<p> Initial DRAFT version</p><p>WATER RESOURCES dynamic and distributed representation of the individual river basins. The underlying dynamic SIMULATION AND (daily) water resources simulation model describes OPTIMIZATION: A WEB the water resources systems at a basin scale including the groundwater system for conjunctive BASED APPROACH use. A coupled dynamic, daily water quality model simulates DO/BOD as well as conservative and DDr. Kurt FEDRA first-order decaying pollutants and sediments with Environmental Software & Services GmbH the full resolution of the basin wide river network A-2352 Gumpoldskirchen, Austria topology. This provides additional criteria for the [email protected] http:// www.ess.co.at optimization.</p><p>The primary unsupervised optimization identifies ABSTRACT sets of feasible and non-dominated pareto-optimal Water resources are one of the limiting factors of solutions in heavily constrained scenarios; these are regional development in many parts of the world, the basis for an interactive discrete multi-criteria and in particular round the Eastern and Southern selection using a reference point approach with the Mediterranean. Scarcity and conflict, but also participation of end users. considerable uncertainty, inter-annual variability, and simple lack of reliable data characterize the The multi-criteria approach covers global and problems of water resource management in many sectoral demand and supply balances, reliability of countries and river basins. supply, access, cost and benefits, as well as compliance with water quality standards Arbitrary Within the Framework of a FW6 sponsored INCO- penalty functions can be used for the valuation of MPC project, OPTIMA, a simulation based water violation of standards and missing targets, both resources planning and optimization system is shortfalls of supply as well as excess (flooding or being developed and applied in case studies in pollution), or exceeding water quality standards. Cyprus, Turkey, Lebanon, Jordan, Palestine and Israel, Tunisia and Morocco. The approach and methodology are demonstrated using the several case studies of the OPTIMA The model system addresses both quantity and project, and in particular the Gediz river basin quality, water demand and supply, reliability, In Turkey which covers about 18,000 km2, surface and groundwater, water technologies and approaches a total population of 2 million, and efficiency of use, allocation strategies, costs and drains into the Aegean Sea. The case demonstrates benefits. A web based client-server implementation the entire range of prototypical water management supports distributed use and easy access, providing problems in the region, and their potential considerable server side computing power without solutions. The case also demonstrates the any demand on the clients other than a standard importance of the institutional and regulatory web browser. This architecture supports a framework, and the need for direct participation of participatory approach involving local stake holders major actors and stakeholders in the planning and for multi-criteria optimisation and decision support. decision making processes which requires an easy to understand interface to the optimization tools. The simulation system uses nested or cascading models that represent various parts of the water INTRODUCTION resources system with a high spatial and temporal Water resources management lends itself to resolution. Component models such as a daily analysis by simulation and optimisation, as the rainfall-runoff model for sub-catchments use an basic principle of mass conservation provides a automatic calibration methodology based on pattern solid framework. Keeping track of the water in a matching and Gestalt rather than traditional river basin as a natural hydrographic unit in a statistical least square methods. This works very dynamic water budget, and estimating the costs and well in situations of limited data availability and benefits of supply and demands met should answer high data uncertainty. many questions. </p><p>The optimization of the overall water resources Of course, in any practical application there are system uses similar heuristics and concepts of uncertainties not only on the physical side, but even genetic programming, based on a realistic, detailed,</p><p>Kurt Fedra Page 1 4/6/2018 Initial DRAFT version more so on the socio-economic and political front Morocco, OPTIMA is extending classical where criteria, objectives and constraints are being optimisation and mathematical programming defined and evaluated. Multiple actors and their methodology in several respects, by: different views and preferences, conflicting  Using a full-featured dynamic and distributed objectives, hidden agenda, plural rationalities all simulation models and genetic programming as contribute to make water resources management a the core to generate feasible and non- challenging field. dominated alternatives. Water technology alternatives including their cost structure, and up-to-date remote-sensing derived land use While simulation modelling of WHAT IF scenarios information are primary inputs; can provide a high degree of detail and realism to problem representation, optimisation can answer  Extending the set of objectives, criteria and the more interesting question of HOW TO. The constraints through expert systems technology to include difficult to quantify environmental challenge is in combining the best of both worlds, and social dimensions; and at the same time integrate the