Non-stationary Statistical Model And Its Applications In Hydro-meteorological Frequency Analysis

Under the influences of climate change and human activities,it has been widely reported that the climatic and hydrologic conditions were changed in most regions of the world.Especially for the urban regions,the changing mechanisms and driving forces of meteorological and hydrological conditions reveal significant variations.Therefore,the tradition frequency analysis,which is based on the assumption that sample series are stationary,is no longer suitable for frequency analysis under environmental variations.By choosing Beijing as the study area,this paper studied the changing patterns of hydrological and meteorological factors by using statistical analysis.Two extreme value distribution(GEV and GP)and GAMLSS model were used to compare the performance between stationary and non-stationary model when fitting different sample series.A non-stationary model has a certain physical basis was built by considering the influence of human activities.The main conclusions are:(1)The inter-annual variations of precipitation and temperature display obvious spatial patterns: urban heat island and urban rain island effects are significant in the central urban area,which intensify the risk of urban waterlogging.The underlying surface in Beijing was changed rapidly,and the impermeable area was increased significantly.As a result,a significant increasing trend can be detected from the annual runoff series.(2)Compared with the traditional estimation methods(L-moments and maximum likelihood estimation),Bayesian,generalized maximum likelihood estimation,Rigby–Stasinopoulos and Cole–Green algorithm can generate more accurate parameter estimators.(3)For a non-stationary time series,the non-stationary model(GEV,GP and GAMLSS)can reflect the changing patterns of sample series by using different functions to describe the relationship between location parameter and time t.Thus,the errors of non-stationary model are less than that of stationary model,and the results of frequency analysis can be better utilized in practice.(4)A non-stationary model,which only considers the relationship between parameter and time t,is merely based on the statistical aspect to fit the sample series,so its physical meaning is unclear.By introducing the influencing factors such as meteorological conditions and human activities,the non-stationary model can better describe the sample series.