, March, June <strong>and</strong> September<br />
respectively. For the noon cases these ratios are<br />
much smaller i.e 2.1, 1.3, 7.2 <strong>and</strong> 1.1. The biggest<br />
957
Month Hour Hourly averages ten min averages<br />
01 0 5.2E-03 2.3E-04<br />
01 12 4.4E-05 2.1E-05<br />
03 0 2.5E-05 5.4E-06<br />
03 12 3.9E-06 2.9E-06<br />
06 0 2.7E-07 3.8E-08<br />
09 12 1.8E-07 2.5E-08<br />
09 0 1.9E-06 1.7E-06<br />
09 12 1.4E-06 1.3E-06<br />
Table 1: Values of maximum concentrations for<br />
hourly averaged winds <strong>and</strong> for ten minutes averaged<br />
winds in the right panels<br />
value is for January at midnight, while the smallest<br />
value is for September at noon. These differences,<br />
presumably come from the differences in the<br />
respective stability regimes <strong>and</strong> wind strengths.<br />
Figures 2 3, 4 <strong>and</strong> 5 are graphical representations<br />
of the concentrations of three hour averages for the<br />
the continuous release with constant rate of the release.<br />
The left panels represent results using hourly<br />
averaged winds while on the right panels we have<br />
results form ten minutes averaged winds. General<br />
characteristic, as seen from these panels is that concentrations<br />
are smaller for ten minutes winds (i.e.<br />
respective areas are wider). We also see quite strong<br />
signal of seasonal dependence as well as diurnal<br />
variations though in smaller magnitude. The differences,<br />
as in the case of corresponding maximums,<br />
presumably come from two reasons. Differences<br />
in the wind strength <strong>and</strong> in stability at that moment.<br />
Comparison between two panels, left <strong>and</strong><br />
right is comparison of differences in the averaging<br />
method only. But that difference has twofold consequence.<br />
First temporal variation in ten minutes wind<br />
might ”stretch” the passive substance <strong>and</strong> secondly<br />
through parameterization of dispersion coefficients<br />
(Pasquill-Gilfford scheme).<br />
Figure 2: Concentration of pollution after three<br />
hours of continuous release. Upper two panels are<br />
for the 15nth of January. Start of the release at<br />
midnight <strong>and</strong> at noon while lower two are for the<br />
15nth of March, with the same starting of the release.<br />
Winds are hourly averages. Please note that<br />
the scales are different for different panels<br />
958
Figure 3: Concentration of pollution after three<br />
hours of continuous release. Upper two panels are<br />
for the 15nth of January. Start of the release at<br />
midnight <strong>and</strong> at noon while lower two are for the<br />
15nth of March, with the same starting of the release.<br />
Winds are ten minutes averages. Please note<br />
that the scales are different for different panels<br />
Figure 4: Concentration of pollution after three<br />
hours of continuous release. Upper two panels are<br />
for the 15nth of June. Start of the release at midnight<br />
<strong>and</strong> at non while lower two are for the 15nth<br />
of September, with the same starting of the release.<br />
Winds are hourly averages<br />
959
4 CONCLUSIONS<br />
The differences in calculations of dispersion of a<br />
wind borne material having ten minutes averages<br />
<strong>and</strong> hourly averages are quite evident. They come<br />
basically from two effects. Firstly there are differences<br />
in the wind intensity <strong>and</strong> in wind direction.<br />
The second difference comes from different states<br />
of the atmosphere for different seasons (stability).<br />
These differences are present also going for midnight<br />
to noon. Ratios of the maximums in these run<br />
are quite different <strong>and</strong> are quite large for stable cases<br />
<strong>and</strong> weaker winds. Having in mind the underlying<br />
physics of a puff model we may say that ten minutes<br />
winds are preferable to the longer period averaged<br />
winds in longer term calculations of concentrations<br />
an an airborne material.<br />
5 INQUIRIES AND CORRESPONDENCE<br />
All inquires concerning papers, at any stage of<br />
the process of preparation, review <strong>and</strong> publication<br />
should be addressed to:<br />
Dr Borivoj Rajković, Department for Meteorology,<br />
College of Physics, Belgrade University Dobračina<br />
16, 11000 Belgrade, Serbia <strong>and</strong> Montenegro<br />
Phone: +381 11 625 981<br />
Fax: +381 11 3282 619<br />
Email: bora@ff.bg.ac.yu<br />
ACKNOWLEDGMENTS<br />
This research was partially sponsored by the Republic<br />
of Serbia, Ministry of science, technologies <strong>and</strong><br />
development under grant no. 1197 <strong>and</strong> by Republican<br />
Meteorological Institute of Serbia.<br />
REFERENCES<br />
[1] Holstag A. A. M., , <strong>and</strong> P. Van Ulden. A simple<br />
scheme for datime estimates of the surface<br />
fluxes from routine wether data. J. Appl. Meteor.,<br />
22:517–529, 1983.<br />
Figure 5: Same as in figure 4 except for the winds,<br />
which are ten minutes averages<br />
[2] Holstag A. A. M. Estimates of diabatic<br />
wind speed profiles from neasr-surface wether<br />
observations. Boundary-Layer Meteorology.,<br />
29:225–250, 1984.<br />
[3] Z. Gršić. A computer program for the pollution<br />
distribution assessment, 1996. Third Interna-<br />
960
tional Symposium <strong>and</strong> exhibition on <strong>Environmental</strong><br />
Contamination in Central <strong>and</strong> Eastern<br />
Europe,Warsaw.<br />
[4] Z. Gršić <strong>and</strong> P. Milutilović. Automated meteorological<br />
station <strong>and</strong> appropriate software for<br />
air pollution distribution assessment, 2000. Air<br />
Pollution <strong>Modelling</strong> <strong>and</strong> Its Application XIII,<br />
edited by S. -E. Gryning <strong>and</strong> E.Batchvarova.<br />
961
Interaction Between Hydrodynamics <strong>and</strong> Mass-<br />
Transfer at the Sediment-Water Interface<br />
Carlo Gualtieri<br />
Hydraulic & <strong>Environmental</strong> Engineering Department Girolamo Ippolito, University of Napoli Federico II<br />
Via Claudio, 21 - 80125, Napoli (Italy), Phone: +390817683433<br />
Email: carlo.gualtieri@unina.it<br />
Abstract: Modeling mass-transfer across the sediment-water interface is a significant issue in<br />
environmental hydraulics. In fact diffusional exchanges of solutes between the bed sediment <strong>and</strong> the<br />
overlying water column could greatly affect water quality. Particularly, diffusional flux of dissolved<br />
oxygen (DO) towards the bed sediments from the water column could be responsible for low <strong>and</strong><br />
unacceptable levels of DO in the ecosystem. This flux depends both on sediment <strong>and</strong> flow characteristics.<br />
The objective of the present paper is to investigate the interaction between flow hydrodynamics <strong>and</strong><br />
dimensionless fluxes of dissolved substances across the sediment-water interface. Therefore, some<br />
literature predictive models are compared with experimental laboratory data collected both in flumes <strong>and</strong><br />
benthic chambers. Also, the influence of turbulent flow features on mass-transfer process is investigated<br />
using the available data. These data demonstrated a significant influence of the friction velocity u* on<br />
solutes flux for both data sets supporting the assumption that vortices in the near-wall region would affect<br />
that flux.<br />
Keywords: <strong>Environmental</strong> hydraulics, sediment-water interface, diffusive transport, sediment oxygen<br />
dem<strong>and</strong>.<br />
1. INTRODUCTION<br />
The benthic boundary layer (BBL), sometimes<br />
termed as bottom boundary layer, is a zone of<br />
paramount importance to the biology, chemistry,<br />
geology <strong>and</strong> physics of the oceans, seas, lakes <strong>and</strong><br />
even rivers. It is formed by those portions of<br />
sediment column <strong>and</strong> water column that are affected<br />
directly in the distribution of their properties <strong>and</strong><br />
processes by the presence of the sediment-water<br />
interface. Its importance is twofold (Lorke et al.,<br />
[2003]). First, within the BBL hydrodynamic energy<br />
is dissipated due to the bottom friction. Second, the<br />
BBL controls the exchange of solutes <strong>and</strong> particles<br />
between the sediment <strong>and</strong> the water. In fact, the bed<br />
could contain various types of chemicals, such as<br />
dissolved oxygen (DO), ammonia, hydrogen<br />
sulphide, organic chemical, heavy metals <strong>and</strong><br />
radionuclides. Such chemicals can be present within<br />
the bed both in dissolved or particulate form, i.e.<br />
attached on the particles forming the bottom<br />
sediments. Thus, chemicals sorbed onto sediments<br />
particles can be exchanged with the overlying water<br />
column through settling <strong>and</strong> resuspension or scour<br />
processes that are also greatly affected by the<br />
hydrodynamics of water flow (Chapra, [1997]).<br />
Dissolved chemicals could be then exchanged<br />
between the pore water <strong>and</strong> the water column across<br />
the sediment-water interface through mass-transfer<br />
processes, which are basically diffusive processes.<br />
Particularly, mass-transfer of dissolved oxygen,<br />
nitrogen <strong>and</strong> inorganic ions is of paramount<br />
importance in water quality <strong>and</strong> waste allocation<br />
load problems. Therefore, mass-transfer modeling<br />
can contribute to assess changes in water quality of<br />
river <strong>and</strong> streams due to the anthropic activities.<br />
The objective of the present paper is to investigate<br />
the interaction between flow hydrodynamics <strong>and</strong><br />
fluxes of dissolved substances across the sediment-<br />
962
˺<br />
is<br />
<strong>and</strong><br />
water interface. Therefore, some literature predictive<br />
models are considered <strong>and</strong> compared. Also, the<br />
influence of turbulent flow features on mass-transfer<br />
process is investigated using experimental laboratory<br />
data collected both in flumes <strong>and</strong> benthic chambers.<br />
2. DIFFUSIONAL EXCHANGE AT<br />
SEDIMENT-WATER INTERFACE<br />
In a lake the transition from the background flow, far<br />
away from lake bottom, to the flow at the sedimentwater<br />
interface is relatively simple due to the<br />
presence of the rigid boundary. The temporal<br />
structure of the BBL is steady as the bottom friction<br />
tends to remove fluctuations (Wüest <strong>and</strong> Lorke,<br />
[2003]). A logarithmic profile structure holds, the<br />
turbulence field is also stationary <strong>and</strong> the TKE<br />
equation is the balance between the production by<br />
Reynolds stresses <strong>and</strong> the dissipation<br />
˾, which<br />
provides the measure for turbulence level as:<br />
u*<br />
= (1)<br />
k z<br />
where u* is the friction velocity [L·T - ¹] that is equal<br />
to u*=(τ 0 /ρ) 0.5 , k is Von Kármán constant k=0.41, z<br />
is water height above the sediment-water interface<br />
[L], τ b is the bottom shear stress [N·L - ²] <strong>and</strong>, ρ is<br />
water density [M·L - ³].<br />
The vertical mass-transport within the turbulent BBL<br />
is a combination of molecular <strong>and</strong> turbulent diffusion<br />
<strong>and</strong> the vertical diffusivity K v [L²·T - ¹] is the sum of<br />
molecular D m [L²·T - ¹] <strong>and</strong> turbulent eddy diffusivity<br />
D t , [L²·T - ¹] which depends on the dissipation rate of<br />
turbulent kinetic energy <strong>and</strong> on the stability of the<br />
density stratification (Lorke et al., [2003]). In the<br />
natural environment, typically it is D t >>D m .<br />
However, D t decreases steeply with the water height<br />
z. In fact as turbulent eddies approach the sedimentwater<br />
interface, this interface tends to damp them as<br />
they approach closer than their length scale.<br />
Therefore, in the external area of the BBL where<br />
eddies move r<strong>and</strong>omly the mass-transport is<br />
dominated by eddy diffusion, whereas moving to the<br />
sediment the influence of turbulent eddy diffusivity<br />
decreases <strong>and</strong> close to the sediment, where<br />
turbulence is low, the vertical transport is dominated<br />
by molecular diffusion. D m values depend mainly on<br />
the solutes exchanged <strong>and</strong> on the water temperature.<br />
Also, classical boundary layer theory states that near<br />
the bottom there is a sublayer, termed viscous<br />
boundary layer (VBL), where the flow is laminar <strong>and</strong><br />
velocity gradient is constant. This sublayer acts as a<br />
region of resistance to the transfer of momentum,<br />
˾<br />
3<br />
heat <strong>and</strong> mass. Within the VBL the momentum<br />
transfer is dominated by viscous forces <strong>and</strong> its<br />
thickness δ v could be defined as the height where D t<br />
in equal to the kinematic viscosity ν [L²·T - ¹]. The<br />
height δ v could be estimated as:<br />
11 ˽v = (2)<br />
* ŭ<br />
<strong>and</strong> δ v is typically δ v ≈10 - ² m. Approaching further to<br />
the sediment-water interface, turbulent diffusivity D t<br />
decreases up to molecular diffusivity D m . This<br />
defines the thickness δ c of the concentration<br />
boundary layer or diffusivity boundary layer (DBL),<br />
where the transport due to the eddies becomes<br />
negligible compared to molecular diffusion. The<br />
diffusive boundary layer is extremely thin, much<br />
smaller than the VBL. According to the dependence<br />
with z of D t , the thickness δ c could be related with<br />
the thickness δ v as:<br />
δ v<br />
δ c = α<br />
Sc<br />
(3)<br />
where Sc=ν/D m is the Schmidt number, ratio of the<br />
kinematic viscosity ν to molecular diffusivity D m <strong>and</strong><br />
a coefficient which is usually assumed to be<br />
between ¼ (Wüest <strong>and</strong> Lorke, [2003]). Eq. (3)<br />
demonstrates that δ c is solute-specific <strong>and</strong> is slighty<br />
temperature-dependent, as both ν <strong>and</strong> D m change<br />
with temperature. If<br />
˺=⅓, eq. (3) shows that δ c is for<br />
the substances of environmental concern range from<br />
1/13 to 1/6 the thickness of the velocity boundary<br />
layer δ v . Sometimes, since Sc≈10³ <strong>and</strong> Sc⅓≈10, δ c is<br />
approximated as δ c =0.1·δ v . Lorke has proposed a<br />
different scaling for δ c , assuming that for lowenergetic<br />
systems, such as lakes <strong>and</strong> reservoirs, the<br />
DBL thickness is forced by the BBL turbulence,<br />
whereas for high-energetic systems, such as streams<br />
<strong>and</strong> estuaries, δ<br />
⅓<br />
c value is controlled by the current<br />
velocity (Lorke et al., [2003]). Therefore, the<br />
Batchelor length scale L B , which describes the<br />
smallest length of turbulent concentration<br />
fluctuations before molecular diffusion smoothes the<br />
remaining gradients, could be used at least as firstorder<br />
approximation to define δ c as δ c =L B . This<br />
assumption leads to<br />
˺=½.<br />
However, the thickness of DBL could not be exactly<br />
defined from a physical point of view because its<br />
boundaries are not sharp (Wüest <strong>and</strong> Lorke, [2003]).<br />
At the upper boundary a transition zone where D t<br />
<strong>and</strong> D m are comparable, exists, while the lower limit,<br />
i.e. the sediment-water interface, is not an horizontal<br />
plate but is sculptured into elaborate l<strong>and</strong>scapes<br />
when viewed at the scale of the DBL [Røy et al.,<br />
963
˽<br />
[2002]). Also, this interface is quite permeable so<br />
horizontal currents could diffuse slightly into the<br />
porewater. Moreover, in shallow waters heating of<br />
the sediments can lead to buoyant porewater<br />
convection, which further increases the exchange<br />
between the sediments <strong>and</strong> the water (Wüest <strong>and</strong><br />
Lorke, [2003]). In these conditions, advection<br />
dominates on diffusive transport. Finally, in<br />
eutrophic waters, advective transport across the<br />
sediment-water interface occurs mainly by methane<br />
<strong>and</strong> carbon dioxide bubbles formed in the anoxic<br />
layer of the sediments (Wüest <strong>and</strong> Lorke, [2003]).<br />
Nevertheless if the sediment-water interface is<br />
treated as an infinite plane crossed by onedimensional<br />
chemical gradient, the classical Fickian<br />
diffusion model could be applied. Using this<br />
approach, if D m is constant, the mass flux across the<br />
sediment-water interface could be modeled as:<br />
dC<br />
J flux = Dm<br />
(4)<br />
dz<br />
where J is the vertical mass flux per unit interfacial<br />
area [M·L - ²·T - ¹] <strong>and</strong> dC/dz is the concentration<br />
gradient over z of the exchanged solute, which is the<br />
driving force of the diffusional process. Assuming<br />
that there is no solute production or consumption<br />
within the DBL, a linear solute concentration profile<br />
exists <strong>and</strong> Eq. (4) could be approximated as:<br />
˝C<br />
J = D<br />
(5)<br />
flux<br />
m<br />
c<br />
where ˝C is the ˝C=C ∞ -C 0 , if C ∞ <strong>and</strong> C 0 are solute<br />
concentration in the bulk water <strong>and</strong> at the sedimentwater<br />
interface, respectively [M·L - ³]. The ratio D m /δ c<br />
is usually replaced with a conductance term, i.e. the<br />
mass-transfer coefficient K m-t [L·T - ¹], that relates the<br />
driving force to the mass flux. Thus, eq. (5) yields:<br />
J flux = K m-t˝C<br />
(6)<br />
The concentrations within the DBL could be<br />
measured using microelectrodes, that allows to work<br />
at very high spatial distribution (Güss, [1998]; Lorke<br />
et al., [2003]), whereas K m-t should be estimated.<br />
Generally, K m-t is a function of the fluid <strong>and</strong> solute<br />
properties, surface geometry, <strong>and</strong> flow conditions<br />
(Steinberger <strong>and</strong> Hondzo, [1999]).<br />
The diffusional transfer of solutes through the BBL<br />
influences a number of important biological <strong>and</strong><br />
geochemical processes in the upper sediments such<br />
as the dissolution of calcium carbonate, the oxidation<br />
of organic matter <strong>and</strong> metals (iron, manganese, etc.),<br />
the removal of reactive nitrogen by denitrification,<br />
the supply of oxygen to obligate-aerobic sedimentdwelling<br />
organisms, the growth of microbial mats,<br />
<strong>and</strong> the release of contaminants from polluted<br />
sediments (Wüest <strong>and</strong> Lorke, [2003]). Particularly,<br />
diffusional flux of dissolved oxygen (DO) towards<br />
the bed sediments from the overlying water column<br />
has been intensively investigated because it is often<br />
responsible for low <strong>and</strong> unacceptable levels of DO in<br />
the ecosystem. This diffusional flux is due to the<br />
production of oxygen-consuming substances, such as<br />
methane <strong>and</strong> ammonium ion, that are then oxidated<br />
in the aerobic layer of the bed sediments resulting in<br />
a sediment oxygen dem<strong>and</strong> (SOD) (Chapra, [1997];<br />
Gualtieri, [2001]). SOD value could be directly<br />
measured or predicted using modeling framework<br />
(Chapra, [1997]). SOD value depends both on<br />
sediment <strong>and</strong> flow over sediment characteristics<br />
(Nakamura, [1994]; Nakamura <strong>and</strong> Stefan, [1994];<br />
Mackenthun <strong>and</strong> Stefan, [1994]; Mackenthun <strong>and</strong><br />
Stefan, [1998]; Josiam <strong>and</strong> Stefan, [1999]; Higashino<br />
et al., [2003]; Gualtieri, [in press]). Laboratory<br />
measurements revealed a significant decrease of δ c<br />
for increasing flow velocities (Gundersen <strong>and</strong><br />
Jørgensen, [1990]; Hondzo, [1998]; Steinberger <strong>and</strong><br />
Hondzo, [1999]). Thus, at low flow velocities,<br />
diffusive transport is the limiting factor of SOD<br />
production, which increases as flow velocity erodes<br />
the DBL. In this case, when near-bottom velocity is<br />
the key parameter, the process is termed as waterside<br />
controlled. As the velocity grows, at some point,<br />
the rate of metabolic <strong>and</strong> chemical reactions are not<br />
limited by the rate of transport of DO through the<br />
DBL. Therefore, biochemical reactions within the<br />
sediments becomes the limiting factor <strong>and</strong> SOD is<br />
independent of the current velocity over the<br />
sediment. In this case, mass-transfer is termed as<br />
sediment-side controlled.<br />
In lakes, where turbulence is often low <strong>and</strong> there is<br />
an high availability of organic matter within the<br />
sediment, the microbiological activity <strong>and</strong> the<br />
consequent SOD flux is usually limited by the<br />
physical constraints of the diffusional transport, i.e. it<br />
is water-side controlled. In the present paper this<br />
case is considered <strong>and</strong> investigated.<br />
3. PREDICTIVE MODELS FOR MASS-<br />
TRANSFER COEFFICIENT K m-t<br />
Several predictive equations have been proposed to<br />
estimate mass-transfer coefficient K m-t . In this<br />
section 4 equations are presented <strong>and</strong> applied.<br />
Nakamura (Nakamura, [1994]), applying the<br />
similarity theory of the bottom shear stress <strong>and</strong> the<br />
turbulent heat o mass transfer, derived:<br />
964
has<br />
is<br />
=8·τ<br />
K<br />
m−t<br />
2<br />
= ⋅ n1<br />
⋅ λ ⋅U<br />
⋅ Sc<br />
π<br />
−0.75<br />
(7)<br />
where n 1 =0.109 <strong>and</strong> the classic friction factor of<br />
Darcy-Weisbach equation, that is 0 /ρ·U², if U<br />
is the mean velocity of the flow over the sediment<br />
[L·T - ¹]. By applying the analysis of heat transfer to<br />
the diffusional mass transfer through the diffusive<br />
boundary layer, predictive equation for K m-t has been<br />
obtained, if n 2 =0.1, as (Higashino <strong>and</strong> K<strong>and</strong>a, 1999):<br />
̄ ̄<br />
K<br />
m−t<br />
3⋅<br />
6<br />
= ⋅ n2<br />
⋅ λ ⋅U<br />
⋅ Sc<br />
8 ⋅π<br />
−0.66<br />
(8)<br />
Hondzo (Hondzo, [1998]) has conducted laboratory<br />
experiments to elucidate dissolved oxygen transfer<br />
mechanism at the sediment-water interface over a<br />
smooth bed. Hondzo has derived:<br />
K<br />
m−t<br />
−2<br />
3<br />
= 0.0558 ⋅ u* ⋅Sc<br />
(9)<br />
A different equation was derived from the same data<br />
set, plotting the data of dimensionless K m-t against<br />
Reynolds number Re for the mean velocity U<br />
(Steinberger <strong>and</strong> Hondzo, [1999]):<br />
Dm<br />
0.89±<br />
A2 1 3<br />
K m− t = ( 0.012 ± A1) ⋅ Re ⋅ Sc (10)<br />
h<br />
where A1=±0.001 <strong>and</strong> A2=±0.05 are the 90%<br />
confidence intervals for the mean coefficient <strong>and</strong> the<br />
mean exponent, respectively.<br />
4. COMPARISON OF AVAILABLE MODELS<br />
In this section, the previously outlined predictive<br />
models for K m-t have been tested for an idealized<br />
scenario to be compared. Assuming a channel reach<br />
with bed slope J b =0.001, mass-transfer coefficient<br />
for dissolved oxygen K m-t through the sedimentwater<br />
interface has been computed using eqs. 7/8/9<br />
<strong>and</strong> 10 as a function of streamflow mean velocity U.<br />
The friction factor been estimated using wellknown<br />
Blasius equation for smooth surfaces as<br />
̄=0.316·R -0.25 . All dissolved oxygen <strong>and</strong> water<br />
parameters involved in the estimation are in Table 1.<br />
Test results are presented in Fig.1a, where the masstransfer<br />
coefficient K m-t has been plotted against<br />
streamflow mean velocity U. The considered range<br />
for U is from 0 to 0.40 m/s. For the Steinberger-<br />
Hondzo equation, three couples of values for A 1 <strong>and</strong><br />
A 2 have been considered. They are termed as S-H<br />
̄<br />
max ,<br />
S-H <strong>and</strong> S-H min , respectively.<br />
Inspection of results shows that for U=0.40 m/s S-<br />
H max <strong>and</strong> S-H min exhibit the highest <strong>and</strong> lowest values<br />
for K m-t . Thus, eq. (10), that was experimentally<br />
derived, encompasses all the remaining equations.<br />
Table 1 – Input data for idealized scenario<br />
Channel<br />
Slope J b 0.001<br />
Dissolved oxygen<br />
Molecular weight M – g/mole 32<br />
Molecular diffusivity D m -m²/s 1.80×10 -9<br />
Schmidt number Sc 557.22<br />
Water<br />
Temperature T - °C 20<br />
Density ρ – kg/m³ 998.15<br />
Specific weight γ – N/m³ 9787.89<br />
Surface tension T s – N/m 0.07276<br />
Kinematic viscosity ν - m²/s 1.003×10 -6<br />
K m-t - m/s<br />
5.E-05<br />
4.E-05<br />
3.E-05<br />
2.E-05<br />
1.E-05<br />
0.E+00<br />
Fig.1a - K m-t vs U<br />
Nakamura<br />
S-Hmax<br />
Higashino<br />
Hondzo<br />
S-H<br />
S-Hmin<br />
0.0 0.1 0.2 0.3 0.4<br />
U - m/s<br />
The range of mass-transfer coefficient K m-t for<br />
U=0.40 m/s is comprised from 1.32×10 -5 m/s to<br />
4.22×10 -5 m/s. Also, discarding value from S-H min ,<br />
for U=0.40 m/s, K m-t is in the range from 1.88×10 -5<br />
m/s to 4.22×10 -5 m/s. Moreover, eq. (7) <strong>and</strong> S-H max<br />
always exhibit similar result. Finally, Higashino eq.<br />
(8) provides intermediate K m-t values.<br />
Fig.1b presents test results in terms of Sherwood<br />
number Sh against Reynolds number Re* based on<br />
the friction velocity u*. The Sherwod number is:<br />
K m-t ⋅ z<br />
Sh = (11)<br />
D<br />
m<br />
<strong>and</strong> it represent a dimensionless mass-transfer flux.<br />
Sh values are in the range from 388 to 1242 for<br />
965
Reynolds number Re*=1200. The S-H max equation<br />
<strong>and</strong> Nakamura eq. (7) closely agree.<br />
Sh<br />
1500<br />
1250<br />
1000<br />
750<br />
500<br />
250<br />
0<br />
Fig.1b - Sh vs Re*<br />
Nakamura<br />
S-Hmax<br />
Higashino<br />
Hondzo<br />
S-H<br />
S-Hmin<br />
0 400 800 1200<br />
Re*<br />
5. ANALYSIS OF EXPERIMENTAL DATA.<br />
DISCUSSION<br />
In this section, experimental data are analyzed to<br />
confirm the influence of friction velocity u* on the<br />
mass-transfer process. Available data refer to SOD<br />
studies (Glud et al., [1995]; Mackenthun <strong>and</strong> Stefan,<br />
[1998]; Steinberger <strong>and</strong> Hondzo, [1999]; House,<br />
[2003]; Røy et al., [2004]; Tengberg et al., [2004]).<br />
Mackentum-Stefan, Hondzo, House <strong>and</strong> Røy data<br />
sets were collected in laboratory flumes, whereas<br />
Glud <strong>and</strong> Tengberg data sets refer to benthic<br />
chamber measurements. Particularly, Tengberg data<br />
were collected in three different types of benthic<br />
chambers (Tengberg et al., [2004]). The temperature<br />
of the considered data is in the range from 8.6 to<br />
23.8 °C <strong>and</strong> the Schmidt number Sc values are<br />
accordingly from 1032 to 454.<br />
The values of Sh are in the range from 75 to 2500,<br />
while Re* values are comprised from 145 to 1645.<br />
These values correspond to mass-transfer rate K m-t<br />
values from 1.74×10 -6 to 3.88×10 -5 m/s <strong>and</strong> friction<br />
velocity u* values in the range from 0.0018 to<br />
0.0190 m/s. Near-bottom current speed are in lakes<br />
usually in the range from 0.02 to 0.1 m/s, but they<br />
can reach also more than 0.2 m/s during storms,<br />
especially in shallow waters (Wüest <strong>and</strong> Lorke,<br />
[2003]). In coastal areas the flow velocity near the<br />
bed could be typically of 0.02-0.04 m/s (Glud et al.,<br />
[1995]). Thus, friction velocity u* typically ranges<br />
from 0.0005 to 0.005 m/s in lakes (Wüest A. <strong>and</strong><br />
Lorke A., [2003]), whereas in streams <strong>and</strong> river u*<br />
belongs to the range from 0.0001 to 0.01 m/s<br />
(Higashino et al., [2004]).<br />
The experimental data are presented in Fig.2, where<br />
K m-t is plotted against u*. Notably, friction velocity<br />
u* data were not available in the Glud data set. They<br />
were calculated using the equation proposed by<br />
Pullin et al. (Pullin et al.) for radial flow impellers:<br />
B<br />
u* = 0.026 N D<br />
(12)<br />
h<br />
where N is the impeller rate of rotation [T - ¹], D is the<br />
impeller diameter [L], B is the width or diameter for<br />
the chamber [L] <strong>and</strong> h is the height from the<br />
sediments to the stirrer [L].<br />
K m-t - m/s<br />
5.E-05<br />
4.E-05<br />
3.E-05<br />
2.E-05<br />
1.E-05<br />
0.E+00<br />
Fig.2 - Experimental data<br />
Glud<br />
Mackentum<br />
Hondzo<br />
House<br />
Røy<br />
Tengberg<br />
y=1.498x<br />
R 2 =0.986<br />
0.000 0.005 0.010 0.015 0.020<br />
u*- m/s<br />
y=0.554x<br />
R²=0.897<br />
The analysis of data generally confirms the<br />
dependence of mass-transfer rate K m-t from the<br />
friction velocity u*. A linear relationship between<br />
K m-t <strong>and</strong> u* is supported both for flumes <strong>and</strong> benthic<br />
chambers data (Fig.2). Data collected in benthic<br />
chambers are generally higher than those taken in<br />
laboratory flumes. However, u* Tengberg data were<br />
obtained through an hydrodynamic characterization<br />
of the chambers where a PVC plate simulated the<br />
sediments surface. Thus, u* values only give<br />
indications about the prevailing hydrodynamic<br />
conditions in the chambers during the sediment<br />
incubation (Tengberg et al., [2004]. Since sediments<br />
surface is rougher than PVC plate, u* <strong>and</strong> also Re*<br />
were underestimated <strong>and</strong> the slope of Tengberg data<br />
in Fig.2 should be lower. As Tengberg data, Hondzo<br />
data set also refers to hydraulically smooth bed of<br />
artificial or riverine sediments (Hondzo, [1998]).<br />
966
Glud data were collected on a rough sediment<br />
surface (Glud et al., [1995]).<br />
Sc -0.5× Sh/Re*<br />
1.00<br />
0.10<br />
0.01<br />
Fig.3 - LE <strong>and</strong> SE models<br />
y = 0.059x -0.030<br />
R 2 = 0.005<br />
y = 0.048x -0.125<br />
R 2 = 0.095<br />
10 100 1000 10000<br />
Re*<br />
Flumes<br />
Benthic chamber<br />
To better underst<strong>and</strong> how hydrodynamics control the<br />
rate of mass-transfer turbulent flow features, such as<br />
turbulent eddies size, should be considered. In a<br />
turbulent flow the length scale of the eddies ranges<br />
from the flow domain, i.e. integral scale eddies, to<br />
smaller sizes, i.e. Kolmogorov scale eddies. At the<br />
integral scale, larger eddies break down into multiple<br />
smaller eddies efficiently transferring their energy<br />
with little loss. At the Kolmogorov scale, viscosity<br />
converts kinetic energy into heat (Pope, 2000. Thus,<br />
it is possible to assume that mass-transfer process at<br />
sediment-water interface would be controlled either<br />
by larger eddies either by smaller eddies. This<br />
hypothesis leads to the large-eddy <strong>and</strong> small-eddy<br />
models, respectively. These models could be<br />
generally represented as:<br />
or as:<br />
K<br />
m-t<br />
u*<br />
1<br />
-0.5<br />
n<br />
= c Sc Re*<br />
(13a)<br />
0.5 n+<br />
1<br />
1 Sc Re*<br />
Sh = c<br />
(13b)<br />
where c 1 is a constant, n=-0.50 for the large-eddy<br />
model <strong>and</strong> n=-0.25 for the small-eddy model.<br />
Available experimental data are also compared with<br />
results from large-eddy (LE) <strong>and</strong> small eddy (SE)<br />
models (Fig.3). Both models are not supported since<br />
n values are -0.125 <strong>and</strong> -0.03 for flumes data <strong>and</strong><br />
chambers data, respectively. This result is consistent<br />
with that was found by Hondzo (Hondzo, [1998]).<br />
Notably, experimental data for the mass-transfer<br />
process at the air-water interface demonstrated that<br />
the small-eddy model should be preferred to largeeddy<br />
model (Moog <strong>and</strong> Jirka, [1999]; Gualtieri <strong>and</strong><br />
Gualtieri [2004]).<br />
Sc -0.5× Sh<br />
120<br />
100<br />
80<br />
60<br />
40<br />
20<br />
Fig.4 - Streamwise vortex model<br />
0<br />
Flumes<br />
Benthic chamber<br />
y = 0.056x<br />
R 2 = 0.922<br />
y = 0.023x<br />
R 2 = 0.922<br />
0 500 1000 1500 2000<br />
Re*<br />
Counter-rotating quasi-streamwise vortices are often<br />
present at the sediment-water interface (Nino <strong>and</strong><br />
Garcia, [1996]). Those vortices would control masstransfer<br />
across that interface through a mechanism<br />
described by Hondzo (Hondzo, [1998]). As a<br />
turbulent motion reach the interface, it renews it to<br />
the bulk water concentration. After that, molecular<br />
diffusion returns the sediment-water interface to bed<br />
concentration. The presence of the streamwise vortex<br />
creates a pumping effect that produces ejection of<br />
low-momentum fluid, with low DO concentration on<br />
one side of the vortex core <strong>and</strong> the injection of highmomentum<br />
fluid, with high DO concentration<br />
toward the bed on the other (Hondzo, [1998]). This<br />
mechanism appears to be common to the flow over<br />
both smooth <strong>and</strong> rough surfaces (Nino <strong>and</strong> Garcia,<br />
[1996]). This vortex has a velocity scale U v of u*, a<br />
length scale L v of<br />
̆/u* <strong>and</strong> a time scale T v of<br />
̆/u*².<br />
Thus, the distance of diffusive transport is given by<br />
(D m ×T v ) 0.5 <strong>and</strong> the mass-transfer coefficient K m-t is :<br />
K<br />
m-t<br />
which yields:<br />
K<br />
m -t<br />
( D × T )<br />
v<br />
0.5<br />
0.5<br />
m<br />
0.5<br />
v<br />
m v D<br />
∝ ≈ (14a)<br />
T T<br />
0.5<br />
m<br />
0.5<br />
D u*<br />
-0.5<br />
∝ ≈ u* Sc (14b)<br />
̆<br />
Therefore, the streamwise vortex (SV) model is:<br />
967
0.5<br />
Sh = c2 Re* Sc<br />
(15)<br />
where c 2 is a numerical constant. This model was<br />
also tested using the 2 sets of available experimental<br />
data (Fig.4). The analysis of data demonstrates a<br />
significant influence of the friction velocity u* on the<br />
dimensionless flux, i.e. Sherwood number Sh,<br />
supporting the assumption that counter-rotating<br />
quasi-streamwise vortices which are present in the<br />
near-wall region could affect mass-transfer process<br />
through the sediment-water interface.<br />
1500<br />
1250<br />
1000<br />
Fig.5 - Sh vs Re*<br />
eq. (7) eq. (8)<br />
S-Hmin<br />
Glud<br />
M-S<br />
S-H<br />
House<br />
Tengberg<br />
δc - mm<br />
1.5<br />
1.3<br />
1.0<br />
0.8<br />
0.5<br />
0.3<br />
0.0<br />
Fig.6 - δ c vs u*<br />
y = 0.002x -1.002<br />
R 2 = 0.878<br />
y = 0.014x -0.551<br />
R 2 = 0.932<br />
y = 0.001x -0.956<br />
R 2 = 0.961<br />
0.000 0.005 0.010 0.015 0.020<br />
u*- m/s<br />
Glud<br />
Mackentum<br />
Hondzo<br />
House<br />
Røy<br />
Tengberg<br />
Sh<br />
750<br />
6. CONCLUDING REMARKS<br />
500<br />
250<br />
0<br />
0 400 800 1200<br />
Re*<br />
Moreover, available experimental data are compared<br />
with eqs. (7) (8) <strong>and</strong> with S-H min equation (Fig.5).<br />
The comparison shows that eq. (7) <strong>and</strong> S-H min<br />
equation encompass all the flumes data. However,<br />
most of the benthic chambers data are higher than<br />
the values predicted by the considered models.<br />
Finally, the influence of the friction velocity u* on<br />
the thickness of diffusive boundary layer δ c was<br />
investigated (Fig.6). The data confirmed δ c erosion<br />
with the increasing friction velocity u*. Particularly,<br />
regression analysis of both Hondzo <strong>and</strong> Tengberg<br />
data, which were collected on a hydraulically smooth<br />
surface, provides the equation:<br />
˽= p<br />
c u<br />
(16)<br />
c 3 *<br />
where c 3 is a numerical constant <strong>and</strong> p is the power<br />
law exponent, which is –1.002 <strong>and</strong> -0.956 for<br />
Hondzo set <strong>and</strong> Tengberg set, respectively. Since the<br />
exponent p is very close to 1, these data sets confirm<br />
the linear relationship predicted by eqs. (2) <strong>and</strong> (3).<br />
However, Glud data, collected on rough surface,<br />
provide a p value of –0.551.<br />
Modeling the fluxes of solutes across the sedimentwater<br />
interface is a relevant contribute water quality<br />
analysis of rivers <strong>and</strong> lakes. The objective of the<br />
present paper was to investigate the interaction<br />
between flow hydrodynamics <strong>and</strong> dimensionless<br />
fluxes of dissolved substances across the sedimentwater<br />
interface. Therefore, some literature predictive<br />
models were compared with experimental laboratory<br />
data collected both in flumes <strong>and</strong> benthic chambers.<br />
These models encompass only flume data, whereas<br />
benthic chambers data exhibit higher values. Also,<br />
the influence of turbulent eddies on mass-transfer<br />
process was investigated using the available data.<br />
These data demonstrated that experimental data<br />
collected both in flumes <strong>and</strong> in benthic chambers are<br />
significantly correlated with flow friction velocity u*<br />
supporting the hypothesis that counter-rotating<br />
streamwise vortices which are present in the nearbed<br />
region could affect mass-transfer process<br />
through the sediment-water interface.<br />
REFERENCES<br />
Chapra S.C. (1997). Surface water quality modeling.<br />
McGraw-Hill, New-York<br />
Glud R.N., Gundersen J.K., Revsbech N.P.,<br />
Jørgensen B.B. <strong>and</strong> H, Hüttel M. (1995).<br />
Calibration <strong>and</strong> performance of the stirred<br />
flux chamber from the benthic l<strong>and</strong>er Elinor.<br />
Deep Sea Research I, vol.42, n.6, pp.1029-<br />
1042<br />
968
Gualtieri C. (2001). Dimensionless steady-state<br />
NSOD model. Proceedings of 29 th IAHR<br />
Congress, Beijing, China, September 16/21<br />
Gualtieri C. <strong>and</strong> Gualtieri P. (2004). Turbulencebased<br />
models for gas transfer analysis with<br />
channel shape factor. <strong>Environmental</strong> Fluid<br />
Mechanics, vol.4, n.3, September 2004,<br />
pp.249-271<br />
Gualtieri C. (in press). Discussion on “M.Higashino,<br />
H.G.Stefan <strong>and</strong> C.J.Gantzer: Periodic<br />
diffusional mass transfer near sediment/water<br />
interface: Theory. J.Env.Eng., ASCE, vol.129,<br />
n.5, May 2003, pp.447-455.” in press<br />
Higashino M. <strong>and</strong> K<strong>and</strong>a T. (1999). Fundamental<br />
studies on release of dissolved substance from<br />
bottom sediment to flowing water.<br />
Proceedings of 28 th IAHR Congress, Graz,<br />
Austria, August 22/27, 1999<br />
Higashino M., Stefan H.G. <strong>and</strong> Gantzer C.J. (2003).<br />
Periodic diffusional mass transfer near<br />
sediment/water interface: Theory. J.Env.Eng.,<br />
ASCE, vol.129, n.5, May 2003, pp.447-455<br />
Higashino M., Gantzer C.J. <strong>and</strong> Stefan H.G. (2004).<br />
Unsteady diffusional mass transfer at the<br />
sediment/water interface: Theory <strong>and</strong><br />
significance for SOD measurements. Water<br />
Research, vol.38, pp.1-12<br />
Hondzo M. (1998). Dissolved oxygen transfer at the<br />
sediment-water interface in a turbulent flow.<br />
Water Resources Research, vol.34, 12,<br />
pp.3525-3533<br />
House W.A. (2003). Factors influencing the extent<br />
<strong>and</strong> development of the oxic zone in<br />
sediments. Bioegeochemistry, vol.63, pp.317-<br />
333<br />
Josiam R.M. <strong>and</strong> Stefan H.G. (1999). Effect of flow<br />
velocity on sediment oxygen dem<strong>and</strong>:<br />
comparison of theory <strong>and</strong> experiments.<br />
J.American Water Resources Association,<br />
vol.35, n.2, pp.433-439<br />
Lorke A., Müller B., Maerki M. <strong>and</strong> Wüest A.<br />
(2003). Breathing sediments: The control of<br />
diffusive transport across the sediment-water<br />
interface by periodic boundary-layer<br />
turbulence. Limnology <strong>and</strong> Oceanography,<br />
vol.48, n.6, pp.2077-2085<br />
Moog D.B. <strong>and</strong> Jirka G.H. (1999). Air-water gas<br />
transfer in uniform channel flow. J.Hydraulic<br />
Engineering, ASCE, 125, 1, January 1999,<br />
pp.3-10<br />
Mackenthun A.A. <strong>and</strong> Stefan H.G. (1994).<br />
Experimental study of sedimentary oxygen<br />
dem<strong>and</strong> in lakes; dependance on near-bottom<br />
flow velocities <strong>and</strong> sediment properties.<br />
University of Minnesota, St.Anthony Falls<br />
Hydraulic Laboratory, Project Report n.358,<br />
Minneapolis, MI<br />
Mackenthun A.A. <strong>and</strong> Stefan H.G. (1998). Effect of<br />
flow velocity on SOD: experiments.<br />
J.Env.Eng.Div. ASCE, vol.124, n.3, pp.222-<br />
230<br />
Nakamura Y. <strong>and</strong> Stefan H.G. (1994). Effect of flow<br />
velocity on SOD: theory. J.Env.Eng.Div.<br />
ASCE, vol.120, n.5, pp.996-1016<br />
Nakamura Y. (1994). Effect of flow velocity on<br />
phosphate release from sediment. Water<br />
Science & Technology, vol.30, n.10, pp.263-<br />
272<br />
Nino Y. <strong>and</strong> Garcia M.H. (1996). Experiments on<br />
particle-turbulence interactions in the nearwall<br />
region of an open channel flow:<br />
Implications for sediments transport. J. Fluid<br />
Mechanics, vol.326, pp.285-319<br />
Pullin C., Oldham C. <strong>and</strong> Ivey G. . A simple<br />
estimation of friction velocities in stirred<br />
benthic chambers. Centre for Water Research<br />
Report ED 1691CP<br />
Røy H, Hüttel M. <strong>and</strong> Jørgensen B.B. (2002). The<br />
role of small-scale sediment topography for<br />
oxygen flux across the diffusive boundary<br />
layer. Limnology <strong>and</strong> Oceanography, vol.47,<br />
n.3, pp.837-847<br />
Røy H, Hüttel M. <strong>and</strong> Jørgensen B.B. (2004).<br />
Transmission of oxygen concentration<br />
fluctuations through the diffusive boundary<br />
layer overlying aquatic sediments. Limnology<br />
<strong>and</strong> Oceanography, vol.49, n.3, pp.686-692<br />
Steinberger N. <strong>and</strong> Hondzo M. (1999). Diffusional<br />
mass transfer at sediment-water interface.<br />
J.Env.Eng.Div. ASCE, vol.125, n.2, pp.192-<br />
200<br />
Tengberg A., Stahl H., Gust G., Müller V., Arning<br />
U., Andersson <strong>and</strong> Hall P.O.J. (2004).<br />
Intercalibration of benthic flux chambers I.<br />
Accuracy of flux measurements <strong>and</strong> influence<br />
of chamber hydrodynamics. Progress in<br />
Oceanography, vol.60, pp.1-28<br />
Wüest A. <strong>and</strong> Lorke A. (2003). Small-scale<br />
hydrodynamics in lakes. Annual Review in<br />
Fluid Mechanics, vol.35, pp.373-412<br />
969
A Spatially-Distributed Conceptual Model For Reactive<br />
Transport Of Phosphorus From Diffuse Sources: An<br />
Object-Oriented Approach<br />
B. Koo, S. Dunn <strong>and</strong> R. Ferrier<br />
The Macaulay Institute, Craigiebuckler, Aberdeen, AB15 8QH, UK<br />
Abstract: This paper presents CAMEL, a spatially-distributed conceptual model for simulating reactive<br />
transport of phosphorus from diffuse sources at the catchment scale. A catchment is represented in the model<br />
using a network of grid cells <strong>and</strong> each grid cell is comprised of various conceptual storages of water,<br />
sediment <strong>and</strong> phosphorus. To allow for reactive transport processes of phosphorus between grid cells, two<br />
cascade routing schemes are used for groundwater <strong>and</strong> channel water flows, respectively. The model has a<br />
modular, object-oriented structure so that it can be easily modified or extended <strong>and</strong>, furthermore, it can even<br />
provide a library of hydrological <strong>and</strong> hydrochemical processes from which the user can select a sub-set of<br />
processes suitable for a particular application. A verification study of the model has been carried out for a<br />
hypothetical catchment with satisfactory results.<br />
Keywords: Phosphorus; Diffuse source; Reactive transport; Spatially-distributed; Object-oriented; CAMEL<br />
1. INTRODUCTION<br />
Phosphorus (P) is an essential element for plant<br />
growth <strong>and</strong> its input to the soil has long been<br />
recognised as necessary to maintain profitable crop<br />
<strong>and</strong> livestock production. Excess inputs of P,<br />
however, may cause eutrophication of fresh<br />
waters. Many st<strong>and</strong>ing waters in the UK have<br />
undergone eutrophication <strong>and</strong> many UK rivers are<br />
heavily polluted with P [Withers <strong>and</strong> Lord, 2002].<br />
This has focussed attention on the pollution of<br />
freshwaters by P loss from agricultural diffuse<br />
sources.<br />
P is readily adsorbed to sediment particles <strong>and</strong><br />
forms insoluble precipitates with cations such as<br />
iron, aluminum <strong>and</strong> calcium [Sample et al., 1980].<br />
Because of this phenomenon, called P retention, P<br />
is strongly associated with sediment particles in<br />
the soil. Consequently, the majority of P from<br />
diffuse sources is transported by surface runoff in<br />
particulate forms. However, surface runoff or subsurface<br />
drainage can also transport significant<br />
amounts of dissolved P particularly if the soil is<br />
overloaded with P <strong>and</strong> the soil/geology has a low<br />
adsorption capacity for P.<br />
During the course of delivery from the soil to the<br />
river system, P undergoes numerous<br />
transformation processes. Important processes<br />
related to P transformation in the stream include:<br />
detachment <strong>and</strong> deposition of sediment particles;<br />
adsorption <strong>and</strong> desorption of soluble P to/from<br />
sediment particles [House et al., 1995]; coprecipitation<br />
of P with calcite in hardwaters<br />
[House <strong>and</strong> Donaldson, 1986; Jarvie et al, 2002];<br />
formation of the ferrous phosphate mineral<br />
vivianite in anoxic sediments [Woodruff et al.,<br />
1999]; <strong>and</strong> P uptake by aquatic plants through<br />
either root or shoot. The combination of all of<br />
these processes, in t<strong>and</strong>em with variations in river<br />
flow <strong>and</strong> other environmental factors, makes the<br />
transport process of P very complicated.<br />
The significance of each of the above processes<br />
varies greatly in space. A small portion of the<br />
catchment may contribute a large proportion of P<br />
load [Gburek <strong>and</strong> Sharpley, 1998]. These areas<br />
have been termed critical source areas <strong>and</strong> are<br />
characterised by having high potential to release P<br />
into surface or subsurface runoff in conjunction<br />
with hydrologic connectivity with streams or<br />
ditches. Targeting critical source areas would<br />
increase the efficiency <strong>and</strong> reduce the economic<br />
costs of control [Needelman et al., 2001].<br />
Therefore, in the context of catchment management,<br />
it is important to identify critical source<br />
areas <strong>and</strong> major transport processes of P from<br />
those areas. A spatially-distributed P transport<br />
model can be a useful tool for these purposes.<br />
970
This paper presents a spatially-distributed<br />
conceptual model, CAMEL (Catchment Analysis<br />
Model for <strong>Environmental</strong> L<strong>and</strong>-uses) v1.0, that has<br />
been developed for the assessment of long-term<br />
effects of agricultural l<strong>and</strong> use changes on water<br />
quality in terms of sediment <strong>and</strong> P concentrations<br />
in the water.<br />
2. A REVIEW OF EXISTING MODELS<br />
There are a number of existing models that can<br />
simulate dynamics of P transport at a catchment<br />
scale in a distributed or semi-distributed manner.<br />
Examples of these models include AnnAGNPS<br />
[Cronshey <strong>and</strong> Theurer, 1998], ANSWERS-2000<br />
[Bouraoui <strong>and</strong> Dillaha, 2000], SWAT-2000<br />
[Neitsch et al., 2001], LASCAM [Viney et al.,<br />
2000] <strong>and</strong> INCA-P [Wade et al., 2002]. These<br />
models have been reviewed to identify critical<br />
requirements for the new model.<br />
In some models that divide a catchment into small<br />
sub-catchments <strong>and</strong> regard them as homogeneous<br />
units, spatial parameters (e.g. ground surface<br />
slope) are aggregated for sub-catchments that are<br />
different in size. The aggregations are therefore<br />
carried out at different scale <strong>and</strong> this can cause<br />
significant errors in simulation results. For<br />
assessing effects of l<strong>and</strong> use changes, the spatial<br />
consistency of simulation results of a model is of<br />
critical importance. In this respect, grid cell<br />
representation of a catchmant is the better option.<br />
A number of models estimate soil erosion using<br />
certain variants of the Universal Soil Loss<br />
Equation [USLE; Wischmeier <strong>and</strong> Smith, 1978] –<br />
namely, Modified USLE [Williams, 1975] <strong>and</strong><br />
Revised USLE [Renard et al., 1997]. However,<br />
these MUSLE/RUSLE-based models are not only<br />
mathematically unsound [Kinnel, 2004] but also<br />
the USLE fails to deal with soils where organic<br />
matter contents are greater than 4 % [Lilly et al.,<br />
2002]. CAMEL is being developed for application<br />
in Scotl<strong>and</strong> where soils with high organic matter<br />
content are common, which means that this is<br />
likely to be a significant issue.<br />
Some models do not take into account in-stream P<br />
transformation processes. Without P in-stream<br />
processes, however, the dynamics of reactive<br />
transport of P cannot be properly simulated. Thus<br />
it would be impossible to identify where channel<br />
reaches are acting as sources or sinks at particular<br />
times. Furthermore, in some cases, simulation of<br />
conservative (non-reactive) transport of P may<br />
result in misleading results. For a comprehensive<br />
catchment management, therefore, it is essential to<br />
simulate reactive transport of P through channel<br />
routing across the catchment.<br />
All the models listed above have a procedureoriented<br />
top-down structure leaving little<br />
autonomy to the user. The resulting lack of<br />
flexibility <strong>and</strong> extensibility may constitute a barrier<br />
for potential users.<br />
3. MODEL OVERVIEW<br />
CAMEL is a dynamic, mass balance model that<br />
employs conceptual storages <strong>and</strong> spatiallydistributed<br />
parameters. CAMEL contains a<br />
mixture of conceptual <strong>and</strong> physics-based<br />
components. Below is a list of the conceptual<br />
storages defined in each of the cells:<br />
• Four water storages – canopy, soil, aquifer <strong>and</strong><br />
channel;<br />
• Two sediment storages – overl<strong>and</strong> <strong>and</strong><br />
channel-bed;<br />
• Five P storages in the soil – active organic,<br />
stable organic, labile, active inorganic <strong>and</strong><br />
stable inorganic;<br />
• Three P storages in the aquifer <strong>and</strong> channel,<br />
respectively – labile, active inorganic <strong>and</strong><br />
stable inorganic.<br />
Unlike most existing models, CAMEL has a<br />
modular, object-oriented structure so that it allows<br />
the user to select from a library of hydrological<br />
<strong>and</strong> hydrochemical processes a sub-set of<br />
processes suitable for a particular application. In<br />
this way, the model provides a flexibility to ‘build<br />
your own’ model. The user is also allowed to<br />
determine appropriate spatial <strong>and</strong> temporal<br />
resolutions of the model.<br />
CAMEL represents a catchment using a network<br />
of square grid cells. A cell can have a maximum of<br />
8 neighbouring cells among which it can have up<br />
to 7 upstream cells <strong>and</strong> one downstream cell.<br />
Every cell represents the corresponding soilaquifer<br />
column of the catchment <strong>and</strong> has a<br />
rectangular stream channel in the middle.<br />
Input data requirements for CAMEL are in four<br />
main categories of parameters – topography, soil<br />
<strong>and</strong> aquifer, l<strong>and</strong> cover <strong>and</strong> weather:<br />
• Topography – ground surface elevation, slope,<br />
flow direction, flow accumulation, channel<br />
dimensions, channel roughness;<br />
• Soil <strong>and</strong> aquifer – soil depth; soil water<br />
contents at saturation, field capacity <strong>and</strong><br />
wilting point; median particle size,<br />
detachability <strong>and</strong> cohesion of the top soil;<br />
971
saturated hydraulic conductivity of soil;<br />
aquifer water contents at saturation <strong>and</strong> field<br />
capacity; saturated hydraulic conductivity of<br />
aquifer for fast <strong>and</strong> slow layers;<br />
• L<strong>and</strong> cover – canopy storage, overl<strong>and</strong><br />
roughness, soil cohesion increase by root<br />
reinforcement; crop height, crop coefficient,<br />
leaf area index at each of 5 crop stage dates;<br />
livestock excretion rates for cattle <strong>and</strong> sheep,<br />
incorporation rate of plant residue <strong>and</strong><br />
application rates of fertiliser <strong>and</strong> manure;<br />
• Weather – rainfall, air temperature, dew-point<br />
temperature, cloud cover, wind speed <strong>and</strong><br />
mean sea level pressure at every time-step.<br />
The current version of CAMEL provides the<br />
following outputs:<br />
• Time-series outputs for a number of selected<br />
cells at every time-step;<br />
• Snapshot outputs for the entire catchment at<br />
specific time-steps <strong>and</strong> cumulative snapshot<br />
outputs for the whole simulation period;<br />
• Mass balance outputs of water, sediment <strong>and</strong><br />
P for the entire catchment at every time-step.<br />
4. INTRA-CELL PROCESSES<br />
In CAMEL, hydrological <strong>and</strong> hydrochemical<br />
processes are calculated in two steps – intra-cell<br />
processes <strong>and</strong> inter-cell processes. Intra-cell<br />
processes include water flows, soil erosion <strong>and</strong> P<br />
transformations <strong>and</strong> are calculated for each of the<br />
cells.<br />
4.1 Water Flows<br />
CAMEL uses four conceptual storages – canopy,<br />
soil, aquifer <strong>and</strong> channel – for calculating water<br />
balance <strong>and</strong> water flows between them (Figure 1).<br />
Included in the canopy storage processes are<br />
rainfall interception, throughfall <strong>and</strong> evaporation.<br />
For the estimation of potential evaporation <strong>and</strong><br />
reference crop evapotranspiration, CAMEL uses<br />
two Penman equation derivatives suggested by<br />
Shuttleworth [1993].<br />
The soil storage processes include saturationexcess<br />
surface runoff, groundwater recharge,<br />
interflow, transpiration <strong>and</strong> soil evaporation. For<br />
estimation of groundwater recharge <strong>and</strong> interflow,<br />
a simple storage routing technique is applied to<br />
each of the 100 vertical sections of the soil column<br />
based on the relationship between soil water<br />
content <strong>and</strong> hydraulic conductivity.<br />
The aquifer storage processes are discharge to the<br />
channel, discharge to the downstream cell <strong>and</strong><br />
groundwater rise to the soil. Each aquifer is<br />
Intercep<br />
-tion<br />
RAINFALL<br />
CANOPY<br />
GW recharge<br />
downstream<br />
discharge<br />
assumed to have two layers – namely, fast <strong>and</strong><br />
slow layers – with different hydraulic<br />
conductivities to accommodate fast flows through<br />
fissure openings of weathered layers near the<br />
ground surface. Groundwater flows are assumed to<br />
be Darcian <strong>and</strong> are estimated based on differences<br />
in hydraulic heads. Thus groundwater flows in the<br />
model can be bi-directional, allowing for an<br />
estimation of channel-aquifer interactions.<br />
For channel water storage processes, channel<br />
evaporation is assumed to occur at the rate of<br />
potential evaporation.<br />
4.2 Soil Erosion<br />
SOIL<br />
AQUIFER<br />
Throughfall<br />
GW rise<br />
OUT-OF-THE-<br />
MODEL<br />
surface runoff<br />
interflow<br />
GW<br />
discharge<br />
evapotranspiration<br />
CHANNEL<br />
DOWNSTREAM CELL<br />
channel<br />
discharge<br />
Figure 1. Conceptual water storages <strong>and</strong><br />
hydrological processes in CAMEL.<br />
For simulating the effect of sediment supply on<br />
sediment transport, two conceptual sediment<br />
storages – overl<strong>and</strong> storage <strong>and</strong> channel storage –<br />
are assumed in CAMEL (Figure 2).<br />
Sediment particles detached by raindrops (splash<br />
concentrations in the rill flow. If the sediment<br />
upstream<br />
in-flux<br />
splash<br />
detachment<br />
SOIL<br />
OVERLAND<br />
CHANNEL<br />
flow<br />
detachment<br />
rill transport<br />
downstream<br />
out-flux<br />
Figure 2. Conceptual sediment storages <strong>and</strong><br />
sediment transport processes in CAMEL.<br />
972
transport capacity of the rill flow is greater than<br />
the initial sediment concentrations, more sediment<br />
particles are detached (flow detachment) <strong>and</strong><br />
transported to the channel. Otherwise, a part or all<br />
of the detached sediment is deposited <strong>and</strong> added to<br />
the overl<strong>and</strong> sediment storage. Sediment<br />
transported to the channel is added to the channel<br />
sediment storage <strong>and</strong> is transported downstream by<br />
channel flows. The equations for splash<br />
detachment <strong>and</strong> flow detachment have been taken<br />
from EUROSEM [Smith et al., 1995], a physicsbased<br />
soil erosion model.<br />
4.3 P Transformations <strong>and</strong> Transport<br />
The structure of the soil P transformation<br />
component of the model has been widely taken<br />
from the EPIC model [Jones et al., 1984] <strong>and</strong> the<br />
SWAT model [Neitsch et al., 2001] <strong>and</strong> then<br />
further simplifications have been made.<br />
For simulating P transformation <strong>and</strong> transport<br />
processes in the soil, aquifer <strong>and</strong> channel, CAMEL<br />
assumes conceptual storages for organic <strong>and</strong><br />
inorganic P (Figure 3). Organic P in the soil is<br />
divided into two storages: the active organic P<br />
storage (P AO ) <strong>and</strong> the stable organic P storage<br />
(P SO ). P AO consists of P in undecomposed plant<br />
residues, livestock excretion, manure <strong>and</strong><br />
microbes, whereas P SO is composed of P in stable<br />
organic matter i.e. humus. Soil inorganic P is<br />
divided into labile P (P LB ), active inorganic P (P AI )<br />
<strong>and</strong> stable inorganic P (P SI ) storages. P LB is in<br />
rapid equilibrium (several days or weeks) with P AI<br />
which in return is in slow equilibrium with P SI .<br />
When inorganic fertiliser P is added to P AI , it<br />
rapidly equilibrates between P LB <strong>and</strong> P AI . The slow<br />
reaction between P AI <strong>and</strong> P SI then follows. It is<br />
assumed P SI is four times larger than P AI . In the<br />
aquifer <strong>and</strong> the channel, only inorganic P storages<br />
(P LB , P AI <strong>and</strong> P SI ) are assumed <strong>and</strong>, therefore, P<br />
sorption is the only process simulated in the<br />
model.<br />
All P transformation rates are calculated using<br />
first-order kinetic equations taking into account the<br />
effect of soil water content <strong>and</strong> temperature. The<br />
soil water content effect on organic matter<br />
decomposition, mineralisation <strong>and</strong> immobilisation<br />
is estimated using a segmented linear function.<br />
Soil temperature is calculated using the approach<br />
of Kang et al. [2000] <strong>and</strong> its effect on<br />
transformation rates is estimated using the Q10<br />
function [Van Clooster et al., 1994]. Plant uptake<br />
of P is assumed to be proportional to transpiration<br />
rate <strong>and</strong> labile P concentrations in the soil water.<br />
Intra-cell P transport processes in the model<br />
include transport of sediment-bound P (P AI <strong>and</strong><br />
SOIL<br />
PLANT<br />
UPTAKE<br />
INORGANIC<br />
FERTILISER<br />
P SI<br />
CHANNEL<br />
P SI<br />
slow<br />
sorption<br />
slow<br />
sorption<br />
Figure 3. Intra-cell P transformation <strong>and</strong><br />
transport processes between conceptual P<br />
storages in CAMEL (P AO = active organic P;<br />
P SO = stable organic P; P LB = labile P; P AI =<br />
active inorganic P; P SI = stable inorganic P).<br />
P SI ) by surface runoff <strong>and</strong> transport of dissolved P<br />
(P LB ) by surface runoff, groundwater recharge <strong>and</strong><br />
groundwater discharge. Transport of sedimentbound<br />
P is estimated using an enrichment ratio that<br />
exponentially decreases with the sediment flux.<br />
Dissolved P transport is estimated using an<br />
extraction ratio that is an exponential function of<br />
water flows.<br />
5. INTER-CELL PROCESSES<br />
Inter-cell processes are calculated after the intracell<br />
processes are evaluated for the entire<br />
catchment. To allow for reactive transport<br />
processes of P between cells, two cascade routing<br />
schemes are used for groundwater (dissolved P)<br />
<strong>and</strong> channel water (particulate <strong>and</strong> dissolved P)<br />
flows, respectively.<br />
5.1 Channel Water Routing<br />
PLANT RESIDUE /<br />
MANURE / EXCRETION<br />
P AO<br />
P AI<br />
AQUIFER<br />
P SI<br />
P AI<br />
decomposition<br />
mineralisation/<br />
immobilisation<br />
rapid<br />
sorption<br />
rapid<br />
sorption<br />
slow<br />
sorption<br />
rapid<br />
sorption<br />
P LB<br />
P AI<br />
P LB<br />
P SO<br />
P LB<br />
973
For routing of the channel water, the spatially<br />
distributed unit hydrograph approach proposed by<br />
Maidment [1993] is adopted in CAMEL with<br />
modifications. The flow travel time from a cell to a<br />
given downstream cell is estimated by assuming<br />
constant flow velocities. The constant flow<br />
velocities for individual cells are estimated using<br />
Manning’s equation on an assumption that channel<br />
water depth is 1/10 of the channel width. The<br />
volume of channel water leaving each of the cells<br />
is then routed to the given downstream cell in a<br />
certain time-step according to the isochrone of<br />
flow travel time to the cell.<br />
5.2 Groundwater <strong>and</strong> P Routing<br />
A Darcian groundwater flow from an upstream cell<br />
is transported to the downstream cell completing<br />
the process in two time-steps. The downstream<br />
out-flow from a cell in the current time-step<br />
contributes to the upstream in-flow of the<br />
downstream cell in the next time-step. Due to this<br />
separation of upstream in-flows <strong>and</strong> downstream<br />
out-flows in time, groundwater is routed<br />
downstream in a fully cascading way, which<br />
allows for P sorption processes in the aquifer of<br />
indi vidual cells. It should be noted, however, that<br />
this routing scheme is valid only when the<br />
groundwater flow velocity does not exceed the cell<br />
length per time-step.<br />
5.3 Channel Sediment <strong>and</strong> P Routing<br />
For reactive transport of sediment <strong>and</strong> P in the<br />
channel, a comprehensive cascade routing scheme<br />
has been developed. For calculating the channel<br />
sediment budget of a given cell, primary cells that<br />
have no upstream cells are first identified. Then<br />
the amount of sediment <strong>and</strong> P leaving the primary<br />
cells are estimated <strong>and</strong> routed downstream cell-bycell<br />
to the given cell taking into account the<br />
isochrones. Sediment transport processes in the<br />
channel (i.e. detachment <strong>and</strong> deposition), P<br />
sorption <strong>and</strong> transport (in both dissolved <strong>and</strong><br />
particulate forms) processes are evaluated using<br />
the same equations applied to rill flows within a<br />
cell.<br />
6. MODEL VERIFICATION<br />
A verification study of the model has been carried<br />
out for a small hypothetical catchment (0.8 km 2 )<br />
with 200 m grid cells for one year period at daily<br />
time-steps. For this study, a set of daily weather<br />
data from a UK meteorological station was used<br />
<strong>and</strong> the catchment was assumed to have a<br />
homogeneous l<strong>and</strong> cover (winter wheat) <strong>and</strong> a<br />
soil/geology layer (s<strong>and</strong>y silt loam underlain by<br />
well-fissured granite). To avoid unnecessary<br />
confusion, it should be noted that no comparison<br />
with field data has been carried out in this<br />
verification study.<br />
Parameter values were initially taken from various<br />
sources <strong>and</strong> adjusted during verification<br />
simulations to obtain reasonable results. Table 1<br />
lists some of the parameter values used for the<br />
final verification simulation.<br />
The hydrological simulation results including<br />
evapotranspiration, soil water content, discharge<br />
<strong>and</strong> groundwater table elevation show strong<br />
seasonal variations as expected. The overl<strong>and</strong><br />
sediment is delivered to the channel mostly by the<br />
first few storms in autumn when the soil is fully<br />
saturated. The comprehensive cascade routing<br />
scheme for channel sediment <strong>and</strong> P works well<br />
with very little mass balance errors.<br />
Simulation results of P transformation processes in<br />
the soil, demonstrated in Figure 4, show strong<br />
temporal variations reflecting the effect of<br />
agricultural practices such as harvest <strong>and</strong><br />
applications of fertiliser <strong>and</strong> manure. For example,<br />
when mineral P fertiliser is applied in spring, P is<br />
rapidly adsorbed to the soil <strong>and</strong> then, as plant<br />
uptake increases in the growing season, P<br />
adsorption rate gradually decreases leading to P<br />
desorption in summer.<br />
The model also reasonably represents P transport<br />
processes. In the model, P is transported to the<br />
channel in both particulate <strong>and</strong> dissolved forms.<br />
However, simulation results show that most of P is<br />
transported in particulate forms (Figure 5), which<br />
Table 1. Selected parameter values used for the<br />
final simulation<br />
Parameter Unit Value<br />
Saturated hydraulic<br />
conductivity for soil m/day 0.56<br />
Saturated hydraulic<br />
conductivity for aquifer:<br />
- fast-flowing layer m/day 2.50<br />
- slow-flowing layer m/day 0.25<br />
Organic matter<br />
decomposition rate 1/day 0.10<br />
Humus decomposition<br />
rate 1/day 0.03<br />
Rapid adsorption rate 1/day 0.50<br />
Slow adsorption rate 1/day 0.01<br />
Fertiliser application rate KgP/ha/y 35.0<br />
Manure application rate KgP/ha/y 5.0<br />
Plant residue<br />
incorporation rate KgP/ha/y 5.0<br />
974
kgP/ha/d<br />
kgP/ha/d<br />
kgP/ha/d<br />
kgP/ha/d<br />
reflects the characteristics of P being adsorbed to<br />
sediment particles. The amount of dissolved P<br />
transported to the catchment outlet is negligible<br />
(0.02 kgP/ha/y) compared to that of particulate P<br />
(2.14 kgP/ha/y).<br />
In calculating dissolved P concentrations in the<br />
channel water, CAMEL uses the size of labile P<br />
storage in the channel to estimate the amount of<br />
dissolved P in water. The assumption here is that<br />
water is fully interacting with the labile P storage.<br />
This is reasonable when enough water flows in the<br />
channel, but this becomes invalid when very little<br />
water flows in the channel. In reality, during low<br />
flow, water occupies a fraction of the channel bed<br />
<strong>and</strong> the interaction between water <strong>and</strong> the labile P<br />
storage is limited. In the model, the channel water<br />
evenly distributes across the channel bed <strong>and</strong> thus<br />
a full interaction is assumed even in very low flow<br />
conditions. This limitation can cause<br />
unrealistically high concentrations of dissolved P<br />
at very low flow conditions as shown in Figure 6.<br />
kgP/ha/d<br />
40<br />
0<br />
0.2<br />
0.0<br />
1.5<br />
1.0<br />
0.5<br />
0.0<br />
-0.5<br />
0.4<br />
0.0<br />
1.0<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0.0<br />
P Mineralisation<br />
Pfertil<br />
Presidu<br />
Pmanure<br />
Figure 4. Simulation results of P transformation<br />
processes in the soil.<br />
PP from Soil to Channel<br />
PP through Channel<br />
P Inputs<br />
P Adsorption<br />
P Uptake<br />
J F M A M J J A S O N D<br />
J F M A M J J A S O N D<br />
Figure 5. Catchment-mean transport rates of<br />
particulate P (PP)<br />
mgP/l<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0.0<br />
J F M A M J J A S O N D<br />
Figure 6. Simulated concentrations of P at the<br />
catchment outlet (PP = particulate P; DP =<br />
dissolved P)<br />
It is anticipated that this problem will be resolved<br />
in the next version of CAMEL.<br />
Despite some limitations, the model simulation<br />
results are generally reasonable <strong>and</strong> the mass<br />
balance errors of water (-4.34E-12 mm/y),<br />
sediment (9.23E-09 kg/ha/y) <strong>and</strong> P (3.26E-13<br />
kgP/ha/y) are negligible. It is therefore considered<br />
that CAMEL has been correctly coded to represent<br />
the conceptual model.<br />
7. CONCLUSIONS<br />
A spatially-distributed conceptual model,<br />
CAMEL, has been developed for simulating<br />
reactive transport of P from diffuse sources at the<br />
catchment scale. Although based on conceptual<br />
storages, CAMEL evaluates the majority of<br />
processes using physics-based equations. The<br />
model has comprehensive cascade routing schemes<br />
that allow for reactive transport of P across the<br />
catchment. Because of its modular <strong>and</strong> objectoriented<br />
structure, CAMEL can be easily modified<br />
or extended. Furthermore, the model provides a<br />
library of hydrological <strong>and</strong> hydro-chemical<br />
processes from which the user can select a sub-set<br />
of processes suitable for a particular application. In<br />
this way, the model provides the user a flexibility<br />
to ‘build your own’ model. A verification study on<br />
a hypothetical catchment has shown that CAMEL<br />
has been correctly coded to represent the<br />
conceptual model.<br />
With a network of self-contained cells <strong>and</strong><br />
comprehensive routing schemes, CAMEL can<br />
identify the critical source areas in a catchment <strong>and</strong><br />
the major transport processes of P from those<br />
areas. This information may then be used for<br />
improving the efficiency <strong>and</strong>/or effectiveness of<br />
catchment management practices.<br />
8. REFERENCE<br />
PP<br />
DP<br />
975
Bouraoui, F. <strong>and</strong> T.A. Dillaha, ANSWERS-2000:<br />
Nonpoint Source Nutrient Transport Model,<br />
J. Env’tal. Eng.,126 (11): 1045-1055, 2000.<br />
Cronshey, R.G. <strong>and</strong> F.D. Theurer, AnnAGNPS -<br />
Non-point pollutant loading model,<br />
Proceedings of the 1st Federal Interagency<br />
Hydrologic Modeling Conference, Las<br />
Vegas, Nevada, April 19-23, 1998, p. 1-9 to<br />
1-16, 1998.<br />
Gburek, W. J. <strong>and</strong> A. N. Sharpley, Hydrologic<br />
controls on phosphorus loss from upl<strong>and</strong><br />
agricultural watersheds, J. Env’tal. Qual., 27:<br />
267-277, 1998.<br />
House, W. A. <strong>and</strong> L. A. Donaldson, Adsorption<br />
<strong>and</strong> coprecipitation of phosphate on calcite, J.<br />
Colloid Interface Sci., 112: 309-324, 1986.<br />
House, W. A., F. A. Denison <strong>and</strong> P. D. Armitage,<br />
Comparison of the uptake of inorganic<br />
phosphorus to a suspended <strong>and</strong> stream bedsediment,<br />
Water Research, 29: 767-779,<br />
1995.<br />
Jarvie, H. P., C. Neal, A. Warwick, J. White, M.<br />
Neal, H. D. Wickham, L. K. Hill <strong>and</strong> M. C.<br />
Andrews, Phosphorus uptake into algal<br />
biofilms in a lowl<strong>and</strong> chalk river, The Sci. of<br />
the Total Env., 282-283: 353-373, 2002.<br />
Jones, C.A., C.V. Cole, A.N. Sharpley <strong>and</strong> J.R.<br />
Williams, A simplified soil <strong>and</strong> plant<br />
phosphorus model, I. Documentation, Soil<br />
Sci. Soc. Am. J., 48(4): 800-805, 1984.<br />
Kang, S, S. Kim, S. Oh <strong>and</strong> D. Lee, Predicting<br />
spatial <strong>and</strong> temporal patterns of soil<br />
temperature based on topography, surface<br />
cover <strong>and</strong> air temperature. Forest Ecology<br />
<strong>and</strong> Management, 136: 173-184, 2000.<br />
Kinnel, P. I. A., The mathematical integrity of<br />
some universal soil loss equation variants,<br />
Soil Sci. Soc. Am. J., 68(1): 336-337, 2004.<br />
Lilly, A., G. Hudson, R.V. Birnie <strong>and</strong> P. L. Horne,<br />
The inherent geomorphological risk of soil<br />
erosion by overl<strong>and</strong> flow in Scotl<strong>and</strong>,<br />
Scottish Natural Heritage Research, Survey<br />
<strong>and</strong> Monitoring Report No. 183, 2002.<br />
Maidment, D. R., Developing a spatially<br />
distributed unit hydrograph by using GIS,<br />
HydroGIS 93: Application of GIS in<br />
Hydrology <strong>and</strong> Water Resources, IAHS Publ.<br />
No. 211, pp. 181-192, 1993.<br />
Needelman, B. A., W. J. Gburek, A. N. Sharpley<br />
<strong>and</strong> G.W. Petersen, <strong>Environmental</strong><br />
management of soil phosphorus: Modeling<br />
spatial variability in small fields, Soil Sci.<br />
Soc. Am. J., 65: 1516-1522, 2001.<br />
Neitsch, S. L., J. G. Arnold, J. R. Kiniry <strong>and</strong> J. R.<br />
Williams, Soil <strong>and</strong> water assessment tool –<br />
Theoretical documentation, Grassl<strong>and</strong>, Soil<br />
<strong>and</strong> Water Research Laboratory, Agricultural<br />
Research Service, Internet web site<br />
http://www.brc.tamus.edu/swat, 2001.<br />
Renard, K. G., G. R. Foster, G. A. Weesies, D. K.<br />
McCool <strong>and</strong> D. C. Yoder, Predicting soil<br />
erosion by water: A guide to conservation<br />
planning with the Revised Universal Soil<br />
Loss Equation (RUSLE), Agric. H<strong>and</strong>book<br />
703, US Gov. Print. Office, Washington DC.,<br />
1997.<br />
Sample, E. C., R. J. Soper <strong>and</strong> G. J. Racz,<br />
Reactions of phosphate fertilizers in soils, In:<br />
F. E. Khasawneh, E. C. Sample <strong>and</strong> E. J.<br />
Kamprath (eds.), The Role of Phosphorus in<br />
Agriculture, Am. Soc. of Agronomy, Crop<br />
Sci. Soc. of Am., Soil Sci. Soc. of Am.,<br />
Madison, Wisconsin, US, pp. 263-310, 1980.<br />
Shuttleworth, W. J., Evaporation. In: D. Maidment<br />
(ed.), H<strong>and</strong>book of Hydrology, McGraw-Hill,<br />
1993.<br />
Smith, R, D. Goodrich <strong>and</strong> J. Quinton, Dynamic,<br />
distributed simulation of watershed erosion:<br />
The KINEROS2 <strong>and</strong> EUROSEM models, J.<br />
Soil Water Conservation 50:517-520, 1995.<br />
Van Clooster, M, P. Viaene, J. Diels <strong>and</strong> K.<br />
Christiaens, WAVE: a mathematical model<br />
for simulating water <strong>and</strong> agrochemicals in<br />
the soil <strong>and</strong> vadose environment, Reference<br />
<strong>and</strong> Users Manual. Institute for L<strong>and</strong> <strong>and</strong><br />
Water Management, Katholieke Universiteit<br />
Leuven, Leuven, Belgium, 1994.<br />
Viney, N. R., M. Sivapalan <strong>and</strong> D. Deeley, A<br />
conceptual model of nutrient mobilisation<br />
<strong>and</strong> transport applicable at large catchment<br />
scales, J. of Hydrol., 240: 23-44, 2000.<br />
Wade, A. J., P. G. Whitehead <strong>and</strong> D. Butterfield,<br />
The integrated catchments model of<br />
phosphorus dynamics (INCA-P), a new<br />
approach for multiple source assessment in<br />
heterogeneous river systems: model structure<br />
<strong>and</strong> equations, Hydrol. Earth Sys. Sci., 6(3):<br />
583-606, 2002.<br />
Williams, J. R., Sediment-yield prediction with<br />
universal equation using runoff energy factor,<br />
In: Present <strong>and</strong> prospective technology for<br />
predicting sediment yield <strong>and</strong> sources, ARS.<br />
S-40, US Gov. Print. Office, Washington<br />
DC., 244-252, 1975.<br />
Wischmeier, W. C. <strong>and</strong> D. D. Smith, Predicting<br />
rainfall erosion losses – A guide to<br />
conservation planning, Agric. H<strong>and</strong>book 537,<br />
US Gov. Print. Office, Washington DC,<br />
1978.<br />
Withers, P. J. A. <strong>and</strong> E. I. Lord, Agricultural<br />
nutrient inputs to rivers <strong>and</strong> groundwaters in<br />
the UK: policy, environmental management<br />
<strong>and</strong> research needs, The Sci. of the Total.<br />
Env., 282-283: 9-24, 2002.<br />
Woodruff, S. L., W. A. House, M. E. Callow <strong>and</strong><br />
B. S. C. Leadbeater, The effects of biofilms<br />
on chemical processes in surfacial sediments,<br />
Freshwater Biol., 41: 73-89, 1999.<br />
976
A probabilistic modelling concept for the quantification of<br />
flood risks <strong>and</strong> associated uncertainties<br />
Heiko Apel a , Annegret H. Thieken a , Bruno Merz a , Günter Blöschl b<br />
a GeoForschungsZentrum Potsdam (GFZ), Section 5.4 Engineering Hydrology, Telegrafenberg, 14473 Potsdam,<br />
Germany (hapel@gfz-potsdam.de)<br />
b Institute of Hydraulics, Hydrology <strong>and</strong> Water Resources Management, Vienna University of Technology,<br />
Karlsplatz 13, A-1040 Wien, Austria<br />
Abstract: Flood disaster mitigation strategies should be based on a comprehensive assessment of the flood risk<br />
combined with a thorough investigation of the uncertainties associated with the risk assessment procedure.<br />
Within the ‘German Research Network of Natural Disasters’ (DFNK) the working group ‘Flood Risk Analysis’<br />
investigated the flood process chain from precipitation, runoff generation <strong>and</strong> concentration in the catchment,<br />
flood routing in the river network, possible failure of flood protection measures, inundation to economic damage.<br />
The working group represented each of these processes by deterministic, spatially distributed models at different<br />
scales. While these models provide the necessary underst<strong>and</strong>ing of the flood process chain, they are not suitable<br />
for risk <strong>and</strong> uncertainty analyses due to their complex nature <strong>and</strong> high CPU-time dem<strong>and</strong>. We have therefore<br />
developed a stochastic flood risk model consisting of simplified model components associated with the<br />
components of the process chain. We parameterised these model components based on the results of the complex<br />
deterministic models <strong>and</strong> used them for the risk <strong>and</strong> uncertainty analysis in a Monte Carlo framework. The<br />
Monte Carlo framework is hierarchically structured in two layers representing two different sources of<br />
uncertainty, aleatory uncertainty (due to natural <strong>and</strong> anthropogenic variability) <strong>and</strong> epistemic uncertainty (due to<br />