tools into the real-world decision making processes, which  Putting specific emphasis on local acceptance requires a simple and intuitive representation and and implementation through the inclusion of stake-holders in an interactive, participatory interface to end users for a participatory approach. decision making process carefully embedded in institutional structures, using a discrete multi- For the modelling approach, one possible solution criteria reference point methodology; is to build sets or cascades on linked models rather than one all encompassing representation (Fedra,  Comparative evaluation and benchmarking across the set of local and regional case studies 2000) and switch from a classical optimisation in 7 countries, namely Cyprus, Turkey, paradigm that minimises or maximises an objective Lebanon, Jordan, Palestine, Tunisia and function to a concept of satisficing (e.g., Byron Morocco around the Southern and Eastern 1998; Simon, 1957; 1969), i.e., finding one or more Mediterranean. feasible solutions for a set of interactively defined constraints. MODELS AND TOOLS The latter requirement, the interactive involvement A set of linked models is used for the simulation of stakeholders in a decision making process, and optimisation of water resources management requires easy access to tools that may require problems: substantial computing power: a web based solution can offer a powerful set of servers organised in a  A dynamic land use change model cluster to end users that only need a standard PC (http://80.120.147.30/LUC/) and web browser.  A sub-basin scale dynamic rainfall-runoff that feeds the water resources model with THE OPTIMA PROJECT flow inputs for ungaged basins; The overall aim of OPTIMA is to develop,  An irrigation water demand model implement, test, critically evaluate, and exploit an estimating supplementary irrigation water innovative, scientifically rigorous yet practical demands as input for the water resources approach to water resources management intended model; to increase efficiencies and to reconcile conflicting  A river basin scale water resources demands. Based on the European Water Framework modelling system operating on a topological Directive (2000/60/EC) the approach equally network of nodes and reaches and linked to considers economic efficiency, environmental aquifers for conjunctive use scenarios: (http://www.ess.co.at/WATERWARE) compatibility, and social equity as the pillars of sustainable development.  A dynamic basin-wide water quality model The project also aims at building a wide for DO/BOD and conservative or first-order dissemination network of expertise and knowledge decaying pollutants, linked to the water resources management information system. exchange sharing its findings, generic data, and best practice examples. The models are linked to data bases that describe each of the cases in terms of GIS layers (Figure 1), For a broad geographical coverage and a good basis monitoring time series of climatic (Figure 2) and for comparative analysis, OPTIMA brings together hydro-meteorological data (Figure 3), and the main partners from Austria, Italy, Greece, Cyprus, Malta, components of the water resources system like Turkey, Lebanon, Jordan, Palestine, Tunisia, and major demand nodes. Linkage between the models</p><p>Kurt Fedra Page 2 4/6/2018 Initial DRAFT version is by means of indicators aggregated by rules that environmental water demands such as for the summarize the output from one model as boundary nourishment of wetlands, as well as flooding conditions for another, translating differences in conditions with arbitrary penalty function. dimensionality and resolution in time and space. The data and related analysis tools are accessible The dynamic, daily water budget is summarised in through the Internet to facilitate the dissemination terms of supply/demand ratios, globally and by of project results, but also to make it easier for the economic sector or administrative unit; reliability of various actors and stakeholders in each of the supply; and the set of violations of any of the regional cases to share a common information constraints defined for the control nodes. Economic basis. Empowerment through information as valuation (expressed as net present values or foreseen by Agenda 21’s chapter 40 is the annualised costs considering investments, operating underlying concept. costs, and project or technology life times) includes estimates of the cost of various alternative water technologies (from non-conventional supply options like desalination, water harvesting, recycling and re-use to new or bigger reservoirs, lining irrigation canals and more efficient irrigation technologies to water saving showers) versus the benefits generated by supplying water for useful demands. Different allocation scenarios but also the use of different water technologies lead to different cost-benefit ratios for the system. </p><p>Figure1: spatial selection of monitoring stations</p><p>Figure3: water resources model network Figure2: Monitoring data display From the set of results generated, any number of WATER RESOURCES MODELLING constraints can be derived for the optimization, both The central model is a dynamic, basin wide water global criteria aggregated over all nodes and a resources model. It is using a topological network yearly simulation run such as overall reliability of representation of a river basin, that consists of water supply, as well as specific node and location various node types and the river reaches and canals specific