Spatio-temporal Data Modeling

It mined new methods in temperature data field in Chinese mainland, the dataset contains daily mean temperature,longitude,latitude and altitude of nearly 800 meteorological stations throughout the country in consecutive 52 years(1961-2012).What the correlation among daily mean temperature and longitude, latitude and altitude shows as followed.It shows weak correlation when longitude is less than 105°E and latitude is lower than 2000m, but it shows negative correlation when longitude is greater than or equal to 105°E and latitude is high than or equal to 2000m.However it shows negative correlation on latitude from north to south throughout the whole study area. Two models, Model I and Model II, are constructed based on the different correlation performance above.Model I:Z(X,Y,Z)= al+I1·b1X+clY+I2·d1Z+v1(X,Y,Z),in which Z(X,Y,Z) stands for daily mean temperature, X/Y/Z stands for longitude, latitude and altitude respectively, v1(X,Y,Z) stands for error part, I1 and I2 stands for conditional function.Model II:Z(X,Y,Z)= a2+b2·+|rx|·X+c2·|rY|·Y+d2·|rZ|·Z+v2(X,Y,Z).The rx/rY/rZ stands for correlation among daily mean temperature and longitude, latitude and altitude.Least squares principle are used in calculating models parameters,and getting two kinds of residual,vl and v2, from Model I and Model Ⅱ respectively.These two kinds of residual are compared in standard deviation(v2<v1),stable standard deviation (v2<vl)and statistics,aslo analyzed in stationarity and heterogeneity. The variable v1’s standard deviation (si) ranges from 3.1-3.3℃, vl’s stable standard deviation (rl) ranges from 2.9-3.1℃, and the variable v2’s standard deviation (s2) ranges from 2.5-2.9℃, v2’s stable standard deviation (r2) ranges from 2.3-2.6℃. Meanwhile it did cross validation in comparing these two ways of modeling precision. The residual v2’s RMSE is 1.754℃,smaller than vl’s RMSE 2.285℃.The residual v2’s ASEis 1.593℃, also smaller than vl’s ASE2.17℃.The residual v2’s MSE is 0.003℃, also smaller than vl’s MSE0.005℃;SEP1.468℃ also smaller than vl’s SEP1.742℃.In conclusion,Model II performs in high precision than Model I.It proposed a new spatial outlier detecting method based on Delaunay triangular natural neighorhood,aslo used the new method to detect the outliers in mean temperature field efficiently.It applied correlation analysis method to discover that daily mean temperature’s stochastic distribution applies to multiplicative noise model.And 96.27% sations applies to negative trend,only 3.73% stations applies to positive trend,and these 27 stations are located in Szechwan Basin,and 77.78%of them with atitude lower than 500m.