Abstract
It is necessary to support that flood forecast system offers river administrators information such as confidence interval of the predictive water level result and probability to exceed water level, because river administrators make a precise decision from flood forecast. Therefore, flood forecasting system with a distributed runoff model in consideration of an error of the predictive precipitation was developed. Two-variation exponential distribution was adopted as distribution of the errors between observed precipitation and predicted one. The accommodative high particle filter to a non-linear phenomenon was adopted as data assimilation method with observation data and was incorporated in flood forecasting system. As a result of having applied to a flood, the observed water surface level entered the confidence interval of the prediction for 3 hours and it was judged that the system was almost proper.
