Assessing Groundwater Storage Changes in Data-Scarce Basins of Afghanistan: A Machine-Learning Based Downscaling of GRACE(-FO) Data

Climate change, rising water demand, and ecosystem stress are intensifying the reliance on groundwater while limiting the capacity of many basins to effectively monitor and manage subsurface water resources. In data-scarce and conflict-affected regions where monitoring networks are sparse, decision-makers increasingly require reliable, high-resolution information to support drought preparedness, climate adaptation, and sustainable groundwater governance. The present study proposes an evidence-based machine-learning framework for the purpose of enhancing the monitoring of groundwater storage anomaly (GWSA) through the process of downscaling GRACE and GRACE-FO observations from ~3° to 0.1°. The reconstruction of monthly GRACE/GRACE-FO gaps was performed using a Seasonal-Trend Decomposition based on Loess (STL), and a Random Forest model was trained with hydroclimatic and land-surface predictors, including soil moisture, snow water equivalent, evapotranspiration, precipitation, land-surface temperature, and the normalized difference vegetation index (NDVI). The performance of the model was evaluated by comparing the model's results with the existing in-situ groundwater-level observations in the Kabul River Basin. The results indicate that satellite-inferred groundwater losses in Afghanistan are persistent, with a rate of −0.71 cm yr^-1, ranging from basin-scale depletion of −0.77 cm yr^-1 in the Helmand River Basin to −0.40 cm yr^-1 in the Northern River Basin. Recent conditions indicate intensified depletion during 2018–2022, with year-sum GWSA declines reaching ~145 cm in the Harirod–Murghab River Basin, while the Northern River Basin shows comparatively lower losses (~80 cm). The 0.1° downscaled product improves agreement with observations (root mean square error (RMSE) reductions up to 77.8%) and reveals spatially heterogeneous hotspots that are not detectable at coarse GRACE resolution. Generally, the proposed framework translates coarse satellite gravimetry into actionable, basin-relevant information for climate-resilient groundwater management, while underscoring the necessity for uncertainty-aware, multi-source monitoring under increasing hydroclimatic extremes. The approach enables the early detection of emerging depletion hotspots, thereby supporting proactive planning for future water security. This includes targeted demand management, drought response, and adaptation investments in groundwater-dependent regions.