Three Years of Near‐Coastal Antarctic Iceberg Distribution From a Machine Learning Approach Applied to SAR Imagery. Issue 9 (10th September 2019)

• Record Type: Journal Article
• Authors: Barbat, Mauro M.; Rackow, Thomas; Hellmer, Hartmut H.; Wesche, Christine; Mata, Mauricio M.
• Is Part Of: Journal of geophysical research. Volume 124:Issue 9(2019)
• Page Start: 6658
• Page End: 6672
• Publication Date: 2019-09-10
• Subjects: icebergs -- SAR imagery -- remote sensing -- machine learning -- Southern Ocean; Oceanography -- Periodicals; 551.4605
• DOI: 10.1029/2019JC015205
• Languages: English
• ISSNs: 2169-9275

Abstract

Abstract: Mass loss around the Antarctic Ice Sheet is driven by basal melting and iceberg calving, which constitute the two dominant paths of freshwater flux into the Southern Ocean. Although of similar magnitude, icebergs play an important and still not fully understood role in the balance of heat and freshwater around Antarctica. This lack of understanding is partly due to operational difficulties in large‐scale monitoring in polar regions, despite observational and remote sensing efforts. In this study, a novel machine learning approach, augmented by visual inspection, was applied to three Synthetic Aperture Radar (SAR) mosaics of the whole Antarctic continent and its adjacent coastal zone. Although originally intended for a mapping of the Antarctic continent, the SAR mosaics allow us to document the evolution and distribution of the size (and mass) of icebergs in the pan‐Antarctic near‐coastal zone for the years 1997, 2000, and 2008. Our novel algorithm identified 7, 649 icebergs in 1997, 13, 712 icebergs in 2000, and 7, 246 icebergs in 2008 with surface areas between 0.1 and 4, 567.82 km 2 and total masses of 4, 641.53, 6, 862.81, and 5, 263.69 Gt, respectively. Large regional variability was observed, although a zonal pattern distribution is present. This has implications for future climate modeling studies that try to estimate the freshwater flux from melting icebergs, which demands a realistic representation of the interannually varying near‐coastal iceberg pattern to initialize the simulations. Plain Language Summary: When icebergs melt in the Southern Ocean, they cool the surrounding ocean. They also distribute freshwater, which potentially impacts the circulation, biological activity, sea‐ice cover, and the formation of the densest waters of the world's oceans. However, all these influences are not fully understood because we are lacking reliable methods to detect icebergs from space via satellites. This study has the main objective to determine how iceberg mass is distributed in the coastal zone around the Antarctic continent and how this distribution changes between individual years. We show that a novel machine learning approach can be applied to this problem, which is capable to identify icebergs in satellite images of the near‐coastal ocean region almost automatically. The method also works in severe conditions, e.g. when the icebergs are affected by ocean waves or when they are surrounded by sea ice. Our results complement the ongoing discussion about the distribution of Antarctic icebergs in open‐ocean regions that are not affected by sea ice. The resulting data can also be used in computer models that simulate the input of iceberg freshwater into the ocean. Key Points: A novel automatic machine learning approach is applied to SAR imagery, with minimal visual inspection Near‐coastal iceberg size distributions are presented for 1997, 2000, and 2008 Mass distribution for smaller icebergs appears to be relatively stable, while less frequent large icebergs strongly bias the distribution