Abstract: In our report, aimed at the island and tropical characteristics and the unique geographical features of Hainan Island, a random forest model based on spatiotemporal independence was designed, the observed data and the forecast data from ECMWF were used for the accurate correction for each station of 7 days forecast, which has 3 h precision; the proposed method was evaluated by the index of ＜2 ℃ accuracy, ＜1 ℃ accuracy and RMSE. The result showed that the proposed method is obviously better than the model of ECMWF, which is much closer to the observed temperature value, and which also revised the data form ECMWF very well.