The purpose of this study is to apply a non-linear filtering methods (Particle Filter) to a flood prediction system to improve the accuracy of flood water level. The uniqueness of the flood prediction system is to estimates the water level considering temporal change of the sandbar collapse at the backwater reach in the Kumano River. A one dimensional hydrodynamic model of unsteady flow was applied to predict the longitudinal profile of the water level at the reach affected by the sandbar collapse of the river mouth. It was shown that the bed deformation height as one of the state quantities could explain the timing of the sandbar collapse. Other state quantities are discharge at the upstream and the lateral discharge from the branches. These were given from the results of a distributed rainfall-runoff model. Error coefficients are included to update the state quantities by using filtering method with the observed water level. The water level at the objective river of the study was predicted by using the updated initial condition after the filtering, which showed a good agreement with observed water level. It is concluded that the precision of the flood prediction system combined the water surface level prediction model are improved more than before.
Read full abstract