Abstract
A real-time flood forecasting system was developed for runoff forecasting at several water gauging points. The system is based on a flood runoff model composed of the upstream part and the downstream part models which are respectively simulating runoff for the upstream and downstream parts of the objective basin. The upstream part model is a lumped rainfall-runoff model, and the downstream part model consists of lumped rainfall-runoff models for hillslopes adjacent to a river channel and a kinematic flow routing model for the river channel. The system is updated by Particle filtering of the downstream part model as well as by the extended Kalman filtering of the upstream part model. The Particle filtering is a simple and powerful updating algorithm for non-linear and non-gaussian system, so that it can be easily applied to the downstream part model without complicated linearization. The system applied to the Yoshii River Basin located in Okayama Prefecture, and flood forecasting accuracy of the developed system was examined. The comparison between the forecasting accuracy of the system with both Particle filtering and extended Kalman filtering and that with only extended Kalman filtering shows that Particle filtering of the downstream part model well improves forecasting accuracy particularly for the basin having relatively large area ratio of the downstream part.
