A Sensitivity Study of L-Band Synthetic Aperture Radar Measurements to the Internal Variations and Evolving Nature of Oil Slicks

Abstract

This thesis focuses on the use of multi-polarization synthetic aperture radar (SAR) for characterization of marine oil spills. In particular, the potential of detecting internal zones within oil slicks in SAR scenes are investigated by a direct within-slick segmentation scheme, along with a sensitivity study of SAR measurements to the evolving nature of oil slicks. A simple, k-means clustering algorithm, along with a Gaussian Mixture Model are separately applied, giving rise to a comparative study of the internal class structures obtained by both strategies. As no optical imagery is available for verification, the within-slick segmentations are evaluated with respect to the behavior of a set of selected polarimetric features, the prevailing wind conditions and weathering processes. In addition, a fake zone detection scheme is established to help determine if the class structures obtained potentially reflect actual internal variations within the slicks. Further, the evolving nature of oil slicks is studied based on the temporal development of a set of selected geometric region descriptors. Two data sets are available for the investigation presented in this thesis, both captured by a full-polarization L-band airborne SAR system with high spatial- and temporal resolution. The results obtained with respect to the zone detection scheme developed supports the hypothesis of the existence of detectable zones within oil spills in SAR scenes. Additionally, the method established for studying the evolving nature of oil slicks is found convenient for accessing the general behavior of the slicks, and simplifies interpretation.