incomplete knowledge of the system). The model allows us to calculate probabilities of occurrence for events of<br />
different magnitudes along with the expected economic damage in a target area in the first layer of the Monte<br />
Carlo framework, i.e. to assess the economic risks, <strong>and</strong> to derive uncertainty bounds associated with these risks<br />
in the second layer. It could be shown that the uncertainty caused by epistemic sources significantly alters the<br />
results obtained with aleatory uncertainty alone. The model was applied to reaches of the river Rhine<br />
downstream of Cologne.<br />
Keywords: flood risk assessment, uncertainty estimation, probabilistic model<br />
1. INTRODUCTION<br />
Flood defence systems are usually designed by<br />
specifying an exceedance probability <strong>and</strong> by<br />
demonstrating that the flood defence system<br />
prevents damage from events corresponding to this<br />
exceedance probability. This concept is limited by a<br />
number of assumptions <strong>and</strong> many researchers have<br />
called for more comprehensive design procedures<br />
(Plate, 1992; Bowles et al., 1996; Berga, 1998;<br />
Vrijling, 2001). The most complete approach is the<br />
risk-based design approach which balances benefits<br />
<strong>and</strong> costs of the design in an explicit manner<br />
(Stewart <strong>and</strong> Melchers, 1997). In the context of<br />
risk-based design, the flood risk consists of the<br />
flood hazard (i.e. extreme events <strong>and</strong> associated<br />
probability) <strong>and</strong> the consequences of flooding (i.e.<br />
property damages). Ideally, a flood risk analysis<br />
should take into account all relevant flooding<br />
scenarios, their associated probabilities <strong>and</strong> possible<br />
damages as well as a thorough investigation of the<br />
uncertainties associated with the risk analysis.<br />
Thus, a flood risk analysis should finally yield a<br />
risk curve, i.e. the full distribution function of the<br />
flood damages in the area under consideration,<br />
ideally accompanied by uncertainty bounds.<br />
Following these concepts the working group ‘Flood<br />
Risk Analysis’ of the German Research Network on<br />
Natural disasters (DFNK) investigated the complete<br />
flood disaster chain from the triggering event to its<br />
consequences: ‘hydrological load – flood routing –<br />
potential failure of flood protection structures –<br />
inundation – property damage’. Complementary to<br />
applied determistic models a simple stochastic<br />
model consisting of modules each representing one<br />
process of the flood disaster chain was developed.<br />
The advantages for flood risk assessment of the<br />
simple approach are mainly: First, significantly less<br />
CPU time is needed which allows application of the<br />
approach in Monte Carlo simulations. Second, the<br />
simpler model structure makes it easier for the<br />
977
analyst to underst<strong>and</strong> the main controls of the<br />
systems.<br />
The simple stochastic model represents two<br />
fundamentally different types of uncertainty,<br />
aleatory <strong>and</strong> epistemic uncertainty. Aleatory<br />
uncertainty refers to quantities that are inherently<br />
variable over time, space, or populations of<br />
individuals or objects. According to Hall (2003) it<br />
can be operationally defined as being a feature of<br />
populations of measurements that conform well to a<br />
probabilistic model. Epistemic uncertainty results<br />
from incomplete knowledge of the object of<br />
investigation <strong>and</strong> is related to our ability to<br />
underst<strong>and</strong>, measure, <strong>and</strong> describe the system under<br />
study.<br />
The simple stochastic model allows the risk <strong>and</strong><br />
uncertainty analysis through a Monte-Carloframework.<br />
In line with the distinction of aleatory<br />
<strong>and</strong> epistemic uncertainties, the Monte-Carloframework<br />
was hierarchically structured, with each<br />
of the two layers representing one of the two types<br />
of uncertainties (two-dimensional or second-order<br />
Monte-Carlo–simulation, Cullen <strong>and</strong> Frey, 1999).<br />
The first layer represents aleatory uncertainty <strong>and</strong><br />
assumes that the variability of the system is<br />
perfectly known <strong>and</strong> correctly quantified, e.g. by<br />
known parameter distributions. The result of this<br />
first layer of Monte Carlo simulation is a risk curve<br />
for the target area. The second layer of Monte Carlo<br />
simulations represents the uncertainty caused by<br />
our incomplete knowledge of the system. This<br />
distinction into the two uncertainty classes has<br />
important implication for the results of the risk<br />
assessment. The uncertainty bounds derived by this<br />
method cannot be interpreted as steady-state <strong>and</strong><br />
may narrow down as more knowledge about the<br />
processes <strong>and</strong> parameters under of the model is<br />
obtained (Ferson <strong>and</strong> Ginzberg, 1996).<br />
In this paper, the feasibility of this modelling<br />
approach combined with the hierarchical<br />
uncertainty analysis is illustrated for a reach of the<br />
river Rhine in Germany.<br />
1.1 Investigation area<br />
The investigation area of this study was the reach of<br />
the Rhine between Cologne <strong>and</strong> Rees with a focus<br />
on the polder at Mehrum. For this polder the actual<br />
risk assessment was performed. The polder at<br />
Mehrum is a confined rural area of 12.5 km², which<br />
is only inundated if the protecting levee system<br />
fails.<br />
Two levee breach locations were exemplarily<br />
selected along the reach for the simulation. They<br />
differ significantly in their storing capacity. At<br />
Krefeld the large unbounded hinterl<strong>and</strong> provides a<br />
retention basin with a practically infinite retention<br />
capacity whereas the polder at Mehrum is strictly<br />
confined to a comparatively small volume. The<br />
levees at the two breach locations are similar in<br />
structure, but at Mehrum the levee crest is higher,<br />
i.e. larger flood waves are required to overtop the<br />
levee at Mehrum as compared to Krefeld (Figure<br />
1).<br />
Through the selection of a longer reach of the main<br />
river along with the main tributaries the risk<br />
assessment implicitly considers the hydrological<br />
behaviour of a complete watershed. Additionally<br />
the selection of the two breach locations with their<br />
different hinterl<strong>and</strong>s enables a risk assessment<br />
under consideration of possible levee breaches <strong>and</strong><br />
their impact on flood wave propagation.<br />
Rhine<br />
Lippe<br />
Ruhr<br />
Figure 1: Sketch of the investigation area with the<br />
main tributaries Ruhr <strong>and</strong> Lippe <strong>and</strong> the selected<br />
breach locations (BL) Krefeld <strong>and</strong> Mehrum<br />
2. MODULES<br />
The risk analysis for the flood disaster chain is<br />
based on the following modules: Hydrological load,<br />
flood routing, levee failure <strong>and</strong> outflow through<br />
levee breach <strong>and</strong> finally the damage estimation. The<br />
following sections describe the modules briefly,<br />
followed by a description of the Monte-Carloframework<br />
in section 3. More details are given in<br />
Apel et al. (2004a) <strong>and</strong> Apel et al. (2004b).<br />
2.1 Hydrological load<br />
The hydrological load was derived from the flood<br />
frequency curve of the gauge Cologne/Rhine based<br />
on the annual maximum series from 1961 to 1995<br />
(AMS 1961-1995). Four distribution functions were<br />
fitted to the AMS 1961-1995: Gumbel, Lognormal,<br />
Weibull <strong>and</strong> the Pearson-III distribution. The four<br />
distribution functions were weighted by a<br />
Maximum Likelihood method to construct a<br />
composite probability distribution function (Wood<br />
<strong>and</strong> Rodríguez-Iturbe, 1975). Figure 2 shows the<br />
four individual distributions along with the<br />
composite distribution.<br />
In order to determine the occurrence of levee<br />
breaches <strong>and</strong> inundation levels of the polders it was<br />
978
necessary to generate flood hydrographs in addition<br />
to the maximum discharge. Hence typical flood<br />
hydrographs (Apel et al., 2004b) were generated for<br />
the gauge Cologne based on non-dimensional<br />
hydrographs in combination with cluster analysis.<br />
Q [m 3 /s]<br />
104 1.8<br />
1.6<br />
1.4<br />
1.2<br />
1<br />
0.8<br />
0.6<br />
Plotting Positions<br />
0.4<br />
Gumbel<br />
LogNorm<br />
0.2<br />
Weibull<br />
Pearson3<br />
composite<br />
0<br />
10 0 10 1 10 2<br />
2 x T [a]<br />
Figure 2: Different distribution functions fitted to<br />
the annual maximum flood series 1961-1995 of the<br />
gauge Cologne/Rhine.<br />
The results of the cluster analysis are seven types of<br />
typical, realistic hydrographs: single peaked<br />
hydrographs <strong>and</strong> various multiple peaked<br />
hydrographs. A similar procedure was applied to<br />
the main tributaries Ruhr <strong>and</strong> Lippe, using the<br />
corresponding flood hydrographs for the chosen<br />
events to the main river.<br />
2.2 Flood routing<br />
The second module of the flood disaster chain is a<br />
routing module consisting of the Muskingum<br />
routing method for flood waves in river channels<br />
(Maidment, 1992). The required parameters were<br />
estimated for the defined river reaches from the 35<br />
flood events of the years 1961-1995.<br />
2.3 Levee failure<br />
In this case study we defined two levee breach<br />
locations <strong>and</strong> derived probabilities of breaches for<br />
these two points. For the calculation of the pointfailure<br />
probability of a levee, a general engineering<br />
technique was applied in which a breach condition<br />
is defined as the exceedance of a load factor over a<br />
resistance factor. This concept was applied to levee<br />
failures caused by overtopping of the levee crest<br />
which is the most common failure mechanism of<br />
modern zonated levees. The breach criterion was<br />
defined as the difference between the actual<br />
overflow q a [m 3 /s] (the load factor) <strong>and</strong> the critical<br />
overflow q crit [m 3 /s] (the resistance factor). For the<br />
calculation of q a <strong>and</strong> q crit the approaches of<br />
Kortenhaus & Oumeraci (2002) <strong>and</strong> Vrijling (2000)<br />
were used, respectively. These are based on<br />
overtopping height <strong>and</strong> overflowing time as<br />
independent variables <strong>and</strong> on the geometry of the<br />
levees. The only non-geometric parameter used in<br />
this formulae is the turf-quality parameter fg<br />
(Vrijling 2002), which is of subjective nature <strong>and</strong><br />
hence was given particular attention in the<br />
uncertainty calculations (cf. section 3).<br />
From this intermediate complex deterministic<br />
model a probabilistic model representing the<br />
conditional failure probability depending on<br />
overtopping height <strong>and</strong> time was derived<br />
analogously to USACE (1999). The outflow<br />
through a levee breach is calculated from an<br />
empirical outflow formula presented in Disse et al.<br />
(2004).<br />
2.4 Damage estimation<br />
The last module estimates direct monetary losses<br />
within the polder at Mehrum. Since the size <strong>and</strong><br />
location of the inundated areas are not estimated<br />
directly by the simple model presented here, a<br />
damage function that relates the damage in the<br />
inundated areas of the polder at Mehrum to the<br />
inflow of water volume after/during a levee failure<br />
had to be determined. This was done by assuming<br />
the filling of the polder in 0.5 m steps up to the<br />
levee crest <strong>and</strong> intersecting each inundation layer<br />
with the l<strong>and</strong> use map. The damage of the<br />
inundated l<strong>and</strong> use types was estimated by<br />
combining assessed replacement values <strong>and</strong> stagedamage<br />
curves.<br />
For all sectors, with the exception of private<br />
housing, unit economic values were determined<br />
from the economic statistics of North Rhine-<br />
Westphalia from 1997 (data of the gross stock of<br />
fixed assets according to the system of national<br />
accounts from 1958 <strong>and</strong> l<strong>and</strong> use information from<br />
the statistical regional authorities in North Rhine-<br />
Westphalia). The replacement values were scaled to<br />
the year 2000 by data on the development of gross<br />
stock of fixed assets in North Rhine-Westphalia <strong>and</strong><br />
adjusted to Mehrum by comparing the gross value<br />
added per employee in that region with that of<br />
entire North Rhine-Westphalia. Damages were<br />
determined using the step-damage-function of<br />
MURL (2000).<br />
3. RISK AND UNCERTAINTY<br />
CALCULATIONS<br />
For the risk <strong>and</strong> uncertainty analysis a hierarchical<br />
Monte Carlo framework was developed. In the first<br />
level of the analysis the Monte Carlo simulations<br />
represent the variability of the system, i.e. the<br />
aleatory uncertainty. This results in frequency<br />
distributions of floods at the outlet of the<br />
investigation area <strong>and</strong> risk curves for the target<br />
area, the polder at Mehrum. We r<strong>and</strong>omised the<br />
following variables in the first level 10 5 times:<br />
979
- the annual maximum discharge of the Rhine<br />
- the correlation of the maximum discharge of<br />
the Rhine with the tributaries Ruhr <strong>and</strong> Lippe<br />
The second level of Monte Carlo simulations<br />
represents the uncertainty associated with the<br />
results of the first level. In this level, uncertainty<br />
distributions of the flood frequency distributions<br />
<strong>and</strong> risk curves were calculated <strong>and</strong> used to<br />
construct the confidence bounds. The uncertainty<br />
sources covered in this analysis were the selection<br />
of the extreme value statistics functions <strong>and</strong> the<br />
parameter estimation of the stage-discharge<br />
relationship at the levee beach locations.<br />
However, it was not possible to include all<br />
uncertainty sources as for some of them only<br />
insufficient information was available. These<br />
uncertainty sources include the width of a levee<br />
breach after a levee failure <strong>and</strong> the turf quality<br />
parameter involved in the calculation of the<br />
probability of failure. In these two cases statistics<br />
such as mean values, coefficients of variation <strong>and</strong><br />
distribution types were not available. Because of<br />
this, the width of the breach <strong>and</strong> the turf quality<br />
parameter were not incorporated in the MCframework<br />
but examined in scenario calculations.<br />
The values for the breach width in the scenarios<br />
were set to 100, 200, 300 <strong>and</strong> 400 meters according<br />
to expert knowledge of the local flood defence<br />
authorities <strong>and</strong> historical records. Additionally a<br />
zero breach scenario for the location Krefeld was<br />
calculated in order to assess the effect of upstream<br />
levee breaches on the risk in the investigation<br />
entirely. The turf quality scenarios were set<br />
according to the minimum, maximum <strong>and</strong> mean of<br />
the range of value given in Vrijling (2000). The<br />
scenarios apply to both levels of MC-simulations.<br />
4. RESULTS<br />
4.1 Risk analysis<br />
Without any upstream breaches (K0), the levee at<br />
Mehrum failed up to 99 times (failure rate 0.99 ‰)<br />
in the Monte Carlo simulations. When breaches at<br />
Krefeld were allowed, this figure was significantly<br />
reduced to only one failure of the levee at Mehrum<br />
in the case of a breach width of 400 m at Krefeld<br />
irrespective of the value of the turf parameter fg. In<br />
addition to the breach width at Krefeld, the turf<br />
quality has an important effect on the number of<br />
breaches, if the breach width is in the range of 100-<br />
200 m: The lower is the turf quality, the higher is<br />
the number of breaches at both locations.<br />
The flood frequency curve at Rees, the most<br />
downstream gauging station of the reach examined<br />
here, is also influenced by the number of upstream<br />
levee breaches <strong>and</strong> the breach width at Krefeld.<br />
Figure 3 shows the flood frequency curves at Rees<br />
derived from the output of the routing module for a<br />
fixed turf quality <strong>and</strong> varying breach widths at<br />
Krefeld. Overall, the exceedance probabilities of<br />
extreme events are reduced by upstream levee<br />
breaches while the exceedance probabilities of<br />
discharges at the critical levels are increased. This<br />
effect is caused by the reduction of the peak flows<br />
of a number of floods that overtop the levee to<br />
discharges below the critical overflowing discharge.<br />
The effect is more pronounced the wider the breach<br />
at Krefeld is assumed.<br />
Q [m 3 /s]<br />
2<br />
1.8<br />
1.6<br />
1.4<br />
1.2<br />
1<br />
0.8<br />
0.6<br />
0.4<br />
K0<br />
K100<br />
K200<br />
0.2<br />
K300<br />
K400<br />
0<br />
10 0 10 1 10 2 10 3 10 4<br />
2.2 x 104 T [a]<br />
Figure 3: Frequency curves at the outlet of the<br />
investigation area (Rees at the Rhine): scenarios of<br />
different breach widths, fg = 1.05<br />
Damage [Mio. Euro]<br />
120<br />
100<br />
80<br />
60<br />
40<br />
20<br />
K0<br />
K100<br />
K200<br />
K300<br />
K400<br />
0<br />
10 0 10 1 10 2 10 3 10 4<br />
Figure 4: Risk curves for the polder Mehrum,<br />
scenarios with different breach widths; fg = 1.05<br />
The risk curve for Mehrum was constructed from<br />
the calculated inflow volume of the polder for the<br />
different scenarios (Figure 4). The step-like<br />
trajectories of the risk curves are a result of the<br />
presence of the flood protection system as the<br />
damages only occur for discharges equal to or in<br />
excess of discharges causing levee failure. For<br />
breach widths at Krefeld larger than 300 m, the risk<br />
of damage at Mehrum is zero up to a return interval<br />
of 10 4 years which is a result of the high retention<br />
capacity of the upstream polder. This, again,<br />
T [a]<br />
980
emphasises the key role of upstream levee failures<br />
for the flood risk downstream.<br />
4.2 Uncertainty analysis<br />
The uncertainty analysis performed by the 2 nd level<br />
of Monte Carlo simulations yielded confidence<br />
bounds for each scenario. As an example, the<br />
annual maximum discharge frequency curve at<br />
Rees for the breach scenarios with fg set to 1.05 are<br />
shown in Figure 5. It suggests that, for large events,<br />
the uncertainty decreases with the width of the<br />
breach at Krefeld. This is due to the large breach<br />
outflow combined with the almost infinite retention<br />
capacity of the polder at Krefeld. Most of the<br />
r<strong>and</strong>omised discharges of the uncertainty<br />
distributions that produce a levee breach are<br />
reduced to the level of the levee base in the case of<br />
a 400 m breach, resulting in the upper confidence<br />
bound approaching the frequency curve at the level<br />
of the critical breach discharge.<br />
maximum discharges with return intervals<br />
> 200 years (cf. Figure 2).<br />
The combination of these two facts results in<br />
uncertainty distributions that are almost binary. For<br />
floods associated with return intervals > 200 years<br />
either levee failures producing very high damages<br />
can occur, or if the levees happen to resist the flood,<br />
the polder is protected from any damage. The<br />
confidence intervals calculated from these<br />
uncertainty distributions are consequently<br />
enormous. For return intervals as high as 10 4 years<br />
it is possible that the levee resists the flood <strong>and</strong><br />
protects the polder or it fails <strong>and</strong> causes disastrous<br />
damages. This enormous uncertainty is mainly<br />
attributed to the uncertainty in the annual maximum<br />
discharge.<br />
x 10 4<br />
x 10 4<br />
2<br />
2<br />
Q [m 3 /s]<br />
1.5<br />
1<br />
K100, fg = 1.05<br />
0.5<br />
95% conf. b<strong>and</strong><br />
uncert. distr. data<br />
0<br />
10 0 10 2 10 4<br />
T [a]<br />
x 10 4<br />
Q [m 3 /s]<br />
1.5<br />
1<br />
K200, fg = 1.05<br />
0.5<br />
95% conf. b<strong>and</strong><br />
uncert. distr. data<br />
0<br />
10 0 10 2 10 4<br />
T [a]<br />
x 10 4<br />
2<br />
2<br />
Q [m 3 /s]<br />
1.5<br />
1<br />
K300, fg = 1.05<br />
0.5<br />
95% conf. b<strong>and</strong><br />
uncert. distr. data<br />
0<br />
10 0 10 2 10 4<br />
T [a]<br />
Q [m 3 /s]<br />
1.5<br />
1<br />
K400, fg = 1.05<br />
0.5<br />
95% conf. b<strong>and</strong><br />
uncert. distr. data<br />
0<br />
10 0 10 2 10 4<br />
T [a]<br />
Figure 5: Uncertainty in the exceedance probability<br />
of annual maximum discharges at Rees caused by<br />
the distribution function type <strong>and</strong> the stagedischarge-relationship<br />
for the 4 breach scenarios. fg<br />
was set to 1.05.<br />
The risk curves associated with the flood frequency<br />
curves in Figure 5 are shown in Figure 6. It shows<br />
that the uncertainty in damage is hardly reduced by<br />
the breach width which is in contrast to the results<br />
of the flood frequency curve. The uncertainty<br />
bounds (dashed lines in Figure 6) cover a wide<br />
range from zero damage to almost maximum<br />
damage above return intervals larger than about 200<br />
years.<br />
The presented results indicate that the uncertainty<br />
of the risk assessment is enormous. This is caused<br />
by two facts:<br />
1. the large magnitude <strong>and</strong> duration of floods<br />
required to cause levee failures,<br />
2. the comparatively large uncertainty in the<br />
extreme value statistics for the annual<br />
Figure 6: Exceedance probability of damage at the<br />
polder at Mehrum (solid lines) <strong>and</strong> associated<br />
uncertainty (dashed lines) caused by distribution<br />
function type <strong>and</strong> stage-discharge-relationship for<br />
the K100 <strong>and</strong> K200 breach scenarios. fg was set to<br />
1.05. The points show the Monte Carlo realisations.<br />
5 Discussion <strong>and</strong> Conclusion<br />
The proposed model allows us to perform a<br />
quantitative flood risk analysis including the effect<br />
of levee failures along with the associated<br />
uncertainty. Because of the simple structure of the<br />
model proposed here, a large number of Monte<br />
Carlo-simulations can be performed in a reasonable<br />
time which cover a wide variety of flood events.<br />
981
The approach is therefore very well suited to<br />
integrated flood risk assessment.<br />
Risk assessment (aleatory uncertainty)<br />
The results obtained here suggest that, in the study<br />
reach, upstream levee failures significantly affect<br />
the failure probability downstream <strong>and</strong>, hence the<br />
risk curve of the target area. The simulations also<br />
illustrate the effect of the retention volume of a<br />
polder. Because of the very large retention capacity<br />
of the hinterl<strong>and</strong> at Krefeld, the levee failure<br />
probability at Mehrum is significantly reduced <strong>and</strong><br />
the flood frequency curve at Rees is attenuated if<br />
levee failures at Krefeld are allowed. The size of<br />
the polder at Mehrum controls the shape of the<br />
flood risk curve. The step-like shape of the risk<br />
curve results from the small volume of the polder at<br />
Mehrum <strong>and</strong> the high magnitude of the events<br />
overtopping the levee. However, in case of<br />
upstream breach widths larger than 300 m at<br />
Krefeld the risk equals zero for return intervals up<br />
to 10 4 years. Taking the zero breach scenario at<br />
Krefeld as the worst case scenario for the target<br />
area, the results indicate that the flood protection<br />
structures at Mehrum are sufficient to resist floods<br />
up to return intervals of >1000 years, if the<br />
uncertainty of the results is neglected.<br />
Uncertainty analysis (epistemic uncertainty)<br />
Due to the large uncertainty caused by the<br />
epistemic uncertainty sources the statement that the<br />
flood protection structures at Mehrum are sufficient<br />
to protect the area from a 1000-year flood has to be<br />
corrected. From the uncertainty bounds of the zero<br />
breach scenario, being the worst case for the polder<br />
Mehrum, <strong>and</strong> the 100 <strong>and</strong> 200 m breach width<br />
scenarios shown in Figure 6 it can be concluded<br />
that the flood protection structures are likely to<br />
protect from floods with return intervals of less than<br />
200 years. For larger floods, the uncertainty is<br />
mainly attributed to the extreme value statistics of<br />
the annual maximum discharge <strong>and</strong> yields that both<br />
complete failure <strong>and</strong> no failure may occur<br />
producing a range of possible damage from zero to<br />
maximum damage.<br />
The results suggest that a more reliable extreme<br />
value statistics is crucially important for reducing<br />
the uncertainty of the risk assessment. A major<br />
prerequisite for that are longer time series of annual<br />
maximum discharges. The used series of 35 years is<br />
clearly too short to obtain reliable risk assessments<br />
of events with associated return intervals of more<br />
the 200 years. Also, the uncertainties associated<br />
with the breach module are considerably large.<br />
Better knowledge about the breach development<br />
<strong>and</strong> the distribution of the turf quality on natural<br />
levee systems would most likely reduce this<br />
unknown component of uncertainty in the risk<br />
assessment. The comparison of the results of the<br />
risk analysis with the results of the uncertainty<br />
analysis clearly emphasises the necessity of<br />
uncertainty analysis in flood risk asessment<br />
procedures.<br />
Due to its modular structure <strong>and</strong> the universal<br />
nature of the methods used here, the proposed<br />
model system should be transferable to other river<br />
systems provided the required data sets are<br />
available. In addition, single parts of the model<br />
system may be applied independently, e.g. to<br />
investigate the probability of levee failure at a given<br />
location. It is therefore believed that the system<br />
may be profitably used for a number of additional<br />
purposes, e.g. as a tool for cost-benefit analysis of<br />
flood protection measures, <strong>and</strong> as a decision<br />
support system for operational flood control.<br />
Another possible application is the flood<br />
management <strong>and</strong> control during a severe flood for<br />
which estimates of the effects of upstream levee<br />
breaches on the shape <strong>and</strong> propagation of the flood<br />
wave <strong>and</strong> thus on inundation risks at the reaches<br />
downstream may be useful. Real time simulations<br />
of such scenarios could facilitate the emergency<br />
management <strong>and</strong> enhance the efficiency of planned<br />
levee failures or weir openings. However, a<br />
prerequisite for these applications is an accurate<br />
calibration of the model system to a given reach.<br />
Clearly, this needs to be done prior to a severe<br />
flood event. This implies that, ideally, the flood<br />
management system should be applicable to both<br />
long-term planning tasks <strong>and</strong> operational decision<br />
support.<br />
References<br />
Apel, H., Thieken, A.H., Merz, B., Blöschl, G.:<br />
Flood Risk Assessment <strong>and</strong> Associated<br />
Uncertainty. Natural Hazards <strong>and</strong> Earth<br />
System Sciences, in print, 2004a.<br />
Apel, H., Thieken, A.H., Merz, B., Blöschl, G.: A<br />
probabilistic modelling system for assessing<br />
flood risks. Natural Hazards, Special Issue<br />
“German Research Network Natural<br />
Disasters”, in print, 2004b.<br />
Berga, L.: New trends in hydrological safety. In:<br />
Berga, L. (ed.): Dam safety. Balkema.<br />
Rotterdam. pp. 1099-1106, 1998.<br />
Bowles, D. S., Anderson, L. R., Glover, T. F.: Risk<br />
assessment approach to dam safety criteria.<br />
Uncertainty in the Geologic Environment:<br />
From Theory to Practice. Geotechnical Special<br />
Publication No. 58, ASCE. pp. 451-473, 1996.<br />
Cullen, A.C., Frey, H.C.: Probabilistic techniques<br />
in exposure assessment – A h<strong>and</strong>book for<br />
dealing with variability <strong>and</strong> uncertainty in<br />
models <strong>and</strong> inputs. Plenum Press, New York,<br />
335 p., 1999.<br />
Disse, M., Kamrath, P., Hammer, M. <strong>and</strong> Köngeter,<br />
J.: Simulation of flood wave propagation <strong>and</strong><br />
inundation areas by considering dam break<br />
982
scenarios. Natural Hazards, Special Issue<br />
“German Research Network Natural<br />
Disasters”, in print, 2004.<br />
Ferson, S., Ginzburg, L.R., Different methods are<br />
needed to propagate ignorance <strong>and</strong> variability.<br />
Reliability Eng. <strong>and</strong> Syst. Safety, 54, 133-144,<br />
1996.<br />
Hall, J.W.: H<strong>and</strong>ling uncertainty in the<br />
hydroinformatic process. Journal of<br />
Hydroinformatics, 05.4, 215-232, 2003.<br />
Kortenhaus, A., Oumeraci, H.: Probabilistische<br />
Bemessungsmethoden für Seedeiche<br />
(ProDeich). Bericht No. 877, Leichtweiss-<br />
Institut für Wasserwirtschaft, TU<br />
Braunschweig, 205 pp. (http://www.tubs.de/institute/lwi/hyku/german/Berichte/LWI_<br />
877.pdf), 2002<br />
Maidment, D.R.: H<strong>and</strong>book of Hydrology.<br />
McGraw-Hill, New York, 1000 p., 1992.<br />
MURL (Ministerium für Umwelt, Raumordnung<br />
und L<strong>and</strong>wirtschaft des L<strong>and</strong>es Nordrhein-<br />
Westfalen), Potentielle Hochwasserschäden am<br />
Rhein in Nordrhein-Westfalen. (unpublished<br />
report), 2000.<br />
Plate, E.J.: Stochastic design in hydraulics:<br />
concepts for a broader application. Proc. Sixth<br />
IAHR Intern. Symposium on Stochastic<br />
Hydraulics, Taipei, 1992.<br />
Stewart, M.G., Melchers, R.E.: Probabilistic risk<br />
assessment of engineering systems. Chapman<br />
<strong>and</strong> Hall, London. 1997.<br />
USACE (U.S. Army Corps of Engineers): Riskbased<br />
analysis in geotechnical engineering for<br />
support of planning studies. Engineer<br />
Technical Letter (ETL) 1110-2-556.<br />
Washington DC, 166 p., 1999.<br />
Vrijling, J.K.: Probabilistic Design – Lecture Notes.<br />
IHE Delft, 145 p., 2000.<br />
Vrijling, J.K.: Probabilistic design of water defense<br />
systems in The Netherl<strong>and</strong>s. Reliability<br />
Engineering <strong>and</strong> System Safety. Vol. 74: 337-<br />
344, 2001.<br />
983
Parameters Estimation Using Some Analytical Solutions<br />
of the Anisotropic Advection-dispersion Model<br />
Federico Catania a,b , Marco Massabò a,b <strong>and</strong> Ombretta Paladino a<br />
a CIMA - Centro di Ricerca in Monitoraggio Ambientale<br />
b DIAM – Dipartimento di Ingegneria Ambientale<br />
Università degli Studi di Genova<br />
Via Cadorna 7, 17100 Savona (I)<br />
Phone: +39019230271<br />
email: federico.catania@cima.unige.it, m.marco@cima.unige.it, paladino@unige.it<br />
Abstract In this paper some analytical solutions of the advection-dispersion equation are proposed <strong>and</strong><br />
adopted to solve a non-linear parameter estimation problem. To test the robustness of the analytical<br />
solutions if adopted in inverse problems, the anisotropic dispersion coefficients are estimated using sets of<br />
experimental data simulated by Monte Carlo techniques. Cylindrical geometry is considered since large<br />
columns are the most common devices adopted to study both dispersion <strong>and</strong> kinetics mechanisms <strong>and</strong>, even<br />
if the solutions are expressed in terms of Bessel function expansion <strong>and</strong> equations solved only for particular<br />
initial <strong>and</strong> boundary conditions, they give very good results in terms of reliability <strong>and</strong> precision of our<br />
estimates. Discussion of results is based on the analysis of residuals, variance-covariance matrix <strong>and</strong> bias<br />
of parameters. The influence of location <strong>and</strong> time of sampling, number of samples <strong>and</strong> data uncertainty on<br />
the dispersion coefficients estimates is also analyzed by means of ANOVA tests.<br />
Keywords: Parameter estimation; Column outflow experiments; Solute transport.<br />
1. INTRODUCTION<br />
Contaminant transport in aquifers has become of<br />
arising interest in the last few years for scientist<br />
working in environmental engineering,<br />
hydrology <strong>and</strong> chemical engineering. Some<br />
analytical solutions of the advection-dispersion<br />
equation have been proposed in literature with<br />
the aim of studying the mechanism of<br />
contaminant transport, the movements of<br />
pollutants in groundwater <strong>and</strong> to estimate<br />
chemical-physical parameters. In particular, the<br />
estimation problem, i.e. the situation in which<br />
unknown parameters are to be estimated from<br />
experimental data, is a very difficult task if the<br />
theoretical physical-mathematical model of the<br />
described process is expressed by a PDE: since<br />
the minimum problem descending from the<br />
optimisation procedure has to be solved under<br />
constraints being represented by the PDE itself,<br />
only numerical computations can be adopted<br />
<strong>and</strong>, moreover, convergence to optimum is not<br />
always reached. For these reasons the<br />
investigation about the possibility of using<br />
analytical solutions not only for prediction but<br />
specially for solving inverse problems is a<br />
challenging task.<br />
The analytical solutions of the mathematical<br />
models describing pollutant transport are rarely<br />
possible if some important hydraulic/chemical<br />
effects are considered together, so two or three<br />
dimensional solutions of the convectiondispersion<br />
equations are given often for nonreacting<br />
contaminants or for simple degradation<br />
or decay <strong>and</strong> isotropic dispersion (Broadbridge et<br />
al. [2002]); otherwise solutions taking into<br />
account of nonlinear chemical<br />
adsorption/desorption are found only for a<br />
984
monodimensional advection-dispersion scheme<br />
(Van der Zee [1990], Bosma <strong>and</strong> Van der Zee<br />
[1993]).<br />
The authors recently studied <strong>and</strong> proposed<br />
(Massabò et al. [2004]) some analytical solutions<br />
for the transport equation in cylindrical geometry<br />
for a reacting solute under chemical decay or<br />
linear adsorption-like reaction by taking into<br />
account of dispersion in both radial <strong>and</strong> axial<br />
directions. Bessel function expansion is used to<br />
solve the second order PDE model with different<br />
initial conditions corresponding to usual<br />
experimental practices. This mathematical model<br />
represents one of the most adopted experimental<br />
device to investigate about pollutant transport<br />
phenomena <strong>and</strong> so it can be used to fit<br />
experimental data from large columns in which<br />
contaminant flows through a saturated porous<br />
media.<br />
The use of analytical solutions of the advectiondispersion<br />
equation in estimation problems is<br />
very limited. Parameter estimation is difficult<br />
when flow <strong>and</strong> transport parameters are to be<br />
optimised simultaneously because of slow<br />
convergence rates <strong>and</strong> unstable estimates. So,<br />
usually the flow parameters are assumed to be<br />
known. Murphy <strong>and</strong> Scott [1977] introduced an<br />
inverse approach in order to estimate the<br />
dispersivities from observed concentration<br />
values. Strecker <strong>and</strong> Chu [1986] were the first to<br />
estimate both flow <strong>and</strong> transport parameters<br />
dividing the optimisation procedure in two<br />
separate stages, while Wagner <strong>and</strong> Gorelick<br />
[1987] estimated flow <strong>and</strong> transport parameters<br />
simultaneously by inverse modelling, making<br />
use of nonlinear regression based on the least<br />
squares method.<br />
In this work the analytical solutions of the<br />
transport equation proposed by the authors are<br />
used for parameter estimation. In order to<br />
investigate solutions suitability if adopted in<br />
inverse problems, only the dispersion<br />
coefficients are considered at this time: the<br />
analytical solutions contain the full Bessel<br />
expansion also in this case; so, disregarding of<br />
the kinetic term does not influence the reliability<br />
of the entire procedure.<br />
Discussion of results is based on the analysis of<br />
residuals, variance-covariance matrix of<br />
parameters <strong>and</strong> variance-covariance matrix of the<br />
experimental data. The influence of sampling<br />
methodology on the parameter estimates is<br />
analyzed in terms of number of samples, their<br />
location in the model space (x, r, t) <strong>and</strong> the<br />
experimental error.<br />
2. THEORETICAL FRAMEWORK<br />
2.1 The estimation problem<br />
Let us consider the situation in which unknown<br />
parameters θ are to be estimated from<br />
experimental data w* <strong>and</strong> the theoretical model<br />
describing the experimental campaign is defined<br />
by an implicit model.<br />
If the theoretical implicit model is expressed by a<br />
PDE, we have the following situation.<br />
Let the experimental data w* be connected by a<br />
linear mapping (or be the same of) to the<br />
solution u of a PDE, expressed as follows:<br />
L ( u)<br />
= 0<br />
(1)<br />
θ r<br />
∂u<br />
A u + B = f ( ξ ) on the boundary S’ (2)<br />
∂n<br />
defined by coordinates ξ <strong>and</strong> where the subscript<br />
θ indicates the vector of the unknown parameters<br />
(the dispersion coefficients here discussed) in the<br />
differential equation, S’ is the domain in which<br />
the generical mixed boundary condition is<br />
defined; n indicates the direction normal to S’.<br />
Let F be the linear mapping:<br />
F :u →<br />
(3)<br />
r w θ<br />
v θ<br />
where w θ is to be compared with the<br />
experimental data w* to carry out some kind of<br />
regression analysis by minimizing some<br />
convenient objective function Φ ; e.g.:<br />
Φ = w r − w *<br />
(4)<br />
θ<br />
The formal solution of problem (1) + (2) is given<br />
by<br />
"<br />
u r ( x)<br />
= G r ( x,<br />
ξ ) f ( ξ)<br />
dξ<br />
(5)<br />
θ<br />
∫<br />
P<br />
θ<br />
where G” indicates the θ-dependent Green’s<br />
function of the second kind <strong>and</strong> x the<br />
independent variables of the domain on which<br />
u is defined. As there are no general solution<br />
methods for complex PDE’s, if exact or<br />
approximate analytical solutions are not<br />
available, only numerical solutions for problem<br />
(1) + (2) can be used to solve the inverse<br />
estimation problem (4). Solving problem (4)<br />
where w θ depends on u θ by the linear mapping<br />
(3) <strong>and</strong> where u θ is the solution of problem (1) +<br />
(2) means an optimization procedure which<br />
cannot be easily solved by numerical algorithms:<br />
we have the typical convergence difficulties of a<br />
constrained optimization algorithm plus the<br />
985
typical numerical errors of the PDEs solver, both<br />
affecting the search for the global minimum.<br />
In fact, if the model remains in implicit form, a<br />
maximum likelihood estimation procedure leads<br />
to a nonlinear, constrained minimization problem<br />
where constraints are given by the equation<br />
model itself, i.e. equations (1) + (2) + (3); <strong>and</strong><br />
w* are the measured values of all the model<br />
variables regardless their kind, i.e. time, spatial<br />
coordinates <strong>and</strong> state variables (t, x, ξ, u) <strong>and</strong> θ<br />
is the vector of the unknown parameters.<br />
Only if:<br />
i) the model can be solved in reduced<br />
form (explicit analytical solution);<br />
ii) the errors on the independent variables<br />
<strong>and</strong> on space <strong>and</strong> time measures are<br />
iii)<br />
neglected;<br />
the errors on the remaining variables<br />
are independent (Bard[1974]);<br />
the estimation problem reduces to an<br />
unconstrained minimization problem of the kind<br />
weighted or not-weighted least squares. Hence<br />
the importance of finding analytical solutions for<br />
problem (1) + (2) so to satisfy condition (i).<br />
Conditions (ii) <strong>and</strong> (iii) can be easily satisfied if<br />
ad hoc experimental conditions are observed.<br />
It is also important to notice that linear<br />
differential models <strong>and</strong> their analytical solutions<br />
are particularly important in data analysis<br />
because the experimental conditions can be often<br />
forced to keep in the linear domain (relaxation<br />
experiments)<br />
mechanisms such as molecular diffusion,<br />
hydrodynamic dispersion, eddy diffusion or<br />
mixing. Using typical dimensionless variables,<br />
the equation becomes:<br />
∂ C ∂C<br />
+ u =<br />
η<br />
∂t<br />
∂x<br />
Pe<br />
where:<br />
Pe<br />
U R<br />
;<br />
D<br />
2<br />
∂ C 1 ∂C<br />
1<br />
+ ) +<br />
2<br />
∂r<br />
r ∂r<br />
Pe<br />
2<br />
∂ C<br />
(<br />
2<br />
R<br />
L<br />
∂x<br />
U L<br />
PeL<br />
= ;<br />
D<br />
R<br />
=<br />
= =<br />
η =<br />
R<br />
L<br />
L<br />
;<br />
R<br />
(7)<br />
(8)<br />
where R <strong>and</strong> L are respectively the radius <strong>and</strong> the<br />
length of the column <strong>and</strong> U 0 <strong>and</strong> C 0 the scales of<br />
velocity <strong>and</strong> concentration.<br />
2.2 Model equations <strong>and</strong> analytical solutions<br />
Under the hypothesis that large columns are<br />
adopted to investigate anisotropic dispersion<br />
(showed in photograph 1), we assume that the<br />
initial conditions do not depend on the angular<br />
variable; as a consequence, the process preserves<br />
symmetry around the longitudinal axis. The<br />
advection-dispersion PDE expressing the mass<br />
balance of a generic solute in terms of<br />
dimensional concentration C(r,x,t), can be<br />
written as follows:<br />
2<br />
∂C*<br />
∂C*<br />
∂ C*<br />
+ u*<br />
= DR<br />
( +<br />
2<br />
∂t*<br />
∂x*<br />
∂r*<br />
1<br />
r*<br />
∂C*<br />
) + D<br />
∂r*<br />
L<br />
2<br />
∂ C*<br />
2<br />
∂x*<br />
(6)<br />
Here the constant advective term is represented<br />
by the average pore water velocity u* <strong>and</strong><br />
anisotropic dispersion is described by means of<br />
the two mechanical dispersion coefficients D R<br />
<strong>and</strong> D L . They represent different dispersion<br />
Photograph 1: Large column adopted to<br />
investigate anisotropic dispersion.<br />
Boundaries <strong>and</strong> initial conditions are necessary<br />
to have a unique solution. We assume:<br />
∂C(<br />
r,<br />
x,<br />
t)<br />
∂r<br />
r=<br />
1<br />
lim C(<br />
r,<br />
x,<br />
t)<br />
= 0;<br />
x→+∞<br />
∂C<br />
lim = 0;<br />
x→+∞<br />
∂x<br />
= 0;<br />
(9)<br />
(10)<br />
(11)<br />
The first condition represents the mathematical<br />
formulation of the impermeability of the column<br />
wall while the other two are the simplest<br />
boundary conditions representing a semi-infinite<br />
system. By considering the following further<br />
initial <strong>and</strong> boundary conditions describing the<br />
pollutant release:<br />
986
C r,0,<br />
t)<br />
= C H(<br />
t);<br />
(<br />
0<br />
C(<br />
r,<br />
x,0)<br />
= 0;<br />
(12)<br />
(13)<br />
where C 0 is the concentration of contaminant in<br />
the inlet section <strong>and</strong> H(t) is the Heavyside<br />
function. For this case a particular Bessel series<br />
expansion of the concentration function is used<br />
giving the following analytical solution:<br />
C(<br />
r,<br />
x,<br />
t)<br />
=∑ +∞<br />
k=<br />
0<br />
1 1 ⌈xuPeL<br />
⌉<br />
A J0(<br />
Z r)<br />
exp<br />
2 2 <br />
⌊ ⌋<br />
⋅<br />
K k<br />
<br />
⌈<br />
⋅exp<br />
x<br />
⌊<br />
⌈<br />
1<br />
⋅erfc<br />
x<br />
2<br />
⌊<br />
⌈<br />
+ exp − x<br />
⌊<br />
⌈<br />
1<br />
⋅erfc<br />
x<br />
2<br />
⌊<br />
2 2<br />
u PeL<br />
+ η<br />
4<br />
PeL<br />
+<br />
t<br />
1 2<br />
[ Z ]<br />
2 2<br />
u PeL<br />
+ η<br />
4<br />
PeL<br />
−<br />
t<br />
k<br />
1 2<br />
[ Z ]<br />
Pe ⌉<br />
L<br />
⋅<br />
PeR<br />
⌋<br />
1 2<br />
[ Z ]<br />
2<br />
U PeLt<br />
k<br />
+ η<br />
4 Pe<br />
k<br />
Pe ⌉<br />
L<br />
<br />
PeR<br />
⌋<br />
R<br />
2<br />
U PeLt<br />
+ η<br />
4 Pe<br />
⌉<br />
t<br />
+<br />
<br />
⌋<br />
[ ] 1 2 ⌉<br />
Zk<br />
<br />
t (14)<br />
R<br />
<br />
<br />
⌋<br />
<br />
1<br />
where J 0 is the zero-order Bessel function, Z is<br />
k<br />
the k-th zeros of the first order Bessel function<br />
<strong>and</strong> the Bessel’s coefficients are:<br />
A<br />
A<br />
k<br />
0<br />
=<br />
=<br />
1<br />
∫<br />
0<br />
1<br />
∫<br />
2 ρf<br />
( ρ)<br />
J ( Z ρ)<br />
dρ<br />
0<br />
k 2<br />
[ J ( Z )]<br />
0<br />
ρf<br />
( ρ)<br />
dρ<br />
0<br />
1<br />
k<br />
1<br />
,<br />
k = 1,2,...<br />
3. THE ESTIMATION PROCEDURE<br />
(15)<br />
Since we have the analytical solution of the PDE<br />
describing the column behaviour (i.e. condition I<br />
of paragraph 2.1), we can estimate the unknown<br />
parameters Pe L <strong>and</strong> Pe R by solving an<br />
unconstrained minimization problem (least<br />
squares) only if also conditions ii) <strong>and</strong> iii) of<br />
paragraph 2.1 hold true. We can easily suppose<br />
that errors on measures of time <strong>and</strong> space are<br />
negligible with respect to measures on<br />
concentration in the body of the large column.<br />
Concentration measures in the liquid phase could<br />
be carried out with HPLC or Atomic Absorption<br />
if we use, for example, hydrocarbons of heavy<br />
metals as contaminants, in case of reactive flow<br />
(not developed in this paper), or bromide if we<br />
refer to conservative tracers. The overall<br />
estimation procedure has been carried out using<br />
simulated experimental data with errors<br />
belonging to a Gaussian distribution with zero<br />
mean <strong>and</strong> 0.01 or 0.05 variance. We considered<br />
different sampling points in the column length<br />
<strong>and</strong> radius (preserving symmetry around the<br />
longitudinal axis) <strong>and</strong> repetitions at different<br />
time as shown in figure 1.<br />
Figure 1: Grid of possible space points in the<br />
soil column.<br />
30 Monte Carlo runs for each experimental<br />
situation hypothesized have been carried out.<br />
The scheme of the adopted data simulation with<br />
two levels on the number of measurement points<br />
in the spatial domain (S) <strong>and</strong> two levels on the<br />
experimental error variance (V) is resumed in<br />
Table 1.<br />
Situation nx nr nt σ 2<br />
S1V1 5 3 4 0.01<br />
S2V1 5 5 4 0.01<br />
S3V1 10 3 4 0.01<br />
S4V1 10 5 4 0.01<br />
S1V2 5 3 4 0.05<br />
S2V2 5 5 4 0.05<br />
S3V2 10 3 4 0.05<br />
S4V2 10 5 4 0.05<br />
Table 1: The experimental situations analyzed<br />
for parameter estimation.<br />
The true values for parameters Pe L <strong>and</strong> Pe R , i.e.<br />
values adopted to generate the experimental<br />
solute concentration by adding experimental<br />
errors to the output of equation (9), are both set<br />
to 100. A Marquardt’s modified algorithm<br />
(Marquardt [1963], Bard [1974]) with analytical<br />
first order derivatives supplied by the user has<br />
been adopted to solve the nonlinear optimization<br />
problem.<br />
r<br />
z<br />
987
In Table 2 the results of the parameter<br />
estimation procedure are briefly summarized,<br />
where Pe* L <strong>and</strong> Pe* R are the mean values (over<br />
the 30 Monte Carlo runs) of the estimated<br />
parameters <strong>and</strong> σ 2 PeL <strong>and</strong> σ 2 PeR are the related<br />
calculated variances.<br />
Situation Pe* L Pe* R σ 2 PeL σ 2 PeR<br />
S1V1 99,94 99,91 1,75 0,26<br />
S2V1 99,96 99,92 1,23 0,20<br />
S3V1 99,81 99,89 0,55 0,11<br />
S4V1 99,87 99,91 0,33 0,08<br />
S1V2 101,44 99,20 31,84 7,18<br />
S2V2 101,35 99,50 21,11 4,90<br />
S3V2 100,45 99,78 20,36 3,60<br />
S4V2 100,09 99,91 13,00 2,69<br />
Table 2: Results of the estimation procedure.<br />
Figure 1: Example of residuals plot, when nx=5,<br />
nr=3, nt=4 <strong>and</strong> σ 2 =0,01.<br />
Figure 2: Example of scatter plot, when nx=5,<br />
nr=3, nt=4 <strong>and</strong> σ 2 =0,01.<br />
For all the situations <strong>and</strong> for each run a complete<br />
analysis of the residuals has been carried out.<br />
Also the computed variance-covariance matrix of<br />
the parameters, obtained from the Hessian matrix<br />
at minimum (Bard 1974), has been compared<br />
with the calculated values of variance of Table 2.<br />
In Figures 1 <strong>and</strong> 2 one example of residuals <strong>and</strong><br />
one scatter plot are shown.<br />
4. VALIDATION AND DISCUSSION OF<br />
RESULTS<br />
From the analysis of the objective function at<br />
minimum <strong>and</strong> its derivatives, residuals, variancecovariance<br />
matrix of the parameters, mean value<br />
<strong>and</strong> bias of the parameters, we can say that the<br />
proposed procedure for the estimation of the<br />
dispersion parameters seems to work very well.<br />
The use of the analytical solution of the<br />
advection-dispersion equation for parameters<br />
estimation gives values of dispersion coefficients<br />
very close to the true ones <strong>and</strong> with low st<strong>and</strong>ard<br />
deviation. No over- or under-estimation has been<br />
found.<br />
In order to analyse the relation between number<br />
of samples <strong>and</strong> data uncertainty on the residuals<br />
of parameter values in comparison to their mean,<br />
these residuals have been also analyzed by<br />
means of ANOVA tests. ANOVA has been done<br />
for all the situations described in Table 1 in order<br />
to establish which is the factor, among those<br />
enumerated above, to which most of the variance<br />
of the residuals is to be ascribed.<br />
Fischer tests on ANOVA results show that we<br />
have a minimum confidence level of 95% only<br />
when we compare the bigger influence of data<br />
uncertainty due to Monte Carlo runs with the<br />
number of points on x-axis.<br />
One example of ANOVA <strong>and</strong> Fischer test results<br />
for Pe L , in the case of nx=5, nr=3, nt=4 <strong>and</strong><br />
σ 2 =0,01, is shown in Table 3 <strong>and</strong> Table 4.<br />
Factor 1:<br />
Data<br />
uncertainty<br />
Factor 2:<br />
N° of<br />
points on<br />
r-axis<br />
Factor 3:<br />
N° of<br />
points on<br />
x-axis<br />
Variance<br />
of mean<br />
deviation<br />
PeR<br />
Degree<br />
of<br />
freedom<br />
Mean<br />
Square<br />
11.8314 29 0.4079<br />
1.0059 1 1.0059<br />
6.6849 1 6.6849<br />
Table 3: ANOVA results for Pe L when nx=5,<br />
nr=3, nt=4 <strong>and</strong> σ 2 =0,01.<br />
988
Factor<br />
comparison<br />
Fischer<br />
distribution<br />
value<br />
Confidence<br />
level<br />
1-2 16.3854 99.96%<br />
1-3 2.4656 87.28%<br />
2-3 6.6453 76.44%<br />
Table 4: Fischer tests results for Pe L when nx=5,<br />
nr=3, nt=4 <strong>and</strong> σ 2 =0,01.<br />
5. CONCLUSIONS<br />
The procedure here proposed to estimate the<br />
longitudinal <strong>and</strong> transversal dispersion<br />
coefficients in pollutant transport problems <strong>and</strong><br />
based on analytical solutions of the advectiondispersion<br />
equation, gave very good results in<br />
terms of: parameters values very closed to the<br />
true ones, low st<strong>and</strong>ard deviation, robustness <strong>and</strong><br />
reliability of the estimation procedure in all the<br />
simulated experimental situations. Even if the<br />
analytical solutions are possible only with simple<br />
boundary conditions, with some expedients the<br />
real experimental conditions can be forced to<br />
keep in that domain.<br />
The influence of sampling methodology on the<br />
parameter estimates has been also analyzed in<br />
terms of number of samples, their location <strong>and</strong><br />
experimental error to give information about the<br />
choice of the sample domain that, especially<br />
when field campaigns are to be performed <strong>and</strong><br />
the position of the piezometric wells are to be<br />
fixed, strongly influence the whole cost of<br />
experimentation.<br />
Further analyses will regard some consideration<br />
on the precision of the analytical model in terms<br />
of model output sensitivity, also using the<br />
available analytical solution (Massabò et al.<br />
[2004]) with kinetic terms. Besides, in future<br />
works, the extension of the estimated parameter<br />
set, by considering, for example, the pore water<br />
velocity, will be analysed. The analytical<br />
solution could be used also to test parameter<br />
estimation procedures carried out using<br />
numerical algorithms for the solution of the 2D<br />
advection dispersion equation.<br />
Dimensional,<br />
Chemically<br />
Heterogeneous Porous Medium, Water<br />
Resources Research, 29(1), 117–<br />
131, 1993.<br />
Broadbridge, P., J. Moitsheki <strong>and</strong> M. P.<br />
Edwards, Analytical Solutions for Two-<br />
Dimensional Solute Transport with<br />
Velocity-Dependent Dispersion,<br />
<strong>Environmental</strong> Mechanics Water, Mass<br />
<strong>and</strong> Energy Transfer in the Biosphere,<br />
Geophysical Monograph Series, Vol.<br />
129, CSIRO Publishing, 2002<br />
Marquardt, D.W., An Algorithm for Least<br />
Squares Estimation of Nonlinear<br />
Parameters, SIAM Journal of Applied<br />
Mathematics, 11, 431-441, 1963.<br />
Massabò, M., R. Cianci <strong>and</strong> O. Paladino, Some<br />
Analytical Solutions for the Dispersion-<br />
Convection – Reaction Equation in<br />
Cylindrical Geometry, submitted.<br />
Murphy, V.V.N., <strong>and</strong> V.H. Scott, Determination<br />
of Transport Model Parameters in<br />
Groundwater Aquifers, Water<br />
Resources Research, 13(6), 941-947,<br />
1977<br />
Strecker, E.W., <strong>and</strong> W. Chu, Parameter<br />
Identification of a Groundwater<br />
Contaminant Transport Model, Ground<br />
Water, 24(1), 56-62, 1986.<br />
Van der Zee, M.A., <strong>and</strong> E. A. T. M. Sjoerd,<br />
Analysis of Solute Redistribution in<br />
Heterogeneous Field, Water Resources<br />
Research, 26(2), 273–278, 1990.<br />
Wagner, B.J., <strong>and</strong> S.M. Gorelick, Optimal<br />
Groundwater Management under<br />
Parameter Uncertainty, Water<br />
Resources Research, 23(7), 1162-1174,<br />
1987.<br />
6. REFERENCES<br />
Bard, Y., Nonlinear Parameter Estimation,<br />
Academic Press, 341 pp., San Diego,<br />
Calif., 1974.<br />
Bosma, W.P., <strong>and</strong> S.E. Van der Zee, Transport<br />
of Reacting Solute in a One-<br />
989
Soil Hydraulics Properties Estimation by Using<br />
Pedotransfer Functions in a Northeastern Semiarid Zone<br />
Catchment, Brazil<br />
L. F. F. Moreira, A. M. Righetto <strong>and</strong> V. M. de A. Medeiros a<br />
a Programa de Pós-graduação em Engenharia Sanitária/ Universidade Federal do Rio Gr<strong>and</strong>e do Norte<br />
Natal/RN Brasil 59072-970<br />
Abstract: Hydrological modeling of the unsaturated soil zone fluxes allows the transfer processes simulation<br />
through the hydrological active soil zone. The pedotransfer functions (FTP) are useful tools in the modeling<br />
process. They contain analytical functions derived through statistic optimization process using a large amount<br />
of soil information data. This paper aims to analyze the level of reliability of two different pedotransfer<br />
functions [Wösten et al. (2001); Hodnett <strong>and</strong> Tomasella (2000)] by using field measurement of soil properties<br />
<strong>and</strong> experimental infiltration data through a disc infiltrometer in an experimental catchment at northeastern<br />
semi arid zone of Brazil. FTP’s used in this study were derived on previous studies by taking into account<br />
soils of different origins. The former FTP considers a large range of soil classes from temperate climate<br />
regions; the latter was derived through a selection of soils from tropical climate regions. The use of these<br />
pedotransfer functions showed a large variation between calculated soil hydraulic parameters. The tropical<br />
climate function seemed a better adjustment to the experimental data. Van Genuchten parameters <strong>and</strong><br />
experimental infiltration data allowed the derivation of the unsaturated hydraulic conductivity <strong>and</strong> soil porewater<br />
tension functions. The derived soil hydraulic parameters showed spatial <strong>and</strong> temporal variation within<br />
the catchment.<br />
Keywords: Pedotransfer function; tropical soils; reliability<br />
1. INTRODUCTION<br />
Hydrological models are important tools in<br />
physical processes research particularly in the<br />
unsaturated media called hydrological active zone<br />
of soils. However, the task of modeling normally<br />
one supposes previous soil hydraulics parameters<br />
determination. Modeling normally requires<br />
derivation of input parameters that characterize<br />
retention <strong>and</strong> flow capacity at soil vadose zone.<br />
They can translate soil hydraulic behavior in<br />
function of soil physical, chemical <strong>and</strong> biological<br />
properties. According to soil hydraulic behavior,<br />
flow through the vadose zone is highly dependent<br />
of the temporal <strong>and</strong> spatial variability in its<br />
characteristics. Actually, one of the questions<br />
concerning the pedotransfer functions (PTF)<br />
development by many researchers deals with its<br />
capacity to predict accurately soil hydraulic<br />
properties. Accuracy implies a high level of<br />
correspondence between measured <strong>and</strong> predicted<br />
data set from which a PTF was derived.<br />
In this sense, pedotransfer functions are very<br />
useful tools in modeling application. PTFs are<br />
analytical functions derived through statistical<br />
optimization involving a wide variety of<br />
information of different soil types. Such<br />
information consists of soil hydraulic properties<br />
database, grouping a large number of soil<br />
horizons throughout the world. These data are<br />
obtained directly in the field by experiments <strong>and</strong><br />
bulk analysis on laboratory: soil composition,<br />
structure, bulk density, percentage of silt, clay <strong>and</strong><br />
organ matter <strong>and</strong> pH.<br />
The great advantage of PTFs use is the possibility<br />
of predicting soil hydraulic parameters directly.<br />
The derived parameters are commonly used to<br />
express soil water retention <strong>and</strong> hydraulic<br />
conductivity as functions of water volume content<br />
[van Genuchten <strong>and</strong> Mualen (1992), Brooks <strong>and</strong><br />
Corey (1964)]. These functions can be<br />
incorporated into hydrologic distributed models<br />
because they can be able to simulate spatial soil<br />
hydraulic behavior variation through the<br />
watershed.<br />
PTFs can be derived by two approaches: class<br />
PTF <strong>and</strong> continuous PTF. Class PTF is developed<br />
separately for each group individually [Wösten et<br />
al (1990)]. Many researchers have developed<br />
methods to group soils, taking into account some<br />
soil properties. On the other h<strong>and</strong>, continuous<br />
PTF is developed without grouping the data, but<br />
using all data set to derive equations. Actually, it<br />
has been discussed the accuracy of each type of<br />
990
PTF [Hodnett <strong>and</strong> Tomasella (2002)]. For<br />
example, use of class PTF can give good results if<br />
soil mineralogical characteristics of the soils as<br />
well as textural class are similar.<br />
Usually, relationships to predict soil properties<br />
deal with many parameters related one to another<br />
in an undirected relation. This explains the low<br />
level of agreement of these relations. They are<br />
obtained by using mathematical methods<br />
involving a number of soil properties. That<br />
procedure generates some errors that can lead to<br />
limited relations in terms of accuracy. Despite of<br />
the use of regression analysis, this technique has<br />
shown a limited capacity to translate the whole<br />
shape of dependence between parameters. The<br />
development of statistical methods has permitted<br />
to improve the level of accuracy. Actually,<br />
artificial neural networks have become a common<br />
technique used by many authors because of its<br />
ability to work with complex systems.<br />
In practice, PTF is an empirical relationship. So,<br />
its use must be limited by the range/type of soils<br />
from the data used to derive it. Most of the PTFs<br />
developed to predict Brooks <strong>and</strong> Corey (1964)<br />
<strong>and</strong> van Genuchten (1980) parameters were<br />
derived using data sets from soils of temperate<br />
regions. Hodnett (1995) warned that PTFs<br />
developed for temperate soils should be applied<br />
with caution to tropical soils. Tomasella et al.<br />
(2000) tested a PTF derived from Brazilian soil<br />
data <strong>and</strong> concluded that it had better results than<br />
using temperate soil PTF. They concluded that<br />
there might have functional differences between<br />
temperate <strong>and</strong> tropical soils caused by some<br />
factors other than texture.<br />
Hodnett <strong>and</strong> Tomasella (2002) derived a PTF<br />
using a data set of soils from tropical regions,<br />
which was contained in IGBT-DIS soil database.<br />
The data was checked <strong>and</strong> reduced to 771<br />
horizons from 21 tropical countries. They were<br />
divided into two groups to be used in parameter<br />
calibration <strong>and</strong> PTF validation. These data<br />
contained eight pairs of points that defined the<br />
observed water release curve for each soil,<br />
volumetric soil water content (m 3 /m 3 ) in function<br />
of soil matric potential (kPa). These data was<br />
used to derive van Genuchten parameters (α, n, θ s<br />
<strong>and</strong> θ r ) using a nonlinear least squares fitting<br />
routing, resulting in a good level of fitting (91%<br />
had R 2 >0,97).<br />
Wösten et al. (2001) compared the use of 21<br />
different PTFs. They applied them to predict<br />
water content at –33 kPa <strong>and</strong> –1500 kPa by using<br />
a measured dataset from Oklahoma. The tests<br />
showed many types of discrepancies between<br />
them, which may confirm the fact that PTF is an<br />
empirical relationship. In this sense, reliable<br />
predictions must consider the fact that its use is<br />
valid only for soil horizons that fall in the same<br />
texture range as the horizons for which they have<br />
been developed. In this study, a PTF obtained by<br />
Tomasella <strong>and</strong> Hodnett (2002), derived from a<br />
data set composed by 614 tropical soil horizons,<br />
showed a partially good fitting for low water<br />
content (Ψ=-1500 kPa) <strong>and</strong> failing for high water<br />
content (Ψ=-33 kPa).<br />
Hodnett <strong>and</strong> Tomasella (2002) found significant<br />
differences in soil characteristics between<br />
temperate <strong>and</strong> tropical soils. A comparison of soil<br />
textural class distribution between tropical <strong>and</strong><br />
temperate datasets showed great difference in<br />
distributions, especially in clay class (far higher in<br />
tropical sets). A comparison of soil samples<br />
properties showed that, in general, tropical soils<br />
had less mean bulk density for each class than<br />
temperate soils. On the same way, a comparison<br />
of van Genuchten parameters derived for tropical<br />
<strong>and</strong> temperate soils showed significant<br />
differences. The values of parameter α for s<strong>and</strong> in<br />
temperate soils were more than twice higher than<br />
values obtained for tropical soils. For clay class,<br />
differences showed that both soil texture <strong>and</strong><br />
mineralogy are important factors to be<br />
considered. The mean values of fitted θ s for<br />
tropical dataset were higher than for the temperate<br />
data for all classes. On the same way, θ r mean<br />
values in the tropical dataset showed to be higher<br />
than for the temperate soils. This difference<br />
between soils properties from tropical <strong>and</strong><br />
temperate datasets explains the occurrence of<br />
marked differences between water release curves.<br />
2. FUNCTIONS FOR WATER<br />
RETENTION CHARACTERISTICS<br />
In unsaturated porous media, Darcy law defines<br />
flow as a result of soil capacity of transmission<br />
(hydraulic properties) <strong>and</strong> the action of an energy<br />
gradient, as follows,<br />
V = −Kns<br />
∇φ<br />
(1)<br />
where<br />
Kns<br />
= f1( ψ ) ; Kns<br />
= f2<br />
( θ)<br />
; ∇φ<br />
=∇(<br />
ψ −Z)<br />
k ns is unsaturated soil hydraulic conductivity<br />
(cm/s), Ψ is soil matric potential (cm), φ is total<br />
energy head, z is measured vertical distance from<br />
soil surface (cm), θ is volumetric soil water<br />
content (cm 3 /cm 3 ).<br />
Hydraulic gradient is the energy used by flowing<br />
water. Solutions of problems governed by (1)<br />
involve functions that express relations of soil<br />
water retention <strong>and</strong> hydraulic conductivity with<br />
soil water content. Richards equation (1931)<br />
describes flow in unsaturated soil as a<br />
combination of Darcy law <strong>and</strong> conservation of<br />
mass law, as follows,<br />
991
¥¤££<br />
©©<br />
¨<br />
§<br />
¢<br />
¡<br />
<br />
¥¤££<br />
¢<br />
¡<br />
<br />
∂<br />
∂y<br />
∂φ<br />
∂ ∂φ<br />
* ∂φ<br />
K ( ψ ) + K ( ψ ) = θ (2)<br />
∂y<br />
∂z<br />
∂z<br />
∂t<br />
* ∂θ<br />
where θ = . θ * can be derived by soil water<br />
∂ψ<br />
release curve. Flow equation solution involves<br />
two analytical functions: soil hydraulic<br />
conductivity, k=f 1 (ψ), soil matric potential,<br />
ψ=f 2 (θ). The relation ψ=f 2 (θ) can be described<br />
empirically by a number of equations. Brooks-<br />
Corey equation (1964) is defined by the following<br />
expression,<br />
λ<br />
S = b<br />
where<br />
ψ<br />
ψ ¦¦<br />
θ − θ<br />
S =<br />
r<br />
(3)<br />
θ s − θ r<br />
S is soil saturation level, θ s is volumetric soil<br />
water content at saturation (cm 3 /cm 3 ) <strong>and</strong> θ r is<br />
residual water content (cm 3 /cm 3 ), defined as the<br />
water that can be extracted from soil at high<br />
temperatures. ψ b is the air entry pressure head <strong>and</strong><br />
λ is the pore distribution index, both empirical<br />
fitting parameters.<br />
Van Genuchten equation (1980) is a reference in<br />
its ability to characterize flow condition in<br />
unsaturated soil. It has shown good results for a<br />
variety of soils, <strong>and</strong> is defined as follows,<br />
S =<br />
1 +<br />
<br />
1<br />
n<br />
( α ψ )<br />
m<br />
, ψ ≤ 0 (4)<br />
Relationship between parameters m <strong>and</strong> n is<br />
1<br />
m = 1 −<br />
n<br />
where n is a dimensionless parameter that<br />
determines the steepness of the water release<br />
curve.<br />
The parameter α is equal to the inverse of Ψ at<br />
the point where the curve is steepest.<br />
Flow capacity in unsaturated soil zone can be well<br />
characterized with a model proposed by van<br />
Genuchten <strong>and</strong> Mualen (1992), using van<br />
Genuchten parameters, defined as follows,<br />
2<br />
0,5 /( 1) n n− (1−<br />
1/ n)<br />
K S)<br />
= Ksat<br />
S 1−(1−<br />
S ) (5)<br />
( <br />
This work aims to study the level of reliability of<br />
Hodnett <strong>and</strong> Tomasella PTF in its application to<br />
soils of semiarid regions. For this purpose, it will<br />
be used data from soil sample analysis on<br />
laboratory <strong>and</strong> measured hydraulic conductivity at<br />
saturation from infiltration experiments on 8<br />
locations within an experimental catchment in<br />
Brazilian northeastern semiarid zone, state of Rio<br />
Gr<strong>and</strong>e do Norte.<br />
3. METHODS<br />
The study area is located approximately 300 km<br />
west from Natal, state of Rio Gr<strong>and</strong>e do Norte. It<br />
is located within the Espinharas river watershed<br />
in northeastern semiarid zone. The catchment area<br />
is 3,82 km 2 (Fig. 1). It has been environmentally<br />
protected for many years by IBAMA (Brazilian<br />
Institute for <strong>Environmental</strong> Protection)<br />
Administration, Brazilian government. It is<br />
topographically located at the Espinharas<br />
watershed boundary. Research area relief is<br />
partially hill slope; a topographically closed basin<br />
presents an alluvium area located upstream a<br />
reservoir used to storage water for human<br />
consumption.<br />
Figure 1. Geographical coordinates of catchment area<br />
992
Survey measurements showed that soil depth<br />
ranges from 0,2 m to 1,2 m; geological rock<br />
formation appears on the surface area at some<br />
points. At the midl<strong>and</strong> part of the catchment, soil<br />
surface erosion is produced by the action of a<br />
potentially high overl<strong>and</strong> flow on flat areas. On<br />
these areas, the existence of pebbles <strong>and</strong> cobbles<br />
rest on soil surface indicates that: a. sediment was<br />
eroded from the surface; b. soil grain size<br />
distribution is bimodal; c. soil horizon is well<br />
consolidated.<br />
Natural drainage network presents, in most area,<br />
boulder formation along an ephemeral stream<br />
talweg. It indicates that most part of sediment<br />
yield may occur on the drainage network<br />
upstream, where local scour erosion process is<br />
caused by channel flow.<br />
Soil samples were collected at 0,15 m depth in 8<br />
different points covering the most part of the<br />
basin area. Laboratory measurements included<br />
grain size analysis, gravimetric water content,<br />
bulk density <strong>and</strong> textural classification. Sample<br />
grain size distribution analysis showed that most<br />
of them are poorly sorted <strong>and</strong> bimodal (fine <strong>and</strong><br />
coarse modes present in the mixture), except the<br />
alluvium soil formation. Bulk density <strong>and</strong><br />
porosity of soil samples fall to a range of 1,53 to<br />
1,75 g/cm 3 <strong>and</strong> 0,20 to 0,35 respectively.<br />
Measured hydraulic conductivity at saturation<br />
ranges from 1,34 x 10 -4 to 4,5 x 10 -3 cm/s (Table<br />
1). Infiltration experiments were performed in 8<br />
measurement locations within the catchment by<br />
using a constant head disc infiltrometer. This<br />
method allows obtaining accurate estimates of<br />
field saturated hydraulic conductivity, k sat (cm/s).<br />
Water was supplied to the soil at a positive head<br />
by using a disc infiltrometer. It allows the air<br />
entrance to a reservoir (PVC tube with a 0,15 m<br />
diameter), with water liberation to a ponded soil<br />
surface. Steady state soil water flow is the result<br />
of integration of gravitational effect, constant<br />
ponded head influence <strong>and</strong> capillary forces. Two<br />
metallic rings were carefully inserted into the<br />
ground to a depth of approximately 3 cm without<br />
removing any natural vegetation. When<br />
measurement started, ground area between rings<br />
was ponded in order to prevent lateral flow<br />
because it misrepresents vertical infiltration). At<br />
each location, infiltration experiments were<br />
performed until steady state infiltration was<br />
attained. The time required for attaining this<br />
condition varied in function of soil hydraulic<br />
characteristics, with 40 minutes on average. At<br />
steady state condition, measured infiltration<br />
remained approximately constant in function of<br />
time. Measured curves showed variations in soil<br />
hydraulic behavior. In three experiments, it was<br />
verified a marked dispersion of infiltration<br />
capacity in function of time, indicating the<br />
existence of preferred flow pathways for<br />
infiltrating water (vertical <strong>and</strong> cylindrical pores)<br />
or local biological activity into the soil. The other<br />
five experiments resulted in a well-shaped<br />
downward curve before infiltration attained<br />
steady state condition <strong>and</strong> a horizontal<br />
configuration could be observed.<br />
4. RESULTS AND DISCUSSION<br />
Measured soil properties <strong>and</strong> hydraulic<br />
conductivity at field-saturated condition data was<br />
used in Hodnett <strong>and</strong> Tomasella (2002) <strong>and</strong><br />
Wösten et al. (2001) PTFs to obtain soil hydraulic<br />
parameters. Soil hydraulic parameters α, n, θ s , θ r<br />
<strong>and</strong> k sat obtained by using these PTFs are listed in<br />
Tables 2 <strong>and</strong> 3. The level of reliability of Hodnett<br />
<strong>and</strong> Tomasella (2002) PTF can be evaluated by<br />
comparing soil hydraulic parameters obtained<br />
from IGPB/T database with those parameters<br />
derived from using the same PTF <strong>and</strong> measured<br />
properties of soil samples collected within the<br />
catchment. For this purpose, the comparative<br />
analysis was made considering soil parameters<br />
derived for each soil texture class.<br />
In every case, the mean values of field bulk<br />
density were higher than the IGBP/T data set for<br />
each texture class, with differences ranging from<br />
17% (s<strong>and</strong>y clay) to 41% (loamy class).<br />
The mean α values derived by using Hodnett <strong>and</strong><br />
Tomasella (2002) PTF to the field soil samples<br />
were compared with those obtained for the<br />
IGBP/T data set. The differences between α<br />
values were less for loamy texture (12%) <strong>and</strong><br />
higher for clay (33%). Higher differences were<br />
observed for the θ r parameter, especially for<br />
s<strong>and</strong>y clay <strong>and</strong> clay classes, 43% <strong>and</strong> 56%<br />
respectively. For the mean n values, the<br />
differences were less, ranging from 1,3% to 10%.<br />
In a second stage, a comparison was made<br />
between Wösten et al. (2001) <strong>and</strong> Hodnett <strong>and</strong><br />
Tomasella (2002) PTFs by using measured soil<br />
sample data as input. There was a marked<br />
difference between soil parameters derived by<br />
these two functions, ranging from 43% to 98%.<br />
The mean α <strong>and</strong> θ s values are overestimated by<br />
Hodnett <strong>and</strong> Tomasella (2002) PTF for every<br />
class, with the higher difference for s<strong>and</strong>y loam<br />
class. Wösten et al. (2001) k sat parameter was<br />
compared to the field saturated hydraulic<br />
conductivity data for each soil class. These values<br />
were overestimated by Wösten et al. (2001)<br />
function for every class, ranging from 12% (clay)<br />
to 65% (loamy class). Marked differences<br />
between parameters obtained by these two PTFs<br />
indicate that soils from different climatic<br />
conditions show hydraulic differences that may be<br />
reflected in the derived parameters [Wösten et al.<br />
2001].<br />
993
Soil sample 1 2 3 4 5 6 7 8<br />
Bulk density<br />
(g/cm 3 )<br />
1.75 1.74 1.67 1.64 2.03 1.88 1.67 1.93<br />
Porosity 0.31 0.31 0.34 0.35 0.20 0.26 0.34 0.24<br />
q initial (cm 3 /cm 3 ) 0.029 0.034 0.025 0.063 0.054 0.079 0.02 0.066<br />
K sat (x10 -4 cm/s) 12 7.3 15 17 1.3 4 43 15<br />
Soil texture S<strong>and</strong>y loam S<strong>and</strong>y loam S<strong>and</strong>y clay S<strong>and</strong>y clay Loam S<strong>and</strong>y loam Clay S<strong>and</strong>y loam<br />
Table 1. Measured soil hydraulic properties at 8 locations within the catchment.<br />
Soil<br />
sample<br />
α n θ sat K sat<br />
A1 0.151 2.465 0.322 7.395<br />
A2 0.111 2.510 0.330 7.871<br />
A3 0.272 2.619 0.331 6.574<br />
A4 1.117 2.510 0.330 7.871<br />
A5 0.188 3.037 0.391 9.902<br />
A6 0.193 2.478 0.316 5.754<br />
A7 0.191 2.750 0.348 6.410<br />
Table 2. Soil hydraulics parameters obtained by<br />
Hodnett <strong>and</strong> Tomasella PTF (2002) application.<br />
Soil<br />
sample<br />
α n θ sat θ r<br />
A1 0.292 1.398 0.569 0.140<br />
A2 0.260 1.361 0.570 0.153<br />
A3 0.359 1.502 0.566 0.112<br />
A4 0.411 1.542 0.561 0.117<br />
A5 0.290 1.375 0.579 0.163<br />
A6 0.321 1.467 0.588 0.112<br />
A7 0.307 1.442 0.587 0.119<br />
A8 0.303 1.437 0.587 0.119<br />
Table 3. Soil hydraulics parameters obtained by<br />
Wösten et al. PTF (2001) application.<br />
0,6<br />
0,5<br />
-10 kPa<br />
-33 kPa<br />
-1500 kPa<br />
theta<br />
0,4<br />
0,3<br />
0,2<br />
0,1<br />
1,E-01 1,E+00 1,E+01 1,E+02 1,E+03 1,E+04 1,E+05 1,E+06<br />
Matric potential (cmH 2 O)<br />
A3 A1 A2<br />
A4 A5 A6<br />
A7 S<strong>and</strong>y Loam S<strong>and</strong>y clay<br />
Loam<br />
Clay<br />
Figure 2. Water release curves for field sample soils <strong>and</strong> IGPB/T soil data set<br />
994
Different geographic regions around the world<br />
exhibit important variations in structural soil<br />
characteristics that must be considered by PTFs.<br />
Regional specificity of a PTF can be clearly<br />
emphasized when results obtained by PTFs from<br />
tropical <strong>and</strong> temperate data sets are compared.<br />
Hodnett <strong>and</strong> Tomasella (2002) PTF <strong>and</strong> van<br />
Genuchten parameters generated from IGBP/T<br />
data for each textural class were used to construct<br />
the water release curves for the soil classes found<br />
in study area (Figure 5). Although it shows a<br />
reasonable agreement with the curves obtained for<br />
each soil sample, the scatter between curves from<br />
the same textural class reflects limitations for<br />
reproducing water release curve for these soils.<br />
This limitation may be due to some factors: a.<br />
functional differences between soils from IGBP/T<br />
data <strong>and</strong> soils from study area; b. specific soil<br />
hydraulic behavior in semiarid region areas that<br />
may need to be better understood on further<br />
investigations; c. the effect of the bimodality on<br />
soil hydraulic functioning was not considered; d.<br />
use of regression analysis to derive relationships<br />
has a limited capacity to represent a complex<br />
system.<br />
5. CONCLUSIONS<br />
IGBP/T data set used by Hodnett <strong>and</strong> Tomasella<br />
(2002) to derive a tropical PTF was composed of<br />
