West Antarctic Ice Sheet (WAIS) behaviour under future global warming scenarios remains a great uncertainty in sea-level projections1. Over recent decades, the WAIS has retreated and thinned at accelerating rates in the Amundsen Sea Embayment and is predicted to continue this trend2,3. Of concern is the ongoing mass loss of Thwaites and Pine Island glaciers (Fig. 1), which together drain a large portion of the Amundsen Sea sector and reach deep into the heart of the WAIS. These glaciers are susceptible to rapid retreat because they are being melted from beneath by warm Circumpolar Deep Water4,5. Moreover, Thwaites and Pine Island glaciers rest below sea level on a retrograde slope with no known major topographic highs on which the glaciers could stabilize, and therefore they may be susceptible to runaway retreat via marine ice-sheet instability6,7. Such retreat is likely to trigger extensive ice loss of the WAIS8,9, contributing as much as 3.4 m to global sea level over the next several centuries10.

Fig. 1: Map of Amundsen Sea Embayment, Antarctica, showing sites mentioned in the text.
figure 1

Our study locations are shown in bold text. Textured grey represents the current ice-sheet surface (from Reference Elevation Model of Antarctica35). Shaded grey areas outlined in blue represent ice shelves. Dashed blue lines indicate former ice flow directions through Pine Island Trough (PIT). Bathymetry (blue and white shades) is from GEBCO2019 global dataset36. The basemap was created from the SCAR Antarctic Digital Database.

Geologic evidence has provided insight into ice behaviour in the Amundsen Sea sector. Existing data show that Thwaites11 and Pine Island12 glaciers previously merged in Pine Island Trough to form a palaeo-ice stream, which extended to the continental shelf edge at the Last Glacial Maximum13,14. Marine radiocarbon and terrestrial exposure-age dating show deglaciation of the outer embayment at 12–9 thousand years ago (kyr)15,16, with the grounding line reaching close to its current position by the early Holocene14,15. Cosmogenic exposure-age studies on nunataks near Pope and Pine Island glaciers indicate that ice in this region thinned to current elevations by the mid-Holocene17,18 (Fig. 1).

However, the lack of evidence for ice behaviour since the mid-Holocene allows two plausible hypotheses: (1) Thwaites and Pine Island glaciers reached their current configuration in the mid-Holocene and have remained stable until very recently, or (2) these glaciers continued to retreat to positions behind present-day margins until subsequent re-advance to near current limits in the late Holocene. The former would suggest that the recent rapid ice loss from these glaciers is unprecedented over the past 5 kyr and could evolve into runaway retreat7,8, whereas the latter would imply that current rapid ice recession could be reversible. Discriminating between these scenarios is becoming increasingly urgent as some have proposed irreversible retreat of Thwaites Glacier may be under way19.

Records of relative sea level (RSL) reflect the combined effect of global ocean volume change and local perturbations to the solid Earth and geoid due to glacial isostatic adjustment (GIA). Thus, determining the timing and pattern of Holocene RSL variation can provide information on local ice history since the Last Glacial Maximum as well as on current and past rates of rebound directly related to ice-mass fluctuations (for example, refs. 20,21). Today, rapid uplift is prevalent in the Amundsen Sea Embayment region, where current rates exceed 40 mm yr–1 in some locations22. Holocene RSL data afford information on whether such rates are anomalous and can identify transgressions or periods of slowing RSL fall, both of which can be attributed to ice expansion when changes in global ocean volumes are considered.

Radiocarbon and exposure-age results

Our study focused on three largely granitic island chains (Lindsey, Schaefer and Edwards) in the Amundsen Sea Embayment (Fig. 1). Many islands display striations and glacially polished bedrock at higher elevations16 (>19 m above present-day sea level (asl)). These erosional features probably restrict past sea level to <19 m elevation since deglaciation because such features are destroyed easily by wave erosion. All three island chains have raised beaches (Fig. 2 and Supplementary Figs. 6 and 7), the highest of which occur on the Lindsey Islands at ~19 m asl. Thus, available evidence indicates the regional marine limit (the highest elevation reached by the post-glacial sea) is at 19 m elevation. At the Schaefer and Edwards islands, the highest discernible beaches are at ~15 m and 11 m elevation, respectively, although the marine limit may well be a few metres higher on the Schaefer Islands, where it was obscured by snow during our visit.

