ACCURACY IMPROVEMENT OF DEEP ARTIFICIAL NEURAL NETWORK RIVER STAGE PREDICTION USING MULTIPOINT OBSERVATION DATA

Abstract

In flood prediction, reduction of the uncertainty is one of the biggest issues. As a prediction model, artificial neural network (ANN) is one of the potent methods. To improve the accuracy and lead-time of the river-stage prediction model, it is desirable to take advantage of as many observation data as possible. However, due to the limitation of the learning ability of ANN model, it is undesirable to use too many input data into the model. As a new learning method of ANN, deep learning is said to have an excellent ability of learning, and able to handle huge input data. In this study, we developed to models, 4 layered deep learning based ANN and 3 layered conventional ANN. We applied 2 models to the 3 rivers in Japan, Ooyodo River, Onga River and Kokai River. Compared to the conventional ANN, deep learning based ANN model could reflect more observational information and could reproduce the past flood well.