Simulation And Analysis Of The Variation Characteristics Of Rainstorm Flood In The Jiniiang Watershed

Rainfall is the primary input to most hydrological systems,and its change characteristics would directly attribute to compositional diversities and complexities of runoff.Extreme storm runoffs often cause flood hazards.The temporal and spatial variation of rainfall is an important factor affecting flood generation,and it is meaningful to study the flood response to temporal and spatial variation of rainfall.The southeastern coastal area of China is the one of region where convective rainstorms and typhoon storms occur frequently.In order to investigate the applicability of the BP neural network model in the simulation of storm-floods in the region,Jinjiang watershed was selected as a study area.In order to further reveal the characteristics of the storm-floods in the area,we take the observed data of the annual maximum flood and its corresponding storm,as well as typhoon information,during 1956-2011 in Xixi watershed of Fujian Province and divide storm-floods into two types,those that occur during typhoon and non-typhoon seasons.Then,the characteristics of spatial and temporal variations of storm-flood are analyzed respectively with indices describing the spatial and temporal heterogeneity in the southeast coastal areas.A multiple linear regression analysis is used to estimate the relationship between these storm indices and flood peak discharges.In order to investigate the applicability of the BP neural network model in the simulation of storm-floods in the regionthe all storm-floods were divided into two types,i.e.the typhoon storm-floods and non-typhoon storm-floods.Then,the relationship between peak discharge and indices for the three storm-floods was estimated respectively with the method of grey correlation analysis.Finally,a variety of BP neural network models were constructed and the performance of the models was compared.In order to improve the flood simulation in spatial variation of river basin,three gauge stations were used for calibration and validation in simulation.The regulation of a large reservoir was also simulated.Ten rainstorm flood events occurring from 1972 to 1975 were used to calibrate the model using a trial-and-error approach,five rainstorm events occurring from 1975 to 1979 were used to validate the model.Accurate diagnosis of the hazard degree is the base for assessing the flood disaster risk.Flood risk based on precipitation characteristics and flood characteristics was calculated,respectively.Precipitation characteristics included the average precipitation volume and average precipitation intensity based on the flood events,while the flood characteristics included the simulated average peak discharge and average flood volume of these flood events in each sub-basin using the HEC-HMS model.Then,the flood risk results based on precipitation and flood characteristics were compared.The results show that 81%of non-typhoon storm-floods occur between April and June,and 70%of typhoon storm-floods occur between July and September.The number of typhoon storm-floods is twice more than that of non-typhoon storm-floods.Different types of rainstorm occur in different seasons,leading to varied responses of flood.There are obvious differences in the characteristics of spatial and temporal variations of typhoon and non-typhoon storm-floods.It also indicates the great significance of distinguishing typhoon and non-typhoon storm-floods in order to further reveal the spatial and temporal variations of storm-floods in the southeast coastal areas.The BP neural network models that described the typhoon storm-floods and non-typhoon storm-floods respectively perform better,which suggests it is necessary to divide the floods into two types of the typhoon and non-typhoon.The performance of the BP neural network model with four main indices determined by using the methods of grey correlation analysis indicates that the four indices models are also applicable for the prediction of peak discharge in the study area.The simulation results were evaluated with the errors of flood peak discharge,flood volume and time of peak discharge,and efficiency coefficient.The average efficiency coefficients for calibration in three gauge stations of Anxi,Honglai and Shilong were 0.892,0.895 and 0.921,respectively.It shown that the result calculated with flood characteristics was more reasonable.The application of the spatial distributions resulted from the hydrological model simulation for the analysis of flood risk was feasible.