The observed demise of Arctic sea-ice in recent decades1 has had major impacts on the climate, the ecosystem, the livelihood and cultural heritage of Indigenous peoples, and global geopolitical interests2,3,4. As anthropogenic warming continues, climate models suggest that Arctic multi-year ice (MYI) will survive the longest along the northern coastline of the Canadian Arctic Archipelago and Greenland5,6,7,8,9,10,11. Today this region, also known as the last ice area (hereafter referred to as LIA), hosts the thickest and oldest sea-ice in the Arctic Ocean12,13. This is due to cold temperatures and the main Arctic Ocean surface circulation, dominated by the Beaufort Gyre and the Transpolar Drift, transporting sea-ice toward the LIA13,14. As a result, the LIA is perennially covered by a mixture of locally formed and transported first year ice (FYI) and MYI with ages greater than 5 years12. The dominant source regions of MYI in the LIA today are traced to the Eurasian and Amerasian shelves15.

Within the LIA, between northeastern Ellesmere Island and Greenland, lies the Lincoln Sea (Fig. 1), one of the least studied marginal seas of the Arctic Ocean16. In recent years, more and more evidence for a changing sea-ice cover in the Lincoln Sea region has emerged, questioning its long-term stability under a continued anthropogenic warming. Prolonged periods of open water are reported from east of the Lincoln Sea and north of Ellesmere Island17,18 and increased ice motion19 and loss of the oldest MYI12 are associated with recurrent failure of the Nares Strait ice arches20,21,22. These phenomena are likely related to both physical and thermodynamic processes, where anomalous atmospheric circulation and break-up and export of sea-ice by winds17,21 causes an increase in the open water and thin ice area fractions enhancing ice melt17.

Fig. 1: Sea-ice concentration trend in the Arctic Ocean and core locations in the Lincoln Sea.
figure 1

a Map of the Arctic Ocean including the annual sea-ice concentration trend from 1979 to 201894 (and references therein) and the major surface circulation features, the Transpolar Drift (TPD) and Beaufort Gyre (BG) (dashed blue arrows). The dashed pink line outlines the region of the last ice area (LIA) that lies within the Arctic Ocean proper. The trend in sea-ice concentration is shown as % per decade, i.e. the decrease/increase of sea-ice concentration per 10-years for the period from 1979 to 2018. b Close-up of the Lincoln Sea (yellow rectangle in a) including locations of core sites Ryder19-12-GC1 (12-GC) and Ryder19-14-GC1 (14-GC).

As climate model projections for the 21st century agree that the LIA will host the last remaining summer sea-ice in the Arctic6,7,8,9,10,11,23,24, by first principle, a transition to seasonal sea-ice in the LIA would be indicative of a sea-ice regime shift in the entire Arctic Ocean. Applying this principle to previous warm periods in the geological past can thus elucidate the resilience of the Arctic Ocean sea-ice cover to climate change.

In 2019, the Swedish icebreaker Oden set out on the Ryder 2019 Expedition to map and retrieve marine sediment cores from the uncharted Sherard Osborn Fjord and southeastern Lincoln Sea (Fig. 1). Here we examine Holocene sea-ice dynamics based on two gravity cores (Ryder19-12-GC1, Ryder19-14-GC1; hereafter referred to as 12-GC and 14-GC) and 15 multi-core surface samples from the southern Lincoln Sea (Fig. 1). The age models are based on lithological investigations and radiocarbon dating, indicating that the entire Holocene period was recovered. The Early Holocene, characterized by a peak in boreal summer insolation25, represents the most recent period in geological history when temperatures were similar to or possibly exceeded modern observations26. Although the causes of modern warming differ, similar climate boundary conditions suggest that the Early Holocene may be a suitable analogue to understand the configuration of the Arctic climate system in a warmer future. Using a coupled atmosphere-sea-ice-ocean column model, Stranne et al.27 suggest the possibility of an Arctic Ocean-wide transition to a seasonal sea-ice regime during the Early Holocene, supported by data from the Eurasian and Amerasian Arctic shelves28,29,30, the Beaufort Sea31, the Greenland Sea32,33,34,35,36, and the central Arctic Ocean37,38. Within the LIA, reduced MYI during the Early Holocene is supported by coastal records of beach ridges and driftwood from northern Greenland and Ellesmere Island39,40. However, no data exist from the marine realm, critical to assess whether the change to a seasonal sea-ice regime in the Early Holocene impacted the LIA and by extension the entire Arctic Ocean.

Results and Discussion

Holocene sea-ice dynamics in the Lincoln Sea

Our reconstruction of Holocene sea-ice dynamics in the Lincoln Sea is based on the analysis of sympagic (IP25, HBI II) and pelagic (brassicasterol) biomarkers41 and the amount of total organic carbon (TOC). Changes in the relative concentration of the individual biomarker groups are indicative of changes in ice algal versus open water habitats through time41, while TOC provides more general information on the accumulation of organic matter in the sediments. In surface sediments underlying the pack ice area of the central Arctic, biomarker concentrations and TOC are low/absent42 as primary productivity is restricted by the availability of nutrients and photosynthetically active radiation43. In general, sympagic biomarker concentrations are highest in surface sediments in regions of seasonal sea-ice and along the ice edge, while they are absent in ice-free regions42. The concentration of brassicasterol is elevated in the seasonal ice zone, as well as in ice-free regions in line with a broad pelagic source42.

