ニューラルネットワークによる阿武隈川洪水予測の基礎的検討

抄録

Because flood predictions based on discharge analysis or water level correlation cannot necessarily reflect the discharge properties that vary for each flood, the prediction of the time of arrival of the critical water level that is important for flood prediction and flood fighting activities is not adequately precise. The purpose of this study is to use a neural network model based on measured values instead of using predicted rainfall for large rivers to construct a system that can obtain prediction data that corresponds to flood predictions performed by experienced river managers and to prepare simple water level prediction maps that anybody can use even though they are not very precise. When this method was verified using 23 floods, it was possible to predict the arrival time of the warning water level of 15 floods that exceeded this water level up to three hours in advance. And simple prediction maps based on this model are sure to be used for a new form of provision of disaster information in the future.