关键词: 水文频率分析, 非一致性, GCM, 时变矩法, 分解合成法

Abstract: Because of global climate change and anthropogenic activities，the assumption that the hydrological time series is stationary will be no longer valid. Thus，the design values calculated with the traditional hydrological frequency analysis method based on the stationarity assumption might be unreliable and thus the nonstationary frequency analysis methods are required. This paper is aimed to study the impacts of the nonstationarity of the annual maximum （AM） 24 h precipitation series of the Weihe River （Weihe） Basin of Shaanxi Province，China on the design values for the given return periods. Both the GCM daily precipitation data series for the time period of 1901-2100，which is obtained by combining the outputs of the 20c3m and sresa1b emission scenarios of model csiro_mk3_5，which are available on the website （http://www-pcmdi.llnl.gov/），and the observed daily precipitation data of 20 rainfall gauges within the Weihe River （Weihe） Basin for the time period of 1951-2011 are collected. With these data，two AM series of 24 h precipitation over the whole Weihe River Basin are prepared，one is the 1901-2100 GCM series and the other is the observed 1951-2011 series. Because of the coarse spatial resolution of the GCM output，there exists great uncertainty with the 1901-2100 GCM series which must be rescaled firstly. Based on the theory of statistical downscaling method and the observed AM 24 h precipitation series of 1951-2011，the AM 24 h GCM precipitation series of the 1901-2100 is rescaled by the mean-correction factor and standard deviation-correction factor to reduce the errors in the GCM outputs. With the rescaled AM 24 h GCM precipitation series，two nonstationary frequency analysis methods，i.e. the method of time-varying moment and the decomposition and combination method，are employed to estimate both the 100 a and 500 a design values of the AM 24 h precipitation for the design years of 2050，2075，and 2100，respectively. The results of the two nonstationary frequency analysis methods have significant discrepancy comparing with those of the traditional method which is based on the stationary hypothesis. Three conclusions are made as follows. First，there are nonstationarities in both the mean and variance of the rescaled AM 24 h GCM precipitation series. Second，the results of the time-varying moment method are obviously different from those of the traditional frequency analysis method. For the year of 2100，the AM 24 h precipitation design values obtained by the former method for the return periods of 100 years and 500 years are 29.09% and 39.99% greater than the design values from the latter method，respectively. Third，for the method of decomposition and combination considering both the nonstationary mean and variance，its design value has the similar trend with those of the time-varying moment method for the same return period，but has a relative lower increasing rate than the latter method for the return periods of 500 years. For the year of 2100，the AM 24 h precipitation design values obtained by the former method for the return periods of 100 years and 500 years are 26.91% and 19.69% greater than the design values from the traditional method，respectively.a

Key words: hydrological frequency analysis, nonstationarity, GCM, the time-varying moment methodthe decomposition and combination method