ニューラルネットワークモデルによる二級河川での洪水予測の基礎的検討

抄録

A flood prediction system able to predict the water level during flooding from 1 to 3 hours in advance was studied in order to contribute to flood predictions on second class rivers required by the revised Flood Fighting Act. The applicability of a neural network model as a flood prediction method was also studied. The rainfall used was actual measured rainfall premised on using short term rain predictions from the Meteorological Agency. The suitability of the system was evaluated with priority on prediction precision of the arrival time of the water level that is the criterion. It guaranteed at least three hours from the predicted time until the actual time that the river reaches the critical water level at floods equal to 100% of the warning water level and equal to 60% of the call out water level. Using it in conjunction with a non-linear discharge calculation model guaranteed 80% at the call out water level.