Study On Non-Stationary Designed Rainfall Based On GAMLSS Model

The design rainfall is an important reference basis for the project planning design,operation management,and the development and utilization of water resources in the basin.Low design standards will lead to disasters and accidents,as a result of loss of life and property.High design standards may increase construction and management costs,thus,result in waste.The traditional calculation of design rainfall is based on the assumption that the rainfall series are statistically stationary.However,under the changing environment,due to the impact of climate change and human activities,the rainfall series has shown a trend of non-stationary.Therefore,it will result in the deviation of the rainfall design value based on the stationary assumption at present.Therefore,the analysis of the non-stationary of rainfall and the further estimation of the rainfall design value under the non-stationary are of great significance to ensure the rationality of the results of rainfall frequency analysis and calculation,and can provide references for the operation and management of hydraulic engineering,the development and utilization of river basin water resources and flood control decision-making.In this thesis,the rainfall data of 9 rain measuring stations in the Suqian Section of the Ancient Yellow River basin were taken as an example to analyze the non-stationary of rainfall series in different periods,which are one year,flood season and non-flood season.The Generalized Additive Models for Location Scale and Shape(GAMLSS)theory is used to build a time-varying parameter model,through which the variation of future rainfall probability distribution can be deduced.The design methods such as the Expected Number of Exceedances(ENE),the Equal Reliability(ER),the Average Design Life Level(ADLL)and the Design Life Level(DLL)were adopted to calculate the non-stationary design rainfall.The Bootstrap method was used to analyze the uncertainty of the non-stationary design value.The main research results include:(1)Based on the constructed time-varying parameter GAMLSS model and the principle of AIC,the best fitting model is selected to analyze the frequency of non-stationary.The results show that the optimal distribution of rainfall series is mainly LOGNO distribution at each rainfall station.For the rainfall series in one year,Jinsuo Station is suitable for the μ timevarying non-stationary model,while other rainfall stations are applicable to the stationary model.For the rainfall series in flood season,all the rainfall stations are applicable to the stationary model.For the rainfall series in non-flood season,Zaohe Station,Buzi Station and Liulaojian Station are suitable for the μ time-varying non-stationary model,while other rainfall stations are applicable to the stationary model.(2)Based on the analysis of non-stationary rainfall frequency,according to the relationship between the model distribution function and time-varying parameters,the four design methods,ENE,ER,ADLL and DLL,are used to calculate the non-stationary design rainfall.The results show that the non-stationary design rainfall values calculated based on the four design methods are significantly higher than the stationary design results,and the two design methods of ER and ADLL are very similar on non-stationary design rainfall values,and relatively reasonable.(3)The Bootstrap method is used to analyze the uncertainty of non-stationary design rainfall.The results show that the uncertainty of the non-stationary design rainfall based on the four design methods of ENE,ER,ADLL and DLL are significantly different from that of the stationary design results.Among them,the uncertainty of the non-stationary design rainfall based on the methods of ER and ADLL are very similar and relatively small.(4)The design rainfall under non-stationary conditions is more reasonable than the result of stationary design,and it can reduce the risk to a certain extent.Based on the calculation and analysis results,it is recommended to use the two design methods of ER and ADLL to calculate the rainfall design values under the non-stationary conditions in the study area.