Development of a Deep Learning Model for Predicting the River Water Level Based on the Rainfall Distribution Data

摘要

Physical models are available for deductively predicting the river water levels. However, these models require many types of input data to set the boundary conditions and the model parameters. In this study, we developed a deep learning model comprising a convolutional neural network (CNN) and recurrent neural network (RNN) for predicting the water level based on the rainfall distribution data. It is expected that the water level can be predicted using only the rainfall data as the input data because CNN and RNN can extract the spatial and temporal features inherent in the input data, respectively. The developed model was applied to study the Katsura river basin. The response of the water level with respect to rainfall was accurately reproduced. However, with respect to the reproduction accuracy, problems, such as the occurrence of an unstable output in case of the ordinary water level and underestimation of the peak water level, can be observed. The accuracy decreased when the spatial information was not considered in the input. Furthermore, the accuracy decreased when RNN was not used because the rainfall history was not considered. The study results demonstrate the necessity of using both CNN and RNN for predicting the river water level.