a large range of tropical soils. However, it was<br />
limited to regions represented by those data.<br />
IGBP/T data set didn’t cover all types of soil<br />
mineralogy of tropical regions. In this study, a<br />
comparison was made between soil parameters<br />
from IGBP/T data set <strong>and</strong> field samples within a<br />
study area located in a semi arid region. Results<br />
showed some important differences. Although<br />
differences were not significant when compared<br />
with Wösten et al. (2001) temperate PTF, they<br />
emphasize the specificity of semi arid regions<br />
soils behavior. The differences may be explained<br />
by some factors related to specificity on soil<br />
hydraulic functioning in these regions, accuracy<br />
level of relationships between soil properties <strong>and</strong><br />
predicted parameters, among other factors.<br />
Despite of these limitations, calculated<br />
differences between parameters showed a<br />
reasonable agreement with the same trend, what is<br />
shown in water release curves for field sample<br />
soils <strong>and</strong> IGPB/T soil data set shown in Figure 2.<br />
7. REFERENCES<br />
Brooks, R.H. <strong>and</strong> Corey, A.T., 1964. Hydraulic<br />
properties of porous media. Hydrologic paper<br />
Nº 3, Civil Engineering Dept., Colorado State<br />
University, Fort Collins, CO.<br />
Van Genuchten, M.Th., 1980. A closed form<br />
equation for predicting hydraulic<br />
conductivity in unsaturated soils. Soil Sci.<br />
Soc. Am., J. 44, 892-898.<br />
Tomasella, J., Hodnett, M.G., <strong>and</strong> Rossato, L.,<br />
2000. Pedotransfer functions for the<br />
estimation of soil water retention in Brazilian<br />
soils. Soil Sci. Soc. Am., J. 64, 327-338.<br />
Tomasella, J., Hodnett, M.G., 1998. Estimating<br />
soil water retention characteristics fron<br />
limited data in Brazilian Amazonia. Soil<br />
Science, 163, 190-202.<br />
Hodnett, M.G. <strong>and</strong> Tomasella, J., 2002. Marked<br />
differences between van Genuchten soil<br />
water retention parameters for temperate <strong>and</strong><br />
tropical soils: a new water retention<br />
pedotransfer functions developed for tropical<br />
soils. Geoderma, 108, 155-180.<br />
Wösten, J.H.M., Pachepsky, Ya. A. <strong>and</strong> Rawls,<br />
W.J., 2001. Pedotransfer functions: bridging<br />
the gap between available basic soil data <strong>and</strong><br />
missing soil hydraulic characteristics. Journal<br />
of Hydrology, 251, 123-150.<br />
Dong, W., Yu Z. <strong>and</strong> Weber, D., 2003.<br />
Simulations on soil water variation in arid<br />
regions. Journal of Hydrology, 275, 162-181.<br />
Wösten, J.H.M., Schuren, C.H.J.E., Bouma, J.<br />
<strong>and</strong> Stein, A., 1990. Functional sensitivity<br />
analysis of four methods to generate soil<br />
hydraulic functions. Soil Sci. Soc. Am., J. 54,<br />
832-836.<br />
Hodnett, M.G., Pimentel da Silva, L., da Rocha,<br />
H.R. <strong>and</strong> Cruz Senna, R., 1995. Seasonal soil<br />
water storage changes beneath central<br />
Amazonian rainforest <strong>and</strong> pasture. Journal of<br />
Hydrology, 170, 233-254.<br />
Richards, L.A., 1931. Capillary conduction of<br />
liquids through porous mediums. Physics, 1,<br />
318-333.<br />
6. ACKKONWLEDGEMENTS<br />
This research was supported by CT-<br />
HIDRO/FINEP Brazilian Government through<br />
the IBESA Project – Implementation of<br />
Experimental Catchment in Brazilian Semi Arid<br />
Region.<br />
995
An Approach for Calculating the Turbulent Transfer<br />
Coefficient Inside the Sparse Tall Vegetation<br />
D. T. Mihailovic a,d , M. Budincevic b , B. Lalic a,d <strong>and</strong> D. Kapor c,d<br />
a Faculty of Agriculture, Research Institute of Field <strong>and</strong> Vegetable <strong>and</strong> Crops,<br />
University of Novi Sad, 21000 Novi Sad, Serbia, guto@polj.ns.ac.yu<br />
b Department of Mathematics, Faculty of Natural Sciences, University of Novi Sad,<br />
21000 Novi Sad, Serbia<br />
c Department of Physics, Faculty of Natural Sciences, University of Novi Sad,<br />
21000 Novi Sad, Serbia<br />
d University Center for Meteorology <strong>and</strong> <strong>Environmental</strong> Modeling,<br />
University of Novi Sad, 21000 Novi Sad, Serbia<br />
Abstract: The sparse tall grass significantly affects the heat <strong>and</strong> moisture exchange in the lower<br />
atmosphere through the turbulent transfer coefficient inside its environment. Common approaches for<br />
calculation of turbulent transfer coefficient inside the tall grass environment are based on the assumption<br />
that it depends either on wind speed or mixing length inside the canopy. In this paper we suggested a new<br />
approach for calculating the turbulent transfer coefficient inside the sparse tall vegetation. In that sense we<br />
first derived an equation for the turbulent transfer coefficient inside the sparse tall grass using the<br />
“s<strong>and</strong>wich” approach for representation of vegetation, then we examined analytically whether its solution is<br />
always positive. Next, we solved the equation numerically using an iterative procedure for calculating the<br />
attenuation factor in the expression for the wind speed inside the canopy assumed to be a linear combination<br />
of an exponential <strong>and</strong> a logarithmic function. The proposed calculation of turbulent transfer coefficient is<br />
tested using the L<strong>and</strong>-Air Parameterization Scheme (LAPS). Model outputs of air temperature inside the<br />
canopy for 11-13 July 2002 are compared with micrometeorological measurements inside a sunflower field<br />
at the Rimski Sancevi experimental site (Serbia).<br />
Keywords: Turbulence inside the tall sparse vegetation; Turbulent transfer coefficient; Mixing length; L<strong>and</strong><br />
surface schemes; <strong>Environmental</strong> modeling<br />
1. INTRODUCTION<br />
Many complex environmental features at small,<br />
medium, or large scales involve processes that<br />
occur both within <strong>and</strong> between environmental<br />
media (e.g., air, surface water, groundwater, soil,<br />
biota). Considerable recent research work<br />
addresses various aspects of modeling these<br />
processes using new methodologies/approaches,<br />
numerical methods, <strong>and</strong> software techniques (e.g.,<br />
Walko et. al. 2000; Br<strong>and</strong>meyer <strong>and</strong> Karimi 2001;<br />
Lalic et al. 2003; Mihailovic et al. 2001; <strong>and</strong><br />
references herein). In environmental models,<br />
calculating turbulent fluxes inside <strong>and</strong> above a<br />
vegetation canopy requires the specification of air<br />
temperature, water vapor pressure, <strong>and</strong> turbulent<br />
transfer coefficients inside the canopy. The<br />
calculation of these quantities inside a tall grass<br />
canopy has been considered by many authors (e.g.,<br />
Sellers <strong>and</strong> Dorman [1987]; Sellers et al. [1986];<br />
Mihailovic et al. [1993]; Raupach et al. [1996]).<br />
Their work has remarkably improved the<br />
parameterization of turbulent fluxes inside tall<br />
grass canopies in l<strong>and</strong> surface schemes. However,<br />
there is not yet a complete approach for modeling<br />
the turbulent fluxes inside tall grass canopies,<br />
particularly in the case of sparse tall grass canopies<br />
(i.e., those in which the plant spacing is of the<br />
order of the canopy height or larger, Wyngaard<br />
[1988]). Such an approach is needed because tall<br />
grass canopies, through key variables like friction<br />
velocity <strong>and</strong> internal air temperature, can<br />
significantly affect heat <strong>and</strong> moisture exchange in<br />
the lower atmosphere.<br />
The objective of this paper is to suggest a new<br />
method for calculating the turbulent transfer<br />
coefficient inside tall grass canopies in l<strong>and</strong>atmosphere<br />
schemes for environmental modelling.<br />
996
̊<br />
̍<br />
with ̊<br />
̌<br />
as<br />
Section 2 includes derivation of (i) equation for<br />
wind profile inside the canopy in the “s<strong>and</strong>wich”<br />
approach (Section 2.1) <strong>and</strong> (ii) equation for<br />
turbulent transfer coefficient profile inside the<br />
sparse tall canopy (2.2). Section 3 is devoted to a<br />
numerical test. Section 3.1 includes description of<br />
numerical procedure for calculating the turbulent<br />
transfer coefficient inside the sparse vegetation;<br />
<strong>and</strong> (ii) numerical simulation of the air temperature<br />
inside a sunflower field for a three-day period,<br />
performed using a l<strong>and</strong> surface scheme, <strong>and</strong> its<br />
comparison with observations (Sections 3.2 <strong>and</strong><br />
3.3). Section 3.4 summarizes results.<br />
2. TURBULENT TRANSFER COEFFICIENT<br />
INSIDE THE TALL CANOPY<br />
2.1 Derivation of Equation for Wind Profile<br />
Inside the Canopy in the “S<strong>and</strong>wich”<br />
Approach<br />
Let us consider an element of the canopy volume<br />
having an area S <strong>and</strong> height H . The loss of air<br />
particles’ momentum due to close contact with the<br />
plant leaves comes from the drag force arising on<br />
the leaf surface. This drag force F d produces a<br />
shearing such that d / dz,<br />
the vertical gradient of<br />
shear stress, ̍, is equal to the drag force per<br />
volume V , i.e.,<br />
dz̍<br />
d F<br />
= d<br />
. (1)<br />
V<br />
The drag force per leaf unit area, S l , is proportional<br />
to the wind speed, u , i.e., the volumetric<br />
2<br />
kinetic energy 1/<br />
2 u the coefficient of<br />
proportionality C , the leaf drag coefficient. So,<br />
d<br />
F d 1 2<br />
= Cd<br />
u , (2)<br />
Sl<br />
2<br />
where ρ is the air density. Note that S l is the<br />
area of all leaves in the considered volume.<br />
Following the definition of leaf area index (LAI ) ,<br />
we can write LAI = Sl<br />
/( 2S)<br />
, since it is defined in<br />
terms of only one side of the leaf ( S is the ground<br />
surface covered with plants). Using<br />
τ = ρK s du / dz , Eqs. (1) <strong>and</strong> (2), <strong>and</strong> keeping in<br />
mind that the volume occupied by plants is<br />
S( H − h) , after some manipulation we arrive at<br />
d du Cd<br />
Ld<br />
( H − h)<br />
2<br />
K s =<br />
u , (3)<br />
dz dz H<br />
where z is the vertical coordinate, K<br />
s<br />
the<br />
turbulent transfer coefficient inside the canopy,<br />
L the area- averaged stem <strong>and</strong> leaf area density,<br />
d<br />
related to LAI as L d ( H − h)<br />
, while h is the<br />
canopy bottom height [Sellers et al., 1986;<br />
Mihailovic <strong>and</strong> Kallos, 1997].<br />
In a sparse tall grass canopy (one in which the<br />
plant spacing is of the order of the canopy height<br />
or larger), K s is strongly affected by processes in<br />
the environmental space, including the plants <strong>and</strong><br />
the space above the bare soil fraction. Therefore,<br />
Eq. (3) can be slightly modified taking into<br />
account fractional vegetation cover, σ (a measure<br />
of how sparse the tall grass is). The modified<br />
equation has the form<br />
d du Cd<br />
Ld<br />
( H − h)<br />
2<br />
K s = σ f<br />
u . (4)<br />
dz dz H<br />
In the case of dense vegetation ( σ = 1), Eq. (4)<br />
reduces to Eq. (3). Otherwise, when σ f = 0, Eq.<br />
(4) represents the turbulent transfer coefficient<br />
over bare soil. We use Eq. (4) to derive the<br />
equation for turbulent transfer coefficient inside<br />
the sparse tall canopy.<br />
2.2 Derivation of Equation for Turbulent<br />
Transfer Coefficient Profile Inside the<br />
Sparse Tall Canopy<br />
A number of assumptions are offered about the<br />
variation of Ks<br />
inside the canopy, as used in Eqs.<br />
(3) <strong>and</strong> (4). For example, according to Sellers et al.<br />
[1986] they are: (i) K s is proportional to the wind<br />
speed u , i.e, K s = σu<br />
with the length scale, ,<br />
an arbitrary constant; the data of Legg <strong>and</strong> Long<br />
[1975] <strong>and</strong> of Denmead [1976] qualitatively<br />
support this relationship that is often exploited in<br />
environmental modeling, (ii) Jarvis [1976] found<br />
that assumption Ks = Ks(H<br />
) is a representative<br />
one for the upper part of a coniferous canopy, <strong>and</strong><br />
2<br />
(iii) Ks = lmdu<br />
/ dz where l m is a mixing length.<br />
Instead to keep the length scale, σ constant, Lalic<br />
et al. [2003] assumed its dependence on z vertical<br />
coordinate, i.e., σ = σ (z)<br />
that can be obtained<br />
from an ordinary differential equation. Inadequacy<br />
of these approaches lies in the fact that the<br />
behavior of K s must be given a priori, i.e.<br />
presupposed by experience.<br />
f<br />
f<br />
In this paper we shall change the order of steps in<br />
calculation of the turbulent transfer coefficient<br />
inside the sparse vegetation, i.e. we shall solve Eq.<br />
(4) for Ks<br />
after assuming a functional form of<br />
solution for wind speed over the sparse tall<br />
vegetation, containing an attenuating parameter β<br />
997
˻<br />
̌<br />
̌<br />
˻ ˻<br />
̌ ̌<br />
̌<br />
̌<br />
that will be obtained iteratively. After taking the<br />
derivative of Eq. (4) over z , we obtain a<br />
differential equation of first order <strong>and</strong> first degree,<br />
where K is an unknown function, i.e.,<br />
du dK s<br />
dz dz<br />
s<br />
2<br />
d u<br />
K<br />
2 s =<br />
d z<br />
Cd<br />
Ld<br />
( H − h)<br />
2<br />
f<br />
u<br />
H<br />
+ . (5)<br />
Solution of this equation can be found if the wind<br />
speed is used to be a linear combination of two<br />
terms, expressing behavior of the wind speed over<br />
dense <strong>and</strong> sparse vegetation. Thus,<br />
u<br />
1 z <br />
β 1<br />
− <br />
2 H u*<br />
( z)<br />
= σ f u(<br />
H ) e<br />
<br />
+ (1 −σ<br />
f )<br />
−<br />
z<br />
ln , (6)<br />
k z<br />
where u(H<br />
) is the wind speed at the canopy<br />
height, β is an unknown constant to be determined,<br />
u * the friction velocity, k the von Karman<br />
constant supposed to be 0.41 <strong>and</strong> z g the roughness<br />
length over non-vegetated surface. The first term<br />
in the expression (6) is used because it fairly well<br />
approximates the wind profile within the dense tall<br />
grass canopy [Brunet et al., 1994; Mihailovic et<br />
al., 2004], while the second term simulates the<br />
shape of wind profile inside the tall sparse<br />
vegetation. After we introduce the expression (6)<br />
into Eq. (5), <strong>and</strong> rearrange it, we reach<br />
dKs + a( z)<br />
Ks<br />
= b(<br />
z)<br />
, (7)<br />
dz<br />
where<br />
H˻̌<br />
1 z − <br />
<br />
1<br />
− 1<br />
2<br />
2 H u 1<br />
f u(<br />
H ) e<br />
<br />
− (1 − f )<br />
*<br />
2<br />
a z H<br />
k<br />
2<br />
( ) = 4<br />
z (8)<br />
1 z <br />
1<br />
1<br />
− − <br />
2 H u 1<br />
f u(<br />
H ) e<br />
<br />
+ (1 − f )<br />
*<br />
2<br />
k z<br />
<strong>and</strong><br />
2<br />
⌈<br />
1 z − β 1<br />
⌉<br />
− <br />
2 H u*<br />
z<br />
b(<br />
z)<br />
=<br />
<br />
σ u( H ) e <br />
f<br />
( 1 σ f ) ln<br />
<br />
<br />
+ −<br />
k z <br />
×<br />
<br />
g<br />
⌊<br />
⌋<br />
Cd<br />
Ld<br />
( H − h)<br />
.(9)<br />
σ f<br />
H<br />
1<br />
z <br />
β 1<br />
1<br />
− − <br />
2 H<br />
u*<br />
1<br />
βσ f u(<br />
H ) e + (1 −σ<br />
f )<br />
2H<br />
k z<br />
Let us analyze the nature of the solution, K s , of<br />
the Eq. (6) with the initial condition defined as<br />
0<br />
K s ( z0 ) = K s > 0, where z 0 is some certain<br />
height inside the canopy: (i) The solution is unique<br />
<strong>and</strong> defined over the interval [ z 0,∞)<br />
, that follows<br />
from the fact that the functions a(z)<br />
<strong>and</strong> b(z)<br />
are<br />
defined <strong>and</strong> continuous over the interval indicated;<br />
(ii) The solution is positive, that comes from the<br />
analysis of the field of directions of the given<br />
equation or more precisely due to b ( z)<br />
> 0 ; (iii)<br />
The solution is stable that can be seen from the<br />
g<br />
following analysis. When z → ∞ we have<br />
a( z)<br />
≈ β /(2H<br />
) <strong>and</strong> b( z)<br />
≈ B exp[ βz<br />
/( 2H<br />
)]. Now, Eq.<br />
(7) takes the form<br />
βz<br />
dKs<br />
β<br />
K H<br />
s = Be 2<br />
+ , (10)<br />
dz 2H<br />
where<br />
2<br />
2<br />
2σ f u ( H ) Cd<br />
Ld<br />
( H − h)<br />
B = . (11)<br />
βH<br />
The particular solution of this equation has the<br />
form Aexp[ β z /( 2H<br />
)], where A is a constant, that<br />
can be obtained after replacing the particular<br />
solution in Eq. (10). If we follow this procedure we<br />
get A = BH / β . So, in this case, i.e., z → ∞ , the<br />
solution of Eq. (7) is asymptotically stable, it<br />
Aexp β z / 2H<br />
for any given A .<br />
behaves as [ ( )]<br />
3. NUMERICAL EXPERIMENTS<br />
To examine how successfully the foregoing<br />
proposed method for calculation of the turbulent<br />
transfer coefficient parameters support simulation of<br />
the air temperature within a tall grass canopy, a test<br />
was performed using the LAPS l<strong>and</strong> surface scheme<br />
described in Mihailovic [1996]. LAPS outputs of air<br />
temperatures inside the canopy for three days<br />
(11-13 July 2002) were compared with single-point<br />
micrometeorological measurements over a sunflower<br />
field at the Rimski Sancevi experimental site in<br />
Serbia. In the numerical experiments we used a<br />
data set from a measurement program that<br />
examined the exchange processes of heat, mass,<br />
<strong>and</strong> momentum just above <strong>and</strong> inside a sunflower<br />
canopy during its growing season.<br />
3.1 Numerical Procedure<br />
For the fixed β Eq. (7) can be solved using the<br />
finite-difference scheme<br />
n−1<br />
s<br />
n<br />
s<br />
n n<br />
{ b ( z)<br />
− a ( z K }<br />
K = K − ∆ z<br />
) , (12)<br />
where n is the number of the spatial step in the<br />
numerical calculating on the interval [ H , h],<br />
while<br />
∆ z is the grid size defined as ∆ z = ( H − h)<br />
/ N<br />
where N is a number indicating an upper limit in<br />
number of grid sizes used. The turbulent transfer<br />
coefficient calculation starts from the canopy top<br />
with an initial condition defined as<br />
N 2<br />
u(<br />
H )<br />
Ks ( H ) = k f ( H − d)<br />
+ k(1<br />
− u H<br />
H d<br />
f ) * (13)<br />
−<br />
ln<br />
zg<br />
then goes backward up to the canopy bottom<br />
height, h , that is defined according to Mihailovic<br />
et al. [2004]. To obtain parameter β we use an<br />
n<br />
s<br />
998
̌<br />
̅<br />
˻<br />
iterative procedure for this parameter that is not<br />
finished until the condition<br />
N N<br />
k + 1 k<br />
∑ui − ∑ui<br />
<<br />
(14)<br />
i=<br />
1 i=<br />
1<br />
is reached, where k is number of iteration while<br />
µ is less then 0.001. Having this parameter we<br />
can calculate the wind profile on the interval [ H , h]<br />
according to the expression (6). Beneath the<br />
canopy bottom height, the wind profile has the<br />
logarithmic shape [Sellers et al., 1986; Mihailovic<br />
et al., 2004], i.e.,<br />
⌈<br />
<br />
u(<br />
z)<br />
= u(<br />
H ) <br />
<br />
<br />
⌊<br />
1 h <br />
− 1<br />
− <br />
2 H<br />
f e<br />
<br />
1−<br />
<br />
f z<br />
+ ln . (15)<br />
Ȟ zg<br />
ln<br />
h<br />
z<br />
g<br />
ln<br />
z<br />
g<br />
⌉<br />
<br />
⌋<br />
while the area-averaged bulk boundary layer<br />
resistance, r b , has the form [Sellers et al., 1986]<br />
1<br />
=<br />
r<br />
b<br />
H<br />
∫<br />
h<br />
a<br />
Ld<br />
u(<br />
z)<br />
dz , (19)<br />
C P<br />
s<br />
s<br />
where C s is the transfer coefficient [Sellers et al.,<br />
1986] <strong>and</strong> P s the leaf shelter factor. The values for<br />
these parameters were taken from Mihailovic <strong>and</strong><br />
Kallos [1997]. Eqs. (17)-(19) can be modified to<br />
take into account the effects of nonneutrality.<br />
According to Sellers et al. [1986], the position of<br />
the canopy source height, h a , can be estimated by<br />
obtaining the center of gravity of the 1 / r integral.<br />
3.3 Experimental Site <strong>and</strong> Details<br />
b<br />
3.2 Calculating the Air Temperature Inside<br />
the Sunflower Canopy<br />
The temperature inside the sunflower air space,<br />
T a , was determined diagnostically from the energy<br />
balance equation. This procedure comes from the<br />
equality of the sensible heat flux from the canopy<br />
to some reference level in the atmosphere, <strong>and</strong> the<br />
sum of the sensible heat fluxes from the ground<br />
<strong>and</strong> from the leaves to the canopy air volume<br />
[Sellers et al., 1986; Mihailovic, 1996], i.e.,<br />
2T<br />
f Tg<br />
Tr<br />
+ +<br />
rb<br />
rd<br />
r<br />
T<br />
a<br />
a = , (16)<br />
2 1 1<br />
+ +<br />
rb<br />
rd<br />
ra<br />
where T f is the foliage temperature, T g the<br />
ground surface temperature, T r the temperature at<br />
reference level, r b the bulk boundary-layer<br />
aerodynamic resistance, r d the aerodynamic<br />
resistance to water vapor <strong>and</strong> heat flow from the<br />
soil surface to air space inside the canopy, <strong>and</strong> r a<br />
the aerodynamic resistance representing the<br />
transfer of heat <strong>and</strong> moisture from the canopy to<br />
the reference level, z r .<br />
The aerodynamic resistance r a between z r <strong>and</strong> h a ,<br />
the water vapor <strong>and</strong> sensible heat source height<br />
[Sellers et al., 1986], can be defined as<br />
H zr<br />
1 1<br />
ra<br />
= dz + dz . (17)<br />
∫ K ∫<br />
s K<br />
ha<br />
H<br />
s<br />
The aerodynamic resistance in canopy air space,<br />
r d , can be written in the form<br />
h<br />
1<br />
rd<br />
= dz +<br />
∫ K<br />
z<br />
g<br />
s<br />
h<br />
a<br />
∫<br />
h<br />
1<br />
dz , (18)<br />
K<br />
s<br />
The experimental site (270 m x 68 m) is located in<br />
the northern part of Serbia (45.3°N, 19.8°E) on a<br />
chernozem soil of the loess terrace of southern<br />
Backa with the following physical <strong>and</strong> water<br />
properties: Clapp-Hornberger constant “B”: 6.50;<br />
ground emissivity: 0.97; heat capacity of the soil<br />
fraction: 780 J kg -1o C -1 ; saturated hydraulic conductivity:<br />
32x10 -6 m s -1 ; soil moisture potential at<br />
saturation: -0.036 m; soil density: 1290 kg m -3 ;<br />
ratio of saturated thermal conductivity to that of<br />
loam: 1.0; volumetric soil moisture content at<br />
saturation: 0.52 m 3 m -3 ; volumetric soil moisture<br />
content at field capacity: 0.36 m 3 m -3 ; wilting point<br />
volumetric soil moisture content: 0.17 m 3 m -3 ; <strong>and</strong><br />
effective ground roughness length: 0.01 m. The<br />
experimental site was surrounded by other<br />
agricultural fields also sown with sunflowers. The<br />
sunflower rows were oriented north to south, with<br />
row spacing of 0.70 m. This data set was chosen<br />
because it was considered typical <strong>and</strong> representative<br />
of a fully developed sunflower crop. For the<br />
11-13 July period the mean estimated LAI was 3.0<br />
m 2 m -2 ; the crop height, H , was around 1.99 m;<br />
<strong>and</strong> the canopy bottom height, h , was 0.100 m.<br />
The extinction factor, β , was calculated using a<br />
numerical procedure that is described in the<br />
previous subsection. While the zero plane<br />
displacement, d , <strong>and</strong> roughness length, z 0 , were<br />
calculated according to Mihailovic <strong>and</strong> Kallos<br />
[1997]. The scaling length, σ , was derived<br />
following Mihailovic et al. [2004]. In these<br />
calculations the area-averaged stem <strong>and</strong> leaf area<br />
density, L d , had a value of 1.59 m 2 m -3 , while a<br />
value of 0.2 was used for the leaf drag coefficient,<br />
C d . Since the minimum stomatal resistance was<br />
not measured, we assumed it to be 40 s m -1 . The<br />
fractional vegetation cover was 0.90. Other<br />
parameters used in the simulation can be found in<br />
Mihailovic et al. [2000]. Canopy source height,<br />
h , was calculated following Mihailovic et al.<br />
a<br />
999
[2004]. Using the above parameter values, we obtained a value of 1.1 m.<br />
40<br />
Air temperature in canopy T a<br />
( o C)<br />
30<br />
20<br />
10<br />
0<br />
Simulated (thin)<br />
Observed (thick)<br />
192 193 194 195<br />
Julian day<br />
Figure 1. The comparison of three-day variation (11-13 July 2002) of the air temperature simulated by LAPS<br />
<strong>and</strong> observed inside a sunflower canopy at the Rimski Sancevi site.<br />
Temperatures were measured using platinum<br />
resistance thermometers (Pt-100) set at 0.95 <strong>and</strong><br />
2.1 m above the ground. The wind speed at the<br />
reference level of z r = 2.1 m was measured using<br />
a Vector Instruments anemometer. A Kipp Zonen<br />
CM5 solarimeter was used to measure incoming<br />
solar radiation, while relative humidity was<br />
recorded using a Greisinger sensor set at 2.1 m.<br />
Precipitation was measured by an electronic rain<br />
gauge manufactured at the Institute of Physics in<br />
Belgrade. Soil temperature was measured at 0.05-,<br />
0.1-, <strong>and</strong> 0.2-m depths. In all data sets, the<br />
atmospheric boundary conditions at z r = 2.1 m<br />
were derived from measurements of global<br />
radiation, precipitation, relative humidity, <strong>and</strong><br />
wind for 24 hours from 0000 LST at 30-min<br />
intervals. The longwave atmospheric counter--<br />
radiation was calculated via an empirical formula<br />
described in Mihailovic et al. [1995], including a<br />
correction for the amount of cloudiness.<br />
Cloudiness data were taken at 30-min intervals<br />
from the nearest st<strong>and</strong>ard meteorological station,<br />
Rimski Sancevi, which is 500 m away from the<br />
experimental site. These values were interpolated<br />
to the beginning of each time step ( ∆ t = 120 s).<br />
The thicknesses of soil layers were defined as<br />
D 1 = 0-0.1 m, D 2 = 0.1-0.5 m, <strong>and</strong> D 3 = 0.5-1 m.<br />
The initial conditions for the volumetric soil<br />
moisture contents corresponding to these layers<br />
were w 1 = 0.1552 m 3 m -3 , w 2 = 0.1484 m 3 m -3 ,<br />
<strong>and</strong> w 3 = 0.1348 m 3 m -3 . At the initial time the<br />
ground temperature was 292.68 K. The initial<br />
condition for atmospheric pressure was<br />
100.53 kPa.<br />
3.4 Comments <strong>and</strong> Further Plans<br />
The validity of the LAPS-simulated air<br />
temperature inside the canopy was tested against<br />
the observations recorded by the platinum<br />
resistance thermometer located at 0.95 m at 30-min<br />
intervals during 11-13 July 2002. Figure 1 shows<br />
the calculated <strong>and</strong> observed diurnal variations of<br />
air temperature inside the sunflower canopy at the<br />
experimental site. After midnight, the simulated<br />
values are lower than the observations, while in the<br />
early afternoon the simulated values are slightly<br />
higher than the observed ones. This situation<br />
occurs because at night LAPS simulates less heat<br />
transfer from the ground into the canopy air space<br />
than the observations indicate. In contrast, during<br />
the afternoon, the scheme calculates a lower<br />
amount of evapotranspiration, which for some<br />
days results in a higher leaf temperature <strong>and</strong><br />
consequently a higher air temperature inside the<br />
sunflower canopy [Eq. (17)]. Apparently, the<br />
calculation of air temperature inside the tall grass<br />
canopy strongly depends on the resistances given<br />
by Eqs. (17)-(19), i.e., on the resistances’<br />
sensitivity to morphological <strong>and</strong> aerodynamic<br />
parameters, which can be sources of uncertainty in<br />
their calculation. In the future we plan to check the<br />
1000
proposed method using more specific tests<br />
including three-dimensional simulations.<br />
Acknowledgments<br />
This research was supported by the New York<br />
State Energy Research <strong>and</strong> Development Authority<br />
under contractual agreement NYSERDA 4914-<br />
ERTER-ES-99. The research was also supported<br />
by the Serbian Ministry for Science <strong>and</strong><br />
Technology under contracts BTR.S.02.0401.B <strong>and</strong><br />
BTR.S.02.0427.B. The investigators would like to<br />
thank Dr. S.T. Rao for his support.<br />
REFERENCES<br />
Br<strong>and</strong>meyer, J., <strong>and</strong> H.A. Karimi, Coupling<br />
methodologies for environmental models,<br />
Environ. Modeling & <strong>Software</strong>, 15(5), 479-<br />
488, 2001.<br />
Brunet, Y., J.J. Finnigan, <strong>and</strong> M.R. Raupach, A<br />
wind tunnel study of air flow in waving<br />
wheat: Single-point velocity statistics,<br />
Boundary-Layer Meteorol., 70, 95-132,<br />
1994.<br />
Denmead, O.T., Temperate cereals. In: Vegetation<br />
<strong>and</strong> the atmosphere- 2nd., J. L. Monteith<br />
(Ed.), Academic Press, pp. 1-31, New<br />
York, 1976.<br />
Jarvis, P.G., The interpretation of leaf water<br />
potential <strong>and</strong> stomatal conductance found in<br />
canopies in the field, Phil. Trans. R. Soc.<br />
London, Ser. B, 273, 593-610, 1976.<br />
Lalic, B., D.T Mihailovic, B. Rajkovic, I.D.<br />
Arsenic, <strong>and</strong> D. Radlovic, Wind profile<br />
within the forest canopy <strong>and</strong> in the<br />
transition layer above it, Environ. Modeling<br />
& <strong>Software</strong>, 18, 947-950, 2003.<br />
Legg, B. J., <strong>and</strong> I. F. Long, Turbulent diffusion<br />
within a wheat canopy II, Quart. J. Roy.<br />
Meteor. Soc., 101, 611-628, 1975.<br />
Mihailovic, D.T., Description of a l<strong>and</strong>-air<br />
parameterization scheme (LAPS), Global<br />
Planet. Change, 13, 207-215, 1996.<br />
Mihailovic, D.T., <strong>and</strong> G. Kallos, A sensitivity<br />
study of a coupled soil-vegetation boundary<br />
layer scheme for use in atmospheric<br />
modeling, Boundary-Layer Meteorol., 82,<br />
283-315, 1997.<br />
Mihailovic, D.T., B. Rajkovic, B. Lalic, <strong>and</strong> LJ.<br />
Dekic, Schemes for parameterizing<br />
evaporation from a non-plant-covered surface<br />
<strong>and</strong> their impact in on partitioning the surface<br />
energy in l<strong>and</strong>-air exchange parameterization,<br />
J. Appl. Meteor., 34, 2462-2475, 1995.<br />
Mihailovic, D.T., R.A. Pielke, B. Rajkovic, T.J.<br />
Lee, <strong>and</strong> M. Jeftic, A resistance<br />
representation of schemes for evaporation<br />
from bare <strong>and</strong> partly plant-covered surfaces<br />
for use in atmospheric models, J. Appl.<br />
Meteor., 32, 1038-1054, 1993.<br />
Mihailovic, D.T, I. Koci, B. Lalic, I. Arsenic, D.<br />
Radlovic, <strong>and</strong> J. Balaz, The main features of<br />
BAHUS-biometeorological system for<br />
messages on the occurrence of diseases in<br />
fruits <strong>and</strong> vines, Environ. Modeling &<br />
<strong>Software</strong>, 16(8), 691-696, 2001.<br />
Mihailovic, D.T., K. Alapaty, B. Lalic, I. Arsenic, B.<br />
Rajkovic, <strong>and</strong> S. Malinovic, Turbulent<br />
transfer coefficients <strong>and</strong> calculation of air<br />
temperature inside tall grass canopies in<br />
l<strong>and</strong>-atmosphere schemes for environmental<br />
modelling, J. Appl. Meteor., 2004. (In<br />
revision)<br />
Mihailovic, D.T., T.J. Lee, R.A., Pielke, B. Lalic,<br />
I. Arsenic, B. Rajkovic, <strong>and</strong> P.L Vidale,<br />
Comparison of different boundary layer<br />
schemes using single point micrometeorological<br />
field data, Theor. Appl.<br />
Climatol., 67, 135-151, 2000.<br />
Raupach, M.R., J.J. Finnigan, <strong>and</strong> Y. Brunet,<br />
Coherent edies <strong>and</strong> turbulence in vegetation<br />
canopies: The mixing-layer analogy,<br />
Boundary-Layer Meteorol., 78, 351-382,<br />
1996.<br />
Sellers, P.J., <strong>and</strong> J.L Dorman, Testing the simple<br />
biosphere model (SiB) using point<br />
micrometeorological <strong>and</strong> biophysical data,<br />
J. Clim. Appl. Meteorol., 26, 622-651, 1987.<br />
Sellers, P. J., Y. Mintz, Y. Sud, A. Dalcher, A<br />
simple biosphere model (SiB) for use within<br />
general circulation model, J. Atmos. Sci.,<br />
43, 506-531, 1986.<br />
Walko, R. L., <strong>and</strong> Coauthors, Coupled<br />
atmosphere-biophysics-hydrology models<br />
for environmental modeling, J. Appl.<br />
Meteor., 39, 931-944, 2000.<br />
Wyngaard, J.C., Convective processes in the lower<br />
atmosphere, In: Flow <strong>and</strong> transport in the<br />
natural environment: Advances <strong>and</strong><br />
applications, W.L. Steffen <strong>and</strong> O.T.<br />
Denmead (Eds.), Springer, Berlin, 240-260,<br />
1988.<br />
1001
The Utility of GIS Delivered Environment Models in<br />
the Characterisation of Surface Water Bodies under<br />
the Water Framework Directive Low Flows 2000 – a<br />
Case Study.<br />
T. H. Goodwin, M. Fry, M. G. R. Holmes, A. R. Young<br />
Centre for Ecology <strong>and</strong> Hydrology (CEH) - Wallingford, Maclean Building, Crowmarsh Gifford.<br />
Wallingford, Oxfordshire, OX10 8BB.<br />
Abstract: The implementation of the Water Framework Directive (WFD), requires Water Resource<br />
Managers to characterise <strong>and</strong> assess the status of surface water bodies <strong>and</strong> apply management practices<br />
in order to achieve Good Ecological Status (GES), at a national scale. GES is assessed by considering<br />
the biological, hydromorphological <strong>and</strong> physico-chemical characteristics of a water body. There is a<br />
need for a suite of tools which will aid in carrying out this assessment quickly <strong>and</strong> reliably at a large<br />
number of sites. Low Flows 2000 (LF2000), combines aquatic environmental <strong>and</strong> hydrological models<br />
within a user-friendly GIS interface. The software can be used to estimate both the natural statistical<br />
properties of river flow <strong>and</strong> the influenced statistical properties, through the integration of water use<br />
pressures relating to abstractions, discharges <strong>and</strong> flow regulation, at any point within a river system<br />
without recourse to calibration. By combining this functionality with a measure of the aquatic<br />
ecological sensitivity to water use pressures LF2000 is being used operationally by the statutory UK<br />
<strong>Environmental</strong> regulators to support the initial characterisation of water use pressures on surface water<br />
bodies under the WFD.This paper will discuss how LF2000 is being used in this context <strong>and</strong> concludes<br />
with a look to the future. A summary of a prototype water quality modelling extension to LF2000 that<br />
has been developed will be presented alongside the planned implementation of generalised rainfall<br />
runoff models. This will illustrate how LF2000 will provide a suite of tools which can contribute to<br />
both initial <strong>and</strong> further characterisation under the WFD <strong>and</strong> prioritisation of a programme of measures.<br />
Keywords: Water Framework Directive, Water Resource Management, Pressures, Hydrological<br />
Models.<br />
1 INTRODUCTION<br />
The Water Framework Directive (WFD)<br />
[2000/60/EC] requires that all water bodies<br />
within member states must achieve or maintain<br />
‘Good Ecological Status’ (GES) by 2015. GES<br />
is assessed by considering the biological,<br />
physico-chemical, hydromorphological <strong>and</strong><br />
characteristics of a water body.<br />
The first stage in this process is the<br />
identification of water bodies which are at risk<br />
of not achieving, or maintaining GES,<br />
hereafter referred to as ‘at risk’. The initial<br />
characterisation <strong>and</strong> identification of these<br />
water bodies is required to be completed by the<br />
end of 2004. Monitoring programmes will then<br />
be established for the water bodies considered<br />
to be ‘at risk’. Following this, programmes of<br />
measures will be implemented.<br />
This paper will focus on the use of Low Flows<br />
2000 (LF2000) to characterise the surface<br />
water pressures which will, combined with<br />
other information, be fed into a risk assessment<br />
methodology in order to identify surface water<br />
bodies which are ‘at risk’, related to<br />
hydromorphological factors.<br />
Within the UK the regulatory body for<br />
Engl<strong>and</strong> <strong>and</strong> Wales, the Environment Agency,<br />
has a wide remit which includes the<br />
management of water resources through<br />
integrated management. Legislation, relating to<br />
the authorisation of abstraction <strong>and</strong> discharge<br />
pressures has existed since the Water<br />
Resources Act of 1963 <strong>and</strong> the Control of<br />
1002
Pollution Act 1974. The Water Act of 2003<br />
updated the aforementioned Acts, <strong>and</strong> the<br />
subsequent Acts relating to them, to be inline<br />
with EU legislation <strong>and</strong> reflect changes in<br />
water resource issues. This legislation allows<br />
the Environment Agency to issue abstraction<br />
licenses <strong>and</strong> consents to discharge <strong>and</strong><br />
facilitates the management <strong>and</strong> control of<br />
pressures within catchments.<br />
In line with the requirements of the WFD the<br />
Environment Agency is moving towards<br />
integrated catchment management. This has<br />
led to the development of Catchment<br />
Abstraction Management Strategies (CAMS)<br />
[Environment Agency, 2002a]. Initiated in<br />
April 2001, the development of CAMS for all<br />
strategic water management units, 129<br />
catchments, within Engl<strong>and</strong> <strong>and</strong> Wales will be<br />
completed by 2008. Within CAMS<br />
information on the pressures within the<br />
catchment is used to establish the resource<br />
status of the catchment. An assessment of the<br />
ecological requirements of the water body is<br />
then used to determine the status of the<br />
catchment with respect to the water resource<br />
requirements. Consultation with local<br />
stakeholders leads to the development of a<br />
catchment scale abstraction licensing strategy.<br />
The CAMS strategy encompasses many of the<br />
overarching hydromorphological concepts of<br />
the WFD. However, the requirement of the<br />
WFD to identify all water bodies ‘at risk’ by<br />
the end of 2004 has provided an additional<br />
challenge to the Environment Agency that the<br />
implementation of current management<br />
strategies is not able to address.<br />
The initial characterisation <strong>and</strong> identification<br />
of ‘at risk’ water bodies requires a large<br />
number of assessments to be made in a rapid<br />
<strong>and</strong> consistent manner. The primary tool used<br />
within the process of identifying surface<br />
waters under pressure resulting from<br />
abstractions, discharges <strong>and</strong> flow regulation<br />
has been LF2000.<br />
2 LF2000 - BACKGROUND<br />
LF2000 [Young et al. 2003] consists of<br />
environmental hydrological models within a<br />
GIS based framework. The PC based software<br />
has the ability to estimate natural <strong>and</strong><br />
influenced flow statistics on any river on the<br />
1:50000 network, within Engl<strong>and</strong> <strong>and</strong> Wales.<br />
Catchments are defined using a digital terrain<br />
model or an analogue approach, whereby grid<br />
cells are assigned to river reaches on a nearest<br />
neighbour basis. The catchment is used to<br />
derive catchment characteristics such as the<br />
runoff <strong>and</strong> hydrogeological characteristics.<br />
These are used to estimate the flow variability,<br />
represented by the relevant flow duration curve<br />
(FDC), at an annual <strong>and</strong> monthly resolution.<br />
2.1 Regionalised Hydrological Models<br />
The hydrological models which underpin<br />
LF2000 consist of a regionalised model to<br />
estimate the natural temporal variability of<br />
flows <strong>and</strong> a regionalised rainfall-runoff model.<br />
At the catchment scale, hydrogeology is the<br />
dominant factor in determining the temporal<br />
variability of natural flows as represented by<br />
FDC, once normalised for size <strong>and</strong><br />
climatology. For example, flows are less<br />
variable in base-flow dominated chalk<br />
catchments than within impermeable clay<br />
catchments. The regionalised model to<br />
estimate the normalised flow variability uses a<br />
Region Of Influence (ROI) approach, whereby<br />
a st<strong>and</strong>ardised annual or monthly FDC is based<br />
on observed data from a selected data pool<br />
[Holmes et al., 2002b]. The datapool is derived<br />
using similarity measures related to the<br />
distribution of HOST classes, a hydrologically<br />
based soil classification system, to derive<br />
catchment similarity [Boorman et al., 1995].<br />
The st<strong>and</strong>ardised FDC is then rescaled using a<br />
value of runoff, estimated using historical<br />
rainfall <strong>and</strong> PE data together with a soil<br />
moisture model, to predict the FDC in units of<br />
cubic metres per second. [Holmes et al.,<br />
2002a].<br />
2.2 LF2000 – Pressure Information<br />
In addition to the hydrological models used to<br />
assess natural flow conditions, the LF2000<br />
database, based on the CEH Water Information<br />
System (WIS), [Moore, 1997] allows pressure<br />
information to be incorporated into the system.<br />
The pressure information corresponding to<br />
abstractions, discharges <strong>and</strong> impoundments is<br />
stored within the database as 12 monthly<br />
volumes.<br />
The 12 monthly volumes for an abstraction<br />
represent the average monthly volumes of<br />
water abstracted from the surface water body<br />
or groundwater unit. User-defined<br />
Transmissivity <strong>and</strong> Storativity values are used<br />
within an analytical solution to the Theis<br />
equation to determine the impact at the surface<br />
1003
water body of groundwater abstractions<br />
[Bullock et al., 1994].<br />
The monthly volumes for discharges represent<br />
average monthly volumes of water discharged<br />
directly to the surface water body, excluding<br />
stormwater runoff that may be intercepted by<br />
sewer systems.<br />
For impoundments the 12 monthly volumes<br />
represent the average monthly compensation<br />
<strong>and</strong>/or regulated release volumes. LF2000<br />
incorporates the impact of impoundments by<br />
omitting the catchment area above the<br />
impoundment from the natural flow estimation<br />
procedure <strong>and</strong> adding the compensation or<br />
regulated volumes to the resultant natural flow<br />
regime [Bullock et al., 1994].<br />
In application, the pressures within the target<br />
catchment area are first identified. The relevant<br />
pressure information is then used in<br />
conjunction with the natural monthly FDC to<br />
produce estimates of the influenced monthly<br />
FDC. These are aggregated to produce an<br />
influenced annual FDC.<br />
3 LF2000 – THE WFD<br />
The Environment Agency has used LF2000 as<br />
part of the CAMS process [Young et al., 2003]<br />
<strong>and</strong> as part of the st<strong>and</strong>ard licensing procedure.<br />
A modified version of LF2000 has been an<br />
integral part of the methodology developed to<br />
implement the first stage of the WFD; the<br />
characterisation <strong>and</strong> identification of surface<br />
water bodies which are at risk due to surface<br />
water pressures.<br />
3.1 Characterising pressures on Surface<br />
Water Bodies within Engl<strong>and</strong> <strong>and</strong><br />
Wales.<br />
The main objective was to characterise the<br />
natural <strong>and</strong> modified flow regimes of over<br />
6000 surface water Assessment Points (AP)<br />
within Engl<strong>and</strong> <strong>and</strong> Wales. The data from this<br />
assessment is used subsequently, with a<br />
measure of ecological sensitivity, as input to<br />
the risk assessment process to assess whether<br />
the water body is ‘at risk’. An assessment of<br />
this risk is also estimated using the pressure<br />
scenarios predicted for 2015.<br />
3.2 Method<br />
Pressure Assessment<br />
Over 6000 locations within the UK were<br />
designated as APs at which the risk of not<br />
achieving GES should be determined. These<br />
were defined, using the guidance output from<br />
the working group of the common<br />
implementation strategy, on the basis of<br />
typology, geographic features <strong>and</strong> ecological<br />
<strong>and</strong> chemical conditions.<br />
An estimate of the natural flow regime was<br />
made, using LF2000, at each AP. Pressure<br />
information, which included abstractions,<br />
discharges <strong>and</strong> impoundments, was then<br />
incorporated within the LF2000 database <strong>and</strong><br />
the impact of this on the natural flow regime<br />
assessed. A number of flow statistics were<br />
used for this assessment, for example, the<br />
difference in the flow regime at exceedence<br />
percentiles of Q95, Q70 <strong>and</strong> Q50.<br />
Ecological Sensitivity <strong>and</strong> Risk Assessment.<br />
The output data from LF2000 was<br />
subsequently used by the Environment<br />
Agency, together with the ecological<br />
sensitivity, as part of the risk assessment<br />
procedure. A brief description of the way in<br />
which the ecological assessment was made,<br />
<strong>and</strong> how this was used with the results from<br />
LF2000 follows.<br />
The ecological model used to define the<br />
ecological sensitivity of each surface water<br />
body to flow derogation resulting from<br />
pressures, which is not the subject of this<br />
paper, was based on the Lotic Invertebrate<br />
Index for Flow Evaluation (LIFE) score<br />
[Extence et al., 1999]. The LIFE score gives an<br />
indication of the types of ecological<br />
community present within a river reach. The<br />
relationship between LIFE score <strong>and</strong> flows<br />
means that it can give an indication of the<br />
ecological sensitivity to changes in the flow<br />
regime. The LIFE score can be estimated at<br />
any site based on empirical relationships with<br />
the physical <strong>and</strong> chemical characteristics<br />
within natural catchments. Each AP was<br />
assigned the LIFE score estimated at the<br />
nearest General Quality Assessment site, at<br />
which the Environment Agency regularly<br />
assess the quality of the water body. The<br />
ecological sensitivity was then derived from<br />
the estimated LIFE score using the criteria<br />
outlined within the technical framework used<br />
for the implementation of CAMS<br />
[Environment Agency 2002b].<br />
Within the risk assessment procedure the<br />
information on the impact of pressures on the<br />
natural flow regime, together with the<br />
1004
ecological sensitivity at each AP is combined<br />
to provide an indication of whether a surface<br />
water body is ‘at risk’. It is probable that<br />
within an ecologically very sensitive river this<br />
risk will be high whatever the impact of<br />
pressures. Similarly an AP with a high impact<br />
of pressures is likely to be ‘at risk’ whatever<br />
the ecological sensitivity of the reach.<br />
This screening methodology identifies water<br />
bodies which are ‘at risk’ <strong>and</strong> allows<br />
subsequent monitoring to be carried out<br />
strategically <strong>and</strong> efficiently.<br />
3.3 Pressure Assessment - The role of<br />
LF2000 in the risk assessment<br />
process<br />
LF2000 is an integral part of this screening<br />
assessment as it has been used to provide<br />
estimates of the natural <strong>and</strong> influenced FDC at<br />
a large number of points within Engl<strong>and</strong> <strong>and</strong><br />
Wales.<br />
Within Engl<strong>and</strong> <strong>and</strong> Wales the Environment<br />
Agency has authorised over 40,000 abstraction<br />
licenses <strong>and</strong> over 86,000 discharge consents.<br />
There are also approximately 2250<br />
impoundments. It is therefore no small task to<br />
characterise <strong>and</strong> model the surface water<br />
pressures within Engl<strong>and</strong> <strong>and</strong> Wales.<br />
Intelligent filtering of abstraction licenses,<br />
whereby only those for which the abstraction<br />
volumes are greater than 5% of the natural<br />
Q95 at the AP are assumed to be significant,<br />
reduced the number of abstractions to be<br />
characterised to 9000. Annual abstraction<br />
volumes for 2001 were available nationally.<br />
These were distributed throughout the year<br />
based on the seasonal patterns within higher<br />
resolution data from example licenses.<br />
Discharges are poorly quantified as volumetric<br />
data does not commonly form part of the<br />
consent compliance checking. The explicit<br />
representation of discharges was therefore<br />
restricted to Sewage Treatment Works.<br />
Industrial consents were included implicitly by<br />
applying percentage returns to abstractions.<br />
Information on impoundment releases is sparse<br />
<strong>and</strong> not collated nationally. There was no<br />
measured data on which to base the monthly<br />
values for just over 200 of the impoundments<br />
considered. Gustard et al., [1987] summarised<br />
the compensation flows of all impoundments<br />
with the UK with capacities greater than<br />
500ML. For the impoundments without<br />
measured data generic rules were developed to<br />
relate the compensation values provided to<br />
monthly release volumes. An assessment of<br />
this method indicated that compensation<br />
releases tended to be overestimated.<br />
3.4 Results<br />
The results of the analysis, together with<br />
assessments of ecological sensitivity are being<br />
used by the Environment Agency to aid in the<br />
initial characterisation of surface water bodies<br />
<strong>and</strong> the assessment of the pressures <strong>and</strong><br />
ecological impacts within river basins to<br />
identify AP ‘at risk’. At the time of writing it is<br />
understood that the results will not be within<br />
the public domain prior to the submission of<br />
these reports to the EU in Autumn 2004.<br />
A preliminary indication of the ability of<br />
LF2000 to achieve the objectives of the<br />
analysis i.e to estimate the pressure impact at<br />
each AP, was achieved by comparing the<br />
natural <strong>and</strong> influenced LF2000 estimated FDC<br />
with flows at gauging stations. An example is<br />
illustrated within Figure 1 for Q95. This<br />
illustrates that the LF2000 influenced Q95<br />
estimates provide a significantly improved<br />
estimate of the gauged Q95 over the LF2000<br />
natural Q95 estimate. This indicates that the<br />
methodology is effectively representing the<br />
water use patterns, hence the impact of<br />
pressures, within the gauged catchments.<br />
Figure 1. LF2000 estimates of natural <strong>and</strong><br />
influenced Q95 values relative to gauged<br />
values of Q95.<br />
3.5 Summary<br />
LF2000 is a tool by which a rapid assessment<br />
of the natural <strong>and</strong> modified flow estimates at a<br />
large number of sites across Engl<strong>and</strong> <strong>and</strong><br />
Wales can be made. This has enabled the first<br />
stage of the WFD, the characterisation <strong>and</strong><br />
subsequent identification of surface water<br />
bodies that are ‘at risk’, to be completed within<br />
a limited time-frame.<br />
3.6 LF2000 – Future Developments<br />
As a consequence of the deficiencies in the<br />
extent <strong>and</strong> quality of data describing water use<br />
the estimation of actual flow regimes has been<br />
a two stage process; the estimation of natural<br />
flows regimes coupled with the application of<br />
simple deterministic procedures for<br />
incorporating the impacts of water use. As our<br />
underst<strong>and</strong>ing of the characteristics of<br />
1005
100<br />
pressures <strong>and</strong> the quality of measured pressure<br />
practitioners to meet the needs of the WFD<br />
with greater certainty.<br />
Q95 predicted (cumecs)<br />
10<br />
1<br />
0.1<br />
0.01<br />
0.001<br />
Natural Estimate<br />
Influenced Estimate<br />
0.0001<br />
0.0001 0.001 0.01 0.1 1 10 100<br />
Q95 observed (cumecs)<br />
data improves holistic modelling of the system<br />
will become possible. This will enable a<br />
greater range of catchment data to be used<br />
within modelling approaches <strong>and</strong> will allow<br />
feedback mechanisms to be built directly into<br />
models.<br />
In addition, the potential for integrating a<br />
water quality model within the current system<br />
has also been explored leading to the<br />
development of a prototype water quality<br />
module within LF2000. This directly couples<br />
hybrid stochastic-deterministic point source<br />
water quality models to the underpinning<br />
hydrological models within LF2000, such that<br />
the interactions between water use - dilution<br />
<strong>and</strong> water quality can be investigated<br />
dynamically. The methodologies within this<br />
software are based on those developed as part<br />
of the GREAT-ER project [Schowanek et al.,<br />
2001].<br />
Whilst the FDC can provide a significant<br />
amount of information which can aid in<br />
managing water resources within a catchment<br />
the limitations of using statistical measures are<br />
recognised. Time series flow data enables<br />
assessments of yield for water resource<br />
schemes, the in-stream flow requirements of<br />
aquatic flora <strong>and</strong> fauna <strong>and</strong> the impacts of<br />
climate change at the catchment scale to be<br />
made. There is therefore a need to develop<br />
models to provide time series of flows within<br />
surface water bodies at ungauged sites.<br />
Regionalised continuous simulation models are<br />
currently being developed. These use<br />
regionalised rainfall runoff models, combined<br />
with parameters which are derived from<br />
catchment characteristics, to develop<br />
continuous time series of flow data [Young,<br />
2002]. The planned incorporation of this<br />
within the LF2000 framework will allow<br />
4 CONCLUSIONS<br />
LF2000 is a user friendly GIS Framework<br />
underpinned by hydrological models <strong>and</strong> a<br />
flexible database.<br />
LF2000 has been used as part of the screening<br />
process within the WFD to identify pressures<br />
on surface water bodies. This assessment is fed<br />
into a risk based approach for identifying water<br />
bodies at risk of not achieving or maintaining<br />
GES. Pressure information was used to<br />
determine the degree of influence, at over 6000<br />
locations within Engl<strong>and</strong> <strong>and</strong> Wales providing<br />
rapid, consistent results on the degree of<br />
modification at each AP. The outputs of this<br />
have been combined with a measure of<br />
ecological sensitivity to changes in the<br />
hydrological regime to provide an assessment<br />
of the risk that the surface water body will not<br />
achieve GES. The Environment Agency for<br />
Engl<strong>and</strong> <strong>and</strong> Wales, using this data, will<br />
therefore achieve their aim of identifying<br />
surface water bodies at risk of not achieving or<br />
maintaining GES by the end of 2004.<br />
In addition, a number of modules which deal<br />
with alternative aspects of the WFD are<br />
currently being developed. A prototype water<br />
quality module has been developed that can<br />
provide information on the impact on water<br />
quality of point source discharges at a<br />
catchment scale. Methods by which continuous<br />
time series of flows can be estimated at the<br />
ungauged site are also being developed.<br />
LF2000 provides a suite of tools, which<br />
enables the regulator to assess <strong>and</strong> mange<br />
water resources at the catchment scale in a<br />
rapid, consistent manner. These tools can play<br />
an important role in allowing the regulatory<br />
body to manage water resource effectively <strong>and</strong><br />
meet the requirements of the WFD.<br />
5 REFERENCES<br />
Boorman, D.B., J.M. Hollis, <strong>and</strong> A. Lilly,<br />
Hydrology of soil types: a hydrologicallybased<br />
classification of the soils of the United<br />
Kingdom. Report 126. Institute of Hydrology.<br />
Wallingford, 1995.<br />
Bullock, A., A. Gustard, K. Irving, A. Sekulin,<br />
<strong>and</strong> A. Young, Low Flow Estimation in<br />
Artificially Influenced Catchments. National<br />
Rivers Authority R&D Note 274, 1994.<br />
1006
Environment Agency. Managing Water<br />
Abstraction: the Catchment Abstraction<br />
Management Strategy Process. Environment<br />
Agency, Bristol, UK. 2002a.<br />
Environment Agency. Resource Assessment<br />
Management Framework. Environment<br />
Agency, Bristol, UK 2002b.<br />
Extence, C.A., D.M. Balbi, <strong>and</strong> R.P. Chadd,<br />
River flow indexing using British benthic<br />
macroinvertebrates: A framework for setting<br />
hydroecological objectives. Regulated Rivers-<br />
Research <strong>and</strong> Management. 15 (6), 543-574,<br />
1999.<br />
Gustard, A.G. Cole, , D. Marshall, <strong>and</strong> A.<br />
Bayliss, A study of compensation flows in the<br />
UK. IH Report No. 99, 1987.<br />
Holmes, M.G.R., A.R. Young, A. Gustard, <strong>and</strong><br />
R.A. Grew, A new approach to estimating<br />
Mean Flow in the UK. Hydrology <strong>and</strong> Earth<br />
System Sciences, 6 (4), 709-720, 2002a.<br />
Holmes, M.G.R., A.R. Young, A. Gustard, <strong>and</strong><br />
R.A. Grew, A region of influence approach to<br />
predicting flow duration curves within<br />
ungauged catchments. Hydrology <strong>and</strong> Earth<br />
System Sciences, 6 (4), 721-731, 2002b.<br />
Moore, R.V., The Logical <strong>and</strong> Physical Design<br />
of the LOIS Database. LOIS Special <strong>Volume</strong>,<br />
Science of the Total Environment, 194, 137 -<br />
146. 1997.<br />
Schowanek, D., K. Fox, M. Holt, , F.R.<br />
Schroeder, V. Koch, G. Cassani, M. Matthies,<br />
G. Boeije, P. Vanrolleghem, A. Young, G.<br />
Morris, C. G<strong>and</strong>olfi, <strong>and</strong> T.C.J. Feijtel,<br />
GREAT-ER: a new tool for management <strong>and</strong><br />
risk assessment of chemicals in river basins -<br />
Contribution to GREAT-ER #10. Water<br />
Science <strong>and</strong> Technology, 43 (2), 179-185. 2001<br />
Young, A.R., R. Grew, <strong>and</strong> M.G.R. Holmes,<br />
LF2000: a national water resources assessment<br />
<strong>and</strong> decision support tool.<br />
Water Science <strong>and</strong> Technology, 48 (10), 119-<br />
126 . 2003.<br />
Young A.R., River flow simulation within<br />
ungauged catchments using a daily rainfallrunoff<br />
model. In British Hydrological Society<br />
Occasional Paper No. 13: Continuous river<br />
flow simulation: methods, applications <strong>and</strong><br />
uncertainties. Ed. Ian Littlewood. 80 pp, 2002.<br />
1007
A Tool for Evaluating Risk to Surface Water Quality Status<br />
Neil McIntyre<br />
Imperial College London, UK. Email n.mcintyre@imperial.ac.uk<br />
Abstract: Water quality Risk Analysis Tool (WaterRAT) is recently developed software for supporting<br />
surface water quality management. The software contains a library of river <strong>and</strong> lake quality models, aiming to<br />
give flexibility over specification of model scope, complexity <strong>and</strong> scale. Various sources of uncertainty can<br />
be included in the analysis, including uncertainty in boundary conditions, initial conditions, parameters,<br />
model structure <strong>and</strong> management objectives. Water quality can then be modelled allowing for these sources of<br />
uncertainty. Important data uncertainties can be indicated, <strong>and</strong> so data collection programmes can be suitably<br />
refined. In this paper, the motivation for the WaterRAT tool <strong>and</strong> the methods it employs are presented, its<br />
features are described, <strong>and</strong> its utility for uncertainty evaluation <strong>and</strong> sensitivity analysis is demonstrated using<br />
a river water quality management problem. Emerging challenges for modellers, which cannot yet be<br />
addressed using WaterRAT, are discussed.<br />
Keywords: Water quality; simulation; uncertainty; Monte Carlo<br />
1. INTRODUCTION<br />
There is increasing motivation <strong>and</strong> opportunity to<br />
use simulation model predictions as a basis for<br />
managing water resources, including the protection<br />
of surface water quality. The motivation comes<br />
from the generally high priority given to protecting<br />
the aquatic environment. In particular, new<br />
directives governing water quality specify that<br />
river basins should be viewed as integrated units,<br />
potentially requiring consideration of large<br />
numbers of factors with complex, interacting<br />
effects on water quality. The opportunity for using<br />
simulation models is coming from at least three<br />
directions: reducing constraints on computer power<br />
<strong>and</strong> therefore model size <strong>and</strong> complexity; new<br />
attempts to observe <strong>and</strong> underst<strong>and</strong> processes<br />
affecting water quality management; <strong>and</strong> the<br />
increasing number of qualified modellers.<br />
Arguably, a fourth reason to be optimistic about a<br />
more useful role for simulation modelling in the<br />
future is the increasing attention that is now being<br />
given towards resolving, or at least assessing,<br />
model uncertainty. The fundamental reason for<br />
uncertainty is our inability to observe <strong>and</strong><br />
underst<strong>and</strong> the controlling processes <strong>and</strong> represent<br />
them numerically at the relevant scales. Implicit to<br />
that are a number of modelling issues that are welldiscussed<br />
elsewhere (see Beck 1987, McIntyre et<br />
al. 2003a) but might be summarised as: the need to<br />
simplify the real environment into a conceptual<br />
model; equally plausible alternative simplifications<br />
lead to different predictions; the need to calibrate<br />
model parameters leading to biases <strong>and</strong><br />
equifinality; limitations in our measurement<br />
techniques leading to errors in point estimates of<br />
data; errors in integrating or interpolating data to<br />
the relevant modelling scale; <strong>and</strong> numerical errors<br />
in solutions to differential equations. While there is<br />
no clear consensus about how these issues should<br />
be addressed, it is generally agreed that the<br />
consequent model uncertainty is high, <strong>and</strong> that<br />
water quality modellers need to give more attention<br />
to estimating <strong>and</strong> reporting this uncertainty.<br />
This paper introduces the Water quality Risk<br />
Analysis Tool (WaterRAT), a tool for exploring<br />
uncertainty in forecasts of river <strong>and</strong> lake quality.<br />
This software was developed as part of a European<br />
Commission project, Total Pollution Load<br />
Estimation <strong>and</strong> Management (TOPLEM), which<br />
aimed to produce a software system for managing<br />
water pollution in a Chinese catchment where<br />
supporting data are sparse. This paper will briefly<br />
describe WaterRAT, summarise a case study, <strong>and</strong><br />
discuss the limitations of this <strong>and</strong> other software in<br />
the context of future modelling needs.<br />
2. DESCRIPTION OF WATERRAT<br />
2.1 Summary<br />
WaterRAT (McIntyre <strong>and</strong> Zeng 2002) is a<br />
spreadsheet-based modelling tool that includes a<br />
library of surface water quality models, presently<br />
including a choice of one-dimensional river models<br />
<strong>and</strong> two-dimensional lake models. WaterRAT is<br />
built within Microsoft Excel, so that WaterRAT’s<br />
1008
own data processing modules can be supplemented<br />
by those of Excel. The input <strong>and</strong> output is via a<br />
series of spreadsheets <strong>and</strong> model specifications are<br />
made via Visual Basic menus <strong>and</strong> dialogue boxes.<br />
The library of simulation models comprises a<br />
series of Dynamic Link Libraries.<br />
The model library includes alternatives for<br />
modelling pollution transport, water temperature<br />
<strong>and</strong> water quality, <strong>and</strong> also offers a choice of<br />
modelled determin<strong>and</strong>s. These include total<br />
organic carbon, biochemical oxygen dem<strong>and</strong>,<br />
phytoplankton, dissolved oxygen, nitrogen <strong>and</strong><br />
phosphorus, a toxic substance, floating <strong>and</strong><br />
suspended oil, <strong>and</strong> total suspended solids. This is<br />
supported by sediment models which include<br />
biochemically <strong>and</strong> physically-driven sedimentwater<br />
interactions. A thermodynamic model<br />
simulates water temperature <strong>and</strong> ice cover.<br />
2.2 Spatial <strong>and</strong> temporal resolution<br />
For river modelling, the river is represented as a<br />
series of well-mixed control volumes between<br />
which pollution transport processes are simulated<br />
using the advection-dispersion equation supported<br />
by two alternative hydraulic models (a quasi-steady<br />
friction formula <strong>and</strong> a non-linear store). Each<br />
control volumes must be prescribed certain<br />
spatially-varying parameters which depend on the<br />
transport model selected. The lake models work on<br />
the same control-volume principle, except that they<br />
are able to represent the vertical variation in water<br />
quality due to effects of thermal stratification as<br />
well as length-wise variations.<br />
The output time-step is defined by the user, <strong>and</strong><br />
may be anything greater than one minute. The<br />
available input time-series data will be<br />
automatically interpolated to this time-scale, using<br />
either linear interpolation, a cubic spline or a step<br />
function, as chosen by the user. The numerical<br />
integration in the time domain uses a Fehlberg<br />
adaptive time-step scheme. This is an important<br />
feature for Monte Carlo simulation, where<br />
r<strong>and</strong>omly sampled inputs lead to numerical<br />
stability <strong>and</strong> accuracy criteria which can vary<br />
widely, both over the time-domain <strong>and</strong> from one<br />
model realisation to the next (McIntyre et al.,<br />
2004). The spatial grid is prespecified, making the<br />
user responsible for reconciling the spatial<br />
resolution with the temporal tolerance, so that<br />
numerical dispersion <strong>and</strong> spatial averaging errors<br />
are not excessive.<br />
2.3 Boundary conditions, initial conditions<br />
<strong>and</strong> model parameters<br />
Dynamic boundary conditions include the<br />
meteorological, pollution <strong>and</strong> flow source <strong>and</strong><br />
abstraction data. All meteorological time-series<br />
(rainfall, evaporation, dew-point, air temperature,<br />
wind speed, <strong>and</strong> surface light intensity are needed<br />
as inputs to various alternative models) are<br />
assumed to be uniform over the river or lake. Any<br />
number of sources of flow <strong>and</strong>/or pollution can be<br />
input, subject to computer memory. A negative<br />
flow is interpreted as a loss, <strong>and</strong> any associated<br />
pollution loads are neglected.<br />
Static boundary conditions are specified for each<br />
control volume. For the river models examples of<br />
these are: channel cross-section shape; a leakage<br />
rate; sediment oxygen dem<strong>and</strong>; active sediment<br />
area; <strong>and</strong> hydraulic or routing parameters. For the<br />
lake models, the bathymetry is defined by a<br />
volume-level relationship for the lake. Initial<br />
conditions can be either entered via a spreadsheet<br />
as a model input, or they can be estimated using a<br />
specified ‘warm-up’ period. During this period the<br />
dynamic boundary conditions are assumed steadystate<br />
at those of the specified start time of the<br />
simulation.<br />
With the exception of meteorology, all model<br />
parameters, initial conditions, boundary conditions<br />
(including sources of flow <strong>and</strong> pollution) can be<br />
considered as uncertain inputs. Prior to running the<br />
model, the user signifies that an input is uncertain<br />
by specifying a distribution instead of an assumed<br />
value. Each distribution may be propagated to<br />
prediction uncertainty, or included in the<br />
calibration or sensitivity analysis. This means that<br />
the model calibration <strong>and</strong> predictions need not be<br />
conditional on the precision <strong>and</strong> reliability of input<br />
data, <strong>and</strong> that the relative significance of input<br />
uncertainties can be revealed through sensitivity<br />
analysis.<br />
2.4 Model conditioning, sensitivity <strong>and</strong><br />
uncertainty analysis<br />
Uncertainty in inputs may be specified as<br />
independent uniform distributions using a<br />
maximum <strong>and</strong> minimum bound, or as any joint<br />
distribution using a series of discrete samples. In<br />
the latter case, each sample is weighted with a<br />
relative probability.<br />
R<strong>and</strong>om sampling from these distributions (i.e.<br />
Monte Carlo simulation) can be used as a basis for<br />
model conditioning <strong>and</strong> sensitivity analysis. For<br />
example, the unconditioned distribution can be<br />
updated by multiplying the prior probability of<br />
each sample by a posterior probability. The<br />
posterior probability associated with a sampled set<br />
of inputs may be defined as their perceived<br />
likelihood based on how well they simulate the<br />
observed data. This is the same as Generalised<br />
Likelihood Uncertainty Estimation (GLUE; Beven<br />
1009
<strong>and</strong> Binley 1992). However, WaterRAT’s facilities<br />
encourage integration of boundary <strong>and</strong> initial<br />
condition uncertainty, which is not normally done<br />
within the GLUE framework. Alternatively, the<br />
posterior may be defined by the probability that the<br />
sampled set of inputs will lead to a successful<br />
outcome in terms of meeting a water quality target.<br />
Thereby, the probability of achieving or failing a<br />
water quality objective due to combinations of<br />
uncertain inputs may be quantified.<br />
Any posterior likelihood may be plotted against<br />
each individual uncertain input as a marginal<br />
distribution. This is a well-established technique<br />
for assessing regional sensitivity of calibration<br />
objective functions to parameter uncertainty.<br />
However, previous applications generally assume<br />
inputs other than parameters to be known with<br />
certainty, <strong>and</strong> all results are conditional on this<br />
assumption. Also, it seems that the same method<br />
has not previously been applied to assess how the<br />
probability of failing management objectives is<br />
sensitive to uncertain inputs. Such inputs may be<br />
manageable in practice (e.g. point sources of<br />
pollution), or less manageable (e.g. initial sediment<br />
quality), or may be essentially unmanageable (e.g.<br />
parameters representing the physical properties of<br />
the environment).<br />
Whereas sensitivity analysis can highlight which<br />
uncertain inputs are most likely to influence the<br />
model results, prediction of space <strong>and</strong> time-series<br />
is needed to show where <strong>and</strong> when this influence is<br />
significant. Using Monte Carlo sampling, a<br />
specified number of samples are taken either from<br />
the prior uniform distributions of inputs, or from<br />
the sets sampled during prior conditioning. In this<br />
latter case the likelihood of the model result<br />
obtained from each sample is weighted by the<br />
likelihood of that sample (as calculated during<br />
conditioning), following the GLUE methodology.<br />
A distribution of model output can then be derived.<br />
WaterRAT also offers first order methods of<br />
sensitivity analysis <strong>and</strong> uncertainty propagation.<br />
3. CASE STUDY<br />
3.1 Background, model <strong>and</strong> data<br />
The case study presented here is of the Charles<br />
River in Massachusetts. This study represents the<br />
data availability that might be expected in the USA<br />
<strong>and</strong> Europe (rather than the original WaterRAT<br />
application in China where data was especially<br />
limited). The Charles River in the 1990s suffered<br />
from undesirable concentrations of phytoplankton,<br />
largely due to pollution with nutrients. The<br />
principal management option was investment in<br />
phosphorus removal at a number of wastewater<br />
treatment works (WWTWs) along the river. This<br />
modelling study revisits that situation, to identify<br />
the key uncertainties affecting the reliability of the<br />
phosphorus removal option, <strong>and</strong> to quantify the<br />
probability of failure due to the effect of data <strong>and</strong><br />
model uncertainties.<br />
The model is of flow, water depth <strong>and</strong> temperature,<br />
<strong>and</strong> nine water quality determin<strong>and</strong>s:<br />
• Phytoplankton, measured as chlorophyll-a<br />
• Slow-reacting organic carbon<br />
• Fast-reacting organic carbon<br />
• Organic nitrogen<br />
• Ammonium<br />
• Nitrate plus nitrite<br />
• Organic phosphorus<br />
• Inorganic phosphorus<br />
• Dissolved oxygen<br />
A system of partial differential equations<br />
represents the interactions between these 12<br />
variables, including 24 uncertain parameters to be<br />
conditioned. A full description of the model is not<br />
important for the objectives of this paper, but is<br />
available in McIntyre et al. [2003b].<br />
The conditioning <strong>and</strong> model assessment were<br />
based on data from the 20 th August 1996 <strong>and</strong> the<br />
10 th October, 1996. On both dates the water quality<br />
was assumed to be at steady-state. Measurements<br />
of the water quality variables were available from<br />
nine sections along the river, <strong>and</strong> daily pollution<br />
loads from the headwater <strong>and</strong> eight sources (sewers<br />
<strong>and</strong> tributaries). Error bounds in all these data were<br />
estimated; this estimate was largely subjective due<br />
to the limited number of measurements.<br />
3.2 Model conditioning<br />
Model conditioning was performed by r<strong>and</strong>om<br />
sampling from the joint prior distribution of<br />
uncertain inputs, <strong>and</strong> assigning a posterior<br />
probability to each sample based on its<br />
performance in meeting the objective. This was<br />
done in two stages – conditioning upon chl-a data<br />
(measured on the 20 th August) to reduce<br />
uncertainty in parameters, <strong>and</strong> then further<br />
conditioning to the constraint chlorophyll-a <<br />
10mgm -3 , to identify the probability of achieving<br />
this objective across a range of phosphorus (P)<br />
load reduction scenarios.<br />
The first stage of conditioning is essentially the<br />
GLUE method, using the objective function (OF 1 )<br />
given in (1). Using (2), OF 1 is multiplied by the<br />
prior probability Lp <strong>and</strong> rescaled to give a relative<br />
measure of likelihood L of the sampled input set<br />
1010
(α). This is based on the simple, subjectivelyfounded<br />
premise that the better the model fits<br />
reality, the more reliable it will be for predictions.<br />
( ) = ( α ) − ( )<br />
∑ ( A ) −<br />
j A j<br />
2<br />
1 α<br />
j<br />
OF α (1)<br />
L<br />
−1<br />
⎡ ⎤<br />
α ⎢<br />
1⎥<br />
OF1<br />
N<br />
( ) Lp ⋅ OF Lp( α ) ( α )<br />
= (2)<br />
∑<br />
⎣ ⎦<br />
where subscript j indicates the j th monitored section<br />
on the river, A is the model output of chl-a, A is<br />
the measured chlorophyll-a, <strong>and</strong> N is the number of<br />
samples (7000 in this case). Importantly, α is a<br />
sampled set of inputs to the objective function<br />
calculation, which includes a sampled set of model<br />
parameters, a sampled realisation of pollution<br />
loads <strong>and</strong> a sampled realisation of A , all from<br />
within their a priori perceived bounds of error. The<br />
joint posterior distribution of the model parameters<br />
may be obtained by integrating over the other<br />
inputs - an improvement upon the normal practice<br />
of fixing the other inputs during conditioning.<br />
The posteriors (L) from (2) become the new priors<br />
(Lp) for the second stage of conditioning. The<br />
7000 sampled parameter sets, each with an<br />
associated Lp, are recalled <strong>and</strong> the model is run<br />
using each. Other model inputs are r<strong>and</strong>omly<br />
sampled from within new ranges that are relevant<br />
to the forecasting problem. In this case, these<br />
ranges are of a feasible P load reduction (W) at a<br />
selected site, together with the estimated<br />
uncertainty in the other inputs. The relevant<br />
objective function is the intended constraint on<br />
chlorophyll-a in August, defined in (3).<br />
OF<br />
OF<br />
2<br />
2<br />
( α ) = 1 for chl - a<br />
( α ) = 0 otherwise<br />
< 10mgm<br />
−3<br />
(3)<br />
Following calculation of OF 2 for the 7000 model<br />
runs, (2) is applied again (but with OF 2 instead of<br />
OF 1 ). Subsequently, each value of L is the<br />
combined probability of a set of inputs <strong>and</strong> the<br />
objective being achieved (given, of course, the<br />
various modelling assumptions that have been<br />
involved to this point). WaterRAT outputs all the<br />
values of L, Lp, corresponding input samples <strong>and</strong><br />
summary statistics of the conditioned distribution.<br />
Integration over all other uncertain inputs allows<br />
the marginal distribution of each input to be<br />
presented. For example, for our investigated point<br />
source reduction W, P( W A < 10)<br />
can be calculated<br />
across the range of W. Baye's theorem allows the<br />
modelled probability of achieving the objective to<br />
be calculated across W:<br />
P<br />
( A < 10 W )<br />
where,<br />
P<br />
=<br />
( A ) = ∑ Lp ⋅<br />
P < OF2<br />
N<br />
<strong>and</strong><br />
( A < 10) . P ( W A < 10)<br />
Lp( W )<br />
(4)<br />
10 (5)<br />
( W ) 1 N<br />
Lp = /<br />
(6)<br />
For the Charles River study, Equations 3-6 were<br />
applied to the constraint A < 10mgm -3<br />
independently at each of nine strategic locations<br />
(A-I) along the river. Various sites for P load<br />
reductions were analysed. For example, Figure 1<br />
shows the probability of achieving the target as a<br />
function of the percentage P load reduction at the<br />
headwater.<br />
