Leveraging explainable machine learning for enhanced management of lake water quality

Area of Science:

• Environmental Science
• Water Quality Management
• Machine Learning Applications

Background:

• Eutrophication in freshwater lakes is a global issue driven by excess nutrient loads (nitrogen and phosphorus) from wastewater and runoff.
• This impacts aquatic ecosystems and public health, necessitating effective monitoring and management strategies.

Purpose of the Study:

• To predict key water quality parameters (total phosphorus, total nitrogen, nitrate + nitrite, turbidity) using machine learning (ML) algorithms.
• To identify spatial patterns of these parameters and determine the environmental drivers influencing turbidity levels in a large subtropical lake.
• To support operational decision-making through station-specific and lake-wide modeling approaches.

Main Methods:

• Employed four ML algorithms: Extreme Gradient Boosting (XGB), Light Gradient-Boosting Machine (LGBM), Support Vector Regression (SVR), and Random Forests (RFs).
• Utilized easily measurable variables to predict total phosphorus (TP), total nitrogen (TN), nitrate + nitrite (NOx-N), and turbidity.
• Applied K-means clustering for spatial analysis and SHapley Additive exPlanations (SHAP) to identify turbidity drivers.

Main Results:

• XGB demonstrated superior performance in predicting water quality parameters.
• Lake stage, water temperature, and turbidity were identified as significant predictors of nutrient levels.
• Spatial analysis revealed three distinct lake regions based on nutrient and turbidity levels.
• Inflows and lake stage were primary drivers of turbidity near inlets; wind speed and air temperature influenced turbidity in the lake's center.

Conclusions:

• ML models can effectively predict lake water quality parameters, aiding in management decisions.
• Turbidity is a crucial factor influencing nutrient dynamics and is driven by various environmental factors.
• Frequent monitoring of turbidity and its drivers is essential for effective lake management and mitigation of eutrophication.