Introduction

Sea level is rising along the world’s densely populated coastlines, driven by ice sheet and glacier mass loss, groundwater fluxes, and seawater thermal expansion1. One of the largest and deepest unknowns in sea level projections is how fast the Antarctic Ice Sheet (AIS) may shed mass over the 21st century1. Much of the spread in AIS projections represents uncertainty about which processes may drive change, how fast those processes operate, and how to represent those processes in projections2.

The need to address these uncertainties has turned attention to past times when the AIS endured climates as warm as Earth now is or may soon be. These intervals include the Last Interglacial (LIG, 129 − 116 ka), Marine Isotope Stage 11c (426 − 396 ka, hereafter MIS-113), and the mid-Piacenzian Warm Period (MPWP, 3.3 − 3 million years ago, Ma). Each interval provides a useful window into AIS behavior in an Earth system whose state differs somewhat from present or historical climates4. During the LIG, when peak global mean sea level (GMSL) reached between 2 and > 9 meters above present (Fig. 1), atmospheric CO2 concentrations remained near pre-industrial (~ 1850 CE) levels and global mean surface air temperature was comparable to the early 21st century (about 0.5–1.5 °C above pre-Industrial levels5), while polar temperatures were at least several degrees Celsius warmer than pre-industrial levels6. During the exceptionally long (> 30,000 years) MIS-11 interglacial, mid-to-high latitude temperatures were slightly colder than the LIG7, but Arctic temperatures were likely exceptionally warm in summer8; ice sheet melt during this period led peak GMSL to exceed 6 meters and possibly reach 13 meters above present9. Finally, the MPWP, when sea levels may have reached 20 meters above present10, represents the most recent time atmospheric CO2 concentrations rivaled those of the early 21st century, with CO2 levels reaching 394 (+34/− 9) ppm during the KM5c interglacial (~ 3.205 Ma11). Differences between the modern and MPWP Earth systems---including in dynamic topography, atmospheric dynamics, subglacial topography and sediment properties, and ocean circulation patterns---render the MPWP an imperfect analog for the 21st century. Nevertheless, Pliocene ice sheet configurations and sea levels are considered the closest analogs for the multi-millennial-scale sea-level rise that present warming may cause12.

Fig. 1: Compilation of Last Interglacial (LIG) reconstructions of peak global mean sea level (GMSL) and Antarctic and Greenland Ice Sheet (AIS, GrIS) contributions.
Fig. 1: Compilation of Last Interglacial (LIG) reconstructions of peak global mean sea level (GMSL) and Antarctic and Greenland Ice Sheet (AIS, GrIS) contributions.
Full size image

Estimates of (a) peak LIG GMSL and GMSL contributions from the (b) GrIS and (c) AIS are plotted in order of publication year from left to right. Colorbar denotes timing of peak contribution in thousands of years (kyr) before 1950. Vertical bars with/without stars denote 95% credible intervals / reported maximum and minimum bounds; circles indicate best estimates; squares mark estimates corrected for glacial isostatic adjustment. AIS estimates are limited to those directly modeled either by an ice sheet simulation or a glacial isostatic adjustment model. Horizontal light gray bands delineate minimum envelopes of agreement among studies, i.e., bounds that encompass or intersect all estimates given their uncertainties. Source information on reconstructions, including the model used, data constraints employed, and uncertainty assessment, is included in the Supplementary data.

While these past warm periods with higher-than-present GMSL differ in astronomical forcing, atmospheric CO2 concentration, and other boundary conditions, they have often been assumed to share one feature: the absence of North American ice sheets. The assumption appears largely justified during the Pliocene: while the Cordilleran Ice Sheet (CIS) may have started to expand at ~ 3.1 Ma, Northern Hemisphere glaciation only intensified at ~ 2.7 Ma as ice on North America expanded to mid-latitudes13 and the CIS reached its largest Plio-Pleistocene extent14. However, mounting evidence suggests that North American ice sheets persisted during the early stages of the Pleistocene’s warmest interglacials. This persistence, which has been noted7,15,16 but remains overlooked in most paleoclimate simulations, has implications for harnessing Earth history to help improve projections of 21st century sea-level rise and ice sheet decay. In this Perspective, we summarize the state of knowledge concerning North American ice sheet (NAIS) presence during past interglacials; discuss how this persistence could alter understanding of sea level, ice sheets, ocean circulation, and atmospheric dynamics during past warm periods; and outline implications for future projections.

