DEVELOPMENT OF RIVER WATER TEMPERATURE PREDICTION MODEL AT THE LOWEST REACH BY RECURRENT NEURAL NETWORK

Abstract

 Prediction of discharge, water temperature and water quality in a river is necessary in evaluation of climate change impact in a coastal region. In this paper, we attempted to develop an evaluation model of hourly river water temperature by Artificial Neural Network (ANN) as an Artificial Intelligence (AI) technique. Since river water temperature is time-series data, the Recurrent Neural Network (RNN) with Long Short-Term Memory (LSTM) is adapted. As a result, a highly accurate prediction was shown using approximate one-year dataset as a learning data. Future prediction of river water temperature under the present and the future projected climate conditions was conducted in the Yabe River by combining with d4PDF and hydrological model. From the result, utilization of the developed model was clarified.