Groundwater quality prediction using novel hybrid classification and regression models.

Area of Science:

• Environmental Science
• Hydrogeology
• Machine Learning Applications

Background:

• Groundwater quality degradation is a critical issue in the North China Plain (NCP), impacting water supply.
• Traditional groundwater quality assessment methods are resource-intensive and time-consuming.
• The NCP relies heavily on groundwater, necessitating efficient quality monitoring and management strategies.

Purpose of the Study:

• To develop and evaluate novel hybrid machine learning models for predicting groundwater quality in Cangzhou, NCP.
• To compare the performance of various hybrid models, including Stacking Classifiers, LightGBM-PSO, GA-RF, and GA-LightGBM.
• To identify key hydrochemical parameters influencing groundwater quality using SHAP analysis.

Main Methods:

• Collected and analyzed 460 groundwater chemistry samples from Cangzhou.
• Applied hybrid machine learning models: XGBoost, LightGBM, LightGBM-PSO, GA-RF, and GA-LightGBM for classification and regression.
• Utilized SHAP analysis to determine feature importance for groundwater quality prediction.

Main Results:

• Hybrid models LightGBM-PSO and GA-RF achieved high prediction accuracies (up to 99%).
• The GA-RF model demonstrated excellent performance with R²=0.99, RMSE=0.01, and MAE=0.02.
• Hydrochemistry revealed HCO₃⁻, Cl⁻, Ca²⁺, and SO₄²⁻ as dominant ions, with water-rock interactions, evaporation, and salinization as key influences. Water Quality Index (WQI) showed 60.28% excellent/good and 40% poor/very poor quality.

Conclusions:

• Hybrid machine learning models are highly effective for accurate groundwater quality prediction.
• Key factors influencing groundwater quality include Na⁺, HCO₃⁻, F⁻, Cl⁻, Ca²⁺, SO₄²⁻, Mg²⁺, TDS, and EC.
• The study provides valuable insights for sustainable groundwater resource management in the NCP and globally.