A Study on the Prediction Algorithm of Sea Fog Dissipation based on Machine Learning

JinHyun Han1, Hyunseok Joo1, Kuk Jin Kim1, Young-Taeg Kim2 and Seok Jae Kwon2, (1)Underwater Survey Technology 21, Inc., Incheon, South Korea, (2)Korea Hydrographic and Oceanographic Agency, Busan, South Korea
Recently, we have experienced severe natural disasters owing to the climate change which are difficult to predict, and also experienced unexpected compound disasters due to the complexity and diversity of the social structure. For example, since navigation safety of almost all ports in the world is vulnerability to sea-fog, there have been many researches on how to predict sea fog occurrence and its dissipation.

Since 2016, KHOA (Korea Hydrographic and Oceanographic Agency) has managed and performed challengeable researches to predict the occurrence of sea fog using Deep Neural Network (DNN), and found its feasibility and applicability to real operation of trading ports in the Republic of Korea. Currently, DNN-based sea fog occurrence prediction is in operation experimentally at six ports. However, the prediction of sea fog dissipation is much more difficult due to not only complexity of sea fog dissipation compared with its occurrence, but also each dissipative condition of various sea fog types.

To predict hourly dissipation of sea fog using DNN, we constructed training data sets of air temperature, sea water temperature, atmospheric pressure, air-sea temperature difference, humidity, wind speed and direction, and visibility, which are obtained only after the sea fog dissipation. We investigated the relative contribution of each data to sea fog dissipation by means of calculating the feature importance of input data to DNN algorithm.

* Corresponding Author : Research Scientist, Department of Oceanographic Forecast, KHOA (Korea Hydrographic and Oceanographic Agency), 351, Haeyang-ro, Yeongdo-gu, Busan, Republic of Korea, zip code 49111, E-mail : kyt5824@korea.kr