North American ice sheet interglacial persistence: evidence

Reconstructing Pleistocene NAIS evolution is challenging because the Last Glacial Maximum (~ 26 ka) NAIS footprint erased nearly all older terrestrial sedimentary deposits. Nevertheless, indirect evidence has accumulated that constrains the ice sheet’s Plio-Pleistocene evolution.

Last interglacial

The strongest evidence for early LIG NAIS presence comes from marine sediments. Western North Atlantic sediments have a red layer estimated to occur at 126.7 ka (120.7–130.6, 95% credible interval; ages reported with ± are 2σ unless stated otherwise) according to new probabilistic age models we produce for several deep sea cores (Zhou and McManus, 2022 and Supplementary Information), or at 126.5 ± 1 ka using a speleothem U-Th based chronology that has a mean 2σ uncertainty of 770 years15. Multiple lines of evidence suggest that this layer has the same geographic origin as a red layer of ~ 8.2 ka age deposited when a proglacial lake flooded under then burst through a remnant Laurentide Ice Sheet (LIS) ice dam to enter the Labrador Sea (Fig. 2a)17,18,19. First, the layer has high iron oxide concentrations, the chief nearby source of which is the hematite-rich Dubawnt Supergroup that outcrops west of Hudson Bay, and high Ca/Sr ratios, which indicate detrital carbonate from Hudson Strait. Though East Greenland and the United Kingdom also host carbonates and hematite-rich strata, the Hudson Strait region’s proximity to the core sites with red layers suggests that it is the source. Second, the distribution of the red layer follows the Northwest Atlantic Mid-Ocean Channel, pointing to the upstream Hudson Strait region as the source. Third, the bottom contact of the red layer is relatively sharp, and 230Th-normalized mass flux reconstruction indicates that the red layer was deposited very rapidly16. The red sediments are unlikely to have been transported to the core sites from a remote source region (such as by ice rafting or by current transport) and still form a sharp bottom contact. It is therefore very unlikely that East Greenland and the United Kingdom are the source regions for the red layers16. Numerical simulations, supported by radiometrically-dated moraines, indicate that the LIS held enough water at 8.2 ka to raise GMSL 3–5 meters20,21. No such direct evidence has been published to constrain the LIG LIS’s size. However, the wide geographic spread of LIG red layers across the North Atlantic could suggest a Hudson Strait outburst flood of comparable or greater size to the 8.2 ka event, which could imply an LIS holding multiple meters of GMSL equivalent ice volume.

Fig. 2: Marine sedimentary evidence for Laurentide Ice Sheet (LIS) persistence into the Last Interglacial (LIG).
Fig. 2: Marine sedimentary evidence for Laurentide Ice Sheet (LIS) persistence into the Last Interglacial (LIG).
Full size image

a Sediment cores with a red layer thought to source from a Last Interglacial Laurentide Outburst (LILO) flood event. White/blue areas denote LIS92 / glacial lake Agassiz-Ojibway100, respectively, at ~ 8.5 ka, prior to the analogous 8.2-kyr event. Red stars denote sediment cores with a red layer marking the LILO event. b Sediment cores with LIG red layer, modified from ref. 22. Cores include International Ocean Discovery Program (IODP) Expedition 303 Sites U1302-C2H6101 and U1305-C4H2102; Ocean Drilling Program (ODP) Leg 105 Site 646-B2H5103,104; Lamont-Doherty Earth Observatory EW9303-37JPC22; and ODP Leg 172 Sites 1063-B4H5105 and 1061-D4H7106. Ages of the red layer—median in bold, 95% credible interval in parentheses, with layer base marked with black lines—are estimated using the BIGMACS stratigraphic alignment library107 (see Supplementary Information). MIS: Marine Isotope Stage. Core images are reproduced without change from the LDEO Core Repository (EW9303-37JPC) and IODP photo archive (all other cores), from which the photos are available on a CC By 4.0 license (https://creativecommons.org/licenses/by/4.0/).

LIS LIG persistence is further supported by Labrador Sea lead isotope records, a proxy for Canadian cratonic erosion, indicating that the LIS lost considerable mass at 126.5 ± 1 ka22. Ref. 22 do not assess formal uncertainties on the lead isotopic evidence itself. However, their measurements are made on a core from IODP Site U1302/3 core (Fig. 2) that contains the red layer, and this layer occurs during the radiogenic isotope spike, suggesting synchrony between the events. Additional evidence of the timing of LIS disappearance, an optical stimulation luminescence date of 122 ± 12 ka from Baffin Island lake sediments, provides only minimal further constraint23. In the following, we adopt 126.5 ± 1 ka for the putative age of LIG LIS disintegration, acknowledging that future work should refine this timing.

