Exploring hydrochemical drivers of drinking water quality in a tropical river basin using self-organizing maps and explainable AI

Groundwater quality assessment is essential for ensuring sustainable water resource management, particularly in regions heavily dependent on groundwater for domestic and agricultural needs. This study aims to investigate the hydrochemical characteristics and assess the drinking water quality of groundwater in the Ithikkara River Basin, a tropical region in southern India, by integrating hydrogeochemical methods with unsupervised learning and explainable artificial intelligence (XAI). A total of 111 groundwater samples were analysed for major ions, hydrochemical facies, and water quality indices. Self-Organizing Maps (SOM) identified three distinct groundwater clusters, each exhibiting unique geochemical signatures. Hydrochemical facies analysis revealed dominant Na⁺-Cl⁻ and mixed Ca²⁺-Na⁺-HCO₃⁻ types, influenced by silicate weathering, cation exchange, and anthropogenic activities. The Entropy Water Quality Index (EWQI) showed that 89.4 % of samples were of excellent to good quality, with moderate-quality zones located near agricultural and industrial areas. A Random Forest-based ensemble model achieved high predictive accuracy (R² = 0.871), and SHAP analysis revealed Na⁺, K⁺, and TDS as the primary contributors to water quality degradation. The integration of SOM clustering with interpretable machine learning offers a powerful framework for understanding groundwater evolution and guiding sustainable water management in tropical river basins.