Oil Spill Detection in Glint-Contaminated Near-Infrared MODIS Imagery

We present a methodology to detect oil spills using MODIS near-infrared sun glittered radiance imagery. The methodology was developed by using a set of seven MODIS images (training dataset) and validated using four other images (validation dataset). The method is based on the ratio image R = L'GN/LGN, where L'GN is the MODIS-retrieved normalized sun glint radiance image and LGN the same quantity, but obtained from the Cox and Munk isotropic (independent of wind direction) sun glint model. We show that in the R image, while clean water pixel values tend to one, oil spills stand out as anomalies. Moreover, we provide a criterion to distinguish between positive and negative oil-water contrast. A pixel in an R image is classified as a potential oil spill or water via a variable threshold Rs as a function of L'GN, where the threshold values are obtained from the slicks of our training dataset. Two different fitting curves are provided for Rs, according to the contrast sign. The selection of the correct fitting curve is based on the contrast type, resulting from the criterion above. Results indicate that the thresholding is able to isolate the spills and that the spills of the validation dataset are successfully detected. Spurious look-alike features, such as clouds, and other non-spill features, e.g., large areas at the glint region border, are also detected as oil spills and must be eliminated. We believe that our methodology represents a novel and promising, though preliminary, approach towards automatic oil spill detection in optical