The detection of Arctic sea ice linear kinematic features using LadderNet

In polar regions, sea ice linear kinematic features (LKFs) play a critical role in the exchange of mass and energy between the ocean and atmosphere. These features also serve as an important reference for navigation decision, highlighting the growing need to accurately monitor and simulate their changes. An identification and labeling method using artificial intelligence (AI) to detect LKFs based on Synthetic Aperture Radar (SAR) data is proposed in this study. This approach uses sea ice deformation data derived from sea ice drift in the SAR observations and employs a specialized encoder–decoder convolutional neural network, known as LadderNet, to segment these fine-grained LKFs. A post-processing algorithm utilizing connected region detection further assigns markers for individual LKFs. Results show that our detection method has a higher accuracy with F1 Scores ranging between 0.6 and 0.7 than that using UNET architecture. Training the AI model with seasonal data effects the detection results slightly. Compared to the classical algorithm, our study also demonstrates more consistent detection results for both numerical model and observations regardless of practical parameters after training, which provides a standardized metric for inter-comparisons between models and observations.