| Xinjiang is far from the sea,with vast territory,surrounded by mountains and rare precipitation.With a high degree of non-normality and randomness in space,the formation of precipitation is not the result of the action of a certain element.However,the precipitation monitoring stations are sparse and unevenly distributed,mainly distributed at the foot of the mountain and at lower altitudes.There are few sites on the top of the mountain and desert areas,especially in the Tarim Basin and the Zhungeer Basin in Xinjiang.Xinjiang's complex geographical location and topography,combined with rare precipitation observation sites,have greatly limited the study of precipitation in Xinjiang.The temporal and spatial distribution of precipitation and its future trendsunder climate change scenarios are important for Xinjiang's future agricultural production,disaster protection and ecological balance construction.Therefore,it is an urgent research topic to combine the precipitation and the easy-to-acquire,long-term observation and wide-ranging auxiliary variables to study and estimate the precipitation.Because of the strong correlation between precipitation and vegetation index(NDVI),and the long-term observation and coverage of satellite remote sensing data vegetation index,indirect estimation of precipitation has become an important means to overcome the above-mentioned site observations.Due to the complex geographical location and topography of Xinjiang,precipitation has spatial heterogeneity in different regions.The distribution of precipitation may be affected by different factors at different time scales and in different regions.In this paper,based on the spatial and temporal sample point observation data of precipitation in Xinjiang,the space-time correlation of precipitation distribution was fully considered.Based on the multivariate spatio-temporal geostatistics theory,combined with the spatio-temporal autocorrelation of precipitation,considering the influencing factors such as vegetation index,topographic factor and water vapor pressure,thespatio-temporalestimationwere completed.The time trends of the annual,seasonal,and monthly time scales,as well as the spatial distribution characteristics of precipitation in Xinjiang were analyzed.According to the cumulative anomaly characteristics of Xinjiang's annual and seasonal precipitation in Xinjiang,the abrupt change year of precipitation trend was determined.In the estimation process of precipitation,not only the relationship between vegetation index and precipitation was considered,but also topographic factors such as altitude,longitude,latitude,slope and aspect were introduced as auxiliary variables.In addition,a time index was added to adjust the seasonal variation of precipitation and vegetation index.And a polynomial multiple regression model was established combined with these auxiliary variables.The residual after regression used the median smoothing method to further extract time effects,spatial effects,and overall effects.The spatiotemporal trend term of precipitation was jointly extracted by the method of polynomial regression and median smoothing.After the residuals after removing the spatio-temporal trend term meet the stationary condition,the spatio-temporal variogram was modeled for the residual,and the precipitation residual was estimatedusing the space-time Kriging model.Estimated precipitation residual results,polynomial regression fitting results and median smoothing overall effects,time effects,and spatial effects were the estimates of precipitation.The results of the time-space regression median smooth kriging,the space-time regression kriging,the space-time median smoothing kriging,the space-time ordinary kriging four models respectively used leave a cross-validation.The cross-validation results and observations values were evaluated for accuracy,and the root mean square error,mean error,correlation coefficient,and mean absolute error evaluation index values between them were calculated.The accuracy of the space-time regression median smooth Kriging is the highest,indicating that the spatio-temporal trend of precipitation is extracted combination with the polynomial regression and median smoothing method,which helps to improve the estimation accuracy of precipitation space-time Kriging.The regression method is used to introduce the auxiliary variables to estimate the space-time Kriging model.Only the traditional statistical correlation between the main variables and the auxiliary variables is considered,and the spatial autocorrelation and continuity of the auxiliary variables in geostatistics are not considered.Cokriging not only considered the correlation between variables,but also considered the spatial autocorrelation and heterogeneity of the auxiliary variables.Through correlation analysis,this study found that the correlation coefficient between water vapor pressure and precipitation is the largest,and the water vapor pressure is used as the covariate of precipitation estimation.And the CoKriging is expanded in time and space,and the construction process of the space-time cross variogram is derived.The space-time precipitation was