| The fourth assessment report of the Intergovernmental Panel on Climate Change believes that global climate change is an indisputable fact.Climate change can influence the current state of the global hydrological cycle and make the frequency of rainstorms much higher than before.Extreme precipitation can cause floods,droughts,landslides,mudslides,diseases and other disasters,which directly cause casualties and property losses.And extreme precipitation seriously threatens human survival and restricts regional socio-economic sustainable development.Due to the influence of human activities,the underlying surface changes drastically,the characteristics of rainfall-runoff change,and heat balance of surface and atmospheric circulation change abnormally,which cause variation in hydrological factors such as precipitation,evapotranspiration and runoff.Under the variation of hydrometeorological elements caused by the background of global climate change,the original uncertainty analysis of the spatiotemporal distribution of rainfall can not meet the existing standards of flood control and drought resistance in China.Therefore,the research on the uncertainty analysis of the spatiotemporal distribution of extreme rainfall under hydrological variability has become a hot issue in the current hydrological researches.For the hydrological variation diagnosis,the Pettitt method,the Mann-Kendall method,the time-segment sliding segmentation comparison sequence method,and the R/S method were used by predecessors.However,single methods like above methods tend to be biased toward the form of trend or jump in hydrological variation diagnosis.And in most single methods,a certain statistical indicator such as mean or variance was adopted to judge the variation point of series.The calculation accuracy of single method can not be guaranteed and different methods always obtain the didderent results.A single method cannot judge the variation form of the series as a whole.For the frequency analysis of a single variable,the distribution function such as P-III distribution,generalized extreme value distribution,extreme value I curve,eneralized Pareto distribution and so on were used by predecessors.And Maximum Likelihood Estimation Method,Optimization Fitline Method,Mixed Moment Method and Linear Moment Method were used to solve the distribution parameters.However,the above functions cannot fit the multimodal distribution,and the above parameter estimation methods are not ideal for solving parameters of the multimodal distribution.For the multivariate joint distribution,two variables are the research object in most of current multivariate joint distribution studies,and the studies of the joint distribution and its parameter solving method,which object are three variables,are less.For the spatial and temporal distribution of extreme precipitation,most scholars study the temporal and spatial trends of rainfall extremes from a statistical point of view,and there are few studies on the the temporal and spatial distribution of rainfall extremes from a probability perspective.Based on the above problems,the monthly maximum precipitation,date and duration of 16 meteorological stations in Sanjiang Plain of Heilongjiang Province were taken as the research object in this paper.The hydrological variation comprehensive diagnosis method was used to diagnose the variation point.The sample for frequency analysis was selected according to the variation point.PIII function,Gaussian mixture distribution and Logistic were used to frequency analysis on each variable respectively.The real-coded genetic algorithm,expectation maximization method and maximum likelihood estimation method were used to solve the parameters of edge distribution.Based on the results of a single variable frequency analysis,Clayton Copula,Frank Copula and Gumbel-Hougaard Copula functions were used to establish a three-variable joint distribution for frequency analysis of the,date and duration.The probability of rainstorms in each meteorological station from June to September was calculated.Moreover,the spatial distribution of the mean monthly maximum precipitation and the coefficient of variation(Cv)of each month and the spatial distribution of the rainstorm probability of each meteorological station were obtained according to the results of univariate and multivariate frequency analysis.And the main conclusions were obtained as follow:(1)The variation point of maximum monthly precipitation series of each meteorological station in Sanjiang Plain is between 1981 and 1996.(2)P-III distribution has a good fitting effect in fitting the frequency distribution of maximum monthly precipitation series,and And the real-coded genetic algorithm solving the parameters of the P-III curve is highly accurate and easy to implement.(3)The Gaussian mixture model solves the problem that the single distribution function cannot fit the multi-peak distribution.Expectation maximization method simultaneously solves Gaussian mixture distribution parameters and distribution weights.The results showed that it is better to apply the Gaussian mixture model and the expectation maximization algorithm to solve the frequency analysis of date series of maximum monthly precipitation.(4)The accuracy of the Logistic distribution function and the maximum likelihood estimation method solving the frequency analysis of duration series of maximum monthly precipitation has a room to improve.(5)The Frank Copula fitting most of the variables selected in this paper is better,the Clayton Copula fitting the remaining variables is better,and the G-H Copula fitting the variables is not good.(6)In the whole year,the mean monthly maximum precipitation is maximum in June,July and August,the mean monthly maximum precipitation of September is less than it in June,July and August,and and the minimum mean monthly maximum precipitation is in December,January and February.In space,the mean monthly maximum precipitation in June-August is the maximum in the northwest,the mean monthly maximum precipitation in the southeast is smaller,and the minimum is in the middle of the plain.The mean monthly maximum precipitation in other months decrease from northwest to southeast.In the whole year,the Cv of monthly maximum precipitation is the minimum in July and August,the Cv of monthly maximum precipitation is the maximum in December,January and February.In space,the Cv of monthly maximum precipitation in March-October is larger in the northwest and southeast,and lower in the middle of the plain.The Cv of monthly maximum precipitation in other month is lower in the northwest and southeast,and higher in the middle of the plain.(7)The probability of rainstorm in July and August is larger,followed by the probability of rainstorm in September,and the probability of rainstorm in June was the smallest in the flood season in the Sanjiang Plain.The probability of rainstorm in the early to mid-June is relatively high.The probability of rainstorm in the mid-to-end of July and August is relatively high.The probability of rainstorm in the middle of September is relatively high.And the probability of rainstorm in June,July and August gradually decreased from northwest to southeast.The probability of of rainstorm in September gradually decreased from the middle to both sides of Snjiang Plain.(8)In the area of stretching range of Wandashan,such as Xinhua,Junchuan,290,Jiansanjiang and Yanjun,the probability of rainstorm is larger in June,July and August,and landslides,mudslides and flash floods should be prevented.In the plain areas such as 859,290,Friendship,Beixing,853 and Qingfeng,the probability of rainstorm in September is larger,and effective measures should be taken to reduce crop yields caused by rainstorm. |
PDF Full Text Request