Improving oil slick detection by SAR imagery using ancillary data

The main trouble. of oil spill detection systems based on synthetic aperture radar image s is the discrimination of true oil slicks from other surface phenomena giving a similar signature. Most of these systems consist of three main stages: dark areas detection, features extraction and classification. The aim of this work is to improve the classification performance by using additional data in order to define a more accurate training set and identifying the features with the highest discrimination capability. It was used 27 ENVISAT ASAR images of the Prestige oil spill together with data from other sources and meteorological or oceanographic models. Results show that the radiometric features seem to work better in order to distinguish between oil slicks and look-alikes, and also that it is possible identify as look-alikes using ancillary data up to 10% of the dark areas previously detected