Fig. 2: Satellite imagery and photographs of study sites in the Amundsen Sea Embayment.
figure 2

ac, Left panel shows Edwards Islands (a), Lindsey Islands (b) and Schaefer Islands (c). Dashed lines in photos on right panels denote raised marine beaches from which shell and bone samples were collected. Photos are from sites corresponding to the red squares in adjacent imagery. Yellow stars in a denote location of additional sampled islands (Photos provided in Supplementary Figs. 2, 3 and 6). Credit: ac (left), WorldView-2/DigitalGlobe, a Maxar Company; ac (right), Scott Braddock.

We derived an RSL curve from 55 calibrated radiocarbon ages of shells (limpets, probably Nacella sp.) and penguin bones (probably Adélie) extracted from raised beaches (Fig. 3a and Supplementary Table 1). We infer that the shells were incorporated into beaches while the molluscs were still alive or shortly thereafter, so radiocarbon dates from shells date beach formation and therefore contemporaneous sea level. However, it is also possible for previously deposited shells to be recycled into younger beaches, so some fraction of shell radiocarbon dates may represent only maximum-limiting ages for beach deposits. In this study, except for three old outliers from the Schaefer Islands, shell radiocarbon ages increase systematically with elevation. Furthermore, ages from similar elevations are tightly clustered, and, as noted in the following, shell ages and a single exposure age at 8 m elevation agree. Thus, we conclude that the shells afford actual—rather than apparent—ages for the beaches and that the true RSL curve must pass through the shell data (Fig. 3a; see Supplementary Information). Extrapolating the trend in shell ages to the marine limit at 19 m indicates that the marine limit dates to 5.5 kyr. In contrast to the shells, we infer that bone samples, mostly surface finds from relict penguin nests, were deposited only after a beach uplifted above sea level. They afford minimum-limiting ages for the beaches, and therefore the RSL curve must lie on or below the elevation of these samples.

Fig. 3: RSL reconstruction and comparison with GIA models for the Amundsen Sea Embayment, Antarctica.
figure 3

a, RSL curve based on radiocarbon ages of shells from raised beaches. All samples are presented as calibrated radiocarbon ages for shells and bones with horizontal bars indicating 2-sigma errors. The solid black line with associated bootstrap confidence intervals is an interpolation of the shell data (see Supplementary Information) using the method of ref. 37. The horizontal dashed black line represents the proposed regional marine limit (elevation of highest sampled beach) at 19 m. Elevation errors (Methods) associated with radiocarbon samples are represented by vertical bars on each data point and range from 0.3 to 0.7 m. Vertical error bars for Lindsey Island samples (0.3 m) are smaller than symbols at this scale. b, Comparison of RSL data with GIA model predictions derived using two different ice-history models (ICE-6G_C27,32 and W1228,33). For each ice-history model, two curves are shown; these represent RSL change assuming either a strong (solid lines, upper mantle viscosity = 5 × 1020 Pa s) or a weak (dashed lines, upper mantle viscosity = 5 × 1019 Pa s) Earth model. The lithosphere thickness is set at 71 km (ref. 34).

