A study of a real-time storm surge forecast system using a neural network at the Sanin Coast, Japan

This study investigates the sensitivity of real-time storm surge forecasts to local measured data in an artificial neural network along the Sanin Coast, Japan. The forecast experiments were conducted by combining the various components for the input data; the sea surface level, the sea level pressure, its depression rate, the wind speed, the wind direction and the typhoon position, increasing the numbers of measurement stations and typhoon events. It is found that the increasing numbers of components, stations and events result in the accurate storm surge prediction in the neural network. In addition, it appears that the predicted storm surges are closer to the observation in the neural network trained with the forecast lead times of 1h and 2h than.