Non-stationary Flood Frequency Analysisand Its Effect On Flood Control

Extreme floods are one of the most threatening natural disasters for human beings. Frequency analysis is an integral part of risk assessment and mitigation in engineering design and commonly relies on the assumption of identically distributed observations. However, increasing evidence demonstrate that large-scale modes of climate variability and land use change exert a significant influence on hydrological extreme events in various regions worldwide. In light of the substantial change of climate, land use/land cover and increased number of soil-water conservation projects in Daqing River Basin, the assumption of stationarity in annual maximum inflow flood series of reservoirs should be reconsidered; otherwise reservoirs will undertake risks and which will further influence the flood-control system in Daqing River Basin. Baiyangdian Lake Basin, which is the principal flood-control part in Daqing River Basin, was taken as the study area in this thesis. Annual maximum inflow series of four upstream reservoirs were analyzed using non-stationary flood frequency method to obtain the significant effect of climate and land use/land cover change on design flood, reservoir flood routing and regional flood control. Meanwhile, climate predictors were identified to predict the extreme flood/rainfall. The main conclusions and innovations are summarized as follows:(1)This thesis proposed two steps to detect the variation in annual maximum inflow series and the second step is named detailed diagnosis to detect the trend and change points of flood series. For change points detection, Pettitt test was employed to identify the variation range and was associated with other methods’ detection results, land use/land cover change and change point of annual maximum 30-day accumulated rainfall series in reservoirs catchments to confirm the most possible change point. The results reveal that the most possible change point of all non-stationary flood series are occurred in 1979 and they have significant temporal trend.(2)Based on the change point detection result, the conditional probability distribution and mixed distribution were analyzed for nonstationary flood frequency calculation. And the conditional probability distribution is inferior to the mixed distribution. Hence, the mixed distribution was employed to analyze the nonstationary flood frequency and the results demonstrated that(a) the mixed distribution provided a more appropriate and superior fit than conventional distribution(P3 distribution);(b) Compared with the design flood values estimated by P3 distribution, design flood values of nonstationary flood series estimated by mixed distribution have different extent of decrease under cases of different return periods. Each reservoir has the similar conclusion.(3)The two design flood hydrographs under series’ non-stationarity and taking no account of series’ non-stationarity were regarded as inflow hydrographs of each reservoir. Then they were used to implement flood routing and to obtain the flood routing results. Compared with flood routing results regardless of series’ non-stationarity, namely, maximum reservoir water level and maximum discharge, the results under environmental change have different degrees of decreases corresponding to different return periods in each reservoir. This thesis estimated the effect of environmental change on reservoir’s flood routing.(4)Similarly, there were two cases for considering series’ non-stationarity and taking no account of series’ non-stationarity. On the basis of discharge hydrographs of each reservoir with two cases, the design flood hydrographs of rivers were added correspondingly and the inflow flood hydrographs of Baiyangdian Lake under two cases are obtained. Furthermore, two inflow flood hydrographs are routed through Baiyangdian Lake to investigate the influence of environmental change on regional flood routing.(5)The Xidayang Reservoir catchment is the largest catchment in Daqing River Basin. The annual maximum daily inflow(AMDI) to the Xidayang reservoir was analyzed to identify the climatic drivers of this variability. Since there were limited land use change data, the annual maximum 30-day rainfall(AMR) series was used to assess whether the AMDI trend can be ascribed to climate. A Bayesian nonstationary model was developed for the AMDI and AMR using lognormal and generalized extreme value distributions with climate predictors. We find that the average May-June-July Sea surface temperature anomalies in the Northern Indian Ocean and Western Pacific Ocean show a high negative correlation with the AMDI and the 30-day AMR series. Next we compare models of AMDI and AMR under three assumptions: a) time-invariant, b) linear temporal trend and c) climate informed, and find that the climate-informed models exhibit the best performance according to Deviance Information Criterion(DIC) and Bayesian coverage rate using 90 th percentile for both AMDI and AMR. Leave one out cross validation(LOOCV) is used to demonstrate that these models can provide useful prediction guidance for flood control and flood/rainfall risk management of reservoirs in Daqing River Basin, before the monsoon season begins, thus facilitating adaptation to a changing climate.