A real-time river stage forecasting method is developed, which is aimed to apply to river sections where a rating curve does not exist. The method uses a particle filter combined with the dynamic wave model. A number of the dynamic wave models (particles) which have different parameter values and boundary conditions run in parallel; a set of well behaved particles are selected in real time according to the observed river stage; and several hours-ahead river stages are predicted with their prediction error at each observation time step. The method is applied to the Katsura River for the flood of the typhoon No. 23 in 2004. The predicted river stage shows a good agreement with the observed one. The predicted maximum river stages along the study river section also well explain the flood marks.
