Abstract
<title>Abstract</title> <p>Groundwater contamination in semi-arid regions is increasingly influenced by complex hydrogeochemical interactions and anthropogenic activities, necessitating adaptive and interpretable environmental intelligence frameworks for sustainable groundwater management. In this study, an integrated GeoAI–Machine Learning–Explainable Artificial Intelligence (GeoAI–ML–XAI) framework was developed to evaluate groundwater quality in 50 villages of Beenagunj Block, Madhya Pradesh, India. Pollution Index of Groundwater (PIG) values were computed using major hydrochemical parameters including TDS, chlorides, fluoride, nitrates, hardness, conductivity, alkalinity, turbidity, and pH. Twenty-eight machine learning regression algorithms were employed for PIG prediction, among which the Trilayered Neural Network demonstrated the highest predictive accuracy (R² = 0.996; RMSE = 0.009), while Gaussian Process Regression models exhibited robust and stable performance. Explainable AI techniques including SHAP and LIME identified TDS, chlorides, conductivity, and total hardness as dominant groundwater contamination drivers. To improve interpretability reliability, a novel Explainability Stability Index (ESI) was proposed to quantify the consistency of hydrochemical feature contributions across multiple algorithms. Furthermore, an Explainable Groundwater Quality Index (XGWQI) and Explainable Groundwater Vulnerability Index (EGVI) were developed by integrating SHAP-based adaptive weighting, spatial vulnerability behavior, and explainability stability into groundwater assessment. The proposed GeoAI-XAI framework substantially improves interpretability transparency, hotspot identification, adaptive groundwater quality assessment, and sustainability-driven environmental decision-support compared with conventional groundwater quality indices.</p>