河川特性を反映したニューラルネットワーク洪水予測の精度向上

抄録

Flood forecast using Neural Network model has operated in Natori River, downstream of Abukuma River since 2003. In this study, we systematically grasped the causes of the error in the current flood forecast based on the accuracy verification using the prior forecasting results. Moreover, we improved and reflected river characteristics in Neural Network model considering the required model accuracy for the practical use.

As a result of accuracy verification of prior forty nine (49) flood forecasts, the practical performance of flood forecast using Neural Network for large river was confirmed. Besides the reflecting the river characteristics, switching scheme of multiple flood forecast models was utilized considering the difference of scale of the flood and its hydrograph. More than three (3) hours’ lead time was ensured for the every flood which exceeded the warning water level utilizing the improved flood forecast model.