Monitoring Land-fast Sea Ice in the Western Antarctic Through Multi-sensor Data Fusion

Land-fast sea ice is almost motionless and fixed to shorelines, contrary to pack ice floating on the sea. As the spatiotemporal distribution of land-fast sea ice is closely related to the global and polar climate systems, it is crucial to accurately monitor land-fast sea ice to better understand the climate systems. Remote sensing can be used to monitor land-fast sea ice as it provides data covering vast areas at high temporal resolution. While remote sensing-based land-fast sea ice monitoring has been investigated in the Arctic areas, very few studies were conducted over the Antarctic areas. In particular, no studies have been conducted in the western Antarctic due to much greater variability of sea ice and more complex topography than the eastern Antarctic.

The purpose of this study was to develop an automated land-fast sea ice monitoring approach using multi-sensor data fusion and machine learning approaches in the entire Antarctic especially focusing on the western part. The basic assumptions of land-fast sea ice with almost 100% of sea ice concentration and almost 0 m/s of sea ice velocity were used. Based on these assumptions, a total of 11 variables including sea ice concentration, 8 dual polarization frequency channels from The Advanced Microwave Scanning Radiometer for EOS (AMSR-E), ice surface temperature from visible/TIR sensor such as the MODerate resolution Imaging Spectroradiometer (MODIS) and ice velocity from Special Sensor Microwave/Imager (SSM/I) were used to identify land-fast sea ice. In addition to the 11 remote sensing-based variables, previous land-fast sea ice results visually identified using MODIS data by Fraser in the eastern Antarctic were used as reference data. Two rule-based machine learning approaches including See5.0 and random forest were used to map land-fast sea ice. Long-term temporal changes of the Antarctic land-fast sea ice distribution were analyzed during the period of 2000 to 2008 at multiple time scales. National Snow & Ice Data Center (NSIDC)-provided 250m MODIS Antarctic ice shelf images and high resolution Synthetic Aperture Radar (SAR) data (Radarsat 2) were used to validate the spatial distribution of the predicted land-fast sea ice.