Advancing Stochastic Water Quality and Simulation-Optimization Techniques for Potable Water Systems Facing Source Water Quality Degradation

Water scarcity in the United States has forced drinking water utilities to treat degraded sources—such as brackish water and wastewater effluent—to meet drinking water demands. Furthermore, land development and long-term changes in climate will likely degrade the quality of traditional water supplies. Optimizing water treatment operations is one way to maintain the reliability of potable water systems in response to decreased source water quality. Current treatment practices focus heavily on engineering experience to inform operational decisions. Computer-based decision-making tools, known as decision support systems (DSSs), can be used to generate operating policies that improve the reliability of drinking water quality and cost effectiveness of treatment. However, limited source water quality data and lack of trusted in water treatment simulation models are major roadblocks for the development and adoption of DSSs. The goal of this research is to address these issues by adapting water resources planning methods (i.e., stochastic streamflow generation, reservoir modeling, and multi-objective optimization) to support potable water system decision making. Specifically, we adapt a multivariate k-nearest neighbor (k-NN) time series resampling technique—widely applied for streamflow and weather generation—to create source water quality scenarios. To simulate extremes in water quality, we implement a modified k-NN which uses random errors to generate unobserved values. We demonstrate the value of this method by applying it to a water quality dataset from a utility in Northern Colorado. The ensembles generate a variety of water quality scenarios and capture the statistical properties of the historical data. Moreover, the ensembles serve as inputs to a DSS composed of a water treatment model coupled with multi-objective optimization system. The DSS enables planners to generate a Pareto optimal operating policies with respect to critical treatment metrics, such as short-term risk, long-term risk, and cost. This coupled stochastic simulation and DSS offers a unique and powerful framework to explore system performance under various influent water quality and operating policy scenarios.