Data-driven modeling of municipal water system responses to hydroclimate extremes

Sustainable western US municipal water system (MWS) management depends on quantifying the impacts of supply and demand dynamics on system infrastructure reliability and vulnerability. Systems modeling can replicate the interactions but extensive parameterization, high complexity, and long development cycles present barriers to widespread adoption. To address these challenges, we develop the Machine Learning Water Systems Model (ML-WSM) - a novel application of data-driven modeling for MWS management. We apply the ML-WSM framework to the Salt Lake City, Utah water system, where we benchmark prediction performance on the seasonal response of reservoir levels, groundwater withdrawal, and imported water requests to climate anomalies at a daily resolution against an existing systems model. The MLWSM accurately predicts the seasonal dynamics of all components; especially during supply-limiting conditions (KGE > 0.88, PBias < +/- 3%). Extreme wet conditions challenged model skill but the ML-WSM communicated the appropriate seasonal trends and relationships to component thresholds (e.g., reservoir dead pool). The model correctly classified nearly all instances of vulnerability (83%) and peak severity (100%), encouraging its use as a guidance tool that complements systems models for evaluating the influences of climate on MWS performance.

Automated summary

The study develops a Machine Learning Water Systems Model (ML-WSM) to address the challenges in sustainable western US municipal water system (MWS) management. By applying the ML-WSM to the Salt Lake City water system, the researchers demonstrate that the model can accurately predict the seasonal dynamics of the different components and classify instances of vulnerability and severity. The findings suggest that the ML-WSM can be used as a guidance tool to evaluate the influences of climate on MWS performance.