Geostationary Ocean Color Imager (GOCI) Marine Fog Detection Algorithm based on Machine Learning Method

Marine fog is a worldwide phenomenon occurring at both coastal and open ocean areas. It can disrupt transportation, shipping and aviation due to reduced visibility lower than 1 km. Also, it can form, adevect and dissipate in a few hour so (near)real-time monitoring is necessary to lessen social and economic damages. Satellite marine fog detection technique is very important because in-situ measurements of visibility are limited over oceans. Purpose of this study is to develop marine fog algorithm for monitoring marine fog around Korean Peninsula with high spatial resolution. GOCI provide high spatial resolution about 500m and GOCI-II will provide 250m resolution, so the performance of marine fog detection will be much improved. In this study, GOCI visible data is applied for the marine fog detection algorithm based on decision tree method. Hourly-available and high-resolution visible channel measurement have advantages in monitoring marine fog along complicated coastal area of Korean Peninsula. In addition to GOCI, Himawari-8 cloud top height data was used as a supplementary data since visible channel cannot provide cloud height information. Decision tree technique is applied to find the relationship between input indices and the existence of the marine fog and to decide proper thresholds. Our validation results show that if there 's no cloud above the fog, marine fog is properly detected with our algorithm. With cloud pixels, hit rate and false alarm rate are 0.61 and 0.21, respectively but when we excluded the cloud pixels, hit rate became 1 within our validation samples.