Leveraging machine learning methods to quantify 50 years of dwindling groundwater in India

Global compilations and regional studies, indicative of the unsustainable extraction and subsequent unremittingly de-pleting groundwater (GW) in India, either provide bulk estimates or are confined to the river basins and therefore con-ceal inferences from a nationwide policymaking perspective. Here, we provide the state-wise past (2000-2020) and future (2030-2050) assessment of dwindling groundwater in India utilizing in-situ groundwater levels (GWL) from 54,112 wells, remote sensing products, and hydrological simulations. By employing three machine learning methods, we show a decline in GWL of over 80% in North India with a notable shift towards the eastern state of Uttar Pradesh and a cumulative groundwater loss (169.96 +/- 19.67 km3) equivalent to the water storage capacity of the world's big -gest dam (Kariba Dam, Zimbabwe). Its likely contribution to sea-level rise (0.47 +/- 0.06 mm) is about 64% of that from annual global glacier melt. Our results typically contrast the GW recovery paradox in South India (e.g., a declining trend of -84.48 +/- 38.81 mm/a (p < 0.05) in Andhra Pradesh during 2000-2020), reveal high seasonal variability (e.g., up to-6 m in Maharashtra), and illustrate the skewed effect of survivor bias in the traditional assessments. We infer the significant impact of underlying hydrogeology and the implementation of water-related policies and pro-jects on the GWL dynamic and variability in the region. Projected GWL reveals a likely water scarcity situation for about 2.8 million km(2) area and one billion residents of the country up to 2050. Our observation-based analysis offers insights into the state-level monthly GW dynamics, which is critical for efficient interstate resource allocation, devel-opment plans, and policy interventions with broad methodological implications for the water-scarce countries.