Neighborhood-aware retrieval of a spatial floe fragmentation index for Arctic sea ice using Sentinel-1 SAR and AMSR2

Arctic sea ice fragmentation involves changes in floe area composition, spatial separation, and geometric complexity, which affect lateral melt rates and ice–ocean heat fluxes. Existing metrics primarily characterize floe size distributions or area fractions, limiting their capacity to resolve variations in the spatial arrangement of floes. We propose a spatial floe fragmentation index (SFFI) that jointly quantifies floe area composition, spatial separation, and geometric complexity. Using 416 Sentinel-1 SAR scenes spanning diverse Arctic regions and ice conditions, we extract floe contours to generate 52,074 SFFI reference labels. For samples with comparable floe area composition, higher SFFI values correspond to greater inter-floe spacing and lower circularity, demonstrating that the index captures spatial and geometric differences not represented by area composition alone. To enable daily Arctic monitoring, we develop a neighborhood-aware retrieval framework integrating AMSR2 radiometry, sea ice concentration (SIC), and neighborhood statistics, and evaluate its performance using random forest, XGBoost, and multilayer perceptron models. Under the 2016–2024 SAR-scene split, XGBoost with the complete feature set achieves the best performance, with an R2 of 0.6877, an RMSE of 0.3454, and an MAE of 0.2536. Neighborhood-aware features provide relative R2 gains of 4.6% to 10.0% over the SIC-inclusive baselines. The selected model maintained robust predictive performance on the independent 2025 dataset, supporting its temporal transferability. Using the best-performing model, we generate a daily Arctic SFFI product on a 12.5 km grid for 2016–2025. Results show that SFFI captures a pronounced seasonal cycle, with high-SFFI areas expanding during the melt season and contracting during refreezing, while revealing regional differences among Arctic marginal seas.