In addition to shells and bones, we also measured cosmogenic 10Be from four bedrock samples to provide additional information on the timing of deglaciation. Two samples from the Lindsey Islands (LD-19-C1 and LD-19-C2), both from striated bedrock above the inferred marine limit, yielded ages of 7.9 ± 0.3 kyr and 8.8 ± 0.2 kyr (Supplementary Table 2), suggesting that the islands became ice free at approximately this time. These ages are younger than two published 10Be bedrock exposure ages from the Lindsey Islands (recalculated 10.5 ± 0.4 kyr, 15.4 ± 0.7 kyr (ref. 16); Supplementary Table 3). The latter of these published ages was excluded by ref. 16 as having previous exposure. The former also could have previous exposure, but alternatively could reflect slightly earlier deglaciation of the other side of the island from our samples. Our ages are the same as that of a similar-elevation sample from the Jaynes Islands (9.0 ± 0.5 kyr; ISL-3 in16), 50 km from the Lindsey Islands.

In contrast to data from the Lindsey Islands, our two exposure ages from the Edwards Islands of bedrock at 19.1 m elevation yielded an average age of 3.9 ± 0.1 kyr (Fig. 3a). These much-younger ages may reflect either delayed deglaciation of the Edwards Islands relative to the Lindsey and Schaefer islands or emergence of the bedrock from the ocean. If the latter, the 10Be ages appear too young when compared with the radiocarbon-derived RSL curve. Moreover, it would require the local marine limit on the Edwards Islands to be at or above 19 m elevation, something for which we did not find evidence. Because of these factors, and the fact that the age of the samples is approximately equal to the age of the observed local marine limit at 11 m elevation (on the basis of the RSL curve), we favour the interpretation of delayed deglaciation, of either local glacier ice or a remnant stagnant ice block, for these ages.

A recalculated age of 2.8 ± 0.2 kyr (Fig. 3a) from an erratic boulder at 8 m elevation on an island 5 km north of the Edwards Islands was previously interpreted as representing retreat of the Canisteo Peninsula ice front23. However, this exposure age is indistinguishable from radiocarbon-dated shells in beach deposits at the same elevation, so must record emergence rather than deglaciation.

Broader implications of RSL data

Regardless of the interpretation of the exposure-age data, the radiocarbon-constrained RSL curve (Fig. 3a) implies uninterrupted RSL fall in the Amundsen Sea Embayment since 5.5 kyr. We attribute this to unloading of the ice mass dominated by Thwaites and Pine Island glaciers. Although the shell data form an approximately linear array, an exponential curve also could fit the data with reasonable confidence. The data indicate monotonic RSL fall, which is consistent with a simple unloading history (for example, ref. 24). More complex curves, including those showing evidence of marine transgression, typically reflect complicated glacial histories including re-advance (for example, refs. 20,21,25). Our RSL curve does not provide evidence for Holocene ice thickness fluctuations large enough to slow or reverse isostatic rebound following early Holocene mass loss. Although we cannot exclude the possibility of minor grounding-line oscillations, these RSL data do not indicate substantial mass fluctuations in the Amundsen Sea Embayment since 5.5 kyr.

Our results show that the Amundsen Sea region experienced on average ~3.5 mm yr–1 of RSL fall between 5.5 and 0.3 kyr (Fig. 3a). By contrast, bedrock uplift rates are currently 15 mm yr–1 at sites 60–95 km from our field sites and as much as 41 mm yr–1 elsewhere in the Amundsen Sea Embayment22. Although RSL change reflects multiple factors in addition to bedrock uplift (for example, ocean volume change), the large difference we observe between modern bedrock uplift rates and Holocene RSL change requires that a recent increase in the rate of viscoelastic uplift is contributing to current high rates of isostatic adjustment. These rates are linked to contemporary ice-mass loss in this region22, which appears to be unprecedented over the past 5.5 kyr.

