The study focuses on the Canadian Beaufort Sea, one of the marginal seas of the Arctic Ocean (Fig. 1), bounded by the glacially excavated Amundsen Gulf to the east, Mackenzie Trough to the west, and the Mackenzie River delta to the south (Carmack et al., 2004). The shelf is a large estuarine setting at the interface between the Arctic Ocean and the Mackenzie River (Omstedt et al., 1994) (Fig. 1). The Mackenzie River is a significant source of freshwater to the Beaufort Sea, with an annual water discharge of 316 km³ yr⁻¹ (Holmes et al., 2012) and is considered the largest Arctic river in terms of sediment flux (124–128 Mt yr⁻¹) (Stein et al., 2004). At the same time, permafrost coastal erosion adds another 8–9 Mt yr⁻¹ of sediment into the Beaufort Sea, including carbon and nutrients (Wegner et al., 2015). Surface water circulation in the Beaufort Sea is primarily characterized by the clockwise Beaufort Gyre, which drives offshore currents towards the west and traps the majority of the Arctic Ocean's freshwater in the Canada Basin (Serreze et al., 2006). There is also a eastward flowing shelf-break current at depths beneath 50 m, which transports Pacific Water (coming from the Bering Strait) along the slope (Pickart, 2004). Sea ice cover on the Beaufort Shelf north of the Mackenzie River Delta varies from year to year, but generally begins to form in mid-October, persisting until ice break up in April–May (Fig. S1 in the Supplement). During ice break up, an open water flaw lead occurs along the outer edge of the landfast ice establishing a spring marginal ice zone over the shelf.

Two sediment cores were analyzed in the study. They were recovered as part of the Permafrost Carbon on the Beaufort Shelf (PeCaBeau) project during the 4th Leg of the 2021 CCGS Amundsen expedition (Bröder et al., 2022, Fig. 1). At station PCB09 (71.1° N, 135.1° W) at a water depth of 675 m on the Beaufort shelf slope, a piston core (PC, length of 420 cm) and multicore (MC, 30 cm) were retrieved (Fig. 1). At station PCB11 (70.6° N, 136.0° W) on the outer Beaufort shelf (74 m water depth) a giant gravity core (GGC, length of 290 cm) and MC (32 cm) were recovered (Fig. 1). PCB09 is found within the modern Atlantic bottom water mass, while PCB11 lies within the Pacific summer water (Fig. S2), the water masses were defined as in Matsuoka et al. (2012). The core tops (0–1 cm) from 22 multicores collected during PeCaBeau, were used to ground truth the hydrogen isotope ratio of C_16:0 fatty acid proxy for reconstructing salinity and test the applicability of the SST reconstructions (Fig. S1).

The age models of cores PCB09 (Fig. 2a) and PCB11 (Fig. 2b) show that they cover the last 13350 ± 690 and 9120 ± 720 cal. yr BP, respectively. PCB11 had a mean sedimentation rate of 35 ± 10 cm kyr⁻¹ with slightly higher sedimentation rates in the Late Holocene (< 4 ka). Conversely, PCB09 had an average sedimentation rate of 50 ± 27 cm kyr⁻¹, with substantially higher sedimentation rates before the Late Holocene.

The upper 300 cm of PCB09 (0–11 ka) display a gradual downcore increase in bulk density, reflecting porosity loss in largely homogenous silty-clay sediments (Fig. 2a). Below 300 cm, slightly higher variability in the bulk density, elevated magnetic susceptibility and higher Zr / Rb all point towards a transition to slightly coarser-grained sediments. Ca / Ti tended to increase downcore becoming more variable below 105 cm (7.6 ± 0.6 cal yr BP). There was a notable stepwise increase in Ca / Ti at 345 cm (11.7 ± 0.4 ka, Fig. 2a). Discrete peaks in Zr / Rb and bulk density, indicative of sediment coarsening, co-occurred with elevated Ca / Ti ratios at depths of 276 cm (10.8 ± 0.4 cal. yr BP), 345–352 cm (11.8 ± 0.4 cal. yr BP) and 402 cm (12.8 ± 0.4 cal. yr BP).

