Nonstationary Flood Frequency Analysis Considering Historical Flood

The robustness of flood extreme event frequency analysis is of great significance to engineering planning and design and flood defense management.Improving the robustness is one of the core issues in flood frequency analysis.However,in the context of climate change and human activity impact,the flood data used for analysis may be nonstationary,which may lead to a large estimation error of the traditional flood frequency analysis based on the stationary assumption.Another problem is that in many river basins,hydrological designers often face the situation that systematic(gauging)data of less than 100 or even 50 years is required to estimate a 100-year flood(the flood with 100-years recurrence interval)or even 1000-year flood.Therefore,in order to improve the robustness of flood frequency analysis,the methods to incorporate the historical flood data into the framework of nonstationary flood frequency analysis are urgently required.The selected study areas include the Weihe,the Hanjiang,and the upper Yangtze River Basins.The key contents and results in the thesis are as follows:(1)The Mann-Kendall and Pettitt tests and single factor correlation analysis are used to conduct preliminary diagnosis and identification of the stationarity of the annual maximum daily flow(AMDF)in the study basin.The results show that each of the Xianyang and Huaxian gauging stations in the Weihe River,the Ankang,Huangjiagang,and Huangzhuang gauging stations in the Hanjiang River,and the Yichang gauging station of the upper Yangtze River has a statistically significant decreasing trend and a statistically significant change point.The AMDF series at the Yichang station has a change point in 1960s,while those at the other five stations have change points in the1980s.For the Weihe Basin,the factors such as climate and land use/cover change(LUCC)have changed in a specific period,and these factors have a significant correlation with floods at the outlet.For the he upper Yangtze River Basin,the climate and reservoir factors have significant changes,and these factors have a significant correlation with the flood at the outlet.The above results indicate that common factors driving the nonstationarity of the AMDF series include climate,LUCC and reservoir.(2)In order to assess the upstream multi-reservoir effects on the nonstationarity of the downstream AMDF series,this paper proposes a rainfall-reservoir composite index(RRCI)containing an rainfall index to reflect the degree of constraint of inflow characteristics on the reservoir flood control.The application of the RRCI index in the Hanjiang Basin shows that the RRCI has increased after the main reservoir completion and has a high negative correlation with the AMDF series.And the correlation coefficients at Ankang,Huangjiagang,and Huangzhuang stations are-0.53,-0.71,and-0.70,respectively.It is indicated that the RRCI can well explain the nonstationarity of the AMDF series in the Hanjiang Basin.(3)To carry out the frequency analysis of the nonstationary AMDF series,a covariate-based nonstationary flood frequency analysis method is used in the Weihe Basin and Hanjiang Basin.The results show that in the Weihe Basin,the optimal model based on climate and LUCC covariates performs better than that only based on climate covariate or only based on LUCC covariates,and the best explanatory variables for the nonstationarity of the Weihe floods are annual summer precipitation anomaly,annual total precipitation anomaly,plowland area,woodland area,and grassland area;in the Hanjiang Basin,the optimal model based on the RRCI index performs better than that based on the RI_s index.(4)To deal with the estimation errors of large floods due to the short length systematically gauging data in nonstationary flood frequency analysis,a method to incorporate the historical flood data into the nonstationary flood frequency analysis was proposed.In this method,based on the time-varying moment method and the likelihood function,the likelihood of the flood censored data(consisting of historical flood survey data and systematic gauging data)with the nonstationarity was described.Take the Yichang station as an example.The results show that under the two conditions,i.e.stationary and nonstationary assumptions,after the incorporation of the historical flood censored data from 1470 to 1881 in the Yichang station,the estimation for the rare flood events is significantly changed,and the uncertainty of the shape parameter which reflects the behavior of the tail of the GEV distribution is reduced.Moreover,due to the incorporation of historical data,under the same covariate values,the flood estimates with the exceeding probabilities of less than 1%are increased,and the lower the probability,the greater the increase of flood estimate.It is concluded that in the nostationary frequency analysis,considering the historical flood can effectively reduce the uncertainty of the large flood estimation resulting from the short length systematically gauging data.(5)To assess the uncertainty of flood estimation in the nonstationary flood frequency considering the historical flood,a method based on variance decomposition to separately deal with the uncertainty of historical flood and parameter estimation is proposed.The results of the uncertainty assessment in the flood estimation at Yichang station under the nonstationary condition show that the parameter estimation and historical flood uncertainty leads to the uncertainty of the flood estimation.And the impact of the former is significantly greater than the latter,which means that if only the accuracy of the data is not too bad,it is suggested to use the data to reduce sampling errors.Considering the effects of the Three Gorges and the other upstream reservoirs,the nonstationary models produce the lower estimation than the stationary model for the floods with the exceeding probabilities of less than 1%.However,the their confidence intervals are relatively large.For example,when the value of summer precipitation anomaly is the 95th quantile and the value of reservoir index is the level in 2017,the flood with the exceeding probability of 1%was estimated to be58237±13265 m~3/s and the flood with the exceeding probability of 0.1%was estimated to be 65312±19903 m~3/s.This indicates that under the current reservoir impact level,the estimates of extreme floods may still be relatively uncertain.