DETERMINATION OF DYNAMIC CRITICAL RAINFALL BASED ON GEOMORPHOLOGICAL INSTANTANEOUS UNIT HYDROGRAPH AND RADIAL BASIS FUNCTION NEURAL NETWORK

To disclose the formation mechanism of flash flood disaster, it is necessary to develop a dynamic critical rainfall model that considers all influencing factors. Targeting the Peihe River watershed in China's Henan Province, this paper designs a runoff convergence calculation plan based on the geomorphological instantaneous unit hydrograph (GIUH), and uses the plan to simulate 8 floods in the target watershed. The simulated results were close to the measured data. Next, the GIUH was adopted to predict the critical rainfalls of 16 floods in the target watershed. The radial basis function neural network (RBFNN) was selected to create a dynamic critical rainfall prediction model, with the preceding rainfall, cumulative rainfall and rainfall intensity as the inputs and the critical rainfall of each event as the output. The model was employed to predict the critical rainfalls of 6 historical floods. The results show that the prewarned critical rainfall reached the pass rates of 100% and 83.3%, respectively, for the 1 h and 3 h periods. Hence, that the GIUH can ensure the calculation accuracy despite the lack of data in regions prone to flash flood; the RBFNN-based dynamic critical rainfall prediction model can effectively improve the accuracy of critical rainfall calculation and the flash flood prewarning.