Abstract
This study proposes a data assimilation method for a distributed rainfall-runoff prediction system. The system is composed by a rainfall-runoff model and a river routing model. Since it is computationally inefficient for updating all the model variables on the real-time basis, the proposed filtering method takes river discharges, which are simulated by Muskingum-Cunge method, as the state variables. It also sequentially estimates and collects prediction biases induced by the rainfall-runoff model. The application to the Katsura river basin shows that the filtering with bias estimation and correction improves the accuracy of flood predictions.