constraints defined as minimum or connecting them Nodes represent objects such as maximal flow (or supply) expectations, again with sub-catchments, reservoirs, wells, diversions and different temporal resolution and aggregation. An confluences and areas of water demand such as example would be the number of all periods of cities, tourist resorts, irrigation districts, and large three or more days when domestic water supply is industries. The surface water network can be less than 50% of the specified water demand for coupled to one or more aquifers to represent any or all of the domestic demand nodes conjunctive use scenarios. Control nodes keep track representing cities or tourist resorts. A maximum of flows versus targets or constraints and can number of such events – which could be as low as 0 represent both minimum flow requirements, – defines one possible constraint.</p><p>Kurt Fedra Page 3 4/6/2018 Initial DRAFT version</p><p>Figure 6: water quality model interface</p><p>Figure 4: Editing water allocation scenarios The models, their display, analysis, and summary pages and the interactive scenario editors are implemented as web-accessible applications: dynamic pages generated server side with cgis and PHP can be viewed and edited from any web browser by a user with the required access privileges and authentication.</p><p>CALIBRATION AND OPTIMIZATION Calibration and optimisation are conceptually similar: we consider the set of models as a transfer function that translates a point in the parameter space (including decision variables) into a point in a behaviour space of high dimensionality. The acceptable sub-region in the behaviour space that defines a feasible scenario can represent both observations as well as expectations of the systems performance (Fedra, 1983). </p><p>Figure 5 : Node specific model results A simple example is the calibration of the rainfall- runoff model: rather than using a least-square The water quality model that post-processes the approach, we define a set of constraints in terms of daily flow regimes from the water resources inequalities or ranges around a given concept that scenarios adds additional criteria for describing and the model has to meet. The constraints can be evaluating a given scenario in terms of meeting or derived from the observation data, but also violating water quality standards in space and time. represent expert knowledge. They can represent The dynamic (daily) model shares the network and ranges or boxes around observed flow, but also the uses the flow scenarios generated by the water relationships between various observation data or resources model, and adds pollutant load data to the derived metrics, e.g., to describe hydrograph return flow from demand nodes (domestic, recession, ratios between minimum maximum industrial, agricultural) as well as from lateral flows, averages or totals over selected periods, and inflow to the individual reaches. in general, any or all patterns that can define plausible expected behaviour or Gestalt of the system modelled rather than some statistical properties of highly uncertain observation data.</p><p>Kurt Fedra Page 4 4/6/2018 Initial DRAFT version</p><p> solutions. Progress is being made by iteratively tightening the constraints, both by making acceptable ranges smaller, and by adding new concepts or dimensions. Since several population can be explored independently, an efficient task parallel implementation on several multi-processor machines is possible to speed up the optimisation process.</p><p>Rather then using a completely random approach, we formulate a set of rules or heuristics that exploit knowledge about the effects of parameter changes: depending on the constraints violated, we can identify sub-sets of the parameter vector and preferred directions for modification that are more likely to succeed: rather than blind mutations and selection, we breed solutions more effectively.</p><p>Figure 7: rainfall-runoff model scenario The results of the optimisation process are several possible solutions that meet all constraints specified – this of course can be interpreted as meeting a minimal value of an objective function, is considerably more flexible and can represent several concepts in their natural units and an easy to understand multi-criteria and multi-objective model. These solutions are now available for a final, interactive step with the direct involvement of end users and stakeholders to define their preference structures in terms of implicit trade-offs, using a reference point approach (Wierzbicki, 1998). An important advantage of the methodology is in its intuitive and simple interface, and the minimisation of assumptions that must be made.