Design Flood Estimation Considering Nonstationarity Due To Climate Change

Design quantile of flood series is an important reference for the design and operation of hydrological project and water resources management. Traditionally, design flood quantile is derived under the assumption of stationarity of the observed flood series. However, under the conditions of climate change and large scale human activities, the stationarity of the flood series is suspected and the corresponding design result made on this basis is questionable. There is a need then for development of effective methods for deriving design flood quantile under nonstationary conditions.Generally, it is customary to adopt only the annual maximum flood series or peaks over threshold series in nonstationary flood frequency analysis (FFA), but nonstationary FFA considering the statistical characteristics of daily flow series is rare seen. In this thesis, a nonstationary FFA method based on the theory of maximum domain of attraction of Gumbel distribution is proposed by considering the statistical characteristics of annual daily flow series. With the theory of time-varying moment, for a given design probability, design quantiles varying from one year to the next can be derived. This design result, however, can hardly be applied in application since it is impossible to change the design standard every year. In the present study, a physically based nonstationary design flood quantile analysis method is proposed by combining the ENE (Expected Number of Exceedances) and EWT (Expected Waiting Time) interpretations of return period with the technique of statistical downscaling of meteorological factors. Three representative regions within China, i.e. the Weihe River basin, the Qingjiang River basin and the Xijiang River basin, are selected as case studies. The main contents and results of this thesis are summarized as follows:(1) A pre-examination of the observed annual maximum flood series of the three representative basins indicates different types of nonstationarity. The results of the nonstationary frequency analysis using the GAMLSS are in accordance with those of the pre-examination. Nonstationary FFA using the GAMLSS with meteorological covariates can not only provide a model with physical meaning but also improve the model performance comparing with taking time as covariate. At the same time, the relations between the norming constants of the Gumbel distribution and the statistical parameters of the parent distribution of annual daily flow series are built based on the maximum domain of attraction of the Gumbel distribution. A nonstationary FFA method considering the statistical characteristics of annual daily flow series is finally proposed and termed as the norming constants method (NCM). The performance of the nonstationary FFA based on the NCM is superior to the GAMLSS model using time as covariate and is close to the GAMLSS model with meteorological covarites.(2) The outputs of 9 GCMs under 3 Representative Concentration Pathways (RCPs), i.e. RCP2.6, RCP4.5 and RCP8.5, are employed for deriving the projections of future precipitation and temperature with the convenience of the statistical downscaling technique. Results indicate that the future evolutions of the temperature for the three representative basins are quite similar. There are increasing trends in the annual mean temperature series under all the 3 RCPs before the year of 2040, and then it remains stable under the RCP2.6 but continues to increase under the RCP8.5. Under the RCP4.5, the annual mean temperature still increases until the year of around 2070 after which it remains stable. While the future evolutions of the precipitation for the three representative basins are quite different. For the Weihe River basin, the annula total precipitation transforms from a slightly decreasing trend to a slightly increasing trend with the increase of the radiative forcing. But for the Qingjiang River and Xijiang River basins, the annual total precipitation exhibits different degrees of decreasing trend under the 3 RCPs, and the decreasing trend becomes more obvious with the increase of the radiative forcing.(3) Physically based nonstationary design flood quantile is derived under the ENE and EWT interpretations of return period based on the nonstationary FFA and statistical downscaling of meteorological factors. Results indicate that for the Weihe River and Xijiang River basins with obvious nonstationarities, the nonstationary design flood quantiles are obviously different from the stationary design results. The nonstationary design results of the Weihe River basin are lower than those of the stationary case, which suggests that the occurance of flood event in the future will ease. Under this circumstances, taking the nonstationary design results as reference standard can avoid the waste of economic cost to some degree. On the contrary, the nonstationary design results of the Xijiang River basin are higher than those of the stationary case, which suggests that the occurance of flood event in the future will get worse. In this case, taking the nonstationary design results as reference standard can reduce the failure risk of hydrological project. For the Qingjiang River basin with no obvious nonstationarity, the nonstationary design results are quite close to those of the stationary case, which suggests that the occurance of flood event in the future will be comparable with historical conditions. The nonstationary design results with meteorological covariates are more plausible comparing with the case of using time as covariate which seems overstated due to unreasonable extrapolation.(4) The statistical uncertainty of the nonstationary design flood quantile is analyzed by the Bootstrap method. Results indicate that for the Weihe River and Xijiang River basins with obvious nonstationarities, the nonstationary design flood quantiles are far out of the 95% confidence intervals of the stationary case, which suggests that the degree of nonstationarities cannot be ignored even though the uncertainty is considered. While for the Qingjiang River basin with no obvious nonstationarity, all the the nonstationary design flood quantiles are involved in the 95% confidence intervals of the stationary case, which suggests that there is no big difference concerning whether the nonstationarity is considered in the design flood quantile analysis.