Accordingly, elevated concentrations of both sympagic and pelagic biomarkers in 12-GC and 14-GC between approximately 11.3 ± 0.6 cal ka BP and 9.7 ± 0.6 cal ka BP indicate a period of seasonal sea-ice during the Early Holocene (Fig. 2, Supplementary Fig. 1). The close correlation of IP25 and brassicasterol throughout this interval suggests that the presence/absence of sea-ice governed both sympagic and pelagic productivity, as is expected in the marginal ice zone or environments with relatively short-lived ice-free periods during summer42,44,45. This is supported by the PBIP25 index (Fig. 2) with a mean value of 0.5 (n = 33), considered to reflect a seasonally ice-free/marginal sea-ice environment. After ~9.7 ± 0.6 cal ka BP, a drop in all biomarker concentrations (Fig. 2) and a decrease in the amount of TOC (Fig. 2) suggests the (re-)establishment of a perennial sea-ice cover in the southern Lincoln Sea. Throughout the mid-to-late Holocene, biomarker concentrations remain low, while TOC increases again to intermediate levels after ca. 8 cal ka BP (Fig. 2). This is in line with a more persistent sea-ice cover compared to the Early Holocene. However, short-term (centennial) variability cannot be resolved throughout this interval due to the relatively lower sedimentation rates (1–40 cm/kyr) compared to the Early Holocene (50–125 cm/kyr). In the topmost samples of both cores (5.5 cm/~2 cal ka BP and 1 cm/~1 cal ka BP in 12-GC and 14-GC, respectively) the concentrations of sympagic and to a lesser extent pelagic biomarkers increase once again (Fig. 2). Due to the low sedimentation rates and bioturbation, these values are likely influenced by modern sea-ice variability and are consistent with the relatively high biomarker concentrations in the 15 surface samples from the region (Supplementary Fig. 2) (Supplementary discussion 2.2). The surface samples yield PBIP25 values of 0.6–0.8 (Supplementary discussion 2.2) (Fig. 2, Supplementary Fig. 2), which are slightly lower than during the mid-to-late Holocene and indicative of seasonal to perennial sea-ice. The decrease in PBIP25 likely reflects recently observed reductions in the LIA sea-ice concentration17,18,20,46 (Figs. 23) and the beginning of the transition towards a more variable/seasonal sea-ice regime in the southern Lincoln Sea.

Fig. 2: Sea-ice biomarkers and TOC from two sites in the Lincoln Sea alongside regional atmospheric temperatures across the Holocene.
figure 2

a PBIP25 in Ryder19-12-GC1 (light blue diamonds) and Ryder19-14-GC1 (black diamonds). b IP25 (green), brassicasterol (yellow), and TOC (orange) in Ryder19-12-GC1. c IP25 (green), brassicasterol (yellow), and TOC (orange) in Ryder19-14-GC1. d Mean annual air temperature anomaly (MAAT) at Agassiz ice cap compared to pre-industrial (PI; Supplementary Methods 1.2) (black) including the 2σ uncertainty envelope (grey)61. MAAT was reconstructed based on the oxygen isotopic signature (δ18O) in the ice61. Additionally the mean of the Early Holocene time period with evidence for seasonal sea-ice (dashed pink) with the corresponding 95% confidence interval (light pink) is shown. e Reconstructed June-August temperatures at Agassiz ice cap (black) based on the melt layer record including the 2σ uncertainty envelope (grey)61 and the mean of the Early Holocene time period with evidence for seasonal sea-ice (dashed pink) with the corresponding 95% confidence interval (light pink). f 21st of June insolation at 80°N95 (blue). The stars above the horizontal dashed line show PBIP25 (light blue, black), IP25 (green), brassicasterol (yellow), and TOC (orange) in the surface samples of Ryder19-12-MC1-8 and Ryder19-13-MC1-5 (Fig. 4). The horizontal light yellow bar marks the Early Holocene interval characterized by seasonal sea-ice in the southern Lincoln Sea.

Fig. 3: Modern sea-ice, mean annual and summer temperatures in the southern Lincoln Sea region.
figure 3

a Annual mean sea-ice concentration in the southern Lincoln Sea (grey) including the LOESS (locally weighted scatterplot smoothing) smooth (black) and the 95% confidence interval (dashed black) from the Pan-Arctic Ice Ocean Modeling and Assimilation System (PIOMAS), 1978–202165. b JJA and mean annual temperatures at a location proximal to 12-GC and 14-GC (82.5 °N, 55 °W, 0 m MSL) from the ERA5 reanalysis (2 m atmospheric temperatures)64 (light grey) and the respective LOESS smooth (black) including the 95% confidence interval (dashed black), 1979–2021. c JJA and mean annual temperatures proximal to the Agassiz ice cores (80.75 °N, 73 °W, 1740 m MSL) from the ERA5 reanalysis (2 m atmospheric temperatures)64 (light grey) and the respective LOESS smooth (black) including the 95% confidence interval (dashed black), 1979–2021. The average of JJA temperatures for the Early Holocene interval characterized by seasonal sea-ice (dashed pink) and the 95% confidence interval (light pink) are included in the plot of JJA temperatures at the Agassiz ice cap.

