Water resources forecasting methods play crucial roles in long-term water resources management, while they have always been affected by unavoidable uncertainties from model structure, parameters, and input data sets. This study integrates 12 individual forecasting methods, such as kNN (k-nearest neighbor) and AR (Auto Regressive), based on the Bayesian Model Averaging approach to construct a probabilistic forecasting model BPF (Bayesian Probability Forecasting). After that, the numerical experimental simulation is issued in the Huai River Basin using data sets from 1997 to 2022. Through statistical and numerical analysis of the forecasting results, it is considered that: 1) there are significant differences among the forecasting results from the above 12 deterministic methods, and no single one can consistently perform more accurately than the others; 2) the mean forecasting results from the BPF model are always close to the most accurate one of the 12, and the BPF is considered to be stable and reliable; 3) the probabilistic forecasting results of the BPF perform with high coverage ratio (CR) and low bias width (BW) in both the calibration period and verification period, and thus considered to be able to describe the posterior probability distribution of water resources reliably.