Probability of achieving objective<br />
1.0<br />
0.8<br />
0.6<br />
0.4<br />
(H)<br />
(D)<br />
(C)<br />
(B)<br />
(A)<br />
0.2<br />
(I)<br />
(E)<br />
(F)<br />
(G)<br />
0.0<br />
-100% -75% -50% -25% 0%<br />
Reduction<br />
Figure 1. Modelled probability of A < 10mgm -3<br />
at nine sections (A-I) on the Charles River as a<br />
function of percentage reduction in P load from<br />
the headwater.<br />
Probability of achieving objective<br />
1.0<br />
0.8<br />
0.6<br />
0.4<br />
(A)<br />
0.2<br />
(E)<br />
(D)<br />
0.0<br />
(B)<br />
(C)<br />
-100% -75% -50% -25% 0%<br />
Reduction<br />
Figure 2. Modelled probability of A < 10mgm -3<br />
at nine sections (A-E) on the Charles River as a<br />
function of percentage reduction in P load from<br />
the CRPCD wastewater treatment works.<br />
1011
Figure 2 shows the same for P load reduction at the<br />
large CRPCD treatment works on the upper<br />
Charles River, given a 50% reduction at the<br />
headwater (F-I are consistently zero in this case<br />
<strong>and</strong> so are not shown). Even substantial reductions<br />
at either location would be a high-risk option for<br />
controlling chlorophyll-a especially at the further<br />
downstream sections, given the uncertainties about<br />
how the system will respond.<br />
3.3 Sensitivity analysis<br />
One method of sensitivity analysis is the wellestablished<br />
method of regional sensitivity analysis<br />
first applied to water quality models by Spear <strong>and</strong><br />
Hornberger [1980]. This summarises the difference<br />
between the prior <strong>and</strong> posterior marginals using the<br />
univariate Kolmogorov-Smirnov (KS) statistic.<br />
An example output is given in Figure 3, the KS<br />
statistic comparing the distributions prior to <strong>and</strong><br />
after conditioning to the A < 10mgm -3 constraint.<br />
In Figure 3, the x-axis contains all the uncertain<br />
inputs, divided into model parameters <strong>and</strong> P<br />
loading rates, <strong>and</strong> the y-axis is the value of the KS<br />
statistic. There are nine trajectories – one for each<br />
of the nine sections (A-I) on the river. For the<br />
purpose of this discussion only the evidently most<br />
important inputs are labeled. It is clear that the<br />
uncertainty in the load of P from the headwater,<br />
<strong>and</strong> that in the model parameters (representing the<br />
biochemistry) dominate the uncertainty in the<br />
outcome. The regional influence of variations in P<br />
loadings from wastewater treatment works is small<br />
in comparison.<br />
4. DISCUSSION<br />
4.1 Review of WaterRAT<br />
The aim of WaterRAT is to allow integration <strong>and</strong><br />
exploration of many sources of uncertainty. For<br />
example, the case study included sampling<br />
realisations of input <strong>and</strong> output data from within<br />
perceived error bounds, so that the parameter<br />
conditioning <strong>and</strong> subsequent analysis were not<br />
conditional on the accuracy of any measured data.<br />
The sensitivity analysis indicated that the reliability<br />
of management decisions is controlled by<br />
parameter uncertainty, <strong>and</strong> uncertainty about the<br />
contribution of the headwater (i.e. distributed<br />
sources from the upper catchment). Judicious<br />
planning might therefore in this case involve<br />
further data collection <strong>and</strong> modelling, prior to<br />
engineering interventions.<br />
The obvious scientific weakness of the case study<br />
is the assumption that the model structure is<br />
correct. A nominal remedy would be to repeat the<br />
analysis with a more complex model from the<br />
WaterRAT library (e.g. including sediment-water<br />
interactions) <strong>and</strong> compare the outcomes. However,<br />
it is speculated that this would lead to the same<br />
overall conclusion, considering the limitations of<br />
the data. Although discrete trials of alternative<br />
model structures is unlikely to resolve the issue of<br />
model structure error, it indicates their potential<br />
significance. Should the data allow, more rigorous<br />
analysis methods of evaluating structural error are<br />
available, outwith WaterRAT (e.g. Wagener et al.<br />
2002).<br />
In WaterRAT, the procedure used for sampling<br />
from uniform priors is Latin hyper-cube sampling.<br />
Initial versions of WaterRAT also included Monte<br />
Carlo Markov Chain sampling methods, which<br />
might be expected to be better for Bayesian model<br />
analysis (Vrugt et al. 2003). However, the<br />
advantage was not evident when using typically<br />
sparse data such as those from the Charles River.<br />
WaterRAT also includes a genetic algorithm for<br />
deterministic optimistation, <strong>and</strong> first order methods<br />
as an alternative to Monte Carlo analysis.<br />
4.2 Current challenges<br />
Amongst others, a primary challenge facing surface<br />
water quality modellers is how to obtain useful<br />
estimates of uncertainty in much larger models<br />
than those in WaterRAT. Spatially distributed<br />
catchment models may include hundreds or<br />
thous<strong>and</strong>s of uncertain model inputs <strong>and</strong> outputs.<br />
Current methods of uncertainty analysis for<br />
complex environmental models are centered<br />
around Monte Carlo simulation, due to the ease of<br />
application to complex non-linear simulation<br />
models. While new computing power is allowing<br />
models to exp<strong>and</strong> in size, it is far from allowing<br />
comprehensive Monte Carlo sampling of all<br />
possible models <strong>and</strong> model input scenarios. This is<br />
despite new algorithms based on Markov Chains<br />
aimed at giving a more efficient exploration of the<br />
uncertainties (e.g. Vrugt et al. 2003). Furthermore,<br />
the importance of extreme values has been largely<br />
overlooked in design of algorithms, which tend to<br />
focus resources on the modes of the posteriors.<br />
Research at Imperial is currently investigating<br />
pathways to resolving these challenges.<br />
1012
1<br />
Headwater<br />
P load<br />
KS statistic<br />
Phytoplankton<br />
growth rate<br />
Organic P<br />
decomposition<br />
rate<br />
Phytoplankton<br />
death<br />
rate<br />
CRPCD<br />
WWTW<br />
P load<br />
Stop River<br />
Tributary<br />
P load<br />
0<br />
Model parameters<br />
Pollution sources<br />
Figure 3. Sensitivity of the probability of achieving target water quality to the various model inputs,<br />
measured by the Kolmogorov-Smirnov (KS) statistic.<br />
5. CONCLUSIONS<br />
The WaterRAT software, developed at Imperial<br />
College has been introduced. A case study (the<br />
Charles River, Massachusetts) has been used to<br />
highlight some capabilities <strong>and</strong> limitations of the<br />
software. The significance of many different<br />
sources of uncertainty can be included in Bayesian<br />
analysis <strong>and</strong> regional sensitivity analysis. The<br />
analyses can be used to indicate priorities for<br />
protecting water quality via further modelling,<br />
data collection <strong>and</strong> engineering interventions. The<br />
main limitation to the WaterRAT software is that<br />
it provides no tools to assess model structure<br />
error, apart from discrete comparisons of<br />
alternative model structures. Also, at present it<br />
only includes models of rivers <strong>and</strong> lakes rather<br />
than of the wider catchment. Finally, the paper<br />
briefly discussed research priorities for water<br />
quality modelling, arguing that new methods of<br />
analysis will be needed to face the challenges of<br />
distributed modelling <strong>and</strong> extreme value analysis.<br />
6. ACKNOWLEDGEMENTS<br />
The European Commission funding funded this<br />
work, <strong>and</strong> thanks also to Steve Chapra <strong>and</strong> Camp<br />
Dresser <strong>and</strong> McKee Inc. for the Charles data.<br />
7. REFERENCES<br />
Beck, M.B., Uncertainty in water quality models,<br />
Water Resources Research, 23(8), 1393-<br />
1441, 1987.<br />
Beven, K.J. <strong>and</strong> Binley, A.M., The future of<br />
distributed models; model calibration <strong>and</strong><br />
predictive uncertainty, Hydrological<br />
Processes, 6, 279-298, 1992.<br />
McIntyre, N. <strong>and</strong> Zeng, S., TOPLEM River <strong>and</strong><br />
lake water quality models. Report.<br />
PL972722 D4.3(10), Imperial College<br />
London, 2002.<br />
McIntyre, N., Wagener, T., Wheater, H.S. <strong>and</strong><br />
Zeng, S., Uncertainty <strong>and</strong> risk in water<br />
quality modelling <strong>and</strong> management, Journal<br />
of Hydroinformatics, 5, 259-274, 2003a<br />
McIntyre, N., Wagener, T., Wheater, H.S.,<br />
Chapra, S., Risk-based modelling of surface<br />
water quality- A case study of the Charles<br />
River, Massachusetts, Journal of<br />
Hydrology., 274, 225-247, 2003b.<br />
McIntyre, N., Jackson, B., Wheater, H.S., Chapra,<br />
S. Numerical efficiency in Monte Carlo<br />
simulations - a case study of a river<br />
thermodynamic model, ASCE Journal of<br />
<strong>Environmental</strong> Engineering, 130, 4, 456-<br />
464, 2004.<br />
Spear, R. C. <strong>and</strong> Hornberger, G. M.,<br />
Eutrophication in peel inlet - 2.<br />
Identification of critical uncertainties via<br />
generalized sensitivity analysis, Water<br />
Research 14(1), 43-49, 1980.<br />
Vrugt, J., Gupta, H., Bouten, W. <strong>and</strong> Sorooshian,<br />
S., A shuffled complex evolution Metropolis<br />
algorithm for optimization <strong>and</strong> uncertainty<br />
assessment of hydrologic model parameters,<br />
Water Resources Research, 39(8), SWC1:1-<br />
1:16, 2003.<br />
Wagener, T., McIntyre, N., Lees, M.J., Wheater,<br />
H.S. <strong>and</strong> Gupta, H.V., Towards reduced<br />
uncertainty in conceptual rainfall-runoff<br />
modeling: dynamic identifiability analysis,<br />
Hydrological Processes, 17(2), 455-476,<br />
2002.<br />
1013
Spatially Distributed Investment Prioritization for<br />
Sediment Control over the Murray Darling Basin,<br />
Australia<br />
Hua Lu ab , Christopher Moran a , Ian Prosser a <strong>and</strong> Ronald DeRose a<br />
a<br />
CSIRO L<strong>and</strong> <strong>and</strong> Water, Canberra, ACT 2601, Australia<br />
b<br />
Institute of Theoretical Geophysics, University of Cambridge, UK<br />
Abstract: Based on a spatially-distributed sediment budget across the Murray Darling Basin, costs of<br />
achieving a range of sediment reduction targets were estimated for a number of locations. Four investment<br />
prioritization scenarios were tested to identify the most cost-effective strategy to control suspended sediment<br />
loads. The impacts of spatial heterogeneity of sediment transport <strong>and</strong> varying the spatial scale of target<br />
locations on cost effectiveness were examined. The results show that: 1) an optimum solution of costeffective<br />
sediment control can be determined through the spatial sediment budget; 2) appropriate<br />
investment prioritization can offer potential large cost savings as the magnitude of the costs can vary by<br />
several times depending on what type of erosion source or sediment delivery is targeted; 3) target settings<br />
which only consider the erosion source rates can potentially result in spending more money than r<strong>and</strong>om<br />
management intervention; <strong>and</strong> 4) prioritization becomes a more cost effective strategy as the area<br />
considered increases because of the spatial heterogeneity of contributing sediment. An interpretation of the<br />
non-linear cost to increasing sediment reduction relationship is also provided.<br />
Keywords: Sediment control; Spatially distributed modelling; Prioritization.<br />
1. INTRODUCTION<br />
World-wide, suspended sediment with attached<br />
nutrients <strong>and</strong> organic matter are significant<br />
contributors to poor water quality in many<br />
waterways. Awareness of water quality<br />
degradation has led to actions in many places.<br />
Part of these actions is the setting of targets to<br />
reduce suspended sediment <strong>and</strong> pollutants. For<br />
instance, in the USA, 40%-50% reductions in<br />
nutrient export have been set [Schleich et al.,<br />
1996; WDNR, 1988]. Nine European countries<br />
have agreed to take joint actions to achieve a 50%<br />
reduction in the total load of nutrients to the<br />
Baltic Sea [HELCOM, 1993]. In Australia, under<br />
the National Action Plan for Salinity <strong>and</strong> Water<br />
Quality, federal <strong>and</strong> state government agencies are<br />
working together to set targets for improving<br />
water quality [NAP, 2003]. A target of reduction<br />
by 30% has been set for the catchments of the<br />
Great Barrier Reef [Environment Australia, 2003].<br />
However, the jurisdictions allocating resources to<br />
achieve the targets need strategic advice. That is,<br />
which areas or/<strong>and</strong> pollutant types require the<br />
greatest investment to achieve the desired<br />
outcome(s)?<br />
Few studies have been carried out on cost<br />
effectiveness of management at a broad spatial<br />
extent. Gianessi <strong>and</strong> Peskin [1981] used a<br />
national water network model which took into<br />
account pollutants from both industrial <strong>and</strong><br />
agricultural activities to simulate the effects of<br />
four policy scenarios on water quality in America.<br />
They concluded that efficient sediment-related<br />
pollution control could be achieved by focusing on<br />
one third of the nation’s agricultural regions.<br />
Schleich et al. [1996] used linear programming to<br />
determine whether the cost of achieving<br />
phosphorus reduction targets was different<br />
depending on the scale of the units over which<br />
management action was considered. They found<br />
that optimizing at the outlets of subcatchments<br />
was more expensive than optimizing from the<br />
basin outlet. The severe eutrophication <strong>and</strong><br />
ecological collapse of the Baltic Sea has led to<br />
internationally-coordinated research activities<br />
seeking cost effective policies of pollutant<br />
reduction [Gren, 2001]. Stochastic approaches<br />
1014
were used to examine the cost changes for a given<br />
probability of achieving a certain pollutant load<br />
target [Gren et al., 2000].<br />
The environmental properties governing pollutant<br />
generation, transport <strong>and</strong> deposition are not<br />
homogeneous over broad areal extents. There is<br />
considerable spatial <strong>and</strong> temporal variation<br />
inherent in topography, climate, soil, vegetation,<br />
management practises <strong>and</strong> l<strong>and</strong> use. While<br />
heterogeneity appears to be difficult for analysis, it<br />
presents a major opportunity, i.e., the possibility<br />
of cost saving through prioritized actions.<br />
Proper representation of the linkage between<br />
location <strong>and</strong> nature of pollutant sources <strong>and</strong> their<br />
downstream impacts is also critical. When<br />
considering sediment in terms of water quality<br />
impact, the management concern is how to control<br />
the sediment load at a point of interest<br />
downstream <strong>and</strong> the erosion sources are often<br />
several hundreds of kilometres upstream. Only a<br />
proportion of soil erosion reaches the channel<br />
network <strong>and</strong> only a proportion of that sediment is<br />
transported downstream as sediment can be<br />
intercepted by riparian vegetation <strong>and</strong> deposited<br />
on foot slopes <strong>and</strong> floodplains <strong>and</strong> in reservoirs<br />
<strong>and</strong> lakes.<br />
This paper proposes a method for spatially<br />
distributed investment prioritization. We consider<br />
a large regional basin – the Murray-Darling Basin<br />
in eastern Australia. Heterogeneity of contributing<br />
sediment <strong>and</strong> linkages between sources <strong>and</strong><br />
targets are explicitly represented through<br />
spatially-resolved sediment budgets [Prosser et<br />
al., 2001]. The spatial accounting of sediment<br />
budgets enables us to distinguish the sediments<br />
that made the way to a sediment control location<br />
from those which deposit before reaching the<br />
control location. By comparing the costeffectiveness<br />
of a range of management strategies,<br />
we show how resources could be allocated<br />
spatially under certain management action.<br />
2. METHODS<br />
2.1 Study Area<br />
The Murray-Darling Basin (MDB) covers an area<br />
of 1.1 × 10 6 km 2 (about 14% of Australia, Fig. 1)<br />
<strong>and</strong> it is an important agricultural centre. It<br />
contains around 75% of Australia’s irrigated l<strong>and</strong>,<br />
accounts for 40% of Australian agricultural<br />
production <strong>and</strong> inhabits two million people, about<br />
10% of the national population [ABS, 2002]. It<br />
also has the three longest rivers in Australia<br />
(Murray, Darling <strong>and</strong> Murrumbidgee). The river<br />
system is showing signs of environmental stress:<br />
salinity, reduction in both water quality <strong>and</strong><br />
quantity, sedimentation, loss of fish species <strong>and</strong><br />
algal blooms [NLWRA, 2001].<br />
AUSTRALIA<br />
140°E<br />
40°S 140°E<br />
150°E<br />
40°S<br />
2.2 Sediment Budget<br />
The investment prioritization analysis was carried<br />
out using the results of spatial modelling of<br />
sediment budgets across the MDB. The sediment<br />
budgets assess current patterns of the major<br />
erosion, river sediment transport <strong>and</strong> deposition<br />
processes in the Basin, using the SedNet model<br />
[Prosser et al., 2001]. SedNet is a set of GIS<br />
programs that define river networks <strong>and</strong> their<br />
associated catchments <strong>and</strong> route sediment through<br />
the network as a function of river hydrology <strong>and</strong><br />
mapping of erosion processes [Prosser et al.,<br />
2001]. The application of SedNet to the MDB is<br />
reported in detail in DeRose et al. [2003].<br />
The river network of the MDB was defined from<br />
the 9” digital elevation model (DEM), Australia<br />
(http://cres.anu.edu.au/dem) <strong>and</strong> divided into river<br />
links, separated by tributary junctions or nodes.<br />
Each link of the river network has an associated<br />
catchment area of around 50 - 100 km 2 . The river<br />
links are the basic elements of the sediment<br />
budget model <strong>and</strong> the area contributing to the link<br />
is referred as link element hereafter. Each link, i,<br />
receives a mean annual supply of suspended<br />
sediment from upstream tributaries (T i ), from<br />
bank erosion along the link itself (B i ), <strong>and</strong> from<br />
gully erosion (G i ) <strong>and</strong> hillslope sheetwash <strong>and</strong> rill<br />
erosion (E i ) in the link element. Rates of each<br />
erosion process were estimated from detailed<br />
150°E<br />
30°S 30°S<br />
0 250 500<br />
125 Km<br />
MDB<br />
Murray River<br />
Darling River<br />
Murrumbidgee River<br />
Goulburn Catchment<br />
Balonne Catchment<br />
Namoi Catchment<br />
Murrumbidgee Catchment<br />
Fig. 1. Location of the Murray–Darling Basin<br />
(MDB) in Australia. A hill-shaded version of the<br />
DEM in the background highlights the low relief<br />
of the MDB.<br />
1015
mapping of the controlling environmental factors.<br />
[Hughes <strong>and</strong> Prosser, 2003; Lu et al., 2003b]. A<br />
fraction of the gross amount of hillslope erosion in<br />
the catchments is delivered to rivers <strong>and</strong> this is<br />
accommodated through calculation of a hillslope<br />
sediment delivery ratio (γ i ) for each link area [Lu<br />
et al., 2003a].<br />
The mean annual yield of suspended sediment<br />
from the link is the total supply of suspended<br />
sediment to the link (S i ) less deposition on<br />
floodplains or in reservoirs (D i ). The suspended<br />
sediment budget for a link is:<br />
( γ )<br />
Y = S − D = T + E + G + B − D (1)<br />
i i i i i i i i i<br />
where the term in brackets is the total sediment<br />
supply (I i ) from the link element i.<br />
The mean annual delivery of sediment from a link<br />
element to a contribution point downstream (λ i ,<br />
t/y) is the sediment supply from the link element<br />
(I i ) multiplied by the sediment delivery efficiency<br />
through all river links (j = 1…M) along the route<br />
to the contribution point:<br />
M Y<br />
λ = I ∏ j<br />
S<br />
i i<br />
j = 1 j<br />
(2)<br />
The suspended sediment yield at a single sediment<br />
control location k can then be calculated by:<br />
N<br />
k<br />
= λ<br />
(3)<br />
i<br />
i=<br />
1<br />
Y<br />
where N is the total number of link elements<br />
contributing to sediment control location k.<br />
2.3 Investment Prioritization Scenarios<br />
We used four scenarios to mimic the types of<br />
management strategies that are currently being<br />
implemented or are under consideration: Scenario<br />
A: r<strong>and</strong>om management, where parts of river<br />
basins <strong>and</strong> particular erosion processes were<br />
chosen at r<strong>and</strong>om for treatment; Scenario B:<br />
investment prioritized to sediment sources, those<br />
places in the catchment with the highest erosion<br />
rates; Scenario C: prioritized to delivery to nearest<br />
streams, by combining information of erosion<br />
sources <strong>and</strong> hillslope sediment delivery, thereby<br />
seeing where it is effective to trap eroding soil, as<br />
opposed to preventing it from eroding upslope;<br />
<strong>and</strong> Scenario D: prioritized to delivery to control<br />
points, by fully utilizing the information resulting<br />
from the sediment budget including broad scale<br />
sediment deposition, i.e., focusing on the areas<br />
with particular erosion processes that contribute<br />
most to the suspended sediment loads.<br />
The four scenarios were implemented for each five<br />
percent incremental reduction (5-100%) in<br />
suspended load from current conditions to the<br />
conditions before European Settlement (minimum<br />
erosion <strong>and</strong> sediment transport activities which<br />
were predominated by natural processes only).<br />
The units to which we applied the control<br />
strategies were the link elements.<br />
2.4 Cost Estimation<br />
The costs of the primary management practices<br />
were obtained from the Goulburn-Broken<br />
Catchment Management Authority, Australia. The<br />
average per unit costs of reducing erosion rate for<br />
three types of erosion sources <strong>and</strong> hillslope<br />
sediment delivery ratio are summarized in the<br />
Table 1.<br />
Table 1. Estimated per unit costs for three types<br />
of erosion sources <strong>and</strong> per 1% of current hillslope<br />
sediment delivery ratio.<br />
Gully (per tonne) 130<br />
Riverbank (per tonne) 34<br />
Hillslope (per tonne) 80<br />
Unit Cost ($)<br />
γ (per 1% of current γ ) 9900<br />
3. RESULTS<br />
To underst<strong>and</strong> the relationship between sediment<br />
sources <strong>and</strong> their linkage to control locations we<br />
examine four catchments in some detail. The<br />
locations of the catchments are shown in Fig. 1.<br />
Suspended sediment contribution (%)<br />
Suspended sediment contribution (%)<br />
100<br />
90<br />
80<br />
70<br />
60<br />
50<br />
40<br />
30<br />
20<br />
10<br />
100<br />
90<br />
80<br />
70<br />
60<br />
50<br />
40<br />
30<br />
20<br />
10<br />
0<br />
0<br />
(a) Goulburn<br />
0 20 40 60 80 100<br />
(c) Murrumbidgee<br />
% contribution<br />
Gully proportion<br />
Riverbank proportion<br />
Sheetwash proportion<br />
0 20 40 60 80 100<br />
Proportion of catchment (%)<br />
100<br />
90<br />
80<br />
70<br />
60<br />
50<br />
40<br />
30<br />
20<br />
10<br />
0<br />
100<br />
90<br />
80<br />
70<br />
60<br />
50<br />
40<br />
30<br />
20<br />
10<br />
0<br />
(b) Namoi<br />
0 20 40 60 80 100<br />
(d) Balonne<br />
0 20 40 60 80 100<br />
Proportion of catchment (%)<br />
Fig. 2. Estimations of accumulative area<br />
contributions of suspended sediment in the (a)<br />
Goulburn, (b) Namoi, (c) Murrumbidgee <strong>and</strong> (d)<br />
Balonne catchments respectively. The relative<br />
proportions of suspended sediment contribution<br />
from each of the main erosion processes are also<br />
shown. The locations of the four catchments can<br />
be found in Fig. 1.<br />
1016
In the Goulburn <strong>and</strong> Murrumbidgee catchments<br />
the sources of sediment are predominantly from<br />
riverbank erosion (Fig. 2a). In the Namoi <strong>and</strong><br />
Balonne catchments the contributing sources are<br />
predominantly from hillslope sheet <strong>and</strong> rill<br />
erosion (Fig. 2b). The Goulburn <strong>and</strong> Namoi<br />
catchments have approximately the same degree<br />
of heterogeneity of the contributing sediment. The<br />
Murrumbidgee (Fig. 2c) has a strong degree of<br />
heterogeneity of sediment contribution compared<br />
to the more homogeneous Balonne (Fig. 2d), as<br />
indicated by the curvature of the accumulative<br />
sediment contribution by area (solid lines).<br />
Each of the four scenarios was run for each of the<br />
four example catchments. Fig. 3 shows the cost<br />
curves derived for each scenario for the four<br />
example catchments. Scenario A was run ten<br />
times for each catchment to give an indication of<br />
the r<strong>and</strong>om error range. For all cases, Scenario D<br />
represents the most cost-effective strategy. For<br />
some cases, Scenarios B <strong>and</strong> C are not necessarily<br />
better than r<strong>and</strong>om selection (Scenario A) (e.g.,<br />
Fig. 3b,d).<br />
When the sources of contributed sediment are<br />
dominantly sheet <strong>and</strong> rill erosion (Namoi <strong>and</strong><br />
Balonne catchments, Fig. 3b,d) scenarios which<br />
only consider the erosion source rates (with <strong>and</strong><br />
without local sediment delivery efficiency) can<br />
result in spending more money than r<strong>and</strong>om<br />
management. However, when the variable linkage<br />
between sediment source <strong>and</strong> the target control<br />
location is taken into account a radical<br />
improvement in cost-effectiveness can be achieved<br />
(Scenario D). This highlights the difference<br />
between erosion control for on-site productivity<br />
maintenance <strong>and</strong> off-site suspended sediment<br />
delivery. When the source is dominantly gully <strong>and</strong><br />
river bank (Goulburn catchment, Fig. 3a),<br />
Scenario A is the least effective (Fig. 3b,d).<br />
Fig. 3. Cost versus sediment reduction curves<br />
(cost curves) for the four example catchments<br />
shown in Fig. 1.<br />
We examine the effect of spatial scale on cost by<br />
altering the position of sediment control locations<br />
(where sediment targets will be set). Separately, in<br />
each catchment, we compared the total<br />
expenditure when sediment control locations are<br />
positioned at the catchment outlet with the case<br />
where they were nested within the catchment at<br />
particular channel sub-nodes. The 10-20 subnodes<br />
were arbitrarily chosen along the major<br />
tributaries within each catchment. Each sub-node<br />
receives sediment from around 30 – 50 up-stream<br />
link elements <strong>and</strong> the aim is to reduce the total<br />
load summed across all the sub-nodes. There some<br />
link elements that directly contribute to the<br />
catchment main control locations rather than any<br />
sub-node. We treated these link elements as an<br />
additional sub-catchment.<br />
1017
Total Cost ( Millions $ )<br />
Total Cost (Millions $)<br />
350<br />
300<br />
250<br />
200<br />
150<br />
100<br />
50<br />
0<br />
400<br />
350<br />
300<br />
250<br />
200<br />
150<br />
100<br />
50<br />
0<br />
Set targets at the whole catchment<br />
Set targets at the sub-nodes<br />
(a) Goulburn<br />
(c) Murrumbidgee<br />
0 0.2 0.4 0.6 0.8 1<br />
Proportion of sediment reduction<br />
300<br />
(b) Namoi<br />
250<br />
200<br />
150<br />
100<br />
0<br />
250<br />
By implementing Scenario D only, Fig. 4 shows<br />
that total expenditure by setting targets at subnode<br />
level is higher than by treating the<br />
catchments as a whole, for all percentage<br />
reductions. These results are consistent with the<br />
findings of Schleich et al. [1996]. Fig. 4 also<br />
50<br />
200<br />
150<br />
100<br />
50<br />
0<br />
(d) Balonne<br />
0 0.2 0.4 0.6 0.8 1<br />
Proportion of sediment reduction<br />
Fig. 4. Comparison of cost curves when control<br />
locations for suspended sediment targets are set<br />
at sub-nodes defining sub-catchments within the<br />
catchment, <strong>and</strong> at the catchment outlet for (a)<br />
Goulburn, (b) Namoi, (c) Murrumbidgee <strong>and</strong> (d)<br />
Balonne catchments respectively.<br />
shows that the difference is greater in some<br />
catchments than others. Larger cost savings are<br />
achieved in the Goulburn <strong>and</strong> Murrumbidgee<br />
catchments by treating the catchment as a whole<br />
(shown in Fig. 4a,c). The differences are caused<br />
by the patterns of the main sediment sources, their<br />
relationship to the control locations, <strong>and</strong> the<br />
choice of control locations themselves. Unlike the<br />
Namoi <strong>and</strong> Balonne catchments (Fig. 4b,d) where<br />
most of the sediment is contributed by sheet <strong>and</strong><br />
rill erosion in upl<strong>and</strong>s, most of the sediment in the<br />
Goulburn <strong>and</strong> Murrumbidgee catchments<br />
(Fig.4a,c) is contributed by bank erosion from the<br />
link elements along the major channels. For<br />
catchments like the Goulburn <strong>and</strong> Murrumbidgee,<br />
setting the same percentage of sediment reduction<br />
targets at sub-nodes within the catchment often<br />
misses the opportunity to prioritize investment<br />
along the main channels, where sediment is<br />
directly transported to the main control locations,<br />
resulting in unnecessary expenditure in the upl<strong>and</strong><br />
areas, in which eroded sediment is deposited<br />
locally. Apart from the internal heterogeneity of<br />
contributing sediment, the relative differences in<br />
total expenditure can be also influenced by other<br />
factors such as the number of reservoirs,<br />
floodplain deposition <strong>and</strong> the amount of regulated<br />
flow for irrigation (e.g. sediment lost in the<br />
system due to the loss of the flow).<br />
1018
Fig. 5. Spatial distribution of investment to achieve a 70% reduction in suspended sediment with<br />
the control location set at the catchment outlet. Two catchments are shown – the one to the left<br />
(Murrumbidgee) has greater heterogeneity of spatial distribution of sediment contribution to the<br />
control location than the one to the right (Balonne). (a) total expenditure, (b) hill slope erosion<br />
reduction (in difference, the same hereafter), (c) hillslope sediment delivery ratio reduction, (d)<br />
gully erosion reduction, (e) bank erosion reduction.<br />
Maps can be produced from each scenario of total<br />
expenditure, <strong>and</strong> reductions of hill slope erosion,<br />
hill slope sediment delivery ratio (where<br />
considered), gully erosion <strong>and</strong> bank erosion. Fig.<br />
5 shows the most cost effective strategy (Scenario<br />
D) for a 70% reduction in suspended sediment<br />
loads at the catchment outlet. The Murrumbidgee<br />
catchment (on the left side in Fig. 5) has a greater<br />
concentration of proposed expenditure than the<br />
Balonne catchment. This reflects that greater<br />
curvature of the accumulative area contribution<br />
function, which indicates a more heterogeneous<br />
sediment contribution, results in a more<br />
concentrated pattern of expenditure.<br />
4. DISCUSSION AND CONCLUSION<br />
We proposed a range of investment prioritization<br />
scenarios to identify the most cost-effective<br />
strategy to control suspended sediment loads. We<br />
demonstrated that a spatially-distributed sediment<br />
budget approach provided a rational basis to<br />
determine an optimum strategy for cost-effective<br />
sediment control. We showed that appropriate<br />
investment prioritization can potentially offer<br />
large cost savings as the magnitude <strong>and</strong><br />
distribution of costs can vary by several times<br />
depending on what type of erosion source or<br />
sediment delivery is targeted in a spatially varying<br />
manner. Target settings which only consider the<br />
1019
erosion source rates can potentially result in<br />
spending more money than r<strong>and</strong>om management<br />
intervention.<br />
Heterogeneity of sediment contribution is the<br />
physical factor leading to potential cost saving.<br />
We have shown that the greater the degree of<br />
internal heterogeneity, the larger the cost saving<br />
through prioritization. It is more cost-effective to<br />
prioritize the investment at large basin area than<br />
at sub-catchment level because it better utilizes<br />
spatial heterogeneity. This raises the prospect that<br />
bodies responsible for setting suspended targets<br />
could benefit greatly from examining the tradeoffs<br />
between cost savings in control measures <strong>and</strong><br />
the costs of installing or moving monitoring<br />
stations, for example. Another consideration is<br />
how the results might be used to inform the<br />
market in provision of the services required to<br />
control sediment sources at different spatial<br />
scales. It is likely that other issues will exhibit<br />
spatial heterogeneity, e.g., pollutant sources, <strong>and</strong><br />
opportunities for maximizing the value from<br />
investment in control could be realized by<br />
considering scale <strong>and</strong> heterogeneity in selecting<br />
locations for target setting.<br />
5. ACKNOWLEDGEMENT<br />
This study was partially funded by the Murray<br />
Darling Basin Commission under Project D10012.<br />
6. REFERENCES<br />
ABS, Environment condition of Australia’s<br />
freshwater resources, Year book Australia,<br />
Australian Bureau of Statistics, 2002.<br />
DeRose R. C., I. P. Prosser, M. Weisse, <strong>and</strong> A. O.<br />
Hughes, Summary of sediment <strong>and</strong> nutrient<br />
budgets for the Murray-Darling Basin,<br />
Technical Report 15/01, CSIRO L<strong>and</strong> <strong>and</strong><br />
Water, Canberra, 2003.<br />
http://www.clw.csiro.au/publications/technical<br />
2003.<br />
Environment Australia, Reef Water Quality<br />
Protection Plan, 2003.<br />
Gianessi, L. P., <strong>and</strong> H. M. Peskin, Analysis of<br />
national water pollution control policies 2.<br />
Agricultural sediment control, Water Resour.<br />
Res., 17, 803-821, 1981.<br />
Gren, I. M., <strong>International</strong> versus national actions<br />
against nitrogen pollution of the Baltic Sea,<br />
Environ. <strong>and</strong> Resour. Econo., 20, 41-59, 2001.<br />
Gren, I. M., G. Destouni, <strong>and</strong> H. Scharin, Cost<br />
effective management of stochastic coastal<br />
water pollution, Environ. <strong>Modelling</strong> <strong>and</strong><br />
Assessment, 5, 193-203, 2000.<br />
HELCOM, The Baltic Sea joint comprehensive<br />
environmental action programme, Baltic Sea<br />
<strong>Environmental</strong> Proceedings, No. 48, Helsinki,<br />
Finl<strong>and</strong>, 1993.<br />
Hughes, A. O., <strong>and</strong> I. P. Prosser, Gully <strong>and</strong><br />
riverbank erosion mapping for the Murray-<br />
Darling Basin, Technical Report 3/03, CSIRO<br />
L<strong>and</strong> <strong>and</strong> Water, Canberra, 20p., 2003.<br />
http://www.clw.csiro.au/publications/technical<br />
2003.<br />
Lu, H., C. J. Moran, <strong>and</strong> I. P. Prosser, <strong>Modelling</strong><br />
sediment delivery ratio over the Murray<br />
Darling Basin. Proceedings of MODSIM 2003,<br />
Townsville, 485-491, 2003a.<br />
Lu, H., I. P. Prosser, C. J. Moran, J. Gallant, G.<br />
Priestley, <strong>and</strong> J. G. Stevenson, Predicting<br />
sheetwash <strong>and</strong> rill erosion over the Australian<br />
continent, Aust. J. Soil Res., 41, 1-26, 2003b.<br />
NAP, National Action Plan for Salinity <strong>and</strong> Water<br />
Quality, Commonwealth of Australia, 2003.<br />
http://www.napswq.gov.au/<br />
NLWRA, Australian Agriculture Assessment.<br />
National L<strong>and</strong> <strong>and</strong> Water Resources Audit,<br />
Canberra, 2001.<br />
Prosser, I. P., B. Young, P. Rustomji, C. Moran,<br />
<strong>and</strong> A.O. Hughes, Constructing river basin<br />
sediment budgets for the National L<strong>and</strong> <strong>and</strong><br />
Water Resources Audit, Technical Report<br />
15/01, CSIRO L<strong>and</strong> <strong>and</strong> Water, Canberra,<br />
2001.http://www.clw.csiro.au/publications/tech<br />
nical2001.<br />
Schleich, J., D. White, <strong>and</strong> K. Stephenson, Cost<br />
implications in achieving alternative water<br />
quality targets, Water Resour. Res., 32, 2879-<br />
2884, 1996.<br />
Wisconsin Department of Natural Resources<br />
(WDNR). Lower Green Bay remedial action<br />
plan: Desired future state. WDNR Doc.<br />
PUBLR-WR-175-87, Madison, 1988.<br />
1020
Appropriate Accuracy of Models for Decision-Support<br />
Systems: Case Example for the Elbe River Basin<br />
Jean-Luc de Kok 1 (j.l.dekok@ctw.utwente.nl), Koen U. van der Wal 1 , <strong>and</strong> Martijn J. Booij 1<br />
1 Department of Water Engineering <strong>and</strong> Management, Faculty of Engineering Technology, University of<br />
Twente, PO Box 217, 7500 AE, Enschede, The Netherl<strong>and</strong>s<br />
Abstract: Given the growing complexity of water-resources management there will be an increasing need<br />
for integrated tools to support policy analysis, communication, <strong>and</strong> research. A key aspect of the design is the<br />
combination of process models from different scientific disciplines in an integrated system. In general these<br />
models differ in sensitivity <strong>and</strong> accuracy, while non-linear <strong>and</strong> qualitative models can be present. The current<br />
practice is that the preferences of the designers of a decision-support system, <strong>and</strong> practical considerations<br />
such as data availability guide the selection of models <strong>and</strong> data. Due to a lack of clear scientific guidelines the<br />
design becomes an ad-hoc process, depending on the case study at h<strong>and</strong>, while selected models can be overly<br />
complex or too coarse for their purpose. Ideally, the design should allow for the ranking of selected<br />
management measures according to the objectives set by end users, without being more complex than<br />
necessary. De Kok <strong>and</strong> Wind [2003] refer to this approach as appropriate modeling. A good case example is<br />
the ongoing pilot project aiming at the design of a decision-support system for the Elbe river basin. Four<br />
functions are accounted for: navigability, floodplain ecology, flooding safety, <strong>and</strong> water quality. This paper<br />
concerns the response of floodplain biotope types to river engineering works <strong>and</strong> changes in the flooding<br />
frequency of the floodplains. The HBV-D conceptual rainfall-runoff model is used to simulate the impact of<br />
climate <strong>and</strong> l<strong>and</strong> use change on the discharge statistics. The question was raised how well this rainfall-runoff<br />
model should be calibrated as compared to the observed discharge data. Sensitivity analyses indicate that a<br />
value of R 2 = 0.87 should be sufficient.<br />
Keywords: decision-support system; river-basin management; appropriate modeling; rainfall-runoff; Elbe<br />
1. INTRODUCTION<br />
The Elbe is one of the largest rivers in Central<br />
Europe. Water quality in the river is affected by<br />
agricultural runoff, while settlements along the<br />
river form important point sources of pollution. In<br />
terms of shipping density the river is second only<br />
after the Rhine in Germany. Planned <strong>and</strong> ongoing<br />
river engineering works aimed at improving the<br />
navigability of the river <strong>and</strong> reducing the risk of<br />
flooding include large-scale dike displacement,<br />
groyne restoration, <strong>and</strong> excavation of the river bed<br />
<strong>and</strong> floodplains. Several sections of the river have<br />
been designated as protected nature reserves with<br />
vegetation types that form a habitat for rare fauna.<br />
It is not yet clear how the hydromorphological<br />
consequences of the river engineering works will<br />
affect the vegetation conditions along the river.<br />
The formulation of an optimal management<br />
strategy requires in-depth underst<strong>and</strong>ing of the<br />
interaction of these measures with the socialeconomic,<br />
ecological, <strong>and</strong> physical processes at<br />
different scale levels. Since a methodology <strong>and</strong> the<br />
instruments for integrated river-basin management<br />
were not available, the German Federal Institute of<br />
Hydrology initiated the project ‘Towards a generic<br />
tool for river basin management’ [De Kok et al.,<br />
2000]. The ultimate goal is to develop a prototype<br />
decision-support system (DSS), which helps the<br />
water managers to formulate an effective strategy<br />
for sustainable management of the Elbe river basin.<br />
The four functions included in the DSS are: inl<strong>and</strong><br />
shipping, water quality, floodplain vegetation, <strong>and</strong><br />
flooding safety. In view of the multi-objective<br />
nature of the prototype <strong>and</strong> scale differences of<br />
models <strong>and</strong> data, the choice has been made for a<br />
modular design with three scale levels: catchment,<br />
main channel (including floodplains), <strong>and</strong> river<br />
section (a section of 20 km). Figure 1 schematizes<br />
how the hydraulic <strong>and</strong> ecological models are<br />
integrated in the Elbe-DSS. The research question<br />
addressed in this paper is how well a hydrological<br />
model should be calibrated in relationship to the<br />
accuracy of the floodplain ecology model. The<br />
example pertains to a section of the Elbe River<br />
near the town of Tangermünde in Saxony-Anhalt,<br />
which is where one of the gauge stations is located.<br />
1021
River Basin<br />
Catchment Module<br />
urban<br />
wastewaters<br />
agriculture<br />
forestry<br />
HYDROLOGY<br />
LAND USE<br />
Point<br />
Diffuse<br />
1D RIVER WATER QUALITY<br />
+ TOXICANTS<br />
flood<br />
risk<br />
socialeconomic<br />
state<br />
1D HYDROMORPHOLOGY MODULE<br />
shipping<br />
River Module<br />
chemical<br />
state<br />
groyne<br />
modification<br />
dike<br />
shifting<br />
flood frequency,<br />
discharge<br />
s<strong>and</strong><br />
suppletion<br />
ecological<br />
state<br />
1D/2D HYDROMORPHOLOGY MODULE<br />
ECOLOGICAL MODULE<br />
(floodplain/river/bank)<br />
Floodplain Module<br />
Figure 1. Outline of the integration of models in the prototype decision-support system for the Elbe river<br />
basin (dotted lines indicating the three scale levels: catchment, main channel, <strong>and</strong> river section)<br />
[De Kok et al., 2000].<br />
1022
are<br />
2 MODEL BASE<br />
2.1 Floodplain Ecology<br />
The response of the biotope types in the<br />
floodplains to changing hydraulic conditions <strong>and</strong><br />
river engineering works is based on the rule-based<br />
model MOVER (MOdel of VEgetation Repsonse)<br />
described by Fuchs et al. [2002]. This model has<br />
been developed by the German Federal Institute<br />
of Hydrology for the floodplains of the river<br />
Rhine <strong>and</strong> is currently extended to the Elbe River.<br />
The model consists of a matrix, with rows<br />
indicating the dominant biotope types <strong>and</strong> the<br />
columns indicating the abiotic parameters.<br />
MOVER is based on a statistical approach, with<br />
the flood duration (total number of flooding days<br />
per year) as key input variable. The flood duration<br />
is determined from the statistical distribution of<br />
the daily average discharge, digital elevation data<br />
<strong>and</strong> water levels in the main channel. The latter<br />
are calculated as a function of the discharge by<br />
means of a 1D stationary flow model which was<br />
calibrated for the Elbe River by Otte-Witte et al.<br />
[2002]. The modeled relationship between water<br />
level h <strong>and</strong> discharge q can be described by a<br />
rating curve:<br />
large area <strong>and</strong> an important role for snow. It is a<br />
relatively simple model, in which the climate data<br />
are transformed into a base-flow discharge <strong>and</strong> a<br />
quick runoff discharge, as shown in Figure 2.<br />
Several versions for more specific situations or for<br />
more differentiated approaches were developed<br />
<strong>and</strong> nowadays a wide range of applications of the<br />
model can be found [Bergström, 1995].<br />
Krysanova et al. (1999) developed the HBV-D<br />
model used in this study, in which a basin can be<br />
subdivided into subbasins, <strong>and</strong> a more<br />
differentiated l<strong>and</strong> use definition is applied. At the<br />
moment, the conceptual hydrological model<br />
HBV-D is calibrated for the Elbe river basin.<br />
b<br />
h ( q)<br />
= a q<br />
(1)<br />
where a <strong>and</strong> b are parameters. The flood duration<br />
is given by<br />
N year<br />
( q )<br />
365 ln<br />
= × 1−<br />
erf <br />
*<br />
2 σ<br />
− µ <br />
<br />
2 <br />
(2)<br />
1/ b<br />
where ( z )<br />
q * = a is the critical discharge for<br />
inundation of a location with elevation z , <strong>and</strong><br />
<strong>and</strong> the mean <strong>and</strong> st<strong>and</strong>ard deviation for the<br />
daily average discharge. The desired accuracy of<br />
the number of flooding days depends on the<br />
sensitivity of the ecological model. The rule<br />
matrix of MOVER distinguishes differences in the<br />
flood duration of ten days per year. For most<br />
biotope types even larger ranges in the flood<br />
duration will lead to identical maps.<br />
̌ ̅<br />
2.2 Rainfall-Runoff<br />
The daily discharge statistics were obtained with<br />
historic time series for the period 1964-1995,<br />
which have been analyzed by Helms et al. [2002].<br />
The HBV model was developed by Bergström in<br />
1976 [1995]. The initial goal of the HBV model<br />
was real-time flood simulation under typical<br />
Swedish conditions, which means basins with a<br />
Figure 2: HBV model structure<br />
[Van der Wal, 2001].<br />
3 DISCHARGE STATISTICS<br />
The hydrological model will be used to generate<br />
discharge time series for the average regime <strong>and</strong><br />
flood events under various climate change<br />
scenarios. The question was raised how accurately<br />
the hydrological model should be calibrated on<br />
existing data. This is important in view of the<br />
effort to be spent on calibration <strong>and</strong> data<br />
collection. A common indicator for the quality of<br />
hydrological models is the Nash-Sutcliffe<br />
coefficient proposed by Nash <strong>and</strong> Sutcliffe [1970]<br />
<strong>and</strong> denoted by:<br />
1023
R<br />
2<br />
=<br />
1<br />
N<br />
N tot<br />
∑<br />
t<br />
tot t = 1<br />
1−<br />
2<br />
2<br />
( Q ' −Q<br />
)<br />
σ<br />
t<br />
(3)<br />
where Q t ’ <strong>and</strong> Q t denote the modeled <strong>and</strong> historic<br />
discharge time series, <strong>and</strong> σ is the st<strong>and</strong>ard<br />
deviation. To simulate the output of the HBV<br />
model different discharge time series can be<br />
generated by adding an auto-correlated noise term<br />
ε to the original data:<br />
t<br />
where<br />
Q ' =<br />
t<br />
Q t<br />
+ ε<br />
t<br />
(4)<br />
compare the time series on the basis of the<br />
percentage of years, for which the total flood<br />
duration did not differ more than ten days from<br />
the value for the historic time series i.e.<br />
∆N<br />
year<br />
≤ 10<br />
(6)<br />
In anticipation of a vegetation succession model it<br />
makes sense to examine the difference at the scale<br />
of months as well. Assuming independence of the<br />
flood duration between different months the<br />
criterion<br />
∆N<br />
can be used.<br />
month<br />
∆N<br />
year<br />
≤<br />
12 ≈ 3<br />
(7)<br />
ε<br />
t<br />
δ Q αε<br />
=<br />
t t<br />
+<br />
t−1<br />
(5)<br />
4. CASE EXAMPLE<br />
with δ<br />
t<br />
a scaling factor drawn from a r<strong>and</strong>om<br />
uniform distribution in the interval [-B, +B], <strong>and</strong><br />
α the autocorrelation of the differenceQ ' −Q<br />
.<br />
In this way the st<strong>and</strong>ard deviation of the time<br />
series can be changed without affecting the mean<br />
of the discharge. This can be expected under<br />
changing climate conditions [Booij, 2002]. The<br />
reason is that the mean of the distribution will be<br />
described correctly by calibration of the water<br />
balance. To obtain a reasonable value the<br />
parameter α was taken from a calibrated HBV<br />
model for the Meuse river basin [Booij, 2002],<br />
because the catchments are similar <strong>and</strong> both the<br />
Elbe <strong>and</strong> Meuse are rainfed rivers. The range B<br />
does not depend on the value of α . Its magnitude<br />
can be varied to generate discharge time series<br />
Q with different values of the quality index R 2 .<br />
'<br />
t<br />
The obvious approach would be to increase the<br />
value B until the difference in flood duration for<br />
the artificial <strong>and</strong> historical discharge time series<br />
reaches a value approximating the ten-day<br />
accuracy required for the MOVER model.<br />
Unfortunately, this approach will not result in<br />
meaningful estimates for R 2 . This can be<br />
addressed to the statistical character of the<br />
ecological model, which does not match the<br />
dynamic nature of the rainfall-runoff model. A<br />
discharge time series of poor quality can have a<br />
st<strong>and</strong>ard deviation close to the value for the<br />
historic data. Sensitivity analyses proved that the<br />
MOVER model was not very sensitive to the<br />
st<strong>and</strong>ard deviation of the time series. For the<br />
selected location substantial differences in the<br />
biotope type distribution occur only for changes in<br />
σ larger than 15 %. Hence, a difference of ten<br />
days would correspond to unrealistically low<br />
values of R 2 . For this reason we decided to<br />
t<br />
t<br />
The parameter values for the selected site near the<br />
Tangermünde gauge station (Elbe km 388.2) are<br />
given in Table 1 below.<br />
parameter<br />
value<br />
µ (m 3 s -1 ) 6.18<br />
σ (m 3 s -1 ) 0.56<br />
a 20.97<br />
b 0.061<br />
z (m + sea level) 32.4<br />
Table 1. Discharge <strong>and</strong> hydraulic parameters for<br />
the study site.<br />
Near Tangermünde the flood plains are relatively<br />
flat with an average elevation of 32.4 m. above<br />
sea level on the right bank. This leads to an<br />
average flood duration of twenty-five days per<br />
year. Artificial discharge time series were<br />
generated by varying the value of B in the range<br />
[0.05, 0.30]. For α the value of 0.82 was found<br />
for the Meuse river basin [Booij, 2002]. Figures 3<br />
<strong>and</strong> 4 show the percentage of years <strong>and</strong> months,<br />
which do not satisfy the criteria of (6) <strong>and</strong> (7)<br />
against the value of R 2 .<br />
1024
Figure 3. Percentage of years with flood duration<br />
that is different according to criterion (6) as a<br />
function of R 2 .<br />
Depending on what percentage of years or month<br />
with different flood duration is accepted one can<br />
decide which value of R 2 is sufficient for the<br />
rainfall-runoff model. For example, a ten percent<br />
difference indicates that R 2 should be around<br />
0.87, which can be considered feasible for the<br />
calibration. The corresponding value of B is<br />
0.20.<br />
Figure 4. Percentage of months with different<br />
flood duration according to (7) as function of R 2 .<br />
The step structure of the curves shown in Figures<br />
3 <strong>and</strong> 4 is a consequence of the definition of R 2 ,<br />
which is based on daily discharge data. Time<br />
series differing in R 2 do not necessarily differ<br />
according to criteria (6) <strong>and</strong> (7). Figure 5 shows a<br />
sample of 365 days for the observed <strong>and</strong><br />
simulated times series for a value of B = 0.20. In<br />
general these results indicate that calibration<br />
should be possible in view of the desired accuracy<br />
of the discharge distribution, provided one accepts<br />
a deviation in the flood duration above the criteria<br />
(6) <strong>and</strong> (7) for 10 % of the months or years. For<br />
comparison the calculation was repeated at the<br />
level of days as well. A value of R 2 = 0.87 turned<br />
out to correspond to falsely predicted flooding in<br />
3 % of the days over the 35-year period.<br />
Figure 5. One-year sample of observed (solid)<br />
<strong>and</strong> simulated (dashed) discharge time series.<br />
5. CONCLUSIONS<br />
There exist no scientific st<strong>and</strong>ards to measure<br />
when it is appropriate to integrate different<br />
models in a decision-support system. This makes<br />
the design an ad-hoc process. This problem<br />
becomes more prominent when statistical <strong>and</strong><br />
dynamic models are used in combination. The<br />
integration of a dynamic rainfall-runoff model<br />
with a statistical model for the biotope types of<br />
the floodplains along the Elbe River served as a<br />
case example to show how the problem could be<br />
addressed. The sensitivity of the ecological<br />
model for changes in the discharge statistics<br />
proved to be low. In this case direct sensitivity<br />
analyses will be of limited use to determine the<br />
required quality of the input discharge time series.<br />
Instead it is better to compare simulated discharge<br />
time series of different quality with historic data.<br />
Depending on the sensitivity of the ecological<br />
model for changes in the discharge distribution<br />
one can formulate a criterion for the acceptable<br />
accuracy of the time series. This will indicate how<br />
good the rainfall-runoff model should be<br />
calibrated. For the study site at Tangermünde<br />
calibration seems feasible at the level required for<br />
predicting vegetation response.<br />
6. REFERENCES<br />
Bergström, S., 1995. The HBV model, V.P.<br />
Singh: Computer Models of watershed<br />
hydrology, Water Resour. Publ.,<br />
Highl<strong>and</strong>s Ranch, 443-476, 1995.<br />
Booij M.J., Appropriate modelling of climate<br />
change impacts on river flooding, Ph.D.-<br />
thesis, University of Twente, Enschede,<br />
The Netherl<strong>and</strong>s, 2002.<br />
De Kok J.L., Wind H.G., Delden H. <strong>and</strong> Verbeek<br />
M., Towards a generic tool for river<br />
basin management. Feasibility<br />
assessment for a prototype DSS for the<br />
Elbe. Feasibility study - report 2/3. Final<br />
report., University of Twente, Enschede,<br />
2000.<br />
De Kok J.L. <strong>and</strong> Wind H.G., Design <strong>and</strong><br />
application of decision-support systems<br />
for integrated water management: lessons<br />
to be learnt, Special Issue Physics <strong>and</strong><br />
Chemistry of the Earth, part B:<br />
Hydrology, Oceans, <strong>and</strong> Atmosphere,<br />
28 (14-15), 571-578, 2003.<br />
1025
Fuchs E., Giebel H., Hettrich A., Huesing V.,<br />
Rosenzweig S. <strong>and</strong> Theis H.-J., Einsatz<br />
von ökologischen Modellen in der<br />
Wasser- und Schifffahrtsverwaltung, das<br />
integrierte Flussauenmodell INFORM. –<br />
BfG – Mitteilung Nr. 25, Koblenz ,2002.<br />
Helms M., Büchele B., Merkel U., <strong>and</strong> Ihringer J.,<br />
Statistical analysis of the flood situation<br />
<strong>and</strong> assessment of the impact of diking<br />
measures along the Elbe (Labe) river,<br />
Journal of Hydrology, <strong>Volume</strong> 267, Issues<br />
1-2, 94-114, 2002.<br />
Krysanova V., Bronstert A., Wohlfeil D.-I.,<br />
<strong>Modelling</strong> river discharge for large<br />
drainage basins: from lumped to<br />
distributed approach, Hydrolog. Sci. J.<br />
44 (2), 313-331, 1999.<br />
Nash J.E. <strong>and</strong> Sutcliffe J.V., River flow<br />
forecasting through conceptual models,<br />
Part I – A discussion of principles,<br />
Journal of Hydrology, 10, 282-290,<br />
1970.<br />
Otte-Witte K., Adam K., Meon G. <strong>and</strong> Rathke K.,<br />
Hydraulisch-morphologische Charakteristika<br />
entlang der Elbe, in:<br />
Morphodynamik der Elbe,<br />
Schlussbericht des BMBF-<br />
Verbundprojektes mit Einzelbeiträgen<br />
der Partner und Anlagen-CD, 203-299,<br />
Nestmann F. <strong>and</strong> Büchele B. (eds.),<br />
Institut für Wasserwirtschaft und<br />
Kulturtechnik der Universität Karlsruhe,<br />
Karlsruhe, 2002.<br />
Van der Wal K.U., Meuse Model Moulding, On<br />
the effect of spatial resolution, MSc. thesis,<br />
Department of Civil Engineering,<br />
University of Twente, The Netherl<strong>and</strong>s,<br />
December 2001.<br />
1026
River Basin Management Plans <strong>and</strong> Decision Support<br />
C. Giupponi a,c , R. Camera b , V. Cogan c , A. Fassio c<br />
a Università degli Studi di Milano, Italy - carlo.giupponi@unimi.it<br />
b<br />
Fondazione Eni Enrico Mattei, Milan, Italy<br />
c Fondazione Eni Enrico Mattei, Venice, Italy<br />
Abstract: The EU Water Framework Directive (WFD) aims at establishing “a framework for the<br />
protection of inl<strong>and</strong> surface waters, transitional waters, coastal waters <strong>and</strong> groundwaters”, (Dir. 2000/60/EC,<br />
art.1) for all European Member States. The extent to which protection <strong>and</strong> management of the water<br />
environment are approached in an integrated <strong>and</strong> holistic way is one of the innovative aspects of the WFD. In<br />
order to implement such approach, the WFD foresees the establishment of a Programme of Measures (PM),<br />
<strong>and</strong> the development of a River Basin Management Plan (RBMP) for each European River Basin District<br />
(RBD) with articles 11 <strong>and</strong> 13 respectively. To fulfil these requirements, planners need a methodology that<br />
integrates environmental, social <strong>and</strong> economic concerns <strong>and</strong> that may involve interested parties in the<br />
formulation of strategies. The MULINO Project (EVK1-2000-00082) has developed a methodology <strong>and</strong> a<br />
Decision Support System (DSS) that tackles such problems. This paper explains how the MULINO<br />
methodology <strong>and</strong> its software tool (mDSS) structure <strong>and</strong> manage contributions from decision makers, experts<br />
<strong>and</strong> stakeholders to elicit environmental objectives, identify pressures, analyse human impacts, <strong>and</strong> make a<br />
choice between alternative measures. Links are made with the planning procedures prescribed in the WFD to<br />
present how the use of MULINO in WFD implementation could help water authorities meet their obligations,<br />
<strong>and</strong> demonstrate a management approach that is coherent with the new requirements.<br />
Keywords: Decision Support System; Integrated Water Management; Multi-Criteria Analysis; Water<br />
Planning.<br />
1. INTRODUCTION<br />
The project MULINO (MULti-sectoral, INtegrated<br />
<strong>and</strong> Operational Decision Support System for<br />
Sustainable Use of Water Resources at the<br />
Catchment Scale) was funded under the Fifth<br />
Framework Programme of the European Union<br />
(EVK1-2000-00082) <strong>and</strong> aimed at developing a<br />
DSS tool to assist water authorities in the<br />
management of water resources 1 . Specific aims<br />
were improving the quality of decision making <strong>and</strong><br />
achieving a truly integrated approach to river basin<br />
management. By supporting the integration of<br />
socio-economic <strong>and</strong> environmental modelling<br />
techniques with GIS functions <strong>and</strong> multi-criteria<br />
1 The MULINO Consortium: Fondazione ENI Enrico Mattei<br />
(Italy), Universidade Atlântica (Portugal), Université<br />
Catholique de Louvain (Belgium), Silsoe Research Institute<br />
(UK), European Commission Joint Research Centre, Centre for<br />
Advanced Studies, Research <strong>and</strong> Development in Sardinia,<br />
(Italy), Research Institute of Soil Science <strong>and</strong> Agrochemistry of<br />
Bucharest (Romania), Fundatia Pentru Tehnologia Informatiei<br />
Aplicate in Mediu, Agricultura si Schimbari Globale<br />
(Romania), Institute of Water <strong>and</strong> Environment, Cranfield<br />
University (UK).<br />
decision aids, the MULINO DSS (mDSS) aspires<br />
to be an operational tool which meets the needs of<br />
European water management authorities <strong>and</strong> which<br />
facilitates the implementation of the EU Water<br />
Framework Directive (WFD).<br />
After a brief introduction to the MULINO<br />
methodology, this paper introduces the general<br />
application context in which project outputs might<br />
be used to support WFD implementation. Specific<br />
reference is made to two of the Common<br />
Implementation Strategy (CIS) guidance<br />
documents that were available at the end of 2003:<br />
the Guidance on the Planning Process [EC, 2003a]<br />
<strong>and</strong> the IMPRESS document for the analysis of<br />
pressures <strong>and</strong> impacts [EC, 2002a].<br />
2. THE MULINO PROJECT<br />
The specific application context for the MULINO<br />
methodology <strong>and</strong> the mDSS software is defined in<br />
terms of a decision which will affect the use of<br />
water resources. Such a decision might be related<br />
to ordinary water management activities or be<br />
1027
connected to unusual events. The methodology has<br />
been designed with water authorities as the target<br />
users, <strong>and</strong> its application would involve decision<br />
makers <strong>and</strong> technicians. The terms “decision<br />
maker” <strong>and</strong> “user” are used indiscriminately. It is<br />
envisaged that MULINO could be applied within<br />
the planning process required for the<br />
implementation of the Water Framework<br />
Directive. In particular, the method might be used<br />
to support the design of the programmes of<br />
measures (PMs) <strong>and</strong> to develop the River Basin<br />
Management Plans (RBMPs) for a River Basin<br />
District (RBD) or specific plans for sub-basins<br />
within the RBD. According to the requirements of<br />
the WFD, the river basin authority should<br />
implement a series of decisional processes at<br />
various scales <strong>and</strong> involve interested stakeholders<br />
during this process (Article 14 of the WFD).<br />
Figure 1. Flowchart of the MULINO methodology<br />
The MULINO methodology takes the user through<br />
a process that begins with describing <strong>and</strong><br />
structuring a water management problem, involves<br />
selected stakeholders in information sharing, <strong>and</strong><br />
culminates in identifying a final choice between<br />
possible actions. By using the mDSS software tool,<br />
the user approaches the choice among a finite set<br />
of options through Multi-Attribute Analysis<br />
(MAA) methods. MAA decision rules are used by<br />
mDSS to identify the “best” option. In particular<br />
mDSS guides the user through three decision<br />
phases: “Conceptual phase”; “Design phase”, <strong>and</strong><br />
“Choice phase” [Simon, 1960].<br />
The mDSS tool is one of the components of the<br />
MULINO methodology, which starts from the<br />
formalisation of a problem which triggers a<br />
decisional process in which various actors are<br />
involved. A typical list of involved parties can<br />
include the decision making body (including<br />
policy makers <strong>and</strong> technicians), other<br />
administrations at higher <strong>and</strong> lower levels,<br />
associations of various economic sectors,<br />
concerned citizens’ groups, research organisations,<br />
environmental groups, <strong>and</strong> water companies.<br />
The MULINO approach anticipates a decisional<br />
process based upon the phases shown in figure 1<br />
above. Various actors can be involved in the<br />
process, their contributions co-ordinated by the<br />
water management authority responsible for<br />
decision implementation. The mDSS can be used<br />
throughout to document the selection of criteria<br />
<strong>and</strong> the preferences of the various parties, as well<br />
as to select the preferred option given the set of<br />
choices that have been made to set up the decision<br />
problem.<br />
Conceptual Phase<br />
In a typical application, the first step is to<br />
identify the study area. Once this has been<br />
done, its socio-economic <strong>and</strong><br />
environmental characteristics are described<br />
according to the DPSIR conceptual<br />
framework (Driving forces, Pressures,<br />
State, Impact <strong>and</strong> Response) [EEA, 1999].<br />
Causal relationships <strong>and</strong> dynamic<br />
interactions within the catchment are<br />
conceptualised in a procedure through<br />
which the user is asked to construct DPS<br />
“chains” in order to identify the main<br />
cause-effect relationships between human<br />
activities <strong>and</strong> the state (or change of state)<br />
of water resources. This first phase is<br />
termed “Conceptual Phase”. The<br />
MULINO methodology introduces a local<br />
network study to be completed through a<br />
series of interviews with selected<br />
stakeholders, <strong>and</strong> the application of<br />
modelling tools to analyse the dynamic aspects of<br />
the water cycle. The decision maker can structure<br />
the problem in collaboration with stakeholders,<br />
through a questionnaire targeted to the decisional<br />
problem in question. The socio-economic <strong>and</strong><br />
environmental information is stored in appropriate<br />
catalogues <strong>and</strong> organised according to the DPSIR<br />
approach in various formats allowing the user to<br />
deal with spatial <strong>and</strong> temporal data series.<br />
The user is then ready to enter the “Design Phase”<br />
where he/she describes the alternative options,<br />
selects the decisional criteria taking into account<br />
the results of the local network analysis <strong>and</strong> the<br />
results of data coming from surveys, census,<br />
monitoring <strong>and</strong> modelling are stored in the<br />
Analysis Matrix (AM). The AM is structured with<br />
options in the columns <strong>and</strong> decisional criteria in<br />
the rows.<br />
Design Phase<br />
Choice Phase<br />
The evaluation, normalisation <strong>and</strong> weighting of<br />
1028
the multidimensional data stored in the AM takes<br />
the decision maker to the “Choice Phase” in<br />
which the Evaluation Matrix (EM) is built <strong>and</strong> one<br />
or more decision rules are applied to identify the<br />
“best” option. Local network questionnaires are<br />
designed to support public participation by<br />
collecting structured information about<br />
stakeholders’ preferences that relate to the decision<br />
problem. These preferences can be combined in<br />
the mDSS’s group decision making routine.<br />
In this final phase, the mDSS software allows the<br />
user to analyse how the variables influence the<br />
selection of the “best” option through the<br />
sensitivity analysis <strong>and</strong>, finally, a “sustainability<br />
chart” is provided to assess the balancing of social,<br />
economic <strong>and</strong> environmental performances of the<br />
various options.<br />
3. THE WATER FRAMEWORK<br />
DIRECTIVE AND THE COMMON<br />
IMPLEMENTATION STRATEGY<br />
The implementation of the Water Framework<br />
Directive is a dem<strong>and</strong>ing process for the EU<br />
Member States. The challenges are numerous: an<br />
extremely dem<strong>and</strong>ing timetable; the complexity of<br />
the text, the diversity of possible solutions to<br />
scientific, technical <strong>and</strong> practical questions; <strong>and</strong><br />
the problem of capacity building; just to name a<br />
few.<br />
In order to support the implementation of the<br />
Directive, a strategic document establishing a<br />
Common Implementation Strategy (CIS) was<br />
drafted <strong>and</strong> finally approved in May 2001. The<br />
CIS was established during an informal meeting of<br />
EU Water Directors <strong>and</strong> the Norwegian Water<br />
Director held in Paris in October 2000. The main<br />
aim of the CIS is to support coherent <strong>and</strong><br />
harmonious implementation of the Water<br />
Framework Directive among Member States, by<br />
establishing a common underst<strong>and</strong>ing <strong>and</strong><br />
guidance about the key aspects of the Directive.<br />
After the Water Directors’ decision to establish the<br />
CIS, a work programme was set up involving ten<br />
working groups <strong>and</strong> three expert advisory fora.<br />
The Water Directors steer <strong>and</strong> drive the whole<br />
process [EC, 2003b].<br />
After an initial phase of setting up organisational<br />
structures <strong>and</strong> modes of operation, the CIS work<br />
gained momentum in late 2001 <strong>and</strong> 2002. By<br />
November 2002 there were around 700 members<br />
in the expert network <strong>and</strong> over seventy working<br />
group <strong>and</strong> expert advisory fora meetings had taken<br />
place. By the end of that year, nine guidance<br />
documents, four reports <strong>and</strong> the pilot river basin<br />
network had been finalised. The first phase of the<br />
strategy was completed successfully <strong>and</strong> had<br />
achieved the establishment of a European expert<br />
network.<br />
Later on the structure was reorganised by grouping<br />
most of the issues together in four working groups:<br />
• WG 2.A “Ecological Status”<br />
• WG 2.B “Integrated River Basin Management”<br />
• WG 2.C “Groundwater”<br />
• WG 2.D “Reporting”<br />
The focus that has been defined for the technical<br />
work in the years 2003 <strong>and</strong> 2004 considers the<br />
following priorities: carrying out the pilot testing<br />
exercise; facilitating the intercalibration;<br />
developing technical guidance on specific<br />
outst<strong>and</strong>ing or new issues; maintaining the<br />
network; <strong>and</strong>, reviewing the guidance documents<br />
for inclusion in a comprehensive “EU manual for<br />
Integrated River Basin Management” [EC, 2003b].<br />
This document is not yet available <strong>and</strong> thus the<br />
MULINO methodology has been developed with<br />
reference to the guidance documents currently<br />
available.<br />
4. HOW MULINO SUPPORTS RIVER<br />
BASIN MANAGEMENT PLANNING<br />
In this section the work that has been done on river<br />
basin planning in the CIS working groups, is<br />
considered to illustrate how the use of MULINO in<br />
WFD implementation could help water authorities<br />
meet their new obligations, <strong>and</strong> demonstrate a<br />
management approach that is coherent with the<br />
new requirements. Four of the central themes that<br />
are dealt with in the official guidance documents<br />
are considered individually <strong>and</strong> evidence is drawn<br />
from one of the MULINO case studies.<br />
4.1 Integration<br />
Several different forms of integration, relevant to<br />
the WFD, are mentioned in the guidence document<br />
on the planning process. In general, integration is<br />
seen “as [a] key to the management of water<br />
protection within the river basin district” [EC,<br />
2003a p. 10]. Relationships with the MULINO<br />
approach are discussed below.<br />
When the MULINO methodology is applied in a<br />
way that documents the opinions <strong>and</strong> preferences<br />
of a range of individuals, it can provide competent<br />
authorities with an operational approach to<br />
combine a range of perspectives to describe <strong>and</strong><br />
assess pressures <strong>and</strong> impacts on water resources.<br />
In MULINO’s Design Phase the DPSIR<br />
conceptual framework provides a common<br />
structure for organising the information collected.<br />
This approach supports the user in managing the<br />
“integration of a wide range of measures,<br />
including pricing <strong>and</strong> economic <strong>and</strong> financial<br />
instruments, in a common […] approach” [EC<br />
2003a p. 10].<br />
mDSS’s capability to integrate modelling tools or<br />
their outputs in the decisional process <strong>and</strong> Multi-<br />
1029
Criteria Analysis functionalities support the<br />
“integration of disciplines, analyses <strong>and</strong><br />
expertise” [EC 2003a p. 10].<br />
The MULINO methodology was developed<br />
through experimentation in eight case studies that<br />
involved water authorities at different levels. Seen<br />
from both a “top-down” <strong>and</strong> a “bottom-up”<br />
perspective, the experience gained during the<br />
project shows the potential for an operational<br />
approach for the “integration of different decisionmaking<br />
levels that influence water resources <strong>and</strong><br />
water status” [EC, 2003a p. 10], be they local,<br />
regional or national. This methodology encourages<br />
the user to consider the priorities of other<br />
authorities in the description of the decision<br />
problem in the conceptual phase <strong>and</strong> in the<br />
definition of options <strong>and</strong> criteria in the design<br />
phase.<br />
4.2 Planning Components <strong>and</strong> Preconditions<br />
Among the considerations for a sound planning<br />
process provided in the guidance documents five<br />
preconditions to river basin planning are included.<br />
The MULINO methodology is proposed here as a<br />
support to achieving some of these preconditions.<br />
Through the mDSS scenario functionality,<br />
MULINO supports the development of “a vision<br />
of what the RBD will be in the future” [EC 2003a<br />
p.22] <strong>and</strong> through the use of the sustainability<br />
chart, “help[s] to determine what measures have<br />
be taken in the perspective of a sustainable<br />
development”. The user can compare the analysis<br />
matrix that has been prepared for the current<br />
conditions, with other matrices in which parameter<br />
values represent expected or possible future<br />
conditions.<br />
Many data formats are compatible with mDSS,<br />
<strong>and</strong> can be used in the “Conceptual” phase of<br />
mDSS tool, facilitating greater access to the<br />
information supporting the decisional process. A<br />
participatory multi-level approach supports<br />
capacity building <strong>and</strong> “the raising of public<br />
awareness”, an “informal transfer of know how<br />
(e.g. through the exchange of experience between<br />
river basin managers)” , <strong>and</strong> “formal training<br />
both internal <strong>and</strong> external” [EC 2003a , pp. 23-24]<br />
Authorities are advised to tackle the planning<br />
process with ‘the appropriate toolbox’. The mDSS<br />
tool could be a useful component of a toolbox that<br />
helps the decision-maker “to make right priorities<br />
concerning the program of measures” <strong>and</strong> to<br />
define <strong>and</strong> evaluate “numerous alternatives that<br />
represent various possible compromises among<br />
conflicting groups, values, <strong>and</strong> management<br />
objectives” [EC 2003a p. 27].<br />
4.3 Planning Process<br />
The specific requirements in the Water Framework<br />
Directive with regards to the planning process<br />
include the “identification of significant pressures<br />
<strong>and</strong> assessment of their impacts” [EC 2003a, p.<br />
31]. The first phase of the MULINO methodology<br />
is compatible with the approach recommended in<br />
the guidance document [EC, 2002a] which is<br />
dedicated to the identification of pressures <strong>and</strong><br />
assessment of impacts <strong>and</strong> developed by <strong>and</strong><br />
informal working group called IMPRESS.<br />
The mDSS adopts the same DPSIR conceptual<br />
framework for analysis, <strong>and</strong> the IMPRESS<br />
catalogues of indicators can be adopted in mDSS<br />
in the conceptual phase, giving the user a tool for<br />
managing the specific information that is provided.<br />
Consequently, the results of the analysis proposed<br />
in the IMPRESS document can be represented in<br />
a) PROBLEM EXPLORATION<br />
d) SENSITIVITY ANALYSIS<br />
c) DECISION RULE: RANKING<br />
e) SUSTAINABILITY ANALYSIS<br />
b) ANALISIS AND EVALUATION MATRICES<br />
Figure 2. Collection of screens representing a typical sequence of mDSS<br />
implementation steps.<br />
1030
the form of DPS chains <strong>and</strong> used by the mDSS<br />
software. The screening models proposed by<br />
IMPRESS can also be used in conjunction with<br />
mDSS <strong>and</strong> their outputs can be included in the<br />
mDSS decision analysis which takes place in the<br />
choice phase.<br />
Another step in the planning process, referred to as<br />
“gap analysis” can be supported by different<br />
analytical tools but “…can not rely on quantitative<br />
information only […] methods should be<br />
transparent <strong>and</strong> flexible, promoting public<br />
participation <strong>and</strong> facilitating negotiation<br />
processes” [EC 2003a, p. 41]. Through the<br />
MULINO method <strong>and</strong> the use of the mDSS<br />
software, the three phases of the decision process<br />
<strong>and</strong> the final outcomes can be described using<br />
charts, graphs <strong>and</strong> matrices, which illustrate how<br />
the decision-maker arrives at the “best” option.<br />
4.5 Planning <strong>and</strong> Public Participation<br />
The mDSS software has been designed to facilitate<br />
the integration of stakeholders <strong>and</strong> the civil society<br />
in decision making by promoting transparency <strong>and</strong><br />
communication about decisional processes. The<br />
guidance document considers planning as “a<br />
systematic, integrative <strong>and</strong> iterative process”<br />
which “culminates when all the relevant<br />
information has been considered <strong>and</strong> a course of<br />
action has been selected” [EC 2003a, p. 13]. The<br />
information <strong>and</strong> consultation of the public, active<br />
involvement <strong>and</strong> consultation of interested parties<br />
