Assessing groundwater storage variations in the Volta River Basin combining remote sensing tools and machine learning downscaling techniques

Water resources, vital for sustaining life and driving socio-economic development globally, face increasing pressure, necessitating accurate monitoring of storage variations. In this study, the water storage changes and its main drivers within the VRB are deeply investigated using remote sensing tools. The Gravity Recovery and Climate Experiment (GRACE) satellite derived terrestrial water storage anomalies (TWSA) is the only tool which vertically integrates all hydrological variables, and is suitable for groundwater storage anomalies (GWSA) changes investigation. The present investigation initially uses the Generalized Three-Corned Hat approach followed by a weighted average to merge four GRACE derived TWSA. Three machine learning techniques including XGBoost, LightGBM and Random Forest are applied to downscale TWSA at a spatial resolution of 0.1°. Results showed that (i) the merged TWSA depicts the lowest uncertainty with a median of 0.94 cm. (ii) The LightGBM model yielded the highest R² (0.99) and the lowest rmse (0.69 cm) in test phase. (iii) The LightGBM downscaled product indicated that GWSA increased (0.32 cm/month) over 2002–2022. (iv) The influence of precipitation and evapotranspiration on GWSA appeared to be rather harmless, while the spatial distribution of GWSA and subsurface runoff showed significant positive trend over the pixels connected with dams, reservoirs, and irrigated areas. This suggests that anthropogenic variable is the main driver of GWSA changes within the VRB. (v) Statistically significant positive trends are observed in downscaled GWSA time series and in-situ GWSA measurements.