Abstract: It is significant to establish high accuracy temperature grid data set with long time series to study climate change，crop growth and development，disaster impact and assessment. Based on observed climate standard values data from 104 meteorological stations and annual temperature data in Northeast China from 1971 to 2000，by using the spatial interpolation methods，the annual mean temperature temporal and spatial distribution in Northeast China were studied and the accuracy of the interpolation methods was evaluated with the error indicators，temperature characteristic values and time series temperature curves. The result showed that，on the spatial scale，the interpolation results based on the Partial Thin Plate Smoothing Splines（PTPSS）with the elevation as the covariate was better than other methods，with the RMSE 0.341℃，and the MAE 0.264℃，followed by Cokriging method（with the elevation as covariates），again as the ordinary Kriging method. On the temporal scale，the time series annual mean temperature curve form 1971 to 2000 based on PTPSS（with the elevation as the covariate）was consistent with the time series temperature curve extracted from the reference grid temperature data set. The correlation coefficients of the mean temperature，maximum temperature and minimum temperature curves extracted from the interpolated temperature grid set based on PTPSS and the corresponding curves from the reference grid temperature set were greater than 0.9. The results indicated that the PTPSS interpolation method with the elevation as the covariate was the most suitable for the construction of the annual mean temperature surface in Northeast China.

Key words: Regional average temperature, Spatial interpolation, Accuracy evaluation