Higher LIG insolation forcing relative to the Holocene has been cited as evidence for pre-LIG LIS disappearance24. Conversely, LIS existence has been hypothesized to depress regional temperatures by increasing albedo and altering atmospheric dynamics25 – effects that could have supported LIG LIS persistence even with higher insolation. Both lines of reasoning miss other relevant feedbacks in the system. For instance, average Northern Hemisphere cooling does not address the relevant summertime cooling around LIS ablation zones; the LIS orographic impact could also cause summertime warming elsewhere in the Northern Hemisphere; and LIS mass loss ultimately occurred because ice marginal surface melt, marine calving, and submarine melt together exceeded snow accumulation. Nevertheless, timeslice experiments using a general circulation model with prescribed ice sheet boundary conditions suggest that a remnant early-LIG LIS likely did cool the Arctic through a combination of higher albedo, the heat sink effect of the ice sheet, and reduced sea ice melt26. This regional cooling is further supported by proxy records of lower-than-present Norwegian Sea surface temperatures (SST) from 128-126.5 ka27; a coincident cold freshwater pulse entering the North Atlantic28; and North Atlantic SST temperature proxies lagging behind ice volume proxies in early MIS-5e29. One caveat to this hypothesis is that the GrIS may have also contributed to early-LIG North Atlantic cooling and freshening.

Terrestrial evidence for LIS collapse timing remains too poorly dated to provide context for persistence but offers opportunities for future advances. For instance, morainal evidence suggests that the penultimate glacial maximum LIS may have exceeded the Last Glacial Maximum LIS in eastern but not western North America30,31. These moraines, if dated, could help bound the evolution of the eastern LIS during the Penultimate glacial cycle, if not its collapse.

Persistence of a substantial LIS until 126.5 ± 1 ka would align with indications that the peak contribution of the Greenland Ice Sheet (GrIS) occurred late in the LIG32, during or after the Northern Hemisphere (65°N) summer insolation maximum at 127 ka (Fig. 1b)33. While peak LIG GMSL estimates remain between 2 and > 9 meters (Fig. 1a), the last decade has seen estimates of the GrIS contribution largely center on 1-3 meters after 124 ka (Fig. 1b), though some estimates reach as high as 5 m32. By contrast, the AIS likely reached its LIG minimum prior to 126 ka (Fig. 1c)34,35,36. The AIS maximum contribution remains uncertain, between 3 and 8 meters (Fig. 1c). A larger contribution aligns with evidence that the West Antarctic ice sheet (WAIS) may have collapsed completely37, accompanied by a contribution from East Antarctica38. If a complete WAIS collapse occurred, the ice sheet likely regrew quickly, as the WAIS configuration after 126 ka requires preservation of the Weddell Sea’s Ronne Ice Shelf35,39.

Finally, the timing of LIG LIS disintegration overlaps with the age (126 ka) of the oldest dateable ice at the Skytrain Ice Rise of West Antarctica’s Ronne Ice Shelf39. The Skytrain Ice Rise abruptly thinned by 400 meters during the 8.2 ka event40, suggesting a link between the Holocene LIS outburst flood and Weddell Sea ice loss. If an LIG outburst flood event also drove West Antarctic thinning just prior to 126 ka, it could explain the lack of Skytrain Ice Rise ice > 126 ka.

Taken together, the above ice histories imply that LIG LIS persistence may have coincided with maximum AIS melt—an implication supported by glacial isostatic adjustment (GIA) modeling showing that simultaneous WAIS collapse and LIS persistence produce a relative sea-level oscillation whose amplitude and geometry resemble observations41. If this coincidence did occur, LIS presence would have masked the full AIS contribution to GMSL in an early (i.e., 129-126.5 ka) LIG sea-level highstand42. The individual ice-sheet histories also suggest that at no point during the LIG did Earth’s ice sheets reach a simultaneous minimum, likely due to the anti-phased summer solar insolation intensity experienced at each pole as well as the bipolar seesaw.

While we focus on the LIS, the Eurasian Ice Sheet (EIS) complex, particularly the Fennoscandian Ice Sheet, is also typically assumed to be absent by the start of the LIG GMSL sea-level highstand at 129 ka. However, extreme North Atlantic cooling is associated with the terminal stadials that typify deglaciations of the 100 kyr world; such cooling43 could have arrested Northern EIS retreat and led a remnant ice sheet to be present during the subsequent interglacial. More quantitative constraints on the timing of EIS disappearance will be needed to verify the timing of the disappearance of the EIS.

