Abstract: Monitoring flood events is essential for early warning, control project scheduling, and risk management. Based on the Gravity Recovery and Climate Experiment (GRACE) satellite data of Apr. 2002 to June 2017, this paper studies variations in terrestrial water storage anomaly (TWSA) in the Pearl River basin, and combines the Global Land Data Assimilation Systems (GLDAS)-simulated TWSA, temperature, and rainfall data to construct a generalized regression neural network (GRNN) model to extend the TWSA time series to the time span of Jan. 2000 to Dec. 2018. A flood potential index (FPI) is calculated to monitor large-scale extreme flood events. Results show that 1) five official GRACE products are highly correlated in this basin, and the TWSA derived from GRACE has a 1-2 month lead time relative to that from water budget. 2) The GRNN-extended TWSA for the basin is consistent with that of GLDAS, indicating strong capability of prediction by the GRNN model. 3) FPI is, when calculated using the GRNN-predicted TWSA, a good indicator in monitoring the basin’s large-scale flood events of Apr. to Dec. 2008.

Key words: GRACE, GRNN, flood event, monitoring, Pearl River basin