Thirty-three years of glacier grounding line retreat in Antarctica 1992-2025

The Grounding Line (GL) - the transition from ice grounded on the continent and ice afloat in the ocean - is a sensitive indicator of glacier stability and mass balance. Using differential SAR interferometry from ERS-1/2, Sentinel-1, RADARSAT-1/2, RCM, ALOS PALSAR-2, COSMO-SkyMed, and ICEYE, we assemble a continental scale record of grounding line migration from 1992 to 2025. Over 77±10% of Antarctic coastal length, we detect no GL migration. Stable areas include the vast Ross, Filchner-Ronne, Amery and West ice shelves, and broad sectors of Coats, Queen Maud, Enderby, and Princess Elizabeth Lands. Retreat is concentrated in (i) the Antarctic Peninsula - 2-18 km along Larsen A-B and 2-6 km along parts of GeorgeVI, and no change on Larsen C-D ice shelves; (ii) Wilkes and GeorgeV lands - 6-10 km on Denman, Totten, Moscow, Frost, Holmes, Mertz, Ninnis, and Cook, and 26 km on Vanderford; and (iii) West Antarctica - 5-7-km on Ferrigno, Fox, and Venable, with extreme retreat in the Amundsen and Getz sectors (Pine Island 33 km, Thwaites 26 km, Haynes 20 km, Pope 23 km, Smith 42 km, Kohler 12 km, East Getz 9 km toward Berry 18 km, Hull 14 km and Land 5 km). The ice sheet lost 12,820±1,873km2of grounded ice in 1996-2025, or 442±64 km2/year, with 62% from West Antarctica and 28% from East Antarctica. Retreat clusters in areas where bathymetry channelizes warm Circumpolar Deep Water toward deep grounding zones where beds are retrograde, except in the northeastern Antarctic Peninsula. The results provide a harmonized benchmark for ice grounding zone-based ice sheet models and identifies gateways where future retreat is likely to accelerate.

Methods
SAR Missions. We use radar interferometry (InSAR) observations from several satellite Synthetic-Aperture Radar (SAR) missions. The European Space Agency (ESA) Earth Remote Sensing ERS-1 collected repeat-pass interferometry at the C-band frequency (5.3 GHz) in 1992 and 1994 (3-day repeat). ESA collected data at a one-day repeat with ERS-1 and ERS-2 in 1995/1996. ERS-1 acquired 3-day repeat data at the end of its mission in 2000. ERS-2 acquired 3-day repeat data at the end of its mission in 2011. ESA launched Sentinel-1a in 2014, followed by Sentinel-1b in 2016, which acquired SAR data at the C-band frequency (5.4 GHz) with a 6 to 12 day repeat until Sentinel-1b ceased operations in 2021. Sentinel-1c was launched in December 2024 and collected data at a one-day repeat with S1a for one month over the Amundsen Sea. The Japanese Space Agency (JAXA) launched the ALOS PALSAR in 2006 to acquire SAR data at the L-band frequency (1.27 GHz) at a 44-day repeat until 2011. In 2014, JAXA launched ALOS-2 PALSAR-2 (1.24 GHz) at a 14-day repeat cycle. The Canadian Space Agency (CSA) launched RADARSAT-1 in 1995 to collect data at a 24-day repeat cycle at the C-band frequency (5.4 GHz) until 2013; RADARSAT-2 in 2007 with the same cycle but left (i.e. south) looking capability; and the RADARSAT Constellation Mission (RCM) in 2019 started to acquire data in Antarctica in 2023 at a 4-day repeat cycle. The Agenzia Spaziale Italiana (ASI) launched the CSK2 and CSK3 satellite part of the CosmoSkyMed Constellation in 2007 and 2008, which started to acquire 1-day repeat track data at the X-band frequency (9.60 GHz) in 2015 until 2021 after which we no longer have access to data. Finally, ICEYE launched X6 and X7 in 2020 to operate at the X-band frequency (9.65 GHz) on a one-day ground track repeat (GTR) starting in 2023. For each SAR, we form interferometric pairs with the natural ground repeat cycle and difference consecutive pairs to obtain a differential interferogram, or DInSAR, that measures the differential displacement between two consecutive SAR interferograms.

