Study on daily mean temperature modeling

Daily mean temperature is the important atmospheric parameter; also have been one of the essential data sets for climate change study. And by now the studies on daily mean temperature mainly focus on Kriging theory based methods, which is to realize gridding, interpolating, extrapolating and analysis in time dimension and/or space dimension. However, it´s supposed that small scale variation part in the spatial data has been met stationary assumption in Kriging based methods. It´s may not be suitable for sorts of atmospheric data, and daily mean temperature included. In this paper, we propose a new method for modeling daily mean temperature and prove it by the daily mean temperature data set downloaded from China Meteorological Data Sharing Service System, which lasts 52a(1961-2012)including over 800 meteorological stations nationwide distributed in China. Considering the spatial correlation among this spatial data and longitudelatitudeheight, to build a new multiplicative multiple regression model based correlation coefficients to describe the big scale expectation function in daily mean temperature data set and meanwhile in comparison with traditional Kriging method. The result shows that the zero-mean residual from this new method, which represents the small scale random function in this data set, doesn´t show spatial correlation with longitudelatitudeheight and its spatial heterogeneity(0°/45°/90°/135°direction semivariograms) performs in nearly isotropy with better stationarity. Meanwhile the standard deviation and stable standard deviation almost smaller than those from traditional Kriging method, 95.63% and 91.80% respectively. In conclusion, the new method models daily mean temperature data set better than traditional Kriging method. After cross validation, the mean error (ME) is -0.003 and root-mean-square error (RMSE) is 1.652 from the new method, and traditional Kriging method results are 0.005 and 1.877 respectively.