Predicting Turbidity and Total Organic Carbon Changes under Climate Change in Water Supply Systems using Machine Learning

Abstract Impacts of climate change on surface water quality are a concern for water utilities. Adapting to these changes requires accurate predictive models that can assess the effects of climate change on water quality. In this study, we developed machine learning models to support climate change impact assessments, due to their low computational cost and reduced complexity in required input data. We integrated a weather generator to produce scenarios of precipitation and air temperature, along with a water demand model, a hydrologic model, a water system model, and a machine-learning based water quality model. Using upstream hydroclimatic and system state variables, we predicted turbidity and total organic carbon at the Tesla Treatment Facility, which is downstream of the Hetch Hetchy Regional Water Supply System. Our results show that changes in precipitation have a stronger impact on water quality than changes in air temperature or water demand. Increased precipitation intensifies the frequency, duration, and severity of water quality extremes, while also shortening their return periods. These findings provide valuable insights for water managers in developing water quality management plans under climate risk.