Long term runoff forecasting is an important field of hydrological forecasting, which is of great significance to the comprehensive management of watershed water resources. The runoff of five hydrological stations in the upper reaches of the Yangtze River in the main flood season (June to August) from 1954 to 2020 are taken as the reseach object. This research is based on the theory of climate teleconnection affecting regional runoff. 10 climate factors are employed as prediction factors for runoff probability prediction based on hierarchical Bayesian model. The results show that the runoff in the upper reaches of the Yangtze River in the main flood season is obviously affected by a variety of large-scale climate factors. North American subtropical high ridge position index et. al. are selected as the model prediction factors. Taking the log normal distribution as the prior distribution of the prediction target, five Markov chains are established, and the parameter posterior distribution is deduced by random sampling in the probability space through MCMC algorithm. The uncertainty interval of the probability prediction results has a high coverage of the measured values. The correlation analysis of prediction results, ROC curve and CRPSS shows that the model effectively captures the information of large-scale climate factors and is suitable for the long-term prediction of runoff in the upper reaches of the Yangtze River.