A zone-specific water quality modelling framework for medium and large lakes using interpretable machine learning.

Area of Science:

• Environmental Science
• Water Quality Management
• Machine Learning Applications

Background:

• Spatial heterogeneity in lakes complicates water quality prediction and management.
• Accurate modeling is crucial for understanding and mitigating pollution in medium and large lakes.

Purpose of the Study:

• To develop an interpretable, zone-specific machine learning (ML) framework for enhanced lake water quality prediction.
• To improve the understanding of dynamic changes in water quality parameters and their drivers.
• To apply the framework to predict total nitrogen (TN) and total phosphorus (TP) in Yangcheng Lake.

Main Methods:

• Developed a zone-specific ML modeling framework using aggregated and segregated strategies.
• Employed Extreme Gradient Boosting (XGBoost), Random Forest (RF), and Support Vector Regression (SVR) algorithms.
• Applied the framework to predict TN and TP concentrations and identify change drivers in Yangcheng Lake.

Main Results:

• The XGBoost algorithm with an aggregated modeling strategy demonstrated optimal performance in predicting water quality across all lake zones.
• Air temperature was the primary driver for TN prediction, while water level was key for TP prediction.
• Tributary inputs and endogenous loads significantly influenced nutrient levels, with varying importance across different lake zones (western, middle, eastern).

Conclusions:

• The proposed ML framework effectively models spatial heterogeneity in lake water quality.
• The findings provide a promising approach for water quality prediction and pollution management in medium and large lakes.
• Understanding zone-specific drivers is crucial for targeted lake management strategies.