Differential Interferometry. One SAR interferogram measures the line of sight displacement of the ice surface over one repeat cycle, combined with topography which we remove automatically using a digital elevation model of Antarctica. In a differential interferogram, or DInSAR, the steady component of the glacier motion cancels out, leaving only the motion associated with tidal flexure of the ice on floating ice or blurbs of subsidence/uplift caused by subglacial water migration on grounded ice, or noise. When the glacier accelerates, the assumption of steady state breaks down, but the corresponding pattern of motion will resemble a pattern of horizontal motion rather than a pattern of tidal flexure. To mitigate the effect of speedup, we tend to use nearest neighbor pairs of interferograms and do not combine interferograms acquired months apart.

GL detection. We do not unwrap the interferometric phase to quantify vertical displacements but use the pattern of fringes (or 360 degrees in phase) instead to detect the GL, as in Rignot et al. (2011), i.e., we detect the first interferometric fringe in the seaward direction, or initial vertical motion indicating that the ice lifts off its bed at high sea level height. For Sentinel-1a/b/c, we use a Machine Learning algorithm to map a large number of GL (Mohajerani et al., 2021) in 2018, 2019 and 2020, but these products still require a final stage of manual filtering.

At the C-band frequency, we have one fringe of signal for every 28 mm of line of sight deformation. This ratio decreases by a factor 4 at the L-band frequency and increases by a factor 2 at the X-band, i.e. for the same tidal signal we get 4 times less fringes at the L-band and 2 times more at the X-band. The operating frequency of the radar, however, does not matter for grounding line mapping, which is based on a differential displacement of the ice surface caused by an increase in basal water pressure measured independent of the radar frequency. Frequency selection affects the noise level of the signal.

Retreat estimates. Because the glaciers are broad, 120 km wide for Thwaites, it is difficult to characterize the retreat with a single number. We measure the retreat along a flow line of the fastest-moving portion of the glacier, or central flow line, between the most retreated position of the GL in the 1990s and the most retreated position of the GL in the latest data (2021-2025). The quotation is therefore approximate. We complement it with an estimation of the total area of retreat, which is more rigorous and well defined. Our estimates of grounded ice loss exclude glaciers terminating in a calving cliffs in the Peninsula, but the difference is small and probably at the 100 km2 level.

Errors in GL detection and areas of ungrounding assume a precision of ±500 m in GL mapping, which is conservative given that at a given epoch using one DInSAR, we pick the GL position within 100 m. The ±500 m accounts for the fact that we do not fully sample the tidal-induced, short-term migration of the GL, especially where we have few tracks. We measure the area of ungrounding in square kilometer. We transform the polygon data from EPSG 4326 (latitude, longitude, WGS84) into EPSG 6932 (x, y, WGS 84, named NSIDC EASE-Grid 2.0 South, which is a Lambert Azimuthal Equal Area centered on the South Pole). We delineate polygons between the most retreated position of the GL in the 1990s and the most retreated position of the GL in the most recent data (2021-2025). All GL positions include the time of acquisition of the data, which makes it possible to correct for changes in tides and atmospheric pressure. We do not attempt to do this correction herein because the relationship is not linear (seawater intrusions can be trapped in between tides (Chen et al., 2023) and the correction is non critical with multiple observations (Rignot et al., 2024).

References

H. Chen, E Rignot, B Scheuchl, S Ehrenfeucht, Grounding Zone of Amery Ice Shelf, Antarctica, From Differential Synthetic-Aperture Radar Interferometry. Geophys. Res. Lett. 50(6), 613 e2022GL102430 (2023).

Y Mohajerani, et al., Automatic delineation of glacier grounding lines in differential interferometric synthetic-aperture radar data using deep learning. Nat. Sci. Rep. 11, 4992 (2021).

E Rignot, J Mouginot, B Scheuchl, Antarctic grounding line mapping from differential satellite radar interferometry. Geophys. Res. Lett. 38, 1–6 (2011).

E Rignot, et al., Widespread seawater intrusions beneath the grounded ice of Thwaites Glacier, 597 West Antarctica. Proc. Nat. Acad. Sci. 121(22), e2404766121 (2024).