Micropaleontological data from 12-GC and 14-GC support a shift from reduced sea-ice during the Early Holocene to a more persistent ice cover during the mid-to-late Holocene47 (Supplementary Fig. 3, Supplementary Fig. 4). Between ca. 11.3 ± 0.6 cal ka BP and 9.7 ± 0.6 cal ka BP the benthic foraminiferal species Cassidulina neoteretis, typically associated with bottom waters of Atlantic origin48, dominates the assemblage in both cores (Supplementary Fig. 3, Supplementary Fig. 4). While C. neoteretis also dominates during the mid-to-late Holocene, other species, such as Oridorsalis umbonatus, become more important (Supplementary Fig. 3, Supplementary Fig. 4)47. In the Arctic Ocean, O. umbonatus is characteristic of areas with a dense sea-ice cover and low export of organic carbon to the seafloor49. An increase in O. umbonatus in the southern Lincoln Sea around 9.7 ± 0.6 cal ka BP (especially pronounced in 14-GC, Supplementary Fig. 4) thus supports the re-establishment of a perennial ice cover around this time. Similar to the benthic foraminifera, the ostracod assemblage during the mid-to-late Holocene suggests the presence of a persistent ice cover (Supplementary Fig. 3, Supplementary Fig. 4)47. Although Polycope spp. is typically described as a genus associated with seasonal/marginal sea-ice environments50,51, it also occurs in surface sediments of the central Arctic Ocean under perennial sea-ice52,53 and may be associated with the upper Arctic Ocean Deep Water and/or lower Arctic Intermediate Water affected by waters of Atlantic origin53,54. In the southern Lincoln Sea, C. neoteretis bears witness to bottom waters of Atlantic origin throughout the entire Holocene, which might explain the presence of Polycope spp. during the mid-to-late Holocene (Supplementary Fig. 3, Supplementary Fig. 4). At the same time, the ostracod species Acetabulastoma arcticum is present in all samples younger than ~9.2 cal ka BP in 14-GC (Supplementary Fig. 4) and in one sample in 12-GC (ca. 6.2 cal ka BP)47. A. arcticum lives parasitically on the under-ice amphipod Gammarus, found exclusively in the brackish surface layer underneath pack ice52. Thus, it can be used as an indicator of a perennial sea-ice cover such as in the central Arctic Ocean today38 and does not occur on continental shelves characterized by seasonally ice-free conditions52. The exclusive (apart from one sample in 14-GC) presence of A. arcticum in samples younger than ca. 9.2 cal ka BP therefore supports the re-establishment of a perennial ice cover in the southern Lincoln Sea during the mid-to-late Holocene.

Other Arctic biomarker records that cover the Early Holocene are from the outer and inner northeast Greenland Shelf35,36, the eastern Fram Strait/Yermak Plateau33,34, the Barents Sea north of Svalbard32, the inner and outer Laptev Sea29,30,55, and the Beaufort Sea31 (Supplementary Fig. 5). All available records show a reduction in the sea-ice extent during the Early Holocene, with most records suggesting this decrease occurred earlier than in the Lincoln Sea (Supplementary Fig. 5). In the Beaufort Sea, reduced sea-ice occurs from the onset of the record at 14 cal ka BP31. In the inner and outer Laptev Sea, biomarker records suggest a reduction in sea-ice extent following the Younger Dryas around 11.8 cal ka BP29,30,55 and in the northernmost Barents Sea sea-ice retreats around 11.6 cal ka BP32 (Supplementary Fig. 5). Three records in proximity to the southern Lincoln Sea are from areas of perennial sea-ice cover in the northernmost Barents Sea32 (JR142-11GC, HH-15-06GC) and on the inner NE Greenland shelf (PS100/270-1), covered by the landfast Norske Øer Ice Barrier36. They show similar temporal characteristics to the southern Lincoln Sea, with increased biomarker concentrations during the Early Holocene and a drop in concentrations/fluxes at 9.1 cal ka BP and 9.6 cal ka BP in the northernmost Barents Sea and on the NE Greenland shelf, respectively32,36 (Supplementary Fig. 5). The re-establishment of a perennial ice cover in the Atlantic sector of the Arctic Ocean/northernmost Greenland Sea thus closely follows that of the southern Lincoln Sea around 9.7 ± 0.6 cal ka BP (Fig. 2, Supplementary Fig. 5). An earlier sea-ice retreat, as well as a later re-establishment of a perennial ice cover in regions outside the LIA agree with the LIA being the last (first) region of MYI in a warming (cooling) climate and suggest that the transition to seasonal sea-ice in the LIA during the early Holocene was equivalent to a regime shift in large parts of Arctic Ocean.

On the other hand, some contrasting sedimentary and micropaleontological evidence exists from the central Arctic Ocean. While de Vernal et al.56 suggest that perennial sea-ice persisted over the western and central Lomonosov Ridge during the Early Holocene, Hanslik et al37. and Cronin et al.38 suggest an interval of reduced sea-ice in the central Arctic Ocean from approximately 11.6–10 ka37. Due to low sedimentation rates in the central Arctic Ocean, however, the Early Holocene is typically only represented by 2–3 datapoints37,38,56, which might not resolve the sea-ice variability. If sea-ice persisted in the central Arctic Ocean during the Early Holocene, seasonal sea-ice in the southern Lincoln Sea was likely restricted to the near-coastal areas, associated with a thinner more mobile sea-ice cover and the breakdown of the landfast sea-ice cover. In this scenario, a thin ice pack may have survived in the central Lincoln Sea and parts of the central Arctic Ocean.

