Frontal ablation is one of the most stubbornly difficult components of glacier mass balance to measure. It bundles together several distinct processes, including iceberg calving, melting where the ice face meets the ocean below the waterline, and melting and sublimation at the ice-air interface above it. Because calving events are sudden and submarine melt is hidden from view, researchers have historically had to settle for coarse temporal resolution, often averaging their estimates over years or even decades. Previous regional studies, such as work on 27 Alaskan glaciers, 38 major glaciers in Patagonia, and 49 tidewater glaciers on the Greenland Ice Sheet, each covered only a subset of the marine-terminating glaciers in their regions and rarely resolved seasonal variability.

The breakthrough in the new study comes from automation. Mapping calving fronts by hand is extraordinarily labor-intensive; earlier research teams manually digitized roughly 4,500 to more than 10,000 front positions to constrain mass change at glacier termini. The Svalbard team instead deployed a deep-learning segmentation model called Tyrion-T-GRU, which performs pixel-wise classification of satellite radar imagery to distinguish ice from ocean. Unlike earlier architectures that process single images in isolation, the multi-temporal model reads entire time series of radar images, giving it wider temporal context and stabilizing its predictions. Pre-trained on the CaFFe benchmark dataset of calving fronts from Antarctica, Alaska, and Greenland, the model was fine-tuned with 4,703 glacier-focused and 213 regional training pairs built from Sentinel-1 radar images and manually curated zone labels for ice, ocean, and land.

The scale of the resulting computation is staggering. The researchers preprocessed, ranked, and generated predictions for 203,294 satellite radar images across the archipelago, ultimately deriving 15,647 monthly-averaged calving front positions. To convert front positions into mass loss, the team combined the frontal area change with two additional components. Ice discharge, the flux of ice flowing through a fixed cross-section upstream of the terminus, was computed monthly from the ITS_LIVE velocity database, ice thickness maps derived from the Copernicus GLO-30 digital elevation model and Svalbard bedrock topography, and careful geometric corrections for flow direction. A climatic mass balance correction, drawn from the MAR regional climate model at 6-kilometer resolution, accounted for surface accumulation and melt within the glacier domain between the fluxgate and the calving front.

The results paint a picture of an archipelago in accelerating retreat. Of the 147 glacier basins, 119 showed net frontal displacement toward retreat over the decade, while only 28 advanced. The combined length of Svalbard’s tidewater glacier fronts grew from roughly 835 kilometers in 2015 to about 900 kilometers in 2024, an increase of approximately 8 percent, which the authors suggest may indicate a destabilization of glacier termini, particularly in the rapid growth observed after 2019. Ice discharge proved remarkably stable between years, averaging 17.75 plus or minus 0.20 gigatonnes per year and accounting for about 82 percent of total frontal ablation, but terminus mass change accelerated sharply, intensifying by roughly 60 percent compared with the 2010 to 2020 period.

One glacier dominates the regional budget. The Austfonna ice cap in northeast Svalbard, driven by the destabilized and surging Austfonna Basin 3, accounts for roughly 48 percent of the archipelago’s total frontal ablation, with Basin 3 alone losing ice at 5.05 plus or minus 0.35 gigatonnes per year. Five key glaciers explain 47 percent of the total decadal rate, and just 27 glaciers account for 80 percent, leaving the remaining 120 glaciers responsible for only a fifth of the loss. Other major contributors include the Kvitøyjøkulen ice cap, whose 101-kilometer calving front spans about 90 percent of its island’s coastline, and Stonebreen on Edgeøya. Negribreen stands out as a highly dynamic outlier: despite a terminus of only about 18 kilometers, its frontal ablation rate surged by roughly 375 percent compared with the previous decade, making it a prime candidate for future process studies.

The monthly resolution exposes seasonal rhythms that annual averages completely obscure. Frontal ablation rates peak between August and December, reaching a maximum in September, and drop to their lowest in April and July. Individual glaciers display striking seasonal cycles, with Austfonna Basin 3 advancing steadily during the first half of each year before retreating rapidly in the second half, a pattern repeated in nearly every year of the record. The dataset also reveals interannual swings, with frontal area change ranging from strong retreat of 12.83 gigatonnes in 2024 to a net advance of 11.63 gigatonnes in 2019, and the peak regional ablation year falling in 2018.

Placed in longer-term context, the new numbers confirm a troubling trajectory. Regional frontal ablation rates rose from 7.62 plus or minus 2.65 gigatonnes per year in 2000 to 2010, to 16.82 plus or minus 2.48 gigatonnes per year in 2010 to 2020, and now to 21.57 gigatonnes per year over 2015 to 2024, an increase of roughly 183 percent since the start of the century. All three long-term estimates fall outside each other’s uncertainty ranges, supporting a genuine escalation in mass loss. The drivers have also shifted: while earlier acceleration was dominated by increased ice discharge, the most recent period shows terminus mass loss accelerating faster than discharge, hinting at changing controls at the ice-ocean boundary.

The methodological achievement is as significant as the measurements themselves. Validation against 89 manually digitized front positions across ten representative glaciers yielded a mean regional spatial uncertainty of 38.1 meters, and comparisons with independent studies showed an 83 percent agreement rate in retreat and advance classifications. The authors note that sea ice and ice mélange, frozen debris that can pin glacier fronts in place, likely degraded model performance in winter months, and they recommend that future work exploit the wider-swath radar mode and interpolate across data gaps to achieve complete temporal coverage.

With the full dataset, including monthly frontal ablation, ice discharge, and calving front time series, publicly available on Zenodo, the study hands the glaciology community an unprecedented resource. Climate modelers can use the high-resolution record to calibrate projections and improve the notoriously poor representation of frontal ablation in ice-dynamical models, while process-oriented researchers can finally probe how ocean temperature, fjord bathymetry, subglacial hydrology, and ice mélange interact to govern the pace of Arctic glacier retreat. As the Arctic continues to warm faster than the rest of the planet, knowing exactly when and where these glaciers lose ice may prove essential for forecasting their contribution to global sea level.

Subject of Research: Monthly frontal ablation and ice discharge at Svalbard tidewater glaciers measured by deep-learning segmentation of satellite radar imagery

Article Title: A decade of monthly frontal ablation at 147 tidewater glaciers in Svalbard

Article References: Pyles, D., Dreier, M., Wendleder, A., Kochtitzky, W., Gourmelon, N., Christlein, V., & Seehaus, T. (2026). A decade of monthly frontal ablation at 147 tidewater glaciers in Svalbard. Earth System Science Data, 18(10), 7319-7344. https://doi.org/10.5194/essd-18-7319-2026

Image Credits: AI Generated

DOI: 10.5194/essd-18-7319-2026

Keywords: Svalbard, tidewater glaciers, frontal ablation, iceberg calving, deep learning, Sentinel-1, ice discharge, glacier mass balance, Austfonna, Arctic cryosphere, remote sensing, sea level