</p><p>DECISION SUPPORT The primary objective of the OPTIMA project is to contribute to better policy and decision making for Figure 8: rainfall-runoff model results resource management in the coastal zone. The basic elements are reliable data and information, The same concept is used for the optimization of exploiting modern information technology, an the water resources model. Using a combination of integration of qualitative and quantitative tools for brute forward Monte Carlo and local stochastic hill rational and scientifically based design and analysis climbing, we can identify sets or populations of of options and alternatives, and the support for scenarios (parameter sets) the meet a given set of wide participation in the decision making process. constraints. These sets are used as the basis of DSS based on optimisation technologies are a several cycles of evolutionary programming, using central element of operations research, and an a standard set of operators on the respective established technology in water resources research parameter vectors. These operators affect both (Loucks et al., 1981). Their practical applicability structural binary changes (choice of structures, for complex problems, however, is limited by the allocation rules and water technologies) as well as fact that efficient optimisation requires a sometimes continuous modification (the scaling of technology gross simplification (usually based on linearization) parameters). Parameters that affect the same object of the problem to arrive at an optimal solution with in the system are grouped together, and are more guaranteed convergence. A secondary problem is likely to be exchanged as a group rather than that the formalisation and related abstraction and individually, thus preserving cross-correlation simplifications make assumptions and results patterns within the set or population of feasible difficult to understand and communicate, which</p><p>Kurt Fedra Page 5 4/6/2018 Initial DRAFT version hinders broad participation in the decision making Management: Risks and Challenges for the 21st process and thuds often generates barriers to the Century. Proceedings of the EWRA Symposium, actual implementation of technically optimal September 2-4, 2004, Izmir, Turkey, Volume I, solutions. Brute forward numerical optimisation, 23-38 pp. based on simulation modelling can retain a Fedra, K. (2002) GIS and simulation models for sufficiently detailed, realistic description. However, Water resources Management: A case study of the combinatorial explosion of alternatives makes the Kelantan River, Malaysia. 39-43 pp., GIS an exhaustive search of the decision space Development, 6/8. impossible for even moderately complex problems. Fedra, K. (2001) Environmental RTD projects and An alternative is the introduction of domain applications over the Internet: Case studies from specific heuristics in a multi-tiered approach, using Thailand, China, Malaysia and Japan. Paper rule-based expert systems, and genetic algorithms, presented at the Asia eco-Best Roundtable in Lisbon, March 2001. which can make the search much more efficient than traditional methods. Iterating between different Fedra, K. (2000) Environmental Decision Support levels of aggregation and representation, Systems: A conceptual framework and application examples. Thése prèsentèe á la evolutionary strategies and local stochastic gradient Facultè des sciences, de l'Universitè de Genéve search, a screening level approach and the use of pour obtenir le grade de Docteur és sciences, evolutionary concepts of good enough rather than mention interdisciplinaire. 368 pp., Imprimerie de optimal can lead to efficient solutions even for very l'Universitè de Genéve, 2000. complex and large-scale problems. Fedra, K. (1983) A Monte Carlo Approach to Estimation and Prediction. In: M.B. Beck and G. This innovative approach to the optimisation of van Straten [eds.]: Uncertainty and Forecasting of complex, non-linear, distributed and dynamic Water Quality, Springer-Verlag, Berlin. pp.259- systems is embedded in a framework of interactive, 291. participatory decision support based on a secondary Fedra, K. and Feoli, E. (1998) GIS Technology and layer of discrete multi-criteria optimisation. Using a Spatial Analysis in Coastal Zone Management. reference point approach to simplify the expression EEZ Technology, Ed. 3, 171-179. of preferences and trade-offs, this combined Fedra, K., and Jamieson, D.G. (1996) An object- method supports interactive, exploratory use and oriented approach to model integration: a river aims at easy integration in the planning and policy basin information system example. In: Kovar, K. making process, facilitating the participation of all and Nachtnebel, H.P. [eds.]: IAHS Publ. no 235, major institutions, actors and stakeholders in water pp. 669-676. resources, which is more or less everybody, i.e., the Jamieson, D.G. and Fedra, K. (1996) The public. WaterWare decision-support system for river basin planning: I. Conceptual Design. Journal of Hydrology, 177/3-4, pp. 163-175. ACKNOWLEDGEMENTS Loucks, D.P., Stedinger, J.R. and Haith, D.A. The OPTIMA project (INCO-CT-2004-509091) is (1981) Water resources Systems Planning and funded, in part, by the European Commission, DG Analysis. 559 pp, Prentice Hall, Englewood Research, under the FW6 INCO-MPC program. Cliffs, NJ. Michalewicz, Z. and Fogel, D.B. (1998) How to Solve It: Modern Heuristics. 467 pp, Springer, Berlin. REFERENCES Byron, M. (1998) Satisficing and Optimality. Ethics Simon, H.A. (1969) The Science of the Artificial. 109, 67-93. MIT Press, Cambridge, MA. Fedra, K. (2004) Coastal Zone Resource Simon, H.A. (1957) Models of Man. Social and Management: Tools for a Participatory Planning Rational. Wiley, NY. and Decision Making Process. In: Green, D.R. et Wierzbicki, A. (1998) Reference Point Methods in al. [eds.]: Delivering Sustainable Coasts: Vector Optimization and Decision Support. IR- Connecting Science and Policy. Proceedings of 98-017, 43 pp ; International Institute for Applied Littoral 2004, September 2004, Aberdeen, Systems Analysis, Laxenburg, Austria. Scotland, UK., Volume 1, 281-286 pp. Fedra, K. (2004) Water Resources Management in the Coastal Zone: Issues of Sustainability. In: Harmancioglu, N.B., Fisitikoglu, O., Dlkilic, Y, and Gul, A. [eds.]: Water Resources</p><p>Kurt Fedra Page 6 4/6/2018</p>