Abstract: We can improve the accuracy of real-time flood forecasting models using data assimilation, which integrates hydrological observation data with the flood forecasting model. We have developed a real-time, probabilistic channel flood forecasting model based on a particle filter. It takes the discharge, stage, and roughness coefficient of cross sections along the river as the basic particles of the flow state, and stage observations at hydrological stations as the required observations. We applied the model to a real flood event, downstream from the Yellow River. Our results show that particle filter algorithm effectively corrected the flow state particles. Additionally, we produced more accurate intervals for the flow's initial condition and roughness coefficient, which can be used in future flood forecasting calculations. These will improve the accuracy of the model's predictions, because the probabilistic intervals are more appropriate. Moreover, the forecasts for different lead times indicate that, as the lead time increases, the positive effect of the data assimilation weakens, reducing the accuracy of the forecasts. The uncertainties of the stage prediction increase over time, because different particles have different roughness coefficients. Additionally, the uncertainties of the discharge predictions decrease over time, because of the given deterministic model boundary conditions. The model can successfully assimilate the original historical stage observation data, which shows that it is practical and can be applied to real flood forecasting tasks.