Irrigation-Driven Groundwater Recharge and Quality Degradation in Semi-Arid Regions: Hydrochemical, GIS-Based, and Explainable Machine Learning Assessment in Central Tunisia

Groundwater resources in semi-arid regions are increasingly threatened by agricultural intensification, irrigation expansion, and climate variability. This study investigates the influence of irrigation practices on groundwater recharge and quality degradation in the semi-arid Regueb Basin, Central Tunisia, using an integrated framework combining hydrochemical analysis, irrigation water quality indices, GIS-based spatial modeling, and explainable machine learning (XAI). Thirty groundwater samples were analyzed for major physicochemical parameters and irrigation suitability indicators, including Electrical Conductivity (EC), Total Dissolved Solids (TDS), Sodium Adsorption Ratio (SAR), sodium percentage (%Na), and Irrigation Water Quality Index (IWQI). Hydrochemical facies are dominated by Ca–Mg–Cl, Na–Cl, and Ca–Mg–SO4 water types, reflecting the combined effects of evaporite dissolution, water–rock interaction, evaporation, and irrigation return flow. Groundwater salinity is generally high, with EC values ranging from 1490 to 8710 µS/cm, while nitrate concentrations frequently exceed the World Health Organization guideline value of 50 mg/L in intensively cultivated zones, indicating significant anthropogenic contamination linked to fertilizer leaching and irrigation practices. GIS-based recharge assessment indicates that irrigation return flow may represent an important component of effective recharge in cultivated areas under semi-arid conditions while simultaneously contributing to salinization and nutrient accumulation within the aquifer system. However, quantitative partitioning of recharge sources requires further investigation using tracer-based approaches and numerical modeling. More than 40% of groundwater samples were classified as unsuitable for irrigation because of elevated salinity and sodicity hazards. To explore the relationships among irrigation water quality indicators, several machine-learning algorithms were evaluated for IWQI estimation and interpretation. Linear Regression achieved the highest performance for IWQI estimation (R2 = 0.9839), reflecting the strong internal relationships among irrigation water quality indicators. SHapley Additive exPlanations (SHAP) analysis identified Residual Sodium Carbonate (RSC) as the most influential parameter controlling irrigation water quality. The results highlight the dual role of irrigation as both a recharge-enhancing mechanism and a driver of groundwater degradation. This study provides an integrated hydrochemical–GIS–XAI framework for identifying vulnerable zones and supporting sustainable groundwater management strategies in semi-arid agricultural regions.