MIS-11 interglacial

Until recently, discussion of MIS-11 ice volumes did not consider North American ice sheets7. The three existing quantitative assessments place peak GMSL late in the interglacial at 6–13 meters (99% credible interval)9, 13 ± 2 meters44, and 8.5–11 meters (99% credible interval)45 above present. These estimates partition contributions between Greenland and Antarctica under the assumption that the LIS and EIS had terminated earlier in the interglacial. Underpinning this inference is the argument that GMSL stability from 410 to 401 ka as identified from the benthic oxygen isotope stack (e.g., ref. 46)—implies an absence of interglacial ice volume changes. This argument enabled ref. 9 to correct Bermudan and Bahamian sea-level indicators for GIA by assuming that these sites, on the LIS peripheral bulge, would have subsided throughout the interglacial.

The assumption that constant GMSL implies no interglacial volume changes is undercut by ice-sheet asynchrony during Plio-Pleistocene interglacials36,47,48,49,50. It is instead possible that stable GMSL masked asynchronous ice sheet minima that largely balanced each other out. Interglacial collapse asymmetry and delayed disappearance was tested by ref. 45 for the MIS-11 Greenland and Antarctic ice sheets, but not for the LIS. However, lead isotope analysis of Labrador Sea sediments suggests LIS persistence past ~ 410 ka and that MIS-11 GMSL may have risen slowly in part because of delayed LIS disintegration51. Updating MIS-11 LIS disappearance timing would likely lead to revisions in existing estimates of MIS-11 GMSL9,45.

Such a revision would not immediately affect future AIS projections because these projections do not currently use MIS-11 ice volumes as a paleo-target, likely because of the dearth of ice volume constraints. In the past decade, however, constraints on MIS-11 Greenland climate and ice extent from cosmogenic52,53 and palynological54 sources have accumulated, and modeling efforts are employing these bounds to quantify GrIS volume55,56. It is therefore likely that projections will, in the future, add MIS-11 to the catalog of paleoclimate calibration targets. When they do, it will be important to account for the possibility of NAIS persistence into MIS-11.

Earlier Pleistocene interglacials

Prior to MIS-11, direct proxy evidence for or against NAIS persistence during interglacials exists but is piecemeal. For instance, glacio-volcanic deposits from British Columbia indicate that a continent-scale CIS existed soon after the peak of the MIS 15 interglacial (598 ± 7.5 ka)57. In addition, coupled climate-ice sheet experiments suggest that a CIS up to ~8 meters GMSL equivalent in size persisted into MIS-13 (530-480 ka)58. Nevertheless, as with the later Pleistocene, indirect evidence is also accumulating that more early-mid Pleistocene interglacial ice existed on North America than once thought. Integrated cosmogenic nuclide concentrations in Labrador Sea sediments are consistent with the LIS fully disappearing only during MIS 5e, 9, and 11 and otherwise persisting in some form through the remaining interglacials of the last million years59. This result is supported by terrestrial cosmogenic exposure records near the centers of LIS loading60 and from indications that the LIS rapidly regained mass after the LIG ended61. Both findings suggest that complete LIS deglaciation was rare—a conclusion bolstered by the scarcity of periods with no LIS-sourced iceberg rafted debris in North Atlantic sediment cores62.

Should future ice sheet projections wish to use earlier interglacials as calibration targets, more direct constraints on NAIS size will be needed. A relatively untapped area of these constraints are the pockets of buried glacial ice that pepper Canada. Though the hottest periods of the mid-Holocene may have degraded Pleistocene permafrost deposits in the western Arctic, mean annual air temperatures in eastern Canada near freezing throughout the Holocene63 could have preserved relict Pleistocene ice masses. These cryospheric time capsules could have survived multiple glacial cycles even when buried beneath LIS ice: cold-based glacial conditions may have minimized mid-late Pleistocene erosion of Arctic Canadian landscapes64. Cold temperatures and minimal erosion across the Pleistocene together raise the prospect that the buried ice found in the North American high Arctic65 that dates to > 700 ka is part of a trove of early-mid-Pleistocene glacial ice and paleo-permafrost deposits66. These deposits would hold clues about NAIS persistence during past warm periods and should be targeted for future work.

Pliocene

While North American glaciations intensified only after 2.7 Ma, evidence continues to emerge for episodes of significant Northern Hemisphere ice earlier in the Pliocene, including on North America. At ~ 4 Ma, iceberg rafted debris (IRD) records suggest that enough ice covered Greenland, Scandinavia, and Alaska to reach the coast67, while glacial tills from the Hudson Bay lowlands place initial LIS growth as early as 3.5 Ma68. IRD pulses in the North Atlantic during Marine Isotope Stage M2 (~ 3.3 Ma) indicate expansion of the Greenland, Iceland, and Scandinavian ice sheets (e.g., ref. 69), consistent with coupled climate-ice sheet simulations suggesting 40-60 m of GMSL drawdown70,71. Terrestrial records from Alaska and northwest Canada reveal a robust CIS and Brooks Range glaciation during the late Gauss polarity Chron (3.05 − 2.60 Ma72,73,), while rare early Pleistocene (~ 2.4 Ma) dates on LIS glacial tills at 39oN13 do not preclude less extensive LIS advances prior to 2.6 Ma.