has particular importance for accessing the<br />
information that is required. At the basis of the<br />
MULINO methodology is the belief that<br />
consulting with stakeholders is an essential step of<br />
decisional processes connected with water<br />
resources management. The involvement of<br />
interested parties is envisaged throughout the<br />
MULINO methodology. In the conceptual phase of<br />
mDSS it is possible to structure the decision<br />
problem with input from stakeholders through the<br />
local network analysis. A questionnaire is designed<br />
to collect structured information from stakeholders<br />
which makes their preferences explicit. In the<br />
choice phase the participation of the stakeholders<br />
can be structured using the group decision making<br />
function that allows the different actors’<br />
preferences to be considered in the evaluation of<br />
options.<br />
Water managers can adopt the MULINO<br />
methodology “to facilitate the interaction <strong>and</strong><br />
discussion among managers <strong>and</strong> stakeholders ”<br />
The problem of developing “a balance between<br />
environmental functioning <strong>and</strong> users with<br />
conflicting aims” can also be approached through<br />
applying the mDSS group decision function.<br />
5. MULINO CASE STUDIES<br />
The MULINO methodology <strong>and</strong> mDSS software<br />
were developed over three years through<br />
experimentation in a selection of European case<br />
studies. In each case, the approach was tested in<br />
the context of a decision problem that was chosen<br />
<strong>and</strong> described by a representative from a water<br />
authority involved with planning for the area (see<br />
Table 1). All of the cases relate to issues that are<br />
relevant to the WFD implementation process.<br />
Romania Bahlui 1950 km 2 National<br />
“What is the best farming strategy to minimise<br />
sediment <strong>and</strong> nitrate loads while preserving living<br />
st<strong>and</strong>ards of irural communities?”<br />
Portugal Caia 780 km 2 National<br />
“What is the optimum level of water retention<br />
(control) in the Caia dam for multi-sectoral water<br />
management?”<br />
UK Yure & Bare 2500 km 2 National<br />
“What are the optimal seasonal water prices for<br />
maximising irrigation while minimising the adverse<br />
ecological impacts on the rivers?”<br />
Belgium Nethan 55 km 2 Regional<br />
“How can we reduce the risk of flooding? If we use<br />
storm basins, how big should they be <strong>and</strong> where<br />
should they be located?”<br />
Italy Vela 100 km 2 Local<br />
“What are the best solutions to reduce the nitrate<br />
discharges to the Venice Lagoon from the rivers of<br />
the Vela sub-basin?”<br />
Italy Cavallino 23 km 2 Local<br />
“How can we substitute groundwater with surface<br />
water for irrigation? Which is the best treatment<br />
method for guaranteeing water quality st<strong>and</strong>ards?”<br />
Italy Arborea 100 km 2 Regional<br />
“What is the best way to reduce the contaminants<br />
entering the phreatic aquifer in Arborea?”<br />
Europe - 3216000 km 2 European<br />
“What is the most efficient option for spatial<br />
implementation of the Nitrate Directive?”<br />
Table 1. The case studies: location, river basin,<br />
surface area, scale <strong>and</strong> decision contexts.<br />
For instance, in the Yare & Bure catchments in the<br />
west of the UK the decision problem is framed in<br />
the following question: “What are the optimal<br />
prices for winter <strong>and</strong> summer abstraction for<br />
maximising irrigation while minimising the<br />
adverse ecological impacts on the rivers?” within a<br />
river basin management approach. The problem<br />
was explored through the consideration of 16<br />
different criteria that represent the interests of a<br />
group of as many stakeholders. The team worked<br />
with the National Environment Agency, which is<br />
responsible for issuing abstraction licences in the<br />
area. In this case the problem was framed in such a<br />
way so as to predict the quantity of extraction for<br />
each option based on farmers’ optimisation<br />
strategies according to a whole farm profitability<br />
model. The options were also assessed for the<br />
ecological flows resulting from the different<br />
extraction patterns resulting from the variations in<br />
1031
price, thus experiencing the implementation of the<br />
WFD which, within the bounds of achieving good<br />
ecological status, is concerned with the assessment<br />
of the recovery of the costs of water services.<br />
The MULINO case study results will not<br />
necessarily have a direct relationship with WFD<br />
implementation in the UK or in the other case<br />
study areas, but the approach adopted could be<br />
useful for that process. The administrations<br />
involved with the MULINO project will probably<br />
play some direct or indirect role in the<br />
implementation process, <strong>and</strong> given their positive<br />
response to the methodology, it seems likely that<br />
the experience will provide some support for the<br />
forthcoming implementation activities.<br />
6. CONCLUDING REMARKS<br />
The specific contributions that have been identified<br />
above illustrate the nexus between the MULINO<br />
project <strong>and</strong> WFD implementation. This was the<br />
original main aim of the project, <strong>and</strong> for this<br />
reason greatest efforts have been made to make the<br />
results of the project compliant with the evolving<br />
guidance documents of the CIS.<br />
The MULINO project started just a few days after<br />
the publication of the WFD <strong>and</strong> has provided<br />
project results in time to allow European<br />
institutions to take advantage of this EU supported<br />
research. On the other h<strong>and</strong>, the methodology was<br />
developed alongside the CIS guidance<br />
documentation, which is still a work in progress.<br />
Competent authorities are already working on<br />
WFD implementation, according to a very tight<br />
schedule. Timing research to coincide with the<br />
developments in real world case studies <strong>and</strong> with<br />
policy implementation in the various EU countries<br />
was one of the main coordination challenges posed<br />
to MULINO.<br />
There is hope that the MULINO methodology will<br />
be adopted in some cases to assist the WFD<br />
implementation process. The positive experiences<br />
with case studies support an optimistic view <strong>and</strong><br />
further adaptations of the software are being<br />
guided by suggestions from the users. In a broad<br />
sense, MULINO can be useful to water managers<br />
because it proposes a framework for integrating (i)<br />
different methodological approaches; (ii) the<br />
preferences of the various actors involved in a<br />
planning process; <strong>and</strong>, (iii) a series of different<br />
modelling tools <strong>and</strong> data formats. The MULINO<br />
methodology was developed <strong>and</strong> tested in case<br />
studies of varying geographical scales from local<br />
to continental. Different decisional contexts in six<br />
countries within the EU <strong>and</strong> abroad have<br />
confirmed the flexibility of the tool. These<br />
applications were driven by the needs of potential<br />
DSS users: authorities competent in the field of<br />
water management. The result is a general<br />
approach <strong>and</strong> a software tool, which may support<br />
decision-makers in conducting a ”flexible,<br />
dynamic, cyclic <strong>and</strong> prospective planning process”<br />
in order to implement the Water Framework<br />
Directive in “a socially acceptable manner”, in<br />
different contexts [EC 2003a, p. 14].<br />
7. REFERENCES<br />
EC Common Strategy on the Implementation of<br />
the Water Framework Directive (2000/60),<br />
Strategic Document as agreed by the Water<br />
Directors under Swedish Presidency, 2001.<br />
http://europa.eu.int/comm/environment/water/waterframework/implementation.html<br />
EC Common Strategy on the Implementation of<br />
the Water Framework Directive (2000/60),<br />
Guidance for the analysis of Pressures <strong>and</strong><br />
Impacts in accordance with the Water<br />
Framework Directive, 2002a, http://forum.<br />
europa.eu.int/Public/irc/env/wfd/library: file WG 2.1-<br />
IMPRESS_Guidance_v5.3_except_Annex V.pdf<br />
EC<br />
EC<br />
Common Strategy on the Implementation of<br />
the Water Framework Directive (2000/60),<br />
Guidance on public participation in<br />
relation to the Water Framework Directive.<br />
Active involvement, consultation <strong>and</strong> public<br />
access to information, 2002b http://forum.<br />
europa.eu.int/Public/irc/env/wfd/library: file PP<br />
Guidance Main Text Final 11-12-2002.pdf<br />
Common Strategy on the Implementation of<br />
the Water Framework Directive (2000/60),<br />
Best practices in river basin planning -<br />
Work Package 2 Guidance on the planning<br />
process, 2003a. http://forum.europa.eu.int/Public/<br />
irc/env/wfd/library: file WG 2.9 - Planning Process<br />
guidance.doc<br />
EC Carrying Forward the Common<br />
Implementation Strategy for the Water<br />
Framework Directive - Progress <strong>and</strong> Work<br />
Programme for 2003 <strong>and</strong> 2004, 2003b.<br />
EEA, <strong>Environmental</strong> indicators: typology <strong>and</strong><br />
overview. Technical Report No 25, EEA,<br />
Copenhagen, 1999.<br />
MULINO Project 2004 Final documents, available<br />
online at: http://www.feem.it/web/loc/mulino/<br />
Simon, H. A., The New Science of Management<br />
Decision. Harper <strong>and</strong> Brothers, New York,<br />
1960.<br />
WWF, Elements of Good Practice in Integrated<br />
River Basin Management- A practical<br />
resource for implementing the EU Water<br />
Framework Directive. WWF_World Wide<br />
Fund For Nature, Brussels, Belgium, 2001<br />
http://www.p<strong>and</strong>a.org/downloads/europe/wfdpraticalre<br />
sourcedocumentenglish.pdf<br />
1032
Introducing River <strong>Modelling</strong> in the Implementation of<br />
the DPSIR Scheme of the Water Framework Directive<br />
Stefano Marsili-Libelli a , Francesco Betti b , Susanna Cavalieri b<br />
a Dept. of Systems <strong>and</strong> Computer Engineering, University of Florence<br />
Via S. Marta, 3 - 50139 Florence, ITALY<br />
Email:marsili@ingfi1.ing.unifi.it<br />
b <strong>Environmental</strong> Protection Agency for Tuscany<br />
Via Nicola Porpora, 20<br />
50122 Florence, ITALY<br />
Email:susanna.cavalieri@arpat.toscana.it<br />
Abstract: In Italy the National <strong>Environmental</strong> Protection Agency (ANPA) is about to adopt the Drivers-<br />
Pressures-States-Impacts-Responses (DPSIR) model introduced by the EC Water Framework Directive<br />
(WFD). This paper reassess the current definitions of Indicators in the light of the WFD, proposes the design<br />
of modular procedures <strong>and</strong> computational practices to determine the most significant State indicators,<br />
integrates the QUAL2E water quality model for the generation of quality data to assess differing DPSIR<br />
scenarios, with the final aim to produce an integrated software, partly based on Excel <strong>and</strong> partly on QUAL2E,<br />
whereby current quality data can be used to generate quality scenarios <strong>and</strong> apply the DPSIR model. The<br />
proposed method is applied the Arno river catchment.<br />
Keywords: Water Framework Directive; DPSIR; water quality models; decision support systems; catchment<br />
planning<br />
1. INTRODUCTION<br />
The central concept of the Water Framework<br />
Directive (WFD, EC 60/2000) is the integration<br />
among the various expertise <strong>and</strong> disciplines aiming<br />
at a better management of water (EC 2002a; E.C.<br />
2002b; E.C. 2002c). This paper presents an<br />
attempt to such integration to relate Pressures <strong>and</strong><br />
Impacts in the Drivers-Pressures - States - Impacts<br />
- Responses (DPSIR) model, as required by Article<br />
5 <strong>and</strong> along the guidelines of Annex II of the<br />
Water Framework Directive. However, its use is<br />
not straightforward given the differing nature of the<br />
data on which it operates. Normally information<br />
about Drivers are supplied by the statistical or<br />
socio-economic departments, whereas the data<br />
from which Impacts are computed from data<br />
directly collected by the authority in charge local<br />
monitoring. Normally the communication <strong>and</strong> data<br />
integration among these structures is weak.<br />
Moreover, in the practical application of the<br />
DPSIR model several obstacles are encountered:<br />
1) Statistical data are related to administrative<br />
boundaries which almost always do not<br />
coincide with the physical boundaries<br />
delimiting the model domain.<br />
2) The distinction between States <strong>and</strong> Impacts is<br />
not fully clear, because often Impacts are<br />
regarded as a further processing of the States.<br />
For river systems, proposals for the st<strong>and</strong>ardisation<br />
of their ecological status have already been<br />
forwarded (Hering <strong>and</strong> Strackbein, 2001) <strong>and</strong><br />
several States have been proposed on a biological<br />
basis, such as the Extended Biotic Index (EBI), to<br />
portray the ecological condition of the river<br />
system, but no practical Impact definition has been<br />
proposed so far.<br />
In the light of these considerations the present<br />
research attempts to:<br />
a) Introduce the use of water quality models,<br />
QUAL2E in particular, for the generation of<br />
quality data to be used in the DPSIR model<br />
<strong>and</strong> produce quality scenarios, both actual <strong>and</strong><br />
projected;<br />
b) Introduce a practical definition of Impacts in<br />
the light of the WFD;<br />
c) Design modular procedures <strong>and</strong> computational<br />
practices to determine the most<br />
significant State indicators <strong>and</strong> produce<br />
Impact information from them.<br />
1033
The final product of the study is an integrated<br />
software, partly based on Excel <strong>and</strong> partly on<br />
QUAL2E, through which current hydraulic <strong>and</strong><br />
river quality data can be used to generate quality<br />
scenarios to be assessed in the DPSIR context. The<br />
procedure is first described in general terms <strong>and</strong><br />
then illustrated in details with an application to the<br />
Arno river system, in central Italy.<br />
2. INTEGRATION OF THE DPSIR SCHEME<br />
The integration between the DPSIR scheme <strong>and</strong><br />
the water quality model consists of a number of<br />
cascaded operations, which are linked as shown in<br />
Figure 1.<br />
1<br />
2<br />
3<br />
4<br />
Hypothetical<br />
loading<br />
functions<br />
Data compatibility<br />
Data-base<br />
harmonisation<br />
Computational<br />
Computational<br />
DPSIR<br />
DPSIR<br />
procedures<br />
procedures<br />
Excel macros for:<br />
Excel macros for:<br />
•input preparation<br />
•input preparation<br />
•evaluation of model output<br />
•evaluation of model output<br />
QUAL2E Interfacing<br />
QUAL2E Interfacing<br />
QUAL2E<br />
Application to the<br />
Arno river catchment<br />
Scenario assessment<br />
in the DPSIR context<br />
States<br />
Computation of<br />
macrodescriptors<br />
Scenario generation<br />
Fig. 1. Collection of procedures required for the<br />
integrated DPSIR scheme.<br />
As Figure 1 shows, there are four main steps<br />
involved:<br />
1) In the preliminary part, the data availability,<br />
consistency <strong>and</strong> compatibility are assessed<br />
<strong>and</strong> the required data-bases are either<br />
harmonised if already existing or constructed<br />
if only raw data are available. It should be<br />
realised that several databases are required to<br />
set up the DPSIR scheme <strong>and</strong> these data are<br />
presently maintained by differing administrations,<br />
hence the need for a preliminary<br />
harmonisation of the available data regarding<br />
river catchment <strong>and</strong> related water quality into<br />
a coherent framework. The result has been a<br />
comprehensive Driver definition;<br />
2) A number of numerical procedures have been<br />
developed to obtain a consistent Pressure<br />
generator from the existing Drivers or from<br />
their hypothetical values assumed in new<br />
scenarios. Other related procedures have been<br />
set-up for the assessment of quality model<br />
output<br />
3) An interface has been developed between the<br />
previous procedures, mainly coded as Excel<br />
macros, <strong>and</strong> the river quality model;<br />
4) QUAL2E was selected as the river quality<br />
model <strong>and</strong> used as a States generator starting<br />
with the input data originating from the<br />
DPSIR context. A downstream processing<br />
section determines the Impacts from the<br />
QUAL2E outputs <strong>and</strong> makes them available<br />
for the scenario assessment procedures in step<br />
2. It also provides the interface for<br />
geographical information system (GIS)<br />
presenting the computation results as colour<br />
codes on the catchment thematic map. More<br />
studies on the interfacing between river<br />
quality models <strong>and</strong> GIS can be found in<br />
Marsili-Libelli et al. (2001).<br />
2.1 Data assessment in view of the DPSIR<br />
scheme<br />
Setting up a DPSIR scheme implies the availability<br />
of a large number of data regarding the river<br />
catchment, which are usually not gathered <strong>and</strong><br />
maintained by the same agency. Therefore, a<br />
preliminary task has been the harmonisation <strong>and</strong><br />
validation of the data: three main Drivers have<br />
been considered: population, agriculture <strong>and</strong><br />
industry. The first is defined as the number of<br />
people consistently living in the area, though in<br />
resort areas seasonal fluctuations have been<br />
accounted for. The agriculture driver was defined<br />
as a combination of the extension of agricultural<br />
l<strong>and</strong> <strong>and</strong> livestock, whereas industry was accounted<br />
for in terms of number of employees, energy bill<br />
<strong>and</strong> water consumption. These Drivers generate<br />
pressures in terms of pollution discharges into the<br />
river systems. Population <strong>and</strong> industry tend to<br />
generate point-source pollution, whose wastewater<br />
is generally collected through a sewage system <strong>and</strong><br />
delivered to a centralised wastewater treatment<br />
plant. The agricultural pressure is more difficult to<br />
quantify since a large part of it generates diffuse<br />
pollution. This can be estimated with specific<br />
software (CRITERIA) which yields the synthetic<br />
pollution load given the agricultural activity <strong>and</strong><br />
the terrain characteristics. At the end of this<br />
preliminary data harmonisation, drivers <strong>and</strong><br />
pressures were defined in coherent terms.<br />
2.2 Integration of a water quality model in the<br />
DSPIR scheme<br />
Having defined Drivers <strong>and</strong> Pressures, the next<br />
problem is the integrating the latter in a water<br />
quality model context. For this, it is required that<br />
Pressures generate inputs compatible with the<br />
water quality model. Under these boundary<br />
conditions the model produces a quality scenario<br />
from which the States are extracted <strong>and</strong> the<br />
1034
Impacts computed. This augmented DPSIR scheme<br />
is shown in Figure 2, with the insertion of the<br />
selected water quality model, QUAL2E (Brown<br />
<strong>and</strong> Barnwell, 1985) buffered by a pre- <strong>and</strong> postprocessing<br />
sections as interfaces to the<br />
conventional DPSIR scheme. In this context<br />
QUAL2E represents a bridge between Pressures<br />
<strong>and</strong> States.<br />
DRIVERS<br />
•Population<br />
•Agriculture<br />
•Industry<br />
QUAL2E input files from the Excel spreadsheets<br />
containing the Drivers <strong>and</strong> Pressures data of the<br />
whole river catchment.<br />
Once the hydraulic <strong>and</strong> quality data were specified,<br />
calibration runs were made in order to select the<br />
kinetic parameters which gave the best agreement<br />
between model response <strong>and</strong> observed quality data.<br />
Given the seasonal variability of several Drivers<br />
<strong>and</strong> related Pressures, the data were grouped into<br />
seasonal matrices <strong>and</strong> the same was done with<br />
QUAL2E parameters. The result was the<br />
availability of four seasonal scenarios for the<br />
whole procedure.<br />
DPSIR Scheme<br />
PRESSURES<br />
•Domestic waste<br />
•Agricultural waste<br />
•Industrial waste<br />
States<br />
Water quality<br />
parameters<br />
Untreated<br />
wastewater<br />
Nonpoint source<br />
pollution<br />
IMPACTS<br />
Synthetic quality indicators<br />
(MPL)<br />
WWTP<br />
Pollution<br />
abatement<br />
Pressure-driven<br />
inputs<br />
•Hydraulic<br />
•Quality<br />
QUAL2E<br />
QUAL2E<br />
river<br />
river<br />
model<br />
model<br />
Treated<br />
wastewater<br />
Point source<br />
pollution<br />
Fig. 2. Integration of water quality modelling in the<br />
DPSIR scheme.<br />
2.3 Scenario generation<br />
Providing the water quality model with the correct<br />
inputs requires a pre-processing stage which,<br />
starting with the Drivers, defines the resulting<br />
Pressures in terms of treated <strong>and</strong> untreated waste,<br />
introduces the abatement of the point- sources<br />
considering the average efficiency of the WWTP,<br />
as shown in Figure 2. All these data must be<br />
formatted in order to be compatible with the<br />
QUAL2E input data format. This procedure can<br />
also be used to assess hypothetical scenarios,<br />
generated by Drivers perturbations around the<br />
current values.<br />
2.5 Impact generation<br />
From the QUAL2E outputs, consisting of a large<br />
number of chemical <strong>and</strong> biological pollution<br />
indicators, some synthetic quality indicators are<br />
now extracted in accordance to Table 1.2 in Annex<br />
II of the WFD defining the ecological status<br />
classifications. The most coherent with the model<br />
output is certainly the Macrodescriptors Pollution<br />
Level (MPL) introduced by the Italian legislation<br />
(D.L. 152/99) in accordance with the WFD, which<br />
can be obtained from the scores of the seven main<br />
pollution indicators shown in Table 1.<br />
Table 1<br />
Definition of Macrodescriptors Pollution Level<br />
MPL Level<br />
Parameter 1 2 3 4 5<br />
100-DO<br />
(% sat.)<br />
BOD 5<br />
( mgO 2/L)<br />
COD<br />
( mgO 2/L)<br />
NH 4<br />
( mgN/L)<br />
NO 3<br />
( mgN/L)<br />
P tot<br />
( mgP/L)<br />
E. coli<br />
(UFC/100<br />
mL)<br />
≤ |10| ≤ |20| ≤ |30| ≤ |50| >|50|<br />
15<br />
25<br />
1,50<br />
10,0<br />
0,60<br />
20000<br />
Score 80 40 20 10 5<br />
2.4 Water quality modelling<br />
QUAL2E was selected as the water quality model<br />
being the most widely used by environmental<br />
agencies around the world <strong>and</strong> having achieved a<br />
high degree of acceptance <strong>and</strong> credibility. Setting<br />
up the input data for a QUAL2E model is not an<br />
easy task, because the river must be partitioned<br />
into reaches of appropriate length, each subdivided<br />
in cells, <strong>and</strong> for each unit both hydraulic <strong>and</strong><br />
quality parameters must be specified. An<br />
automated procedure was coded to generate the<br />
1035
Summing the scores for each variable yields the<br />
MPL value, which is then translated into a fivezone<br />
colour code, according to the ranges of Table<br />
2. If one of the variables could not be measured, a<br />
reduced, 6-variable, MPL can be computed with<br />
scaled ranges.<br />
Quality<br />
MPL<br />
Table 2<br />
MPL ranges<br />
Score<br />
7 variables 6 variables<br />
High 1 560-480 480-440<br />
Good 2 475-240 420-220<br />
Moderate 3 235-120 215-110<br />
Poor 4 115-60 105-55<br />
Bad 5 < 60 < 55<br />
A collection of Excel macros provide the required<br />
post-processing procedures to computes the MPL<br />
from the QUAL2E model outputs <strong>and</strong> present it on<br />
the cartography using the pertinent colour codes.<br />
concept of Population Equivalent (PE) for<br />
domestic pollution, Employee Equivalent (EE) for<br />
industrial pollution <strong>and</strong> Fertiliser Consumption<br />
(FC) for agriculture. The numerical values of these<br />
correspondence were obtained from demographic<br />
<strong>and</strong> socio-economic studies regarding the human<br />
<strong>and</strong> economic activities in Tuscany. The first two<br />
of these data represent the input to the wastewater<br />
treatment compartment, whereas the third<br />
represents the diffuse pollution, which should be<br />
estimated with specific tools, e.g. Criteria. From<br />
the WWTP operating records, the average removal<br />
efficiency is obtained <strong>and</strong> this represents the<br />
transfer function between Pressures <strong>and</strong> actual<br />
quality inputs to the river model, whose outputs<br />
define the States of the system, globally referred to<br />
as river quality. The last stage is the computation<br />
of the synthetic quality index MPL, representing<br />
the Impact resulting from the application of the<br />
known Pressures.<br />
Given the seasonal variability of two of the three<br />
Drivers, population <strong>and</strong> agriculture, together with<br />
the climatic changes <strong>and</strong> ensuing variation in river<br />
self-purification dynamics (Brown <strong>and</strong> Barnwell,<br />
1987; Chapra, 1997), it was decided to generate<br />
four pressure matrices, one for each season.<br />
Drivers<br />
Population<br />
Industry<br />
Agriculture<br />
3. APPLICATION TO THE ARNO CATCHMENT<br />
The above procedure was implemented in the<br />
database system of ARPAT, the regional<br />
environmental protection agency in Tuscany, <strong>and</strong><br />
applied to the river Arno catchment, shown in<br />
Figure 3, together with the main tributaries,<br />
wastewater treatment plants, flow gauges with a<br />
rating curve <strong>and</strong> the water quality monitoring<br />
stations.<br />
Pressures<br />
States<br />
PE Load<br />
Point-source<br />
Load<br />
WWTP<br />
efficiency<br />
QUAL2E<br />
River<br />
quality<br />
EE Load<br />
Diffuse<br />
Load<br />
FC Load<br />
Florence<br />
Impacts<br />
MPL<br />
Excel color code generation<br />
ArcView color segmentation<br />
Fig. 4. Computational scheme relating Drivers,<br />
Pressures, States <strong>and</strong> Impacts in the proposed<br />
model.<br />
WWTP<br />
Water quality monitoring stations<br />
Flow gauges<br />
Arno river<br />
Tributaries<br />
Catchment<br />
Fig. 3. Arno river catchment.<br />
The first step was to analyse the Drivers <strong>and</strong><br />
generate the Pressures. From the three main<br />
Drivers the pressures were derived introducing the<br />
3.1 Water quality model calibration<br />
The data from the water quality monitoring<br />
stations, indicated with squares in Figure 3 were<br />
used to perform a rough calibration of QUAL2E.<br />
At this stage a precise calibration was not possible,<br />
nor advisable, because:<br />
1) No fully validated quality model for the<br />
Arno river system exists to date. Several<br />
aspects of the Arno river systems are not yet<br />
fully understood, let alone modelled;<br />
1036
2) Quality data, either from the river<br />
monitoring stations or from the WWTPs,<br />
need further validation <strong>and</strong> are not always<br />
closely linked to hydraulic data;<br />
3) Diffuse pollution data <strong>and</strong> projections are<br />
still incomplete.<br />
Even with these sources of uncertainty a water<br />
quality model such as QUAL2E can still be used in<br />
this context as an enhancement to the existing databases<br />
in a more comprehensive scheme with the<br />
final aim of Impact computation. This is currently<br />
expressed as the MPL, divided in five ranges rather<br />
than sharp numerical values. Hence the use of a<br />
preliminary calibrated QUAL2E model can be<br />
justified for indicating a new approach <strong>and</strong><br />
applying the method previously outlined.<br />
Figures 5 - 9 show the effect of perturbing the<br />
Population Driver with a 20% increase of the<br />
domestic pollution over its current value. The<br />
results shown were obtained for the summer<br />
scenario, but similar results were produced for the<br />
other seasonal settings.<br />
BOD (mg L -1 )<br />
9.00<br />
8.00<br />
7.00<br />
6.00<br />
5.00<br />
4.00<br />
3.00<br />
2.00<br />
1.00<br />
0.00<br />
0.00 50.00 100.00 150.00 200.00<br />
River length (km)<br />
Reach boundary<br />
Measured quantity<br />
QUAL2E output (reference)<br />
QUAL2E output (perturbed)<br />
Fig. 5. BOD model output in the reference <strong>and</strong><br />
perturbed Summer scenario, together with the<br />
calibration data.<br />
COD (mg L -1 )<br />
25<br />
20<br />
15<br />
10<br />
Reach boundary<br />
Measured quantity<br />
QUAL2E output (reference)<br />
QUAL2E output (perturbed)<br />
DO (mg L -1 )<br />
N tot (mg L -1 )<br />
P tot (mg L -1 )<br />
9.00<br />
8.00<br />
7.00<br />
6.00<br />
5.00<br />
4.00<br />
3.00<br />
2.00<br />
1.00<br />
0.00<br />
Reach boundary<br />
Measured quantity<br />
QUAL2E output (reference)<br />
QUAL2E output (perturbed)<br />
0.00 50.00 100.00 150.00 200.00<br />
River length (km)<br />
Fig. 7. DO model output in the reference <strong>and</strong><br />
perturbed summer scenario, together with the<br />
calibration data.<br />
8<br />
7<br />
6<br />
5<br />
4<br />
3<br />
2<br />
1<br />
0<br />
Reach boundary<br />
Measured quantity<br />
QUAL2E output (reference)<br />
QUAL2E output (perturbed)<br />
0.00 50.00 100.00 150.00 200.00<br />
River length (km)<br />
Fig. 8. Total N model output in the reference <strong>and</strong><br />
perturbed summer scenario, together with the<br />
calibration data.<br />
0.4<br />
0.35<br />
0.3<br />
0.25<br />
0.2<br />
0.15<br />
0.1<br />
Reach boundary<br />
Measured quantity<br />
QUAL2E output (reference)<br />
QUAL2E output (perturbed)<br />
5<br />
0<br />
0.00 50.00 100.00 150.00 200.00<br />
River length (km)<br />
Fig. 6. COD model output in the reference <strong>and</strong><br />
perturbed summer scenario, together with the<br />
calibration data.<br />
0.05<br />
0<br />
0.00 50.00 100.00 150.00 200.00<br />
River length (km)<br />
Fig. 9. Total P model output in the reference <strong>and</strong><br />
perturbed summer scenario, together with the<br />
calibration data.<br />
1037
Florence<br />
Impact from the QUAL2E model output in terms<br />
of the Macrodescriptors Pollution Level, used to<br />
qualify the water quality in the WFD context <strong>and</strong><br />
provide the corresponding colour codes to the GIS<br />
environment depicting the catchment situation.<br />
MPL = 4<br />
MPL = 3<br />
MPL = 2<br />
River catchment<br />
PS = Point source<br />
W = Withdrawal<br />
D = Dam<br />
Fig. 10. Colour segmentation of the reference<br />
scenario.<br />
In addition to producing stationary values along the<br />
river course, the software computes the MPL b<strong>and</strong>s<br />
<strong>and</strong> places the corresponding colours on the river<br />
reaches in the GIS catchment map. The resulting<br />
quality scenarios are compared in Figure 10,<br />
showing the reference scenario, <strong>and</strong> Figure 11<br />
showing the perturbed situation. It can be seen that<br />
the river quality is decreased by one level,<br />
particularly in the middle <strong>and</strong> lower reaches,<br />
downstream of the dam, where the quality was<br />
already critical.<br />
MPL = 4<br />
MPL = 3<br />
MPL = 2<br />
River catchment<br />
Florence<br />
PS = Point source<br />
W = Withdrawal<br />
D = Dam<br />
Fig. 11. Colour segmentation of the perturbed<br />
scenario.<br />
4. CONCLUSIONS<br />
This paper has presented a computational<br />
procedure for the integration of a widely used river<br />
water quality model, QUAL2E, into the DPSIR<br />
scheme for the assessment of water quality in the<br />
guidelines of the Water Framework Directive of<br />
the European Community. The benefits of the<br />
integration consist of a better integration of the<br />
many databases required to prepare the input data<br />
to QUAL2E. It also provides a set of automated<br />
procedures to launch model simulations directly<br />
from the quality data spreadsheets. Another set of<br />
Excel macros was developed to compute the<br />
5. ACKNOWLEDGEMENTS<br />
This research was supported by the Regional<br />
<strong>Environmental</strong> Protection Agency (ARPAT).<br />
6. REFERENCES<br />
Brown, L.C., <strong>and</strong> T.O. Barnwell, The Enhanced<br />
Stream Water Quality Models QUAL2E <strong>and</strong><br />
QUAL2E-UNCAS: Documentation <strong>and</strong><br />
User Manual, EPA 300/3-87/007, EPA<br />
<strong>Environmental</strong> Research Laboratory,<br />
Athens, GA., 1987.<br />
Chapra, S., Surface Water-quality Modeling,<br />
McGraw Hill, 844 pp., New York, 1997.<br />
E.C., Guidance for the analysis of Pressures <strong>and</strong><br />
Impacts in accordance with the Water<br />
Framework Directive, Board of Water<br />
Directors, Nov. 2002a.<br />
E.C., Analysis of pressures <strong>and</strong> impacts the key<br />
implementation requirements of the water<br />
framework directive policy summary to the<br />
guidance document, Board of Water<br />
Directors, Nov. 2002b.<br />
E.C., Pressure <strong>and</strong> Impacts Analysis, Board of<br />
Water Directors, Dec. 2002c.<br />
Hering, D. <strong>and</strong> J., Strackbein, St<strong>and</strong>ardisation of<br />
river classifications: Framework method<br />
for calibrating different biological survey<br />
results against ecological quality<br />
classifications to be developed for the<br />
Water Framework Directive (Contract No:<br />
EVK1-CT 2001-00089: STAR stream types<br />
<strong>and</strong> sampling sites), 2001.<br />
Marsili-Libelli, S., Caporali, E., Arrighi, S., <strong>and</strong><br />
C., Becattelli. A georeferenced water<br />
quality model. Water Sci. Tech., 43 (7): 223<br />
– 230, 2001.<br />
Vogt, J., Implementing the GIS element of the<br />
WFD. EC Working Group GIS, WFD<br />
Common Implementation Strategies,<br />
Brussels 2002.<br />
1038
Sensitivity Analysis of a Network-Based, Catchment-<br />
Scale Water-Quality Model<br />
J.P. Norton 1, 2, 3 , L.T.H. Newham 1 <strong>and</strong> F.T. Andrews 1<br />
1 Integrated Catchment Assessment <strong>and</strong> Management Centre, School of Resources Environment <strong>and</strong><br />
Society, The Australian National University, Canberra, ACT 0200, Australia<br />
2 Dept. of Mathematics, The Australian National University, Canberra, ACT 0200, Australia<br />
3 School of Engineering, The University of Birmingham, Birmingham B15 2TT, UK<br />
Abstract: Careful consideration of the uncertainties <strong>and</strong> sensitivities associated with model outputs is<br />
essential when critical decisions are made on the basis of such results. Consideration of uncertainty is<br />
particularly important in the context of natural resource management, where models are often used to<br />
tackle complex <strong>and</strong> conflicting issues across multiple scales, as is the case in evaluating management<br />
options to reduce surface water pollution. This paper describes an analysis of uncertainty in the<br />
catchment-scale integrated hydrologic, economic, stream sediment <strong>and</strong> nutrient export model known as<br />
CatchMODS. The paper briefly describes the linked components of CatchMODS <strong>and</strong> its application in<br />
the Ben Chifley Dam catchment, Australia. An initial investigation to investigate some of the most<br />
important sources of output uncertainty is described. First-order sensitivities to selected model<br />
parameters are found analytically by linearising parts of the model <strong>and</strong> used, together with knowledge<br />
of where non-linearity has most effect, to point to conditions to be investigated further. The extent of<br />
non-linear effects is also checked by comparing the analytical results with the results of parameterperturbation<br />
tests. Results from the analysis are used to prioritise continuing model development <strong>and</strong><br />
data-collection activities. The results are also to be incorporated into a decision-analysis framework to<br />
evaluate management options to reduce surface water pollution. The decision-analysis framework <strong>and</strong><br />
incorporation of uncertainty analysis into it are outlined.<br />
Keywords: Sensitivity analysis; Water quality modelling; CatchMODS model; Decision analysis<br />
framework.<br />
1. INTRODUCTION<br />
As natural-resource managers increase their<br />
reliance on the outputs of complex environmental<br />
models, there is an increasing need for better<br />
underst<strong>and</strong>ing of model behaviour, particularly<br />
the impacts of uncertainties. Such is the case<br />
where decisions are based on the outputs of<br />
hydrologic <strong>and</strong> water-quality models.<br />
This paper describes sensitivity analysis (SA) on<br />
the catchment-scale integrated hydrologic, stream<br />
sediment <strong>and</strong> nutrient export model known as<br />
CatchMODS. The aims are first to improve<br />
underst<strong>and</strong>ing of the behaviour of CatchMODS<br />
<strong>and</strong> second to examine SA techniques appropriate<br />
for such models. The results of the analysis are to<br />
be further considered in a decision-analysis<br />
framework [Myšiak et al 2004] for evaluating the<br />
efficacy of management options in controlling<br />
diffuse-source pollution.<br />
Section 2 gives a brief description of<br />
CatchMODS. The next section briefly describes<br />
the SA techniques used. The results of the SA are<br />
presented in Section 4, initially by algebra-based<br />
SA of the two-parameter non-linear nitrogen<br />
routing submodel in detail. It shows that a range<br />
of useful information is obtainable in this way<br />
with very few model runs. Experimental results<br />
from perturbations of the parameters are then<br />
used to check the extent of non-linear effects on<br />
sensitivity. Next the ease of algebraic SA for a<br />
linear dynamical model is illustrated by analysis<br />
of the linear part of the hydrological submodel.<br />
Finally, the implications of the SA results for<br />
multi-criteria decision analysis are discussed.<br />
2. CatchMODS MODEL<br />
The Catchment-scale Management of Diffuse<br />
Sources (CatchMODS) model simulates current<br />
conditions <strong>and</strong> the effects of l<strong>and</strong> <strong>and</strong> water<br />
management activities on diffuse-source pollutant<br />
1039
loads. CatchMODS links several components: a<br />
regionalised hydrologic model based on the<br />
IHACRES rainfall-runoff model [Jakeman et al.<br />
1990, Croke <strong>and</strong> Jakeman 2003], a suspendedsediment<br />
model developed from the SedNet<br />
model [Prosser et al. 2001], <strong>and</strong> simple empirical<br />
total phosphorus <strong>and</strong> total nitrogen models. The<br />
model also incorporates a simple cost-accounting<br />
component to enable the tradeoffs between<br />
environmental remediation costs (fixed <strong>and</strong><br />
continuing) <strong>and</strong> environmental benefits (pollutant<br />
load reductions) to be explored. To provide a<br />
catchment-scale perspective, CatchMODS has a<br />
node-link spatial structure, with upstream<br />
subcatchments (typically 20-50km 2 in area) <strong>and</strong><br />
river reaches (typically 7-12km long) providing<br />
input to downstream elements, so that pollutants<br />
can be routed through the stream network.<br />
Outputs are available for each subcatchment <strong>and</strong><br />
the downstream end of each reach.<br />
The data dependencies in CatchMODS, shown in<br />
Figure 1, are relatively simple, representing the<br />
influence of the drivers of the physical processes.<br />
Figure 1. Data dependencies in CatchMODS.<br />
CatchMODS has been applied in the Ben Chifley<br />
Dam catchment in New South Wales, Australia,<br />
as part of a project to improve the management of<br />
diffuse-source pollutants. A description of that<br />
application <strong>and</strong> greater detail on the model can be<br />
found in Newham et al. [2004].<br />
Several features of CatchMODS make it a useful<br />
example for investigating SA techniques for<br />
environmental models: its application is networkbased,<br />
allowing cascade (routing) effects to be<br />
investigated; it incorporates components with a<br />
range of complexity; the data dependencies<br />
between submodels are not complicated <strong>and</strong><br />
submodels can be largely assessed individually;<br />
<strong>and</strong> SA of CatchMODS is part of a process of<br />
iterative model development, with CatchMODS<br />
incorporating many of the modifications<br />
suggested by Newham et al. [2003] following<br />
analysis of the SedNet model.<br />
3. SENSITIVITY/UNCERTAINTY<br />
ANALYSIS<br />
3.1 Experiment <strong>and</strong> analysis for SA<br />
It is now widely accepted that the outputs of<br />
environmental prediction models to aid decisionmaking<br />
should be accompanied by quantitative<br />
assessment of their uncertainty. This ideal may<br />
not be realisable, not least because of difficulty in<br />
quantifying the contributing uncertainties. Input<br />
uncertainties are likely to include unpredicted<br />
disturbances <strong>and</strong> slow, poorly identified trends.<br />
Estimation of parameter uncertainty is usually<br />
either subjective or dependent on restrictive <strong>and</strong><br />
perhaps unjustified probabilistic assumptions.<br />
Systematic modelling error, although assessable<br />
to some extent from the historical fit of the model<br />
to observations, is likely to be inhomogeneous,<br />
making extrapolation dubious. The next best thing<br />
to an uncertainty analysis is a sensitivity<br />
assessment, which can show which input features<br />
<strong>and</strong> model parameters influence the output<br />
behaviour most strongly <strong>and</strong> require most careful<br />
attention.<br />
SA usually treats the model as a “black box”,<br />
investigated by Monte Carlo trials or systematic<br />
perturbation of parameter values. The latter relies<br />
on calculating (approximately) some of the<br />
derivatives in the Taylor series<br />
p<br />
p p<br />
2<br />
∂y i<br />
∂ y<br />
δy<br />
i<br />
i = δθ<br />
j<br />
j + 1<br />
δθ<br />
j k<br />
jδθ<br />
= 1 ∂ θ<br />
2<br />
k<br />
j = 1 = 1 ∂ θ j ∂ θ<br />
k<br />
+ higher-order terms (1)<br />
for the change in a scalar output<br />
y i due to<br />
variations in parameters θ<br />
1<br />
to θ p , assuming that<br />
the derivatives exist. For small enough variations<br />
in a model without sharp non-linearity, all<br />
individual cause-effect relations may be almost<br />
linear. The linear part of the variation of<br />
input or model parameter<br />
∂ yi<br />
y i with<br />
θ j , determined by<br />
∂θ<br />
j , defines the conventional sensitivity<br />
y<br />
i<br />
δy i / y θ i j ∂y S<br />
i<br />
θ<br />
≡ lim = , (2)<br />
j δθ →0<br />
δθ j / θ j y i ∂θ<br />
j<br />
normalised (relating proportional changes in<br />
<strong>and</strong>θ<br />
j ) to remove dependence on the units<br />
employed. A vector y of outputs or θ of<br />
parameters merely requires the sensitivities to be<br />
found for all outputs <strong>and</strong> parameters. If the output<br />
is a time series, the sensitivity is an influence<br />
function of time. In all cases it can be found<br />
y i<br />
1040
approximately by noting the output change when<br />
the parameter undergoes a small perturbation.<br />
However, interaction between two or more<br />
parameters may affect the output, even if the<br />
output is linear in each parameter. To check for<br />
two-parameter interaction, for example through<br />
bilinear terms<br />
p<br />
ijk<br />
θ jθ<br />
k<br />
, all second derivatives<br />
∂ 2 yi<br />
∂θ j∂θ<br />
k<br />
, j ≠ k must be found. To check<br />
the influences of terms up to total degree m in the<br />
parameters, including interaction between up to m<br />
parameters, all derivatives up to the mth must be<br />
found. Higher-order differences of results from<br />
more perturbation runs give them approximately.<br />
If m is high enough, this approach shows the<br />
effects of smooth non-linearities over specific<br />
perturbation ranges. In practice, the computing<br />
load to find all possibly significant derivatives<br />
may well be excessive. Moreover, sharp nonlinearity<br />
may make Taylor-series approximation<br />
of the output variation impracticable. An<br />
alternative such as Monte Carlo (MC) trials over<br />
the parameter-uncertainty ranges will then be<br />
needed. There is a large literature on how best to<br />
arrange MC trials (Saltelli et al., 2000), but they<br />
incur an inevitable risk of missing significant<br />
behaviour.<br />
So far, the model has been treated as a “black<br />
box”, assuming very little prior knowledge.<br />
However, a simulation model is not a black box;<br />
its constituent relations are known, if<br />
complicated. This knowledge may help to guide<br />
SA in several ways: to look for significant<br />
interactions; to see what non-linearities are<br />
present <strong>and</strong> where they are sharp; to see what<br />
aspects of output behaviour are sensitive to<br />
particular parameter groups; <strong>and</strong> to focus<br />
successively on parts of the model with known<br />
connections to the rest, instead of considering all<br />
parameters at once. Catchment models, with<br />
relatively simple structure defined by the stream<br />
network (cascades <strong>and</strong> confluences), offer such<br />
opportunities.<br />
Two of the components of CatchMODS will be<br />
investigated in detail: the dissolved-nitrogen<br />
transport submodel <strong>and</strong> the linear module of the<br />
IHACRES rainfall-runoff model.<br />
3.2 SA of dissolved-nitrogen transport model<br />
The stream network is divided into stream<br />
reaches, numbered (h,i) as shown in Figure 2,<br />
where h counts down, reach by reach, from the<br />
maximum number of reaches from the catchment<br />
outlet to the headwaters <strong>and</strong> i is odd if the stream<br />
is the left-h<strong>and</strong> tributary, even if the right-h<strong>and</strong>, at<br />
the confluence at the lower end of the reach.<br />
Figure 2. Example of numbering for stream<br />
reaches.<br />
The submodel for mean annual dissolved nitrogen<br />
N at the bottom of reach (h,i) is<br />
N ′<br />
hi<br />
=<br />
N<br />
hi<br />
gG<br />
hi<br />
= N ′<br />
hi<br />
exp( −C<br />
+ N<br />
h−1,2<br />
i−1<br />
+ N<br />
h−1,2<br />
i<br />
N<br />
hi<br />
′<br />
hi<br />
/ Q<br />
hi<br />
)<br />
<br />
<br />
<br />
(3)<br />
The first equation accounts for nitrogen<br />
introduced in reach (h,i), proportional to baseflow<br />
increase G , <strong>and</strong> from the tributaries. The<br />
hi<br />
second accounts for denitrification. Here C hi is the<br />
channel area (reach length × width), Q hi the mean<br />
annual flow <strong>and</strong> g the parameter, assumed<br />
common to all reaches, to which the sensitivity of<br />
N at the outlet to the dam is required.<br />
Differentiating (3),<br />
∂N<br />
′<br />
hi<br />
∂g<br />
∂N<br />
so<br />
∂<br />
hi<br />
g<br />
=<br />
= (1 −<br />
1<br />
= (<br />
N ′<br />
hi<br />
∂N<br />
∂<br />
hi<br />
g<br />
G<br />
hi<br />
−<br />
∂N<br />
h−1,2<br />
i−1<br />
∂N<br />
h−1,2<br />
i<br />
+<br />
+<br />
∂g<br />
∂g<br />
C<br />
hi<br />
N<br />
Q<br />
hi<br />
′ ′<br />
hi<br />
∂N<br />
hi<br />
C<br />
) exp( −<br />
∂g<br />
C ∂ ′<br />
hi<br />
N<br />
hi<br />
) N<br />
Q<br />
hi<br />
hi<br />
∂g<br />
N<br />
hi<br />
Q<br />
hi<br />
1<br />
= (<br />
−<br />
gG<br />
hi<br />
+ N<br />
h−1,2<br />
i−1<br />
+ N<br />
h−1,2<br />
i<br />
′<br />
hi<br />
<br />
<br />
<br />
<br />
) <br />
<br />
<br />
<br />
<br />
<br />
∂N<br />
h−1,2<br />
i−1<br />
∂N<br />
h−1,2<br />
i<br />
N<br />
hi<br />
( G<br />
hi<br />
+<br />
+ )<br />
∂g<br />
∂g<br />
(4)<br />
C<br />
hi<br />
).<br />
Q<br />
hi (5)<br />
1041
N<br />
1<br />
C<br />
hi<br />
S hi g = (<br />
− ).<br />
gG<br />
hi<br />
+ N<br />
h−1,2<br />
i−1<br />
+ N<br />
h−1,2<br />
i<br />
Q<br />
hi<br />
N<br />
N<br />
h−1,2<br />
i−1<br />
h−1,2<br />
i<br />
( gG<br />
hi<br />
+ N<br />
h−1,2<br />
i−1<br />
S g + N<br />
h−1,2<br />
i<br />
S g )<br />
(6)<br />
N<br />
Here ∂ N<br />
hi<br />
∂g<br />
<strong>and</strong> S hi<br />
g are more complicated<br />
functions of g than appears at first sight, as all the<br />
N’s depend on g.<br />
These expressions indicate that:<br />
(i) the recursion (6) can be used to find all the<br />
sensitivities to g, starting at the top of the<br />
catchment, after a single run to get the nominal<br />
values of all G’s, Q’s <strong>and</strong> N’s. Generally, all firstorder<br />
sensitivities to any one parameter at any<br />
operating point can be generated (exactly but for<br />
finite precision) by one simulation run <strong>and</strong> one<br />
run of (6), so long as the derivatives exist;<br />
(ii) both (5) <strong>and</strong> (6) are inconsistent with an<br />
assumption that N is a finite-degree polynomial in<br />
g: after rationalisation, it is not possible to match<br />
coefficients of all powers of g on the two sides.<br />
This is not surprising, as the denitrification<br />
equation in (3) is of infinite degree in g;<br />
(iii) if the exponential in (3) is not far below unity<br />
(i.e. if denitrification is by a small percentage),<br />
(3) can be approximated by<br />
N<br />
hi<br />
≅ N′<br />
hi<br />
(1 − C<br />
hi<br />
N′<br />
hi<br />
/ Q<br />
hi ) , then substituting<br />
into (5) <strong>and</strong> equating highest-degree terms in g on<br />
each side, the highest (significant) degree in g in<br />
N<br />
hi<br />
is found to be twice the higher of the highest<br />
degrees in N h−1,2<br />
i−1<br />
<strong>and</strong> N<br />
h−1,2<br />
i<br />
. However, g<br />
is typically small <strong>and</strong> low-degree terms dominate;<br />
(iv) in (5), the contributions of reach (h,i) <strong>and</strong> the<br />
immediately upstream tributaries to ∂ N<br />
hi<br />
∂g<br />
,<br />
by G , ∂N<br />
g<br />
hi h−1,2<br />
i−1<br />
∂ <strong>and</strong> ∂N<br />
h−1,2<br />
i<br />
∂g<br />
, are<br />
additive <strong>and</strong> equally weighted, so after running<br />
(5) it is easy to see the relative importance of each<br />
source in each reach.<br />
To illustrate, a nominal run followed by recursive<br />
solution of (5) <strong>and</strong> (6) gives the results shown in<br />
Table 1 for the lower ends of one reach <strong>and</strong> its<br />
tributaries. Finite-difference results from a 10%<br />
perturbation of g are also shown.<br />
Table 1. Analytical <strong>and</strong> perturbation results<br />
relating to reach (5,1).<br />
∂ N<br />
Reach N ∂ g<br />
δ S<br />
N<br />
g<br />
N δg<br />
5,1 0.4905 192.3 167.3 0.1960<br />
4,1 0.2181 94.47 91.74 0.2165<br />
4,2 0.0185 -108.5 -95.71 -2.931<br />
G<br />
51<br />
= 1590, gG<br />
51<br />
=<br />
Several points emerge:<br />
0.7949, C<br />
51<br />
Q<br />
51<br />
= 0.7207<br />
• G 51 heavily dominates ∂N<br />
41<br />
∂g<br />
<strong>and</strong><br />
∂N 42<br />
∂g in determining ∂N<br />
51<br />
∂g<br />
in (5),<br />
<strong>and</strong> the same is true in (6), determining<br />
N<br />
S 51<br />
g<br />
• the derivatives <strong>and</strong> sensitivities from the<br />
perturbation test differ noticeably from those<br />
from (5) <strong>and</strong> (6), because of non-linearity<br />
• one derivative is negative, indicating that in<br />
one or more higher reaches, the effect of g on<br />
the exponent in the denitrification equation<br />
dominates its effect in increasing N ′ .<br />
An important but less than obvious point revealed<br />
by comparing analytical <strong>and</strong> perturbation results<br />
is that ill conditioning can reduce the accuracy of<br />
computed normalised sensitivities. Figure 3<br />
shows the analytical <strong>and</strong> perturbation-derived<br />
sensitivities<br />
S for all reaches, ordered by N.<br />
N<br />
g<br />
The discrepancies are due partly to non-linearity,<br />
but depend on N, being much larger at very low N<br />
because of rounding. Large proportional error at<br />
very low values of N is, of course, unlikely to be<br />
serious.<br />
2 2<br />
A recursion for ∂ N<br />
hi<br />
∂g<br />
is easily derived<br />
but algebraically complex enough not to yield<br />
easy conclusions about sensitivity.<br />
3.3 SA of effective-rainfall / runoff model<br />
By contrast to the submodel above, the<br />
hydrological part has several (7) parameters <strong>and</strong><br />
is not cascaded, as runoff is found for each<br />
subcatchment <strong>and</strong> for the catchment as a whole by<br />
applying a single IHACRES model calibrated for<br />
whichever area is represented. The part of the<br />
submodel relating flow to effective rainfall is<br />
straightforward to analyse. It can be written as the<br />
temporal recursion<br />
1042
45<br />
40<br />
35<br />
30<br />
Analytical Value<br />
10% Perturbation<br />
5% Perturbation<br />
Sensitivity<br />
25<br />
20<br />
15<br />
10<br />
5<br />
0<br />
1.00E-17 1.00E-14 1.00E-11 1.00E-08 1.00E-05 1.00E-02 1.00E+01<br />
Figure 3. Comparison of analytical<br />
Dissolved Nitrogen (t/y)<br />
N<br />
S g <strong>and</strong><br />
N<br />
S g by perturbation (5 <strong>and</strong> 10%), with N. Note log scale on<br />
horizontal axis <strong>and</strong> that absolute sensitivity values are shown.<br />
Q<br />
k<br />
= −a<br />
1<br />
Q<br />
k−1<br />
− a<br />
2<br />
Q<br />
k−2<br />
+ b<br />
0<br />
E<br />
k<br />
+ b<br />
1<br />
E<br />
k−1<br />
(7)<br />
where E k is effective rainfall in day k <strong>and</strong> Q k flow<br />
at the end of day k. With the parameter vector<br />
θ ≡ a<br />
1<br />
a<br />
2<br />
b<br />
0<br />
b<br />
1<br />
, the vector<br />
influence function ∂Q ∂θ<br />
is given by<br />
defined as [ ] T<br />
∂Q<br />
k<br />
∂θ<br />
<br />
∂Q<br />
∂ <br />
− −<br />
−1<br />
Q<br />
Q<br />
k<br />
−<br />
k−2<br />
k−1<br />
a<br />
1<br />
a <br />
∂a<br />
2<br />
1<br />
∂a<br />
1 <br />
∂Q<br />
∂<br />
− <br />
−<br />
−1<br />
Q<br />
a<br />
k<br />
−<br />
−<br />
−<br />
2<br />
1<br />
Q<br />
<br />
∂<br />
2<br />
a<br />
k<br />
a<br />
k 2<br />
∂<br />
= 2<br />
a<br />
2 (8)<br />
∂Q<br />
−<br />
∂<br />
− <br />
−<br />
1<br />
Q<br />
a<br />
k<br />
−<br />
2<br />
1<br />
a<br />
k<br />
+ <br />
∂<br />
2<br />
E<br />
b<br />
0<br />
∂b<br />
k<br />
0 <br />
∂Q<br />
−<br />
∂<br />
− <br />
−<br />
k 1<br />
Q<br />
a −<br />
k 2<br />
1<br />
a +<br />
− <br />
∂<br />
2<br />
E<br />
b ∂<br />
k 1<br />
1<br />
b<br />
1 <br />
<strong>and</strong> it is easy to see that the sequences { Q ∂a 1<br />
}<br />
{ ∂ Q ∂a },{ ∂ Q ∂b } <strong>and</strong> { Q ∂b }<br />
∂ ,<br />
2 1<br />
∂<br />
0<br />
are all<br />
outputs of the same dynamical process, driven<br />
− Q delayed by one<br />
respectively by { Q}<br />
day, { E } <strong>and</strong> { }<br />
− , { }<br />
E delayed by one day. The time<br />
constants of the dynamical process are the quick<strong>and</strong><br />
slow-flow time constants of the rainfallrunoff<br />
relation. Once the effects of differences in<br />
initial conditions have faded, { ∂Q ∂a<br />
2}<br />
essentially { Q ∂a 1<br />
}<br />
similarly for { ∂ Q ∂b } <strong>and</strong> { ∂ Q ∂b }<br />
∂ delayed by one day, <strong>and</strong><br />
is<br />
, so the<br />
1<br />
0<br />
parameter sensitivities of the mean flow over a<br />
year are<br />
∂Q<br />
365 ∂Q<br />
=<br />
1 k<br />
365<br />
≅<br />
∂a<br />
1<br />
k = 1 ∂ a<br />
1<br />
∂Q<br />
∂Q<br />
≅<br />
∂b<br />
0<br />
∂b<br />
1<br />
1<br />
365<br />
365 ∂Q<br />
k<br />
<br />
k = 1 ∂ a<br />
2<br />
∂Q<br />
= ;<br />
∂a<br />
2<br />
(9)<br />
It is also not difficult to see that, with the time<br />
constants very much less than a year,<br />
∂Q<br />
∂Q<br />
∂Q<br />
<br />
≅ −Q<br />
− a − <br />
∂<br />
1<br />
a<br />
a<br />
∂<br />
2<br />
1<br />
a<br />
1<br />
∂a<br />
1 <br />
∂Q<br />
Q ∂Q<br />
<br />
so ≅ −<br />
≅ <br />
∂a<br />
1<br />
1+<br />
a<br />
1<br />
+ a<br />
2<br />
∂a<br />
2 <br />
∂Q<br />
∂Q<br />
E <br />
≅ ≅<br />
<br />
∂b<br />
0<br />
∂b<br />
1<br />
1+<br />
a<br />
1<br />
+ a<br />
2 <br />
(10)<br />
so these sensitivities can be found without<br />
performing a run.<br />
1043
4. DISCUSSION AND CONCLUSIONS<br />
This paper has described SA on the catchmentscale<br />
integrated hydrologic <strong>and</strong> water-quality<br />
model, CatchMODS. The analysis is an important<br />
step in model development <strong>and</strong> has been very<br />
useful for improving underst<strong>and</strong>ing of the<br />
behaviour of the model particularly with respect<br />
to cascading sensitivity effects. It has contributed<br />
to improving management outcomes by<br />
developing techniques to identify significant<br />
sources of uncertainty in model predictions i.e.<br />
where uncertainty in inputs have greatest impact<br />
on model prediction.<br />
The results presented here also illustrate several<br />
general factors in the analysis of complex <strong>and</strong>/or<br />
cascading environmental models. Cascading<br />
makes the overall effect of even very simple nonlinearities<br />
on sensitivity difficult to assess without<br />
either an algebraic analysis or considerable<br />
experimentation. Algebraic analysis plus a very<br />
modest amount of computing can yield a good<br />
deal of insight not easily obtainable by<br />
experiment. Rounding errors can give rise to<br />
significant errors in the estimation of sensitivities.<br />
In sensitivity-propagation recursions such as (5)<br />
<strong>and</strong> (6), ill-conditioning may arise (<strong>and</strong> indeed<br />
sometimes does in the Ben Chifley catchment)<br />
when the contributing terms are individually not<br />
small. As seen in the effective-rainfall/runoff<br />
submodel, sensitivity analysis is straightforward<br />
when the recursion is linear.<br />
SA is necessary to support the decision analysis<br />
framework for the Ben Chifley Dam catchment<br />
developed in parallel with the construction of<br />
CatchMODS. The aim of the decision analysis<br />
framework is to incorporate a broader view into<br />
evaluation of the performance of various<br />
management options to reduce surface-water<br />
pollution. A preliminary description of the<br />
decision analysis framework, which tries to<br />
reconcile the ecological <strong>and</strong> economic effects of<br />
remediation actions using multicriteria decision<br />
analysis, is available (Myšiak et al. 2004). Our<br />
intention is to investigate further the influence of<br />
model-input uncertainties in this framework on<br />
potential management recommendations.<br />
CatchMODS includes refinements to the SedNet<br />
sediment-transport model described in Newham<br />
et al. (2003). SA of both models has had a role in<br />
the iterative process of model development <strong>and</strong><br />
testing, providing insight into the overall effects<br />
of components of the models <strong>and</strong> clarifying their<br />
relative importance <strong>and</strong> their interactions.<br />
Difficulties exist in communicating the need for,<br />
<strong>and</strong> techniques of, SA to end-users especially<br />
non-technical managers. These difficulties present<br />
possibly greater limitations than SA techniques.<br />
As part of the continuing process of SA <strong>and</strong><br />
continued model development, more complete SA<br />
is recommended for the CatchMODS model. This<br />
might include SA across the multiple components<br />
of the model to determine the effects of parameter<br />
interactions, using Monte-Carlo sampling<br />
techniques such as Fourier Amplitude Sensitivity<br />
Testing (Saltelli et al., 2000) as necessary. More<br />
fundamental SA, using algebraic analysis where<br />
possible, is also planned to determine the effects<br />
of spatially local variations in parameter values.<br />
Such investigations may allow alternative model<br />
structures, providing adequate resolution at<br />
minimum computational cost, to be identified.<br />
5. REFERENCES<br />
Croke, B.F.W. <strong>and</strong> A.J. Jakeman, A Catchment<br />
Moisture Deficit Module for the IHACRES<br />
Rainfall-Runoff Model, <strong>Environmental</strong><br />
<strong>Modelling</strong> <strong>and</strong> <strong>Software</strong>, vol. 19, 1, pp. 1-5,<br />
2004.<br />
Jakeman, A.J., Littlewood, I.G. <strong>and</strong> P. G.<br />
Whitehead, Computation of the<br />
Instantaneous Unit Hydrograph <strong>and</strong><br />
Identifiable Component Flows with<br />
Application to Two Small Upl<strong>and</strong><br />
Catchments, Journal of Hydrology, vol. 117,<br />
pp. 275-300, 1990.<br />
Myšiak, J., Newham, L.T.H <strong>and</strong> R.A. Letcher,<br />
Integrated <strong>Modelling</strong> <strong>and</strong> Decision Making<br />
Analysis for Water Quality Management:<br />
Ben Chifley Dam Catchment Case Study, in<br />
Proceedings of Integrated Water<br />
Management of Transboundary Catchments:<br />
A Contribution form Transcat, Palazzo<br />
Zorzi, Venice, Italy, 24-26 March 2004.<br />
Newham, L.T.H., Letcher, R.A., Jakeman, A.J.<br />
<strong>and</strong> T. Kobayashi, A Framework for<br />
Integrated Hydrologic, Sediment <strong>and</strong><br />
Nutrient Export <strong>Modelling</strong> for Catchment-<br />
Scale Management, <strong>Environmental</strong><br />
<strong>Modelling</strong> <strong>and</strong> <strong>Software</strong>, 2004, in press.<br />
Newham, L.T.H., Norton, J.P., Prosser, I.P.,<br />
Croke, B.F.W. <strong>and</strong> A.J. Jakeman, Sensitivity<br />
Analysis for Assessing the Behaviour of a<br />
L<strong>and</strong>scape-Based Sediment Source <strong>and</strong><br />
Transport Model, <strong>Environmental</strong> <strong>Modelling</strong><br />
<strong>and</strong> <strong>Software</strong>, vol. 18, pp. 741-751, 2003.<br />
Prosser, I.P., Rustomji, P., Young, W.J., Moran,<br />
C. <strong>and</strong> A. Hughes, Constructing River Basin<br />
Sediment Budgets for the National L<strong>and</strong> <strong>and</strong><br />
Water Resources Audit, CSIRO Technical<br />
Report 15/01, CSIRO L<strong>and</strong> <strong>and</strong> Water,<br />
Canberra, 2001.<br />
Saltelli, A., Chan, K. <strong>and</strong> E. M. Scott (Eds.),<br />
Sensitivity Analysis, Wiley, Chichester,<br />
2000.<br />
1044
Dealing with unidentifiable sources of uncertainty within<br />
environmental models<br />
A. van Griensven a,b <strong>and</strong> T. Meixner b<br />
a<br />
BIOMATH, Ghent University, Belgium (ann.vangriensven@biomath.ugent.be)<br />
b<br />
University of California Riverside, California, USA (tmeixner@mail.ucr.edu)<br />
Abstract: Sources of Uncertainty Global Assessment using Split SamplES (SUNGLASSES) is a method for<br />
assessing model global uncertainty to aid in the development of integrated models. The method is<br />
complementary to the commonly investigated input <strong>and</strong> parameter uncertainty, as it accounts for errors that<br />
may arise due to unknown or unassessable sources of uncertainty, such as model hypothesis errors,<br />
simplifications, scaling effects or the lack of the observation period to represent long-term variability <strong>and</strong><br />
fluctuations in the system. Such sources are typically dominant for most environmental models <strong>and</strong> they<br />
undermine the reliability of environmental models.<br />
The SUNGLASSES algorithm directly estimates the overall predictive uncertainty without<br />
identifying or quantifying the underlying sources of uncertainties. The method uses the split sample approach<br />
to generate an estimate of model output uncertainty by selecting a threshold below which model simulations<br />
are determined to be acceptable. Where this methodology differs from other methods that use a threshold, is<br />
that the threshold is determined by evaluating the confidence bounds on model outputs during an evaluation<br />
time period of data that was not used to initially calibrate the model <strong>and</strong> generate parameter estimates. Where<br />
parameter uncertainty is often assessed using some goodness-of-fit criterion such as the mean squared errors,<br />
SUNGLASSES focuses on a criterion that evaluates the correctness of the model output values to be used<br />
directly in decision making, such as total mass balance assessments or violations of st<strong>and</strong>ards as imposed by<br />
legislation. The described method is applied to the integrated water quality modelling tool, SWAT2003,<br />
applied to Honey Creek, a tributary of the S<strong>and</strong>usky catchment in Ohio. Water flow <strong>and</strong> sediment loads are<br />
analysed. The incorporation of the split sample approach in the methodology produces a reasonable error<br />
bound that captures most of the observations during both the initial calibration period <strong>and</strong> during the<br />
evaluation period.<br />
Keywords: Uncertainty; Catchment; Water quality; <strong>Modelling</strong><br />
1. INTRODUCTION<br />
Applications of environmental models are<br />
important tools for decision making. But, the<br />
model results can be highly uncertain <strong>and</strong> may<br />
therefore adversely impact decisions. Therefore, it<br />
is important to know the reliability of model<br />
results. In this context, it is important to distinguish<br />
between confidence intervals for model results <strong>and</strong><br />
reliability. Confidence intervals are the results of<br />
an uncertainty analysis, while reliability depends<br />
on the completeness of the uncertainty analysis that<br />
should ideally cover all sources of uncertainty, i.e.,<br />
physical input uncertainty, parameter input<br />
uncertainty, model structure <strong>and</strong> code hypothesis<br />
uncertainty. Therefore, it is important to assign<br />
reliability levels on the model predictions of<br />
interest (i.e. the model outputs being used for<br />
decision making) even when the sources of<br />
uncertainty are not completely known or<br />
understood.<br />
2. METHODS<br />
2. 1 Introduction<br />
Two methods for assessing uncertainty are<br />
presented: ParaSol <strong>and</strong> SUNGLASSES. ParaSol is<br />
an optimisation <strong>and</strong> statistical method for the<br />
assessment of parameter uncertainty. In the<br />
literature, many methods for parameter uncertainty<br />
exist. Compared to these methods, ParaSol can be<br />
classified as being global, efficient <strong>and</strong> being able<br />
to deal with multiple objectives. These<br />
requirements for an uncertainty method are typical<br />
for environmental models of many types.<br />
On top of ParaSol, SUNGLASSES uses all<br />
parameter sets <strong>and</strong> simulations that were generated<br />
by ParaSol. SUNGLASSES aims at detecting<br />
1045
additional sources of uncertainty by using an<br />
evaluation period in addition to the calibration<br />
period.<br />
2.2 ParaSol<br />
ParaSol – Parameter Solutions - (van Griensven<br />
<strong>and</strong> Meixner, 2004a) operates by a parameter<br />
search method for model parameter optimisation –<br />
a modified version of the SCE-UA method (Duan<br />
et al., 1992) – followed by a statistical method that<br />
uses the model runs that were performed during the<br />
optimisation to provide parameter uncertainty<br />
bounds <strong>and</strong> the corresponding uncertainty bounds<br />
on the model outputs.<br />
The Shuffled complex evolution algorithm<br />
This algorithm conducts a global minimisation of a<br />
single function for up to 16 parameters (Duan et<br />
al., 1992). In a first step (zero-loop), SCE-UA<br />
selects an initial ‘population’ by r<strong>and</strong>om sampling<br />
throughout the feasible parameters space for p<br />
parameters to be optimised (delineated by given<br />
parameter ranges). The population is portioned into<br />
several “complexes” that consist of 2p+1 points.<br />
Each complex evolves independently using the<br />
simplex algorithm. The complexes are periodically<br />
shuffled to form new complexes in order to share<br />
the gained information. SCE-UA has been widely<br />
used in watershed model calibration <strong>and</strong> other<br />
areas of hydrology such as soil erosion, subsurface<br />
hydrology, remote sensing <strong>and</strong> l<strong>and</strong> surface<br />
modelling (Duan, 2003). It has been found to be<br />
robust, effective <strong>and</strong> efficient (Duan, 2003).<br />
Objective functions<br />
Sum of the squares of the residuals (SSQ): This<br />
objective function is similar to the Mean Square<br />
Error function (MSE) <strong>and</strong> aims at estimating the<br />
matching of a simulated series to a measured time<br />
series.<br />
SSQ =<br />
∑[ TF(<br />
x<br />
measured<br />
− TF x<br />
simulated<br />
]<br />
n ,<br />
) (<br />
n ,<br />
)<br />
n=<br />
1, N<br />
(1)<br />
with n the number of pairs of measured (x measured )<br />
<strong>and</strong> simulated (x simulated ) variables <strong>and</strong> TF a user<br />
defined transformation function.<br />
The sum of the squares of the difference of the<br />
measured <strong>and</strong> simulated values after ranking<br />
(SSQR): The SSQR method aims at the fitting of<br />
the frequency distributions of the observed <strong>and</strong> the<br />
simulated series. As opposed to the SSQ method,<br />
the time of occurrence of a given value of the<br />
variable is not accounted for in the SSQR method<br />
(van Griensven <strong>and</strong> Bauwens, 2003).<br />
2<br />
After independent ranking of the measured <strong>and</strong> the<br />
simulated values, new pairs are formed <strong>and</strong> the<br />
SSQR is calculated as<br />
SSQR =<br />
2<br />
∑[ x rank measured<br />
−<br />
simulated<br />
]<br />
r ,<br />
x<br />
rank r ,<br />
r=<br />
1, N<br />
where r represents the rank.<br />
Multi-objective optimisation<br />
(2)<br />
Several SSQ’s or SSQR’s can be combined to a<br />
Global Optimisation Criterion (GOC) using (van<br />
Griensven <strong>and</strong> Meixner, 2004):<br />
GOC<br />
=<br />
M<br />
∑<br />
(3)<br />
The probability of a given parameter solution<br />
being the best one is related to the GOC according<br />
to (van Griensven <strong>and</strong> Meixner, 2004):<br />
p(θ<br />
| Yobs ) ∝ exp −<br />
SSQm<br />
* N<br />
SSQ<br />
m=<br />
1 m,<br />
min<br />
[ GOC]<br />
(4)<br />
Thus the sum of the squares of the residuals get<br />
weights that are equal to the number of<br />
observations divided by the minimum. This<br />
equation allows also for uncertainty analysis as<br />
described below.<br />
Parameter change options<br />
In this optimisation <strong>and</strong> uncertainty algorithm<br />
parameters affecting hydrology or pollution can be<br />
changed either in a lumped way (over the entire<br />
catchment), or in a distributed way (for selected<br />
subbasins or HRU’s). They can be modified by<br />
replacement, by addition of an absolute change or<br />
by a multiplication of a relative change. A<br />
parameter is never allowed to go beyond<br />
predefined parameter ranges. A relative change<br />
allows for a lumped calibration of distributed<br />
parameters while they keep their relative physical<br />
meaning (soil conductivity of s<strong>and</strong> will be higher<br />
than soil conductivity of clay). This last method of<br />
relative change is the method utilised here.<br />
Uncertainty analysis method<br />
The uncertainty analysis divides the simulations<br />
that have been performed by the SCE-UA<br />
optimisation into ‘good’ simulations <strong>and</strong> ‘not<br />
good’ simulations. The simulations gathered by<br />
SCE-UA are very valuable as the algorithm<br />
samples over the entire parameter space with a<br />
focus of solutions near the optimum/optima. There<br />
are two separation techniques, both are based on a<br />
threshold value for the objective function (or<br />
global optimisation criterion) to select the ‘good’<br />
simulations by considering all the simulations that<br />
give an objective function below this threshold.<br />
m<br />
1046
is<br />
The threshold value can be defined by ̐2 -statistics<br />
where the selected simulations correspond to the<br />
confidence region (CR) or Bayesian statistics that<br />
are able to identify the high probability density<br />
region (HPD) for the parameters or the model<br />
outputs (figure 1).<br />
̐2 -method<br />
For a single objective calibration for the SSQ, the<br />
SCE-UA will find a parameter set ¡* consisting of<br />
the p free parameters (¢*1, ¢*2,… ¢*p), that<br />
corresponds to SSQ min , the minimum of the sum<br />
the square SSQ. According to ̐2 statistics, we can<br />
define a threshold “c” for “good’ parameter sets<br />
using equation:<br />
c = SSQ<br />
(5)<br />
whereby n is the number of observations <strong>and</strong> p the<br />
number of free parameters. The<br />
̐2<br />
p,0.95 gets a higher<br />
value for more free parameters p.<br />
For multi-objective calibration, the selections are<br />
made using the GOC of equation (3) . A threshold<br />
for the GOC is the calculated by:<br />
c = GOC<br />
(6)<br />
with NTOT the total number of observations for all<br />
the objective functions considered in the GOC.<br />
All parameter sets that give simulations with a<br />
GOC below the value “c’ will be selected as<br />
“good” parameter sets.<br />
Smax<br />
200<br />
180<br />
160<br />
140<br />
120<br />
100<br />
80<br />
60<br />
40<br />
20<br />
min<br />
min<br />
2<br />
χ<br />
* (1 +<br />
n<br />
p<br />
−<br />
,0.95<br />
2<br />
χ p,0.95<br />
*(1 + )<br />
NTOT − p<br />
Results for 743 ParaSol simulations<br />
ParaSol runs̐2- Confidence region Bayesian confidence region<br />
0<br />
0.0 0.2 0.4 0.6 0.8 1.0<br />
Figure 1: Confidence region for the ̐2 -statistics<br />
<strong>and</strong> the Bayesian statistics for the 2 parameters<br />
Smax <strong>and</strong> k of a simple 2-parameter model.<br />
Bayesian method (Box <strong>and</strong> Tiao, 1974)<br />
p<br />
k<br />
)<br />
This option is described briefly since it is not<br />
chosen for the case study discussed in this paper.<br />
In accordance to the Bayesian theorem, the<br />
probability p(́|Yobs) of a parameter set<br />
proportional to the GOC (equation 4) upon the<br />
assumption that the initial parameter distribution is<br />
equal to the uniform distribution. After<br />
normalizing the probabilities (to ensure that the<br />
integral over the entire parameter space is equal to<br />
1) a cumulative distributions can be made <strong>and</strong><br />
hence a 95% confidence regions can be defined.<br />
As the parameters sets were not sampled r<strong>and</strong>omly<br />
but were more densely sampled near the optimum<br />
during SCE-UA optimisation, it is necessary to<br />
avoid having the densely sampled regions<br />
dominate the results. This problem is prevented by<br />
determining a weight for each parameter set<br />
́<br />
́i by<br />
the following calculations (van Griensven <strong>and</strong><br />
Meixner, 2004).<br />
2.3 SUNGLASSES<br />
To develop a stronger evaluation of the model<br />
prediction power, the Sources of Uncertainty<br />
Global Assessment using Split SamplES<br />
(SUNGLASSES) was designed to assess predictive<br />
uncertainty that is not captured by the parameter<br />
uncertainty estimated by Parasol. The<br />
SUNGLASSES method accounts for strong<br />
increases in errors when simulations are done<br />
outside the calibration period by using a splitsample<br />
strategy whereby the validation period is<br />
used to set uncertainty ranges.<br />
These uncertainty ranges depend on the GOC, used<br />
during a calibration period representing the<br />
objective functions, <strong>and</strong> an evaluation criterion (to<br />
be used in decision making) used during an<br />
evaluation period. The GOC is used to assess the<br />
degree of error in the process dynamics, while the<br />
evaluation criterion defines a threshold on the<br />
GOC. This threshold should be as small as<br />
possible, but the uncertainty ranges on the criteria<br />
should include the “true” value for both the<br />
calibration <strong>and</strong> the validation period, e.g. when<br />
mass balance is used as criteria, these “true” values<br />
are a model bias equal to zero. Thus, the threshold<br />
is increased till the uncertainty ranges on the mass<br />
balance bias includes zero. SUNGLASSES<br />
operates by ranking the GOCs (Figure 2).<br />
Statistical methods can be used to define a<br />
threshold considering parameter uncertainty. In this<br />
case, ParaSol was used to define such a threshold.<br />
However, when we look at the predictions, it is<br />
possible that unbiased simulations are not within<br />
the ParaSol uncertainty range, which means that<br />
there are some more unknown uncertainties acting<br />
on the model outputs (Figure 3). Thus, a new,<br />
higher threshold is needed in order to have<br />
1047
unbiased simulations included in the uncertainty<br />
bounds (figure 2 <strong>and</strong> 3).<br />
GOC (log-scale)<br />
1.E+07<br />
1.E+06<br />
1.E+05<br />
1.E+04<br />
1.E+03<br />
Ranked GOCs for all SCE-UA simulations<br />
ParaSol threshold<br />
1.E+02<br />
1 2001 4001 6001 8001 10001 12001 14001 16001 18001<br />
rank<br />
Figure 2: Selection of good parameter sets<br />
using a threshold imposed by ParaSol or by SUN<br />
GLASSES<br />
Model bias for the sediment loads (%)<br />
SUNGLASSES threshold<br />
SUNGLASSES are programmed by the authors<br />
within the SWAT2003 version.<br />
16<br />
6 4<br />
17<br />
7<br />
3<br />
S<strong>and</strong>/Fremont<br />
2<br />
12<br />
14<br />
Tyomchlee Crk<br />
11<br />
15<br />
5<br />
9<br />
8<br />
10<br />
24<br />
Rock Crk<br />
1<br />
25<br />
Honey Creek<br />
Honey Crk<br />
26<br />
Honey Creek<br />
13<br />
18<br />
240<br />
200<br />
160<br />
120<br />
80<br />
40<br />
0<br />
-40<br />
1998-1999 2000-2001 1998-1999 2000-2001<br />
22<br />
27<br />
23<br />
20<br />
21<br />
19<br />
S<strong>and</strong>/Bucyrus<br />
28<br />
ParaSol<br />
SUNGLASSES<br />
Figure 3: Confidence regions for the sediment<br />
loads calculations according to ParaSol <strong>and</strong><br />
SUNGLASSES<br />
3. CASE STUDY<br />
The methods are applied to a river basin model of<br />
