Abstract:The multi-model integration test of the Bayesian model averaging (BMA) method is carried out for the forecast after correcting the errors of 2 m temperature, 10 m wind speed, and 2 m relative humidity from 1 December 2020 to 15 March 2021 in the Beijing-Tianjin-Hebei Region based on the four models （CMA-GFS, CMA-REPS, CMA-MESO 〖KG-*5〗3 km, and CMA-MESO 〖KG-*5〗1 km）. The results show that the root-mean-square error of each model’s element is significantly reduced after error calibration. The prediction effect of the BMA multi-model blending is much better than that of calibrated output of every participant model. Compared with the original errors of several models, the improvement of the 2 m temperature integration forecast is between 0.5-1.4℃, and the improvement rate of the root-mean-square error is about 20%-40%. In the meantime, the root-mean-square error of 10 m wind speed and 2 m relative humidity improved by 12%-45% and 25%-35%, respectively. The horizontal root mean square error distribution of each element is significantly different at different terrain heights, and the error distribution of different elements has been significantly reduced throughout the region. In addition, BMA can obtain the full probability density function, which can quantitatively predict the uncertainty of each element.