A Multi-Scale Flux Density Model for Sea Ice Lead Extraction from Optical Satellite Images

The monitoring of sea ice leads is crucial for researching atmospheric circulation and climate change. Remote sensing (RS) technology plays a vital role in monitoring sea ice leads. However, RS data are often limited by the weak contrast between leads and surrounding sea ice, as well as noise under adverse weather conditions. Although high resolution satellite data can better extract lead structures, their limited swath and orbit coverage prevent full and continuous observation of the Arctic region. In contrast, medium-low resolution sensors such as MODIS and Sentinel-3 offer wide coverage and high revisit frequency, making them particularly suitable for large-scale and rapid lead extraction. Unsupervised methods, including threshold-based techniques, remain widely used but are highly sensitive to noise, illumination variability, and rapidly changing cloud conditions. In contrast, supervised methods, like machine-learning and deep-learning methods, require large amounts of labeled samples and have limitations in model interpretability and robustness. Existing methods still struggle to reliably detect weak, narrow, and rapidly evolving leads, especially in low-contrast and medium-low resolution imagery. Sea ice leads typically show sharp local intensity transitions relative to the surrounding ice, whereas the sea ice surface contains smoother and more homogeneous gradient patterns. To exploit this characteristic, this study proposes a multi-scale flux density (MFD) framework for extracting sea-ice leads from optical satellite imagery. The approach employs a multi-scale circular-ring flux-density operator to calculate the gradient flux across concentric annuli, capturing the distinctive flux-density feature associated with lead structures. Lead regions generate dominant outward gradient flux due to strong radiometric contrasts, while adjacent ice areas exhibit inward dominated flux responses, making the lead-ice separation physically interpretable. By analyzing the scale-dependent evolution of MFD responses, the method reliably distinguishes sea-ice leads from the background ice surface based on flux-density differences. Furthermore, multi-scale max pooling and differential flux responses are jointly exploited to construct a scale-adaptive contrast metric for capturing leads with varying widths. Finally, minimal thresholding and morphological operations are applied to produce clean and topologically continuous lead structures.Experiments conducted on Sentinel-3 and MODIS data over the Arctic demonstrate that the proposed method successfully extracts leads with varying widths and contrast levels, particularly under weak contrast conditions. The MFD framework directly characterizes gradient flux patterns induced by radiometric contrasts between leads and surrounding sea ice, providing a physically interpretable mechanism for lead detection. It achieves superior structural continuity, enhanced noise robustness, and improved sensitivity to low-contrast leads, even in scenes with uneven illumination. Overall, the framework is computationally efficient, physically interpretable, and well suited for large-scale operational sea-ice monitoring, polar navigation support, and climate-related analysis.