GEOSPATIAL MODELING OF GROUNDWATER DEPLETION AND ITS IMPACT IN PARTS OF NORTHWEST INDIA

Abstract

Groundwater is a critical resource for agriculture, industrial growth, and domestic use, particularly in semi-arid and arid regions of Northwest India, where surface water is limited. However, in recent decades, over-extraction of groundwater, primarily for irrigation, has led to significant groundwater depletion in areas such as the Malwa region of Punjab. This research employs geospatial modeling, machine learning, and remote sensing techniques to assess groundwater depletion trends and predict future impacts, offering a systematic analysis over a 21-year period (1997–2018) using groundwater level (GWL) data from 90 wells in the region. The study aims to provide insight into spatial variations in depletion rates, the potential for groundwater recharge, and the implications of water resource management practices. The analysis reveals an alarming trend of groundwater decline, with over 30% of wells in Malwa experiencing depletion at an average rate of approximately 40 cm per year. Seasonal patterns of groundwater levels, influenced by monsoon variability, show that while most areas report consistent declines, certain regions in southwestern Malwa experience waterlogging during monsoon periods, resulting in localized groundwater rise. This dichotomy between groundwater depletion and localized waterlogging reflects the complex hydrological challenges faced in this region, emphasizing the need for tailored water management strategies. To quantify the spatial and temporal trends in groundwater levels, statistical tools such as the Modified Mann-Kendall (MMK) test and Sen’s slope estimator were applied, allowing for a more nuanced understanding of depletion patterns. Hierarchical cluster analysis further enabled classification of wells based on depletion rates, providing a clear spatial framework for targeted groundwater management interventions. To forecast groundwater trends under varying future scenarios, machine learning (ML) models, including Random Forest (RF), Bagging-REPTree, and Bagging-DSTree, were applied. Among these, RF emerged as the most robust model, demonstrating high predictive accuracy across multiple statistical metrics, including Root Mean Square Error (RMSE), Nash-Sutcliffe Efficiency (NSE), and correlation coefficient (CC). By offering reliable predictive capability, the RF model proves valuable for informing water resource policies and planning in high-demand agricultural zones. This study highlights the utility of ML techniques for groundwater forecasting, advocating for their broader application in resource-scarce and over-exploited regions.