A similar pattern is seen in PCB11, where below 240 cm (7.4 ± 0.6 cal. yr BP) there was an abrupt increase in bulk density, magnetic susceptibility, Zr / Rb and Ca / Ti. TOC concentrations and the Br / Cl ratio (which mirrors small scale changes in the TOC) also decreased notably through this interval (Figs. 2b, 3a). This lithologic transition post-dates the Late glacial and Early Holocene detrital carbonate inputs in cores recovered from the Beaufort Sea slope (Klotsko et al., 2019). It is likely that this coarser basal facies is related to the inundation of the shelf during transgression.

Bulk sediment δ13C of PCB09 was lowest in the Early Holocene at approximately -26.3 ‰ until 8.7 ± 0.4 ka, before increasing during the Middle and Late Holocene to -24.7 ‰ (Fig. 3b). The trend in δ13C in PCB11 is similar to PCB09 showing a steady increase over time from -26.5 ‰ to -25.8 ‰.

IP₂₅ and HBI II (C_25:2) concentrations were generally low (< 2 ng g⁻¹) in the Early Holocene (Fig. 3c, d). IP₂₅ in both cores increased throughout the Middle to Late Holocene. During the Late Holocene, IP₂₅ and HBI II concentrations dropped in PCB09 around 1.8 ± 1.5 ka. Concentrations of both biomarkers were higher in PCB11 than in PCB09 after 3 ka, reaching modern values of 40 and 150 ng g⁻¹ (IP₂₅ and HBI II). PIP₂₅ values in both cores increased from the Early to the Mid-Holocene (Fig. 4a).

HBI III (C_25:3) and HBI IV (C_25:4) were low in both cores with values below 8 ng g⁻¹ (Fig. 3e, f). Concentrations were higher in PCB11 than in PCB09 after 4 ka.

The concentration of isoGDGTs and OH-GDGTs followed a similar pattern throughout the Holocene (Fig. 3g, h). IsoGDGTs and OH-GDGT concentrations in PCB09 were stable during the Early Holocene at around 400 and 25 ng g⁻¹, respectively. At around 8.5 ka, the isoGDGTs and OH-GDGT amounts doubled. Throughout the Mid-Holocene, isoGDGTs and OH-GDGT concentrations were variable but above 500 ng g⁻¹. A drop in PCB09 to almost below detection limits occurred between 1–1.5 ka. IsoGDGTs and OH-GDGTs in PCB11 showed a steady increase from around 100 and 10 ng g⁻¹, respectively, in the Early Holocene to > 500 and > 50 ng g⁻¹ (Fig. 3g, h).

BrGDGTs concentrations in PCB09 were below 100 ng g⁻¹ throughout the cores except for peaks during the Early and Mid-Holocene at 11.2 ± 0.3, 8.2 ± 0.5 and 5.7 ± 0.5 ka, the latter was also seen in PCB11 (albeit concentrations were higher in PCB11) (Fig. 3i). BIT values varied from 0.1 to 0.4 in PCB09 and from 0.1 and 0.5 in PCB11. The #ring_tetra values were always < 0.7 for both cores. Terrestrial sterol concentrations in PCB09 were relatively stable throughout the core except for short-lived peaks at 9.5 ± 0.4, 8.2 ± 0.5 and 5.0 ± 0.9 ka (Fig. 3j). In PCB11, the concentration remained stable throughout the core after an initial increase at 6.9 ± 0.6 ka and a peak in the surface sediment.

Although biomarker concentrations expressed per gram of sediment may be influenced by mineral dilution or post-depositional degradation, TOC contents in our cores varied only slightly (0.9 %–1.3 % in PCB09 and 1.1 %–1.3 % in PCB11; Fig. 3a). Such limited variability indicates that TOC normalization would not substantially alter the observed trends. The distinct downcore patterns among lipid biomarker classes therefore likely reflect differences in source input and preservation rather than a uniform effect of degradation or dilution.

