TWO-DIMENSIONAL SIGNIFICANT WAVE FORECAST BY USING CONVOLUTIONAL LSTM AND ERA5 DATA

Wave forecasting has been used in various situations, such as feasibility decision of construction works. In addition, the demand for short-term wave forecasting for several hours to several days lead time is expected to increase under the situation where development of offshore wind farms in Japan. In addition to the operational global ocean surface wave models, machine learning has been employed for wave forecasting as well. This is mainly because of lower computational cost compared with the numerical model. However, wave prediction using deep learning is mostly based on fixed-point prediction, and there are few researches regarding two dimensional field of surface wave forecasting. In this study, we attempted to predict the 2D wave field using convolutional LSTM with training dataset based on the 5th generation ECMWF atmospheric reanalysis (ERA5). The evaluation of predicted significant wave height was carried out on semi-closed Japan sea area.