The Semi-parametric Spatial Extreme Value Model With The Application Of Precipitation

An important goal of studying extreme precipitation data is to forecast the trend of precipitation in the region.Based on the study of distribution of extreme precipitation data in meteorological stations,it is expected to interpolate the distribution of precipitation in other regions.We can make appropriate preparations in advance to reduce the loss caused by extreme precipitation.It is an important reference for disaster prevention and control.Because the meteorological stations are discrete and the data amount is a set of finite points,and there may be missing data in the data sets of each station,we can not simply conclude with the time series analysis.We need to establish the space-time model to find the appropriate interpolation method.In this paper,we study the annual precipitation data continuously in Jiangsu Province.Based on the generalized extreme value distribution,we propose a semi-parametric or semi-parametric spatial extreme value model.Because it is based on the idea of generalized partial linear model,we first need to know the attributes of the parameters: whether it is a parameter or a non-parameter.In the premise that the extreme distribution parameters are related to the information of latitude and longitude of weather stations,we establish the semi-parametric spatial extreme value model.The model involves the use of kernel functions.Taking into account the distance from the spatial point to the fixed point.The farther the distance is,the smaller the value of kernel function is.In order to reflect the distance between points,the information of latitude and longitude of weather stations are used in the kernel function.In order to reflect the distance between stations,the latitude and longitude information of weather stations is used in the kernel function.The spatial correlation between the data is fully considered,it is helpful to explore more accurate conclusions.In this paper,we use the daily precipitation records of 12 meteorological stations in Jiangsu Province from 1951 to 2013.After preprocessing,the data is converted into annual maximum precipitation data to analyze its distribution.The three parameters of the extreme value distribution are obtained by the full-data with semi-parametric spatial extreme value model,the fitting results are comparable to the maximum likelihood estimation.And in some cases are even better.First,we assume that there is no precipitation data in one station,thanwe set up the semi-parametric spatial extreme value model to get the parameter information of this station.Second,we use maximum likelihood estimate with the parameter information of certain station and use spatial smoothing method to get the parameter information of the station.We gain the conclusions.We think that the model can be used to analyze historical precipitation data to get the better estimation.It is suitable to use the semi-parametric spatial extreme value model to fit the distribution of precipitation of areas which have no observed value.