Spatio‐Temporal Groundwater Vulnerability Assessment in the Paler Watershed Using GIS‐Based DRASTIC and Multialgorithm Machine Learning Approaches

ABSTRACT Evaluating groundwater vulnerability is crucial for sustainable water resource management, particularly where groundwater is heavily utilized for domestic and agricultural purposes. This paper presents a methodology combining machine learning with the GIS‐based DRASTIC model for the Paler watershed, lower Krishna subbasin, southern India. The DRASTIC model assesses seven hydrogeological characteristics: depth to water table, recharge, aquifer media, soil media, topography, vadose zone influence and aquifer conductivity. We enhanced forecast accuracy using random forest, support vector machines, XG boost, artificial neural networks and long short‐term memory networks. The study area, with 859 mm annual rainfall, was characterized using 104 observation wells over 15 years; five modelling approaches are compared with modified DRASTIC‐L across land use datasets for 2017, 2021, 2023 and 2024. SVM and XG Boost outperformed other methods, achieving R 2 of 0.76–0.81, whereas LSTM and ANN performed poorly with negative NSE values. High‐vulnerability zones expanded from 35% in 2017 to 43% in 2024, driven by water‐table decline and agricultural intensification. SVM‐predicted vulnerability indices correlated strongly with observed nitrate concentrations ( r = 0.82, p < 0.001), confirming predictive validity. Integrating GIS‐based DRASTIC with machine learning provides a transferable framework for groundwater protection across data‐limited semiarid regions.