Machine Learning and GIS-Integrated Framework for Irrigation Water Quality Index (IWQI) Mapping and Hydrochemical Clustering in an Algerian Semi-Arid Watershed

Abstract In the Ain Temouchent region, surface water is severely limited, leaving groundwater as the primary source for both domestic supply and agricultural irrigation. This study evaluates the suitability of El Melah watershed groundwater for irrigation by computing the Irrigation Water Quality Index (IWQI) following the Meireles approach, mapping its spatial distribution through GIS, and complementing the analysis with two machine learning methods: (i) a Random Forest (RF) model with Leave-One-Out Cross-Validation (LOOCV) and SHAP explainability analysis to identify the dominant drivers of IWQI degradation; and (ii) K-means and hierarchical clustering to delineate hydrochemical facies and provide a statistically grounded alternative to the conventional Wilcox diagram. Salinity is elevated throughout the study area (EC: 1,034–5,010 µS cm⁻¹). The RF model achieved LOOCV R² = 0.61, with Na⁺ and EC confirmed as the dominant IWQI degradation drivers by SHAP analysis. Hierarchical clustering resolved three hydrochemical facies: a Ca²⁺-HCO₃⁻ continental group, a Na⁺-Cl⁻ evaporite-influenced group, and a seawater-intrusion group, all confirmed as statistically distinct by one-way ANOVA (p < 0.05). Overall, groundwater quality in the watershed ranges from moderate to severe restriction for irrigation use, with the northern sector most critically affected by marine infiltration.