Satellite and Grace‐Based Analysis of Agricultural Groundwater Stress Using Machine Learning

Résumé fourni par la source

ABSTRACT Groundwater sustainability has become a pressing concern in intensively cultivated regions, particularly where agricultural expansion and climatic variability exert increasing pressure on subsurface water reserves. This study presents a machine learning‐enhanced assessment of groundwater‐agriculture interactions using multi‐temporal satellite‐derived indices and GRACE groundwater storage data from 2020 to 2024. A comprehensive analysis was conducted using supervised classification algorithms alongside MODIS‐based vegetation indices (NDVI, EVI), water indices (MNDWI, NDWI), soil moisture indicators (SMI, NDMI), evapotranspiration, precipitation variability, drought severity (SPI), groundwater anomalies (GRACE), and vegetation condition index (VCI). Results reveal substantial groundwater depletion of 5.1 cm water equivalent thickness, with GRACE minimum values declining from −7.32 to −12.58 over the study period. Surface water availability also declined sharply, with maximum MNDWI and NDWI values decreasing by 13% and 42%, respectively. Land use analysis showed a 3531.6 km 2 (3.9%) reduction in cropland and an alarming 611.2 km 2 (27.1%) increase in bare ground, indicating widespread land degradation. Correlation analysis identified a strengthening negative relationship between groundwater decline and vegetation health ( r = −0.06, p < 0.05). The machine learning model successfully identified groundwater‐stressed agricultural zones with 87.3% classification accuracy. These findings provide robust, data‐driven insights for policymakers and resource managers working toward sustainable groundwater use in rapidly changing agricultural landscapes.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.