A decision support system for optimization of regional drinking water supply

Abstract

Finding a strategy that allows economically efficient drinking water production in regional supply systems at minimal environmental cost is often a complex task. In order to determine the optimal spatial production configuration, a systematic trade off among costs and benefits of possible strategies is required. Such a trade-off involves the handling of pronounced non-linear relations between quantitative aspects of strategies and their corresponding impacts. We developed a computer-based methodology for multiple objective optimisation of drinking water production by combining 'Min Cost Flow' and Genetic Algorithms (GA). The impact of production strategies is assessed by environmental, economic and geo-hydrologic modelling. Finding the optimal solution requires valuation of objective categories by translating impacts into a common scale and/or by definition of constraints that are specific for a particular category. If the impact of a category cannot be converted a priori to a common scale, a Pareto frontier of non-inferior solutions is calculated. Thus, the interdependency of impact categories can be clarified and decision makers and stakeholders are facilitated in the selection of appropriate production strategies. The approach was implemented in a GIS-based decision support system in order handle all spatial relations efficiently and to offer decision makers an adequate access to the methodology. Groundwater quality prediction studies are frequently carried out within the framework of drinking water supply in order to assess the future composition of groundwater that will be pumped at production wells. These prediction studies help to assure a safe supply of drinking water in the future. Regional drinking water companies typically exploit numerous pumping wells and need to decide on research priorities for these wells as budgets are limited. Assessment of the uncertainty of prediction studies has been a scientific topic for many years, particularly when numerical models are used as predictive tools. Sophisticated techniques for the quantification of the uncertainty of model results have been developed over the past decades. In sharp contrast to the progress on the level of model uncertainty is prioritisation of prediction studies still generally based upon ‘expert judgement’. Very few studies have focussed on the question how uncertainty of predictions on the composition of pumped groundwater should be used for management decisions on research priorities. However, deciding on these research strategies has become more complex, due to the increased size and interdependency of regional drinking water supply systems. Consequently, there is a need for decision support methods in order to avoid sub-optimal strategies. This report presents a framework that is based on the above-mentioned methodology for multipleobjective optimisation of drinking water production. It enables decision support for allocation of research priorities to groundwater quality prediction studies. Rational research strategies on groundwater quality prediction seek to minimize the risk of well failure due to contamination of groundwater (breakthrough). There are 3 elements that form the basis of our approach: • the quantification of risks for drinking water supply due to groundwater pollution • an operational quantification of the reliability of predictions • the anticipated marginal precision efficiency of additional prediction studies The minimal negative impact of well failure in both economic and environmental terms is assessed by using genetic algorithms