The exposure ages from the Lindsey Islands, along with the radiocarbon date of 10.7 kyr from a reworked shell from the Schaefer Islands, indicate that the eastern Amundsen Sea Embayment continental shelf was deglaciated at 11–9 kyr. These ages are consistent with marine geological records that show retreat of the grounding line to its modern position by the early Holocene15. However, beaches apparently did not develop until ~5.5 kyr. Given the proximity of the islands to the current ice margin on the Canisteo Peninsula (<2 km), a possible explanation for the delay is the presence of an ice shelf until ~5.5 kyr, as suggested by ref. 17. An ice shelf that surrounded (but did not cover) the islands would prevent beaches from forming but allow marine organisms to live offshore and 10Be to accumulate in exposed bedrock. This hypothesis is consistent with evidence from marine sediment cores that indicates a remnant ice shelf persisted in the Amundsen Sea Embayment until at least 7.5 kyr (ref. 26). Perennial land-fast sea ice could have the same effect of preventing beach formation.

Comparison with GIA models

The GIA models for Antarctica typically predict RSL rise in the early Holocene, a marine limit at ~20 m asl and RSL fall in the mid to late Holocene27,28. These GIA model results agree relatively well with observations where extensive RSL data exist (for example, the Ross Sea24,29 and Marguerite Bay30,31). Existing RSL data commonly show raised beach deposits recording exponential RSL fall in the mid to late Holocene. Our RSL data from Pine Island Bay display similar features. We compared our RSL data with GIA model predictions generated using two different Antarctic ice-history models, ICE-6G_C27,32 and W1228,33 (Fig. 3b). In W12, the islands in this study become ice free between 11 and 10.5 kyr, and in ICE6G_C, between 8.5 and 8.0 kyr. Predictions were generated separately for all three island groups, but since all sites lie within 30 km, model predictions differ by only ~2 m at all sites for the late Holocene. For this reason, Fig. 3b shows the averaged-RSL prediction for all three island groups. Results are shown in Fig. 3b for a strong (upper mantle viscosity = 5 × 1020 Pa s) and a weak (upper mantle viscosity = 5 × 1019 Pa s) Earth rheology model. The Amundsen Sea Embayment is known to have a relatively thin lithosphere34 with a very low upper mantle viscosity22. The ICE-6C_C model, combined with the weaker Earth rheology model, provides the best fit to our RSL data. In addition, the exponential curve produced by the weaker Earth rheology model fits within the confidence bounds of our RSL reconstruction, whereas that of the strong Earth rheology model does not and overestimates regional Holocene RSL. However, none of the models is consistent with the exposure ages on bedrock above the marine limit or with our interpretation of the elevation of the marine limit based on the upper limit of marine sediments and the lower limit of glacial polish and striations. All of the GIA models in Fig. 3b predict that the exposure-age samples at 19–26 m elevation should have been under water at the time of deglaciation and consequently not exposed to cosmic radiation. Therefore, although the ICE-6G_C (weak) model fits our RSL data between 0 and 4 kyr (Fig. 3b), we infer that none of the models considered here accurately captures the mid- to late Holocene RSL history of the region.

In summary, the Holocene RSL record presented here puts into context the current rapid bedrock uplift near the grounding lines of Thwaites and Pine Island glaciers and shows a monotonic fall in RSL since the mid-Holocene. Although the resolution of the RSL curve does not preclude the possibility of minor marginal fluctuations of Thwaites and Pine Island glaciers during the past ~5.5 kyr, this record is best explained by the hypothesis that ice reached close to its current margin in the mid-Holocene and has remained in the vicinity of that position since, without large-scale glacier recession or re-advance. Thus, there remains no direct evidence that Thwaites and Pine Island glaciers were substantially smaller than present during the present interglacial period, and the present-day rate of ice recession appears to be unprecedented in the past ~5.5 yr.


Field sampling

Field operations were part of cruise NBP1902 conducted from the US research vessel the Nathaniel B. Palmer. Marine organic material (shells, probably from the limpet Nacella, and bones, probably from Adélie penguins (Supplementary Fig. 7)) were collected from raised marine beaches. Samples were found beneath boulders (up to 25 cm diameter) or in ~0.5 m deep excavations. We collected a minimum of three samples from each beach where possible. We avoided surface samples because they are more likely to be modern in an area heavily colonized by penguins at the present day. Samples for exposure-age dating were collected from bedrock outcrops using a hammer and chisel. There was no topographic shielding.

