Floods are a significant and pervasive threat globally, exacerbated by climate change and increasing extreme weather events. The Gravity Recovery and Climate Experiment (GRACE) and its follow-on mission (GRACE-FO) provide crucial insights into terrestrial water storage anomalies (TWSA), which are vital for understanding flood dynamics. However, the observational gap between these missions presents challenges for flood monitoring, affecting the estimation of long-term trends and limiting the analysis of interannual variability, thereby impacting overall analysis accuracy. Reconstructing the missing data between GRACE and GRACE-FO is essential for systematically understanding the spatiotemporal distribution characteristics and driving mechanisms of interannual changes in regional water reserves. In this study, the Generative Adversarial Imputation Network (GAIN) is applied to improve the monitoring capability for flood events in the Pearl River Basin (PRB). First, the GRACE/GRACE-FO TWSA data gap is imputed with GAIN and compared with long short-term memory (LSTM) and k-Nearest Neighbors (KNN) methods. Using the reconstructed data, we develop the Flood Potential Index (FPI) by integrating GRACE-based TWSA with precipitation data and analyze key characteristics of FPI variability against actual flood events. The results indicate that GAIN effectively predicts the GRACE/GRACE-FO TWSA gap, with an average improvement of approximately 50.94% over LSTM and 68.27% over KNN. The reconstructed FPI proves effective in monitoring flood events in the PRB, validating the reliability of the reconstructed TWSA. Additionally, the FPI achieves a predictive accuracy of 79.7% for real flood events, indicating that short-term flood characteristics are better captured using TWSA. This study demonstrates the effectiveness of GAIN in enhancing data continuity, providing a reliable framework for large-scale flood risk assessment and offering valuable insights for flood management in vulnerable regions.
Read full abstract