Aiming at the uncertainties and low accuracy of current water level prediction models,this paper proposes a probabilistic forecast framework of channel water level by integrating machine learning and Bayesian model averaging method.Taking Jingjiang River section of the upper Yangtze River channel as the research area, random forest (RF),support vector machine (SVM) and artificial neural network (ANN) are used to predict the water level of Shashi station and Xinchang station.The results show that RF has the highest prediction accuracy,followed by SVM,and ANN has a relatively lower prediction accuracy,but all of the three machine learning models provide relatively excellent prediction accuracy.Based on the prediction results of three machine learning models,Bayesian model averaging method is used to forecast the water level ensemble probability.Compared with RF,BMA water level prediction results are further improved,and the possible range of future water level under 90% probability is accurately obtained.The method proposed in this study effectively improves the forecast accuracy of waterway water level and realizes probabilistic forecast,which can provide technical support for ship navigation safety.