Reconstruction of Water Storage Anomalies and Drought Identification Using an Integrated Attention Mechanism

Objectives: The identification of lake hydrological droughts is crucial for the scientific evaluation of drought scale, severity, and comprehensive management. Long-term monitoring of terrestrial water storage anomalies (TWSA) enables the quantification of drought characteristics in lakes and their catchments. Compared to traditional measurement methods, the Gravity Recovery and Climate Experiment (GRACE) satellite can precisely monitor changes in TWSA, providing a spatial observation tool for identifying lake hydrological droughts. Methods: In this study, Poyang Lake was taken as a case study. Hydrological data with seasonal components and linear trends removed, along with TWSA estimated from GRACE, were used as input values. A long short-term memory (LSTM) network model integrated with a convolutional neural network (CNN) and attention mechanism was employed to reconstruct the TWSA time series from 1982 to 2002. Finally, combining GRACE observations and the reconstructed TWSA time series, hydrological drought events in the Poyang Lake basin from 1982 to 2023 were identified, and the potential recovery time of each drought event was estimated. Results: The results indicate that the reconstructed data successfully identified hydrological drought events in the Poyang Lake basin and quantitatively analyzed their characteristics. Conclusions: The duration and recovery time of hydrological droughts were found to have an exponential relationship with their overall severity. This method provides a new approach for drought monitoring research in medium and small-scale basins.