Machine Learning Based Water Quality Evolution and Pollution Identification in Reservoir Type Rivers

Area of Science:

• Environmental Science
• Water Resource Management
• Remote Sensing
• Machine Learning Applications

Background:

• Quantifying pollutant transport and transformation in reservoir-regulated rivers is challenging due to complex hydrodynamic and biogeochemical interactions.
• Understanding these dynamics is crucial for managing water quality in vital river systems like the Yulin River, a tributary of the Three Gorges Reservoir.

Purpose of the Study:

• To explore the spatiotemporal dynamics of water quality parameters (WQPs) in the Yulin River using field monitoring and satellite imagery.
• To evaluate the performance of advanced machine learning algorithms for retrieving WQPs, including chemical oxygen demand (COD), total phosphorus (TP), total nitrogen (TN), and chlorophyll-a (Chla).
• To assess the influence of hydrological regulation and meteorological factors on pollutant dynamics.

Main Methods:

• Combined 48 months of high-frequency field monitoring data (January 2020 to December 2023) with Sentinel-2 multispectral imagery.
• Systematically evaluated four machine learning algorithms: Extreme Gradient Boosting (XGBoost), Random Forest (RF), Categorical Boosting (CatBoost), and Gradient Boosted Decision Trees (GBDT).
• Analyzed the impact of reservoir impoundment and hydrological conditions (low-flow vs. high-flow) on WQPs.

Main Results:

• XGBoost demonstrated superior performance in WQP retrieval, with R² values from 0.9154 to 0.9488 and low RMSE.
• Reservoir impoundment caused significant Chla surges (100%-1000% increase in 56.2% of the area) and moderate TN fluctuations (≤40% growth in 73% of the area).
• Hydrological conditions strongly influenced COD and Chla in estuarine regions, with higher concentrations during low-flow periods. Meteorological factors showed weak correlations (|r| < 0.41).

Conclusions:

• XGBoost is a robust tool for large-scale WQP retrieval and watershed-scale monitoring, offering high precision (MAE 0.0201-1.4277 mg/L).
• Hydrological regulation, particularly reservoir impoundment, is the predominant driver of pollutant dynamics in the Yulin River.
• The findings provide critical insights for effective management of river ecosystems influenced by reservoir operations.