Groundwater serves as a critical resource for India’s water security and agricultural sustainability. However, increasing urbanization and changing climatic patterns have intensified pressure on this vital reserve, especially in rapidly growing urban centers. This study investigates groundwater level (GWL) fluctuations across 54 Indian cities with populations exceeding one million, analyzing their interactions with key climatic parameters, such as temperature, precipitation, and evapotranspiration, along with population growth. The model performance varied substantially across R 2 values ranging from 0.507 to 0.991 reflecting strong predictability in climate driven aquifer systems but reduced skill in cities where groundwater dynamics are dominated by intensive abstraction, complex hydrogeology or unobserved anthropogenic controls. We developed a city specific machine learning framework using Extreme Gradient Boosting (XG Boost), incorporating lagged climatic variables, drought indices and population proxies, and benchmarked its performance against linear regression and random forest models. Unlike previous localized or single model studies, this work provides the first consistent, pan-Indian urban groundwater assessment that explicitly evaluates model skill heterogeneity across climatic regimes and level of urban influence. Feature importance analysis revealed population and its temporal lags as dominant predictors (40-85% contribution), while lagged drought indices (PDSI lag2, PDSI Anomaly lag2) emerged as critical secondary drivers accounting for 35-40% of climate-driven importance, demonstrating 2-3 months aquifer response timescales. Spatial analysis identified northern alluvial plains as climate insensitive, population dominated systems (mean R 2 = 0.94), while peninsular hard-rock aquifers retained strong climate groundwater coupling, indicating regional vulnerability to combined anthropogenic climate stress. These findings highlight the pressing need for adaptive and data-driven groundwater management policies in urban India. The XGBoost model provides a powerful tool for understanding and predicting groundwater dynamics, offering valuable insights for policymakers and urban planners aiming to ensure long-term water sustainability. • Optimized XG Boost model achieved exceptional groundwater prediction performance (R 2 = 0.507-0.991) across 54 Indian cities, with 45 cities exceeding R 2 = 0.80, demonstrating robust capture of spatially heterogeneous aquifer dynamics. • Population and its temporal lags dominated feature importance (40-85% contribution), with permutation analysis revealing catastrophic model collapse ( R 2 > 1.0) in 8 cities upon population removal, establishing anthropogenic extraction as the primary groundwater depletion driver. • Lagged drought indices (PDSI lag2, PDSI Anomaly lag2) emerged as critical secondary predictors, accounting for 35-40% of climate-driven importance and reflecting 2-3 month aquifer response timescales to precipitation anomalies and recharge processes. • Spatial analysis revealed northern alluvial plains exhibit climate-insensitive, population-dominated systems (mean R 2 = 0.94), while peninsular hard-rock aquifers retain strong climate-groundwater coupling, indicating regional vulnerability to combined climate-extraction stress • The study argues against a one-size-fits-all approach and stresses that groundwater management must be tailored to each region’s unique climate and geological context.
Maurya et al. (Sun,) studied this question.