the Honey creek, a tributary of the S<strong>and</strong>usky river,<br />
Ohio (Figure 4) using the modelling tool “SWAT”.<br />
3.1 SWAT<br />
The Soil <strong>and</strong> Water Assessment Tool (SWAT)<br />
[Arnold et al., 1998] is a semi-distributed <strong>and</strong><br />
semi-conceptual program that calculates water,<br />
nutrient <strong>and</strong> pesticide transport at the catchment<br />
scale on a daily time step. It represents hydrology<br />
by interception, evapo-transpiration, surface runoff<br />
(SCS curve number method [USDA Soil<br />
conservation Service, 1972]), soil percolation,<br />
lateral flow <strong>and</strong> groundwater flow <strong>and</strong> river routing<br />
(variable storage coefficient method [Williams,<br />
1969]) processes. Other processes include<br />
nutrient, erosion, crop <strong>and</strong> pesticide, in-stream<br />
water quality processes. The catchment is divided<br />
into sub-basins, river reaches <strong>and</strong> Hydrological<br />
Response Units (HRU’s). While the sub-basins<br />
can be delineated <strong>and</strong> located spatially, the further<br />
sub-division into HRU’s is performed in a<br />
statistical way by considering a certain percentage<br />
of sub-basin area, without any specified location in<br />
the sub-basin. The methods ParaSol <strong>and</strong><br />
Figure 4: Location of the Honey creek within the<br />
S<strong>and</strong>usky basin.<br />
3.2 Model description<br />
A simple SWAT2003 model for Honey creek<br />
that covers 338 km 2 <strong>and</strong> consists of 1 subbasin (5<br />
HRU’s), 1 river reach <strong>and</strong> 1 point source near the<br />
mouth of the creek (van Griensven et al., 2004).<br />
Daily data for water flow <strong>and</strong> sediment<br />
concentrations were used for calibration <strong>and</strong><br />
evaluation of the model. Table 1 lists the 10 most<br />
important parameters for water flow <strong>and</strong> sediments<br />
concentrations, according to the results of a<br />
sensitivity analysis (van Griensven et al., 2004).<br />
Parameter<br />
SMFMX<br />
ALPHA_BF<br />
ch_k2<br />
USLE-P<br />
CN2<br />
sol_awc<br />
surlag<br />
SFTMP<br />
SMTMP<br />
Sol_z<br />
Table 1: parameters used in calibration<br />
Description<br />
Maximum melt rate for snow during<br />
(mm/°C/day)<br />
Baseflow alpha factor (days).<br />
Channel conductivity (mm/hr)<br />
USLE equation support practice (P)<br />
factor.<br />
SCS runoff curve number for moisture<br />
condition II.<br />
Available water capacity of the soil<br />
layer (mm/mm soil).<br />
Surface runoff lag coefficient<br />
Snowfall temperature (°C)<br />
Snow melt base temperature (°C)<br />
Soil depth<br />
1048
(a)<br />
Figure 5: Confidence regions for the time series of the daily sediment loads (Box-Cox transformation for<br />
y-axis) according to ParaSol (a) <strong>and</strong> SUNGLASSES (b)<br />
(b)<br />
3.3 Objective functions<br />
SWAT was applied to the Honey Creek catchment<br />
to estimate sediment export from the catchment.<br />
Therefore, the joint calibration included the SSQ<br />
<strong>and</strong> SSQR for streamflow <strong>and</strong> SSQR for sediment<br />
loads, with a Box-Cox transformation to reduce the<br />
heteroscedastic nature of the residuals. The results<br />
allow an investigation of the joint uncertainty when<br />
both flow <strong>and</strong> water quality variables are used for<br />
model calibration as should be common practice<br />
for water quality models.<br />
3.4 Evaluation criterion<br />
Based on the assumption that the model purpose<br />
was to assess global fluxes of sediments load at the<br />
outlet of the creek, the evaluation criteria was<br />
described by the model biases on the mass flux that<br />
were calculated as:<br />
N<br />
N<br />
⌈<br />
⌉<br />
∑<br />
SIM<br />
n<br />
−∑OBS<br />
n <br />
n=<br />
1<br />
n=<br />
1<br />
BIAS = <br />
*100. (8)<br />
N<br />
<br />
<br />
∑OBSn<br />
⌊<br />
<br />
n=<br />
1 ⌋<br />
for N the number of pairs (simulation,<br />
observation), SIM n the simulation at day n <strong>and</strong><br />
OBS n the observation of day n. The bias was<br />
calculated for the water flow <strong>and</strong> the sediment<br />
loads in the calibration <strong>and</strong> validation period.<br />
4 RESULTS AND DISCUSSION<br />
The confidence region for the sediment load<br />
calculations for ParaSol using the option ̐2 -<br />
statistics with<br />
̐2<br />
10,0.97.5 (Figure 5a) is much<br />
narrower <strong>and</strong> captures fewer observations than the<br />
confidence region for SUNGLASSES (Figure 5b).<br />
This result suggests that a traditional parameter<br />
uncertainty only covers a small share of the total<br />
uncertainty for cases where enough observations<br />
exist. Similarly, the confidence regions for the bias<br />
on the outputs of interest, i.e. the total loads, is<br />
much larger under SUNGLASSES than under<br />
ParaSol (Figure 3).<br />
The result that SUNGLASSES has a much larger<br />
uncertainty bound than the ParaSol method<br />
indicates that other causes of uncertainty are<br />
involved including: the inappropriateness of the<br />
data set to identify the important processes, model<br />
structural errors, <strong>and</strong> model discretisation errors.<br />
The latter sources are likely true of most<br />
distributed environmental models as they share<br />
many of the attributes of distributed water quality<br />
models (processes scaled up from point scale to<br />
l<strong>and</strong>scape scale, multiple criteria to meet, <strong>and</strong><br />
inadequate data availability to properly<br />
parameterise these models). Erosion processes<br />
require thus an even higher physically based<br />
analysis of the system in order to define proper<br />
processes <strong>and</strong> scaling. Erosion processes are as<br />
well dem<strong>and</strong>ing for the underlying hydrological<br />
processes, where a proper representation of the<br />
small scale processes is needed rather than a just<br />
some good curve fitting.<br />
The result that model structural error is a critical<br />
problem in water quality models is not unexpected<br />
as others have shown that structural changes in<br />
models can dramatically improve simulation results<br />
when focused on predicting floods [Boyle et al.,<br />
2001], predicting the effects of l<strong>and</strong> use change on<br />
streamflow <strong>and</strong> salinity [Kuczera <strong>and</strong><br />
Mroczkowski, 1998; Mroczkowski et al., 1997], or<br />
in finding flaws in models of stream chemical<br />
composition [Meixner et al., 2002]. Given this<br />
past experience of success in altering model<br />
structure <strong>and</strong> improving prediction results it is not<br />
surprising that model structural uncertainty is the<br />
1049
major source of predictive uncertainty when using<br />
water quality models. The result is thus reassuring<br />
since it indicates that what is needed for these<br />
models are a better way to represent the processes<br />
in them.<br />
SUNGLASSES somehow operates not only as a<br />
validation procedure for the model structure but<br />
also as a validation of the parameter uncertainty<br />
procedure of the model (in this case the application<br />
of ParaSol on the Honey creek model). This<br />
validation is related to model structure since a<br />
good model structure should require less data to<br />
capture all dynamics <strong>and</strong> to average out the errors<br />
than is the case for a poor model structure. In<br />
general, if all underlying assumptions of the<br />
parameter uncertainty method are correct <strong>and</strong> if the<br />
dataset is adequate to translate the variability of the<br />
system into a model, SUNGLASSES should not<br />
lead to larger uncertainty bounds for the model<br />
outputs.<br />
5 CONCLUSIONS<br />
The ParaSol results show an important drawback in<br />
traditional statistical uncertainty methods: these do<br />
account for the number of observations, but do not<br />
consider additional sources of uncertainty that are<br />
in general not known <strong>and</strong> not quantifiable, such as<br />
model hypothesis errors, simplifications, scaling<br />
effects or the lack of the observation period to<br />
represent, in the model, the long-term variability<br />
<strong>and</strong> fluctuations of the real world. These<br />
uncertainties lead to wrong assessments of<br />
indicators (like global mass balances) that might be<br />
used in decision making. Therefore,<br />
SUNGLASSES is proposed for assessing total<br />
uncertainty to aid in the development of integrated<br />
models. It reveals problems of bias in the model<br />
outputs to be used for decision making by<br />
evaluating predictions outside the calibration<br />
period. SUNGLASSES leads to more selections of<br />
parameter combinations <strong>and</strong> much wider<br />
uncertainty ranges. SUNGLASSES enables thus to<br />
assess predictive uncertainty <strong>and</strong> helps decision<br />
makers underst<strong>and</strong> how uncertain their models are<br />
so that they can put the proper level of trust in<br />
computational models of the environment as they<br />
move forward to make decisions. The results here<br />
indicate that the main concern should be about the<br />
uncertainty associated with model structural error<br />
<strong>and</strong> less so on model parametric uncertainty.<br />
6 ACKNOWLEDGEMENTS<br />
Support for this work was provided by the National<br />
Science Foundation (EAR-0094312 ). The authors<br />
are grateful to Sabine Grunwald <strong>and</strong> Tom Bishop<br />
of the University of Florida for sharing the<br />
S<strong>and</strong>usky catchment model <strong>and</strong> R. Srinivasan <strong>and</strong><br />
J. Arnold for their support of this research.<br />
REFERENCES<br />
Arnold J. G., R. Srinivasan, R. S. Muttiah, <strong>and</strong> J.<br />
R. Williams. Large area hydrologic modeling<br />
<strong>and</strong> assessment part I: model development.<br />
Journal of the American Water Resources<br />
Association, 34(1), 73-89, 1998.<br />
Box, G.E.P., <strong>and</strong> G.C.Tiao. Bayesian Inference in<br />
Statistical Analysis, Addison-Wesley-<br />
Longman, Reading, Mass, 1973.<br />
Boyle, D. P., H. V. Gupta, S. Sorooshian, V.<br />
Koren, Z. Zhang, <strong>and</strong> M. Smith, Toward<br />
improved streamflow forecasts: Value of<br />
semidistributed modeling, Water. Resourc.<br />
Res., 37:2749-2759, 2001.<br />
Duan, Q., V. K. Gupta, <strong>and</strong> S. Sorooshian,<br />
Effective <strong>and</strong> efficient global optimization for<br />
conceptual rainfall-runoff models, Water.<br />
Resourc. Res., 28:1015-1031, 1992.<br />
Duan, Q., S. Sorooshian, H. V. Gupta, A. N.<br />
Rousseau, <strong>and</strong> R. Turcotte, Advances in<br />
Calibration of Watershed Models,AGU,<br />
Washington, DC, 2003.<br />
Kuczera, G. <strong>and</strong> M. Mroczkowski, Assessment of<br />
hydrologic parameter uncertainty <strong>and</strong> the<br />
worth of multiresponse data, Water. Resourc.<br />
Res., 34:1481-1489, 1998.<br />
Mroczkowski, M., G. P. Raper, <strong>and</strong> G. Kuczera,<br />
The quest for more powerful validation of<br />
conceptual catchment models, Water.<br />
Resourc. Res., 33:2325-2335, 1997.<br />
Meixner, T., L. A. Bastidas, H. V. Gupta, <strong>and</strong> R.<br />
C. Bales, Multi-criteria parameter estimation<br />
for models of stream chemical composition,<br />
Water. Resourc. Res., 38(3):9-1-9-9, 2002.<br />
USDA Soil Conservation Service. National<br />
Engineering H<strong>and</strong>book Section 4 Hydrology,<br />
Chapter 19, 1983.<br />
van Griensven A. <strong>and</strong> W. Bauwens. Multiobjective<br />
auto-calibration for semi-distributed<br />
water quality models, Water. Resourc. Res.,<br />
in press, 2004.<br />
van Griensven A., T. Meixner, S. Grunwald, T.<br />
Bishop <strong>and</strong> R. Srinivasan. A global sensitivity<br />
analysis method for the parameters of multivariable<br />
watershed models, Journ. Hyrol.,<br />
submitted, 2004.<br />
van Griensven A. <strong>and</strong> T. Meixner. A global <strong>and</strong><br />
efficient multi-objective auto-calibration <strong>and</strong><br />
uncertainty method for water quality<br />
catchment models. Water Resources<br />
Research, submitted, 2004.<br />
Williams, J.R. Flood routing with variable travel<br />
time or variable storage coefficients. Trans.<br />
ASAE 12(1), 100-103.<br />
1050
Assessing SWAT model performance in the evaluation of<br />
management actions for the implementation of the<br />
Water Framework Directive in a Finnish catchment<br />
I. Bärlund a , T. Kirkkala b , O. Malve a <strong>and</strong> J. Kämäri a<br />
a<br />
Finnish Environment Institute (SYKE), P.O.Box 140, FI-00251 Helsinki<br />
b<br />
Southwest Finl<strong>and</strong> Regional Environment Centre, P.O.Box 47, FI-20801 Turku<br />
Abstract: The ecological status of Lake Pyhäjärvi may be classified as moderate due to its elevated nutrient<br />
concentrations <strong>and</strong> algal biomass production. Therefore, the Eurajoki river basin, including Lake Pyhäjärvi,<br />
has been chosen as the Finnish test catchment in an ongoing EU project on benchmarking models for the<br />
Water Framework Directive. One aim of the project is to test the suitability of models for the assessment of<br />
management options proposed to meet the surface water quality targets. The catchment model SWAT is<br />
currently being tested for its applicability for analysing the effectiveness of proposed measures to reduce<br />
agricultural <strong>and</strong> sparse settlement nutrient loading. The model is being applied to the river Yläneenjoki<br />
catchment draining to Lake Pyhäjärvi. First results indicate that SWAT can be calibrated for flow <strong>and</strong><br />
sediment yield using catchment scale parameters. For nutrients, however, parameters describing more<br />
detailed catchment processes have to be calibrated. The preliminary essay on measures such as buffer strips<br />
indicate that SWAT includes relevant management options that affect nutrient leaching. However, the<br />
descriptions of these management options require some modifications in order to describe correctly the<br />
reduction efficiency in local conditions.<br />
Keywords: Catchment; Nutrients; Agricultural practices; Water Framework Directive; SWAT<br />
1. INTRODUCTION<br />
The EU Water Framework Directive (WFD)<br />
m<strong>and</strong>ates Member States to develop river basin<br />
management plans for each river basin district. To<br />
achieve this the responsible authorities must have<br />
tools to assess alternative management options.<br />
One aim of the EU-funded project Benchmark<br />
models for the Water Framework Directive<br />
(BMW) is to establish a set of criteria to assess the<br />
appropriateness of models for the use in the<br />
implementation of WFD (Saloranta et al. 2003).<br />
Effects of environmental conditions <strong>and</strong><br />
agricultural practices on nutrient leaching have<br />
been studied in several field trials in Finl<strong>and</strong> (e.g.<br />
Puustinen 1994; Turtola <strong>and</strong> Kemppainen 1998).<br />
Due to complexity of the soil-water-plant<br />
interactions, the direct up-scaling of results from<br />
these singular field scale experiments to regional<br />
assessments of losses can be misleading.<br />
Therefore, mathematical modelling tools have<br />
been developed <strong>and</strong> modelling strategies set up to<br />
generalise the effect of environmental conditions<br />
<strong>and</strong> agricultural practices on nutrient losses on<br />
field <strong>and</strong> catchment scale. Models like<br />
SOIL/SOILN, GLEAMS <strong>and</strong> ICECREAM have<br />
been used to assess phosphorus (P) <strong>and</strong> nitrogen<br />
(N) losses from agricultural l<strong>and</strong> in Finl<strong>and</strong><br />
(Granlund et al. 2000; Knisel <strong>and</strong> Turtola 2000;<br />
Tattari et al. 2001). The SWAT model has been<br />
applied to the Vantaanjoki basin to estimate<br />
retention of total N <strong>and</strong> P in this Finnish river<br />
basin. The model performance was found to be<br />
satisfactory, the Nash-Suthcliffe coefficient for the<br />
simulation of flow <strong>and</strong> N <strong>and</strong> P loads ranged for<br />
validation from 0.43 to 0.57 (Grizzetti et al. 2003).<br />
The Finnish test case in the BMW project is based<br />
on linking models: first the lake model LakeState<br />
is used for setting the targets for the loading<br />
reduction for Lake Pyhäjärvi. Based on these<br />
results, the catchment model SWAT will be used<br />
for analysing the effectiveness of proposed<br />
measures to reduce agricultural <strong>and</strong> sparse<br />
settlement nutrient loading. In order to test the<br />
applicability of SWAT for this purpose, the model<br />
is being applied to the river Yläneenjoki catchment<br />
draining directly to Lake Pyhäjärvi <strong>and</strong><br />
1051
contributing over 50% of the phosphorus load<br />
reaching the lake. The modelling approach<br />
comprises three distinct phases: 1) the evaluation<br />
of the SWAT model utilising the available<br />
monitoring data along the Yläneenjoki reach <strong>and</strong><br />
its main tributaries, 2) linking the SWAT model to<br />
the lake model <strong>and</strong> to a simple economic costeffectiveness<br />
analysis to rank 3-5 management<br />
options, <strong>and</strong> 3) participation of the Finnish<br />
national <strong>and</strong> local stakeholders in the modelling<br />
process <strong>and</strong> communication of the analysis results.<br />
The third phase is particularly important since the<br />
Yläneenjoki catchment has been intensively<br />
studied by local water managers <strong>and</strong> thus one<br />
additional aim is to utilise the stakeholder knowhow<br />
in data interpretation <strong>and</strong> model<br />
parameterisation, <strong>and</strong> finally for the interpretation<br />
of results. In this paper first results for the phases<br />
1 <strong>and</strong> 3 are presented.<br />
2. MATERIAL AND METHODS<br />
2. 1 The research area<br />
Lake Pyhäjärvi, situated in the municipalities of<br />
Säkylä, Eura <strong>and</strong> Yläne in southwestern Finl<strong>and</strong>, is<br />
one of the most widely studied lakes in Finl<strong>and</strong>. In<br />
the 1970s, the water quality of Lake Pyhäjärvi was<br />
classified as excellent, but in the classification<br />
carried out in the 1990s, the water quality was<br />
only estimated as good. The eutrophication of the<br />
lake has progressed at a rapid pace over the last<br />
few years. Lake Pyhäjärvi is currently<br />
mesotrophic. The greatest threat to the lake is the<br />
nutrient load which exceeds the tolerance limit of<br />
the lake. According to studies <strong>and</strong> mathematical<br />
models, the phosphorus load to Lake Pyhäjärvi<br />
should be reduced to almost half of the present<br />
amount in order to stop the eutrophication process<br />
<strong>and</strong> to gradually improve water quality.<br />
The major inflows to Lake Pyhäjärvi are the rivers<br />
Yläneenjoki <strong>and</strong> Pyhäjoki, which cover 68% of the<br />
drainage basin. Of the total area 22% is cultivated,<br />
the remainder comprises forest, peatl<strong>and</strong> <strong>and</strong><br />
housing areas. Residential water leaves Lake<br />
Pyhäjärvi via the River Eurajoki.<br />
Field cultivation <strong>and</strong> animal husb<strong>and</strong>ry comprise<br />
55% <strong>and</strong> 39% of the external phosphorus (P) <strong>and</strong><br />
nitrogen (N) load to Lake Pyhäjärvi, respectively.<br />
Since the drainage basin of the lake is relatively<br />
small, atmospheric deposition to the lake is also an<br />
important component of the external load: it makes<br />
up 20% of the P load <strong>and</strong> 33% of the N load.<br />
Other external nutrient sources include the rural<br />
population, summer cottages <strong>and</strong> forestry.<br />
2. 2 The modelling tools<br />
Ecological response of Lake Pyhäjärvi to external<br />
nutrient loading is modelled using the LakeState<br />
model (Statistical Lake Assimilation Capacity<br />
Analysis Model). It is the enhanced version of the<br />
previous model of Lake Pyhäjärvi with improved<br />
re-suspension dynamics, loading optimisation<br />
subroutines <strong>and</strong> Bayesian Markov chain Monte<br />
Carlo sampling methods for the parameter<br />
estimation <strong>and</strong> prediction uncertainty analysis. The<br />
LakeState model is a dynamic CSTR<br />
(Continuously Stirred Tank Reactor) model that<br />
calculates total phosphorus <strong>and</strong> total nitrogen<br />
concentrations in the lake water body <strong>and</strong> their<br />
bulk mass in the active surface sediment layer on<br />
erosion, on transportation <strong>and</strong> on sedimentation<br />
bottom. The model calculates also the biomass of<br />
Diatomophycea, Chrysophycea, nitrogen fixing<br />
Blue Green Algae <strong>and</strong> minor groups of<br />
phytoplankton.<br />
The SWAT model (Soil <strong>and</strong> Water Assessment<br />
Tool) is a continuous time model that operates on<br />
a daily time step at catchment scale (Arnold et al.<br />
1998; Neitsch et al. 2001). It can be used to<br />
simulate water <strong>and</strong> nutrient cycles in agriculturally<br />
dominated l<strong>and</strong>scapes. The catchment is generally<br />
partitioned into a number of subbasins where the<br />
smallest unit of discretisation is a hydrologic<br />
response unit (HRU). SWAT is a process based<br />
model, including also empirical relationships. One<br />
objective of such a model is to assess long-term<br />
impacts of management practices. The model has<br />
been widely used but also further developed in<br />
Europe (e.g. Eckhardt et al. 2002; Krysanova et al.<br />
1999; van Griensven et al. 2002). SWAT was<br />
chosen for this case study for three main reasons:<br />
its ability to simulate both P <strong>and</strong> N on catchment<br />
scale, its European wide use <strong>and</strong> its potential to<br />
include agricultural management actions. Also,<br />
SWAT was evaluated against the diffuse pollution<br />
benchmark criteria developed by the BMW project<br />
<strong>and</strong> it was found to have potential with respect to<br />
the Water Framework Directive requirements in<br />
Scotl<strong>and</strong> (Dilks et al. 2003).<br />
2. 3 The modelling approach<br />
The modelling approach in this case study is based<br />
on linking the Yläneenjoki catchment with Lake<br />
Pyhäjärvi (Figure 1). The lake model LakeState is<br />
utilised to set the load reduction target which is<br />
required to improve water quality of Lake<br />
Pyhäjärvi. The task of the SWAT model is to<br />
assess the possibility to reach this target using a<br />
variety of management options such as buffer<br />
strips or changes in fertilisation practices.<br />
1052
Load<br />
reduction<br />
target<br />
Yläneenjoki catchment<br />
Lake Pyhäjärvi<br />
Effect of<br />
management<br />
practices<br />
Figure 1. The modelling set-up.<br />
2. 4 Setting up the SWAT model<br />
The regular monitoring of water quality of river<br />
loads has been started as early as 1970s.<br />
Monitoring of ditches <strong>and</strong> brooks entering the<br />
rivers or lake started at the beginning of 1990s.<br />
The nutrient load has been monitored in the<br />
Yläneenjoki river by taking <strong>and</strong> analysing, in<br />
general bi-weekly, water samples <strong>and</strong> measuring<br />
the daily water flow at one point (Vanhakartano).<br />
Furthermore, water quality was monitored on a<br />
monthly basis in three additional points in the<br />
main channel <strong>and</strong> in 13 open ditches running into<br />
the river Yläneenjoki in the 1990s.<br />
For the SWAT simulations the available data on<br />
l<strong>and</strong> use <strong>and</strong> soil types had to be aggregated.<br />
Forests in Finl<strong>and</strong> are classified according to their<br />
stage of growth <strong>and</strong> dominant tree <strong>and</strong> soil type<br />
into ca. 50 classes. Since the parameterisation of<br />
all these would be an overwhelming task the<br />
classes were regrouped according to their stage of<br />
growth. Similar regrouping was required for<br />
coarse soils which show a great variety but only<br />
patchwork locations within the catchment: tills, till<br />
ridges, eskers, gravel <strong>and</strong> coarse s<strong>and</strong> were<br />
grouped <strong>and</strong> parameterised according to till<br />
characteristics which is the dominant type. In<br />
conclusion, the SWAT parameterisation was<br />
performed for 7 l<strong>and</strong> use types (water, field, forest<br />
cuts <strong>and</strong> recently planted forest, active forest, old<br />
forest, peat bog <strong>and</strong> sealed areas) <strong>and</strong> 6 soil types<br />
(open bedrock, till <strong>and</strong> other coarse soils, silt, clay<br />
<strong>and</strong> turf). The first classification of the<br />
Yläneenjoki catchment resulted in 30 subbasins.<br />
With a threshold value of 20% for l<strong>and</strong> use <strong>and</strong><br />
10% for soil types the number of HRU's is 116.<br />
This approach that resulted in few HRU's was<br />
chosen in this first calibration essay to keep an<br />
overview of the calibration process. This meant,<br />
however, that l<strong>and</strong> use as well as soil types were<br />
reduced to three. The average size of the HRU's is<br />
867 ha. The preliminary tests of the SWAT model<br />
showed that the available GIS data on soils <strong>and</strong><br />
l<strong>and</strong> use is suitable for the basic model set up but<br />
requires reclassification.<br />
The first step taken was to perform an uncalibrated<br />
model run for stream flow <strong>and</strong> nutrient<br />
concentrations at the measurement point<br />
Vanhakartano, which is situated ca. 4 km from the<br />
river mouth. This was performed for the years<br />
1990-1994. The parameterisation was based on<br />
measurements, expert judgement <strong>and</strong> previous<br />
field scale modelling work (i.e. using the<br />
ICECREAM model). Clear information gaps for<br />
the Yläneenjoki data set concerned a wide range of<br />
parameters (ca. 30) where model default values<br />
are now used. For certain parameters like PSP (P<br />
availability index) <strong>and</strong> PHOSKD (P soil<br />
partitioning coefficient) estimates for field soils<br />
exist but no information about their values in<br />
forest soils is available.<br />
3. RESULTS AND DISCUSSION<br />
3. 1 The environmental target<br />
Posterior parameter distributions for the LakeState<br />
model were estimated <strong>and</strong> cross validated for the<br />
years 1992-1999. Clear correlation between<br />
parameters <strong>and</strong> parameter unidentifiability<br />
problems were revealed. The difference between<br />
observed <strong>and</strong> simulated lake total phosphorus<br />
dynamics <strong>and</strong> algal biomass peaks was acceptable<br />
while simulated total nitrogen did not perform as<br />
well. Average parameter values were utilised for<br />
the optimisation of nutrient loading.<br />
The ecological target for the lake nutrient load<br />
reduction was set thus that the biomass of nitrogen<br />
fixing Blue Green algae in the lake did not exceed<br />
0.2 mg l -1 . External total phosphorus <strong>and</strong> total<br />
nitrogen loadings were optimised with lake data<br />
for the years 1992-1999 to obtain the target. It was<br />
concluded that 40% reduction of external loading<br />
is necessary.<br />
3. 2 Calibration <strong>and</strong> model evaluation<br />
The uncalibrated SWAT run showed clear faults in<br />
the ability to describe observed processes. For<br />
discharge this concerned mainly three<br />
phenomenon: too much snow melt during winter<br />
months, timing <strong>and</strong> amount of snow melt in spring<br />
<strong>and</strong> too many <strong>and</strong> partially over-predicted peaks<br />
during summer (Figure 2a). These were tackled in<br />
the calibration procedure, where in the first phase<br />
only basin-wide parameters were changed. A<br />
reasonable fit was acquired using four parameters<br />
(Figure 2b): SURLAG (surface runoff lag<br />
coefficient), SFTMP (snowfall temperature),<br />
SMTMP (snow melt base temperature) <strong>and</strong> TIMP<br />
1053
(snow pack temperature lag factor). The Nash-<br />
Suthcliffe coefficient for the period 1990-1994<br />
was 0.44. An improvement is still needed to<br />
correct the slow reduction of flow after the snow<br />
melt period in May. This is probably due to<br />
inadequate interaction between surface water <strong>and</strong><br />
groundwater. This will be the next calibration step<br />
on subbasin level.<br />
a) 35<br />
30<br />
25<br />
20<br />
15<br />
10<br />
5<br />
0<br />
01-93 04-93 06-93 09-93 12-93 03-94 06-94 09-94 12-94<br />
[m 3 s -1 ]<br />
b)<br />
[m 3 s -1 ]<br />
35<br />
30<br />
25<br />
20<br />
15<br />
10<br />
5<br />
Q_uncal<br />
0<br />
01-93 04-93 06-93 09-93 12-93 03-94 06-94 09-94 12-94<br />
obs<br />
Figure 2. Comparison of the uncalibrated<br />
(Q_uncal, a) <strong>and</strong> calibrated (Q_cal, b) flow at<br />
Vanhakartano (obs) for the years 1993-1994.<br />
The water balance components at basin scale in<br />
average for 1990-1995 were roughly in line with<br />
expert judgement. The actual evapotranspiration<br />
was assessed to be too low, whereas surface runoff<br />
was perceived as being too high. This means that<br />
parameters governing surface runoff, e.g. the SCS<br />
runoff curve numbers, need a closer scrutiny on<br />
HRU level.<br />
Calibration improved the sediment concentration<br />
simulation result only in regard to one basin-wide<br />
parameter: PRF (peak rate adjustment factor for<br />
sediment routing in the main channel). The Nash-<br />
Suthcliffe coefficient value 0.22 indicates,<br />
however, further need of improvement.<br />
No basin-wide parameter improved the simulation<br />
of the nutrient concentrations. Generally, the<br />
overall PO 4 -P <strong>and</strong> total P peak concentrations are<br />
overestimated <strong>and</strong> the nitrogen concentrations<br />
(NO 3 +NO 2 -N, NH 4 -N <strong>and</strong> totN) underestimated.<br />
The measurements indicate also clearly less<br />
variability in the concentration values between the<br />
months <strong>and</strong> seasons than what is the first simulated<br />
impression.<br />
The excellent distribution of monitoring points for<br />
water quality variables within the catchment can<br />
be utilised to assess the most significant areas, e.g.<br />
critical soil-crop combinations, for the calibration<br />
process. For sediment concentration, for example,<br />
obs<br />
Q_cal<br />
a)<br />
the higher concentrations at the point most<br />
upstream (P4) caused by intensive agriculture can<br />
be very well depicted by the simulation (Figure<br />
3a). The rise in concentration between P2<br />
(Vanhakartano) <strong>and</strong> P1 closest to the lake,<br />
however, is very difficult to explain since the<br />
distance between the points is only ca. 2 km <strong>and</strong> in<br />
the simulation the main channel reaches belong to<br />
the same channel type. There might be an error in<br />
the subbasin discretisation just above P1. On the<br />
other h<strong>and</strong> the measurement point at P1 is<br />
occasionally influenced by rising lake water <strong>and</strong><br />
there is measured data for only the one year 1994<br />
(Figure 3b).<br />
Figure 3. Simulated (a) <strong>and</strong> measured (b) average<br />
annual sediment concentrations in four points<br />
along the main channel.<br />
Another way to utilise the monitoring data<br />
available is to make a comparison between<br />
different types of subbasins. There is a clear<br />
difference in the measured average total nitrogen<br />
concentrations between forest dominated<br />
subbasins F1-F3 <strong>and</strong> agriculture dominated<br />
subbasins A1-A3 (Figure 4a). On this scale the<br />
simulated concentrations are clearly<br />
underestimated but additionally it can be noted<br />
that the difference between forest dominated <strong>and</strong><br />
agriculture dominated subbasins in the simulation<br />
is clearly higher.<br />
meas. average conc. [ug l -1 ]<br />
a)<br />
average conc [mg l -1 ]<br />
b)<br />
ann. average [mg l -1 ]<br />
3500<br />
3000<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
200<br />
150<br />
100<br />
50<br />
0<br />
200<br />
150<br />
100<br />
0<br />
50<br />
0<br />
1991 1992 1993 1994<br />
1991 1992 1993 1994<br />
F1 F2 F3 A1 A2 A3<br />
b)<br />
SED(P1)_uncal<br />
SED(P2)_cal<br />
SED(P3)_uncal<br />
SED(P4)_uncal<br />
SED(P1)<br />
SED(P2)<br />
SED(P3)<br />
SED(P4)<br />
Figure 4. Measured (a) <strong>and</strong> simulated (b) average<br />
annual total nitrogen concentrations in the outlet of<br />
three forest dominated subbasins (F1-F3) <strong>and</strong> in<br />
three agriculture dominated subbasins (A1-A3).<br />
The main difference to the observed set-up is that<br />
the SWAT is at present rather roughly discretised:<br />
the areas F1-F3 consist of only forest <strong>and</strong> they are<br />
parameterised in a way that very little erosion <strong>and</strong><br />
sim. average conc. [ug l -1 ]<br />
3500<br />
3000<br />
2500<br />
2000<br />
1500<br />
1000<br />
500<br />
0<br />
F1 F2 F3 A1 A2 A3<br />
1054
nutrient leaching should occur, which is according<br />
to experimental experience. In reality all areas<br />
include some agriculture, <strong>and</strong> it might be situated<br />
close to the monitoring point. The areas A1-A3 are<br />
in relation rather well depicted. The<br />
underestimation may be corrected by calibration,<br />
e.g. by testing parameters governing sediment loss,<br />
N mineralisation <strong>and</strong> fertilisation distribution. The<br />
question to what extent the discretisation should be<br />
improved has to be discussed with the local water<br />
manager in order to assess the role of agriculture<br />
in the forest dominated subbasins, i.e. should the<br />
emphasis be on improving the discretisation or the<br />
parameterisation. One clear model deficiency in<br />
assessing nitrogen leaching from forested areas is<br />
the absence of soluble organic N as an output<br />
variable. The major part of total N leached from<br />
forest soils is in soluble organic form.<br />
3. 3 Management options<br />
Despite the fact that the presented SWAT set-up is<br />
still uncalibrated for nutrients a simple analysis of<br />
the management options to reduce nutrient loading<br />
was performed. Model performance on including<br />
management options plays a conclusive role in<br />
evaluating SWAT since this is the main interest of<br />
the local water manager.<br />
The option essayed was to include 15m wide<br />
buffer strips for all agricultural HRUs on clay or<br />
silt soil (mainly located in the river valleys).<br />
The only input parameter governing buffer strip<br />
efficiency is its width. The efficiency is calculated<br />
internally using the equation:<br />
TRAPEFF = 0.<br />
367⋅WIDTH<br />
The emerging curve was compared with average<br />
efficiency values from two Finnish data sets (Uusi-<br />
Kämppä et al. 1996 & 2000; Puustinen 1999)<br />
(Figure 5).<br />
efficiency [-]<br />
02967 .<br />
1.0<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
0 2 4 6 8 10 12 14 16 18 20 22<br />
width of the strip [m]<br />
trapeff<br />
eff(sed)<br />
eff(sedP)<br />
eff(totP)<br />
eff(solP)<br />
eff(no3)<br />
eff(totN)<br />
eff(sed)<br />
eff(sedP)<br />
eff(solP)<br />
eff(no3)<br />
eff(totN)<br />
Figure 5. Efficiency of buffer strips to reduce<br />
sediment <strong>and</strong> nutrient runoff according to SWAT<br />
(trapeff) <strong>and</strong> two sets of measurements.<br />
It can be seen that the efficiency according to<br />
SWAT for a certain buffer strip width is clearly<br />
overestimated when compared to Finnish field<br />
scale experience. More important still, the<br />
measurements indicate that the buffer strip<br />
performs differently depending on the variable<br />
studied. In SWAT the efficiency is the same for all<br />
output variables on HRU scale.<br />
The effect of the buffer strip option on annual<br />
nutrient loading was tested using the parameter set<br />
according to the current calibration status (Table<br />
1). A SWAT simulation with buffer strips was<br />
compared to a simulation without them.<br />
Table 1. Change in selected annual output<br />
variables in the 5 th year after start of the<br />
simulation at Vanhakartano compared to<br />
simulation results without buffer strips.<br />
change in %<br />
discharge 0<br />
sediment -0.34<br />
orgN* -80<br />
orgP* -80<br />
NO 3 -N -56<br />
NH 4 -N 0<br />
soluble P -64<br />
*orgN/P: sediment bound N <strong>and</strong> P<br />
This small analysis reveals that there is a<br />
discrepancy between the sediment concentration<br />
<strong>and</strong> the sediment bound organic N <strong>and</strong> P at<br />
Vanhakartano. The orgN/P change seems to be the<br />
same as the change in the HRUs. There is<br />
probably no change in the surface waters for these<br />
variables, i.e. a lack in parameterisation. The NH 4 -<br />
N concentration in the river seems not to be<br />
connected to agriculture at present. This<br />
information will be utilised in further calibration<br />
<strong>and</strong> testing work.<br />
4. CONCLUSIONS AND OUTLOOK<br />
The overall objective of the BMW project is to<br />
evaluate the suitability of models for the use in the<br />
implementation of the WFD. Therefore, the<br />
intention of this application is not to produce<br />
precise simulations for management purposes,<br />
rather the objective is to test the model's<br />
applicability to assess the effectiveness of potential<br />
management options. SWAT includes relevant<br />
management options that affect nutrient leaching.<br />
However, the descriptions of these management<br />
options (e.g. buffer strips) require some<br />
modifications in order to describe correctly the<br />
reduction efficiency in local conditions. SWAT<br />
applicability for Finnish conditions can be<br />
improved additionally by creating a national<br />
parameter data base collating experiences from all<br />
applications in similar conditions. Scenario runs<br />
1055
increase the underst<strong>and</strong>ing of processes in the<br />
model that should be improved.<br />
The SWAT model includes presently a variety of<br />
parameters for which there is no information<br />
available. In order to focus on the most significant<br />
ones, a systematic sensitivity analysis is needed.<br />
This would also aid in successful calibration of the<br />
water quality variables. Neither is much known<br />
about model performance for forest soils <strong>and</strong> thus<br />
forest dominated subbasins. This could be tested in<br />
a basin growing only forest.<br />
5. ACKNOWLEDGEMENTS<br />
Prof. Jouko Sarvala from Turku University<br />
(Department of Biology, Section of Ecology) is<br />
acknowledged for providing lake ecological data<br />
which has been used for lake modelling. The<br />
financial support of the BMW project through the<br />
European Commission's 5th Framework<br />
Programme (contract EVK1-CT-2001-00093) is<br />
gratefully acknowledged.<br />
6. REFERENCES<br />
Arnold J.G., R. Srinivasan, R.S. Muttiah <strong>and</strong> J.R.<br />
Williams, Large area hydrologic modelling<br />
<strong>and</strong> Assessment part I: model development,<br />
Journal of American Water Resources<br />
Association, 34(1), 73-89, 1998.<br />
Dilks C.F., S.M. Dunn <strong>and</strong> R.C. Ferrier,<br />
Benchmarking models for the Water<br />
Framework Directive: evaluation of SWAT<br />
for use in the Ythan catchment, UK, IN:<br />
Arnold, J. et al. (eds.), Condensed abstracts<br />
of the 2nd <strong>International</strong> SWAT Conference,<br />
1.-4.7.2003, Bari, Italy, 63-66, 2003.<br />
Eckhardt K., S. Haverkamp, N. Fohrer <strong>and</strong> H.-G.<br />
Frede, SWAT-G, a version of SWAT99.2<br />
modified for application to low mountain<br />
range catchments, Physics <strong>and</strong> Chemistry of<br />
the Earth 27, 641-644, 2002.<br />
Granlund K., S. Rekolainen, J. Grönroos, A.<br />
Nik<strong>and</strong>er <strong>and</strong> Y. Laine, Estimation of the<br />
impact of fertilisation rate on nitrate leaching<br />
in Finl<strong>and</strong> using a mathematical simulation<br />
model, Agriculture, Ecosystems <strong>and</strong><br />
Environment 80, 1-13, 2000.<br />
Grizzetti B., F. Bouraoui, K. Granlund, S.<br />
Rekolainen <strong>and</strong> G. Bidoglio, <strong>Modelling</strong><br />
diffuse emission <strong>and</strong> retention of nutrients in<br />
the Vantaanjoki watershed (Finl<strong>and</strong>) using<br />
the SWAT model, Ecological <strong>Modelling</strong> 169,<br />
25-38, 2003.<br />
Knisel W.G. <strong>and</strong> E. Turtola, GLEAMS model<br />
application on a heavy clay soil in Finl<strong>and</strong>.<br />
Agric. Water Manage., 43, 285-309, 2000.<br />
Krysanova V., A. Bronstert <strong>and</strong> D.-I. Müller-<br />
Wohlfeil, <strong>Modelling</strong> river discharge for large<br />
drainage basins: from lumped to distributed<br />
approach, Hydrological Sciences-Journal des<br />
Sciences Hydrologiqes, 44(2), 313-331,<br />
1999.<br />
Neitsch S.L., J.G. Arnold, J.R. Kiniry <strong>and</strong> J.R.<br />
Williams, Soil <strong>and</strong> Water Assessment Tool –<br />
Theoretical Documentation - Version 2000,<br />
Blackl<strong>and</strong> Research Center – Agricultural<br />
Research Service, Texas – USA, 2001.<br />
Puustinen M., Effect of soil tillage on erosion <strong>and</strong><br />
nutrient transport in plough layer runoff,<br />
Publ. of the Water <strong>and</strong> Environment<br />
Research Institute, 17, 71-90, 1994.<br />
Saloranta T.M., J. Kämäri, S. Rekolainen <strong>and</strong> O.<br />
Malve, Benchmark criteria: a tool for<br />
selecting appropriate models in the field of<br />
water management, <strong>Environmental</strong><br />
Management, 32(3), 322-333, 2003.<br />
Tattari S., I. Bärlund, S. Rekolainen, M. Posch, K.<br />
Siimes, H.-R. Tuhkanen <strong>and</strong> M. Yli-Halla,<br />
<strong>Modelling</strong> sediment yield <strong>and</strong> phosphorus<br />
transport in Finnish clayey soils.<br />
Transactions of ASAE, 44(2), 297-307, 2001.<br />
Turtola E. <strong>and</strong> E. Kemppainen, Nitrogen <strong>and</strong><br />
phosphorus losses in surface <strong>and</strong> drainage<br />
water after application of slurry <strong>and</strong> mineral<br />
fertilizer to perennial grass ley. Agric. Food<br />
Sci. Finl., 7, 569-581, 1998.<br />
Uusi-Kämppä J., E. Turtola, H. Hartikainen <strong>and</strong> T<br />
Yläranta, The ineractions of buffer zones <strong>and</strong><br />
phosphorus runoff, Quest <strong>Environmental</strong>,<br />
Buffer zones: Their processes <strong>and</strong> potential<br />
in water protection (eds. N.E. Haycock, T.P.<br />
Burt, K.W.T. Goulding <strong>and</strong> G. Pinay), 43-53,<br />
1996.<br />
Uusi-Kämppä J., B. Braskerud, H. Jansson, N.<br />
Syversen <strong>and</strong> R. Uusitalo, Buffer zones <strong>and</strong><br />
constructed wetl<strong>and</strong>s as filters for agricultural<br />
phosphorus. Journal of <strong>Environmental</strong><br />
Quality, 29(1), 151-158, 2000.<br />
Van Griensven A., A. Francos <strong>and</strong> W. Bauwens.<br />
Sensitivity analysis <strong>and</strong> autocalibration of an<br />
integral dynamic model for river water<br />
quality, Water Science <strong>and</strong> Technology,<br />
45(5), 321-328, 2002.<br />
1056
Assessing the Effects of Agricultural Change on Nitrogen<br />
Fluxes Using the Integrated Nitrogen CAtchment<br />
(INCA) Model<br />
Katri Rankinen a , Heikki Lehtonen b , Kirsti Granlund a <strong>and</strong> Ilona Bärlund a<br />
a Finnish Environment Institute, P.O.Box 140, FIN-00251 Helsinki, Finl<strong>and</strong><br />
e-mail: katri.rankinen@ymparisto.fi<br />
b MTT Economic Research, Agrifood Research Finl<strong>and</strong>, Luutnantintie 13, FIN-00410 Helsinki, Finl<strong>and</strong><br />
Abstract: The INCA (Integrated Nitrogen CAtchment) model is a semi-distributed, dynamic nitrogen model<br />
which simulates nitrogen fluxes in catchments. Sources of nitrogen can be atmospheric deposition, the terrestrial<br />
environment or direct discharges. The model can simulate nitrogen processes in six l<strong>and</strong> use classes.<br />
There are three components included; the hydrological model, the catchment nitrogen process model <strong>and</strong> the<br />
river nitrogen process model. The aim of this study is to compare the effects of three different agricultural<br />
policy scenarios on inorganic nitrogen flux to the sea from Finnish catchments. Target years of these scenarios<br />
are 2010 <strong>and</strong> 2020. The changes in agricultural production in different scenarios of agricultural policy are<br />
evaluated using the DREMFIA model (Dynamic Regional Model of Finnish Agriculture). DREMFIA is a<br />
dynamic dis-equilibrium model based on an evolutionary scheme of technology diffusion which considers<br />
farm investments, evolving farm size structure <strong>and</strong> technological change explicitly. In the first phase of the<br />
study the INCA model is applied to the Simojoki river basin in northern Finl<strong>and</strong>, where main anthropogenic<br />
influences are agriculture, atmospheric deposition <strong>and</strong> forestry. At the Simojoki river basin agriculture is<br />
mainly animal husb<strong>and</strong>ry <strong>and</strong> grass cultivation. The river Simojoki discharges to the Bothnian Bay. The<br />
predicted changes in agricultural production <strong>and</strong> l<strong>and</strong> use at Simojoki river basin prove to have more effect<br />
on inorganic nitrogen flux to the sea than changes in forestry practices or atmospheric deposition. This result<br />
stems from the specific location, ecosystem type <strong>and</strong> characteristics of farm l<strong>and</strong> in Simojoki basin. Next the<br />
INCA model will be applied to a river basin in southern Finl<strong>and</strong>, where the main l<strong>and</strong> use form is agriculture.<br />
Keywords: Agricultural policy scenarios; Agricultural sector modelling; Semi-distributed modelling; N<br />
leaching; Northern river basin<br />
1. INTRODUCTION<br />
Eutrophication of surface waters due to increased<br />
nutrient loading during the last decades is one of<br />
the main environmental concerns in Finl<strong>and</strong>. Agriculture<br />
comprises the largest single source of nutrients<br />
to surface waters. Municipal <strong>and</strong> industrial<br />
waste water purification has effectively decreased<br />
nutrient load from point sources leading to improved<br />
water quality, but no clear effects of decreasing<br />
non-point loading are found [Räike et al.<br />
2003].<br />
There is agriculture all over the Finl<strong>and</strong> despite of<br />
the northern location of the country, though main<br />
production areas are in southern <strong>and</strong> western parts<br />
of the country. Finnish agricultural products come<br />
mainly from family farms. The location of the<br />
different production lines <strong>and</strong> use of arable l<strong>and</strong><br />
are dictated by the climatic conditions. Most of the<br />
crop production is in the south whereas cattle<br />
breeding is concentrated in central, eastern <strong>and</strong><br />
northern parts.<br />
Since 1995, the Finnish agricultural support measures<br />
have been based on the Common Agricultural<br />
Policy (CAP) of the EU. There are three kind of<br />
agricultural supports in Finl<strong>and</strong>. CAP supports for<br />
arable crops <strong>and</strong> animals are closely linked to the<br />
market arrangements of the CAP, <strong>and</strong> these are<br />
1057
Lake Simojärvi<br />
.<br />
4<br />
Water quality monitoring station<br />
Discharge gauging station<br />
Simojoki outlet 0 20 40 km<br />
#<br />
Helsinki<br />
Figure 1. Location of the Simojoki river basin in northern Finl<strong>and</strong><br />
financed in full from the EU budget. In 2003 these<br />
accounted for 26% of all agricultural support. The<br />
share of support for rural development co-financed<br />
by Finl<strong>and</strong> <strong>and</strong> the EU was 41%. National aid<br />
accounted for 34%.<br />
Typically different mathematical models <strong>and</strong> decision<br />
support systems are used for evaluating policy<br />
measures like EU's Agenda 2000 reform [e.g. Wier<br />
et al. 2002, Pacini et al. 2004]. Forsman et al.<br />
[2003] described a generic framework in which<br />
economical models can be linked with N transport<br />
<strong>and</strong> transformation models at different scales.<br />
Their principles are followed in this study when<br />
linking an agricultural policy model to a catchment<br />
scale N model. A physical plot scale model would<br />
give more detailed information of N processes but<br />
is more problematic to be applied on a catchment<br />
scale. Also, Quinn [2004] argued that a complex<br />
physical model is not needed when studying nitrate<br />
pollution problem at the catchment scale.<br />
The aim of this study is to compare the effects of<br />
agricultural policy scenarios on inorganic nitrogen<br />
(N) flux to the sea from a Finnish catchment. The<br />
changes in agricultural production in different<br />
scenarios of agricultural policy are evaluated using<br />
the DREMFIA model (Dynamic Regional Model<br />
of Finnish Agriculture) [Lehtonen 2001]. The<br />
effects of agricultural production scenarios on<br />
inorganic N processes are simulated by the dynamic,<br />
semi-distributed INCA model (Integrated<br />
Nitrogen in CAtchments) [Wade et al. 2002,<br />
Whitehead et al. 1998]. The Finnish Agri-<br />
<strong>Environmental</strong> Progamme (FAEP) interview research<br />
of the farmers is used to derive typical cultivation<br />
practices [Palva et al. 2001].<br />
In this first phase of the study these models are<br />
applied to the Simojoki river basin in northern<br />
Finl<strong>and</strong>. Agricultural fields cover less than 3% of<br />
the total area, <strong>and</strong> agricultural production is mainly<br />
grass cultivation for animal production. Influences<br />
of changes in agricultural production on N leaching<br />
were compared with influences of changes in<br />
atmospheric deposition <strong>and</strong> forestry practices<br />
which were previously evaluated by Rankinen et<br />
al. [2004]. In the next phase of the study the INCA<br />
model will be applied to a river basin in southern<br />
Finl<strong>and</strong>, where the main l<strong>and</strong> use form is agriculture<br />
<strong>and</strong> production lines are more variable.<br />
2. MATERIALS AND METHODS<br />
2.1 The Simojoki river basin<br />
The Simojoki river basin (3160 km 2 ) can be subdivided<br />
into nine sub-basins (Fig. 1.). Over the period<br />
1961-1975, annual precipitation varied between<br />
650-750 mm <strong>and</strong> annual runoff between<br />
350-450. The mean annual temperature is +0.5 -<br />
+1.5 o C. The duration of the snow cover is from<br />
the middle of November to early May. According<br />
to the Finnish Meteorological Institute growing<br />
season started on an average on 10 th May in years<br />
1961-1990. Length of the growing season was on<br />
average 140 days.<br />
The river Simojoki is a salmon river in nearnatural<br />
state, <strong>and</strong> the dominant human impacts in<br />
the area are forestry, agriculture <strong>and</strong> atmospheric<br />
deposition. An average 0.5% of the total catchment<br />
area is felled annually. In 1995 there were 1365 ha<br />
of peat mining area (0.43% of the catchment area).<br />
Urban areas cover only 0.06% <strong>and</strong> agricultural<br />
1058
fields 2.7% of the catchment area [Perkkiö et al.<br />
1995]. Grass cultivation for animal husb<strong>and</strong>ry is<br />
the most common form of agricultural production.<br />
2.2 The DREMFIA model<br />
The changes in agricultural production are evaluated<br />
by the Dynamic Regional Model of Finnish<br />
Agriculture [Lehtonen 2001, 2004]. DREMFIA is<br />
a dynamic dis-equilibrium model based on an<br />
evolutionary scheme of technology diffusion<br />
which considers farm investments explicitly. The<br />
model can be used to evaluate the effects of different<br />
agricultural policies both on production <strong>and</strong> on<br />
agricultural income. Agriculture in the Simojoki<br />
river basin is modelled as one region in DREMFIA<br />
model which covers 17 other agricultural regions<br />
in Finl<strong>and</strong>.<br />
The core of the model is an optimisation block<br />
which maximises the producer <strong>and</strong> consumer surplus.<br />
Endogenous investments determine animal<br />
<strong>and</strong> crop production volume in the long-term, but<br />
short-term changes in crop production are constrained<br />
by flexibility constraints. The constraints<br />
are validated on the basis of average crop production<br />
data from 1990-2002. Changing agricultural<br />
policy <strong>and</strong> consumption trends are given exogenously.<br />
All foreign trade flows are assumed to <strong>and</strong><br />
from the EU (Armington assumption is used).<br />
Fertilization <strong>and</strong> yield levels are dependent on crop<br />
<strong>and</strong> fertilizer prices. Feeding of animals may<br />
change due to production <strong>and</strong> animal nutrition<br />
requirements. The average milk yield depends on<br />
feedstuffs used in feeding. Thus, the price of milk<br />
<strong>and</strong> feedstuffs affect the nitrogen fertilization level<br />
<strong>and</strong> milk yield of dairy cows. The model is validated<br />
to observed production levels in 1995-2002.<br />
In this study the scenarios of agricultural policy are<br />
based on Agenda 2000 agreement of CAP. Base<br />
(BASE) scenario follows Agenda 2000 reform<br />
which is assumed to stay unchanged up to 2020.<br />
National support, investment support <strong>and</strong> environmental<br />
support are assumed to stay at the level<br />
in which they were in year 2002. Prices of dairy<br />
products as well as the producer price of milk is<br />
assumed to fall (
Table 1. Used max. N fertilization either for mineral<br />
fertilizers or for organic fertilizers<br />
1. fertilization 2. fertilization<br />
Mineral<br />
NH 4 -N 70 kg ha -1 30 kg ha -1<br />
NO 3 -N 30 kg ha -1 24 kg ha -1<br />
Organic<br />
Soluble N 105 kg ha -1 60 kg ha -1<br />
Fertilization is given to the INCA model as time<br />
series in which half of the fertilization is assumed<br />
to be organic <strong>and</strong> half mineral. Mineral fertilizers<br />
are dissolved at rate 0.15 day -1 . Soluble N of organic<br />
fertilizers is assumed to be NH 4 -N, which<br />
rapidly nitrifies in soil. Both NH 4 -N <strong>and</strong> NO 3 -N<br />
pools are dissolved at rate 0.15 day -1 . Long-term<br />
effect of organic fertilizers on soil fertility is taken<br />
into account by allowing a high mineralization rate<br />
from the soil organic N pool. N process parameters<br />
for l<strong>and</strong> use class Agricultural fields are presented<br />
in Table 2.<br />
Table 2. Parameter values for N processes at the<br />
l<strong>and</strong> use class Agricultural fields<br />
Process Value Unit<br />
Mineralisation 1.2 kg ha -1 day -1<br />
Nitrification 0.2 day -1<br />
Denitrification 0.015 day -1<br />
Growth season start day 136 -<br />
Growth period 120 days<br />
Nitrate uptake rate 0.25 day -1<br />
Ammonium uptake rate 0.2 day -1<br />
Maximum N uptake 250 kg N yr -1 ha -1<br />
3. RESULTS AND DISCUSSION<br />
3.1 Assessed changes in agricultural practises<br />
The output of the DREMFIA model includes yearto-year<br />
variation in the total area of agricultural<br />
l<strong>and</strong> as well as in the area of main crops <strong>and</strong> the<br />
fertilization levels. The total area of agricultural<br />
l<strong>and</strong> is limited by the area of suitable soil types for<br />
agriculture.<br />
Grass cultivation stays as the main production<br />
form at the Simojoki river basin (Fig. 2). In BASEscenario<br />
the total area of agricultural l<strong>and</strong> increases<br />
by 25% by the year 2010 <strong>and</strong> then starts<br />
slowly to decrease. In MTR-scenario the total area<br />
of agricultural l<strong>and</strong> increases by 35% by the year<br />
2010 <strong>and</strong> then levels off. In reality the utilised<br />
area [1000 ha]<br />
area [1000 ha]<br />
agricultural area has increased by 11% in northern<br />
Finl<strong>and</strong> in 1996-2003.<br />
After the year 2010 the milk production volume<br />
<strong>and</strong> the area of grass cultivation starts to decrease<br />
<strong>and</strong> the area of green fallow to increase. This is<br />
because decreasing milk price <strong>and</strong> de-coupled<br />
payments decrease dairy investments. Fertilization<br />
levels including nitrogen from manure are presented<br />
in Table 3.<br />
8<br />
6<br />
4<br />
2<br />
0<br />
8<br />
6<br />
4<br />
2<br />
0<br />
1995<br />
1995<br />
1997<br />
1997<br />
1999<br />
1999<br />
2001<br />
2001<br />
2003<br />
2003<br />
2005<br />
2005<br />
Figure 2. Assessed changes in agricultural l<strong>and</strong><br />
use<br />
Table 3. Simulated fertilization levels for grass<br />
BASE<br />
BASE<br />
All Cereals Grass Fallow<br />
MTR<br />
1995 160 kg N ha -1 160 kg N ha -1<br />
2010 151 kg N ha -1 143 kg N ha -1<br />
2020 153 kg N ha -1 151 kg N ha -1<br />
3.2 Effect of agricultural policy on N leaching<br />
2007<br />
MTR<br />
2007<br />
In the INCA model simulations meteorological<br />
data from years 1994-1996 is used. Observed <strong>and</strong><br />
simulated discharge <strong>and</strong> inorganic N concentrations<br />
at the river Simojoki outlet in years 1994-<br />
1996 are presented in Figure 3. These years are<br />
used as a base-line for scenarios when evaluating<br />
inorganic N leaching.<br />
Total area of cultivated l<strong>and</strong>, main crop types <strong>and</strong><br />
fertilization levels simulated by the DREMFIA<br />
model are used as input to the INCA model.<br />
Changes in parameterised l<strong>and</strong> use types happen<br />
2009<br />
2009<br />
2011<br />
2011<br />
2013<br />
2013<br />
2015<br />
2015<br />
2017<br />
All Cereals Grass Fallow<br />
2017<br />
2019<br />
2019<br />
1060
Q [m 3 s -1 ]<br />
600<br />
400<br />
Observed<br />
Simulated<br />
200<br />
0<br />
1.1.1994 1.7.1994 1.1.1995 1.7.1995 1.1.1996 1.7.1996<br />
date<br />
NO 3 -N [mg l -1 ]<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
1.1.1994 20.7.1994 5.2.1995 24.8.1995 11.3.1996 27.9.1996<br />
date<br />
NH 4 -N [mg l -1 ]<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
Simulated<br />
Observed<br />
0<br />
1.1.1994 20.7.1994 5.2.1995 24.8.1995 11.3.1996 27.9.1996<br />
date<br />
Figure 3. Observed <strong>and</strong> simulated discharge <strong>and</strong> inorganic N concentrations at the outlet of the Simojoki<br />
river<br />
between agricultural fields <strong>and</strong> forest on mineral<br />
soil when area of agricultural fields increases.<br />
Green fallow is simulated as unfertilized set-aside<br />
l<strong>and</strong> (forest cut on mineral soil).<br />
In BASE scenario the total area of agricultural l<strong>and</strong><br />
increases by 9% by the year 2020 from 1995 <strong>and</strong><br />
grass stays as main crop but that change is not<br />
enough to increase total inorganic N flux to the<br />
sea. In MTR scenario the total area of utilised<br />
agricultural fields increases by 35%, but the share<br />
of green fallow increases at the expense of grass<br />
cultivation which leads decreasing inorganic N<br />
flux to the sea by 2.5%. Inorganic N fluxes to the<br />
sea are highest in the year 2010 when the total area<br />
of both agricultural l<strong>and</strong> <strong>and</strong> grass cultivation is<br />
largest.<br />
The scenarios of forestry <strong>and</strong> atmospheric deposition<br />
[Rankinen et al. 2004] were compared to the<br />
scenarios of agriculture. Forest cut areas are assumed<br />
to increase by 20% (Cut+20 scenario) <strong>and</strong><br />
deposition level is assumed to decrease from 2.3<br />
kg N ha -1 yr -1 to 2 kg N ha -1 yr -1 . In Cut-<br />
100_NoPeat scenario no forest cut areas <strong>and</strong> no<br />
peat mining is assumed. In Figure 4. the effects of<br />
different scenarios on N flux to the sea in the year<br />
2010 were compared.<br />
Even though the total area of agricultural l<strong>and</strong> at<br />
the Simojoki river basin is only a couple of percents,<br />
changes in it has more pronounced effect on<br />
inorganic N flux than changes in forestry practices<br />
or atmospheric N deposition. Inorganic N load is<br />
typically about ten times higher from cultivated<br />
fields than from forests. Expected changes in atmospheric<br />
N deposition are very low <strong>and</strong> forest cut<br />
areas are assumed to represent situation several<br />
years after treatment when disturbance is abated.<br />
Direct <strong>and</strong> local effects of forest cut may be more<br />
extensive. Forest cut areas are located mainly in<br />
uppermost areas of the river basin, so inorganic N<br />
may be denitrified in river water before it reaches<br />
the outlet of the river. Agricultural fields on the<br />
other h<strong>and</strong> are mainly located on river deposits<br />
along the river near outlet.<br />
Agriculture MTR<br />
Agriculture<br />
BASE<br />
Cut+20%<br />
Deposition<br />
Cut-100% <strong>and</strong><br />
no peatmining<br />
-10.0 -5.0 0.0 5.0 % 10.0<br />
Figure 4. Changes in inorganic N flux to the<br />
Bothnian Bay according to different scenarios in<br />
2010<br />
1061
4. CONCLUSIONS<br />
This study shows that changes in agricultural production<br />
volume <strong>and</strong> production practises derived<br />
from the DREMFIA model can be combined with<br />
the INCA model to evaluate inorganic N flux from<br />
terrestrial areas to surface waters. With the catchment<br />
scale INCA model the effects of changes in<br />
agricultural production can be compared to the<br />
effects of other changes in the river basin. When<br />
including evaluation of environmental goals <strong>and</strong><br />
social <strong>and</strong> economic impacts this approach can be<br />
exp<strong>and</strong>ed to fulfil the whole generic structure of a<br />
decision support system described by Forsman et<br />
al. [2003].<br />
Assessed changes in agricultural l<strong>and</strong> use in the<br />
Simojoki river basin can alter the inorganic N flux<br />
to the sea up to 5%. Agricultural activities at this<br />
northern river basin, which is considered less favourable<br />
production area, are clearly policy driven.<br />
Any significant reduction in milk price <strong>and</strong> decoupling<br />
of agricultural support from production is<br />
likely to decrease the intensity <strong>and</strong> scale of production.<br />
These are contradictory to the results of Winter<br />
<strong>and</strong> Gaskell [1998] <strong>and</strong> Weir et al. [2002] who<br />
did not find any significant environmental effect of<br />
the Agenda 2000 CAP reform in Great Britain <strong>and</strong><br />
Denmark.<br />
The next step is to continue this study by applying<br />
the models to a catchment in southern Finl<strong>and</strong><br />
where the main l<strong>and</strong> use form is agriculture <strong>and</strong><br />
the agricultural production lines are more variable.<br />
5. ACKNOWLEDGEMENTS<br />
This study was supported by the Commission of<br />
the European Union, the INCA project (EVK1-<br />
CT-1999-00011). The financial support of the<br />
SUSAGFU project (contract 76724) <strong>and</strong> NU-<br />
TRIBA project (contract 202421) through the<br />
Academy of Finl<strong>and</strong> is gratefully acknowledged.<br />
6. REFERENCES<br />
CEC, http://europa.eu.int/comm/<br />
agriculture/ mtr/index_en.htm, 2003.<br />
Forsman, Å., A. Grimvall, J. Scholtes <strong>and</strong> H.B.<br />
Wittgren, Generic structures of decision support<br />
systems for evaluation of policy measures to reduce<br />
catchment-scale nitrogen fluxes, Physics<br />
<strong>and</strong> Chemistry of the Earth, 28, 589-598, 2003<br />
Lehtonen, H., Principles, structure <strong>and</strong> application<br />
of dynamic regional sector model of Finnish agriculture.<br />
PhD thesis, Systems Analysis Laboratory,<br />
Helsinki University of Technology. Agrifood<br />
Research Finl<strong>and</strong>, Economic Research<br />
(MTTL) Publ. 98. Helsinki, 2001.<br />
Lehtonen, H., Impacts of de-coupling agricultural<br />
support on dairy investments <strong>and</strong> milk production<br />
volume in Finl<strong>and</strong>. Forthcoming in Acta<br />
Agriculturae Sc<strong>and</strong>inavica, Section C: Food<br />
Economics, 2004.<br />
Pacini, C., G. Giesen, A. Wossink, L. Omodei-<br />
Zorini <strong>and</strong> R. Huirne, The EU's Agenda 2000<br />
reform <strong>and</strong> the sustainability of organic farming<br />
in Tuscany: ecological-economic modelling at<br />
field <strong>and</strong> farm level, Agricultural Systems 80,<br />
171-197, 2004.<br />
Palva, R., K. Rankinen, K. Granlund, J. Grönroos,<br />
A. Nik<strong>and</strong>er <strong>and</strong> S. Rekolainen, Maatalouden<br />
ympäristötuen toimenpiteiden toteutuminen ja<br />
vaikutukset vesistökuormitukseen vuosina 1995-<br />
1999. MYTVAS-projektin loppuraportti.<br />
Suomen ympäristö. Ympäristönsuojelu. pp 92.<br />
2001<br />
Perkkiö, S., E. Huttula <strong>and</strong> M. Nenonen, Simojoen<br />
vesistön vesiensuojelusuunnitelma. Vesi- ja<br />
ympäristöhallinnon julkaisuja-sarja A 200, 1995.<br />
Quinn, P. Scale appropriate modelling: representing<br />
cause-<strong>and</strong>-effect relationships in nitrate pollution<br />
at the catchment scale for the purpose of<br />
catchment scale planning, Journal of Hydrology<br />
291, 197-217, 2004.<br />
Rankinen, K., K. Granlund <strong>and</strong> A. Lepistö, Integrated<br />
nitrogen <strong>and</strong> flow modelling (INCA) in a<br />
boreal river basin dominated by forestry: scenarios<br />
of environmental change. Water, Air <strong>and</strong> Soil<br />
Pollution Focus, (Accepted), 2004.<br />
Räike, A., O.-P. Pietiläinen, S. Rekolainen, P.<br />
Kauppila, H. Pitkänen, J. Niemi, A. Raatel<strong>and</strong><br />
<strong>and</strong> J. Vuorenmaa, Trends of phosphorus, nitrogen<br />
<strong>and</strong> clorophyll a concentrations in Finnish<br />
rivers <strong>and</strong> lakes in 1975-2000. the Science of the<br />
Total Environment 310, 47-59, 2003.<br />
Wade, A., P. Dur<strong>and</strong>, V. Beaujoan, W. Wessels,<br />
K. Raat, P.G. Whitehead, D. Butterfield, K.<br />
Rankinen <strong>and</strong> A. Lepistö, Towards a generic nitrogen<br />
model of European ecosystems: New<br />
model structure <strong>and</strong> equations. Hydrology <strong>and</strong><br />
Earth System Sciences 6, 559-582, 2002.<br />
Whitehead, P.G., E.J. Wilson <strong>and</strong> D. Butterfield, A<br />
semi-distributed Integrated Nitrogen model for<br />
multiple source assessment in Catchments<br />
(INCA): Part I-model structure <strong>and</strong> process<br />
equations. the Science of the Total Environment<br />
210/211, 547-558, 1998.<br />
Wier, M., J.M. Andersen, J. Jensen <strong>and</strong> T.C. Jensen,<br />
The EU's Agenda 2000 reform for the agricultural<br />
sector: environmental <strong>and</strong> economic effects<br />
in Denmark, Ecologicl Economics 41, 345-<br />
359, 2002.<br />
Winter, M. <strong>and</strong> P. Gaskell, The Agenda 2000 debate<br />
<strong>and</strong> CAP reform in Great Britain. Is the en-<br />
1062
vironment being sidelined? L<strong>and</strong> Use Policy 15, 217-231, 1998.<br />
1063
Implications of Complexity <strong>and</strong> Uncertainty for<br />
Integrated <strong>Modelling</strong> <strong>and</strong> Impact Assessment in River<br />
Basins<br />
Valentina Krysanova, Fred Hattermann, <strong>and</strong> Frank Wechsung<br />
krysanova@pik-potsdam.de<br />
Abstract: The paper focuses on implications of uncertainty in climate change impact assessment at the<br />
river basin <strong>and</strong> regional scales. The study was performed using the process-based ecohydrological spatially<br />
distributed model SWIM (Soil <strong>and</strong> Water Integrated Model). The model integrates hydrological processes,<br />
vegetation/crop growth, erosion <strong>and</strong> nutrient dynamics in river basins. It was developed from the SWAT <strong>and</strong><br />
MATSALU models for climate <strong>and</strong> l<strong>and</strong> use change impact assessment. The study area is the German part of<br />
the Elbe River basin (about 100.000 km 2 ). It is representative for semi-humid l<strong>and</strong>scapes in Europe, where<br />
water availability during the summer season is the limiting factor for plant growth <strong>and</strong> crop yield. The<br />
validation method followed the multi-scale, multi-site <strong>and</strong> multi-criteria approach <strong>and</strong> enabled to reproduce<br />
(a) water discharge <strong>and</strong> nutrient load at the river outlet along with (b) local ecohydrological processes like<br />
water table dynamics in subbasins, nutrient fluxes <strong>and</strong> vegetation growth dynamics at multiple scales <strong>and</strong><br />
sites. The uncertainty of climate impacts was evaluated using comprehensive Monte Carlo simulation<br />
experiments.<br />
Keywords: integrated modelling, ecohydrological model, climate impact, uncertainty, Elbe River.<br />
1. THE ELBE RIVER BASIN AS A CASE<br />
STUDY<br />
The case study area provides an example of a river<br />
basin, where the current regional trend in<br />
precipitation differs from the global trend resulting<br />
from GCMs (General Circulation Models). The<br />
study area is the German part of the Elbe River<br />
basin (about 100.000 km 2 ). The long-term mean<br />
annual precipitation in the study area is 659 mm.<br />
The long-term mean discharge of the Elbe River is<br />
716 m 3 s -1 at the mouth, <strong>and</strong> the specific discharge<br />
is 6.2 l s -1 km -2 , which corresponds to the mean<br />
annual runoff of 10.06 x 10 9 m 3 (29.7 % of the<br />
annual precipitation).<br />
A primary reason for selecting this river basin as<br />
case study region is its vulnerability against water<br />
stress in dry periods. The basin is located around<br />
the boundary between the relatively wet maritime<br />
climate in western Europe <strong>and</strong> the more continental<br />
climate in eastern Europe with longer dry periods,<br />
<strong>and</strong> the annual long-term average precipitation in<br />
the area is relatively small. Therefore the Elbe<br />
River basin is classified as the driest among the<br />
five largest river basins in Germany (Rhine,<br />
Danube, Elbe, Weser <strong>and</strong> Ems) with all resulting<br />
problems <strong>and</strong> conflicts. The region is<br />
representative of semi-humid l<strong>and</strong>scapes in<br />
Europe, where water availability during the<br />
summer season is the limiting factor for plant<br />
growth <strong>and</strong> crop yield. The basin is densely<br />
populated, <strong>and</strong> has the second lowest water<br />
availability per capita within Europe. Due to<br />
expected change in circulation pattern <strong>and</strong> local<br />
orographical conditions the amount of precipitation<br />
will most likely decrease in the Elbe drainage basin<br />
(Werner & Gerstengarbe, 1997).<br />
2. MODEL SWIM<br />
The process-based ecohydrological model SWIM<br />
(Soil <strong>and</strong> Water Integrated Model) (Krysanova et<br />
al., 1998 & 2000) was used in the study. SWIM is<br />
a continuous-time spatially distributed model,<br />
integrating hydrological processes, vegetation<br />
growth (agricultural crops <strong>and</strong> natural vegetation),<br />
1064
nutrient cycling (nitrogen, N <strong>and</strong> phosphorus, P),<br />
<strong>and</strong> sediment transport at the river basin scale. The<br />
modelling system includes an interface to the<br />
Geographic Information System GRASS<br />
(Geographic Resources Analysis Support System)<br />
(GRASS4.1, 1993). The spatial disaggregation<br />
scheme has three levels: basin – subbasins –<br />
hydrotopes. The subbasin map can be produced by<br />
using the r.watershed operation in GRASS or input<br />
from other sources, <strong>and</strong> the hydrotope map is<br />
usually produced by overlaying the subbasin, l<strong>and</strong><br />
use <strong>and</strong> soil maps. The SWIM/GRASS interface<br />
allows to extract spatially distributed parameters of<br />
elevation, l<strong>and</strong> use, soil <strong>and</strong> vegetation, <strong>and</strong> to<br />
derive the hydrotope structure <strong>and</strong> the routing<br />
structure for the basin under study.<br />
3. MODEL VALIDATION APPROACH<br />
The need for powerful validation techniques for<br />
distributed hydrological <strong>and</strong> ecohydrological<br />
models has often been pointed out. While the<br />
primary idea of distributed hydrological modelling<br />
is to reproduce water fluxes in subbasins <strong>and</strong><br />
hydrotopes along with river discharge, the models<br />
are often validated using only observed river<br />
discharge at the basin outlet, <strong>and</strong> multi-scale<br />
validation is rather exceptional. This is especially<br />
true for macro-scale basins. The river discharge is<br />
an integral attribute of hydrological processes in<br />
the river basin, but its correct representation by the<br />
model does not guarantee adequacy in spatial <strong>and</strong><br />
temporal dynamics of all water components in the<br />
basin.<br />
Ideally, the validation has to be multi-scale, multisite<br />
<strong>and</strong> multi-criteria <strong>and</strong> based on sensitivity <strong>and</strong><br />
uncertainty analyses performed in advance, if the<br />
model has to be further applied at the regional<br />
scale <strong>and</strong>/or for climate or l<strong>and</strong> use change impact<br />
assessment. The multi-scale <strong>and</strong> multi-site<br />
validation should include several basins of<br />
different size <strong>and</strong> located in different<br />
regions/subregions with various topographical<br />
conditions, l<strong>and</strong> use composition <strong>and</strong> soils. At least<br />
some of the basins should be nested, in order to<br />
allow a special test, whether the model or some of<br />
its parameters or variables are scale-dependent.<br />
The multi-criteria validation should include<br />
different statistical criteria of fit <strong>and</strong> spatially<br />
distributed hydrological characteristics (like soil<br />
moisture, groundwater table, snow distribution)<br />
beside commonly used Nash <strong>and</strong> Sutcliffe<br />
efficiency <strong>and</strong> water discharge at the basin outlet.<br />
Besides, an ecohydrological model must be<br />
validated also for vegetation dynamics, crop yield,<br />
nutrient fluxes in soil, nutrient load, <strong>and</strong> sediment<br />
yield.<br />
This method of validation has been successfully<br />
applied to the model SWIM used in this study. The<br />
model was extensively validated in more than 20<br />
subbasins (partly nested) of the Elbe River basin<br />
(Krysanova et al., 1998, Hattermann et al., 2004)<br />
using the multi-scale, multi-criteria <strong>and</strong> multi-site<br />
validation method. It has been proven that SWIM<br />
is able to reproduce satisfactory the observed river<br />
discharge, spatio-temporal groundwater table<br />
dynamics, nitrogen <strong>and</strong> phosphorus cycling in<br />
soils, nutrient loads at the basin scale, <strong>and</strong> regional<br />
crop yields. The method <strong>and</strong> its results are<br />
presented in Hattermann et al., 2004. The<br />
comprehensive model validation increases the<br />
reliability of the model, <strong>and</strong> creates a sound basis<br />
for subsequent climate impact assessment.<br />
4. METHODS OF CLIMATE<br />
DOWNSCALING<br />
Nowadays General Circulation Models (GCMs)<br />
are used for better underst<strong>and</strong>ing of the<br />
development of the earth climate system <strong>and</strong><br />
prediction of future climate change. However their<br />
current resolution is too rough for correct<br />
representation of hydrological cycle variations<br />
within river catchments.<br />
The 10 km resolution of climate model is a critical<br />
threshold, since at this scale the climate model<br />
outputs become comparable with the scale of<br />
variation within river catchments, <strong>and</strong> climate<br />
variables (like air temperature, precipitation) could<br />
be predicted directly without the need for<br />
downscaling. It is also important that at this scale<br />
vegetation, soil <strong>and</strong> geology can be represented<br />
explicitly without the need for upscaling.<br />
Therefore, only at this scale the climate <strong>and</strong><br />
hydrological models could be directly linked, <strong>and</strong><br />
the major source of uncertainty in climate impacts<br />
assessment would be removed. However, this<br />
resolution is not yet achieved in current GCMs.<br />
The problem can be partly solved by applying<br />
downscaling methods to transform the GCM<br />
outputs into climate input parameters at the<br />
regional <strong>and</strong> river basin scale. Two main types of<br />
downscaling methods are in use: the deterministic<br />
dynamical downscaling method <strong>and</strong> the statistical<br />
downscaling method.<br />
The deterministic models have basically the same<br />
mathematical framework as the global climate<br />
models, but a finer grid resolution. The<br />
deterministic downscaling models are applied by<br />
nesting their grid structure into the grid structure of<br />
GCMs (the outputs of GCMs are taken as<br />
boundary conditions to calculate climate input data<br />
1065
for regional applications). They are physically<br />
based <strong>and</strong> can be solved numerically. The<br />
disadvantage of the deterministic downscaling<br />
models is their large data <strong>and</strong> computation<br />
dem<strong>and</strong>. In addition, the physics of the atmosphere<br />
is mathematically extremely complex in such<br />
models, so that this type of models is still under<br />
development.<br />
The second type of downscaling methods makes<br />
use of the correlation between the large-scale<br />
climate patterns (where the results of GCMs are<br />
relatively reliable) <strong>and</strong> their regional<br />
representation, considering consistency in<br />