Mechanisms destabilizing the sea-ice cover in the Lincoln Sea across the Holocene

Following the last deglaciation, the Early Holocene was a time of rapid climate and environmental change. Rising summer insolation in the northern hemisphere resulted in Early Holocene warmth in the Arctic and subarctic realm57. Atmospheric warming was most pronounced during summer, with large parts of Greenland experiencing summer temperatures warmer than those of the mid-twentieth century by about 10 ka58. A review of available ice core records suggests that the northwestern and central Greenland Ice Sheet (GrIS) may have experienced summer temperatures 3–5 °C warmer compared to the mid-twentieth century, while the southern GrIS may only have experienced warming of up to 1–2 °C58. In addition to the spatial heterogeneity of Early Holocene warmth, the timing varies depending on latitude and in continental versus marine records59,60. In general, a Holocene Thermal Maximum (HTM) occurs earlier in high latitude marine records (11–7 ka) than at lower latitudes and in continental records (8–4 ka)60. Ice cores from the Agassiz ice cap on Ellesmere Island provide a record of Holocene mean annual air temperature anomalies (MAAT) (compared to PI; Supplementary Methods 1.2) and June-August temperatures (TJJA) in close proximity to the southern Lincoln Sea61. These demonstrate a regional HTM between 11 ka and 8 ka, during which temperatures regularly exceeded present-day values61 (Fig. 2). Peak Holocene temperatures at Agassiz ice cap thus occur earlier compared to ice core records from Greenland58, in line with an early HTM at higher latitudes60. Maximum Early Holocene temperatures at Agassiz ice cap are associated with the largest accumulation rate of pollen grains61, indicating a more productive terrestrial ecosystem in close proximity to the ice cap.

The regional HTM as recorded at Agassiz ice cap (11–8 ka) is concomitant with the evidence of seasonal sea-ice in the southern Lincoln Sea between approximately 11.3 ± 0.6 cal ka BP and 9.7 ± 0.6 cal ka BP (Fig. 2), suggesting that atmospheric temperatures likely played an important role for Early Holocene sea-ice conditions. This is supported by climate model simulations, which show that warming from increased radiative forcing may have been enough to shift the Arctic Ocean from a perennial to a predominantly seasonal sea-ice regime during the Early Holocene27. Other models, however, only show a reduction in sea-ice concentration in the Arctic marginal seas, while sea-ice thickness decreases everywhere in the Arctic when the Early Holocene radiative forcing is applied62. Thus, in addition to atmospheric warming other factors may have contributed to decrease the Early Holocene sea-ice extent in the southern Lincoln Sea. Biomarker-based sea-ice reconstructions from the Amerasian and Eurasian Arctic shelves indicate a reduction in sea-ice extent already between 14 ka and 11.8 ka28,29,30,63, earlier than in the Lincoln Sea (Fig. 2, Supplementary Fig. 5). Today, these regions comprise the dominant source areas of MYI to the Lincoln Sea15. Thus, an early decline in sea-ice extent likely caused a reduction in the net transport of sea-ice towards the LIA from as early as 14 ka. Other mechanisms, which might have contributed to a reduction in the sea-ice extent in the Lincoln Sea, include increased freshwater discharge from the GrIS and enhanced export of sea-ice through Fram Strait (Supplementary discussion 2.1).

Based on the LIA dataset and in combination with previously published records of Holocene sea-ice variability from across the Arctic and subarctic realm28,29,30,32,33,34,35,36,37,38,56 we thus argue that the Arctic Ocean likely transitioned to a predominantly seasonal sea-ice regime during the Early Holocene, while some areas in the central Lincoln Sea and adjacent central Arctic Ocean may have retained a thin ice pack. The early decline in sea-ice extent in the LIA MYI source regions and the associated reduction in net sea-ice transport towards the Lincoln Sea likely resulted in a thinner, younger, and more mobile ice cover, which made the LIA more vulnerable to atmospheric and oceanic forcing during the regional HTM resulting in the observed seasonal sea-ice signal.

This is consistent with modelling experiments of the MYI source regions to the LIA under future warming scenarios15. In a high warming scenario, corresponding to the representative concentration pathway 8.5, net sea-ice transport to the LIA ceases and the LIA transitions from predominantly MYI to FYI with ice free summers15. In a low warming scenario, consistent with a global warming threshold of 2 °C compared to the PI, on the other hand, the central Arctic Ocean becomes the dominant source region together with increased in situ formation of sea-ice15. Both scenarios are associated with a more mobile and thinner ice pack in the LIA compared to the historical record15.

Implications for future sea-ice dynamics in the Lincoln Sea

In the following section, we are discussing the implications of the early Holocene sea-ice conditions in the Lincoln Sea for the future stability of sea-ice in the LIA. Given the correlation of sea-ice and atmospheric temperatures during the early Holocene, we are focusing on atmospheric temperatures as the main driver of sea-ice change, not considering additional drivers such as changes in the Arctic Ocean surface circulation, effects of freshwater addition to the Arctic Ocean, and local sea-ice dynamics, including the formation of ice arches in Nares Strait.

The interval of seasonal sea-ice in the southern Lincoln Sea from ~11.3 ± 0.6 to 9.7 ± 0.6 cal ka BP was characterized by mean TJJA −5.1 ± 1.1 °C and MAAT (compared to PI; Supplementary Methods 1.2) of 3.6 ± 1.1 °C as reconstructed from the Agassiz ice cores61 (Fig. 2). Modern (2009 CE) MAAT at Agassiz ice cap are ~3.1 °C (compared to PI)61, while the reconstructed TJJA time series stops at 1934 CE. Thus, to further compare Early Holocene and modern temperature and sea-ice dynamics, the mean annual temperature (MAT) and TJJA from 1979–2021 were extracted from the ERA5 reanalysis64 (Fig. 3) at two locations corresponding to Agassiz ice cap (80.75 °N, 73 °W, 1740 m MSL) and sea-level in close proximity to the coring sites (82.5 °N, 55 °W, 0 m MSL) (Fig. 4). The annual mean sea-ice concentration (1978–2021) was extracted from the Pan-Arctic Ice Ocean Modeling and Assimilation System (PIOMAS)65 (Fig. 3). Sea-ice concentration data were averaged over an area corresponding to the southern Lincoln Sea (Fig. 4).

