Published Paper Details:

Groundwater forms an important freshwater commodity in the semi-arid and hard-rock provinces of India, with its distribution governed by complex relationships between geological, geomorphological, topographical and climatic factors. The present investigation outlines a comprehensive framework for spatially prognosticating groundwater potential zones (GWPZ) in a selected section of the Mahanadi River Basin, Sambalpur district, Odisha, using GIS-based thematic analysis in conjunction with comparative machine learning methodologies. Twelve conditioning variables relevant to groundwater availability, including geology, geomorphology, land use/land cover, soil characteristics, elevation, slope, curvature, topographic position index (TPI), topographic wetness index (TWI), lineament density, drainage density and precipitation, were processed and analysed as part of a GIS environment. The identification of groundwater potential was completed through the Analytical Hierarchy Process (AHP) using the combination of seven predictive modelling paradigms: Frequency Ratio (FR), Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), k-Nearest Neighbors (KNN), Artificial Neural Network (ANN), and XGBoost. Model efficacy was evaluated based on receiver operating characteristic-area under the curve (ROC-AUC), ground truth validation accuracy and Cohen's Kappa statistics. Findings show that the Artificial Neural Network (ANN; Accuracy= 93.75% and Kappa= 0.90) and Random Forest (RF; Accuracy= 91.25% and Kappa= 0.86) models outperform the other approaches in groundwater potential forecast, whereas the Frequency Ratio (FR), although it shows a discernible trade-off of the AUC, has lost its spatial reliability. Good groundwater potential zones are mostly coincident with landscapes that have low slope, high lineament density, good geomorphologic characteristics, and moderate drainage density. These results demonstrate that the combination of GIS and machine learning models significantly enhances the accuracy and reliability of groundwater potential mapping and therefore provides an effective decision support tool for sustainable groundwater management and site selection for groundwater extraction systems in river basin areas.