Taken together, these observations suggest that North America’s gradual cryospheric expansion into the Pleistocene likely significantly impacted sea level. If this expansion aligned with Antarctic mass loss episodes, the muted GMSL fluctuations in contemporaneous oxygen isotope records could be obscuring larger ice-sheet fluctuations, as likely occurred for GrIS and AIS47,50,74.

Implications for future climate projections

It has been the general practice in many paleoclimate simulations to assume that when past GMSL is above present, the LIS is gone and that only the GrIS and AIS are influencing the Earth system, including by changing sea level. For the LIG, this assumption was not enforced by ref. 75, whose ice sheet ensemble permitted LIS melt until ~126 ka, but has been a standard assumption since then. For instance, refs. 76 and 77 used an ensemble of GIA models produced using ice sheet scenarios in which the LIS disappears by 128 ka. Similarly, ref. 78 assume that the LIS is not present when peak LIG GMSL occurs, and the LIG sea-level budget proposed in ref. 79 also assumed the LIS was absent.

However, this assumption may not hold for past warm periods. Evidence is consistent with the LIS being present during the early portions of the LIG (i.e., prior to 126.5 ka), MIS-11 (i.e., prior to 405 ka), and possibly other warm periods, which may have masked loss from other ice sheets such that peak interglacial GMSL would not be a reliable indicator of total ice sheet contribution. This means that the AIS may be more sensitive to past warming than previously interpreted47,80. AIS sensitivity to past warming is important to know because some AIS models that run into the future get tested by their ability to reproduce ice sheet mass loss dynamics during Plio-Pleistocene interglacials as inferred from proxies35,36,81.

The assumption of LIS absence from prior interglacials has influenced future projections in two ways. First, some models that project future AIS mass loss (e.g., refs. 36,81) have been tuned to hit LIG mass loss targets based on sea-level reconstructions that assumed an absent LIS. Hence, those targets and the resulting future AIS projections may be biased towards lower AIS contributions. Second, forward simulations of LIG cryospheric evolution use climate forcing that itself assumes pre-LIG LIS termination. For instance, ref. 82 fed their LIG AIS models atmospheric and oceanic boundary conditions produced with an Earth System model forced by ice geometries from the last deglaciation, which reach present-day conditions before the LIG begins.

Including LIS persistence during past interglacials would have several effects on future climate projections. The most obvious effect comes from the calculation used to determine AIS melt based on peak GMSL. Both for the LIG (e.g., refs. 12,79, Fig. 3a) and for earlier warm periods10, AIS melt is routinely calculated as the residual when peak GrIS melt, mountain glacier melt, and thermosteric expansion are subtracted from peak interglacial GMSL – a technique hereafter referred to as the ‘subtraction method’ (e.g., ref. 79). The calculation of the subtraction method hinges on simultaneous peak melt contributions, which likely did not occur either during the LIG48 or during the Pliocene50,83—and on the absence of ice on North America. Nevertheless, AIS melt contribution assessments produced using the subtraction method generally do not account for the time-dependence of the AIS/GrIS tradeoff, in part because of uncertainties about individual ice sheet timings, in part because differences between the early- vs. late-LIG GrIS contribution may be modest (< 2 meters GMSL equivalent) and smaller than uncertainties from other sources.

Fig. 3: Schematic diagram of effect that Laurentide Ice Sheet (LIS) persistence into the Last Interglacial (LIG) would have on LIG sea-level budget.
Fig. 3: Schematic diagram of effect that Laurentide Ice Sheet (LIS) persistence into the Last Interglacial (LIG) would have on LIG sea-level budget.
Full size image

a–d Scenarios with high [low] peak GMSL estimates in which the LIS (a, c) disappears prior to the LIG or (b, d) persists into the interglacial. Black lines with uncertainty bars indicate examples of peak LIG GMSL (e.g., 7.6 m ± 1.7)11. Blue/green envelopes mark Antarctic/Greenland Ice Sheet (AIS/GrIS) contribution; purple envelope shows extra AIS contribution possible with > 3 meters of LIS volume in the early LIG22,51. Gray bar marks the combined contribution of thermal expansion and mountain glaciers, estimated at ~ 1 meter following ref. 11. Colored area below the y = 0 axis is equal to the persistent LIS contribution. Note that these illustrative scenarios assume constant GMSL from 128 to 120 ka so as to emphasize the effect of LIS persistence; paleo-constraints allow for GMSL variation during the LIG.

