Groundwater vulnerability under climate change : a machine learning framework

Abstract

Groundwater quality assessment is essential for sustainable water management but remains challenging owing to hydrogeological complexity, climate variability, and anthropogenic pressures. This study evaluates groundwater quality in the East Nile River region of Sohag, Egypt, by integrating hydrochemical parameters with machine learning (ML) approaches. The aim is to identify spatial contamination risks, assess water suitability, and explore the implications of environmental change. In total, 78 groundwater samples were collected and analyzed for 15 hydrochemical and ion-derived parameters. Two ML algorithms, Artificial Neural Networks (ANN) and Extreme Gradient Boosting (XGBoost), were applied, with datasets divided into training (70%) and testing (30%) subsets. The models generated probability maps of groundwater vulnerability, highlighting central areas, particularly near agricultural zones with shallow water tables, as most at risk. ANN outperformed XGBoost (R² = 0.89 vs. 0.86), demonstrating higher predictive accuracy for groundwater quality assessment. Spatial variability was linked to hydrogeological processes and intensified urban and agricultural activities, while climate-related stressors are expected to exacerbate groundwater deterioration. The proposed ML framework provides a reliable, cost-effective tool for groundwater monitoring and early warning, supporting improved decision-making for water sustainability. These findings contribute to global efforts toward SDG 6 (Clean Water and Sanitation), SDG 12 (Responsible Consumption and Production), and SDG 13 (Climate Action).