Abstract
Real-time rainfall-runoff prediction systems using a particle filter are developed. The systems update state variables (storage variables and model parameters) using observed variables and predict future river discharge on a real-time basis under a parallel computing environment. The systems are applied to the Sonohara River basin (492km2) in the upper part of the Tone River basin and three different prediction systems, a model parameter update system, a storage variable update system, and a parameter and storage variable update system are compared in terms of prediction accuracy. Results obtained are as follows: a storage variable update system provides stable prediction results; it is applicable for floods with different magnitudes; the variance of the system noise influences prediction results and the standard deviation as 10 percent of the storage variable gave appropriate prediction results.
