Evaluating groundwater potential with the synergistic use of geospatial methods and advanced machine learning approaches

The rapid increase in population, urbanization, and industrial activity in developing countries is intensifying pressure on groundwater resources, leading to severe water shortages. This study aims to evaluate and compare the predictive capabilities of six ensemble machine learning (ML) models; i.e., Random Forest (RF), AdaBoost, Neural Network, Decision Tree, k-Nearest Neighbors and Extreme Gradient Boosting. For the delineating groundwater potential zones by integrating ML algorithms with Geographic Information System (GIS) tools, offering a novel approach for groundwater resource mapping. Eleven conditioning factors, including elevation, slope, soil types, geomorphology, degree of aspect, rainfall, land use/land cover, stream power index, topographic wetness index, and land surface temperature, were used as input parameters. Model performance was evaluated using multiple metrics, including Area Under the Curve (AUC), Classification Accuracy, F1 Score, Precision, Recall, and Matthews Correlation Coefficient (MCC). The results revealed that RF was the most accurate model AUC (0.91), mapping the largest areas for very high 346 sq. km and low 486 sq. km zones. AdaBoost, effective with imbalanced data, achieved the highest MCC (0.672). Sensitivity analysis revealed that geomorphology, elevation, and rainfall were the most influential parameters for groundwater potential zoning. This study highlights the potential of ensemble ML models in advancing groundwater resource assessment and offers a foundation for further exploration in urban regions facing water scarcity challenges, and identifies priority areas for sustainable water use and planning.