Emily Kumpel, Casey Brown

Publication Date



Nearly 1 billion people around the world experience intermittent water supply (IWS), including about 70% of residents in the Mexico City area. Households with IWS often rely on multiple sources of water to meet their needs, including municipal piped water, trucked water, and rainwater. When calculating water costs and reliability of supply, models of these systems must account for household decision-making regarding the volume of water to use from each different source each day. Modeling these household decisions (or “control policies”) is challenging, especially when households use rainwater as a water source, due to the complexity of the input variables involved (e.g. intermittent water schedule, season, day of week), but is critical to understanding the role that household-level interventions, such as household storage and rainwater harvesting, may play in water access. Universal approximators provide a solution to this challenge by allowing for flexible shaping of these control polices. This study uses Radial Basis Function Networks to determine optimal household water management decisions, maximizing reliability of water supply while minimizing costs for an arbitrary household in Mexico City. The model design is informed using data collected during interviews with households in the city. The model produces Paretooptimal solution sets that demonstrate which household-level investments are most effective for improving the reliability of water access. Results show that household storage tanks are a critical component of water access, especially in households with very low access to the municipal piped water supply. A tank volume of around 1500-2500 liters can provide most of the savings, depending on the availability of municipal water, although a larger tank is better able to collect rainwater. IWS households with sufficient storage are able to meet their water needs with piped water nearly as reliably as those with continuous water supply, as long as a minimal threshold of water is delivered. When household storage is limited, households are more vulnerable to disruptions in the piped network, and costs increase if supply is not delivered consistently. Rainwater harvesting systems are shown to be economically viable at the DocuSign Envelope ID: 3F0D957B-51D2-42AE-9AE7-C1921FE89C34 iv household level regardless of the frequency of municipal piped water service. The techniques presented in this study are a crucial step in modeling water resources in cities with IWS.