Exploring Nonlinear Dynamics of Sea Ice Deformation Using Deep Learning-Based Optical Flow

Recent advances in high-resolution deep learning-based optical flow and radar imaging have opened new opportunities to analyze sea ice deformation at unprecedented spatiotemporal scales. Building on our novel deep neural network-based motion-tracking method we can now provide insights into the fundamental processes driving sea-ice deformation. For the ship radar data from the MOSAiC expedition, our approach resolves deformation events at scales down to 10 meters and 10-minute intervals across a 10 km × 10 km domain, generating on the order of 10^8 deformation-rate estimates per day with accuracy comparable to, or exceeding, that of traditional measurements. By quantifying the presence of nonlinear dynamics in intermediate-scale ice dynamics, our analysis refines established scaling laws and reveals emergent behaviors in deformation processes. Our findings emphasize the importance of seasonality and spatial heterogeneity in determining the mechanical response of Arctic sea ice under changing climate conditions.