High effectiveness of GRACE data in daily-scale flood modeling: case study in the Xijiang River Basin, China

The modeling and forecasting of short-duration and high-intensity floods are of importance for flood defenses and adaptations. One of the conventional ways to model or forecast such events is to utilize hydrological models driven by meteorological and hydrological station data. However, this suffers from complicated parameter specification and large uncertainties, particularly in regions with very few gauged stations. Based on the daily downscaled Gravity Recovery and Climate Experiment (GRACE) solutions, this study employed three different machine learning models and two hydrological models for flood modeling at the daily timescale by taking the Xijiang River Basin in China as a case study. The results show that: (1) the uncertainty of daily GRACE solutions alone governs the difference between GRACE data and hydrological simulations; (2) there is a strong correlation between the high-frequency components of runoff anomalies and terrestrial water storage anomaly (TWSA), and runoff plays a dominant role in TWSA variation during floods; (3) the developed machine learning models can model runoff during floods effectively and outperform the hydrological models. The proposed comprehensive method based on remote sensing satellites provides a potential new way for flood modeling, particularly for poorly gauged regions.