Fig. 4: Southern Lincoln Sea and northern Nares Strait including locations discussed in the manuscript.
figure 4

Surface samples (yellow symbols with area-coded shape) are numbered according to the respective multi-corer station. The locations of gravity core 12-GC and 14-GC are marked with red dots. Additionally the bathymetry (IBCAO v4)96 and velocity of the GrIS97 are shown and the major glaciers of the region and locations of presented temperature records (blue stars) are marked. The dashed black line marks the approximate boundary of sand-/siltstones and carbonates in the region surrounding Sherard Osborn Fjord75.

MAT and TJJA at Agassiz ice cap and in the southern Lincoln Sea are significantly correlated (MAT Agassiz ~ Lincoln Sea: R2 = 0.77, P < 0.001; TJJA Agassiz ~ Lincoln Sea: R2 = 0.70, P < 0.001) with a mean difference of 4.9 ± 0.7 °C and 6.9 ± 0.9 °C for MAT and TJJA, respectively. This suggests that both the MAT and summer temperature at Agassiz ice cap are representative of those in the southern Lincoln Sea. Between 1979 and 2021, MAT/TJJA show a significant warming trend of 0.5/0.6 °C (P < 0.001) per decade at Agassiz ice cap and 0.4/0.2 °C (P < 0.001) per decade in the Lincoln Sea (Fig. 3). Throughout the record, TJJA at Agassiz ice cap exceeded the Early Holocene mean of −5.1 ± 1.1 °C in 12 out of 43 years, with six occasions during the last decade (Fig. 3). This indicates that both the MAAT (3.1 °C as reconstructed from Agassiz ice core61) and TJJA (from ERA5) at Agassiz ice cap are approaching levels last seen during the Early Holocene.

Simultaneously, the mean annual sea-ice concentration in the southern Lincoln Sea65 decreased with −1.4 % per decade (P < 0.001) from 1979–2021 and with −2.7 % per decade (P < 0.001) from 2000 onwards (Fig. 3). This is consistent with the observed signs of a changing sea-ice cover in the wider LIA region during recent years17,18,21,46 and the decrease in PBIP25 in surface sediments compared to the mid-to-late Holocene (Fig. 2, Supplementary Fig. 2) (Supplementary discussion 2.2). Lincoln Sea MAT and TJJA and the annual mean sea-ice concentration show a significant negative correlation between 1979 and 2021 (MAT ~ sea-ice: R2 = −0.56, P < 0.001; TJJA ~ sea-ice: R2 = −0.62, P < 0.001) (Supplementary Fig. 6), supporting a prominent role of atmospheric temperatures (both MAT and TJJA) for sea-ice dynamics in the southern Lincoln Sea.

At the 21st conference of the United Nations Framework Convention on Climate Change (COP21), 196 parties adopted ‘The Paris Agreement’ to limit mean global warming to 1.5–2 °C (compared to PI levels; 1850–1900). As a result of Arctic amplification, a global mean warming of 2 °C compared to PI would result in a mean annual air temperature increase of 3–5 °C (compared to the PI) in the Arctic66. Different estimates exist for Arctic warming during the Early Holocene. Based on 16 terrestrial sites from the western Arctic, Kaufman et al.67 suggest that summer temperatures warmed on average 1.6 ± 0.8 °C (compared to the 20th century average). Climate model compilations suggest summer warming of up to 1 °C (compared to PI) between 30–90°N during the Early Holocene68, which compares to a (marine and terrestrial) temperature proxy stack for the same latitudinal band (relative to 1961–1990)69. Mean annual temperatures from model simulations, on the other hand, are colder than the PI reference until ~9 ka68. For central Iceland, Flowers et al.70 find a maximum Early Holocene warming of 3.5 °C (compared to 1961–1990). Similar to average MAAT (compared to PI) at Agassiz ice cap for the interval of seasonal sea-ice during the Early Holocene (~11.3 ± 0.6 to 9.7 ± 0.6 cal ka BP), which are 3.6 ± 1.1 °C (Fig. 2). This highlights the complexity of Early Holocene temperature reconstructions in the Northern Hemisphere, which likely arises from proxy biases and regional differences in both the timing and expression of Early Holocene warmth.

Nonetheless, an MAAT of 3.6 ± 1.1 °C at Agassiz ice cap between ~11.3 cal ka BP and 9.7 cal ka BP, falls within the range of projected mean annual Arctic air temperature anomalies (3–5 °C) for a global warming threshold of 2 °C (compared to PI). While these numbers are not directly comparable, as one is representative of the whole Arctic, while the other represents a regional signal over land at an altitude > 1000 m, MAAT during the Early Holocene in the wider Lincoln Sea region were most likely within the range of projected Arctic temperatures for a mean global warming of 2 °C (compared to PI).

Our results thus suggest that we can expect a transition to seasonal sea-ice in the southern Lincoln Sea for a global mean warming of 2 °C compared to the PI. Under high greenhouse gas emission scenarios (SSP2-4.6, SSP3-7.0, and SSP5-8.5), 2 °C warming (compared to PI) will likely be reached by the middle of this century71. While the Early Holocene record also demonstrates that a sea-ice regime shift in the southern Lincoln Sea is reversible within the reconstructed temperature range, unforeseen feedbacks may occur (e.g., ice arch failure in Nares Strait (Supplementary Discussion 2.1)). This highlights the importance of the 1.5 °C global warming target and the significance of adhering to ‘The Paris Agreement’ for the protection of the unique environments of the Arctic Ocean.