We determined the elevation relative to sea level (here defined as the EGM96 geoid) of at least one location on each distinct raised beach using a Septentrio APS3 high-precision Global Positioning System (GPS) unit with differential correction relative to a temporary base station deployed during each sampling period and located additional sample locations using uncorrected handheld GPS. To determine the elevations of sample locations lacking DGPS positions, but that are still located on corresponding beach ridges, we used photogrammetric digital elevation models (DEMs) of each island group prepared by the Polar Geospatial Center from DigitalGlobe satellite imagery. We registered each DEM to the Differential GPS (DGPS) data by applying a vertical offset determined to minimize mean square differences between DEM and DGPS elevations for DGPS survey points, and then sampled elevations from the corrected DEMs for all sample locations. The elevation uncertainty is estimated from the distribution of the residuals between the DGPS points and the elevations of the corrected DEMs at those locations and ranges from 0.3 to 0.7 m. We applied the DEM-derived elevation uncertainties to all samples in Fig. 3 to account for undulations in beach ridges.

Radiocarbon dates

Radiocarbon samples were prepared at the University of Maine Glacial Geology and Geochronology Laboratory. We cleaned shell samples in ultrapure water overnight in an ultrasonic bath before drying, weighing and shipping them to the National Ocean Sciences Accelerator Mass Spectrometry (NOSAMS) facility for dating. For bone samples, we extracted collagen following procedures modified from the University of California Santa Cruz Stable Isotope Laboratory ( Pieces (50–100 mg) of bone were washed in ultrapure water for 24 hours in a sonicator and dried overnight. Then samples were decalcified in a 0.5 N HCL solution for 24–48 hours. Next, we rinsed the samples and let them sit in ~10 ml of 0.1 N NaOH solution overnight to remove contaminants. To defat samples, they were subjected to multiple baths in petroleum ether while in an ultrasonic bath. Last, we rinsed and dried the collagen samples before sending them to the Environmental Geochemistry Laboratory at Bates College for carbon/nitrogen ratio (C/N) analysis. Samples with a C/N ratio between 2.9 and 3.6 were considered to have sufficiently well-purified collagen to be suitable for radiocarbon dating38. We sent collagen samples within the appropriate C/N range to the NOSAMS facility for radiocarbon dating. Resultant radiocarbon ages were converted to calendar yr bp using CALIB version 8.139 and the Marine20 calibration dataset40. Although this calibration dataset is not optimized for polar regions40, it is the best correction available at present (P. Reimer, personal communication). We applied a delta-R value of 610 ± 110 yr, recalculated from the Holocene Antarctic coral dataset41 for specific use with Marine2020.

Cosmogenic exposure-age dating

Four bedrock samples were processed at the University of Maine Cosmogenic Isotope Laboratory. We crushed and sieved samples and subjected them to froth flotation, followed by a sequence of HF/HNO3 leaches to isolate pure quartz (as determined by ICP-OES). We weighed approximately 30 g of pure quartz for each sample and spiked them with an in-house 10Be carrier made from phenakite. Following standard ion-exchange chemistry42, samples were precipitated as hydroxides, converted to BeO and then packed in cathodes with niobium. The 10Be/9Be ratios were measured at the Center for Accelerator Mass Spectrometry at Lawrence Livermore National Laboratory. Samples ran with high 9Be currents (20–30 microamps). The full 10Be/9Be procedural blank for the batch was 3.6E−16. We calculated ages with version 3 of the online exposure-age calculator described by ref. 43 and subsequently updated using the default calibration dataset44 and ‘St’ scaling45,46. Ages in the text are given with 1-sigma internal errors, whereas those plotted on the RSL curve are shown with 1-sigma external errors that incorporate uncertainties in production rate and scaling for more realistic comparison to radiocarbon ages.