Automated Weddell Sea iceberg tracking with a machine learning approach applied to SAR imagery

Mauro Barbat1, Thomas Rackow2, Christine Wesche2, Hartmut H Hellmer3 and Mauricio M Mata4, (1)Federal University of Rio Grande-FURG, Institute of Oceanography, Rio Grande, Brazil, (2)Alfred Wegener Institute Helmholtz-Center for Polar and Marine Research Bremerhaven, Bremerhaven, Germany, (3)Alfred Wegener Institute, Bremerhaven, Germany, (4)Federal University of Rio Grande, Instituto de Oceanografia, Rio Grande, Brazil
Icebergs are continental ice fragments ranging in size from few meters to kilometers that may drift far away from their region of origin. Mainly driven by ocean and atmospheric dynamics, drifting icebergs represent a significant hazard for polar navigation and are able to influence the ocean environment around them. Freshwater flux from melting icebergs can locally decrease salinity and temperature, thus affecting ocean circulation, biological activity, sea ice and ---on larger scales--- the climate system. However, despite their potential impact, the large-scale monitoring of drifting icebergs in sea ice-covered regions is often restricted to giant icebergs due to difficulties in accurately identifying and following the motion of much smaller features in polar ocean regions. So far, tracking of smaller icebergs from satellite imagery thus has been limited to open-ocean regions not affected by sea ice. In this study, an automated iceberg motion tracking method based on a machine learning-approach for automatic iceberg detection was applied to 7,146 Advanced Synthetic Aperture Radar (ASAR) images acquired between 2002-2012 for the entire Weddell Sea, Antarctica. Overall, 415 individual icebergs with surface area between 3.4 and 3,612 km2 were tracked and the main drifting patterns of icebergs of various different sizes in this region could be identified automatically. The majority of the tracked icebergs drifted between 1.02 and 2,679.2 km westward around the continent at an average drift speed of 3.1 ± 4.4 km day-1, following the Antarctic Coastal Current and the Weddell Gyre. It was also possible to estimate an average disintegration rate of 25.9 ± 18.8% for icebergs with drifting times longer than 30 days. Our machine-learning approach is a robust alternative for detection and tracking of highly dynamic objects such as icebergs, even under ambiguous ASAR background signatures. It allows a monitoring of icebergs even in the challenging near-coastal environment, where the presence of sea ice and coastal ocean dynamics usually pose major obstacles for other approaches.