The Gravity Recovery and Climate Experiment (GRACE) mission and its Follow-On mission (GRACE-FO) exhibit substantial uncertainties and coarse resolution, limiting their capability for precise Terrestrial Water Storage Anomaly (TWSA) monitoring. In order to overcome these challenges, the Bayesian Three-Cornered Hat (BTCH) method was initially used to fuse multi-source GRACE solutions and evaluate its uncertainties. Subsequently, the spatial resolution of the fused TWSA product was improved from 0.5–0.05° using two kinds of downscaling models, namely Geographically Weighted Regression (GWR) and Random Forest (RF). A High-resolution Drought Severity Index (HDSI) was then constructed based on the downscaled TWSA, and its applicability was verified by comparing with traditional drought indices (scPDSI and SPEI). The results showed that the uncertainty of the fused TWSA product was significantly reduced to 13.42 mm, representing a 37.4% reduction compared to single-institution products. The GWR model achieved superior performance (MAE: 0.63 mm; RMSE: 0.49 mm) compared to the RF-based results. The HDSI revealed that the longest drought event occurred between 2015 and 2017, affecting over 90% of the basin area at its peak. Compared to traditional drought indices such as the scPDSI and the SPEI, the HDSI accurately recognizes the occurrence of drought events and identifies more drought areas with finer spatial delineation. This research provides high-resolution data for water resource management and an early warning system for drought in the YRB, while also highlighting regional differences in the applicability of the drought index.
Zhang et al. (Sun,) studied this question.