Abstract: Based on the ECMWF,Grapes_GFS and SWCWARMS 2m temperature forecast within 168 hour-period of validity from September to December 2017,the bi-weight moving mean method is conducted separately to each model to reduce the systematic deviation.Then an integrated 2m temperature correction field is obtained from the three models.The results show that:(1) There is a clear daily change in the forecast of all three models,and the forecast of the EC is optimal.(2) The forecast on both the lowest and highest temperature in the three models are generally of negative systematic bias,especially within the plateau and transitional areas.(3) The accuracy rate of forecasting has been significantly improved,with the root mean square error decreasing by about 1.4-2.5℃,and the average error between ± 0.5℃ in most areas.Generally,the bias correction is more effective in areas with large deviations.(4) The variable-weighted and the equal-weighted integration schemes have similar effects,and both of them are better than the correction field of the three models.