Filling the data gap between GRACE and GRACE-FO based on a two-step reconstruction method

Terrestrial water storage represents both surface and subsurface water resources and plays a crucial role in the global hydrological cycle. The Gravity Recovery and Climate Experiment (GRACE) satellite mission provides large-scale and high-stability terrestrial water storage anomaly (TWSA) data for water resource analysis. However, an 11-month data gap exists between the two generations of gravity satellites, GRACE and GRACE Follow-On (July 2017 to May 2018), posing challenges for TWSA analysis and applications. This study proposed a long short-term memory (LSTM) and Bayesian convolutional neural network (BCNN) combined in an LSTM-BCNN reconstruction model. By extracting long-term temporal change features and recent environmental spatial features, the model reconstructs and corrects the trend and random terms of the TWSA decomposition to reconstruct the TWSA data during the gap period. The model was evaluated on a global scale across 40 basins and at a grid scale. At the grid scale, the LSTM-BCNN model achieves a correlation coefficient (CC) of 0.89 ± 0.06, Nash-Sutcliffe efficiency coefficient (NSE) of 0.78 ± 0.12, and normalized root mean square error (NRMSE) of 0.11 ± 0.02. Across the 40 basins, the LSTM-BCNN model effectively reconstructed the TWSA data during the gap period.