Surface sediments δ2H C_16:0 values from the Beaufort Sea (Figs. S1, S4) range from -275 ‰ to -200 ‰, comparable with the values obtained by the preliminary study of Sachs et al. (2018). δ2H C_16:0 values of all sediments correlate with summer salinity (r2 = 0.63, p < 0.001) and the calibration equation is the same as the one obtained by Sachs et al. (2018). It is to be noted that the uncertainty associated with the analysis and calibration reaches 7 psu, which is quite large for salinity changes over glacial-deglacial timescales. This is to the contrary to what Allan et al. (2023) and Wu et al. (2025) observed in a set of surface sediments in Baffin Bay and a downcore record of the Beaufort Sea, where no relationship with salinity is observed. This contrast for the surface sediments could come from the different environment as Baffin Bay is a much more enclosed basin compared to the Beaufort Sea, or that δ2H C_16:0 values encompassed a too small range of salinity (28–34 psu). The downcore Beaufort Sea record yielded no notable salinity response of the δ2H C_16:0 when large freshwater influence were expected, which was attributed to the production of C_16:0 by heterotrophs (Wu et al., 2025).

Sea surface salinity inferred from δ2H C_16:0 values at PCB09 increased from 27 ± 7 psu during the Early Holocene to 30 ± 7 psu during the Mid-Holocene, and remained stable until 3 ka, increasing to 32 ± 7 psu during the Late Holocene (Fig. 4b). Reconstructed salinity at PCB11 was more stable during the Mid-Holocene and Late Holocene (28 ± 7 psu) until an increase to 30 ± 7 psu in the last centuries. (Fig. 4b). Although uncertainties associated with reconstructed salinity are large (±7 psu) the salinity trend between both locations agree with modern observations of lower salinities at PCB11 than further offshore at PCB09 (Fig. S2).

Two different sets of SSTs were reconstructed using the OH-GDGT only (RI-OH^′) or a combination of OH- and isoGDGT (TEX-OH) (Figs. 4d, S4, S5a, b). SSTs were only reconstructed when the BIT index was below 0.3 (Fig. 4c) as both calibrations are sensitive to terrestrial input (Varma et al., 2025). RI-OH^′ in the surface sediments varies from 0.05 to 0.17 while TEX-OH varies from 0.08 to 0.32. Both indexes plot in the global calibration curves from (Varma et al., 2024) and the reconstructed SST varies from 0.9 to 4.0 and -0.1 to 11.6 °C, respectively. TEX-OH reconstructed SSTs in PCB09 varied between 7 ± 2.6 °C in the Early Holocene, remained stable during the Mid-Holocene (∼ 3 ± 2.6 °C), decreased to 0 ± 2.6 °C between 1–1.5 ka and after which they increase to 5 ± 2.6 °C, close to modern summer SST (Locarnini et al., 2024). The reconstructed SST in the multicore is -1 ± 2.6 °C, similar, within the calibration error, to modern annual SST (Locarnini et al., 2024) (Fig. 4d). RI-OH^′ reconstructed SSTs in PCB09 produce high temperatures (∼ 12 ± 2.6 °C) between 7 and 10 ka, maybe due to non-thermal impact on the production of OH-GDGT (Harning and Sepúlveda, 2024). For both cores, reconstructed SST using RI-OH^′ is stable around 3 °C (Fig. S5b). PCB11 TEX-OH reconstructed SSTs are stable during the Early to Mid-Holocene (∼ 5.0 ± 2.6 °C). However, after 1 ka the reconstruction produces unrealistically large variation (5–15 °C, Fig. S5a).