Last interglacial

With three or more ice sheets during interglacials, however, pinning down the timing of each ice sheet’s maximum GMSL contribution becomes crucial—both for top-down assessment of ice volumes based on modeled GMSL and for bottom-up assessment of GMSL via ice-volume histories. For instance, if peak AIS melt occurred early in an interglacial, LIS persistence would require more AIS melt to reach the same peak GMSL. This effect, illustrated in Fig. 3 for the LIG, applies regardless of peak GMSL estimate. With the 7.6 ± 1.7 meter estimate of ref. 79, LIS persistence could enable an Antarctic contribution of > 8 meters (Fig. 3b) if the excess contribution is assumed to not come from Greenland—which approaches the maximum likely AIS contribution as assessed by most sources (Fig. 1c). With the 1.2 − 5.3 meter (95% credible interval) maximum LIG GMSL of ref. 76., LIS persistence could permit 4-5 m of AIS melt. Furthermore, in the lower GMSL scenarios of refs. 76,77, pre-LIG LIS disappearance would require only minimal (< 2 m) AIS contribution, which disagrees with most sources (Fig. 1c). In summary, any LIG LIS persistence would require increased contribution from elsewhere, in this case, likely the AIS. Hence, the presence of NAIS ice may be masking the true sensitivity of the AIS.

The importance of LIS persistence coinciding with Antarctic melt additionally depends on when LIG GMSL exceeded present levels. When combined with LIS persistence and the late-LIG timing of Greenland excess melt, earlier GMSL exceedance enhances the likelihood of excess Antarctic melt, while later GMSL exceedance attenuates it. Reconstructions differ on exact LIG GMSL exceedance timing, but recent work has placed exceedance prior to 128 ka (P > 0.95)42,48,76,77,84. Future LIG work that more conclusively constrains GMSL exceedance timing will have the added benefit of clarifying the importance of LIS persistence.

LIS LIG persistence could also affect AIS and GrIS dynamics via climate mechanisms. Increased AIS melt could have occurred with a persistent LIG LIS via an oceanic teleconnection82. Oceanographic evidence suggests that pulses of cold glacial water, likely of LIS origin, may have affected North Atlantic circulation into the early LIG85. North Atlantic circulation modifications of this kind would have warmed the Southern Ocean via the classic Bipolar Seesaw. Southern Ocean warming in turn likely influenced Antarctica by encouraging intrusion of circumpolar deep water onto the continental shelf86. This mechanism may have driven AIS change throughout the Penultimate deglaciation and into the early LIG82,87. A persistent LIS might also have affected GrIS melt via a climate mechanism, since LIS topography appears to raise Northern Hemisphere Arctic air temperatures by encouraging atmospheric stationary waves that increase poleward energy flux over Greenland via jet stream perturbations88.

Whether through climate teleconnections or the changed sea-level budget of a three or four-ice-sheet interglacial, LIS LIG persistence would favor AIS melt scenarios on the high side of the range (3.5–7.4 m GMSL) used by ref. 81 to tune their future AIS projections. We quantify the impact of LIS persistence on future projections using the approach of ref. 80, who built a statistical emulator to relate LIG simulations to 21st century AIS projections. The simulations were produced by varying two variables: proportional sensitivity of model hydrofracturing, and maximum rate of ice cliff wastage. Ref. 80 produced the emulator by applying Gaussian process regression to the relationship between these variables and past and future ice volume (under a very high emissions scenario, RCP8.5). They then conditioned the emulated future projections on the peak LIG Antarctic contributions of refs. 48,75,81 and others to demonstrate how shifting paleoclimate targets could influence future projections.

We employ the ref. 80 emulator to test the effect of LIG LIS persistence (Fig. 4). With peak LIG GMSL below ~ 5.3 meters76, no LIS volume on land during peak LIG GMSL leads to future projections with highest likelihood of < 0.2 m of Antarctic contribution by 2100 and no likelihood of a contribution > 0.65 m (Fig. 4a). By contrast, with 6 meters of LIS volume concurrent with peak Antarctic melt, the range of possible AIS contributions by 2100 reaches ~ 0.83 m. That probability distribution is virtually identical to the projected AIS contribution under a high LIG sea-level scenario (7.6 ± 1.779) with no LIS grounded ice (Fig. 4b). Under this higher sea-level scenario, LIS volume of 6 meters during peak GMSL elevates the highest probability in AIS melt exceeding 0.7 meters by 2100 and precludes AIS projections below ~ 0.25 m.

