Downscaled GRACE/GRACE-FO observations for spatial and temporal monitoring of groundwater storage variations at the local scale using machine learning

Abstract

Groundwater utilization for several purposes such as irrigation in agriculture, industry, and domestic use substantially impacts water storage. Groundwater Storage Anomaly (GWSA) estimates have improved owing to the Gravity Recovery and Climate Experiment (GRACE) and GRACE-Follow On (GRACE-FO) advancements. However, the characterization of GWSA fluctuation hotspots has been hindered by the coarse resolution of GRACE data. To better measure groundwater storage and depletion variations throughout an area and identify GWSA variation hotspots, a fine spatial resolution of GWSA estimations is required. Therefore, due to the coarse resolution of GRACE measurements, the eXtreme Gradient Boosting (XGBoost) model was developed to simulate fine resolution 0.1° GWSA combining climatic variables (soil moisture storage, evapotranspiration, temperature, surface runoff, and rainfall) from improved spatial high resolution FLDAS (Famine Early Warning Systems Network Land Data Assimilation System) model derived data and geospatial variables (elevation, slope, and aspect) extracted from Digital Elevation Model (DEM). A correlation of 0.98 demonstrated that the XGBoost model successfully simulated groundwater storage at a finer scale over the Upper Indus Plain Aquifer (UIPA). The findings suggested that the UIPA's groundwater storage has been depleted at an annual rate of 0.44 km3/yr which was 7.94 km3 in total between 2003 and 2020. According to the results, there seems to be consistency between the downscaled and original GWSA regarding temporal and spatial variability. The results were verified to show an improved correlation of 0.77 between the downscaled and the in-situ GWSA, compared to 0.75 between the GRACE-derived and the in-situ GWSA.