Based on the chaotic characteristics of the atmosphere,it has become a tendency that the deterministic forecast gradually turns to be the probabilistic forecast with uncertainty.This paper systematically reviews the development background of probabilistic weather forecast,and introduces the concept,and advances of the probabilistic forecast research at home and abroad.Reviews are primarily focused on two kinds of multimodel ensemble probabilistic forecast methods,namely the Bayesian model averaging(BMA) and the multivariate Gaussian ensemble kernel dressing(GEKD).Cases of the BMA and GEKD probabilistic forecasts are shown.The deviation of the BMA probabilistic forecast becomes smaller,and it reduces the uncertainty of the forecast.Hence,the forecast is close to the real value of the atmosphere.As another multimodel ensemble probabilistic forecast approach,the GEKD probabilistic hindcast of the mean value of surface temperature anomaly and its trend basically coincides with the observed data.By using the GEKD approach,a multimodel ensemble probabilistic prediction model is established to predict the interdecadal variability of surface air temperature and to extract interdecadal climate variation features.It is of important application value for the interdecadal variability prediction over the East Asian monsoon region.