| Many cities around the world have experienced rapid urbanization recently,which caused significantly adverse impacts on urban hydrological processes,leading to the frequent occurrence of urban water problems.Urban flood is one of the most typical and common water security problems.Urban floods can cause major disasters in both water quantity and quality and trigger other urban water security problems such as waterlogging,first flush pollution,and contanmination of water bodies,which seriously threatens human life and property.Accurately simulating and forecasting the water quantity and quality processes of flood is the basis for urban flood warning and management,which can mitigate the potential losses.However,due to the complicated underlying surfaces of urban areas,frequent anthropogenic activities,and limited observation data,there are still three key scientific problems unresolved in urban flood simulation and forecasting,which are(i)the insufficient identification of rainfall-runoff relationship under complicated land covers of urban areas,(ii)low accuracy of urban floods prediction in ungauged urban areas,and(iii)large uncertainties in identifying influencing factors of first flush effect.Therefore,this thesis attempts to tackle the above three issues by considering the rainfall-runoff nonlinear relationship in urban areas with complex land covers as the key point,with the logical orders from water quantity to water quality as well as from gauged urban areas to ungauged urban areas.The study area comprises 37 urban catchments with separate drainage and sewage pipeline system and available observations in Fenghuangcheng region,Guangming District,Shenzhen City.The main results and findings are summarized as below:(1)The new urban time variant gain model(TVGM＿Urban)that can be used for rainfall-runoff processes simulation considering the complicated underlying surfaces in urban areas is developed.Based on the shortcomings of insufficient consideration on the rainfall runoff response relationships of different land covers in urban areas in the runoff generation module of most current urban hydrological models,this thesis develops a new urban hydrological model,the TVGM＿Urban model,on the basis of the time variant gain model(TVGM).The TVGM＿Urban model can reflect rainfall-runoff responses of different categories of urban surfaces.This thesis evaluates the performance of the TVGM＿Urban model in urban floods simulation for the study area by comparing with eight widely used urban hydrological models such as Horton infiltration model,SWMM(Storm Water Management Model)etc.The results indicate that the TVGM＿Urban model has the best performance in urban floods simulation,as it has better simulation results in NSE,runoff depth,and time-to-peak than other models.Only the SWMM model performs slightly better than the TVGM＿Urban model in peak flow simulation,which may be attributed to the differences in runoff routing module.However,the TVGM＿Urban model is proved to present better peak flow simulation results than the SWMM model if it adopts the same runoff routing module as the SWMM model.This means that the TVGM＿Urban model has the best performance in urban floods simulation compared to the models with the same runoff routing module.The uncertainty analysis results of different models indicate that the TVGM＿Urban model has higher parameter uncertainty,lower structure uncertainty,and lower prediction uncertainty of floods than the other models.According to the characteristics of the urban catchments with the best simulation results in both calibration and validation period of each model,the TVGM＿Urban is more appropriate for urban floods simulation in urban catchments with an area of 0～0.2km~2,slope of 6～8%,width of 80～100 m,and center-to-outlet length of 200～270 m,while the SWMM model is better for the areas with larger area,steeper slope,and longer conduit length.(2)The optimal regionalization method of TVGM＿Urban model is developed and verified,and the main influential variables of each parameter are identified.This thesis compares the performance of four distance-based methods and four regression methods in urban floods simulation for study area from the aspect of global regionalization,which means takes all urban catchments as one whole for the regionalization of parameters in the TVGM＿Urban model.The results show that the spatial proximity method with average output option has the best performance while the random forest method also performs well.In the regional regionalization that divides all urban catchments into three homogeneous groups for separate regionalization,the optimal regionalization methods in three homogeneous groups are both regression methods.Based on the results of regional regionalization,the combination of the optimal regression methods in each homogeneous group in the regional regionalization is developed and identified as the best regionalization method of the TVGM＿Urban model through comparing with all methods in global and regional regionalization.To evaluate the applicability of the optimal regionalization method,this thesis applies it to urban floods prediction in two urban catchments far from the main study area.The results show that the combination of the optimal regression methods can well predict urban floods in ungauged urban catchments and have high applicability.According to the optimal regression methods in global and regional regionalization,this thesis analyzes the greatest influential variables for each parameter in the TVGM＿Urban model.The results show that the most influential variables of each parameter are consistent with the physical meaning of parameters.The CN value,area ratio of each land cover,geometry variables such as slope and width,and soil moisture related variables are the main influential variables of runoff generation parameters.The variables related to elevation and length of urban catchment are the main variables affecting runoff routing parameters.(3)The most important variables of first flush effect are identified,and a method for first flush pollution prediction in ungauged urban areas is proposed.This thesis compares the performance of the TVGM＿Urban for water quantity and quality model(TVGM＿UQQ)that couples the TVGM＿Urban model with built-up and wash-off modules,SWMM,random forest and regression tree methods in first flush effect simulation.The results show that the random forest method presents the best simulation performance in first flush pollution.Compared to SWMM,the TVGM＿UQQ model has better performance in the simulation of pollutant concentration process,but two water quantity and quality models,especially the SWMM model,both show unsatisfying results in the simulation of first flush pollution.The variable importance analysis of the random forest method indicates that the antecedent rainfall variables,catchment attributes,and rainfall-runoff related variables are the most important variables regulating the first flush pollution.In addition,the importances of antecedent rainfall variables and rainfall-runoff related variables show an increasing tendency with time,while the importances of catchment attributes present a descend trend.Based on the results of main influential variables of first flush pollution and the optimal regionalization method of the TVGM＿Urban model,this thesis develops the random forest method for first flush pollution prediction in the ungauged urban catchments and evaluates its accuracy in first flush pollution prediction in ungauged urban catchments.This thesis aims to propose a new urban hydrological model to improve the accuracy of urban floods simulation and a reasonable regionalization method for urban floods forecasting in the ungauged urban catchments,as well as to identify the main variables influencing the first flush effect pollution for more accurate forecasting of the first flush effect.In short,this thesis can provide scientific support for the warning,predicting,and emergency control of urban floods as well as competent management of first flush effect pollution. |
