Machine learning approach towards explaining water quality dynamics in an urbanised river

Area of Science:

• Environmental Science
• Ecology
• Water Quality Monitoring

Background:

• Human activities significantly impact urban river ecosystems.
• Understanding drivers of water quality is crucial for ecological restoration.
• Urban rivers face unique challenges due to human pressures.

Purpose of the Study:

• To quantify human-induced drivers of spatio-temporal water quality patterns in urban rivers.
• To assess the effectiveness of machine learning in analyzing high-frequency water quality data.
• To identify key factors influencing the ecological health of the River Chess.

Main Methods:

• Utilized high-frequency electrical conductivity and temperature data from a Citizen Science project on the River Chess.
• Employed machine learning models, including boosted trees and Generalized Additive Models (GAM).
• Applied SHapley Additive exPlanations (SHAP) to interpret model results and variable importance.

Main Results:

• Boosted trees model accurately described water quality dynamics with <1% error, outperforming GAM.
• Wastewater Treatment Works (WWTW) outflows were identified as a significant driver of diurnal electrical conductivity variations.
• WWTW outflows caused a 1°C increase in water temperature downstream during low flows.

Conclusions:

• High-frequency water quality monitoring via Citizen Science, coupled with machine learning, effectively identifies urban river pollution sources.
• Wastewater Treatment Works (WWTW) have a measurable and consistent impact on river temperature and conductivity.
• This approach provides valuable insights for managing and improving the ecological health of urban rivers.