Groundwater has a significant contribution to water storage and is considered to be one of the sources for agricultural irrigation; industrial; and domestic water use. The Gravity Recovery and Climate Experiment (GRACE) satellite provides a unique opportunity to evaluate terrestrial water storage (TWS) and groundwater storage (GWS) at a large spatial scale. However; the coarse resolution of GRACE limits its ability to investigate the water storage change at a small scale. It is; therefore; needed to improve the resolution of GRACE data at a spatial scale applicable for regional-level studies. In this study; a machine-learning-based downscaling random forest model (RFM) and artificial neural network (ANN) model were developed to downscale GRACE data (TWS and GWS) from 1° to a higher resolution (0.25°). The spatial maps of downscaled TWS and GWS were generated over the Indus basin irrigation system (IBIS). Variations in TWS of GRACE in combination with geospatial variables; including digital elevation model (DEM), slope; aspect; and hydrological variables; including soil moisture; evapotranspiration; rainfall; surface runoff; canopy water; and temperature; were used. The geospatial and hydrological variables could potentially contribute to; or correlate with; GRACE TWS. The RFM outperformed the ANN model and results show Pearson correlation coefficient (R) (0.97), root mean square error (RMSE) (11.83 mm), mean absolute error (MAE) (7.71 mm), and Nash–Sutcliffe efficiency (NSE) (0.94) while comparing with the training dataset from 2003 to 2016. These results indicate the suitability of RFM to downscale GRACE data at a regional scale. The downscaled GWS data were analyzed; and we observed that the region has lost GWS of about −9.54 ± 1.27 km³ at the rate of −0.68 ± 0.09 km³/year from 2003 to 2016. The validation results showed that R between downscaled GWS and observational wells GWS are 0.67 and 0.77 at seasonal and annual scales with a confidence level of 95%, respectively. It can; therefore; be concluded that the RFM has the potential to downscale GRACE data at a spatial scale suitable to predict GWS at regional scales.

Climate change may cause severe hydrological droughts, leading to water shortages which will require to be assessed using high-resolution data. Gravity Recovery and Climate Experiment (GRACE) satellite Terrestrial Water Storage (TWSA) estimates offer a promising solution to monitor hydrological drought, but its coarse resolution (1°) limits its applications to small regions of the Indus Basin Irrigation System (IBIS). Here we employed machine learning models such as Extreme Gradient Boosting (XGBoost) and Artificial Neural Network (ANN) to downscale GRACE TWSA from 1° to 0.25°. The findings revealed that the XGBoost model outperformed the ANN model with Nash Sutcliff Efficiency (NSE) (0.99), Pearson correlation (R) (0.99), Root Mean Square Error (RMSE) (5.22 mm), and Mean Absolute Error (MAE) (2.75 mm) between the predicted and GRACE-derived TWSA. Further, Water Storage Deficit Index (WSDI) and WSD (Water Storage Deficit) were used to determine the severity and episodes of droughts, respectively. The results of WSDI exhibited a strong agreement when compared with the Standardized Precipitation Evapotranspiration Index (SPEI) at different time scales (1-, 3-, and 6-months) and self-calibrated Palmer Drought Severity Index (sc-PDSI). Moreover, the IBIS had experienced increasing drought episodes, e.g., eight drought episodes were detected within the years 2010 and 2016 with WSDI of −1.20 and −1.28 and total WSD of −496.99 mm and −734.01 mm, respectively. The Partial Least Square Regression (PLSR) model between WSDI and climatic variables indicated that potential evaporation had the largest influence on drought after precipitation. The findings of this study will be helpful for drought-related decision-making in IBIS.

High-resolution soil moisture (SM) information is essential for regional to global hydrological and agricultural applications. The Soil Moisture Active Passive (SMAP) offers daily global composites of SM at coarse-resolution 9 and 36 km, with data gaps limiting its local application to depict SM distribution in detail. To overcome the aforementioned problem, a downscaling and gap-filling novel approach was adopted, using random forest (RF) and artificial neural network (ANN) algorithms to downscale SMAP SM data, using land-surface variables from moderate-resolution imaging spectroradiometer (MODIS) onboard Aqua and Terra satellites from the years 2018 to 2019. Firstly, four combinations (RF+Aqua, RF+Terra, ANN+Aqua, and ANN+Terra) were developed. Each combination downscaled SMAP SM at a high resolution (1 km). These combinations were evaluated by using error matrices and in situ SM at different scales in the ShanDian River (SDR) Basin. The combination RF+Terra showed a better performance, with a low averaged unbiased root mean square error (ubRMSE) of 0.034 ${\mathrm{m}}^3$/${\mathrm{m}}^3$ and high averaged correlation (R) of 0.54 against the small-, medium-, and large-scale in situ SM. Secondly, the impact of various land covers was examined by using downscaled SMAP and in situ SM. Vegetation attenuation makes woodland more error-prone and less correlated than grassland and farmland. Finally, the RF+Terra and ANN+Terra combinations were selected for their higher accuracy in gap filling of downscaled SMAP SM. The gap-filled downscaled SMAP SM results were compared spatially with China Land Data Assimilation System (CLDAS) SM and in situ SM. The RF+Terra combination outcomes were more humid than ANN+Terra combination results in the SDR basin. Overall, the RF+Terra combination gap-filled data showed high R (0.40) and less ubRMSE (0.064 ${\mathrm{m}}^3$/${\mathrm{m}}^3$) against in situ SM, which was close to CLDAS SM. This study showed that the proposed RF- and ANN-based downscaling methods have a potential to improve the spatial resolution and gap-filling of SMAP SM at a high resolution (1 km).