STUDY ON IMPROVING THE ACCURACY OF WAVE ESTIMATION USING WEATHER INFORMATION AND COASTAL IMAGES BY DEEP LEARNING

Abstract

 It is important to understand wave information in coastal management. In this study, a wave estimation model with high accuracy was constructed at the heavily eroding Shichiri-Mihama-Ida beach by applying deep learning to orthomosaic images and meteorological information. It is shown that the third-generation wave estimation model SWAN can estimate the waves at the NOWPHAS observation point off the coast of Mie-Owase, and that a CNN can be applied to both the SWAN estimation results and coastal images to obtain a good estimation of significant wave heights. The results also showed that applying the CNN to both the waves estimated by applying LSTM to the wind speed data and the coastal image improved the accuracy of the significant wave height and wave direction and gave the good estimation of the significant wave period compared to the results obtained by applying the CNN only to the coastal image.