Fig. 4: Influence of Last Interglacial (LIG) Laurentide Ice Sheet (LIS) persistence on projections of future Antarctic Ice Sheet (AIS) mass loss.
Fig. 4: Influence of Last Interglacial (LIG) Laurentide Ice Sheet (LIS) persistence on projections of future Antarctic Ice Sheet (AIS) mass loss.
Full size image

a, b Projections of AIS mass loss by 2100 under a very high emissions scenario (RCP8.5) for LIS of ≤6 meters GMSL equivalent for (a) low—1.2-5.3 meters, 95% credible interval76—and (b) high—μ = 7.6, σ = 1.679,—LIG GMSL scenarios. Projections follow ref. 79 in assuming that during peak GMSL, thermosteric and mountain glaciers contribute 1 ± 0.2 m and the Greenland Ice Sheet contributes 2 ± 1.5 m. Calculations are performed using the Gaussian process emulation framework of ref. 80.

These scenarios were produced assuming that the GrIS contributes 2 ± 1.5 m during peak GMSL and that thermosteric expansion and glaciers (TMG) contribute 1 ± 0.2 m79. Sensitivity tests of these variables (not shown) reveal that though the 21st century AIS projections vary depending on the absolute values of GrIS and TMG contribution, the trend towards higher AIS projections with more LIS persistence remains unchanged regardless of assumed GrIS and TMG contribution.

Earlier warm periods

During MIS-11, protracted LIS deglaciation would have slowed interglacial unloading of the North American continent, which would have expanded the LIS peripheral bulge and delayed peripheral bulge relaxation throughout the interglacial. Delayed peripheral bulge relaxation could have reduced the amount that the Bahamas and Bermuda subsided during early MIS-11, lessening the GIA correction that ref. 9 applied to the Bermudan and Bahamian sea-level indicators. This, in turn, would increase peak GMSL assessed from those indicators to higher than the 6 to 13 m range they infer. Rigorous estimation of this increase would require quantitative modeling outside the scope of this study. Nevertheless, based on the amplitude of intra-LIG RSL fall and rise modeled by ref. 41, we speculate that LIS persistence during MIS-11 could elevate the peak GMSL estimate between 0.5 and 2 m.

On the other hand, the effect of a delayed LIS disappearance on local sea level in South Africa, the subject of refs. 44,45, could be to decrease the amount of local sea-level fall, which would reduce the 8.5 − 11 m GMSL highstand inferred from those data. It is difficult to know exactly the effect that those two opposing corrections would have on a joint inversion for MIS-11 GMSL, and quantifying it with GIA modeling is outside the scope of this study. Still, because GIA effects grow larger with proximity to ice sheets, it is likely that the positive correction to the Bahamas and Bermuda data would outweigh the negative South Africa correction. If so, LIS persistence would push peak MIS-11 GMSL towards the upper end of the 6 − 13 m range, which would imply large GrIS and AIS contributions, as recent findings from both poles suggest occurred53.

Concerning the Pliocene, the full LIS was likely absent during Pliocene warm periods. Nevertheless, assumptions about MPWP synchrony between AIS, GrIS, and/or CIS can undermine the use of Pliocene sea level and ice sheet reconstructions in future projections. For example, in assessing peak MPWP GMSL, ref. 10 follow ref. 50 in assuming that the GrIS’s peak MPWP contribution was 5 ± 1 meters GMSL equivalent during the two largest MPWP sea-level highstands: MIS K1 (3.082-3.038 Ma) and MIS KM5c (3.224-3.126 Ma). However, the modeled AIS and GrIS contributions in ref. 50 are asynchronous. As a result, peak GrIS contributions during MIS KM1 and MIS KM5c coincide with AIS contributions ~ 2 m below their peak MPWP values; and outside of the MIS KM1 and MIS KM5c highstands, peaks in MPWP AIS sea-level contribution can correlate with GrIS contributions as low as 2 m.

The assumption of considerable GrIS melt during MPWP highstands led ref. 10 to revise the AIS projections of ref. 81 from 0.36 ± 0.16 m to 0.07 ± 0.02 m for 2100 and from 9.94 ± 2.97 m to 5.17 ± 3.53 m for 2300. However, if MPWP ice-sheet volumes were more strongly antiphased than implied by ref. 50, AIS minima could have coincided with smaller GrIS contributions89,90, resulting in larger AIS contributions to MPWP GMSL maxima83 and—if it led to GrIS volumes equivalent to modern values during those times—implying that the high-end projections of ref. 81 should remain viable.