estimated using space-time CoKriging method from January 1960 to December 2013 for 54 observation stations in Xinjiang.Before constructing the spatiotemporal variogram,time series analysis of precipitation and vapour pressure was carried out to remove the time period term of precipitation and vapour pressure,leaving random items and trend items.Then,the precipitation and the vapor pressure after the removal of the period are analyzed for spatial trend,and the spatial trend was removed.Next,the spatiotemporal variogram model wasrespectively established for the precipitation and water vapor pressure after removing the spatiotemporal trend.In the modeling process,only the spatiotemporal domain sample in the range was selected to reduce the calculation amount of the program,and the spatiotemporaldirect variograms and spatiotemporal cross variograms of precipitation and water vapor pressure were constructed.Then,the weight coefficients of precipitation and water vapor pressure were solved by a matrixconstructed of spatiotemporal variograms.And the interpolation of the residual of theprecipitation was completed using the spatio-temporal CoKriging.Finally,the estimation result of the residual was added to the periodic term of the time and the spatial trend term to obtain the estimation result of the precipitation.The model cross-validation results show that the root mean square error of the spce-timeCoKriging model is at least 9.68 mm,and the correlation coefficient of the space-time regression median smooth Kriging method is 77.72%,which is higher than the space-time CoKriging.And the correlation coefficient of the space-time regression Kriging is higher than that of the space-time median smooth Kriging.The value of the mean absolute error of the space-time median smooth Kriging method is at least 0.057 mm,and the mean absolute error of the space-time regression median smooth Kriging method is at least 7.07 mm.On the whole,the space-time CoKriging method and the space-time regression median smooth Kriging method are superior to the other three interpolation methods.The time series of annual and seasonal precipitation in Xinjiang,northern Xinjiang,southern Xinjiang and Tianshan Mountains and their cumulative anomalies were established using the sub-regional and seasonal methods.The results show that the average monthly precipitation in the Tianshan Mountains is the highest,followed by Northern Xinjiang and the lowest in Southern Xinjiang.The time series of precipitation in Xinjiang,northern Xinjiang,Tianshan Mountains and southern Xinjiang show an insignificant growth trend.Precipitation has a slight downward trend before 1986 and has a slight growth trend after 1986.Through the analysis of the precipitation time series from January to December 2013,it is found that the precipitation in Xinjiang from October to March is in a low value area,that is,the freezing period in Xinjiang.Precipitation begins to rise in April and rises to a certain high level in May.The precipitation,while in June-August(summer)reaches the maximum in a year.The precipitation in the annual and seasonal precipitation of all climatic zones is the largest in summer,and the precipitation in winter is the smallest.The precipitation in spring is slightly larger than that in autumn,indicating that the contribution of summer precipitation to annual precipitation and local climate anomalies cannot be ignored.The spatial distribution characteristics of annual average and seasonal average monthly precipitation were studied in January-December 2013.The results show that the characteristics in the spatial distribution of the annual average precipitation in Xinjiang is,that the north is more than the south,the west is more than the east,and the mountainous area is more than the plain.The spatial distribution characteristics of summer precipitation are the closest to the spatial distribution characteristics of annual average precipitation.Summer precipitation is more prominent and wider range than the annual average precipitation in the central and western precipitation areas of Tianshan Mountain.However,the above spatial distribution feature of precipitation in spring and autumn is gradually weakened,while the precipitation in winter appears to be less in the mountainous areas than in the plains and less in the west than in the east.Therefore,the feature that the north is more than the south,the west is more than the east,and the mountainous area is more than the plain is only applicable to summer precipitation and annual average precipitation.In summary,the spatial and temporal trends of precipitation are extracted by combining polynomial regression and median smoothing methods,which helps to improve the estimation accuracy of precipitation space-time Kriging.The introduction of highly correlated water vapor pressure as a covariate for precipitation estimation can also improve the estimation accuracy of the space-time CoKriging interpolation method.The estimation results of space-time precipitation are of great significance for studying the temporal and spatial distribution characteristics of precipitation in Xinjiang,China. |
PDF Full Text Request