Nonstationary Flood Frequency Analysis Based On Covariates

In a changing environment,the impacts of climate change and frequent human activities may cause the alteration of global water cycle,ice melting and the condition of land use/land cover,and further affect the genetic mechanism of hydrologic variables.Traditional frequency analysis was implemented base on the assumption of stationarity,which is now open to doubt.The Wa ngkuai Reservoir,which was under the substantial change of climate,land use/land cover and increased number of soil-water conservation projects in Daqing River Basin,was taken as the study area in this thesis.The annual maximum flood series of Wangkuai reservoir were adopted to identify the nonstationarity and the associated impact factors.Then,the nonstationary flood frequency analysis of univariate and bivariate-joint distribution were taken and worked out the flood quantiles and the combined values of flood peak and certain flood volumes under certain exceedance probability.The main conclusions and innovations are summarized as follows:(1)The variations in annual maximum flood series of Wangkuai reservoir were detected by change-point test and trend test.First,Pettitt test was employed to identify the most likely change-point,then Mann-Kendall test was adopted to identify the trend of the flood series and the subseries divided by change-point.Finally,the nonstationary form of the flood series was determined to be change-point 1979,which was estimated through a comprehensive analysis taking the conditions of land use/land cover into consideration.(2)The impact factors of the flood series were analyzed in two aspects.On the one hand,some climate indices which would have a certain effect on the study area were selected,and the teleconnections between these climate indices and the flood series were then analyzed through Kendall's tau test and Spearman's rho test.The results showed that Ni?o3 index with lead time of 3-months and NPO index with lead time of 6-months yielded a 95% significant level in the correlation test,and these two indices were determined to be the candidate climate indices(CI).On the other hand,the storage capacity and the catchment area of the soil-water conservation projects were calculated to be 3.95 mm and 1248.72km2,respectively,based on the rainstorm and flood data.And on this basis,flood restoring calculation was employed to estimate the extent of soil-water conservation project impacts on the flood events.Then,the impact factor of the soil-water conservation projects(CDI)was estimated.(3)This thesis addressed two models for nonstationary flood frequency analysis based on GAMLSS theory.In the model with time as covariate(model 1),the Log-Normal distribution offered the best overall performance,and was selected as the most suitable distribution.While in the model with CI and CDI as covariates(model 2),the Weibull distribution presented the best fitting r esults,similar to model 1.The modeling results of centile plots in model 1 showed that all flood series exhibited a steep decreasing trend;while the cetile plots in model 2 revealed undulating behaviors which tended to fit the flood data reasonably well,thus the flood value should be a dynamic changing process under the effected of certain climate and human activities.In addition,flood quantiles were employed to describe the dynamic changing process,and the scope of 90%,95%,98% flood quantiles were took as paradigms;refer to the scope,the fixed design flood which was approved by WRHPDGI would be safer and more conservative under low flow years,while it can be just the opposite under high flow years.(4)Taking the best modeling distributions in(3)as marginal distributions,copula functions were adopted to build the dependence structure of flood peak and annual maximum 1-day flood volume,and the results showed that only Gumbel-Hougaard Copula can provide the best joint distribution.In the model with time as covariate of the marginal distribution(model 3),the combination values of flood peak and volume and the associated “most likely” design events exhibited a decreasing trend under the same exceedance probability,while in the model with CI and CDI as covariates of the marginal distribution(model 4),the joint distribution can describe the undulating behaviors of different years reasonably well,and the combination values revealed some more complicated cases,caused by the intersection of contours of different years.Under the same exceedance probability,the “most likely” design events were smaller than that estimated by the marginal distributions with fixed covariates in most of the study period,while it was the opposite in 1956,1959 and 1963.