Spatial downscaling of GRACE-derived groundwater storage changes across diverse climates and human interventions with Random Forests

Abstract

The Gravity Recovery and Climate Experiment (GRACE) satellites launched in 2002, which were followed by the GRACE Follow-On mission in 2018, represented a significant advancement in global groundwater storage monitoring. However, the inherent limitation of a relatively low spatial resolution confines its applicability primarily to large river basins and aquifers. This study introduces a novel statistical downscaling approach using Random Forests (RF) to enhance the spatial resolution from 3° × 3° to 0.25° × 0.25° for GRACE-derived groundwater storage (GWS) changes. Focusing on sub-Saharan Africa, the Middle East, and South Asia across diverse climates and human interventions, the analysis spans from February 2003 to December 2022. Predictors encompass hydrological fluxes and their accumulation, including IMERG precipitation, GLEAM evapotranspiration, and GLDAS runoff, alongside GLDAS soil moisture and MODIS NDVI. RF models show robust performance across 114 aquifers, with a Pearson’s correlation coefficient (CC) exceeding 0.92 for training sets and 0.85 for validation sets. Validation against in-situ groundwater level observations from wells in India reveals satisfactory performance, with positive Spearman’s CC values between in-situ groundwater levels and downscaled terrestrial water storage anomaly (TWSA) for a majority of dug and bore wells. This study concludes with the generation of monthly downscaled GWS changes (0.25° × 0.25°) spanning from February 2003 to December 2022. These datasets serve as a foundation for facilitating finer-scale hydrological studies and formulating local groundwater management strategies.