Validation of Sentinel-1-derived sea ice cover data in the Arctic

The Sentinel-1 synthetic aperture radar (SAR) instrument can image the polar regions with a large area and high temporal and spatial resolutions, which is particularly suitable for sea ice monitoring in the Arctic. In a previous study, a support vector machine (SVM)-based method was developed to automatically extract the sea ice cover from Sentinel-1 SAR Extra Wide (EW) swath mode data in cross-polarization (horizontal-vertical, HV, or vertical-horizontal, VH). For validation, 468 Sentinel-1 EW HV-polarized images acquired in the Arctic are used to derive the sea ice cover with a spatial resolution of 0.96 km. We then compared the SAR-derived sea ice cover with the Ice Mapping System (IMS) dataset, which has a resolution of 1.0 km, and the AMSR-2 sea ice concentration (SIC, 15% is used as a threshold for deriving sea ice cover) dataset, which has a resolution of 3.125 km, based on pixel-by-pixel matching. The accuracies of the comparisons with the IMS and SIC data are 85.01% and 88.69%, respectively. The comparison shows that the IMS dataset is mostly overestimates the sea ice cover, while the AMSR2 dataset is relatively underestimates the sea ice extent. On the other hand, the sea ice cover information derived from the Sentinel-1 images by using the proposed method well characterizes the details of the marginal ice zone (MIZ). Furthermore, through the analysis of individual cases, the main factors that affect accuracy are drift (or pack) ice and a high land proportion in some areas.