The BIT index showed a steady decrease in PCB11 throughout the Holocene and until 3 ka in PCB09 (Fig. 4c). In PCB09, this decrease was interrupted at 9 and at 1.5 ka with BIT index values reaching 0.3 and 0.4, respectively. The increase at 9 ka was likely due to a relative decrease in crenarchaeol concentration (Fig. S5a) whereas the 1.5 ka increase was likely due to a decrease in brGDGT concentration (Fig. 4i).

Almost all of the planktonic foraminifera (99 %–100 % in abundance relative to other species) observed in PCB09 are N. pachyderma, formerly N. pachyderma sinistra. This is expected since this species has been found to dominate polar water masses (e.g. Eynaud 2011; Moller, Schulz, and Kucera 2013). Planktonic foraminifera are mostly absent in PCB11, consistent with data from plankton tows indicating that planktic foraminifera are rare on the Canadian shelf where surface waters are influenced by Mackenzie River discharge (Vilks, 1989). In PCB09, the foraminiferal shells appear white and fragmented in sections with abundant light-colored and sand-sized ice-rafted debris and other detrital materials (Fig. S6). Foraminifera are more abundant in samples that have relatively more mud aggregates than sand-sized debris (Fig. 2a, b). There is almost zero accumulation rate (per mm yr⁻¹) of N. pachyderma within the shelf slope from 10 ka.

The Late glacial to Early Holocene is only recorded at the shelf slope location (PCB09, Fig. 1). This period is characterized by low concentrations of sea ice biomarkers (Fig. 3a, b) resulting in low PIP₂₅ values (Fig. 4a). The low concentration suggests that this area had intermittent sea ice coverage during the Late glacial to Early Holocene, but the presence of HBI III and HBI IV (Fig. 3e, f) indicate that the region was only under seasonal ice cover until spring allowing late spring/summer open-water diatom primary production (Belt et al., 2015). Heterotrophic production in the shelf slope region during this period is relatively low (as suggested by the presence of ammonium oxidizer Thaumarchaea-derived isoGDGTs; Schouten et al., 2013; Fig. 3g, h) but increased and peaked at 8.2 ± 0.5 ka. During 12–8.2 ka, SSTs were elevated in comparison with the rest of the Holocene (Fig. 4d) which coincided with peak 21 June insolation (Fig. 4f) (Clemens et al., 2010; Laskar et al., 2004). The warmer surface waters might have inhibited the development of sea ice over the Beaufort Shelf.

During the Late glacial to Early Holocene, large freshwater inputs to the Beaufort Shelf, inferred from the low reconstructed salinity (Fig. 4b) likely originated from the decaying Laurentide Ice Sheet. Such water masses derived from drainage regions that had undergone minimal weathering would have released low amounts of nutrients. The influx of low-salinity freshwater may have intensified salinity-driven stratification on the shelf, reducing the upwelling of nutrient-rich saline Pacific waters to the surface which also limited nutrient availability. This stratification and lower nutrient availability likely limited primary productivity and the presence of ammonia-oxidizers on the Beaufort Shelf. It is important to note that sea level on the Beaufort Shelf was > 60 m lower in the Early Holocene than it is today (Fig. 4e). Implying that between 10–12 ka, the Beaufort Sea was a shallow estuarine environment (Fig. 4g; Hill et al., 1993).

