Automatic Sargassum detection using anomaly detection technique: A case study for western coastal area of Korea with GOCI data

Abstract

Sargassum has disruptive impacts on both fisheries and tourism, giving rise to fish kills and beach fouling in the coastal and near-shore areas. This harmful substance increased dramatically in the west coast of Korea, particularly in 2015, and was detected by Geostationary Ocean Color Imager (GOCI) during the period between January and June. Since GOCI data are acquired every hour during the daytime, allowing near real time ocean monitoring, it can be effectively used to observe the temporal variations of the Sargassum distribution. The reflectance of a Sargassum patch floating on the oceanic water is higher than that of the background water in the Near-infrared (NIR) bands, which allows the Sargassum patches to be detected by anomaly detection techniques. In this study, we used Reed-Xiaoli Anomaly Detection (RXD) algorithm, which is one of the most widely used anomaly detection algorithms, for the automatic detection of Sargassum in GOCI data. The several GOCI data in May 2015 were used for the sturdy which showed the widespread spatial distribution of Sargassum in the ocean areas of various optical characteristics. A preliminary result showed that the straightforward application of the RXD algorithm to the whole ocean extent in the images caused the problem that the sensitivity of the result varied significantly by the background reflectance of the adjacent ocean waters. To solve the problem, we employed a 3-step approach that consists of (1) clustering of the ocean area into homogeneous regions, (2) applying the RXD algorithm to each homogeneous region, and (3) distinguishing Sargassum from all the other anomalies base on surface spectral reflectance.