Site Ryder19-12-GC1 and Ryder19-14-GC1

Sites Ryder19-12-GC1 (82.58°N, −52.53°W, 867 m water depth, 273.5 cm sediment recovery) and Ryder19-14-GC1 (82.51°N, −54.45°W, 580 m water depth, 306.5 cm sediment recovery) are located in the glacially-carved through system that extends northward from Nares Strait and the northern Greenland fjords onto the Lincoln Sea shelf (Fig. 4). 12-GC is located north of Sherard Osborn Fjord, on glacial lineations coming out of Victoria Fjord, suggesting that the site is influenced by the activities of both Ryder Glacier, which drains about 1.6% of the GrIS, and C.H. Ostenfeld Glacier, which drains about 0.6% of the GrIS by area72. 14-GC, on the other hand, lies further to the West, north of St Georges Fjord where Steensby Glacier drains about 0.3% of the GrIS73 (Fig. 4). Under present day conditions, both sites are bathed in modified Atlantic Water at the bottom, with a cold and fresh mixed layer (core at 50 m) at the surface74.


In 12-GC four major lithological units (LU) can be distinguished, while 14-GC only recovered three of the four LU’s defined in 12-GC (Supplementary table 1). 12-GC ends in a glacial diamict (LU4), characterized by unsorted sediment and large clasts (Fig. 5). LU4 thus marks the retreat of grounded ice from the site. Above the diamict, distinctly laminated silty-clay characterizes LU3 (Fig. 5), suggesting sedimentation of glacial melt-water plumes in a proximal to distal glacial marine environment. In the bottom of LU3, laminae are irregular, with layers/lenses of coarser material (LU3b) (Fig. 5), suggesting deposition in a more proximal glacial environment. The same type of laminations is present in the bottom of 14-GC (Fig. 5). The transition from LU3b to LU3a around 202.5 cm in 12-GC and 277 cm in 14-GC, is marked by the onset of pronounced cross-cutting and less coarse laminae (Fig. 5). Reduced input of coarse material is in line with the deglacial retreat of local glaciers into the fjords75, while cross-cutting indicates the influence of currents on laminae deposition. This can be either currents related to meltwater discharge at the grounding zone of local glaciers or ocean currents over the Lincoln Sea shelf. Towards the top of LU3a the character of laminations changes to wavy laminae without pronounced cross-cutting in 14-GC and wavy to planar laminae in 12-GC (Fig. 5), in line with a transition from proximal to more distal glacial marine sedimentation.

Fig. 5: Lithology of 12-GC and 14-GC.
figure 5

Radiograph, core photo, XRF Ca/Ti (grey, 20-pt average in black) and bulk density (black) of (a) 12-GC and (b) 14-GC, including prominent sedimentary features (dropstones, cross-cutting, IRD/clasts). Key lithological boundaries can be correlated between 12-GC and 14-GC, for the respective depth of LU boundaries see Supplementary table 1. Uncalibrated radiocarbon ages and their respective depth are marked with yellow triangles. Red triangles mark outliers, which are excluded from the final age model.

Above LU3, both cores record a transitional unit (LU2) of silty clay with a lighter sediment color, high Ca/Ti ratios and increased bulk densities (Fig. 5). While Ca/Ti ratios in 12-GC increase steadily throughout LU3, the increase in 14-GC Ca/Ti is abrupt at the LU3/LU2 boundary (Fig. 5). During the deglacial and early Holocene the delivery of Ca-rich sediments to the fjords and Lincoln Sea shelf was controlled by the retreat of contributing glaciers behind the zone of Silurian sand/siltstones close to the fjord mouths and into the Cambrian/Silurian carbonates of the Franklinian Basin, characterizing the hinterland75,76 (Fig. 4). Differences in Ca/Ti in 12-GC1 and 14-GC1 throughout LU3 thus point to (partly) different sediment sources such as Ryder, Steensby, and C.H. Ostenfeld Glacier (Fig. 4). The transitional unit LU2 is a consistent feature in Holocene sediment cores from the area75,77,78. Inside Sherard Osborn Fjord, LU2 corresponds to laminated, meltwater/suspension settling derived Ca-rich sediments, consistent with glacial erosion of the surrounding platform carbonates in the hinterland of Ryder Glacier75. In 12-GC and 14-GC, higher bulk densities, slightly coarser material, and more mottled/lenticular sedimentation, indicate a higher energy depositional environment in LU2 compared to LU3. It is unclear, however, if this is related to oceanic currents, or regional glacier dynamics. LU1 can be divided in sub-facies LU1a and LU1b. LU1a is characterized by bioturbated homogenous sediments, while LU1b is characterized by a distinct brown color, accompanied by a drop in both XRF Ca/Ti and bulk densities (Fig. 5). LU1b is especially pronounced in 12-GC (Fig. 5). In the lower part of LU1, both cores show evidence for deposition of ice rafted debris (IRD). In 12-GC, IRD is constrained to the LU2/LU1 boundary, while it is evident from the base of LU1 to ca. 22 cm in 14-GC (Fig. 5).


