STUDY ON THE APPLICABILITY OF FLOOD PREDICTING METHOD TO SUBSTANTIAL RIVERS ON SPRING SNOWMENT PERIOD USING METEOROLOGICAL DISTRIBUTION IMAGES

Abstract

 In order to evacuate industrial machines and construction materials in times of flooding at river work sites, it is necessary to perceive the water level increase more than half a day beforehand. In this paper, the authors created a method that predicts the water level up to 1 day ahead by deep learning using meteorological distribution images. The maximum water level of a day at Tsukigata Observatory along the Ishikari River from 2008 to 2020 was predicted. Of the 26 actual flood events, the method was able to predict 17 events 1 day ahead. The method was used to predict the water level of three additional rivers and was able to forecast flood events one day ahead.The technique is be effective for safety management of the river construction.