frequency distribution, annual <strong>and</strong> inter-annual<br />
variability <strong>and</strong> persistency of the main climate<br />
characteristics. The advantage of these methods is<br />
relative robustness of their results as long as the<br />
basic climate correlations in the observed <strong>and</strong><br />
scenario periods do not differ.<br />
Both methods take the results of GCMs as<br />
boundary <strong>and</strong> initial conditions, <strong>and</strong> therefore the<br />
inherent uncertainty in the GCM outputs is<br />
transferred to the regional scale as well.<br />
5. CLIMATE CHANGE SCENARIO<br />
The applied climate scenario was produced in PIK<br />
(F.-W. Gerstengarbe & P.C. Werner) by the<br />
statistical downscaling method described in<br />
Werner & Gerstengarbe (1997) from the<br />
ECHAM4-OPYC3 GCM, which was driven by the<br />
IPCC emission scenario A1. The climate change<br />
scenario is characterized by an increase in<br />
temperature by 1.4°C until 2050 (Fig. 1), <strong>and</strong> a<br />
moderate decrease in mean annual precipitation<br />
(on average -17% in the basin) corresponding to<br />
the observed regional climate trend.<br />
Mean annual temperature<br />
13<br />
12<br />
11<br />
10<br />
9<br />
8<br />
7<br />
6<br />
5<br />
4<br />
1950 1970 1990 2010 2030 2050<br />
Fig. 1. The mean annual observed temperature<br />
(black line: 1950 - 1995), the mean annual<br />
temperature for six scenario realizations (1996 –<br />
2055) for the Elbe basin. The linear trend is shown<br />
as a thick grey line<br />
The applied statistical downscaling method<br />
maintains the stability of the main statistical<br />
characteristics (variability, frequency distribution,<br />
annual cycle, persistence). Climate scenario is<br />
developed using a special cluster analysis<br />
algorithm, which guarantees temporal, spatial <strong>and</strong><br />
physical consistency of the considered<br />
meteorological parameters. First, the series of the<br />
reference variable (temperature) are constructed in<br />
several steps. Once the daily mean values of a<br />
long-term observed time series are obtained, it is<br />
possible to impose the assumed trend onto the<br />
series <strong>and</strong> to create the simulated series. Then the<br />
other meteorological variables are related to the<br />
reference one.<br />
In addition, a conditioned Monte Carlo simulation<br />
was implemented in the downscaling procedure, so<br />
that 100 realizations of the scenario were produced<br />
to investigate the uncertainty of the method. Fig. 1<br />
demonstrates the dynamics of mean annual<br />
temperature for the whole Elbe basin in the<br />
reference <strong>and</strong> scenario periods (for 6 selected<br />
scenario realizations), <strong>and</strong> the corresponding trend.<br />
6. CLIMATE IMPACT ASSESSMENT<br />
WITH UNCERTAINTY<br />
The main objective of the study was to investigate<br />
the vulnerability of water resources <strong>and</strong> agriculture<br />
in the Elbe basin against expected climate change.<br />
The crop spectrum was restricted to three major<br />
crops in the region: winter wheat, winter barley <strong>and</strong><br />
silage maize. The adjustment of net photosynthesis<br />
<strong>and</strong> evapotranspiration to altered atmospheric CO 2<br />
concentration was studied considering two<br />
additional factors (see full description in<br />
Krysanova et al, 1999):<br />
• adjustment of the potential growth rate per<br />
unit of intercepted PAR by a temperature<br />
dependent correction factor alpha based on<br />
experimental data for C3 <strong>and</strong> C4 crops; <strong>and</strong><br />
• assuming a CO 2 influence on transpiration at<br />
the regional scale (factor beta), which is<br />
coupled to the direct CO 2 effect of radiation<br />
use efficiency (factor alpha).<br />
Simulation runs have been carried out in three<br />
variants: (1) only climate change without CO 2<br />
adjustment, (2) with adjustment of net<br />
photosynthesis, <strong>and</strong> (3) with adjustment of net<br />
photosynthesis <strong>and</strong> transpiration. In this way we<br />
accounted for current uncertainty regarding<br />
significance of stomatal effects on higher CO 2 for<br />
1066
egional evapotranspiration. Here only results<br />
related to variant (1) are discussed.<br />
In order to evaluate the direct climate-induced<br />
uncertainty of impact assessment, the climate<br />
scenario was used along with its 100 realizations.<br />
In other words, the modelling with SWIM was<br />
used to transform the uncertainties in climate input<br />
represented by 100 realizations into ecohydrological<br />
responses like evapotranspiration,<br />
surface <strong>and</strong> subsurface runoff, river discharge,<br />
groundwater recharge, <strong>and</strong> crop yield. The model<br />
results were subsequently analyzed considering<br />
seasonal dynamics, trends, histograms for the set of<br />
100 simulations, <strong>and</strong> spatial patterns in different<br />
sub-regions.<br />
According to the simulation results, actual<br />
evapotranspiration is expected to decrease on<br />
average by 4%, with significant subregional<br />
differences. Namely, a moderate increase up to 103<br />
mm y -1 is expected in north-western part of the<br />
basin, <strong>and</strong> a decrease up to 126 mm y -1 is simulated<br />
for the loess subregion located in Saxony-Anhalt<br />
(the central part of basin). Runoff <strong>and</strong> groundwater<br />
recharge show a decreasing trend, whereas<br />
groundwater recharge responded most sensitively<br />
to the anticipated climate change (-37% on<br />
average). Groundwater recharge decreased<br />
practically everywhere, whereas lower absolute<br />
changes are simulated in the loess area, where it is<br />
very low anyway due to soil properties.<br />
Fig. 2 Distributions of surface runoff <strong>and</strong><br />
groundwater recharge in the Elbe basin in 2000-<br />
2005 (upper parts) <strong>and</strong> in 2050-2055 (lower parts)<br />
for a set of 100 climate scenario realizations<br />
The uncertainty in hydrological response under<br />
climate change is quite high. For example, the<br />
histograms in Fig. 2 built on 100 scenario<br />
realizations compare surface runoff <strong>and</strong><br />
groundwater recharge for the Elbe basin in 2000-<br />
2005 (upper parts) <strong>and</strong> in 2050-2055 (lower parts).<br />
The hydrological responses <strong>and</strong> the propagation of<br />
uncertainty differ in three main Elbe subregions:<br />
the mountainous area, the loess subregion, <strong>and</strong> the<br />
lowl<strong>and</strong> area due to differences in geomorphological<br />
<strong>and</strong> climate conditions. According to the<br />
modeling results, the uncertainty in hydrological<br />
responses in lowl<strong>and</strong> is higher than that in<br />
mountainous area.<br />
w. wheat<br />
10%<br />
5%<br />
0%<br />
-5%<br />
-10%<br />
-15%<br />
-20%<br />
-25%<br />
sch lsax brb sax ELBE<br />
sch lsax brb sax ELBE<br />
10%<br />
5%<br />
0%<br />
w. barley<br />
-5%<br />
-10%<br />
-15%<br />
-20%<br />
-25%<br />
sch lsax brb sax ELBE<br />
25%<br />
20%<br />
15%<br />
s. maize<br />
10%<br />
5%<br />
0%<br />
-5%<br />
-10%<br />
Fig. 3 Change in crop yield with the confidence<br />
intervals under climate change scenario for the<br />
1067
Elbe basin <strong>and</strong> its four subregions: Schleswig-<br />
Holstein (sch), Lower Saxony (lsax), Br<strong>and</strong>enburg<br />
(brb) <strong>and</strong> Saxony-Anhalt (sax) in 2046-2055 in<br />
relation to the reference period 1960-1990<br />
The changes in crop yield (Fig. 3) were evaluated<br />
for the whole area on average, <strong>and</strong> for its four<br />
subregions: Schleswig-Holstein <strong>and</strong> Lower Saxony<br />
located in north-western part, Br<strong>and</strong>enburg located<br />
in eastern part, <strong>and</strong> Saxony-Anhalt located in the<br />
central part of the basin.<br />
The results are depicted in Fig. 3 as changes in<br />
percent related to the reference period with the<br />
confidence interval of 95%. According to the<br />
scenario, yield of winter wheat is expected to<br />
decrease in all subregions, with lowest results for<br />
Br<strong>and</strong>enburg <strong>and</strong> Saxony. Yield of winter barley<br />
would decrease rather moderately, whereas<br />
changes for Schleswig-Holstein <strong>and</strong> Lower Saxony<br />
are practically within ±5% interval. For silage<br />
maize, positive response is expected in northwestern<br />
part of the basin, whereas changes in<br />
Br<strong>and</strong>enbug <strong>and</strong> Saxony-Anhalt are negligible. The<br />
confidence intervals for winter barley are the most<br />
narrow (±4 to ±4.9%), whereas they are the largest<br />
for silage maize (±7.3 to ±8.1%).<br />
7. CONCLUSIONS<br />
The overall result of the study is that the mean<br />
water discharge <strong>and</strong> the mean groundwater<br />
recharge in the Elbe basin will be most likely<br />
decreased under expected climate change, but the<br />
uncertainty in hydrological response to changing<br />
climate is generally higher than the uncertainty in<br />
climate input. Crop yield is expected to decrease<br />
for cereals (winter wheat <strong>and</strong> winter barley), <strong>and</strong><br />
moderately increase for silage maize, with<br />
significant subregional differences. A multi-criteria<br />
validation <strong>and</strong> adjustment of model parameters can<br />
reduce the uncertainty level of the model<br />
predictions. The method used in the study is<br />
transferable to other river basins.<br />
REFERENCES<br />
GRASS4.1. Reference Manual, US Army Corps of<br />
Engineers. Construction Engineering<br />
Research Laboatories, Champaign, Illinois,<br />
1993.<br />
Hattermann, F., V.Krysanova, F. Wechsung, M.<br />
Wattenbach, Multiscale <strong>and</strong> multicriterial<br />
hydrological validation of the<br />
ecohydrological model SWIM. In: A.E.<br />
Rizzoli, A.J. Jakeman (eds.), Integrated<br />
assessment <strong>and</strong> decision support. Proc. of<br />
the 1st biennial meeting of the Int. Env.<br />
<strong>Modelling</strong> <strong>and</strong> <strong>Software</strong> Society, vol. 1,<br />
281-286, 2002.<br />
Krysanova, V., Müller-Wohlfeil, D.I. & Becker,<br />
A., Development <strong>and</strong> test of a spatially<br />
distributed hydrological / water quality<br />
model for mesoscale watersheds. Ecological<br />
<strong>Modelling</strong>, 106, 261-289, 1998.<br />
Krysanova, V., Wechsung, F., Becker, A.,<br />
Poschenrieder, W., Graefe, J., Mesoscale<br />
ecohydrological modelling to analyse<br />
regional effects of climate change.<br />
<strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong> Assessment,<br />
4, 4, 259-271, 1999.<br />
Krysanova, F. Wechsung, J. Arnold, R. Srinivasan,<br />
J. Williams, PIK Report Nr. 69 "SWIM<br />
(Soil <strong>and</strong> Water Integrated Model), User<br />
Manual", 239p., 2000.<br />
Werner, P.C. & F.-W. Gerstengarbe, A proposal<br />
for the development of climate scenarios.<br />
Climate Change, 8, 3, 171-182, 1997.<br />
1068
Coupling Surface And Ground Water Processes For Water<br />
Resources Simulation In Irrigated Alluvial Basins<br />
C. G<strong>and</strong>olfi a , A. Facchi a , D. Maggi a , B. Ortuani a<br />
a<br />
Institute of Agricultural Hydraulics, University of Milan, Italy (claudio.g<strong>and</strong>olfi@unimi.it)<br />
Abstract:<br />
Underst<strong>and</strong>ing the interaction between soil, vegetation <strong>and</strong> atmosphere processes <strong>and</strong> groundwater dynamics<br />
is of paramount importance in water resources planning <strong>and</strong> management in many practical applications. This<br />
is the case, for example, of the most important agricultural <strong>and</strong> industrial area in Italy, the Padana Plain,<br />
where intensive exploitation of groundwater for domestic <strong>and</strong> industrial supply coexists with massive<br />
diversions from surface water bodies, providing abundant irrigation to one of the most productive agricultural<br />
districts in Europe. Hydrological models of such complex systems need to include a number of components<br />
<strong>and</strong> should therefore seek a balance between capturing all relevant processes <strong>and</strong> maintaining data<br />
requirement <strong>and</strong> computing time at an affordable level. Water transfer through the unsaturated zone is a key<br />
hydrological process, at the interface between surface <strong>and</strong> ground water. The paper focuses on the analysis<br />
of the modelling approaches that are generally used to describe the soil water dynamics in hydrological<br />
models of water resources systems. A physically based approach, using numerical solutions of Richards<br />
equation, <strong>and</strong> two conceptual models, based on reservoirs cascade schemes, are compared. The analysis is<br />
part of a comprehensive modelling study of water resources in a 700 km 2 irrigation district in northern Italy.<br />
Simulations are carried out using ten years of rainfall data <strong>and</strong> a number of soil profiles that are representative<br />
of the pedological characteristics of the study area. Based on the analysis of results, showing significant<br />
differences in the simulated patterns of output variables, a number of remarks are drawn.<br />
Keywords: Unsaturated zone; Richards equation; reservoir cascade; SWAT; SWRRB<br />
1. INTRODUCTION<br />
Integrated modelling approaches have a great<br />
potential for application to water resources<br />
planning <strong>and</strong> management. This is especially true<br />
for densely settled irrigated plains, where the<br />
interaction between surface <strong>and</strong> ground waters<br />
plays a dominant role. In the case of the Padana<br />
Plain (Northern Italy), for example, water supply<br />
for irrigation is based on surface water diversions,<br />
while domestic <strong>and</strong> industrial supply is provided<br />
by groundwater abstractions. A significant portion<br />
of the total recharge to the aquifers is due to the<br />
percolation of water used for irrigation, which is<br />
mostly performed with traditional, low efficiency<br />
methods.<br />
A number of modelling tools have been proposed<br />
in the last decades, often including quite different<br />
representations of the individual hydrological<br />
processes. Water transfer through the unsaturated<br />
zone is one of the key process, as it determines<br />
evapotranspiration rates <strong>and</strong>, eventually, crop<br />
water stress on one side, <strong>and</strong> recharge to the<br />
aquifers on the other. Two main approaches are<br />
widely used for the mathematica representation of<br />
water flow in the unsaturated zone: numerical<br />
solutions of the Richards equation <strong>and</strong> reservoir<br />
cascade schemes.<br />
Conceptual, reservoir cascade schemes are<br />
included in many hydrological models, both at the<br />
field <strong>and</strong> basin scale [CREAMS, Knisel, 1980;<br />
ANSWERS, Beasley <strong>and</strong> Huggins, 1981; SWRRB,<br />
Williams et al., 1985; GLEAMS, Leonard et al.,<br />
1987; AGNPS, Young et al., 1987; KINEROS,<br />
Woolhiser et al., 1990; EPIC, Sharpley <strong>and</strong><br />
Williams, 1990; WEPP, Flanagan et al., 1995;<br />
SWAT, Neitsch et al., 1999; among the others].<br />
On the other h<strong>and</strong>, physically based approaches,<br />
using numerical solutions of Richards equations,<br />
first adopted at the local scale, have been also<br />
incorporated into basin scale hydrological models<br />
[HYDRUS, Simunek et al., 1998; SWAP, Van<br />
Dam et al., 1997; SWIF, Bouten, 1992; SOIL,<br />
Johnson <strong>and</strong> Jansson, 1991; SHE, Abbott et al.,<br />
1986; ONZAT, Van Drecht, 1983]..<br />
The objective of this paper is to present <strong>and</strong><br />
discuss the results of the comparison of three<br />
different unsaturated flow models in view of the<br />
implementation of an integrated model of water<br />
1069
esources in a large irrigation district (the 700 km 2<br />
Muzza district) in northern Italy [Facchi et al.,<br />
2004].<br />
2. UNSATURATED FLOW MODELS<br />
Three models of water flow in unsaturated soil<br />
were considered: SWAP [Van Dam et al., 1997]<br />
<strong>and</strong> the two conceptual, reservoir-type components<br />
included in SWRRB [Williams et al., 1985] <strong>and</strong><br />
ALHyMUS [Facchi, 2004].<br />
SWRRB belongs to the suite of USDA models,<br />
including among others EPIC, CREAMS, WEPP,<br />
which use the same basic approach to represent<br />
water flow in the unsaturated zone. This consists<br />
of a non-linear reservoirs cascade; the time<br />
constant of each reservoir is inversely proportional<br />
to the unsaturated hydraulic conductivity K, which<br />
is expressed by equation:<br />
⎛ θ ⎞<br />
K ( θ ) = K sat<br />
⎜<br />
θ<br />
⎟<br />
(1a)<br />
⎝ sat ⎠<br />
where θ sat [L 3 L -3 ] is the water content at saturation,<br />
K sat [LT -1 ] is the corresponding hydraulic<br />
conductivity, <strong>and</strong> β is a shape coefficient given by<br />
−2.655<br />
β = (1b)<br />
θ FC<br />
log<br />
θsat<br />
θ FC [L 3 L -3 ] being the water content at field<br />
capacity. This ensures that K=0.002 K sat at θ=θ FC .<br />
Outflow from each reservoir decreases with<br />
decreasing θ <strong>and</strong> is assumed to be null for θ ≤θ FC .<br />
The unsaturated flow component of ALHyMUS is<br />
also based on a non linear reservoir cascade<br />
scheme, including two reservoirs in the root-zone<br />
<strong>and</strong> one additional reservoir from the root-zone to<br />
the groundwater table. Outflow from each<br />
reservoir is proportional to hydraulic conductivity<br />
K, as expressed by Brooks & Corey equation<br />
n<br />
⎛ θ −θr<br />
⎞<br />
K( θ ) = K sat<br />
⎜<br />
⎟ (2)<br />
⎝ θsat<br />
−θr<br />
⎠<br />
where θ r [L 3 L -3 ] is the residual water content <strong>and</strong> n<br />
a shape coefficient.<br />
Finally, SWAP is a widely applied <strong>and</strong> well<br />
documented model, based on a finite difference<br />
solution of Richards equation. Van Genuchten <strong>and</strong><br />
Brooks & Corey equations were used here to<br />
describe water retention <strong>and</strong> conductivity curves,<br />
respectively.<br />
3. MATERIALS AND METHODS<br />
Four different soils were considered in the tests,<br />
covering a wide range of hydraulic characteristics<br />
(see Table 1): BSC, BLV, RAM, SCH,<br />
respectively characterised by silty loam; loam;<br />
s<strong>and</strong>y loam <strong>and</strong> s<strong>and</strong>y textures. For each soil three<br />
β<br />
different homogeneous profiles were considered<br />
with depths of the groundwater table from the<br />
ground surface of 1, 2 <strong>and</strong> 10 m.<br />
Table 1. Hydraulic parameters of the test soils<br />
Soil<br />
θ sat<br />
(m 3 m -3 )<br />
θ r<br />
(m 3 m -3 )<br />
K sat<br />
(cm h -1 )<br />
n<br />
(-)<br />
θ FC<br />
(m 3 m -3 )<br />
BSC 0.52 0.02 1.85 8.65 0.30<br />
BLV 0.50 0.07 1.43 8.85 0.29<br />
RAM 0.44 0.05 7.03 7.73 0.23<br />
SCH 0.42 0.04 35.18 6.96 0.16<br />
Inputs to the top of the profile are rainfall plus<br />
irrigation. A ten year (1993-2002) series of daily<br />
rainfall was used; comparison with historical data<br />
shows that dry (Q 90 ) years are well represented. A number of<br />
irrigations, ranging from three to five per year<br />
depending on weather conditions, was also<br />
supplied, according to the normal irrigation<br />
practices in the area.<br />
No interception, evapotranspiration <strong>and</strong> surface<br />
runoff are assumed to take place, the focus being<br />
only on the water accumulation <strong>and</strong> transport<br />
processes into the unsaturated profile.<br />
4. RESULTS<br />
Simulation results are compared in terms of:<br />
• water content in the top soil (i.e. the first 1 m of<br />
each profile);<br />
• water flow at the bottom of the profile, i.e. the<br />
recharge to the saturated groundwater layer.<br />
The former is the key variable for determining<br />
evapotranspiration rate <strong>and</strong> crop water stress,<br />
while the latter is crucial for the analysis <strong>and</strong><br />
modelling of surface–ground water interactions.<br />
4.1 Soil water content in the root zone<br />
A first observation is that the influence of the<br />
saturated surface depth (i.e. of the boundary<br />
condition at the bottom of the profile) cannot be<br />
captured by reservoir cascade models. Unless a<br />
specific component for capillary rise is included in<br />
these models (which however implies increasing<br />
the number of parameters) it does not make any<br />
difference wether the profile is 1, 2 or 10 m deep.<br />
On the other h<strong>and</strong>, however, the results show that<br />
this influence of the lower boundary condition is<br />
significant only for shallow profiles <strong>and</strong> fine<br />
textured soils. In practice, only for soil BSC <strong>and</strong><br />
profile depths up to very few meters the water<br />
content in the top soil changes significantly.<br />
Figure 1 shows that the difference in water content<br />
values is large between the 1 m <strong>and</strong> the 2 m profile,<br />
while it’s much smaller (soil BSC) or negligible<br />
(for coarser soils) between the latter <strong>and</strong> the 10 m<br />
profile. When the lower boundary condition is not<br />
1070
influential, both SWRRB <strong>and</strong> ALHyMUS<br />
generally show a good agreement with SWAP. For<br />
coarser soils, however, water contents lower than<br />
field capacity are often computed by SWAP <strong>and</strong><br />
ALHyMUS, while no flux underneath field<br />
capacity is allowed by SWRRB (Figure 2). This<br />
assumption apparently limits the flexibility of<br />
SWRRB in simulating soil water dynamics at low<br />
water contents. These observations are confirmed<br />
by Table 2, which reports values of some statistical<br />
<strong>and</strong> fitting indices<br />
Table 2. Average values <strong>and</strong> coefficient of variation for<br />
the simulation runs. Nash-Sutcliffe fitting index was<br />
computed using SWAP simulations as reference series<br />
Average value (m 3 m -3 )<br />
BSC BLV RAM SCH<br />
SWAP-10m 0.350 0.319 0.213 0.155<br />
ALHyMUS 0.345 0.318 0.214 0.156<br />
SWRRB 0.342 0.329 0.239 0.169<br />
CV (%)<br />
SWAP-10m 9.143 7.577 10.017 11.664<br />
ALHyMUS 9.484 7.188 10.247 11.670<br />
SWRRB 9.493 6.722 8.604 11.384<br />
Nash-Sutcliffe index (-)<br />
ALHyMUS 0.630 0.660 0.567 0.485<br />
SWRRB 0.568 0.587 -1.209 -0.757<br />
4.2 Outflow from the profile<br />
Figure 3 shows the outflow at the bottom of the<br />
profile for the loamy soil BLV. At increasing<br />
profile deptht, the effects of smoothing <strong>and</strong><br />
delaying of the input signal caused by increased<br />
soil depths can be seen very clearly. Both effects<br />
are magnified in finer textured soils, while for<br />
coarse soils the depth of the profile has a smaller<br />
influence on the output signal.<br />
Non linear behaviour is also very clear <strong>and</strong> can<br />
only partly be captured by reservoir cascade<br />
schemes (see Figure 4 <strong>and</strong> Table 3). Both<br />
SWRRB <strong>and</strong> ALHyMUS overestimate the<br />
smoothing effect when the thinner (1 m) soil<br />
profiles are considered, with a much more<br />
pronounced attenuation of peaks compared to<br />
SWAP, especially in the case of finer soils. On the<br />
contrary, when thicker profiles are analysed, the<br />
amplitude of the signal is reasonably well captured,<br />
but the phase is poorly described.<br />
Special attention should be paid when using<br />
SWRRB with coarse soils, due to possible<br />
undesired effects of the discontinuity in the<br />
hydraulic conductivity function at water content<br />
equal to the field capacity. For example, the s<strong>and</strong>y<br />
soil SCH drops from a relatively high K=16.9<br />
mm/d at field capacity (Eq. 1) to zero at lower<br />
water contents. This abrupt change in conductivity<br />
produces a typical pulsating pattern of recharge,<br />
which does not seem to have any physical reason.<br />
The agreement between SWAP <strong>and</strong> ALHyMUS or<br />
SWRRB can be improved if the number of<br />
reservoir in the cascade is allowed to vary with<br />
changing profile depth, as illustrated in Figure 5,<br />
where ALHyMUS has been run for the 10 m<br />
profile, using up to five reservoirs for the<br />
percolation layer underlying the root zone (for<br />
which always two reservoirs were used). However,<br />
no well established <strong>and</strong> scientifically sound rule for<br />
fixing the reservoir number is available, <strong>and</strong> only<br />
empirical indications can be found in literature<br />
(see, e.g. Besbes <strong>and</strong> De Marsily, 1984).<br />
Since the comparison is carried out in view of<br />
coupling unsaturated zone model with a regional<br />
groundwater flow model, it was deemed important<br />
to check to which extent the differences in recharge<br />
are mitigated by considering longer simulation<br />
time steps, which are generally used in the latter<br />
models. As it could be expected the differences<br />
become smaller as the time step increases. Figure<br />
6 <strong>and</strong> Table 3 show, for example, the results<br />
aggregated at three months time step, which is the<br />
stress period used for groundwater simulation in<br />
the Muzza case study.<br />
Table 3. Nash-Sutcliffe indices for the daily <strong>and</strong> threemonthly<br />
recharge patterns for all the selected scenarios<br />
daily<br />
three-monthly<br />
ALHyMUS SWRRB ALHyMUS SWRRB<br />
1m-BSC 0.627 0.501 0.889 0.939<br />
1m-BLV 0.419 0.689 0.938 0.935<br />
1m-RAM 0.678 0.917 0.956 0.998<br />
1m-SCH 0.717 0.817 0.973 0.996<br />
2m-BSC 0.491 0.561 0.831 0.878<br />
2m-BLV 0.735 0.749 0.939 0.942<br />
2m-RAM 0.674 0.616 0.964 0.938<br />
2m-SCH 0.753 -0.206 0.979 0.925<br />
10m-BSC 0.494 0.473 0.669 0.652<br />
10m-BLV 0.227 0.219 0.674 0.611<br />
10m-RAM 0.221 -3.304 0.721 -0.640<br />
10m-SCH 0.219 -11.154 0.726 -0.833<br />
1071
0.55<br />
0.5<br />
0.45<br />
a<br />
0.4<br />
0.35<br />
0.3<br />
0.25<br />
1 m 2 m 10 m<br />
1/1/97<br />
1/2/97<br />
1/3/97<br />
1/4/97<br />
1/5/97<br />
1/6/97<br />
1/7/97<br />
1/8/97<br />
1/9/97<br />
1/10/97<br />
1/11/97<br />
1/12/97<br />
1/1/98<br />
1/2/98<br />
1/3/98<br />
1/4/98<br />
1/5/98<br />
1/6/98<br />
1/7/98<br />
1/8/98<br />
1/9/98<br />
1/10/98<br />
1/11/98<br />
1/12/98<br />
0.4<br />
0.35<br />
1 m 2 m 10 m<br />
0.3<br />
b<br />
0.25<br />
0.2<br />
0.15<br />
0.1<br />
1/1/97<br />
1/2/97<br />
1/3/97<br />
1/4/97<br />
1/5/97<br />
1/6/97<br />
1/7/97<br />
1/8/97<br />
1/9/97<br />
1/10/97<br />
1/11/97<br />
1/12/97<br />
1/1/98<br />
1/2/98<br />
1/3/98<br />
1/4/98<br />
1/5/98<br />
1/6/98<br />
1/7/98<br />
1/8/98<br />
1/9/98<br />
1/10/98<br />
1/11/98<br />
1/12/98<br />
Figure 1. Daily patterns of soil water content [L 3 L -3 ] in the top soil, simulated with SWAP with<br />
water table depth of 1, 2, <strong>and</strong> 10 m; (a) soil BSC, (b) soil SCH; years years 1997-1998<br />
0.4<br />
0.3<br />
ALHyMUS<br />
SWAP<br />
DR-SWRRB<br />
field capacity<br />
a<br />
0.2<br />
0.1<br />
1/1/97<br />
1/2/97<br />
1/3/97<br />
1/4/97<br />
1/5/97<br />
1/6/97<br />
1/7/97<br />
1/8/97<br />
1/9/97<br />
1/10/97<br />
1/11/97<br />
1/12/97<br />
1/1/98<br />
1/2/98<br />
1/3/98<br />
1/4/98<br />
1/5/98<br />
1/6/98<br />
1/7/98<br />
1/8/98<br />
1/9/98<br />
1/10/98<br />
1/11/98<br />
1/12/98<br />
b<br />
0.4<br />
0.3<br />
0.2<br />
ALHyMUS<br />
SWAp<br />
DR_SWRRB<br />
field capacity<br />
0.1<br />
1/1/97<br />
1/2/97<br />
1/3/97<br />
1/4/97<br />
1/5/97<br />
1/6/97<br />
1/7/97<br />
1/8/97<br />
1/9/97<br />
1/10/97<br />
1/11/97<br />
1/12/97<br />
1/1/98<br />
1/2/98<br />
1/3/98<br />
1/4/98<br />
1/5/98<br />
1/6/98<br />
1/7/98<br />
1/8/98<br />
1/9/98<br />
1/10/98<br />
1/11/98<br />
1/12/98<br />
Figure 2. Daily patterns of soil water content [L 3 L -3 ] in the top soil, simulated with SWAP, ALHyMUS<br />
<strong>and</strong> SWRRB for soil SCH <strong>and</strong> water table depth at (a) 1 m <strong>and</strong> (b) 10 m; years years 1997-1998<br />
100<br />
80<br />
60<br />
1 m<br />
2 m<br />
10 m<br />
40<br />
20<br />
0<br />
1/1/97<br />
1/2/97<br />
1/3/97<br />
1/4/97<br />
1/5/97<br />
1/6/97<br />
1/7/97<br />
1/8/97<br />
1/9/97<br />
1/10/97<br />
1/11/97<br />
1/12/97<br />
1/1/98<br />
1/2/98<br />
1/3/98<br />
1/4/98<br />
1/5/98<br />
1/6/98<br />
1/7/98<br />
1/8/98<br />
1/9/98<br />
1/10/98<br />
1/11/98<br />
1/12/98<br />
Figure 3. Daily patterns of the outflow at the bottom of the profile (mm d -1 ) simulated with SWAP with<br />
water table depth of 1, 2, <strong>and</strong> 10 m; soil BSC; years 1997-1998<br />
1072
30<br />
25<br />
20<br />
SWAP<br />
SWRRB<br />
ALHyMUS<br />
a<br />
15<br />
10<br />
5<br />
0<br />
1/1/97<br />
1/2/97<br />
1/3/97<br />
1/4/97<br />
1/5/97<br />
1/6/97<br />
1/7/97<br />
1/8/97<br />
1/9/97<br />
1/10/97<br />
1/11/97<br />
1/12/97<br />
1/1/98<br />
1/2/98<br />
1/3/98<br />
1/4/98<br />
1/5/98<br />
1/6/98<br />
1/7/98<br />
1/8/98<br />
1/9/98<br />
1/10/98<br />
1/11/98<br />
1/12/98<br />
10<br />
8<br />
6<br />
SWAP<br />
SWRRB<br />
ALHyMUS<br />
b<br />
4<br />
2<br />
0<br />
1/1/97<br />
1/2/97<br />
1/3/97<br />
1/4/97<br />
1/5/97<br />
1/6/97<br />
1/7/97<br />
1/8/97<br />
1/9/97<br />
1/10/97<br />
1/11/97<br />
1/12/97<br />
1/1/98<br />
1/2/98<br />
1/3/98<br />
1/4/98<br />
1/5/98<br />
1/6/98<br />
1/7/98<br />
1/8/98<br />
1/9/98<br />
1/10/98<br />
1/11/98<br />
1/12/98<br />
Figure 4. Daily patterns of the outflow at the bottom of the profile (mm d -1 ) simulated with SWAP<br />
with water table depth of 1 <strong>and</strong> 10 m; soil BSC; years 1997-1998<br />
10<br />
8<br />
6<br />
SWAP<br />
ALHyMUS_1x10<br />
ALHyMUS_2X5<br />
ALHyMUS_5x2<br />
4<br />
2<br />
0<br />
1/1/97<br />
1/2/97<br />
1/3/97<br />
1/4/97<br />
1/5/97<br />
1/6/97<br />
1/7/97<br />
1/8/97<br />
1/9/97<br />
1/10/97<br />
1/11/97<br />
1/12/97<br />
1/1/98<br />
1/2/98<br />
1/3/98<br />
1/4/98<br />
1/5/98<br />
1/6/98<br />
1/7/98<br />
1/8/98<br />
1/9/98<br />
1/10/98<br />
1/11/98<br />
1/12/98<br />
Figure 5. Daily outflow from a 10 m profile, BSC soil, simulated by SWAP <strong>and</strong> ALHyMUS, at changing<br />
of the reservoir number in the cascade; years 1997-1998<br />
10<br />
8<br />
6<br />
SWAP<br />
DR-SWRRB<br />
ALHyMUS<br />
a<br />
4<br />
2<br />
0<br />
1/1/97<br />
1/2/97<br />
1/3/97<br />
1/4/97<br />
1/5/97<br />
1/6/97<br />
1/7/97<br />
1/8/97<br />
1/9/97<br />
1/10/97<br />
1/11/97<br />
1/12/97<br />
1/1/98<br />
1/2/98<br />
1/3/98<br />
1/4/98<br />
1/5/98<br />
1/6/98<br />
1/7/98<br />
1/8/98<br />
1/9/98<br />
1/10/98<br />
1/11/98<br />
1/12/98<br />
500<br />
400<br />
300<br />
DR-SWRRB<br />
ALHyMUS<br />
SWAP<br />
b<br />
200<br />
100<br />
0<br />
1/1/97<br />
1/2/97<br />
4/3/97<br />
4/4/97<br />
5/5/97<br />
5/6/97<br />
6/7/97<br />
6/8/97<br />
6/9/97<br />
7/10/97<br />
7/11/97<br />
8/12/97<br />
8/1/98<br />
8/2/98<br />
11/3/98<br />
11/4/98<br />
12/5/98<br />
12/6/98<br />
13/7/98<br />
13/8/98<br />
13/9/98<br />
14/10/98<br />
14/11/98<br />
15/12/98<br />
Figure 6. Daily (a) <strong>and</strong> three-monthly (b) patterns of of the outflow at the bottom of the profile,<br />
simulated with SWAP, ALHyMUS <strong>and</strong> SWRRB for soil SCH <strong>and</strong> water table depth of 10 m; years 1997-1998<br />
1073
5. CONCLUDING REMARKS<br />
In the paper results of the comparison of a widely<br />
applied model based on Richards equation –<br />
SWAP – with two reservoir models – SWRRB <strong>and</strong><br />
HALyMUS – were presented. The analysis was<br />
carried out in the framework of an comprehensive<br />
modelling study of water resources in a 700 km 2<br />
irrigation district in northern Italy [Facchi et al.,<br />
2004]. A number of soil profiles, representative of<br />
the study area characteristics, were selected <strong>and</strong><br />
simulations were carried out using 10 years of<br />
daily rainfall observation plus the normal irrigation<br />
supply, according to the current practices in the<br />
area. Soil types range from silty loam to s<strong>and</strong> <strong>and</strong><br />
profile depths from 1 to 10 m.<br />
The comparison was focused on two output<br />
variables, i.e. water content of the top soil (first 1<br />
m) <strong>and</strong> outflow at the bottom of the profile.<br />
A first observation is that reservoir models cannot<br />
capture the influence of water table depth on the<br />
soil water profile. This may extend to the top soil<br />
when the water table depth is small <strong>and</strong> soil texture<br />
is fine. A site-specific analysis needs therefore to<br />
be carried out: in the Muzza study area, for<br />
example, the water table effects turned out to be<br />
relevant only for the finest soil (BSC) <strong>and</strong> for<br />
water table depths of 3-4 m, which are not very<br />
common in the area. When the lower boundary is<br />
not influential the results of the three models are<br />
generally in good agreement. For coarser soils,<br />
however, while water contents computed by SWAP<br />
<strong>and</strong> ALHyMUS often drop below field capacity,<br />
SWRRB tends to overestimate the soil water<br />
content due to the assumption of no flux at water<br />
content lower than field capacity. This is reflected<br />
also in the pattern of the outflow from the profile,<br />
which may show a typical pulsating behaviour as a<br />
result of the sudden drop of hydraulic conductivity<br />
when field capacity is reached.<br />
The ability of reservoir models to mimic the<br />
nonlinear effects on input-output transformation is<br />
rather poor. In general, both SWRRB <strong>and</strong><br />
ALHyMUS may overestimate the smoothing effect<br />
when thin soil profiles are considered, with a more<br />
pronounced attenuation of peaks compared to<br />
SWAP. On the contrary, when thicker profiles are<br />
analysed, the amplitude of the signal is reasonably<br />
well captured, but the phase is poorly described.<br />
Only when outflow is aggregated over longer time<br />
intervals (e.g. months) the differences decrease<br />
significantly. This observation may be relevant in<br />
practical applications, when the unsaturated flow<br />
model is coupled with a regional aquifer model,<br />
providing recharge fluxes over the aquifer stress<br />
periods, which are often of the order of decades or<br />
months.<br />
Acknowledgements<br />
The research was financed with funds of D.G. Agricoltura-<br />
Regione Lombardia <strong>and</strong> of the MIUR-PRIN project<br />
“Assimilazione di osservazione satellitare e modellistica<br />
idrologica nel monitoraggio delle risorse idriche in agricoltura”,<br />
which are gratefully acknowledged. The authors wish to thank<br />
Consorzio Muzza Bassa Lodigiana for support <strong>and</strong> cooperation<br />
<strong>and</strong> ERSAF for providing meteorological <strong>and</strong> soil data.<br />
6. REFERENCES<br />
Abbot M.B., Bathurst J.C., Cunge J.A., O’Connel P.E.,<br />
Rasmussen J. (1986) An introduction to the European<br />
hydrological system “SHE”, 2. Structure of a physically<br />
based, distributed modelling system. Journal of Hydrology<br />
87, 61-77<br />
Beasley D.B., Huggins L.F. (1981) ANSWERS - Users<br />
Manual. EPA-905/9-82-001, U.S. EPA, Region V, Chicago,<br />
IL, USA<br />
Besbes M., de Marsily G. (1984) From infiltration to recharge:<br />
use of a parametric transfer function. Journal of Hydrology<br />
74, 271-293<br />
Bouten W. (1992) Monitoring <strong>and</strong> modelling forest<br />
hydrological processes in support of acidification research.<br />
Ph.D. thesis, Universteit van Amsterdam, The Netherl<strong>and</strong>s<br />
Facchi A. (2004) Nuove tecnologie per la pianificazione<br />
dell’irrigazione a scala di bacino, PhD Thesis, University of<br />
Milan, Italy<br />
Facchi A., Ortuani B., Maggi D., G<strong>and</strong>olfi C.(2004) Coupled<br />
SVAT-groundwater model for water resources simulation in<br />
irrigated alluvial plains, <strong>Environmental</strong> <strong>Modelling</strong> <strong>and</strong><br />
<strong>Software</strong> [in press]<br />
Flanagan, D.C., Livingston S.J. (1995) WEPP User Summary.<br />
NSERL Report No.11. West Lafayette, National Soil Erosion<br />
Research Laboratory<br />
Johnson H., Jansson P.E. (1991) Water balance <strong>and</strong> soil<br />
moisture dynamics of field plots with barley <strong>and</strong> grass ley.<br />
Journal of Hydrology, 129, pp. 149-173<br />
Knisel W.G. (1980) CREAMS: a field scale model for<br />
chemicals, runoff <strong>and</strong> erosion from agricultural management<br />
systems. Conservation Research Report No. 26, USDA<br />
Leonard R.A., Knisel W.G., Still D.A. (1987) GLEAMS:<br />
Groundwater Loading Effects of Agricultural Managment<br />
System. Trans. ASAE, 30, 1403-1418<br />
Neitsch S.L., Arnold J.G., Kiniry J.R., Williams J.R. (1999)<br />
SWAT Soil <strong>and</strong> Water Assessment Tool Theoretical<br />
Documentation<br />
Sharpley A.N., Williams J.R. (1990) EPIC - Erosion<br />
Productivity Impact Calculator: 1, Model documentation,<br />
USDA, ARS-31<br />
Simunek J., Huang K., van Genuchten M. Th. (1998) The<br />
HYDRUS code. Research Report, No. 144, U.S. Salinity<br />
Laboratory Agricultural Research Service, U.S. Department<br />
of Agriculture, Riverside, California<br />
Van Dam J.C., Huygen J., Wesseling J.G., Feddes R.A., Kabat<br />
P., van Walsum P.E.V., Groenendijk P., van Diepen C.A.<br />
(1997). SWAP version 2.0, Theory. Technical Document 45,<br />
DLO Win<strong>and</strong> Staring Centre, Report 71, Department Water<br />
Resources, Agricultural University, Wageningen<br />
Van Drecht G. (1983) Simulatie van het vertikale, nietstationaire<br />
transport van water en een daarin opgeloste stof in<br />
de grond. Rid meded, 83-140<br />
Williams J.R., Nicks A.D., Arnold J.G. (1985) Simulator for<br />
Water Resources in Rural Basins. ASCE J. Hydraulic<br />
Engineering 111, 970-986<br />
Woolhiser D.A., Smith R.E., Goodrich W.E. (1990) KINEROS:<br />
a kinematic runoff <strong>and</strong> erosion model - documentation <strong>and</strong><br />
user manual, ARS-77<br />
Young R.A., Onstad C.A., Bosch D.D. (1987) AGPNS: An<br />
agricultural nonpoint source pollution model: a watershed<br />
analysis tool. USDA Conservation Research Report 35<br />
1074
Investigating Spatial Pattern Comparison Methods for<br />
Distributed Hydrological Model Assessment<br />
S.R. Weal<strong>and</strong>s a, b , R.B. Grayson a <strong>and</strong> J.P. Walker b<br />
a Cooperative Research Centre for Catchment Hydrology, Australia<br />
b Department of Civil <strong>and</strong> <strong>Environmental</strong> Engineering, The University of Melbourne, Australia<br />
Abstract: Distributed hydrological models combine observations <strong>and</strong> knowledge about a hydrological<br />
system to make spatial predictions of hydrological attributes. These models require methods to assess their<br />
performance at spatial prediction. The current practice for assessment is simplistic. For qualitative<br />
assessment, simulated spatial patterns are compared visually against an observed pattern to assess their<br />
spatial similarity. To obtain a quantitative measure of similarity, each individual location is numerically<br />
compared to produce either a mean squared error (MSE) or correlation statistic. Both of these comparisons<br />
have their limitations. The visual comparison is subjective <strong>and</strong> the numerical comparison generally ignores<br />
the spatial structure of the patterns. There is demonstrable need for repeatable methods that can capture <strong>and</strong><br />
quantify the important aspects of visual comparison. This paper demonstrates such a method from the image<br />
processing literature. It is a modification of the MSE statistic, called the information mean squared error<br />
(IMSE). This method weights each location in the spatial pattern by the ‘informativeness’ of ‘an event’ at<br />
that location. The weighted spatial patterns are then compared using a st<strong>and</strong>ard MSE statistic. IMSE aims to<br />
emulate human vision by more heavily weighting informative pixels. This paper applies IMSE to spatial<br />
patterns of soil moisture content. It is found to work well when using local variance as the ‘event’, as this<br />
helps enhance the general spatial trends that humans readily recognise. However, when the two spatial<br />
patterns are vastly different, IMSE proves to be less reliable due to the inconsistent weightings calculated for<br />
each spatial pattern.<br />
Keywords: Spatial pattern; Model assessment; Comparison; Hydrology; Distributed models; Self information<br />
1. INTRODUCTION<br />
There is increasing recognition in hydrological<br />
modelling of a need for improved methods for<br />
comparing spatial patterns [Grayson et al., 2002;<br />
Jetten et al., 2003]. This has arisen from the<br />
increased availability of observed spatial patterns<br />
for testing the spatial predictions from distributed<br />
models. Grayson <strong>and</strong> Blöschl [2000] provide<br />
many examples of modelling projects where<br />
spatial patterns have been observed, with a<br />
purpose to assess the spatial component of the<br />
model predictions. However, within these<br />
projects there has been little use of new spatial<br />
pattern comparison methods. Most studies rely<br />
on st<strong>and</strong>ard statistical measures (like mean<br />
squared error) or subjective visual comparisons to<br />
tell the story of how well the model is predicting<br />
the spatial patterns. This work pursues new<br />
approaches for the spatial pattern comparison<br />
task.<br />
Spatial patterns in hydrology are usually gridbased<br />
representations of an area (or catchment),<br />
with a value provided at every grid cell (or pixel).<br />
Each grid cell is square <strong>and</strong> has dimensions<br />
specified as the cell size (or resolution). Spatial<br />
patterns are effectively the same as grey level<br />
images, although the number of discrete pixel<br />
values is usually larger than in a st<strong>and</strong>ard (8-bit)<br />
image b<strong>and</strong>. Observed spatial patterns can be<br />
obtained via grid-based field measurements, by<br />
interpolation of sparse field measurements or<br />
from remote sensing. These measurements then<br />
require processing to ensure they are consistent<br />
with the predicted spatial patterns from a<br />
hydrological model (i.e. with equivalent support,<br />
spacing, extent). The spatial patterns must be<br />
carefully prepared so that they are comparable.<br />
This work does not focus on this aspect of spatial<br />
patterns, preferring to concentrate on the<br />
comparison once the spatial patterns have been<br />
prepared correctly.<br />
2. COMPARING SPATIAL PATTERNS<br />
When comparing spatial patterns, the ability to<br />
obtain a measure of similarity is essential.<br />
Methods that provide a quantitative measure can<br />
1075
e used to compare an observation with multiple<br />
simulations. The resultant measure can then be<br />
used to determine which of the simulations are<br />
more similar. To underst<strong>and</strong> which features of<br />
the simulations are more similar, the user needs to<br />
underst<strong>and</strong> how the comparison method<br />
computed the similarity measure. By interpreting<br />
the performance of these measures with<br />
hydrological spatial patterns, certain methods may<br />
emerge that are more suitable.<br />
A review of the current suite of methods used for<br />
spatial pattern comparison in hydrology identifies<br />
the features that are currently compared. The<br />
spatial pattern comparisons presented in Grayson<br />
<strong>and</strong> Blöschl [2000] <strong>and</strong> Grayson et al. [2002]<br />
provide a comprehensive cross-section of the<br />
commonly used methods. The most widely used<br />
method is visual comparison, which allows the<br />
user to draw on their background knowledge<br />
about the study area <strong>and</strong> model structure to<br />
interpret the similarity. This method will always<br />
be used when presented with two figures<br />
depicting spatial patterns. It is also used to<br />
compare time series data or transects that have<br />
been extracted from the observed <strong>and</strong> predicted<br />
spatial patterns. This type of comparison is too<br />
subjective for repeatable <strong>and</strong> rigorous<br />
comparison. It is also very time-consuming,<br />
unable to interpret large spatial patterns<br />
completely, <strong>and</strong> it cannot produce a quantitative<br />
measure.<br />
Most common quantitative measures used are<br />
global measures, which characterise the spatial<br />
pattern first using statistics or indices (e.g. mean<br />
error to identify bias, spatial correlation length to<br />
compare spatial statistical structure). These<br />
summaries are then compared numerically. For<br />
local measures, pixel-by-pixel comparisons such<br />
as mean squared error (MSE) (to assess the local<br />
agreement between values) are the dominant<br />
measures. Here, the residuals are computed<br />
between two spatial patterns <strong>and</strong> then squared <strong>and</strong><br />
averaged. The residuals are usually analysed to<br />
detect relationships with topographic variables.<br />
For spatial patterns to be judged as being similar<br />
with all of these local measures, there must be<br />
close agreement between the pixel values at<br />
coincident locations, so these techniques are very<br />
sensitive to minor shifts. They are also<br />
influenced by disagreement between coincident<br />
pixels, even though there may be close agreement<br />
with neighbouring pixels. This is especially<br />
evident when the ‘support’ of the two spatial<br />
patterns does not match (i.e. the observed pixel<br />
value has a support of much less than the cell<br />
size, whereas the predicted value represents the<br />
average of the entire pixel). In these situations,<br />
small-scale variance can mask the overall pattern<br />
<strong>and</strong> thus lead to poor results for local similarity<br />
measures.<br />
These current methods are useful for the analysis<br />
of similarity between spatial patterns, but are<br />
limited in their ability to measure certain aspects<br />
of similarity. Other methods for characterising<br />
<strong>and</strong> comparing more detailed features of spatial<br />
patterns are necessary. By underst<strong>and</strong>ing the<br />
strengths of new similarity measures <strong>and</strong><br />
experimenting with hydrological data sets,<br />
alternative methods for the comparison of spatial<br />
patterns can be further developed.<br />
3. OTHER COMPARISON METHODS<br />
There is a large amount of research in other<br />
disciplines, such as image processing <strong>and</strong><br />
computer vision, that can suggest methods for<br />
working with spatial patterns in hydrology (e.g.<br />
segmentation, image filtering) [Scheibe, 1993].<br />
However, not all techniques in other fields are<br />
applicable to hydrological patterns. For example,<br />
with face recognition, it is usual for the observed<br />
image (a face) to be processed down to a set of<br />
features (such as eye locations) that are stable in<br />
all conditions (e.g. different lighting). This set is<br />
then compared against a large database of features<br />
to find a match. The spatial patterns present in<br />
hydrology rarely have any known features <strong>and</strong><br />
therefore need solutions that are more generic. A<br />
review of this literature is given in Weal<strong>and</strong>s et al.<br />
[submitted; 2003]. In this paper, the focus is on<br />
an approach for image comparison that was<br />
initially developed for assessing the effect of<br />
image filtering on the original image.<br />
3.1. Information Mean Squared Error<br />
When an image is filtered (<strong>and</strong> often distorted), a<br />
measure of its similarity to the original is desired.<br />
In Tompa et al. [2000], a measure called the<br />
information mean squared error (IMSE) is<br />
developed. This measure aims to reflect the level<br />
of similarity that a human observer would<br />
perceive. When humans compare images, it is<br />
common for differences in the main features to be<br />
weighted more heavily than differences in the<br />
background values. Similarly, for spatial pattern<br />
comparisons, the less common values (e.g. highs<br />
or lows) attract more attention during visual<br />
comparison. The basic premise of the IMSE is<br />
quite simple – weight ‘events’ that occur less<br />
frequently in the spatial pattern more highly in the<br />
comparison. The basic events that occur within a<br />
spatial pattern are the actual pixel values.<br />
However, other events like local variance (i.e. the<br />
variance of pixel values within a neighbourhood)<br />
can also be used for the calculation of weights.<br />
1076
The size of the neighbourhood is related to the<br />
size of the features (e.g. small features have high<br />
variance in small neighbourhoods).<br />
The level of weighting applied in this method<br />
represents the ‘informativenss’ of the particular<br />
event. This is measured using Shannon’s selfinformation<br />
measure, as applied in Topper <strong>and</strong><br />
Jernigan [1989]. This is defined to be<br />
I(x) = −log<br />
P<br />
n<br />
N<br />
x<br />
( x) = ,<br />
Y<br />
P(x)<br />
(1)<br />
where I(x) is the self-information for event x, Y is<br />
the base for the logarithm (base e is used here),<br />
P(x) is the probability of event x occurring, n x is<br />
the number of pixels with event equal to x, <strong>and</strong> N<br />
is the total number of pixels in the spatial pattern.<br />
Due to the logarithm, these weights are maximum<br />
when P(x) is close to zero, <strong>and</strong> minimum when<br />
P(x) approaches one. These effects are desirable,<br />
so that an event that is almost everywhere in the<br />
spatial pattern would contain little information<br />
(i.e. it is the background), while the most<br />
infrequent events would have the maximum. The<br />
weights produced can vary from almost zero to<br />
very large numbers, depending on the number of<br />
pixels in the spatial pattern (<strong>and</strong> the base of the<br />
logarithm).<br />
Once the self-information is computed for each<br />
pixel, the original spatial pattern is multiplied by<br />
the weights. The calculation of weights is done<br />
for all spatial patterns being compared. To<br />
compute the IMSE similarity measure, a st<strong>and</strong>ard<br />
MSE calculation is done between two weighted<br />
spatial patterns.<br />
3.2. Selecting the Event for Weighting<br />
The event used for calculating self-information<br />
measures does not have to be the actual pixel<br />
value. Rather, the event chosen should be the<br />
characteristic of the spatial pattern that is<br />
responsible for separating the features of interest<br />
from the background. For example, in a spatial<br />
pattern with a few areas of very high pixel values<br />
on a background of very low values, then pixel<br />
value is a good event. For a more homogeneous<br />
spatial pattern (with less obvious features), local<br />
variance is better, as this is the visual cue for<br />
something of interest in the spatial pattern (as<br />
there is variance within the neighbourhood).<br />
Another consideration in selecting the event is the<br />
number of distinct categories in which the event<br />
occurs. In Tompa et al. [2000], the images were<br />
always single b<strong>and</strong>, 8-bit images (i.e. having 256<br />
individual values). With spatial patterns, there<br />
NON-CATEGORISED<br />
are often thous<strong>and</strong>s of different values, which<br />
should be categorised (or quantised) prior to<br />
having the weights calculated. If this is not done,<br />
a spatial pattern containing 400 pixels could have<br />
400 different values, resulting in every pixel<br />
being weighted equally. As this method was<br />
developed to represent perceptual similarity, the<br />
number of different categories in the spatial<br />
pattern should correspond to the number of<br />
categories the human observer can discern in the<br />
spatial pattern. A firm value cannot be placed on<br />
this, as it varies with the observer. As such, this<br />
should be determined empirically. In Figure 1, a<br />
human observer can probably discern about 10-20<br />
individual categories (when displayed like this),<br />
while in fact there are 146 different values in the<br />
non-categorised spatial pattern.<br />
3.3. Using the IMSE Measure<br />
16 CATEGORIES<br />
8 CATEGORIES 32 CATEGORIES<br />
Figure 1. Observed spatial pattern of soil<br />
moisture displayed in different categories.<br />
The increasing level of grey denotes higher<br />
soil moisture content.<br />
The method described produces a measure that<br />
indicates the level of similarity between the<br />
spatial patterns, with high weights assigned to the<br />
‘more informative’ pixels. In the resulting<br />
measure, a smaller IMSE value denotes more<br />
similar spatial patterns. As the self-information<br />
weights are specific to the spatial pattern being<br />
compared, they cannot be used for intercomparison.<br />
For example, this measure cannot be<br />
used to compare a pair of spatial patterns from<br />
spring, then a pair from winter, with a view to<br />
stating which pair of spatial patterns are more<br />
similar. Instead, this method is suitable for<br />
comparing multiple spatial patterns of the same<br />
event.<br />
One example that is common in modelling<br />
projects is for a single observed spatial pattern to<br />
be compared with many different model<br />
simulations (with different parameter sets). By<br />
obtaining measures of similarity between the<br />
observed spatial pattern <strong>and</strong> each simulation, the<br />
modeller can help decide which parameter sets<br />
lead to best agreement.<br />
1077
4. COMPARISON DEMONSTRATION<br />
This section investigates the use of IMSE for<br />
hydrological spatial patterns. It will be presented<br />
along with measures of bias, correlation <strong>and</strong> root<br />
mean squared error to help interpret the results.<br />
The aim of this demonstration is to characterise<br />
which simulated spatial patterns are most similar<br />
to the observed spatial pattern for two different<br />
dates. The methods all provide measures that can<br />
be used to judge different aspects of similarity.<br />
This demonstration will undertake analysis of the<br />
similarity measures first <strong>and</strong> then look at the<br />
spatial patterns afterwards to discuss the<br />
performance.<br />
4.1. Observations <strong>and</strong> Simulations<br />
The observed spatial patterns analysed here are<br />
from the Tarrawarra project [Western et al.,<br />
2000]. In this study, soil moisture was measured<br />
in the field at regularly spaced grid intervals.<br />
This data has then been smoothed using<br />
geostatistical methods, to make the support of the<br />
field measurements compatible with the model<br />
simulations <strong>and</strong> to add in variability of the<br />
measurement technique (with a nugget effect on<br />
the variogram used for smoothing), the details of<br />
which are in Western <strong>and</strong> Grayson [2000]. The<br />
two observed spatial patterns represent vastly<br />
different soil moisture conditions related to the<br />
season in which they were measured.<br />
Simulations have been produced using the Thales<br />
modelling framework, with the details of these<br />
particular simulations given in Western <strong>and</strong><br />
Grayson [2000]. The 10 different simulations<br />
represent different parameterisations for the<br />
model. The simulations numbered 1-3 ignore<br />
spatially variable evapotranspiration (ET), while<br />
the others allow spatially variable ET (which is<br />
often related to slope <strong>and</strong> aspect). All simulations<br />
have been resampled from an element-based<br />
network onto a regular grid to correspond with the<br />
observed spatial patterns.<br />
4.2. Similarity Measures<br />
With each comparison (between the observed <strong>and</strong><br />
simulated patterns) there are five similarity<br />
measures computed (Table 1). These are bias, R 2<br />
correlation (which ignores bias), root mean<br />
squared error (RMSE), IMSE using pixel value<br />
(pv) <strong>and</strong> IMSE using local variance (lv) (within a<br />
3 pixel square window). Using a small window<br />
ensures that the variance is only computed for the<br />
pixel <strong>and</strong> its 8 neighbours. Two different IMSE<br />
measures are given to highlight the impact of the<br />
event chosen. Bias is used to assess if one spatial<br />
pattern has an overall higher or lower value. This<br />
could be removed from subsequent comparison if<br />
desired. RMSE provides an overall summary of<br />
the difference between the spatial patterns at each<br />
location, with a penalty for large discrepancies<br />
(the squaring of residuals). IMSE also measures<br />
the difference between the spatial patterns at each<br />
location, but with an emphasis on areas having<br />
high information content. If a pixel is in a<br />
frequently occurring event category, its value will<br />
be reduced. If the pixel is in a rarely occurring<br />
event category, its value will be increased.<br />
During the subsequent comparison, if the high<br />
information areas between the spatial patterns are<br />
vastly different, then the measure will be high,<br />
while differences between the ‘low information’<br />
pixels have far less effect.<br />
4.3. Similarity Measures for April<br />
The different measures are interpreted to identify<br />
which simulations are judged more similar to the<br />
observed spatial pattern. All of the simulations<br />
for April had minimal bias. R 2 correlation is best<br />
between the observation <strong>and</strong> simulation 09. The<br />
RMSE values are all around 2.5% V/V, apart<br />
from simulations 01-03 <strong>and</strong> 07. RMSE finds<br />
simulation 09 to be the best match. IMSE with<br />
pixel value as the event finds simulations 04 <strong>and</strong><br />
10 to be the best, with 08 also close. With local<br />
Table 1. Comparison of the observed spatial<br />
pattern to 10 different model simulations for two<br />
occasions in 1996 (Apr, Oct). Bias <strong>and</strong> RMSE<br />
values are in % V/V. The most similar measures<br />
are in bold, other similar ones are in italic.<br />
13-Apr-96<br />
Sim. Bias Corr. RMSE IMSE (pv) IMSE (lv)<br />
01 -0.48 0.29 2.84 3087 1666<br />
02 -0.51 0.19 2.84 3745 2190<br />
03 -0.54 0.12 2.86 4664 2807<br />
04 0.19 0.29 2.53 2185 1278<br />
05 0.15 0.31 2.51 2659 1152<br />
06 0.11 0.31 2.52 3148 1269<br />
07 0.67 0.00 3.60 5442 1730<br />
08 0.02 0.25 2.56 2271 1212<br />
09 0.16 0.40 2.39 4313 1058<br />
10 0.04 0.27 2.55 2146 1266<br />
25-Oct-96<br />
Sim. Bias Corr. RMSE IMSE (pv) IMSE (lv)<br />
01 5.96 0.25 6.98 2172 3778<br />
02 5.51 0.38 6.60 1843 4222<br />
03 4.62 0.46 5.99 1814 4369<br />
04 6.24 0.32 7.19 1983 3731<br />
05 5.85 0.44 6.83 1845 4044<br />
06 5.04 0.52 6.23 1872 4236<br />
07 4.13 0.53 6.13 1678 4380<br />
08 4.30 0.40 5.49 1773 3946<br />
09 5.61 0.42 6.84 1668 4286<br />
10 4.73 0.45 5.81 1858 3736<br />
1078
variance as the event, simulations 05 <strong>and</strong> 09 are<br />
best. Further inspection of 04, 05 <strong>and</strong> 08-10<br />
would be recommended, as these were judged<br />
similar by multiple measures.<br />
4.4. Similarity Measures for October<br />
The simulations for October are all biased, over<br />
predicting by about 4-5% V/V. However,<br />
reasonable R 2 was present with 06 <strong>and</strong> 07. The<br />
RMSE measure, which incorporates bias <strong>and</strong><br />
other errors, found simulation 08 as the best.<br />
IMSE with pixel value suggests 07-09 to be most<br />
similar, while IMSE with local variance finds 04<br />
or 10. On these findings, further inspection of 07-<br />
10 would be suggested.<br />
4.5. Visually Assessing the IMSE Measures<br />
Figure 2 provides a visual representation of how<br />
the IMSE weightings influence the st<strong>and</strong>ard MSE<br />
similarity measure. In the first column, there are<br />
two simulations shown for the soil moisture<br />
content on 13 April 1996. The other columns<br />
contain the IMSE weighted spatial patterns that<br />
are subsequently compared to produce the IMSE<br />
measures in Table 1. By visually inspecting the<br />
first column of spatial patterns, it appears that<br />
simulation 09 does a better job than 10 of<br />
reproducing the linear high-moisture feature.<br />
These high pixel values occur less frequently <strong>and</strong><br />
thus receive a high weighting in the second<br />
column. In simulation 10, there is not such a<br />
distinct difference between the high <strong>and</strong> low pixel<br />
values, resulting in a more even weighting across<br />
the spatial pattern. For local variance, both<br />
simulations are enhanced. The variable areas are<br />
given higher weightings than the homogeneous<br />
background. This appears a more suitable event<br />
when comparing these spatial patterns, as it helps<br />
discern the feature (i.e. the pixels with more<br />
information) from the background. This is a<br />
logical characteristic of spatial patterns to use for<br />
weighting, as human vision is often drawn to<br />
these areas of larger variation. This is similar to<br />
the use of ‘edges’ by Topper <strong>and</strong> Jernigan [1989],<br />
which are a measurement of local gradient widely<br />
used in image processing.<br />
The degree of weighting for local variance that is<br />
applied to the spatial patterns is more pronounced<br />
for simulation 09 than 10. In 09, there are many<br />
areas with low variance, but only a few with very<br />
high variance. As such, the few pixels are heavily<br />
enhanced, while the remainder are heavily<br />
reduced. For 10, there is certainly more<br />
weighting for the areas with high variance, but the<br />
distribution of variances is not as extreme. As a<br />
result, the feature is not as greatly enhanced,<br />
which visually appears more correct.<br />
As with the st<strong>and</strong>ard MSE statistic, even when<br />
two spatial patterns look quite similar (e.g. the<br />
OBSERVED SPATIAL PATTERN - 13 APRIL 1996<br />
ORIGINAL (UNWEIGHTED) IMSE (PIXEL VALUE) IMSE (LOCAL VARIANCE)<br />
SIMULATION 09<br />
RMSE = 2.39% V/V IMSE = 4313<br />
IMSE = 1058<br />
SIMULATION 10<br />
RMSE = 2.55% V/V<br />
IMSE = 2146 IMSE = 1266<br />
Figure 2. Spatial patterns of soil moisture from 13 April 1996. The observed spatial patterns are shown for<br />
two alternative simulations. The original <strong>and</strong> IMSE weighted spatial patterns are shown to help interpret the<br />
meaning of similarity measures. Comparable grey scales are used, with darker greys denoting higher values.<br />
Values have been placed into 20 equal interval categories.<br />
1079
IMSE pixel values for observed <strong>and</strong> simulation<br />
09), if there is not local agreement between the<br />
pixel values then the measure states that they are<br />
dissimilar. This happens here, with simulation 09<br />
having a similarity measure that is twice<br />
simulation 10, although it appears more similar.<br />
At present, the IMSE measure does not account<br />
for minor shifts between pixels, although this<br />
could be one avenue for improvement.<br />
5. DISCUSSION<br />
The use of IMSE with hydrological spatial<br />
patterns relies on choosing an event that can<br />
discern the features from the background. Figure<br />
2 illustrates that local variance can be useful for<br />
discerning the features of interest within an<br />
otherwise homogeneous spatial pattern. This is<br />
also a logical surrogate for human vision, which<br />
uses variation as a means to identify features<br />
[Topper <strong>and</strong> Jernigan, 1989].<br />
The nature of the spatial pattern <strong>and</strong> its<br />
complexity can also make a large difference to the<br />
use of the IMSE measure. In spatial patterns with<br />
a larger extent, the number of different events<br />
occurring can be far greater (due to having many<br />
more pixels). Here, the choice of the number of<br />
categories will influence how well the measure<br />
applies the weightings. Too few categories will<br />
result in rare events being lumped together with<br />
common events, whereas too many categories can<br />
lead to every event being treated as rare.<br />
The application for which this method was<br />
initially developed looks at comparing a distorted<br />
image with the original. Both images therefore<br />
have a similar distribution of values (i.e.<br />
histogram), with some minor changes in the<br />
histogram of the distorted image. If there is a<br />
difference between the histograms for a common<br />
pixel value, this difference will be less influential.<br />
However, if the difference occurs in a rare pixel<br />
value, the weighting highlights the difference.<br />
With applying this same idea to hydrological<br />
spatial patterns, the original image is synonymous<br />
with the observed spatial pattern, while the<br />
simulations should be like distortions of the<br />
original. In reality, the simulations are attempts at<br />
recreating the observations based on an<br />
underst<strong>and</strong>ing of the processes <strong>and</strong> forcings of the<br />
hydrological system. This can result in very<br />
different histograms for the observed <strong>and</strong><br />
simulated spatial patterns. When the IMSE<br />
weights are calculated for each pixel value, there<br />
can be a large difference between the weightings<br />
applied to the observed <strong>and</strong> simulated spatial<br />
patterns. As such, IMSE appears more suitable<br />
for comparison when the spatial patterns have<br />
similar distributions of values. This measure<br />
could also be modified to reduce the impact of<br />
large differences. By calculating the mean<br />
absolute error rather than mean squared error, the<br />
impact of large residuals would be reduced.<br />
The IMSE measure highlights the possibility for<br />
using weightings to make a st<strong>and</strong>ard MSE<br />
statistic compare something different. While<br />
Shannon’s self-information has been used to<br />
define the weights here, other measures (e.g.<br />
terrain related measures) could alternatively be<br />
used to define informative locations.<br />
6. CONCLUSION<br />
This brief look at a method for comparing<br />
distorted images provides a number of ideas for<br />
the comparison of spatial patterns in hydrology.<br />
This method works predominantly in the<br />
measurement domain, but by using local variance<br />
as the event, some spatial characteristics can be<br />
incorporated. Further application of this method<br />
to spatial patterns from hydrological models will<br />
help in assessing its suitability for comparing<br />
spatial patterns.<br />
7. REFERENCES<br />
Grayson, R.B. <strong>and</strong> G. Blöschl eds., Spatial Patterns in<br />
Catchment Hydrology: Observations <strong>and</strong> <strong>Modelling</strong>,<br />
Cambridge University Press, 404 pp., Cambridge,<br />
2000.<br />
Grayson, R.B., G. Blöschl, A.W. Western <strong>and</strong> T.A. McMahon,<br />
Advances in the use of observed spatial patterns of<br />
catchment hydrological response, Advances in Water<br />
Resources, 25(8-12), 1313-1334, 2002.<br />
Jetten, V., G. Govers <strong>and</strong> R. Hessel, Erosion models: quality<br />
of spatial predictions, Hydrological Processes, 17(5),<br />
887-900, 2003.<br />
Scheibe, T.D., Characterization of the spatial structuring of<br />
natural porous media <strong>and</strong> its impacts on subsurface<br />
flow <strong>and</strong> transport, PhD thesis, Department of Civil<br />
Engineering, Stanford University, 1993.<br />
Tompa, D., J. Morton <strong>and</strong> E. Jernigan, Perceptually based<br />
image comparison, <strong>International</strong> Conference on<br />
Image Processing, Vancouver, BC, Canada,<br />
September 10-13, 2000.<br />
Topper, T.N. <strong>and</strong> M.E. Jernigan, On the informativeness of<br />
edges, IEEE <strong>International</strong> Conference on Systems,<br />
Man <strong>and</strong> Cybernetics, Cambridge, MA, USA, 1989.<br />
Weal<strong>and</strong>s, S.R., R.B. Grayson <strong>and</strong> J.P. Walker, Hydrologic<br />
Model Assessment from Automated Spatial Pattern<br />
Comparison Techniques, <strong>International</strong> Congress on<br />
<strong>Modelling</strong> <strong>and</strong> Simulation, Townsville, Australia, 14-<br />
17 July, 2003.<br />
Weal<strong>and</strong>s, S.R., R.B. Grayson <strong>and</strong> J.P. Walker, Quantitative<br />
comparison of spatial patterns for hydrological model<br />
assessment - some promising approaches, Advances<br />
in Water Resources, submitted.<br />
Western, A.W. <strong>and</strong> R.B. Grayson, Soil Moisture <strong>and</strong> Runoff<br />
Processes at Tarrawarra, In: Grayson, R. <strong>and</strong> G.<br />
Blöschl eds., Soil Moisture <strong>and</strong> Runoff Processes at<br />
Tarrawarra, Cambridge University Press, p. 209-246,<br />
Cambridge, 2000.<br />
1080
Reduced Models of the Retention of Nitrogen in<br />
Catchments<br />
K. Wahlin 1 , D. Shahsavani 1 , A. Grimvall 1 , A. J. Wade 2 , D. Butterfield 2 , <strong>and</strong> H. P. Jarvie 3<br />
1) Department of Mathematics, Linköping University, 58183 Linköping, Sweden<br />
2) Aquatic Environments Research Centre, School of Human <strong>and</strong> <strong>Environmental</strong> Sciences, The University of<br />
Reading, UK, RG6 6AB.<br />
3) Centre for Ecology <strong>and</strong> Hydrology, Wallingford, UK, OX10 8BB.<br />
Abstract: Process-oriented models of the retention of nitrogen in catchments are by necessity rather<br />
complex. We introduced several types of ensemble runs that can provide informative summaries of<br />
meteorologically normalised model outputs <strong>and</strong> also clarify the extent to which such outputs are related to<br />
various model parameters. Thereafter we employed this technique to examine policy-relevant outputs of the<br />
catchment model INCA-N. In particular, we examined how long it will take for changes in the application of<br />
fertilisers on cultivated l<strong>and</strong> to affect the predicted riverine loads of nitrogen. The results showed that the<br />
magnitude of the total intervention effect was influenced mainly by the parameters governing the turnover of<br />
nitrogen in soil, whereas the temporal distribution of the water quality response was determined primarily by<br />
the hydromechanical model parameters. This raises the question of whether the soil nitrogen processes<br />
included in the model are elaborate enough to correctly explain the widespread observations of slow water<br />
quality responses to changes in agricultural practices.<br />
Keywords: Model reduction; Ensemble runs; Catchment; Nitrogen; Retention.<br />
1. INTRODUCTION<br />
Numerous process-oriented deterministic models<br />
have been developed to explain <strong>and</strong> predict the<br />
flow of nitrogen through catchments (e.g.,<br />
Arheimer & Br<strong>and</strong>t, 1998; Heng & Nikolaidis,<br />
1998; Kroes & Roelsma, 1998; Whitehead et al.,<br />
1998a,b; Refsgaard et al., 1999). In general, such<br />
models can satisfactorily describe prevailing<br />
spatial distributions of riverine loads of nitrogen.<br />
Also, they can usually clarify most of the seasonal<br />
variation <strong>and</strong> a considerable fraction of the shortterm<br />
temporal fluctuations in the nitrogen loads.<br />
However, it is less certain how well the models can<br />
predict several-year-long lags in the water quality<br />
response to interventions in a drainage area. In<br />
addition, the complexity of the cited models can<br />
make it difficult to comprehend the relationships<br />
between model parameters <strong>and</strong> the predicted<br />
impact of interventions.<br />
The present study was devoted to model reductions<br />
that can help extract the essence of complex<br />
process-oriented models driven by meteorological<br />
data. Specifically, different types of ensemble runs<br />
were introduced in which natural fluctuations in the<br />
model output were suppressed by computing the<br />
average output for a collection of artificially<br />
generated time series of rainfall <strong>and</strong> temperature<br />
data. Some of these ensemble runs were designed<br />
to elucidate the fate of nitrogen applied on the soil<br />
surface. Another group of simulation experiments<br />
aimed to clarify water travel times in the<br />
unsaturated <strong>and</strong> saturated zones.<br />
The above-mentioned techniques were used to<br />
determine how changes in fertiliser applications<br />
affect the riverine loads of inorganic nitrogen<br />
predicted by the Integrated Nitrogen in Catchments<br />
(INCA-N) model (Whitehead et al., 1998b). Time<br />
series of meteorologically normalised nitrogen<br />
loads were computed, <strong>and</strong> the results were<br />
summarised in impulse-response functions. We<br />
also examined which model parameters had the<br />
greatest influence on the total response <strong>and</strong> the<br />
time lag between intervention <strong>and</strong> response.<br />
1081
2. THE INCA-N MODEL<br />
The INCA-N is a semi-distributed, process-based<br />
model of the flow of water <strong>and</strong> nitrogen through<br />
catchments (Wade et al., 2002). INCA-N simulates<br />
the key factors <strong>and</strong> processes that affect the amount<br />
of NO 3 <strong>and</strong> NH 4 stored in the soil <strong>and</strong> groundwater<br />
systems, <strong>and</strong> it feeds the output from these systems<br />
into a multi-reach river model. The final output of<br />
the INCA-N model consists of daily estimates of<br />
water discharge <strong>and</strong> NO 3 <strong>and</strong> NH 4 concentrations<br />
in stream water at discrete points along the main<br />
channel of the river.<br />
INCA-N takes the following input fluxes into<br />
account: atmospheric deposition of ammonium <strong>and</strong><br />
nitrate (wet <strong>and</strong> dry), application of NO 3 <strong>and</strong> NH 4<br />
fertlisers, mineralisation of organic matter<br />
(yielding NH 4 ), nitrification (yielding NO 3 ), <strong>and</strong><br />
nitrogen fixation. From these data various output<br />
fluxes (plant uptake, immobilisation, <strong>and</strong><br />
denitrification) are substracted before the amount<br />
available for stream output is calculated.<br />
Whenever relevant, inputs <strong>and</strong> outputs are<br />
differentiated by l<strong>and</strong>scape type <strong>and</strong> varied<br />
according to environmental conditions (soil<br />
moisture <strong>and</strong> temperature). The model also<br />
simulates the flow of water from different kinds of<br />
l<strong>and</strong> use through the plant/soil system in order to<br />
deliver the nitrogen load to the river system. The<br />
load is then routed downstream, after accounting<br />
for direct effluent discharges, <strong>and</strong> in-stream<br />
nitrification <strong>and</strong> denitrification.<br />
3. STUDY AREA<br />
The empirical data we used were collected in the<br />
Tweed River Basin which is located in Scotl<strong>and</strong><br />
(4300 km 2 ) <strong>and</strong> Engl<strong>and</strong> (680 km 2 ). The l<strong>and</strong>phase<br />
data included information about l<strong>and</strong> use in<br />
23 sub-basins, whereas the meteorological inputs<br />
(air temperature <strong>and</strong> precipitation) were assumed<br />
to be the same for the entire Tweed Basin (Jarvie et<br />
al., 2002).<br />
The catchment of the River Tweed consists of a<br />
horse-shoe-shaped rim of hills composed of older,<br />
harder rocks which surround a relatively flat basin<br />
of younger rocks covered with a thick layer of<br />
glacial debris. The l<strong>and</strong> cover ranges from heather<br />
moorl<strong>and</strong>s <strong>and</strong> rough grazing on the hills,<br />
improved pastures on the lower slopes to arable<br />
l<strong>and</strong> in the lowl<strong>and</strong>s, <strong>and</strong> the average application of<br />
inorganic nitrogen on cultivated l<strong>and</strong> is 106<br />
kg/ha/yr. Average rainfall is about 650 mm in the<br />
lower reaches of the catchment <strong>and</strong> considerably<br />
higher in the highl<strong>and</strong>s. The base-flow index is<br />
estimated to approximately 0.5 for all sub-basins.<br />
4. SIMULATION METHODS<br />
4.1. Notation<br />
From a mathematical point of view, the INCA-N<br />
model <strong>and</strong> other deterministic substance transport<br />
models can be regarded as functions<br />
y = f(x)<br />
The output is a scalar or a vector of moderately<br />
high dimension, whereas x can contain a very large<br />
number of components. We introduce the notation<br />
x(t, z) to show that at least some of the components<br />
of x can depend on time (t) <strong>and</strong> location (z).<br />
Moreover, we write<br />
z<br />
j<br />
p z k<br />
to indicate that z j is located upstream of z k <strong>and</strong><br />
y( t,<br />
zk<br />
) = f ( x(<br />
s,<br />
z<br />
j<br />
), s ≤ t,<br />
z<br />
j<br />
p zk<br />
)<br />
to indicate that the output at time t is a function of<br />
both current <strong>and</strong> previous inputs to all sub-basins<br />
upstream of the location under consideration.<br />
Different types of model inputs are separated by<br />
setting<br />
where<br />
x(s, z) = (u(t 0 , z), v(s, z), w(s, z), θ(z))<br />
u(t 0 , z) defines the state of the system at time t 0 ;<br />
v(s, z) is a vector representing the anthropogenic<br />
forcing of the system;<br />
w(s, z) is a vector representing the meteorological<br />
forcing of the system; <strong>and</strong><br />
θ(z) is a vector of model parameters.<br />
The vector u(t 0 , z) contains information about<br />
water content <strong>and</strong> concentrations of different<br />
nitrogen species in different parts of the system at<br />
the onset of the observation period. Information<br />
about fertiliser use can exemplify the content of<br />
v(s, z), <strong>and</strong> w(s, z) can contain data on air<br />
temperature <strong>and</strong> precipitation. The vector θ(z)<br />
includes hydrogeological parameters <strong>and</strong> rate<br />
coefficients for nitrogen transformation processes.<br />
Unless otherwise stated, we regard riverine loads<br />
of inorganic nitrogen (NO 3 -N + NH 4 -N) as the<br />
primary response variable.<br />