The chronology of 12-GC and 14-GC is based on seven (Supplementary table 2) and five (Supplementary table 3) benthic foraminiferal radiocarbon dates, respectively, and has been developed further from Cronin et al.47. Radiocarbon dates were substantiated with two age-depth tie points based on macroscopic observations of lithological transitions and consideration of linear sedimentation rate patterns. Radiocarbon measurements were performed at the National Ocean Sciences Accelerator Mass Spectrometry (NOSAMS) facility at Woods Hole Oceanographic Institution, Massachusetts, USA. Three radiocarbon dates were excluded from the final age model. In 12-GC the radiocarbon date at 94 cm (OS-159561) suggests an age reversal (Supplementary table 2), similar to the radiocarbon date at 49 cm in 14-GC (OS-159564) (Supplementary table 3). Both of these dates are from the transitional unit LU2 (Fig. 6). The distinct change towards lighter colors in LU2 (Fig. 5) might suggest a different sediment source, with the older radiocarbon ages indicating re-deposition of previously deposited sediments. Additionally, the radiocarbon date from the core catcher (CC) of 14-GC (OS-152304) has been excluded, as it is younger than the preceding date at 180 cm (Supplementary table 3). A too young CC age, likely results from the coring process, where younger foraminifera are incorporated into the core catcher when the gravity corer penetrates the sediment.

Fig. 6: Age model and correlation of 12-GC and 14-GC.
figure 6

Core photos, schematic of major lithological units (as in Fig. 5), XRF Ca/Ti (black, 20-pt average in red-brown), and age-depth relationship of (a) 12-GC and (b) 14-GC, including the correlation between both cores (horizontal black dashed lines). The depth of individual radiocarbon ages is marked with triangles on the core photos and black dots on the age-depth relationship graph. Boundaries used for age modelling (see text) are marked by yellow crosses, age-depth tie points with green diamonds, and the top of each core is indicated with a pink star. The likelihood distribution (pink) is shown for outlier radiocarbon ages. Both age models include the mean (blue line) based on a ΔR of 300 ± 0 and the uncertainty envelope (light blue) based on a ΔR uncertainty of 300 (ΔR = 0 ± 0 and ΔR = 600 ± 0). The dashed blue lines indicated the 1σ uncertainty of the smallest/largest ΔR, respectively.

Radiocarbon ages were calibrated in OxCal v.4.479 using the Marine20 calibration curve80. Since the cores are in close proximity, radiocarbon ages were modelled in a combined approach using additional age-depth control points. The top of both cores was assumed to be modern, using a uniform distribution between 0 cal yrs BP and −70 cal years BP, for the age modelling. Additionally, two age-depth tie points were included in the model (Supplementary table 4). One at the top of the Ca/Ti-rich transitional unit and one at the LU3a/b boundary (Fig. 5). For both tie points, the age was extracted from 12-GC due to better radiocarbon age control and used as age control points in the 14-GC age model. Lastly, the top of LU3 at 104 cm and 61 cm in 12-GC and 14-GC respectively, was added as a boundary79 to the age model. Throughout LU3, 12-GC and 14-GC have similar sedimentation rates (0.08 cm/yr and 0.1 cm/yr, respectively), suggesting regionally consistent sedimentation patterns. However, while radiocarbon dates in 12-GC indicate a drop in sedimentation rate at the LU3/LU2 transition, in 14-GC this drop occurs already within LU3, possibly due to the lack of radiocarbon age control at the LU3/LU2 transition. Adding a boundary into the age model forces the drop in sedimentation rates to occur at the LU3/LU2 transition, as indicated by radiocarbon dates in 12-GC (Fig. 6).

There is no information on the marine reservoir correction (ΔR) for the Lincoln Sea, with published estimates from the wider northern Greenland region ranging from 0 to 770 years using the Marine09 and Marine13 calibration curves75. O’Regan et al.75 conclude that given the large uncertainty in published reservoir corrections, a ΔR of 300 ± 300 years is a suitable approximation for Lincoln Sea sites bathed in modified AW. Using Marine20, a ΔR of 300 ± 300 years encompasses previously published reservoir corrections, corresponding to a ΔR of 150–750 years when applying Marine09 or Marine13. Further, applying the same ΔR as O’Regan et al.75 allows direct comparison of the new data with previously published results from the Ryder 2019 research cruise.

Modelling the ages of both cores together, including a ΔR of 300 ± 300, allows the modelled ΔR to shift considerably (408 ± 12 (1σ), n = 10), resulting from the large uncertainty in the provided reservoir age. To avoid this shift, which is not based on available knowledge but merely a modelling artifact, three separate age-depth models with a ΔR of 300 ± 0, 0 ± 0, and 600 ± 0 were calculated, where the reservoir correction was subtracted from the 14C age prior to modelling, rather than including a ΔR term in the model. This keeps the reservoir correction fixed and provides an envelope of potential ages, based on the uncertainty in the regional reservoir correction. The final age model suggests that 12-GC covers the time span from the present to ca. 12.7 cal ka BP, while 14-GC reaches back to approximately 11.5 cal ka BP (Fig. 6).

Biomarker extraction and analysis

Sea-ice conditions in the southern Lincoln Sea were reconstructed based on the identification of source-specific highly branched isoprenoid (HBI) and common sterol biomarkers. Mono- and di-unsaturated HBIs, also called IP25 (Ice proxy with 25 carbon atoms) and HBI II, respectively, are sympagic biomarkers, produced by sea-ice dwelling diatoms during spring41,81. Their occurrence in sediments is thus used as an indicator of the presence of seasonal sea-ice in the overlying surface ocean. The source of HBI II is less constrained compared to IP25, although studies suggest co-production of IP25 and HBI II in the sympagic diatom species Haslea spicula, H. kjellmanii, and Pleurosigma stuxbergii var. rhomboids81. Typically, HBI II and IP25 co-vary in Arctic sediments with HBI II occurring in higher concentrations42. In addition to sea-ice, salinity might affect the production of IP2582. This is especially important in fjord systems where meltwater discharge might cause large changes in the surface ocean salinity82.

