FLOOD FORECAST USING PREDICTION LEARNING OF SOIL WATER INDEX

Abstract

 A highly accurate and practical flood prediction model is required to carry out disaster prevention operations and evacuation actions with a margin in small watersheds with a flood arrival time of less than one hour. In this paper, we propose to build a deep learning model that predicts the water level by using prediction learning of the soil water index. The prediction accuracy was verified at the Nakatsu River water level observatory point (basin area 42.37km², flood arrival time less than 1 hour) in the upstream area of Miyagase Dam. As a result, the predicted water level up to 6 hours ahead could be predicted with high accuracy unless the current state of the soil water index deviates significantly from the predicted value.