Oil and hazardous and noxious substance (HNS) spills in marine environments cause severe ecological and socio-economic damage. Rapid monitoring is therefore essential, yet direct access to spill sites is often difficult due to the toxicity and dangero...
Oil and hazardous and noxious substance (HNS) spills in marine environments cause severe ecological and socio-economic damage. Rapid monitoring is therefore essential, yet direct access to spill sites is often difficult due to the toxicity and dangerous nature of oil and HNS materials. Satellite remote sensing offers an effective solution for wide-area, rapid monitoring, and optical satellites in particular provide the advantage of spectrally distinguishing pollutants through spectral analysis. In this study, Sentinel-2 optical satellite imagery was used to detect, classify, and assess the environmental impacts of four real spill events: the diesel spill in Jeju (Republic of Korea), the BTEX and petroleum spills in Houston (USA), the sunflower oil spill in Mykolaiv (Ukraine), and the ferric chloride spill at the Lavéra refinery (France). A refined preprocessing workflow—comprising land masking, cloud removal, and ship removal—was established, after which N-FINDR was applied to extract endmembers and separate pollutants based on their spectral characteristics to derive spill extents. Furthermore, the N-FINDR detection results were used as training data to train three AI algorithms—CNN, SVM, and Random Forest—and then applied to satellite images to detect pollutants. The analysis results showed that N-FINDR effectively distinguished the spectral characteristics of the spilled pollutants and surrounding substances such as seawater, and successfully detected the pollutants. Among the AI algorithms based on the N-FINDR algorithm detection results, Random Forest showed the best performance even under limited training data conditions. This is because Random Forest can effectively classify data with a small training data set because it learns through multiple decision trees. In the case of the Jeju Island oil spill, a satellite-based high-resolution bathymetric map was generated using the Stumpf model, which confirmed the presence of a small seamount that was not visible in the existing bathymetric map. These small coastal geographic features were found to significantly influence the dispersion of spilled oil. Furthermore, Normalized Difference Chlorophyll Index(NDCI) was applied to satellite imagery to monitor changes in chlorophyll concentration after a pollutant spill, and the Particulate Organic Carbon Index(POCI) was used to observe changes in organic carbon concentration. The analysis results confirmed that chlorophyll and organic carbon concentrations changed significantly after a pollutant spill depending on the type of pollutant, demonstrating that pollutant spills clearly impact the ecosystem. This study observed coastal pollutant spills of diesel, BTEX, sunflower oil, and ferric chloride using Sentinel-2 satellite imagery. To the authors' knowledge, this is the first time an actual HNS spill has been monitored and analyzed using a high-resolution optical satellite. The spectral characteristics of the pollutants were analyzed using the N-FINDR spectral mixture analysis technique. Additionally, clustering algorithms such as Kmeans and FCM, as well as artificial intelligence algorithms such as CNN, SVM, and Random Forest, were utilized to detect the spilled pollutants and cross-validate the detection results with those of NFINDR. This demonstrates the broad applicability of pollutant monitoring techniques using high-resolution optical satellites to various pollutants and spill environments.