Bayesian model averaging (BMA) probability forecast models were established through calibration of their parameters using 24-h ensemble forecasts of average daily surface air temperature provided by single-center ensemble prediction systems (EPSs) from the following agencies: the European Centre for Medium-Range Weather Forecasts (ECMWF), the United Kingdom Meteorological Office (UKMO), the China Meteorological Administration (CMA), and the United States National Center for Environmental Prediction (NCEP) and its multi-center model grand-ensemble (GE) EPSs in the THORPEX Interactive Grand Global Ensemble (TIGGE), and observations in the Huaihe basin. The BMA probability forecasts of average daily surface air temperature for different EPSs were assessed by comparison with observations in the Huaihe basin. The results suggest that performance was better in the BMA predictive models than that in raw ensemble forecasts. The BMA predictive models for the four single-center EPSs all had good forecast skills; among them, the ECMWF EPS had the best. The BMA predictive models for the GE EPS performed better than any of the four single-center EPSs; those for the GE EPS with exchangeable members (EGE) quickened the computation rate and had the best forecast skill in BMA models for all EPSs. The mean absolute error (MAE) and continuous ranked probability score (CRPS) skills of the BMA models for EGE improved approximately 7% and 10%, respectively, compared with those of raw ensemble forecasts. On the basis of percentile forecasts from the BMA predictive models for EGE, an extreme scorching weather warning scheme was proposed in the study area, which is of significant importance for precautionary measures against such weather conditions.