Research On Uncerntainty Analysis Of Flood Frequency And Flood Forecast

Flood is one of the most serious disasters in China, which has raised barriers to the sustainable development of economy and society. Due to the impact of climate change and human activities, the evolution and driving mechanism of flood has changed significantly, thus increasing the uncertainty of flood. Uncertainty analysis for flood frequency analysis and flood forecast under the changing environment may help to evaluate the regional flood risk and mitigate the risk by reservoir regulation.In this paper, Dadu River basin is selected as the study area. The technical framework of flood uncertainty is proposed from the perspective of flood frequency analysis and flood forecast. The main contents and conclusions of this paper are as follows:(1) The uncertainty analysis for flood frequency based on Bayesian theory has been put forwards. The parameters of flood frequency distribution are considered as random variables, and Bayesian Markov chain Monte Carlo (MCMC) method based on Metropolis-Hastings algorithm is used to evaluate the posterior distributions of GEV distribution parameters and flood quantiles. The results show that the Bayesian MCMC method is an effective tool to estimate the parameters of the flood frequency estimation which gets a better fitting effect than the other parameter methods. With the help of prior information which is unrelated to asymptotic property of likelihood function, posterior distribution of the parameters and quantiles of GEV distribution obtained from Bayesian estimation includes more information compared with the classical statistical methods in flood frequency analysis, and the uncertainty caused by uncertainty of parameters can be expressed. In addition, confidence intervals obtained by Bayesian MCMC method have been compared with those obtained by Delta method, which shows that the width of confidence interval by Bayesian MCMC method is narrower. The distance between upper confidence limit and estimated value of is greater than the distance between lower confidence limit and estimated value, which is more close to the actual situation.(2) The nonstationary analysis for flood frequency based on the Generalized Additive Models for Location, Scale and Shape parameters (GAMLSS) has been put forwards. The GAMLSS were fitted using three spline interpolation method with time, weather circulation and reservoir indicators as explanatory variables. The results show the presence of clear nonstationarities in the flood of Dadu River basin. The model with weather circulation and reservoir indicators as explanatory variables fits best. The nonstationary quantiles for the return period of 50 years by three models are compared with those calculated by P-Ⅲ. Little difference is found by the consistency model, followed by weather circulation and reservoir indicators and time indicators. The application of nonstationary analysis shows the differences between the nonstationary quantiles and their stationary equivalents, which suggeststhe urgent need for nonstationary modeling of flood in Dadu River basin.(3) The parameter optimization algorithms and parameter uncertainty method are introduced. Based on the genetic algorithm (GA) and the particle swarm optimization algorithm (PSO), GA-PSO algorithm is used for parameter optimization. The results show that the GA-PSO algorithm not only has the advantages of genetic algorithm with global convergence, but also uses the particle swarm algorithm to speed up the convergence rate. In addition, parameter uncertainty is analyzed using GLUE method.30000 groups of six parameters are randomly generated to the range of uncertainty of flood at the confidence level of 95%. The ranges of uncertainty contain the flow which shows that the GLUE method is suitable for the uncertainty estimation of parameters of hydrologic model.(4) The Bayesian processor of forecasting (BPF) and hydrologic uncertainty processor (HUP) are introduced and applied in the Dadu River Basin. To avoid the normal quantile transform, a Bayesian probabilistic flood forecasting model based on Copula function (C-BFS) is proposed. Copula function is used to describe the prior density function and the likehood function. The results were compared with BPF and HUP, the results are demonstrated that C-BFS model improves the forecasting accuracy compared with deterministic model and is slightly better than the other models (BPF and HUP) in terms of posterior median forecasting. In addition, the confidence interval of the C-BFS is more excellent than the other two models. The proposed C-BFS model needs neither linear nor normal distribution assumption, and is more flexible in practice.Based on the above research, the uncertainty analysis theory of flood frequency analysis and flood forecast will be further developed under the changing environment, which provides scientific basis for flood risk management.