Spatially Optimised Approach for Predicting Water Quality in a Heterogeneous Agricultural Watershed

Area of Science:

• Environmental Science
• Water Resource Management
• Machine Learning Applications

Background:

• Predicting water quality in heterogeneous watersheds is complex due to spatial variations in parameters and accuracy.
• Spatially adaptive machine learning models offer a solution for localized water quality prediction.

Purpose of the Study:

• To develop and evaluate spatially adaptive machine learning models for predicting irrigation water quality.
• To identify key water quality parameters for enhancing prediction accuracy in different watershed clusters.

Main Methods:

• Calculated the Irrigated Water Quality Index (IWQI).
• Identified spatial clusters of water quality stations based on physiochemical characteristics.
• Developed and assessed six machine learning models (SVM, RF, ET, XGBoost, DT, BRT) for each cluster.
• Conducted sensitivity analysis to determine key prediction parameters.

Main Results:

• Three distinct clusters of water quality stations were identified based on water characteristics.
• The XGBoost model demonstrated the highest accuracy across all clusters (R²=0.99, RMSE=0.02-0.05).
• Locally optimized models effectively predicted water quality using specific parameters within each cluster.

Conclusions:

• Spatially adaptive models are effective for predicting watershed-specific irrigation water quality.
• Findings support timely alerts for water quality management in adaptive agricultural development.
• XGBoost shows superior performance for localized water quality prediction in heterogeneous environments.