Applicability of high-resolution numerical rainfall forecasts in flood forecasting over small and medium sized river basin

The leading-time of flood forecasts can be effectively extended by combining numerical meteorological forecasts with hydrological models, which is of great significance in warning of flash floods over small and medium sized river basins. In order to investigate the potential of using high-resolution numerical meteorological forecasting models in the simulation of flash flood processes, the Taoxi River Basin in Fujian Province was taken as the research object to evaluate the ability of a high spatial and temporal resolution numerical weather prediction model (GRAPES-RAFS) (Rapid Analysis and Forecast System) in short-term rainfall forecasting at different reporting times. Two bias correction methods (LS and QM methods) are further used to correct the systematic bias of GRAPES-RAFS simulated rainfall. The corrected rainfall forecasts are also used to drive the Xin’ anjiang hydrological model with pre corrected and post corrected data to evaluate their applicability in flood forecasts. The evaluation results show that GRAPES-RAFS performs well in simulating rainfall process, but significantly overestimate rainfall amounts. All the evaluation indicators have a good consistency. Both bias correction methods can significantly improve the accuracy of rainfall forecasts, reducing the average relative bias from 60.30% to 18.00% (LS method) and 21.33% QM method), respectively across 12 rainfall events. The performance of uncorrected GRAPES-RAFS in predicting rainfall directly for flood forecasting is poor, and flood peaks are significantly overestimated. However, rainfall bias correction can significantly improve the accuracy of flood. Overall, the rainfall events that perform better after bias correction correspond to the flood events that perform better, and the two bias correction methods perform similarly forecasting.