Spatial downscaling of GRACE terrestrial water storage anomalies for drought and flood potential assessment

• A robust framework based on the LSTM model to downscale GRACE TWSA. • Downscaled TWSA can better capture drought condition at sub-basin scale. • Downscaled TWSA improved the capability to indicate flood potential at county level. Terrestrial water storage anomaly (TWSA) from the Gravity Recovery and Climate Experiment (GRACE) mission provides invaluable information for quantifying changes in freshwater availability. However, the coarse spatial resolution of GRACE TWSA limits its application to sub-regional studies. The study proposed a systematic framework to spatially downscale GRACE TWSA retrievals using a long short-term memory (LSTM) model over the Texas-Gulf Basin. A synthetic experiment was conducted to demonstrate the robustness of the downscaling framework. The real-world experiment revealed that the downscaled TWSA from LSTM can represent the variation of TWSA at the basin (R LSTM = 0.91) and sub-basin scales. The LSTM-based TWSA can better represent the early recovery from extreme droughts for the sub-basins along the coast. Moreover, the LSTM-based TWSA outperformed model-based TWSA in characterizing groundwater variation, especially for sub-basins with deep groundwater levels in the west. The flood analysis showed that the downscaled TWSA from LSTM yielded improved skill in predicting county-level floods, providing a larger true positive rate relative to GRACE TWSA retrievals (TPR LSTM = 0.36 and TPR GRACE = 0.31). Additionally, the trained LSTM models were used to predict fine-resolution TWSA without requiring GRACE observations. Results demonstrated that the accuracy of LSTM-based TWSA forecasts was slightly inferior to the downscaling case, but they still provided useful information for drought and flood predictions at sub-basin to local scales.