Alongside IP25 and HBI II, brassicasterol was analyzed. Sterols are common compounds in eukaryotic cell membranes, occurring in both marine and terrestrial organic matter, albeit in different relative concentrations83,84,85,86,87. In the marine realm, brassicasterol is typically derived from phytoplankton sources, including diatoms and dinoflagellates84,85,87.

Biomarkers were extracted from 6 ± 0.1 g freeze dried sediment. A procedural blank and reference sediment sample (~ 3 g) with known biomarker concentrations were added to every extraction batch (n = 10). Prior to extraction 0.1 µg of 9-Octylheptadec-8-ene (9-OHD) and 5α-androst-16-en-3α-ol were added to each sample, blank and reference sediment as internal standards for HBIs and sterols, respectively. The samples were extracted three times with dichloromethane/methanol (DCM/MeOH, 2:1 v/v) using sonication (15 min) on the first step and vortexing (1 min) on the second and third step. Following centrifuging the supernatant of each extraction step was transferred to a new sample vial and dried under a gentle stream of N2. The different lipid classes were separated according to polarity using silica column chromatography. Nonpolar lipids, such as IP25 and HBI II were eluted with hexane, while the more polar sterols were eluted with DCM:MeOH (1: 1, v/v) (~ 7 mL, each). Prior to analysis the sterol fractions were derivatized using N,O- bis(trimethylsilyl)trifluoroacetamide (50 µL, 70 °C, 1 h) and diluted with 0.5 mL DCM. HBI fractions were diluted with 50 µL hexane.

All samples were analysed at Aarhus University using an Agilent 7890B gas chromatograph (GC) fitted with an HP-5ms Ultra Inert column (30 m × 0.25 mm × 0.25 µm) coupled to a 5977 A series mass selective detector and equipped with a Gerstel multipurpose sampler. For GC–MS operating conditions see Supplementary table 5. The identification of individual lipids is based on the characteristic retention indices and mass spectra of the individual compounds in the reference sediment. In the samples, HBI concentrations were too low to derive a characteristic mass spectrum in GC-MS scan mode. Quantification is achieved by comparing the integrated peak area (PA) of the selected ion for each biomarker (Supplementary table 5) to the PA of the respective internal standard88 under consideration of an instrumental response factor (based on the reference sediment) and the mass or the total organic carbon (TOC) concentrations of the extracted sediment88 (Supplementary Methods 1.1).

IP25 and HBI II are present in the samples in very low concentrations. Thus, for quality control, only samples where both a distinct IP25 and HBI II peak could be identified were quantified. In all other samples, the concentrations of both biomarkers were set to zero, in this case meaning below the level of quantification.

IP25 and brassicasterol were combined in the PBIP25 index (Eq. 1), a semi-quantitative index of past sea-ice concentration89,90. PBIP25 varies between 0 and 1, with values close to 1 suggesting extensive to perennial sea-ice cover, while values between 0.2 and 0.6 have been suggested to reflect seasonal sea-ice conditions with various seasonal duration of the ice cover42.

$${P}_{B}{{IP}}_{25}={{IP}}_{25}/({{IP}}_{25}+\left({brassicasterol}* c\right))$$

TOC, TN and δ13Corg analyses

For the analysis of TOC, 20–30 mg of freeze-dried, homogenized sediment were folded into Ag cups, which were gently rewetted and treated with HCl vapor according to the method outlined in Harris et al.91. The decalcified sediments were then freeze-dried an analyzed on a Flash 2000 Thermo Scientific elemental analyzer coupled in continuous flow mode to a Thermo Delta V Advantage isotope ratio mass spectrometer at Copenhagen University. Soil standards (Elemental Microanalysis, Okehampton, UK) were used for elemental analyser mass calibration. As working standard for isotope ratio analysis pure gases of CO2 calibrated against certified reference materials of 13C-sucrose and 13C-benzoic acid (IAEA, Vienna, Austria) were used.

Reanalysis data

June-August temperature (TJJA) data between 1979 and 2019 are from the ERA5 reanalysis64 available from the Copernicus Climate Data Store (!/dataset/reanalysis-era5-single-levels-monthly-means?tab=overview). Data were extracted for the grid points closest to the Agassiz ice cores (80.75 °N, 73 °W, 1740 m MSL) and the core sites of 12-GC and 14-GC (82.5 °N, 55 °W, 0 m MSL). Monthly mean values of June, July and August were averaged for every given year. The shown data (Fig. 3) represent the 2 m air temperature over the land/ocean surface.

Annual mean sea-ice concentration data between 1978 and 2021 are from the Pan-Arctic Ice Ocean Modeling and Assimilation System (PIOMAS) available from the Polar Sciences Center at the University of Washington ( Data were averaged over the area shown in Fig. 4, with the corner coordinates given in Supplementary table 6. PIOMAS is a coupled ice ocean modeling and assimilation system, combining the Parallel Ocean Program (POP) ocean model with a 12-category thickness and enthalpy distribution (TED) sea-ice model with a viscous-plastic ice rheology65. The model grid is a generalized orthogonal curvilinear coordinate system with the northern grid pole located over Greenland (76 °N, 40 °W), giving the highest model resolution in the Greenland Sea, Baffin Bay, and the eastern Canadian Arctic Archipelago65. Sea-ice data are validated against buoy, submarine, and satellite observations65,92,93.