Kaihong Jiao: Grounded icebergs around Antarctica: A high-resolution dataset derived from deep learning and Sentinel-1 synthetic aperture radar
Icebergs frequently ground on Antarctic shoals, anchoring landfast sea ice, driving coastal marine productivity, and shaping benthic ecosystems. Despite their importance, accurate, continent-wide automated mapping of grounded icebergs has been lacking. To address this gap, we developed an automated detection framework using Sentinel-1 Synthetic Aperture Radar imagery. The framework integrates a novel ResUNet deep-learning network with a multi-temporal algorithm, incorporating physical constraints from bathymetry and sea ice concentration to eliminate false positives. The method demonstrates strong robustness in complex coastal conditions, achieving an F1 score above 0.91 and lowering the minimum detectable iceberg size to 0.016 km². Leveraging the unprecedented low fast-ice extent in early 2025 (late February to early April), we constructed the first high-resolution, continent-wide dataset of grounded icebergs around Antarctica. A total of 39,696 stationary icebergs were identified across the continental shelf, with 67.4% classified as high-confidence grounded icebergs. Tiny icebergs (< 1 km²) numerically dominate (92.6% of targets) and contribute 53.9% of the total grounded area (13,435 km²). Dense clusters are concentrated on shallow shelves and downstream of disintegrating ice-shelf fronts, forming discontinuous chains along 56.5% of the coastline. Crucially, these tiny icebergs exert a potential “picket fence effect”, serving as physical anchors that stabilize fast ice. This dataset provides an essential quantitative baseline for modeling fast-ice stability, assessing nutrient fluxes, and mapping benthic habitats.