Moving forward

Paleoclimate investigations fall along a spectrum between the view that present Earth system events and processes adequately represent the past, and the view that some past events and processes have no present analog. Though paleoclimate research has often treated past interglacials as similar to our current two-ice-sheet world—if not in climate forcing then at least in ice sheet configuration9,91—we argue that this framing needs revision. During the Holocene, the LIS persisted for ~ 3500 years after the Northern Hemisphere reached interglacial warmth92. During the LIG, mean top-of-atmosphere summer (JJA) isolation at 60 °N surpassed early Holocene (10 ka) insolation at 129.5 ka93, 3000 years prior to the 126.5 ± 1 ka date of putative LIG LIS disintegration and under approximately similar atmospheric greenhouse gas concentrations to the pre-industrial Holocene. The symmetry in this delay is unsurprising: in idealized simulations, large ice sheets consistently lag several thousand years behind insolation and surface air temperature throughout the Quaternary94.

Had later LIS advances not scraped North America clean of traces of ice sheet persistence during earlier warm periods, this feature of icehouse Earth might be part of the accepted paradigm. As it is, the evidence of Pleistocene interglacial persistence is subtler than the Holocene evidence. It is hard to ignore the Holocene moraine belts that rib Canada. It is easier to overlook marine sedimentary evidence of interglacial LIS proglacial lake outburst floods, cratonic erosion, freshwater plumes, and iceberg rafting. Nevertheless, now that this subtler evidence has accumulated, the paleoclimate community should treat LIS regularly persisting far enough into Quaternary interglacials to coincide with when GMSL has exceeded present levels as a realistic scenario to be considered.

To close, we propose concrete research directions to help the paleoclimate community pivot towards the paradigm of NAIS interglacial persistence.

First, recent ocean and atmosphere simulations have modeled the LIG using ice configurations in which the LIS disappeared before 127 ka in part because the Paleoclimate Modeling Intercomparison Project (PMIP) endorsed this timing4,95 and made the recommended ice histories available. The next PMIP, for which planning is underway, should consider sensitivity tests in which the North American ice history for the penultimate deglaciation persists deeper into the LIG. We also suggest that these tests should include scenarios in which the ice that persists until 126.5 ka has a size similar to the LIS at 8.2 ka. This configuration is implied by the timing of the deposition of the red sediment layer that represents an outburst flood as the LIS broke apart into separate domes. And whether or not the PMIP protocols are revised, we recommend that future work that uses the existing PMIP protocols should at a minimum employ the recommended LIG ice sheet geometries from ref. 95 rather than adopting pre-industrial configurations (sensu ref. 96).

Second, the continuing spread of estimates for LIG ice sheet contributions to GMSL attests to the limits of inferring ice volume changes from GMSL data and other indirect proxies. Direct nearfield evidence of the timing and magnitude of ice sheet extents and volumes has no substitute. Promising results from lake drilling23,97, paleo-permafrost studies (e.g., ref. 65), marine sediment cores22,51, and marine geophysics98 indicate that these and other proxies may hold the key to confident reconstruction of Quaternary LIS persistence during past warm periods. They also hold the key to understanding how LIS and EIS volumes during peak glacials, which may influence their respective likelihoods of enduring into subsequent interglacials, depend on the sequence of ice sheet nucleation during early glacials99. Additional near-field data from Antarctica, either from sedimentary38 or genomic37 sources, would yield further knowledge advances. Future efforts should aim at collecting more of these data.

Third, ice simulations require validation by collections of geologic proxies that are standardized, assigned robust uncertainties, and have unified age models. Production of such databases has slowed even as funding agencies increasingly support science that uses legacy data. Encouraging production of more such databases will enable intercomparison of data, which will support future work to characterize NAIS dynamics during past interglacials.

Lastly, differences in deglacial ice histories are important in comparing past to future warm periods. The differences we trace between past and present interglacials undermine the accuracy of attempts to connect paleo-GMSL estimates directly to estimates of long-term future GMSL commitment as a function of warming level, as was done in the Intergovernmental Panel on Climate Change’s Sixth Assessment Report1. By contrast, past LIS masking of AIS sensitivity implies that the AIS potential to retreat rapidly under future warming scenarios may be underestimated. In that light, the paleoclimate community should focus attention on quantifying rates, amounts, and timings of individual ice sheet mass loss during past warm periods, as progress on this front will advance understanding of how Earth’s remaining ice sheets may respond to warming.