The concentration of brGDGTs and terrestrial sterols in the shelf slope location during the Early Holocene peaked at 11.3 ± 0.3 and 8.2 ± 0.5 ka (Fig. 3i, j), which agrees with meltwater outflow from the LIS and freshly deglaciated surfaces as seen in nearby cores (Klotsko et al., 2019; Wu et al., 2020a). Additionally, increased freshwater input may have transported more detrital calcium (Ca), as indicated by elevated Ca / Ti ratios (Fig. 2a), which could have enhanced the preservation of foraminifera by buffering the water column and limiting carbonate dissolution, in sediments along the shelf slope. Murton et al. (2010) used optically stimulated luminescence (OSL) dating to identify two major meltwater pulses through the Mackenzie River system between 13 and 11.7 ka and between 11.7 and 9.3 ka. This timing is supported by sedimentary and isotopic records from the Beaufort Sea indicating a major Lake Agassiz flood route through the Mackenzie system (Keigwin et al., 2018; Klotsko et al., 2019). These meltwater events coincide with events (11.3 ± 0.3, 8.2 ± 0.5 ka) in the biomarker records from this study (Fig. 3), and one event at 10.1 ± 0.4 ka is recorded in the reconstructed salinity (Fig. 4b), suggesting enhanced freshwater forcing impacted ocean circulation and increased sea ice extent. The massive meltwater discharge from the LIS (at least ∼ 9000 km³) into its surrounding oceans has been the major cause for eustatic sea level rise from 10 to 6 ka (Moran and Bryson, 1969).

After 8.2 ka, environmental conditions at the shelf slope (PCB09) and the outer shelf (PCB11) diverged markedly, reflecting their contrasting depositional and oceanographic settings. At PCB09, SSTs cooled from ∼ 6 to 3 °C (Fig. 4d), while steadily increasing sea-ice biomarker concentrations led to PIP₂₅ > 0.5 by 7–6 ka, indicating the establishment of stable ice-edge or polynya conditions at the slope. The higher salinity (Fig. 4b) and greater distance from the coast at this site suggest enhanced influence of offshore Pacific-derived waters and reduced terrestrial input.

In contrast, at the outer shelf site PCB11, sea-ice biomarkers also increased after 8.5 ka but were accompanied by persistently high concentrations of open-water diatom markers, implying continued seasonal sea-ice and productive flaw-lead conditions, i.e., an open-water or newly formed sea-ice zone between landfast ice and sea ice. The flaw-lead today occurs about 80 km away from shore (Fig. S1) (Carmack et al., 2004). The proximity of PCB11 to the coastline before 6 ka (Fig. 4g) likely favored landfast-ice diatom assemblages and greater sensitivity to freshwater discharge. Only after 4 ± 0.5 ka did PCB11 reach PIP₂₅ values comparable to the slope, indicating a delayed transition to stable seasonal sea ice, approximately 2 kyr later than at PCB09. Thus, while both sites record a Middle-Holocene trend toward increasing sea-ice cover, the slope experienced earlier stabilization and reduced productivity linked to offshore cooling and stratification, whereas the outer shelf remained a dynamic, seasonally open-water environment likely sustained by coastal flaw-lead formation, and strong riverine influence.

In the Late Holocene (4.2 ka to present), the two Beaufort sites again show distinct sea-ice and productivity trajectories reflecting their different hydrographic settings. At the slope site, perennial sea ice cover developed after 3 ka as indicated by higher PIP₂₅ values (> 0.8), increased reconstructed salinity (Fig. 4a, b), as well as a decline in open-water diatom biomarkers (Fig. 3e, f). A sharp decrease in PIP₂₅ and open-water diatom biomarkers at 1.5 ± 1.5 ka suggests a shift to year-round ice cover coincident with low SST (∼ 0 °C, Fig. 4d). The parallel reduction in heterotrophic archaeal production may reflect strengthened water-column stratification or reduced nutrient supply, consistent with restricted shelf-break upwelling (Schulze and Pickart, 2012). These changes are broadly consistent with the timing of the regional cooling associated with the Little Ice Age (Mann et al., 2009), although the resolution of the biomarker record does not allow precise attribution to centennial-scale events. In contrast, on the outer shelf, seasonal sea-ice conditions persisted longer and sea ice cover expanded gradually and became well established after about 2 ± 0.6 ka. Even as sea-ice biomarkers increased, open-water diatom markers remained relatively abundant, implying continued flaw-lead or marginal-ice-zone productivity sustained by intermittent open-water formation and coastal influence.

