Arctic sea ice cover data from spaceborne synthetic aperture radar by deep learning

Abstract. Widely used sea ice concentration and cover in polar regions are derived mainly from spaceborne microwave radiometer scatterometer data, the typical spatial resolution of these products ranges several to dozens kilometers. Due dramatic changes ice, high-resolution data drawing increasing attention for navigation, environmental research, offshore operations. In this paper, we focused on developing an approach deriving a product Arctic using Sentinel-1 (S1) dual-polarization (horizontal-horizontal, HH, horizontal-vertical, HV) extra wide swath (EW) mode. The discriminating open water by synthetic aperture radar (SAR) is based modified U-Net architecture, deep learning network. By employing integrated stacking model combine multiple classifiers with diverse specializations, segmentation achieved superior accuracy over any individual classifier. We applied proposed 28 000 S1 EW images acquired 2019 obtain high 400 m. validation 96 cases visual interpretation results shows overall 96.10 %. S1-derived was converted then compared Advanced Microwave Scanning Radiometer 2 (AMSR2) showing average absolute difference 5.55 % seasonal fluctuations. A direct comparison Interactive Multisensor Snow Ice Mapping System (IMS) daily achieves 93.98 These show that developed comparable AMSR IMS terms but presenting detailed information, particularly marginal zone (MIZ). Data available at https://doi.org/10.11922/sciencedb.00273 (Wang Li, 2020).