Detection of Mussel Beds Using Airborne Polarimetric SAR Data

Abstract

As an important source of food and protective habitat for other species living in the eulitoral, mussel beds have an important impact on the coastal ecosystem. To realize a frequent monitoring, we proposes the use of airborne polarimetric Synthetic Aperture Radar (SAR) data to meet the challenge of limited time windows for data acquisition. For distinguishing mussel beds from surrounding tidal flats, the suitability of six different polarimetric features, based on the Freeman Durden decomposition, the Kennaugh element framework and target depolarization effects, are analyzed. Based on the most promising features, pixel-based classification is performed using an encoder-decoder Convolutional Neural Network (CNN). The resulting regions identified as mussel covered are compared to verified reference data, generated three and five years before. Detected changes are additionally compared by visual alignment with current optical aerial imagery. The study demonstrates the potential of airborne polarimetric SAR data for the generation of up-to-date maps of mussel beds.