4.2. Meteorological normalisation<br />
Meteorological normalisation aims to remove or<br />
suppress the impact that r<strong>and</strong>om variation in<br />
weather conditions has on the model output. We<br />
performed so-called conditional normalisation, i.e.,<br />
the predicted riverine load of inorganic nitrogen<br />
was averaged over different meteorological<br />
1082
́<br />
́<br />
forcings, while the anthropogenic inputs were fixed<br />
(Grimvall et al., 2001; Forsman & Grimvall,<br />
2003).<br />
A set {w i , i = 1, … , n} of artificial meteorological<br />
inputs with approximately the same statistical<br />
properties as the original data series was created by<br />
resampling blocks of observed weather records<br />
(Lahiri, 1999). Specifically, 30-day-long blocks<br />
were sampled, each of which was r<strong>and</strong>omly<br />
selected from the different observation years with a<br />
shift of up to 15 days in the Julian day. Thereafter,<br />
the model was run for each element of {w i , i = 1,<br />
… , n}, <strong>and</strong> the mean output<br />
n<br />
1<br />
y( t,<br />
zk<br />
) = ∑ f (( u,<br />
v,<br />
wi<br />
, ), s ≤ t,<br />
z<br />
j<br />
p zk<br />
)<br />
n<br />
i=<br />
1<br />
was computed for each time point t. A total of 400<br />
replicates of the meteorological forcing was found<br />
to be sufficient to remove the weather-dependent<br />
variation in the model output.<br />
4.3. Ensemble runs mimicking the transport of<br />
labelled nitrogen species<br />
Laboratory <strong>and</strong> field experiments involving<br />
labelled nitrogen species have contributed<br />
substantially to current knowledge regarding the<br />
turnover of nitrogen in soil (e.g., Shen et al.,<br />
1989). Any process-oriented model that can<br />
accommodate user-defined time series of fertiliser<br />
inputs can be employed to mimic important<br />
features of such experiments.<br />
Let v(s, ⋅) designate a given fertilisation scheme<br />
<strong>and</strong> let ∆v(s, ⋅) denote a minor change in that<br />
scheme. We can then compute the difference<br />
∆ f = f ( u,<br />
v + ∆v,<br />
w,<br />
) − f ( u,<br />
v,<br />
w,<br />
) (1) ́<br />
for each time point t. If ∆v(s, ⋅) = 0 for the second<br />
year <strong>and</strong> onwards, such calculations can provide<br />
information about the fate of the nitrogen applied<br />
during the first year. Moreover, we can compute<br />
impulse response functions for the impact of<br />
fertiliser application on riverine loads of nitrogen.<br />
4.4. Ensemble runs mimicking the transport of<br />
inert substances <strong>and</strong> water<br />
If all processes involving transformation or<br />
immobilisation of nitrogen are switched off, the<br />
ensemble runs mentioned in the previous section<br />
can provide information about the travel time of an<br />
inert substance. In that case the flow of water is the<br />
only transport mechanism, thus such ensemble runs<br />
also reveal the travel times of water through the<br />
unsaturated <strong>and</strong> saturated zones. In particular, it<br />
can be established whether the nitrogen delivered<br />
from l<strong>and</strong> to surface water is younger or older than<br />
the water reaching the stream.<br />
5. RESULTS<br />
Ensemble runs were made for a variety of systems<br />
ranging from a soil column to entire catchments.<br />
The simplest systems were defined as catchments<br />
with a single sub-basin <strong>and</strong> a single l<strong>and</strong>-use<br />
category. Furthermore, in some of the ensemble<br />
runs, all in-stream processes, including abstraction<br />
of river water <strong>and</strong> direct emissions to the river,<br />
were switched off. The base-flow index was varied<br />
from zero to one in order to highlight the role of<br />
groundwater in the riverine loads of nitrogen.<br />
5.1. Nitrogen retention in simple systems<br />
Figure 1 shows the meteorologically normalised<br />
riverine loads of inorganic nitrogen obtained from<br />
the INCA-N model to simulate a system consisting<br />
of a single sub-basin comprising only arable l<strong>and</strong>.<br />
The weather-dependent interannual variation in<br />
riverine loads was removed, but, despite that, the<br />
values obtained for the different years are not<br />
identical due to memory effects of the initial state<br />
of the simulated system.<br />
Figure 1. Meteorologically normalised riverine<br />
Normalised load of<br />
inorganic N (kg/ha/yr)<br />
120<br />
100<br />
80<br />
60<br />
40<br />
20<br />
0<br />
1 2 3 4 5 6 7<br />
Year<br />
loads of inorganic nitrogen for an artificial<br />
catchment consisting of a single sub-basin<br />
comprising only arable l<strong>and</strong> <strong>and</strong> receiving a<br />
constant level of ammonium <strong>and</strong> nitrate fertliser<br />
(combined total 106 kg N/ha/yr). The base-flow<br />
index was set to zero, <strong>and</strong> all in-stream processes<br />
in the INCA-N model were switched off.<br />
When the application of fertilisers was increased<br />
by 1% the first year, the values of ∆ f (Eq. 1) were<br />
positive for a sequence of years. Figure 2 illustrates<br />
the delay in the water quality response in a system<br />
with base-flow index zero. Also, it can be seen that<br />
(due to removal of nitrogen through harvesting <strong>and</strong><br />
denitrification) the cumulated increase in riverine<br />
loads was considerably smaller than the increase in<br />
fertiliser application.<br />
As expected, the time lag in the water quality<br />
response increased with the base-flow index due to<br />
the increased influence of groundwater (Figure 3).<br />
However, the total intervention effect was<br />
unchanged, because the INCA-N model does not<br />
1083
include any transformation or immobilisation of<br />
nitrogen in the saturated zone.<br />
Cumulated response<br />
Relative frequency<br />
Figure 2. Predicted response of riverine loads of<br />
inorganic nitrogen to an impulse (1% increase) in<br />
fertiliser application during the first year of the<br />
study period. The diagrams show the following:<br />
(top) the ratio of the cumulated increase in riverine<br />
loads to the increase in fertiliser application;<br />
(bottom) the relative frequency of travel times for<br />
the applied nitrogen fertiliser. The simulated<br />
system was the same as in Figure 1.<br />
Cumulated response<br />
Relative frequency<br />
1.0<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0.0<br />
1.0<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0.0<br />
1.0<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0.0<br />
0.20<br />
0.15<br />
0.10<br />
0.05<br />
0.00<br />
1<br />
1 2 3 4 5 6 7<br />
Year<br />
1 2 3 4 5 6 7<br />
Year<br />
1 4 7 10 13 16 19 22 25 28<br />
Year<br />
4<br />
7<br />
10<br />
13<br />
16<br />
Year<br />
19<br />
22<br />
25<br />
28<br />
Figure 3. Predicted response of riverine loads of<br />
inorganic nitrogen to an impulse in fertiliser<br />
application during the first year. The base-flow<br />
index was set to 1. All other conditions were the<br />
same as in Figure 2.<br />
5.2. Water residence times in simple systems<br />
Ensemble runs of the type defined in section 4.4<br />
were undertaken to elucidate the water residence<br />
times in the unsaturated <strong>and</strong> saturated zones. The<br />
results obtained for two simple systems are<br />
illustrated in Figure 4. It is especially noticeable<br />
that, on average, the inorganic nitrogen reaching<br />
the river has a shorter travel time than the water in<br />
which it is dissolved. This is due to the fact that<br />
denitrification <strong>and</strong> plant uptake result in<br />
preferential removal (or uptake) of the nitrogen<br />
that has unusually long residence times.<br />
Relative frequency<br />
Relative frequency<br />
1.0<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0.0<br />
0.20<br />
0.15<br />
0.10<br />
0.05<br />
0.00<br />
1<br />
1 2 3 4 5 6 7<br />
Year<br />
4<br />
Water<br />
Water<br />
7<br />
10<br />
Year<br />
Inorg. N<br />
13<br />
Inorg. N<br />
16<br />
19<br />
Figure 4. Predicted relative frequencies of travel<br />
times for inorganic nitrogen <strong>and</strong> water in the<br />
systems defined in Figures 2 <strong>and</strong> 3. The base-flow<br />
index was set to zero (top) or one (bottom).<br />
5.3. Sensitivity analysis of the predicted<br />
response to changes in fertiliser application<br />
The INCA-N model does not include any<br />
transformation or immobilisation of nitrogen in the<br />
saturated zone. Under such circumstances it is<br />
obvious that long residence times in groundwater<br />
will cause long time lags in the water quality<br />
response to l<strong>and</strong>-use interventions in the drainage<br />
area. It is also clear that high rates of<br />
denitrification in soil <strong>and</strong> uptake by plants will<br />
reduce the total intervention effect. However, it is<br />
not as apparent how the parameters governing<br />
nitrogen turnover in soil influence the travel time<br />
for the nitrogen that is leached from l<strong>and</strong> to surface<br />
water.<br />
Figure 5 illustrates that, in the INCA-N model, the<br />
length of the delay in water quality response is<br />
1084
independent of the mineralisation rate. Further<br />
information about the sensitivity of predicted<br />
intervention effects to selected model parameters is<br />
given in Table 1. The total effect values in the<br />
table represent the percentage of the nitrogen<br />
applied on arable l<strong>and</strong> that (eventually) reaches the<br />
river. The relative importance of (almost) direct<br />
response is expressed as p 1 +p 2 , where {p i , i = 1, 2,<br />
…} is the probability distribution of the time lags<br />
for the nitrogen that reaches the river.<br />
Relative frequency<br />
Figure 5. Predicted travel times of nitrogen in a<br />
system consisting of a single sub-basin comprising<br />
only arable l<strong>and</strong> <strong>and</strong> with a base-flow index of 0.5.<br />
Mineralisation rates (kg N ha -1 yr -1 ) are indicated<br />
in the graph.<br />
Table 1. Total effect of the intervention <strong>and</strong><br />
relative importance of almost direct response to<br />
changes in fertiliser application in relation to the<br />
rates of different natural processes (M,<br />
mineralisation; U, plant uptake; D, denitrification).<br />
The base-flow index was 0.5 in all model runs.<br />
M<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0.0<br />
(kg N<br />
ha -1 yr -1 )<br />
1<br />
4<br />
Process rate<br />
U<br />
(m d -<br />
1 )<br />
7<br />
0.50 0.25<br />
10<br />
D<br />
(m d -1 )<br />
Year<br />
13<br />
16<br />
Total<br />
effect<br />
(%)<br />
19<br />
Almost<br />
direct<br />
response<br />
(p 1 + p 2)<br />
0.5 0.02 0.001 47 0.46<br />
0.25 0.02 0.001 47 0.46<br />
0.5 0.01 0.001 59 0.54<br />
0.25 0.01 0.001 59 0.54<br />
0.5 0.02 0.005 40 0.42<br />
0.25 0.02 0.005 40 0.42<br />
0.5 0.01 0.005 49 0.48<br />
0.25 0.01 0.005 49 0.48<br />
5.4. Simulations of catchment-scale retention<br />
When the INCA-N model is used to simulate the<br />
flow of water <strong>and</strong> nitrogen through a whole<br />
catchment, the total delivery of nitrogen to the<br />
river is computed by summing all inputs to the<br />
different parts of the river. The load of inorganic<br />
nitrogen at the mouth of the river can be<br />
considerably smaller due to in-stream processes<br />
(Behrendt & Opitz, 1999). In addition, point<br />
emissions, atmospheric deposition on water<br />
surfaces, <strong>and</strong> abstraction of water can have an<br />
impact on the riverine loads of nitrogen <strong>and</strong> the<br />
response to interventions in the drainage area.<br />
There are no major lakes in the Tweed Basin,<br />
hence in-stream processes will have only a small<br />
effect on the total travel times of nitrogen <strong>and</strong><br />
water through the catchment. Figure 6 illustrates<br />
that the in-stream processes also have only a very<br />
small impact on the cumulated response of riverine<br />
loads to an impulse in fertiliser application.<br />
Cumulated response<br />
1.0<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0.0<br />
Figure 6. Predicted response in riverine loads of<br />
nitrogen to an impulse (1% increase) in fertiliser<br />
application in the entire Tweed Basin during the<br />
first year of the study period. The two diagrams<br />
were derived from ensemble runs in which instream<br />
processes <strong>and</strong> direct inputs to water were<br />
switched off (top) or all processes in the INCA-N<br />
model were active (bottom).<br />
6. DISCUSSION<br />
Switched off<br />
Active<br />
1 2 3 4 5 6 7<br />
Year<br />
This study shows that ensemble runs involving<br />
artificially generated meteorological inputs can be<br />
employed to extract model features that might<br />
otherwise be hidden by the total variation in the<br />
model output. Introducing ensemble runs clarified<br />
how water <strong>and</strong> nitrogen travel times in the<br />
saturated <strong>and</strong> unsaturated zones contribute to time<br />
lags in the river response to interventions in the<br />
drainage area.<br />
The simulation techniques described here facilitate<br />
comparative studies of different catchment models.<br />
Also, ensemble runs provide useful input to<br />
sensitivity analyses of model outputs. The results<br />
obtained with the INCA-N model indicate that the<br />
total intervention effect was influenced mainly by<br />
the parameters governing the turnover of nitrogen<br />
in soil, whereas the temporal distribution of the<br />
water quality response was determined primarily<br />
by the hydromechanical model parameters.<br />
1085
Moreover, we found that, almost regardless of the<br />
model parameters, there was a relatively rapid<br />
response to interventions in the drainage area. This<br />
seems to contradict the absence of an unambiguous<br />
water quality response in many Eastern European<br />
river basins, where agricultural practices changed<br />
dramatically in the early 1990s (Stålnacke et al.,<br />
2003).<br />
Two potential explanations for our observations<br />
call for further discussion. The first of these is<br />
hydrogeological in nature <strong>and</strong> concerns the fact<br />
that the groundwater residence times are rather<br />
long in many river basins in the Baltic Republics<br />
<strong>and</strong> Pol<strong>and</strong>, <strong>and</strong> the monitoring programmes may<br />
have failed to detect the water quality changes that<br />
have actually taken place. The second explanation<br />
is directly related to the INCA-N model. Analyses<br />
of 15 N-labelled fertiliser residues in the soil have<br />
clearly demonstrated that the dominating pathway<br />
of inorganic nitrogen in soil includes uptake by<br />
plants <strong>and</strong> subsequent mineralisation of plant<br />
residues (Shen et al., 1989), <strong>and</strong> these conditions<br />
can apparently prolong the travel time of nitrogen<br />
in the unsaturated zone. However, INCA-N is<br />
unable to model such decoupling phenomena.<br />
7. ACKNOWLEDGEMENTS<br />
The authors are grateful to the Swedish<br />
<strong>Environmental</strong> Protection Agency <strong>and</strong> the Swedish<br />
Research Council for financial support, <strong>and</strong> to the<br />
Scottish Environment Protection Agency <strong>and</strong> the<br />
NERC L<strong>and</strong> Ocean Interaction Study for the<br />
provision of data.<br />
8. REFERENCES<br />
Arheimer, B. <strong>and</strong> Br<strong>and</strong>t, M. 1998. <strong>Modelling</strong><br />
nitrogen transport <strong>and</strong> retention in the catchments<br />
of southern Sweden, Ambio, 27, 471-480.<br />
Behrendt, H. <strong>and</strong> Opitz, D. 1999. Retention of<br />
nutrients in river systems: dependence on specific<br />
runoff <strong>and</strong> hydraulic load. Hydrobiol. 410, 111-<br />
122.<br />
Forsman, Å. <strong>and</strong> Grimvall, A. 2003. Reduced<br />
models for efficient simulation of spatially<br />
integrated outputs of one-dimensional substance<br />
transport models. Environ. Modell. Softw., 18,<br />
319-327.<br />
Grimvall, A., Wackernagel, H., <strong>and</strong> Lajaunie, C.<br />
2001. Normalisation of environmental quality data.<br />
In: L.M. Hilty, P.W. Gilgen (eds) "Sustainability<br />
in the Information Society", pp. 581-590,<br />
Metropolis-Verlag, Marburg.<br />
Heng, H.H. <strong>and</strong> Nikolaidis, N.P. 1998. Modeling<br />
of nonpoint source pollution of nitrogen at the<br />
watershed scale. J. Amer. Water Resour. Assoc.,<br />
34, 359-374.<br />
Jarvie, H. P., Wade, A. J., Butterfield, D.,<br />
Whitehead, P. G., Tindall, C. I., Virtue, W. A.,<br />
Dryburgh, W. <strong>and</strong> McCraw, A. 2002. <strong>Modelling</strong><br />
nitrogen dynamics <strong>and</strong> distributions in the River<br />
Tweed, Scotl<strong>and</strong>: an application of the INCA<br />
model. Hydrol. Earth Sys. Sci., 6, 443-453.<br />
Kroes, J.G. <strong>and</strong> Roelsma, J. 1998. ANIMO 3.5:<br />
user's guide for the ANIMO version 3.5 nutrient<br />
leaching model. Wageningen, SC-DLO, Techn.<br />
Rep. 46, 98 pp.<br />
Lahiri, S.N. 1999. Theoretical comparisons of<br />
block bootstrap methods. Ann. Stat. , 27, 386-404.<br />
Refsgaard, J.C., Thorsen, M., Jensen, J.B.,<br />
Kleeschulte, S., <strong>and</strong> Hansen, S. 1999. Large scale<br />
modelling of groundwater contamination from<br />
nitrate leaching, J. Hydrol. 221, 117-140.<br />
Shen, S.M., Hart, P.B.S., Powlson, D.S., <strong>and</strong><br />
Jenkinson, D.S. 1989. The nitrogen cycle in the<br />
Broadbalk Wheat Experiment:<br />
15 N labeled<br />
fertilizer residues in the soil <strong>and</strong> in the soil<br />
microbial biomass. Soil Biol. Biochem., 21, 529-<br />
533.<br />
Stålnacke, P., Grimvall, A., Libiseller, C., Laznik,<br />
M., <strong>and</strong> Kokorite, I. 2003. Trends in nutrient<br />
concentrations in Latvian rivers <strong>and</strong> the response to<br />
the dramatic change in agriculture. J. Hydrol. 283,<br />
184-205.<br />
Wade, A. J., Dur<strong>and</strong>, P., Beaujouan, V., Wessel,<br />
W. W., Raat, K. J., Whitehead, P. G., Butterfield,<br />
D., Rankinen, K. <strong>and</strong> Lepisto, A. 2002. A nitrogen<br />
model for European catchments: INCA, new model<br />
structure <strong>and</strong> equations. Hydrol. Earth Sys. Sci., 6,<br />
559-582.<br />
Whitehead, P. G., Wilson, E. J., <strong>and</strong> Butterfield, D.<br />
1998a. A semi-distributed nitrogen model for<br />
multiple source assessments in catchments (INCA):<br />
Model structure <strong>and</strong> process equations. Sci. Tot.<br />
Environ., 210/211, 547-558.<br />
Whitehead, P. G., Wilson, E. J., Butterfield, D.,<br />
<strong>and</strong> Seed, K. 1998b. A semi-distributed integrated<br />
flow <strong>and</strong> nitrogen model for multiple source<br />
assessment in catchments (INCA): Application to<br />
large river basins in South Wales <strong>and</strong> Eastern<br />
Engl<strong>and</strong>. Sci. Tot. Environ., 210/211, 559-583.<br />
1086
The Evaluation of Uncertainty Propagation<br />
into River Water Quality Predictions<br />
to Guide Future Monitoring Campaigns.<br />
V<strong>and</strong>enberghe V.*, Bauwens W.** <strong>and</strong> Vanrolleghem P.A.*<br />
* Ghent University, Department of Applied Mathematics, Biometrics <strong>and</strong> Process Contro , BIOMATH<br />
Coupure Links 653, B-9000 Ghent, Belgium (Email:veronique.v<strong>and</strong>enberghe@biomath.Ugent.be)<br />
**Free University of Brussels, Laboratory of Hydrology <strong>and</strong> Hydraulic Engineering, Pleinlaan 2, B-<br />
1050 Brussels, Belgium<br />
Abstract: To evaluate the future state of river water in view of actual loading or different management<br />
options, water quality models are a useful tool. However, the uncertainty on the model predictions is<br />
sometimes too high to draw proper conclusions. It is of high importance to modellers to minimise the<br />
uncertainty of the model predictions. Therefor different research is needed according to the origin of<br />
the uncertainty. If the uncertainty stems from input data uncertainty or from parameter uncertainty,<br />
more reliable results can be obtained by performing specific measurement campaigns. To guide these<br />
measurement campaigns, an uncertainty analysis can give important information.<br />
In this article an overview of different techniques that give valuable information for the reduction of<br />
input <strong>and</strong> parameter uncertainty is given. The practical case study is the river Dender in Fl<strong>and</strong>ers,<br />
Belgium.<br />
First a global sensitivity analysis shows the importance of the different uncertainty sources. Here it is<br />
seen that the parameters influence the model results more than the input data. Further an analysis in<br />
time <strong>and</strong> space of the uncertainty b<strong>and</strong>s is performed to find differences in uncertainty between certain<br />
periods or places. More measurements are needed during periods or on places with high uncertainty.<br />
This research also shows that finding a link between periods with high uncertainty <strong>and</strong> specific<br />
circumstances (climatological, eco-regional, etc…) can help in gathering data for the calibration of<br />
submodels (eg. diffuse pollution vs. point pollution). The methods can be used for every variable under<br />
study <strong>and</strong> for all kind of rivers but the conclusions made for the practical case study are only applicable<br />
for the Dender.<br />
Keywords: monitoring, optimal experimental design, river water quality modelling, uncertainty analysis<br />
1. INTRODUCTION<br />
In the field of environmental modelling <strong>and</strong><br />
assessment, uncertainty analysis (UA) is a<br />
necessary tool to provide, next to the simulation<br />
results, also a quantitative expression of the<br />
reliability of those results. Next to the expression<br />
of uncertainty bounds on the results, uncertainty<br />
studies have mainly been used to provide insight in<br />
the parameter uncertainty. However, uncertainty<br />
analysis can also be a means to prioritise<br />
uncertainties <strong>and</strong> focus research efforts on the<br />
most problematic points of a model. As such, it<br />
helps to prepare future measurement campaigns<br />
<strong>and</strong> to guide policy decisions.<br />
In this study, the use of an UA as an evaluation<br />
tool is assumed to be applied on an already<br />
calibrated model that can simulate measured data<br />
well but with an unacceptably high uncertainty.<br />
We only consider parameter <strong>and</strong> input uncertainty<br />
that can be minimised by gathering additional data.<br />
Model uncertainty <strong>and</strong> mathematical uncertainty<br />
are not taken into consideration. The aim of this<br />
research is to show how UA can be used to guide<br />
future monitoring campaigns to make model<br />
results more reliable by minimising the parameter<br />
<strong>and</strong> input data uncertainty of the model.<br />
The practical case study is the river Dender in<br />
Fl<strong>and</strong>ers, Belgium.<br />
2. CASE STUDY: THE DENDER BASIN<br />
The Dender river, a tributary of the river Scheldt<br />
in Belgium, drains an area of 1384 km 2 . The main<br />
channel is partly canalised <strong>and</strong> contains 14 sluices.<br />
The river is heavily polluted by domestic,<br />
industrial <strong>and</strong> agricultural pollution.<br />
A water quantity <strong>and</strong> quality model for the river<br />
Dender for 1994 was implemented in ESWAT.<br />
ESWAT is an extension of SWAT (van Griensven<br />
<strong>and</strong> Bauwens, 2000), the Soil <strong>and</strong> Water<br />
Assessment Tool developed by the USDA (Arnold<br />
et al., 1998). ESWAT was developed to allow for<br />
an integral modelling of the water quantity <strong>and</strong><br />
quality processes in river basins.<br />
1087
3. METHODS<br />
To reduce the overall uncertainty on the model<br />
results for a certain variable the following steps are<br />
proposed.<br />
1. Identify which sources contribute mainly to<br />
the overall uncertainty on the model results<br />
2. Estimate or calculate the uncertainty related to<br />
those main contributors<br />
3. Propagate the uncertainty through the model<br />
4. Analyse the model results to set up a future<br />
monitoring campaign<br />
5. Perform the measurements<br />
6. Recalibrate the model with new inputs<br />
7. Repeat step 3 till 6 until satisfying results are<br />
obtained<br />
For every step of this process different techniques<br />
exist that can be chosen among according to the<br />
experience of the modeller. In the practical<br />
example the methods we used will be described.<br />
Step 1: Identification of the main uncertainty<br />
contributors, uncertainty characterisation.<br />
This step is mainly carried out via a global or local<br />
sensitivity analysis. Because it is assumed that an<br />
already calibrated model is available, a local<br />
sensitivity analysis will certainly identify the most<br />
important parameters <strong>and</strong> data of the model.<br />
Indeed, local analysis is done around an a priori<br />
assumed value of the parameter. For a local<br />
sensitivity analysis the following methods exist:<br />
finite difference method, (b) the direct differential<br />
method, (c) the Green’s function method, (d) the<br />
polynomial approximation method <strong>and</strong> (e)<br />
automatic differentiation.<br />
For a detailed review of existing sensitivity<br />
techniques reference is made to the reviews of<br />
Turanyi (1990) <strong>and</strong> Rabitz et al. (1983)<br />
Step 2: Estimation or calculation of uncertainty<br />
Parameter uncertainty can be calculated using the<br />
covariance matrix obtained during the local<br />
sensitivity analysis or the calibration process.<br />
(Beck, 1987)<br />
If no direct calculations are possible, e.g. for the<br />
uncertainty on the inputs, it is best to estimate the<br />
uncertainty for this. One can divide the parameters<br />
<strong>and</strong> data in uncertainty classes (accurately known,<br />
very poorly known <strong>and</strong> an intermediate class) <strong>and</strong><br />
assign a percentage uncertainty to them. A similar<br />
approach was adopted by Reichert <strong>and</strong><br />
Vanrolleghem, 2001.<br />
Step 3: Propagate the uncertainty through the<br />
model<br />
For this step Monte Carlo methods can be used, in<br />
which the input data or parameters are sampled<br />
between the uncertainty bounds that are detected in<br />
the previous step. Another option is to apply linear<br />
error propagation. The advantage of the latter is<br />
computational efficiency. However, if model nonlinearities<br />
are significant within the uncertainty<br />
range, the results will be inaccurate. Monte Carlo<br />
simulation is a simple technique but requires a<br />
large number of model runs, which is<br />
computationally very dem<strong>and</strong>ing. Less runs with<br />
the same results as ‘ad r<strong>and</strong>om sampling’ are<br />
needed with ‘the Latin Hypercube sampling’<br />
(McKay et al., 1988).<br />
Step 4: Analyse the model results to set up a future<br />
measurement campaign<br />
Two different approaches can be used according to<br />
the aim for which the additional measurements are<br />
collected. If it is the aim to reduce parameter<br />
uncertainty an automated optimal experimental<br />
design method that is explained in V<strong>and</strong>enberghe<br />
et al (2002) can be used. It is based on<br />
maximisation of the determinant of the Fisher<br />
Information Matrix, which corresponds to the<br />
minimisation of the variance of the parameters.<br />
This method requires a lot of simulation runs but is<br />
totally automated <strong>and</strong> as such requires no<br />
additional information or knowledge from the<br />
modeller.<br />
However, when only focussing on the input data<br />
uncertainty that leads to output uncertainty expert<br />
–knowledge is required. It is then the aim to find a<br />
link between periods of high/low uncertainty <strong>and</strong><br />
external circumstances (rain, discharge points,<br />
seasons, solar radiation,…) This information is<br />
then used to make decisions about, place, period,<br />
frequency,… of future measurements.<br />
Step 5: Perform the measurements<br />
At this stage it is essential to ensure a good quality<br />
control on the measurements to minimise<br />
measurement errors. Important is also to carefully<br />
add information concerning hour, place <strong>and</strong> depth<br />
of the sample.<br />
Step 6: Recalibrate the model with new inputs<br />
An important issue here is that the calibration<br />
method has to be able to find the optimum. First, a<br />
choice is made between manual <strong>and</strong> automated<br />
methods. The former depends totally on the<br />
1088
experience of the modeller. Automated methods<br />
can differ in search method: global search methods<br />
scan the whole parameter space <strong>and</strong> are as such<br />
able to find the global optimum, but do not provide<br />
uncertainty measures. Local search methods start<br />
on a certain point in parameter space <strong>and</strong> end when<br />
they find an optimum. However, there is no<br />
assurance that this is the global optimum, so it is<br />
best to start in the neighbourhood of the optimum<br />
for those methods. With these methods covariance<br />
matrices for the optimum parameters are often<br />
calculated.<br />
Step 7: Repeat step 3 till 6 until satisfying results<br />
are obtained<br />
The stop criterion for this trial <strong>and</strong> error method is<br />
dictated by an ‘a priori’ desired reliability of the<br />
model results. In practice however, personnel, time<br />
<strong>and</strong> equipment matters will be the limiting factor<br />
<strong>and</strong> will indicate when this process stops.<br />
4. RESULTS AND DISCUSSION<br />
The seven steps are now demonstrated on a case<br />
study: simulations of the water quality of the river<br />
Dender, Fl<strong>and</strong>ers, Belgium for 1994. The<br />
evaluation of the uncertainty on model results is<br />
performed for Nitrate in the river water.<br />
Step 1: Identification of the main uncertainty<br />
contributors.<br />
We evaluate the sensitivity of the model on the<br />
following result: the time that NO 3 is higher than 3<br />
mg/l at Denderbelle, near the mouth of the river in<br />
1994. A sensitivity analysis for all input data <strong>and</strong><br />
parameters in the ESWAT model is too complex<br />
for the program we use: UNCSAM (Janssen et al,<br />
1992). This program cannot h<strong>and</strong>le more than 50<br />
parameters at the time. So we split the problem in<br />
different parts: 1) sensitivity to model parameters<br />
2) sensitivity to point pollution input <strong>and</strong> 3)<br />
sensitivity to diffuse pollution input. Each sub<br />
problem gives a ranking of the parameters by using<br />
the St<strong>and</strong>ardised Regression Coefficient (SRC) (1)<br />
SRC i =<br />
∆y<br />
/ S<br />
∆x<br />
i<br />
/ S<br />
y<br />
x i<br />
(1) with ∆ y / ∆xi<br />
= change<br />
in output due to a change in an input factor <strong>and</strong><br />
S , S the st<strong>and</strong>ard deviation of respectively the<br />
y<br />
x i<br />
output <strong>and</strong> the input. The input st<strong>and</strong>ard deviation<br />
S is specified by the user.<br />
x i<br />
The technique is explained in V<strong>and</strong>enberghe et al.<br />
(2001). For each of the subproblems the<br />
parameters or data that contribute significantly to<br />
the output (5 % level) are then taken together in<br />
one overall sensitivity analysis to compare the<br />
contribution of the different outputs. The column<br />
with the SRC as a result of that analysis is<br />
indicated in table 1 with “combined parameterinput”.<br />
Table 1: Results of the sensitivity analysis for the<br />
model output “hours NO 3 >3mg/l” at Denderbelle,<br />
1994. (pa16 = Amount of fertilisation on pasture in subbasin<br />
16; fa4 = Amount of fertilisation on farming l<strong>and</strong> in subbasin 4;<br />
gropa = growth date of pasture; plfa = Plant date on farming<br />
l<strong>and</strong>; co5 = Amount of fertilisation on corn in subbasin 5; co15<br />
= Amount of fertilisation on corn in subbasin 15; pa12 =<br />
Amount of fertilisation on pasture in subbasin 12; co11 =<br />
Amount of fertilisation on corn in subbasin 11; ai5 = O 2 uptake<br />
per unit of NH 3 oxidation; rk5 = denitrification rate; rk2 =<br />
oxygen reaeration rate; ai6 = O 2 uptake per unit of HNO 2<br />
oxidation; bc2 = rate NO 2 to NO 3; rk3 = rate of loss of bod due<br />
to settling; ai4 = O 2 uptake per unit of algae respiration; Rs5 =<br />
organic phophorous settling rate)<br />
Diffuse pollution<br />
input<br />
Pa1<br />
6<br />
SRC<br />
Point pollution input<br />
-0.30 BOD<br />
point<br />
6<br />
Fa4 0.23 NO3<br />
point<br />
7<br />
gro<br />
pa<br />
-0.18 BOD<br />
point<br />
5<br />
plfa 0.17 BOD<br />
point<br />
8<br />
Co5 -0.17 NH3<br />
point<br />
1<br />
Co<br />
15<br />
Pa<br />
12<br />
Co<br />
11<br />
-0.16 BOD<br />
point<br />
3<br />
0.16 BOD<br />
point<br />
7<br />
0.15 BOD<br />
point<br />
1<br />
NO3<br />
point<br />
5<br />
BOD<br />
point<br />
4<br />
NH3<br />
point<br />
2<br />
BOD<br />
point<br />
2<br />
NH3<br />
point<br />
3<br />
SRC<br />
parameter<br />
SRC<br />
Combined<br />
Parameter-input<br />
SRC<br />
-0.61 Ai5 -0.7 Ai5 -0.51<br />
0.42 Rk5 -0.34 Ai6 -0.50<br />
-0.38 Rk2 0.32 Rk5 -0.40<br />
-0.24 Ai6 -0.21 Bc2 0.38<br />
0.23 Bc2 -0.2 Ai4 -0.31<br />
-0.23 Rk3 0.17 Rk2 0.12<br />
-0.22 Ai4 0.12 plfa -0.08<br />
-0.14 Rs5 -0.09 BOD<br />
point<br />
6<br />
-0.07<br />
0.11 0.07 BOD -0.07<br />
point<br />
1<br />
-0.09 Pa16 0.07<br />
0.09<br />
-0.08<br />
0.06<br />
For the parameters, the sampling for the sensitivity<br />
analysis was based on own experience <strong>and</strong><br />
1089
literature ranges. The ranges for the diffuse<br />
pollution inputs are given in table 2 <strong>and</strong> the way<br />
they are determined is explained in V<strong>and</strong>enberghe<br />
et al. (2003). For the point pollution inputs we<br />
sampled uniform between halve <strong>and</strong> double the<br />
values, as we decided that those inputs belong to<br />
the uncertainty class 'poorly known’, indeed, the<br />
loads coming from point pollution were only<br />
available as yearly averages.<br />
Table 2. Uncertainty ranges for diffuse pollution<br />
input.<br />
Input<br />
Uncertainty<br />
Plant date for the crops<br />
+/- 1 month<br />
Harvest date of the crops<br />
+/- 1 month<br />
Amount of fertiliser applied per +/-25%<br />
subbasin <strong>and</strong> per crop (kg/ha)<br />
The global sensitivity of the parameters <strong>and</strong> the<br />
inputs shows that some parameters, O 2 uptake per<br />
unit of NH 3 oxidation, O 2 uptake per unit of HNO 2<br />
oxidation, denitrification rate, rate NO 2 to NO 3 , O 2<br />
uptake per unit of algae respiration <strong>and</strong> the<br />
reaeration rate are most influencing followed by<br />
the input data, plant date on farming l<strong>and</strong>, Amount<br />
of fertilisation on pasture in subbasin 12 <strong>and</strong> bod<br />
loads from point 1 <strong>and</strong> 6. This could not be seen<br />
from the separate analyses of inputs <strong>and</strong><br />
parameters. So the parameters can make the model<br />
give different results that are not much influenced<br />
by the input data. This again shows the importance<br />
of a well-calibrated model.<br />
Step 2: Estimation or calculation of uncertainty<br />
For both the point <strong>and</strong> diffuse pollution input the<br />
same uncertainties were taken as the sampling<br />
range used for the sensitivity analysis because we<br />
obtained no new information between the SA <strong>and</strong><br />
the UA. For the uncertainty on the parameters a<br />
recalibration with the most influencing parameters<br />
so that uncertainty ranges can be calculated with<br />
the covariance matrix is best, but is not done here.<br />
Uncertainties of 50 % were assigned to each of the<br />
parameters.<br />
Step 3: Propagation of the uncertainty through the<br />
model<br />
Here again the uncertainties are split: parameter<br />
uncertainty, diffuse pollution uncertainty <strong>and</strong> point<br />
pollution uncertainty.<br />
Then for each an uncertainty analysis was<br />
performed in which all of the uncertainty sources<br />
are varied at the same time to see the effects of the<br />
uncertainty on parameters <strong>and</strong> inputs. For this<br />
analysis we calculate the uncertainty b<strong>and</strong>s (i.e. the<br />
5% <strong>and</strong> 95% percentiles) for the results of the time<br />
series.<br />
Figure 2 <strong>and</strong> 3 shows the time series of nitrate in<br />
the river water at Denderbelle, situated near the<br />
mouth, with the 5% <strong>and</strong> 95% uncertainty bounds<br />
with resp. uncertainty on diffuse input <strong>and</strong> point<br />
pollution input. Figure 1 shows the uncertainty<br />
bounds for nitrate at the same location due to<br />
parameter uncertainty.<br />
Step 4: Analyse the model results to set up a future<br />
measurement campaign<br />
Figure 1 shows the propagation in time of the<br />
parameter uncertainty for Nitrate in the river at<br />
Denderbelle, 1994. Parameter uncertainty becomes<br />
at certain moments.<br />
To cope with the parameter uncertainty optimal<br />
experimental design based on the Fisher<br />
Information Matrix should be done (as explained<br />
in the methods section) as this is the most<br />
objective method to find important measurement<br />
places to better estimate the parameters. This<br />
design of new experiments is not presented here as<br />
we focus here on the uncertainty analysis <strong>and</strong> what<br />
information can be revealed from it.<br />
Nitrate (mg/l)<br />
10<br />
8<br />
6<br />
4<br />
2<br />
mean<br />
5% percentile<br />
95% percentile measured nitrate<br />
0<br />
1 26 52 78 104 130 156 182 208 233 259 285 311 337 363<br />
time (days)<br />
Figure 1. Simulation of nitrate with confidence<br />
intervals related to parameter uncertainty at<br />
Denderbelle, 1994.<br />
Figure 2 <strong>and</strong> 3 give shows the simulations <strong>and</strong><br />
their confidence intervals related to the uncertainty<br />
on the model inputs.<br />
Nitrate (mg/l)<br />
Nitrate (mg/l)<br />
mean<br />
5% percentile<br />
95% percentile measured nitrate<br />
9<br />
8<br />
7<br />
6<br />
5<br />
4<br />
3<br />
2<br />
1<br />
0<br />
1 26 51 77 102 127 153 178 204 229 254 280 305 330 356<br />
time (days)<br />
mean<br />
5% percentile<br />
95% percentile measured nitrate<br />
18<br />
16<br />
14<br />
12<br />
10<br />
8<br />
6<br />
4<br />
2<br />
0<br />
1 26 52 78 104 130 156 182 208 234 260 286 312 337 363<br />
time (days)<br />
1090
Figure 2 <strong>and</strong> 3. Simulation of nitrate with<br />
confidence intervals related to diffuse <strong>and</strong> point<br />
pollution input uncertainty at Denderbelle, 1994.<br />
rainfall (mm)<br />
rainfall<br />
flow<br />
5<br />
4.5<br />
4<br />
3.5<br />
3<br />
2.5<br />
2<br />
1.5<br />
1<br />
0.5<br />
0<br />
1 27 53 79 105 131 157 183 209 235 261 287 313 339 365<br />
time (days)<br />
100<br />
90<br />
80<br />
70<br />
60<br />
50<br />
40<br />
30<br />
20<br />
10<br />
0<br />
Figure 4. Rainfall <strong>and</strong> Flow in 1994 at<br />
Denderbelle.<br />
Linking the obtained results in step 3 to the<br />
external circumstances, rain <strong>and</strong> flow (fig.4), we<br />
can see that diffuse pollution inputs are important<br />
during periods with high rainfall <strong>and</strong> high flows.<br />
During dry weather flows, the input uncertainty of<br />
the loads is also propagated. Hence this UA learns<br />
that we can obtain a better calibration for the<br />
diffuse pollution part of the model with data that<br />
are taken during wet periods with high flows,<br />
because the model output nitrate is more sensitive<br />
towards inputs of diffuse pollution in those<br />
periods. If one focusses on calibrating the instream<br />
behaviour <strong>and</strong> point pollution then<br />
measurements during dry periods are needed, as<br />
the model is in such conditions not sensitive<br />
towards input of diffuse pollution.<br />
Further it is seen on fig. 1 that the 95 % bounds<br />
show much higher peaks than the mean<br />
concentrations time series. This means that some<br />
peak values of nitrate in the river water at<br />
Denderbelle may not be predicted properly due to<br />
an underestimation of the amount of fertiliser used.<br />
Those peaks (eg. day 156 <strong>and</strong> 260) are<br />
significantly higher than the levels of nitrate for<br />
basic water quality.<br />
It is also of intrest to know how the uncertainty is<br />
propagated from one place to the other. This<br />
analysis was done for the uncertainty propagation<br />
due to diffuse pollution inputs. The amount of time<br />
that NO 3 was higher than 3 mg/l was calculated.<br />
This was done for the time series of the mean, the<br />
5 % - bound <strong>and</strong> the 95% - bound (Fig. 5). The<br />
uncertainty bounds become larger when<br />
approaching the mouth due to the summation of<br />
the uncertainties on all diffuse pollution inputs that<br />
enter the river. However, it is interesting to see that<br />
with the available quality of input data no<br />
conclusions can be drawn concerning the question<br />
whether the diffuse pollution causes more hours<br />
nitrate exceedance downstream than upstream.<br />
More accurate data are needed to draw good<br />
conclusions from the model results.<br />
flow (m 3 /s)<br />
8000<br />
7000<br />
6000<br />
5000<br />
4000<br />
3000<br />
2000<br />
1000<br />
0<br />
Geraadsbergen<br />
Idegem<br />
Pollare<br />
teralfene<br />
Aalst<br />
Dendermonde<br />
Denderbelle<br />
95%<br />
mean<br />
figure 5. Uncertainty propagation from upstream to<br />
the mouth of the Dender in 1994 related to diffuse<br />
pollution input uncertainty.<br />
Step 5: Perform the measurements<br />
Step 6: Recalibrate the model with new inputs<br />
Step 7: Repeat step 3 till 6 until satisfying results<br />
are obtained<br />
Those three steps are only relevant for future<br />
measurement campaigns. However, no additional<br />
measurements were done until now.<br />
5. CONCLUSIONS & RECOMMENDATIONS<br />
The results of uncertainty analysis were here<br />
evaluated to guide future monitoring campaigns.<br />
Diffuse <strong>and</strong> point pollution inputs are considered<br />
separately <strong>and</strong> give information of the model<br />
sensitivity to the inputs. Measurements during dry<br />
periods can be used to better calibrate the model<br />
for point source pollution because the inputs of<br />
diffuse pollution are not important then. On the<br />
other h<strong>and</strong>, periods with rainfall <strong>and</strong> high flows are<br />
needed for the calibration of the model with<br />
diffuse pollution because the model output nitrate<br />
is then very sensitive towards the inputs related to<br />
farmer’s practices.<br />
When comparing the influence of the uncertainty<br />
of the diffuse pollution inputs, the uncertainty<br />
bounds appeared to be too high to draw reliable<br />
conclusions from the model results. So, it showed<br />
the importance of accurate measurements <strong>and</strong><br />
input data if the model results serve for decision<br />
support.<br />
It is obvious from the comparison between the<br />
global sensitivity analysis for the subgroups <strong>and</strong><br />
for all most influencing parameters together that<br />
the parameters are most important. This shows that<br />
it is best to start with a good calibration of your<br />
model <strong>and</strong> then focus on more accurate input data.<br />
Too often a model is calibrated with only one<br />
comprehensive measurement campaign. This is<br />
mostly not the most efficient way. When e.g. only<br />
measurements during dry periods are made, the<br />
model cannot be well calibrated for the diffuse<br />
pollution part. So it is better to perform two<br />
separate smaller measurement campaigns with the<br />
5%<br />
1091
first one being ‘exploring’, while the second<br />
campaign is guided by previous analysis of the<br />
model results. The combination of the two<br />
monitoring campaigns can assure that at least some<br />
measurements are performed at ‘the right<br />
moment’, making the calibration process easier<br />
<strong>and</strong> more reliable.<br />
It is necessary to combine all previous uncertainty<br />
analysis to evaluate the total uncertainty on the<br />
model results <strong>and</strong> to compare them with the<br />
measurements. In this way, model structure<br />
uncertainty can also be quantified (Willems <strong>and</strong><br />
Berlamont, 2002).<br />
In this research the second monitoring campaign is<br />
missing <strong>and</strong> could have shown the possibilities of<br />
the proposed succession of steps.<br />
8. ACKNOWLEDGEMENT<br />
The authors give special thanks to the financial<br />
support of the EU Harmoni-CA project ( EVKI-<br />
CT-2002-20003) <strong>and</strong> the EU CD4WC project<br />
(EVK1-CT-2002-00118).<br />
V<strong>and</strong>enberghe V., van Griensven A., Bauwens W.<br />
(2002). Detection of the most optimal<br />
measuring points for water quality variables:<br />
Application to the river water quality model of<br />
the river Dender in ESWAT. Wat.Sci.Tech.,<br />
46(3),1-7.<br />
V<strong>and</strong>enberghe V., van Griensven A., Bauwens W.<br />
<strong>and</strong> Vanrolleghem P.A. (2003). Propagation of<br />
uncertainty in diffuse pollution into water<br />
quality predictions: Application to the river<br />
Dender in Fl<strong>and</strong>ers, Belgium. In: Proceedings<br />
of the 7 th <strong>International</strong> Specialised Conference<br />
on Diffuse Pollution <strong>and</strong> Basin Management,<br />
17-22 August 2003, Dublin, Irel<strong>and</strong>.<br />
Van Griensven A. <strong>and</strong> Bauwens W. (2000).<br />
Integral modelling of catchments.<br />
Wat.Sci.Tech,, 43(7), 321-328.<br />
Willems P. <strong>and</strong> Berlamont J., (2002). Probabilistic<br />
emission <strong>and</strong> immission modelling: case-study<br />
of the combined sewer-WWTP-receiving<br />
water system at Dessel (Belgium). Wat.Sci.<br />
Tech., 45(3), 117-124.<br />
7. LITERATURE<br />
Arnold J.G., Williams J.R., Srivnivasan R. <strong>and</strong><br />
King K.W. (1996). SWAT manual. USDA,<br />
Agricultural Research Service <strong>and</strong> Blackl<strong>and</strong><br />
Research Center, Texas.<br />
Beck M.B. (1987). Water quality modeling: A<br />
review of the analysis of uncertainty. Wat.<br />
Res. Res., 23(5), 1393-1441.<br />
Demuynck C., Bauwens W., De Pauw N.,<br />
Dobbelaere I. <strong>and</strong> Poelman E. (1997).<br />
Evaluation of pollution reduction scenarios<br />
in a river basin: application of long term<br />
water quality simulations. Wat.Sci.Tech.,<br />
35(9), 65-75.<br />
Janssen P.H.M., Heuberger, P.S.C. <strong>and</strong> S<strong>and</strong>ers S.<br />
(1992). Manual Uncsam 1.1., a software<br />
package for sensitivity <strong>and</strong> uncertainty<br />
analysis. Bilthoven, The Netherl<strong>and</strong>s.<br />
McKay M.D. (1988). Sensitivity <strong>and</strong> uncertainty<br />
analysis using a statistical sample of input<br />
values. In: Uncertainty analysis, Y. Ronen,<br />
ed., CRC Press, Inc., Boca Raton, Florida,<br />
145-186.<br />
Rabitz H., Kramer M. <strong>and</strong> Dacol D. (1983).<br />
Sensitivity analysis in chemical kinetics.<br />
Ann. Rev. Phys. Chem., 34 , 419–461.<br />
Reichert P. <strong>and</strong> Vanrollegem P.A. (2001).<br />
Identifiability analysis of the River Water<br />
Quality Model No. 1 (RWQM1). Wat. Sci.<br />
Tech., 43(7), 329-338.<br />
Turanyi T. (1990), Sensitivity analysis of complex<br />
kinetic systems, tools <strong>and</strong> applications. J.<br />
Math. Chem., 5, 203–248.<br />
1092
Index of authors<br />
Ablan M. ..............................II 840<br />
Acevedo M. ........................I 196<br />
Adam S. ................................III 1276<br />
Agostini P.............................II 629<br />
Agustin E.O ........................II 623<br />
Ahuja L.R. ..........................I 409<br />
Al-Taiee T.M. ....................III 1111<br />
Andretta M. ........................II 513<br />
Andrews F.T. ......................II 1039<br />
Antonucci A. ......................I 98<br />
Apel H...................................I 977<br />
ApSimon H. ........................III 1252<br />
Arauco E. ............................II 543<br />
Argent R.M. ........................I 365<br />
Aronica G. ..........................III 1147, 1183<br />
Arrigucci S. ........................I 63<br />
Arsenic ′ I.D. ........................II 939<br />
Ascough IIº J.C. ................I 409<br />
Astalos J. ..............................I 480<br />
Athanasiadis I.N. ..............II 531, 643<br />
Bagnera A. ..........................II 693<br />
Balent G. ..............................II 699<br />
Balogh P. ..............................II 803<br />
B<strong>and</strong>ini S. ............................I 277, 289<br />
Banks C. ..............................I 69<br />
Barceló D.............................III 1241<br />
Bärlund I. ............................II 723, 1051,<br />
1057<br />
Barros R. ..............................II 840<br />
Bartelt J.L. ..........................II 858<br />
Basset-Mens C...................I 319<br />
Basson L...............................I 313<br />
Bastos E.T. ..........................III 1357<br />
Batten D. ..............................I 203<br />
Baumberger N. ..................III 1276<br />
Bauwens W. ........................II 1087<br />
Bazzani G.M.......................II 599<br />
Beigl P. ..................................II 468, 711<br />
Bellocchi G. ........................II 656<br />
Bellot J. ................................II 846<br />
Berlekamp J. ......................II 593<br />
Bernasconi G. ....................II 513<br />
Bernhardt K. ......................I 57<br />
Betti F. ..................................II 1033<br />
Beyer A.................................III 1229<br />
Biedermann R. ..................II 933<br />
Binder C.R...........................II 791<br />
Binner E. ..............................I 209<br />
Blind M.................................I,II 346, 456,<br />
635<br />
Blöschl G. ............................II 977<br />
Bocquillon C. ....................III 1159<br />
Bolte J.P. ..............................I 1<br />
Bondeau A...........................I 397<br />
Bonet A.................................II 846<br />
Bongartz K. ........................II 562<br />
Booij M.J. ............................II 556, 611,<br />
1021<br />
Borsuk M.E.........................I,II,III 421, 550,<br />
1468<br />
Boumans R. ........................II 783<br />
Bouwer L. ............................II 783<br />
Boyle D.P. ............................III 1135<br />
Brilhante V...........................I 75<br />
Bryan B.A. ..........................II 680<br />
Buchan K. ............................II 656<br />
Budincevic M.....................II 996<br />
Bull C.M...............................II 895<br />
Burkhardt-Holm P. ..........III 1468<br />
Butler D. ..............................I 116<br />
Butterfield D.......................II 1081<br />
Cabanillas D. ......................I 45<br />
Callicott B. ..........................I 196<br />
Callies U...............................III 1129<br />
Calver A. ..............................III 1214<br />
Camera R. ............................II 1027<br />
Campos dos Santos J.L. I 75<br />
C<strong>and</strong>ela A. ..........................III 1147, 1183<br />
Cappy S. ..............................II 736<br />
Carlon C. ..............................II 629<br />
Carrick N. ............................II 574<br />
Castilla M. ..........................I 301<br />
Catania F...............................II 984, 1493<br />
Cavalieri S. ..........................II 1033<br />
Ceccaroni L.........................I 45<br />
Cecchetti M.........................I 93<br />
Cerdeira R. ..........................III 1270<br />
Cernesson F.........................II 662<br />
Chan F. ..................................III 1338, 1423,<br />
1436, 1455<br />
Chiew F.H.S. ......................III 1511<br />
Cleij P. ..................................II 742<br />
Cline J.C...............................II 810<br />
Coelho L.M.R. ..................III 1270<br />
Cogan V. ..............................II 1027<br />
Cogdill T...............................I 196<br />
Comas J. ..............................I 45<br />
Coomber L. ........................III 1363<br />
Corani G. ..............................I,III 93, 1301<br />
Cortés U. ..............................III 1099<br />
Corti G. ................................II 513<br />
Courdier R. ..........................II 462<br />
Cox P.A.................................II 858<br />
Craps M. ..............................II 662<br />
Critto A. ................................II 629<br />
Croke B.F.W. ......................II,III 433, 1201,<br />
1208<br />
Crooks S. ..............................III 1214<br />
Crossman N.D. ..................II 680<br />
Dacombe P...........................I 69<br />
Dasic T. ................................I 153<br />
David O. ..............................I 358, 403,<br />
409, 439<br />
Dávila J. ................................II 840<br />
De A. Medeiros V.M. ......II 990<br />
De A. Vitola M. ................II 828<br />
De Jong C. ..........................II 736<br />
De Kok J.-L. ......................II 1021<br />
De Kort I.A.T. ....................II 556<br />
De Luca S.J.........................II 828<br />
Deconchat M. ....................II 699<br />
DeRose R.............................II 1014<br />
Devireddy V.K. ..................I 128<br />
Devoy R. ..............................III 1417<br />
Djordjevic B. ......................I 153<br />
Djurdjevic V. ......................II 956<br />
Dobrucky M. ......................II 480<br />
Doglioni A...........................I 134<br />
Donatelli M.........................II 656<br />
Drogue G. ............................III 1177<br />
Dunn S. ................................II 970<br />
Dunne D. ..............................III 1417<br />
Dur<strong>and</strong> P...............................I 319<br />
Dussaillant A.R. ................III 1105<br />
Ebenhöh E. ..........................I 177<br />
Eisenack K. ........................I 104<br />
Eisenhuth D. ......................II 846<br />
El-Idrissi A. ........................III 1177<br />
Engelen G. ..........................I 340<br />
Ernst T. ..................................II 474<br />
Evans A.J. ............................II 914<br />
Evans B.................................III 1123<br />
Facchi A. ..............................II 1069<br />
Fassio A. ..............................II 1027<br />
Fatai K...................................III 1398, 1405<br />
Fath B.D. ..............................II 822<br />
Feás J. ....................................II 617<br />
Fenner K...............................III 1229<br />
Ferr<strong>and</strong> N.............................II 662<br />
Ferret R. ................................I 301<br />
Ferrier R. ..............................II 970<br />
Finér L...................................II 883<br />
Fiorucci P. ............................II 717, 748<br />
Flores X. ..............................I 51<br />
Flügel W.A. ........................I,II 265, 562<br />
Foit K.....................................II 945<br />
Förster R...............................I 233<br />
Franken R.O.G. ................II 742<br />
Friedrich R. ........................I 122<br />
Fritchel P.E. ........................III 1135<br />
Fry M. ....................................III 1002<br />
Fukiharu T. ..........................III 1387<br />
Gaddis E.J. ..........................I 227<br />
Gaetani F. ............................II 717, 748<br />
Gal G. ....................................II 506<br />
G<strong>and</strong>olfi C...........................II 1069<br />
Garcia B. ..............................III 1357<br />
Garcia J.M. ..........................III 1270<br />
Garcia L. ..............................I 358<br />
Gasques J.G. ......................III 1357<br />
Gatial E. ................................II 480<br />
Gault J. ..................................III 1417<br />
Gavardinas C. ....................III 1282<br />
Gebetsroither E. ................I 283<br />
Geisler G. ............................I 295<br />
Gerner D...............................II 680<br />
Gibert K. ..............................I 51<br />
Gigler U. ..............................I 271<br />
Giglio D. ..............................II 605<br />
Gilbert R.O. ........................III 1499<br />
Giordano R. ........................I 247<br />
Giove S. ................................II 629<br />
Giupponi C. ........................II 617, 1027<br />
Giusti E. ................................I 110<br />
Giustolisi O.........................I 134<br />
González R.M. ..................III 1264<br />
Goodwin T.H. ....................III 1002<br />
Gouveia C. ..........................III 1270<br />
Graf N. ..................................II 593
Granlund K. ........................II 1057<br />
Grant W.E. ..........................II 822<br />
Grasso M. ............................III 1462<br />
Grayson R.B. ......................II 1075<br />
Gregersen J.B.....................I 346, 456<br />
Gregory S.V. ......................I 1<br />
Greiner R. ............................II 705<br />
Griffioen J. ..........................III 1235<br />
Grimvall A...........................II 519, 1081<br />
Grossinho A. ......................III 1252<br />
Grsic ˇ ′ Z. ................................II 956<br />
Gualtieri C...........................II 962<br />
Guanghuo W.......................II 623<br />
Guariso G.............................I,III 93, 1301<br />
Guerrin F. ............................II 462<br />
Haas A...................................I 450<br />
Habala O. ............................II 480<br />
Habeck A. ............................II 920<br />
Hall N. ..................................I,II 215, 705<br />
Hare M. ................................I 190<br />
Harvey C.F. ........................III 1093<br />
Hasan A.A. ..........................III 1111<br />
Hashimoto T. ......................II 870<br />
Hattermann F.F. ................II 920, 1064<br />
Heaven S. ............................I 69<br />
Heijungs R...........................I 332<br />
Helbig A. ..............................III 1314<br />
Hellmuth M.........................II 797<br />
Hellweg S. ..........................I 295<br />
Hengsdijk H. ......................II 623<br />
Hernández J.M...................III 1351<br />
Hess O...................................II 593<br />
Hesselmann J. ....................I 159<br />
Hesser F.B. ..........................I 444<br />
Hilden M. ............................II 723<br />
Hilker F.M. ..........................II 902<br />
Hinkel J.................................I 352<br />
Hinsch M. ............................II 902<br />
Hluchy L. ............................II 480<br />
Hoffmann L.........................III 1177<br />
Hofman D. ..........................I 371<br />
Hoheisel A...........................II 474, 500<br />
Holman I.P...........................III 1165<br />
Holmes M.G.R. ................II 1002<br />
Hostmann M. ......................II 550<br />
Hoti S.....................................III 1345, 1436,<br />
1449, 1474, 1505<br />
Hreiche A.............................III 1159<br />
Hristov T...............................III 1123<br />
Hu B.......................................III 1326, 1411<br />
Huang P.................................III 1381<br />
Huijbregts M.A.J. ............I 332<br />
Hulse D.W. ..........................I 1<br />
Hungerbühler K. ..............I 295<br />
Huth A...................................II 889<br />
Iffly J.-F. ..............................III 1177<br />
Ioncheva V...........................III 1123<br />
Ito Y. ......................................II 834<br />
Itoh Y. ....................................II 852<br />
Jaeger C. ..............................I 450<br />
Jakeman A.J. ......................I,III 433, 1511<br />
Jakeman T. ..........................I 215<br />
Jarvie H.P. ............................II 1081<br />
Jeffery K.G. ........................I 491<br />
Jeffrey P. ..............................II 537, 668<br />
Ji M.........................................I 196<br />
Jolliet O. ..............................I 307<br />
Jones D.A.............................III 1214<br />
Junk J. ....................................III 1314<br />
Junqueira I.C. ....................II 828<br />
Ka'eo Duarte T...................III 1093<br />
Kabat P. ................................I 11<br />
Kalkuta S. ............................III 1259<br />
Kallache M. ........................III 1517<br />
Kämäri J. ..............................II 1051<br />
Kapor D. ..............................II 939, 996<br />
Karatzas K. ..........................II 525<br />
Kassahun A. ........................III 1282, 1288<br />
Kathirgamanathan P.........III 1247<br />
Kaufmann A. ......................I 283<br />
Kay A.L. ..............................III 1214<br />
Keedwell E. ........................I 141<br />
Kemp-Benedict E. ............II 765<br />
Khu S.T.................................I 141, 147<br />
Kingston G.B. ....................I 87<br />
Kirkkala T. ..........................II 1051<br />
Kjeldsen T. ..........................III 1214<br />
Kleyer M. ............................II 933<br />
Knoflacher M. ....................I 271<br />
Koivusalo H. ......................II 883<br />
Kok K. ..................................II 754<br />
Kokkonen T.........................II 883<br />
Koo B. ..................................II 970<br />
Korobochkina S. ..............III 1487<br />
Kralisch S. ..........................I 403<br />
Krause P. ..............................I 403<br />
Krivtsov V. ..........................I 69<br />
Krol M.S...............................II 760<br />
Kropp J. ................................I,III 104, 1517<br />
Kryazhimskiy A. ..............III 1487<br />
Krysanova V. ......................II 730, 920,<br />
1064<br />
Krywkow J. ........................I 184<br />
Kudo E. ................................II 580<br />
Kuo C.-C. ............................III 1219<br />
Kushida M. ..........................II 834<br />
Kytzia S. ..............................I 233<br />
Laborte A.G. ......................II 623<br />
Lacorte S. ............................III 1241<br />
Ladet S. ................................II 699<br />
Lai N.X. ................................II 623<br />
Lalic B...................................II 996<br />
Lam D. ..................................I,III 427, 1381,<br />
1393<br />
Lambert M.F.......................I 87<br />
Lamorey G. ........................III 1135<br />
Last R. ..................................II 705<br />
Laurén A...............................II 883<br />
Lautenbach S. ....................II 593<br />
Leavesley G. ......................II 736<br />
Leavesley G.H. ..................I 439<br />
Lee H. ....................................III 1171<br />
Lehtonen H. ........................II 723, 1057<br />
León C.J. ..............................III 1345, 1351<br />
Leon L.F. ..............................I 427<br />
Lertsirivorakul R. ............II 705<br />
Letcher R.A.........................I,III 81, 433,<br />
1511<br />
Leterme P. ............................I 319<br />
Lim C.....................................III 1332, 1338,<br />
1363<br />
Lindenschmidt K.-E. ......I 444<br />
Lindquist C. ........................I 196<br />
Lischke H.............................II 908<br />
Littlewood I.G. ..................III 1153<br />
Liu X. ....................................III 1276<br />
Liu Y.......................................I 147<br />
Lizuma L. ............................III 1225<br />
Llorens E. ............................I 45<br />
Löfving E.............................II 519<br />
Lorenz J.J.............................II 810<br />
Lotze-Campen H. ............I 397<br />
Low Choy S. ......................II 927<br />
Lu H. ......................................II,III 1014, 1117<br />
Lucht W. ..............................I 397<br />
Luja P.....................................I 340<br />
Machauer R.........................II 736<br />
MacKay M. ........................III 1381, 1393<br />
MacLeod M.J. ....................III 1229<br />
Madsen H.............................I 147<br />
Maggi D. ..............................II 1069<br />
Maier H.R. ..........................I 87<br />
Makropoulos C.K.............I 116<br />
Maksimov V. ......................III 1487<br />
Malinovic ′ S.........................II 939<br />
Maliska M. ..........................II 480<br />
Malve O. ..............................II 1051<br />
Manca A. ..............................II 771<br />
Manera M. ..........................III 1462<br />
Manzoni S. ..........................I 289<br />
Marcomini A. ....................II 629<br />
Maréchal D. ........................III 1165<br />
Marinova D.........................III 1423<br />
Märker M.............................II 562<br />
Markstrom S. ......................III 1135<br />
Marsili-Libelli S. ..............I,II 63, 110,<br />
1033<br />
Martin-Clouaire R. ..........I,II 166, 699<br />
Masouras A. ........................II 525<br />
Massabò M. ........................II,III 693, 984,<br />
1493<br />
Mastropietro R...................II 513<br />
Matejicek L.........................I 391<br />
Matgen P...............................III 1177<br />
Matthews K.B. ..................II 656<br />
Matthies M. ........................II 593<br />
Maurel P. ..............................II 662<br />
McAleer M. ........................III 1320, 1332,<br />
1338, 1345, 1368, 1423, 1436, 1442, 1449,<br />
1455, 1462, 1474, 1505, 1253<br />
McIntosh B.S. ....................II 537, 668<br />
McIntyre N. ........................II,III 1008, 1171<br />
Médoc J.-M.........................II 462<br />
Meiwirth K. ........................II 951<br />
Meixner T.............................II 1045<br />
Mendoza G. ........................I 301<br />
Mermoud A.........................II 951<br />
Merz B. ................................II 977<br />
Mihailovic ′ D.T. ................II 939, 996
Mijatovic ′ Z. ........................II 939<br />
Milne-Home W. ................II 705<br />
Minciardi R. ........................II,III 605, 693,<br />
717, 748, 1493<br />
Mitkas P.A. ..........................II 531, 643<br />
Miyazaki T...........................II 858, 870<br />
Moeller A.............................I 379<br />
Molini L. ..............................II 693<br />
Möltgen J. ............................III 1294<br />
Monticino M.......................I 196<br />
Moran C. ..............................II,III 1014, 1117<br />
Moreira L.F.F. ....................II 990<br />
Moreno N.............................II 840<br />
Morgan D.............................II 914<br />
Morimune K. ......................III 1307<br />
Morley M.S.........................I 116<br />
Müller C. ..............................I 397<br />
Mysiak ˇ J. ..............................II 674<br />
Najem W...............................III 1159<br />
Nakamori Y. ........................I 385<br />
Nancarrow B.E. ................I 172<br />
Newham L.T.H. ................I,II 81, 1039<br />
Newig J. ................................I 159<br />
Nicholson A. ......................I 215<br />
Nishiyama Y. ......................III 1307<br />
Nitter S. ................................I 122<br />
Nogueira M. ........................III 1270<br />
Normatov I.S. ....................II 487<br />
Norton J.P.............................II,III 1039, 1201,<br />
1208<br />
Notten P.J. ............................I 326<br />
O'Mahony C. ......................III 1417<br />
Oliver M. ..............................I 215<br />
Ortuani B. ............................II 1069<br />
Osada S. ................................III 1307<br />
Ostendorf B.........................II 574, 680<br />
Ostrowski M. ......................II 580<br />
Oxlev L. ................................I,III 17, 1398,<br />
1405<br />
Oxley T. ................................III 1252<br />
Pahl-Wostl C.......................I 177, 190,<br />
240<br />
Paillat J.-M. ........................II 462<br />
Paladino O. ..........................II,III 984, 1493<br />
Pan H. ....................................III 1375<br />
Panebianco S. ....................I 240<br />
Parparov A...........................II 506<br />
Parry H. ................................II 914<br />
Passarella G. ......................I 247<br />
Passier H. ............................III 1235<br />
Pauwels L.L. ......................III 1474, 1505<br />
Pavesi G. ..............................I 277<br />
Pedrollo O.C.......................II 828<br />
Pennington D.W. ..............I 307<br />
Penttinen S. ........................II 883<br />
Pereira D. ............................II 828<br />
Peré-Trepat E. ....................III 1241<br />
Pérez J.L...............................III 1264<br />
Perry L.M.............................II 680<br />
Petrie J.G. ............................I 313, 326<br />
Pettit C. ................................I 253<br />
Pfister L. ..............................III 1177<br />
Phal-Wostl C.......................I 25<br />
Piirainen S. ..........................II 883<br />
Pizzorni D. ..........................II 605<br />
Poch M. ................................I,II 45, 1099<br />
Poethke H.J. ........................II 902<br />
Pollard O. ............................I 141<br />
Popov Z. ..............................II 956<br />
Post D.A. ..............................III 1195<br />
Post J. ....................................II 730<br />
Promburom P. ....................I 221<br />
Prosser I. ..............................II,III 1014, 1117<br />
Pullar D.................................I,II 253, 415,<br />
927<br />
Pulsipher B.A.....................III 1499<br />
Purwasih W. ........................I 385<br />
Quintero R. ..........................II 840<br />
Rajkovic B...........................II 956<br />
Rankinen K. ........................II 1057<br />
Reed P.M. ............................I 128<br />
Refsgaard J.C. ....................III 1288<br />
Reichert B. ..........................II 736<br />
Reichert P.............................I,II,III 421, 550,<br />
1468<br />
Reimer S...............................II 593<br />
Reineking B. ......................II 889<br />
Reis S.....................................I 122<br />
Rellier J.-P. ..........................I 166<br />
Reusser D.E. ......................I 190<br />
Reynard N.S. ......................III 1189, 1214<br />
Richter O. ............................II 945<br />
Righetto A.M. ....................II 990<br />
Rinke K.................................III 1294<br />
Rivington M. ......................II 656<br />
Rizzoli A.E. ........................I 365<br />
Robba M...............................II,III 693, 1093,<br />
1493<br />
Rochester W. ......................II 927<br />
Rode M. ................................I 444<br />
Rodriguez-Roda I. ............I 51<br />
Romanowicz R.J...............III 1129, 1141<br />
Rosato P. ..............................II 617<br />
Rosenbaum R. ....................I 307<br />
Rotmans J. ..........................I 184<br />
Rötter R. ..............................II 623<br />
Rouse W. ..............................III 1381, 1393<br />
Rozemberg T. ....................II 506<br />
Rozemeijer J. ......................III 1235<br />
Rudari R. ..............................II 605<br />
Rudner M.............................II 933<br />
Rüger N. ..............................II 586<br />
Rust H. ..................................III 1517<br />
Sabbadin R. ........................II 699<br />
Sacile R.................................II 605, 693,<br />
717<br />
Sadoddin A. ........................I 81<br />
Salhofer S. ..........................I,II 209, 711<br />
Salvetti A. ............................I 98<br />
San José R. ..........................III 1264<br />
Sánchez J.R.........................II 846<br />
Sànchez-Marré M.............I 51<br />
S<strong>and</strong>erson W.......................II 797<br />
Sarkar R.R. ..........................II 876<br />
Savic D.A.............................I 116, 134,<br />
147<br />
Scheffran J...........................I 104<br />
Scheringer M. ....................III 1229<br />
Schertzer W.........................I,III 427, 1381,<br />
1393<br />
Schlüter M. ..........................II 586<br />
Schneider F. ........................II 711<br />
Schneider I.W.....................I 358, 439<br />
Scholten H. ..........................III 1282, 1288<br />
Scholz R.W. ........................II 791<br />
Schröder B...........................II 933<br />
Schweizer S. ......................I,II 421, 550<br />
Scrimgeour F.G. ................III 1398, 1405<br />
Seaton R.A.F.......................II 668<br />
Sechi G.M. ..........................II 771<br />
Sendzimir J. ........................II 797, 803<br />
Seppelt R. ............................II 945<br />
Shadananan Nair K. ........I 259<br />
Shahsavani D. ....................II 1081<br />
Shareef R. ............................III 1368, 1449<br />
Sharma V. ............................III 1381, 1393<br />
Shi J. ......................................II 568<br />
Shiyomi M...........................II 816<br />
Sigel K. ................................II 674<br />
Simo B. ................................II 480<br />
Sivapalan M. ......................III 1117<br />
Slottje D. ..............................III 1320<br />
Smith C. ................................I 1<br />
Smith P. ................................I 397<br />
Sojda R.S. ............................II 649<br />
Sommaruga L.....................II 543<br />
Son T.T. ................................II 623<br />
Spate J.M. ............................III 1208<br />
Spiller S.H. ..........................I 69<br />
Spörri C. ..............................II 550<br />
Stabel E.................................II 686<br />
Starr M. ................................II 883<br />
Steinberga I. ........................III 1225<br />
Stroebe M. ..........................III 1229<br />
Strzepek K. ..........................II 797<br />
Suckow F. ............................II 730<br />
Suzuki T. ..............................II 816, 870<br />
Swain E.D. ..........................II 810<br />
Swayne D.............................I,II,III 427, 568,<br />
1381, 1393<br />
Swinford A. ........................II 537<br />
Syme G.J. ............................I 172<br />
Tailliez C. ............................III 1177<br />
Tainaka K.............................II 816, 834,<br />
852, 870<br />
Tattari S. ..............................II 723<br />
Tauler R. ..............................III 1241<br />
Tenhumberg B. ..................II 864, 895<br />
Thieken A.H. ......................II 977<br />
Tietje O. ................................II 777<br />
Timoshevskii A. ................III 1259<br />
Tockner K. ..........................II 550<br />
Togashi T. ............................II 858, 870<br />
Tomasoni A.........................II 605<br />
Tonella G. ............................II 840<br />
Tran V.D. ..............................II 480<br />
Trasforini E. ........................II 605, 717<br />
Truffer B...............................II 550<br />
Turon C.................................III 1099
Tymoshevska L. ................III 1259<br />
Tyre A.J.................................II 895<br />
Unger N. ..............................II 468<br />
Unger S. ................................II 474<br />
Uricchio V.F. ......................I 247<br />
Valkering P. ........................I,II 184, 662<br />
Van Dam J.D. ....................II 742<br />
Van Delden H.....................I,II 340, 754<br />
Van den Berg M. ..............II 623<br />
Van der Grift B. ................III 1235<br />
Van der Veen A. ................I 34, 184<br />
Van der Wal K.U...............II 1021<br />
Van der Werf H.M.G.......I 319<br />
Van Griensven A...............II 1045<br />
Van Ittersum M. ................II 623<br />
Van Keulen H.....................II 623<br />
Van Oel P. ............................II 760<br />
Van Wezel A.P. ..................II 742<br />
V<strong>and</strong>enberghe V. ..............II 1087<br />
Vanrolleghem P.A.............II 1087<br />
Vári A. ..................................II 803<br />
Vartalas P. ............................II 643<br />
Veiga B. ................................III 1442, 1523<br />
Viger R.J...............................I,II 358, 736<br />
Vijaykumar N.....................III 1417<br />
Viviani G. ............................III 1183<br />
Vizzari G. ............................I 289<br />
Vladich H.............................I 227<br />
Voinov A...............................I 227<br />
Vurro M. ..............................I 247<br />
Wade A.J. ............................II 1081<br />
Wagner U. ............................II 506<br />
Wahlin K. ............................II 1081<br />
Walker J.P. ..........................II 1075<br />
Wang Y. ................................III 1332<br />
Wang Y-C.............................III 1219<br />
Wassermann G...................I,II 209, 468,<br />
711<br />
Watson B. ............................I 215<br />
Weal<strong>and</strong>s S.R. ....................II 1075<br />
Webb B.W. ..........................I 134<br />
Wechsung F.........................II 1064<br />
Werners S.............................II 783<br />
Wheater H. ..........................III 1171<br />
Willmott S. ..........................I 45<br />
Wilson J.E. ..........................III 1499<br />
Wirtz K.W. ..........................I,III 57, 1276<br />
Wissel C. ..............................II 889<br />
Wolf J.....................................II 623<br />
Xu Y. ......................................II 611<br />
Yates D. ................................II 797<br />
Yeo J.......................................III 1430, 1481<br />
Yeremin V. ..........................III 1259<br />
Yongvanit S.........................II 705<br />
Yoshimura J. ......................II 816, 834,<br />
852, 858, 870<br />
Young A. ..............................III 1171<br />
Young A.R. ..........................II,III 1002, 1189<br />
Young P.C. ..........................III 1129, 1141<br />
Yu P.-S...................................III 1219<br />
Yuvaniyama A. ..................II 705<br />
Zaffalon M...........................II 98<br />
Zompanakis G. ..................III 1282<br />
Zuddas P. ..............................II 771