Environmental drivers and predictive modeling of E. coli variability in an urban stream using machine learning approaches.

Surface water quality in urban streams is shaped by complex and spatially variable environmental interactions, limiting the effectiveness of conventional monitoring approaches. This study integrates over 25 years of hydrological, climatic, and water quality data with interpretable machine learning to identify key drivers of Escherichia coli variability in the Las Vegas Wash, a wastewater-influenced urban system. Ensemble models revealed distinct spatial controls on microbial dynamics. At the downstream site, electrical conductivity, total dissolved and suspended solids, climate variables, and nutrient concentrations were dominant predictors, reflecting the influence of treated wastewater and associated ionic and nutrient signatures. In contrast, upstream variability was more strongly governed by storm events, flow, nutrients, and perchlorate, indicating greater sensitivity to episodic hydrological inputs and localized sources. XGBoost outperformed Random Forest, achieving R2 values of 0.67 and 0.81 at downstream and upstream sites, respectively. These findings demonstrate that environmental drivers of surface water quality can diverge significantly within a single connected system due to spatial heterogeneity in source contributions and transport processes. This study highlights the value of interpretable machine learning for resolving complex environmental interactions and provides a framework for improving predictive water quality management in urban and water reuse-dominated systems.