Automatic Mapping of Hydrocarbon Pollution Based on Hyperspectral Imaging

Large-scale mapping of coastal oil spills and their monitoring over time is a major issue that can be adressed by using hyperspectral images and dedicated processing. Previous researches have shown that it is possible to map the polluted coastline caused by the explosion of the Deepwater Horizon (DwH) platform from AVIRIS images (AVIRIS: Airborne Visible/InfraRed Imaging Spectrometer). But the detection processes required either ground truth or laboratory spectra of hydrocarbons or were not fully automatic.In this paper we focused on an AVIRIS image which covers The Bay Jimmy, located south of New Orleans, and particularly impacted by oil pollution. Two automatic methods were developed to detect oiled coasts. In the first one, we have developed a new spectral index able to detect directly hydrocarbon and less sensitive to noise than indices proposed in previous works. The second one extracts endmembers via Orthogonal Subspace Projection, and then sorts the endmembers in terms of hydrocarbon indices scores, in descending order. Then, the detection map or the abundance map corresponding to the best endmember is used to map oiled areas. Both approaches give results consistent with those of studies previously conducted on the same image, and with maps built from field observations.