By using climate trend coefficient analysis the variation of observed and model predicted China summer temperature in the past 30 years. The dates come from NCEP/NCAR daily reanalysis date and National Climatic Center monthly models temperature date. Through empirical mode decomposition method fitting the raise trend of observed temperature and remove the trend in the context of global warming. Combined with systematic error correction and seasonal prediction method - statistical method of combining power to analysis the trend on the impact of summer temperatures forecast. The results show that : most of the area in China has a significant increase trend in summer temperature in last 30 years. In most areas the climate trend coefficient get through 0. 01 significance level. But mode temperature almost have no change in trend in last 30 years, there is obviously inadequate for global warming. In order to avoid this kind of model simulation of observed temperature the overall trend is insufficient to predict effect and the effect of the optimal predicted results. Using the empirical mode decomposition method can effectively fitting and remove the trend of observed temperature. When forecasting the temperature we remove the trend first and add the trend in the last. Trough tests find that the method can significantly improve the result compare with direct forecast result in most area. And solve the problem of forecast result is lower than observed result. All the results show that it is necessary to remove the observed temperature increase trend in numerical prediction model results in post - processing.