抄録
This paper describes a problem and a method to solve a problem on the particle filer which is applied for the water level prediction. A water level prediction model at the Yodo River has been constructed using the particle filter with the state variables such as the upstream and lateral discharge. We found the prediction model needs improvement from the results of the water level prediction by two floods, Typhoon No. 18 (2013) and Typhoon No. 11 (2014). Not only the upstream and lateral discharges, we newly added the Manning's coefficient to state variables, and then we found the improvement of the accuracy of the estimation of the upstream and lateral discharge. Moreover, we found that an increase of the number of state variables does not assure the improvement of the accuracy of the estimation of the boundary discharges even if the number of particle is increased. In a real-time water level prediction, it is important to consider the balance of the setting of state variable and the number of particle.
