SPATIAL AND TEMPORAL DEEP LEARNING OF WEATHER FORECASTFOR 11 DAY WAVE PREDICTION

In marine an port construction, wave forecasting one day to one week ahead is indispensable to determine whether construction is feasible, to manage the process, and to ensure safety. In recent year, many studies on wave prediction have been conducted using deep learning, showing the potential for practical, accurate, and low-cost wave prediction models. In previous studies, spatial and temporal learning of weather fields have been found to be effective, respectively, but no study has yet taken both into account. In this study, we proposed a model that can learn both in space and in time and clarified its effectiveness. Furthermore, we attempted to forecast wave in the coastal areas of Japan for 11 days by using the forecast data from the GPV global numerical prediction model of the Japan Meteorological Agency (JMA).