Abstract: The Nares Strait is one of the most important passages for sea ice outflow from the Arctic Ocean. It plays a crucial role in the formation of deep water in the Baffin Bay and Labrador Sea and in the Arctic sea ice mass balance. Because of the narrow entrance and low sea ice concentration, it is difficult to retrieve valid sea ice motion information from satellite imagery using existing sea ice motion tracking algorithms. Therefore, there are limited sea ice motion data for this region. In this study, a sea ice motion retrieval algorithm based on ice floe shape features is proposed. It consists of three main steps: image segmentation, ice floe extraction and ice floe matching. Sea ice motion data in the Nares Strait from 20 February to 27 February 2019 were retrieved using the proposed algorithm, and were compared with manually produced validation data to evaluate the accuracy of the ice floe extraction and matching steps. Results show that the mean error of the ice floe shape features ranges from 4.3% to 8.48%, the overall accuracy of the ice floe motion vectors retrieved by the algorithm is 84%, and the root mean square error of the matched centroid and velocity is 0.171 km and 6.313 cm·s^–1, respectively. Compared with existing feature tracking algorithms, including SIFT, ORB, and AKAZE, sea ice motion coverage and accuracy derived from the proposed algorithm are higher by 27.7% and 39%, respectively. The proposed algorithm is not only an effective way to retrieve sea ice motion data for the Nares Strait, but provides some reference for sea ice motion retrieval in other regions (e.g., marginal ice zone).

Key words: Nares Strait, sea ice motion, Sentinel-1, image segmentation, ice floe matching