Simulating flash flood hydrographs and behavior metrics across China: Implications for flash flood management.

China frequently suffers from considerable and disastrous flash floods with wide areal coverage and high frequency. Obtaining useful information to support flash flood management and decision-making is challenging for massive flash flood events that vary greatly in spatio-temporal characteristics. In this study, hydrological modelling approach (CNFF) and cluster analysis were integrated to assess simulation reliability of entire flash flood processes including both hydrographs and behavior characteristics in a manner of similarity classification, rather than at event scale. A total of 207 hourly events from 13 mountainous catchments with diverse physiographic and meteorological characteristics across China were selected for study. Representative flash flood types were classified using normalized hydrographs with diverse spatio-temporal patterns by k-means clustering algorithm. For individual flash flood types, simulation reliability of CNFF was assessed in capturing corresponding hydrographs, seven behavior metrics measuring flash flood magnitude, intensity, occurrence time, flood timescale, rates of change and variability, and their uncertainties. Results showed that three (fast, intermediate and slow) flash flood types were identified from all the flash flood events with overall average silhouette index of 0.45. Hourly hydrographs of three flash flood types were well reproduced by CNFF, with absolute average relative error of runoff within 15% and Nash-Sutcliffe Efficiency above 0.55. All the behavior metrics were the most accurately reproduced for slow flash flood type with the least average relative root-mean-square error (0.30), followed by intermediate (0.52) and fast (0.58) types. Moreover, the slow flash flood type had the most reliable but greatest uncertainty interval of both hydrograph and behavior metrics, with average relative interval length being 1.24 and 71.96%, and 93.10% and 100% of observations contained in 95% confidence interval, respectively. This study provided efficient and detailed information for flash flood management, and extended application scope of hydrological models to encompass flash flood types and behavior metrics.