By the last 0.4 ka, both sites reached modern configurations, but with clear spatial differences: PCB09 records persistently thick, multi-year ice and limited primary production, whereas PCB11 retains higher concentrations of both sea-ice and open-water diatom biomarkers, consistent with modern flaw-lead dynamics observed near the outer shelf (Fig. S1).

Previous studies using IP₂₅ to reconstruct sea ice variability in Arctic marginal seas have reported largely open-water conditions with significant freshwater influence during the Late glacial to Early Holocene.

The nearby cores JPC15 (Keigwin et al., 2018), ARAC20 (Wu et al., 2020a) (Fig. 1) recorded similar environmental changes (sea ice cover, freshwater input) as in PCB09 but different from those recorded in the shallow PCB11 site, highlighting the differences between shelf slope and outer shelf and the spatial variation of the polynya position. Aside from the close-by cores (Keigwin et al., 2018; Klotsko et al., 2019; Wu et al., 2020a), other Arctic records in the Canadian Archipelago (Vare et al., 2009), East Siberian (Dong et al., 2022), Kara (Hörner et al., 2018) , Chukchi (Stein et al., 2017), Laptev (Fahl and Stein, 2012; Hörner et al., 2016), and Lincoln (Detlef et al., 2023) Seas and along the Lomonosov Ridge (Stein and Fahl, 2012), report minimum sea-ice cover during the Early Holocene (centered around 10 ka) (Fig. 5). Detlef et al. (2023) reconstructed sea ice conditions from a sediment core covering the last 11 ka, showing that while the Lincoln Sea currently experiences perennial sea ice cover (PIP₂₅ = 0), it underwent a shift to seasonal sea ice during the Early Holocene (around 10 ka) due to significantly warmer conditions (PIP₂₅ > 0.5). This period of reduced sea ice cover is associated with increased marine productivity and meltwater input indicated by biomarker and sedimentary facies. The Northern Greenland (Detlef et al., 2023) and the site offshore the Laptev Sea on the Lomonosov Ridge (Fahl and Stein, 2012; Hörner et al., 2016) are the first regions to record permanent sea-ice cover beginning around 9 ka after the Early Holocene minimum. The Beaufort Sea (this study; Wu et al., 2020a) showed permanent sea-ice cover over the slope after 3 ka. Seasonal sea-ice cover in the shallower regions of the Laptev and Beaufort Seas (PS51-159 and PCB11) was developed after 5 and 3 ka, respectively. The Chukchi Sea (ARA2B) had seasonal sea-ice from the Mid-Holocene onwards, with an increase after 4.5 ka (Stein et al., 2017). The variations in sea ice cover and primary production in the Chukchi Sea were attributed to differences in solar insolation and variability in Pacific water inflow, which brought increased heat flux and episodic declines in sea ice cover. In the Canadian Archipelago, a record that did not include the Early Holocene (Belt et al., 2010; Vare et al., 2009), increased sea-ice cover exists from from 7 to 3 ka.

In contrast, studies using dinocyst assemblages from around the Arctic Ocean (see the review of de Vernal et al., 2013) report constant sea ice cover for the Early to Mid-Holocene with a clear decrease around 6 ka, followed by a return to pre-6 ka conditions until an increase toward modern times. This could be due to a warm-bias in the dinocyst estimate or a non-representative training set (de Vernal et al., 2013).

Together, many of the biomarker studies provide a consistent narrative of spring sea ice development during the Holocene across the Arctic Ocean following a period of high insolation in the Early Holocene. The transition from largely open-water and freshwater-influenced conditions during the Late glacial to Early Holocene to increasing sea ice cover from the Mid-Holocene onward is a shared feature across the Arctic shelf seas, although regional variations in sea ice dynamics and productivity are observed due to local freshwater input and oceanographic conditions. Evidence from areas with permanent sea ice, such as the Lincoln Sea, shows that the minimum ice cover during the Late glacial extended even into the high Arctic, offering insights into the extent of sea ice reduction during this time.