Environmental Drivers and Predictive Modeling of E. coli Variability in an Urban Stream Using Machine Learning...

Area of Science:

• Environmental Science
• Machine Learning Applications
• Water Quality Management

Background:

• Urban stream water quality is complex and variable, challenging traditional monitoring.
• Wastewater-influenced systems like the Las Vegas Wash require advanced analysis for effective management.

Purpose of the Study:

• To identify key drivers of Escherichia coli (E. coli) variability in a wastewater-influenced urban stream.
• To compare the performance of different machine learning models in predicting E. coli levels.
• To understand spatial differences in environmental controls on water quality.

Main Methods:

• Integration of over 25 years of hydrological, climatic, and water quality data.
• Application of interpretable machine learning (XGBoost and Random Forest) to analyze E. coli drivers.
• Spatial analysis of predictor variables at downstream and upstream monitoring sites.

Main Results:

• Ensemble models revealed distinct spatial controls on E. coli dynamics.
• Downstream E. coli was influenced by electrical conductivity, solids, climate, and nutrients.
• Upstream E. coli was more sensitive to storm events, flow, nutrients, and perchlorate.
• XGBoost model achieved higher accuracy (R²=0.67 downstream, R²=0.81 upstream) than Random Forest.

Conclusions:

• Environmental drivers of surface water quality vary significantly within connected urban systems.
• Interpretable machine learning is valuable for understanding complex environmental interactions.
• Findings provide a framework for improved water quality management in urban and water reuse systems.