Flash floods present significant heterogeneity over both space and time due to diverse topographic,geomorphologic,and hydro-meteorological conditions of catchments.Accurate identification and simulation of typical fla...Flash floods present significant heterogeneity over both space and time due to diverse topographic,geomorphologic,and hydro-meteorological conditions of catchments.Accurate identification and simulation of typical flash flood types are of great significance for the mitigation of flash flood disasters at the national scale.Three flood peak indices and dynamic indices were adopted to characterize the behavioral variability of flash floods.The typical flash flood types and corresponding behavior indices were identified and simulated using statistical analysis(i.e.,principal component analysis,dynamic K-means clustering,and analysis of similarity)and hydrological modelling(i.e.,HEC and XAJ models).There were 177 flash flood events at the hourly scale being selected for case study from eight catchments with various climatic and geographic characteristics.Results showed that all the flash flood events were clustered into three types(named Types 1,2,and 3).Type 1 was characterized by low peak flow intensity,early flood peak occurrence time,and thin flood process with short duration.Type 2 was characterized by low peak flow intensity,late flood peak occurrence time,and flat flood process with long duration.Type 3 was characterized by high peak flow intensity and late flood peak occurrence time.Flash flood types showed high consistency with their influencing factors(e.g.,catchment forest ratio and drainage area,occurrence time and magnitude of maximum storm intensity,and concentration of a storm event).The simulation performances were basically the same for HEC and XAJ models.As for flash flood event simulations,the average relative error varied from 23.25%to 27.98%,from 11.95%to 18.19%,and from 8.30%to18.25%for Types 1,2 and 3,respectively.The average Nash-Sutcliffe efficiency coefficient varied from 0.39 to 0.54,from 0.76 to 0.85,and from 0.86 to 0.91,respectively.As for the six flash flood behavior indices simulations,the average relative rootmean-square error(RMSEr)varied from 0.37 to 0.69,from 0.37 to 0.41,and from 0.18 to 0.25 for Types 1,2,and 3,respectively.The average correlation coefficient(r)varied from 0.52 to 0.68,from 0.78 to 0.85,and from 0.88 to 0.94,respectively.The flood peak indices were the best simulated for Types 2 and 3 with RMSEr varying from 0.18 to 0.28 and r varying from 0.86 to 0.91.The flood dynamic indices were the best simulated for Type 3 with RMSEr varying from 0.19 to 0.21 and r varying from 0.91 to0.97.The study provided detailed flood information supports for flood management at catchment scale,and also provided new insights into flash flood simulations in small and medium-sized catchments from perspective of flood behavioral processes.展开更多
基金supported by the National Natural Science Foundation of China(Grant No.41807171)the National Key Research and Development Program of China(Grant No.2016YFC0400902)the China National Flash Flood Disaster Prevention and Control Project(Grant No.JZ0145B2017)。
文摘Flash floods present significant heterogeneity over both space and time due to diverse topographic,geomorphologic,and hydro-meteorological conditions of catchments.Accurate identification and simulation of typical flash flood types are of great significance for the mitigation of flash flood disasters at the national scale.Three flood peak indices and dynamic indices were adopted to characterize the behavioral variability of flash floods.The typical flash flood types and corresponding behavior indices were identified and simulated using statistical analysis(i.e.,principal component analysis,dynamic K-means clustering,and analysis of similarity)and hydrological modelling(i.e.,HEC and XAJ models).There were 177 flash flood events at the hourly scale being selected for case study from eight catchments with various climatic and geographic characteristics.Results showed that all the flash flood events were clustered into three types(named Types 1,2,and 3).Type 1 was characterized by low peak flow intensity,early flood peak occurrence time,and thin flood process with short duration.Type 2 was characterized by low peak flow intensity,late flood peak occurrence time,and flat flood process with long duration.Type 3 was characterized by high peak flow intensity and late flood peak occurrence time.Flash flood types showed high consistency with their influencing factors(e.g.,catchment forest ratio and drainage area,occurrence time and magnitude of maximum storm intensity,and concentration of a storm event).The simulation performances were basically the same for HEC and XAJ models.As for flash flood event simulations,the average relative error varied from 23.25%to 27.98%,from 11.95%to 18.19%,and from 8.30%to18.25%for Types 1,2 and 3,respectively.The average Nash-Sutcliffe efficiency coefficient varied from 0.39 to 0.54,from 0.76 to 0.85,and from 0.86 to 0.91,respectively.As for the six flash flood behavior indices simulations,the average relative rootmean-square error(RMSEr)varied from 0.37 to 0.69,from 0.37 to 0.41,and from 0.18 to 0.25 for Types 1,2,and 3,respectively.The average correlation coefficient(r)varied from 0.52 to 0.68,from 0.78 to 0.85,and from 0.88 to 0.94,respectively.The flood peak indices were the best simulated for Types 2 and 3 with RMSEr varying from 0.18 to 0.28 and r varying from 0.86 to 0.91.The flood dynamic indices were the best simulated for Type 3 with RMSEr varying from 0.19 to 0.21 and r varying from 0.91 to0.97.The study provided detailed flood information supports for flood management at catchment scale,and also provided new insights into flash flood simulations in small and medium-sized catchments from perspective of flood behavioral processes.
