Seasonal sea ice persisted through the Holocene Thermal Maximum at 80°N

The cryospheric response to climatic warming responsible for recent Arctic sea ice decline can be elucidated using marine geological archives which offer an important long-term perspective. The Holocene Thermal Maximum, between 10 and 6 thousand years ago, provides an opportunity to investigate sea ice during a warmer-than-present interval. Here we use organic biomarkers and benthic foraminiferal stable isotope data from two sediment cores in the northernmost Barents Sea (>80 °N) to reconstruct seasonal sea ice between 11.7 and 9.1 thousand years ago. We identify the continued persistence of sea-ice biomarkers which suggest spring sea ice concentrations as high as 55%. During the same period, high foraminiferal oxygen stable isotopes and elevated phytoplankton biomarker concentrations indicate the influence of warm Atlantic-derived bottom water and peak biological productivity, respectively. We conclude that seasonal sea ice persisted in the northern Barents Sea during the Holocene Thermal Maximum, despite warmer-than-present conditions and Atlantic Water inflow. The northern Barents Sea contained seasonal sea ice between 11.7 and 9.1 thousand years ago, even during the warmer-than-present Holocene Thermal Maximum and despite warm Atlantic Water inflow, according to biomarker and benthic foraminiferal stable isotope records.

T he dramatic decline in Arctic sea ice over recent decades represents one of the most poignant signs of contemporary, anthropogenically forced climatic change 1 . This loss raises considerable concerns for environments, ecosystems and communities at both regional and global scales because of the pivotal role that sea ice plays in the Earth's climate system via energy budgets, feedback loops, atmosphere-ocean exchange and oceanographic circulation 1,2 . Accordingly, the consequences of Northern Hemisphere sea-ice loss will cascade through the Earth System, thereby impacting lower latitudes, for example via changing atmospheric circulation and more frequent extreme weather events 3 . The detailed response of sea-ice cover to, and its role in, an increasingly warmer future climate remains uncertain, however, as numerical-model-based future projections vary considerably between regions and different models [4][5][6] . For example, a winter ice-free Barents Sea is projected by 2061-2088, but there is considerable spread between model-based sea-ice extent projections for the next few decades, for instance ranging from 0.1 (GFDL CM3 model) to 0.6 × 10 6 km 2 (NorESM1-M) for the winter of 2030 6 . Similarly, CMIP5 model simulations of the timing of an ice-free Arctic Ocean vary from the 2010s to the late 2070s, depending on the model used 4 . Such variability in modelbased forecasts stems from differences in parametrization, model approaches, and future emission scenarios 6,7 , but is also attributable to the inherent complexities of internal climate variability. At present, these factors limit the accuracy of future projections 4,8 .
While model simulations offer some insight into potential future sea-ice states in the Arctic, geological archives provide critical information on how sea ice responded to substantial climatic warming in the past, especially during periods characterized by warmer-than-present conditions. Detailed sea-ice histories documenting longer term variability past the observational period thus contribute significantly to the understanding of underlying causes and the contextualization of recent and potential future changes. Such a longer term, palaeo aspect is particularly relevant considering the current trend towards a warmer global climate in tandem with Arctic amplification 9 . The early Holocene Barents Sea provides a unique opportunity to study Arctic sea ice during the most recent past warm period in an oceanographically dynamic area that serves as a focus for Arctic-Atlantic Ocean interaction 7 and that has been identified as a hotspot for current Arctic warming 10 . Here, we present the first high-resolution, decadal-scale early Holocene sea-ice reconstruction from the high Arctic Barents Sea from two sites ( Fig. 1) based on the sea-ice proxy IP 25 and other biomarkers, alongside foraminiferal stable isotopes (δ 18 O, δ 13 C). Our reconstructions unequivocally demonstrate that, despite warm climatic conditions and Atlantic Water influence, seasonal sea ice persisted throughout the early Holocene (up to~9.1 cal ka BP) in the northern Barents Sea at the border to the Arctic Ocean. We further show that high biological productivity characterized this region during that time, with sporadic phytoplankton blooms occurring along the Marginal Ice Zone (MIZ). Our results raise important questions about the response to, and long-term fate of, Arctic sea ice in an increasingly warming climate, in a region of high climatic, environmental and economic importance.

Results and discussion
Seasonal sea ice persists throughout the warm early Holocene. Our reconstructions reveal the continued persistence of seasonal sea ice during the early Holocene in the northern Barents Sea, evidenced by the continuous occurrence of the sea-ice biomarkers IP 25 and IPSO 25 in both investigated cores (Fig. 2). Both these sympagic biomarkers show elevated concentrations following the Younger Dryas cold period (YD, 12.9-11.7 cal ka BP) and into the early Holocene, until~9.1 cal ka BP, when their concentrations begin to decline. This pattern is also mirrored in the two biomarkers produced by open-water phytoplankton, HBI III and HBI IV, which display peak concentrations at 11.7-9.1 cal ka BP and decrease thereafter (Fig. 2). Further, by combining IP 25 concentrations with those of the pelagic biomarker HBI III (see 'Methods'), we identify predominantly intermediate (i.e. 10-50%; sea-ice categories sensu Köseoğlu et al. 11 , see 'Methods') spring sea-ice concentrations (SpSIC) following the YD, until~9.1 cal ka BP (Figs. 2 and 3). Both sites in the northern Barents Sea have similar SpSIC trends during the period 11.7-9.1 cal ka BP, although 06GC shows slightly higher sea-ice concentrations (~20-55%) compared to the more northerly 11GC (~10-40%). Sea ice increases at both sites after~9.1 cal ka BP, shifting from intermediate to predominantly extensive (i.e., >50%) by~8.3 cal ka BP (Fig. 2).
The orbitally forced summer insolation maximum during the Holocene Thermal Maximum (HTM,~10-6 ka BP) resulted in warmer air and ocean temperatures globally at that time 12 , though these signals may have been primarily seasonally driven 13 . At higher latitudes, increased temperatures may have also acted as an effective mechanism to thin sea ice and/or reduce its extent by lowering albedo in a positive feedback loop 14,15 . At 80°N, mean insolation increased by 5 W m −2 (annual) or~50 W m −2 (June) compared to present 16 (Fig. 3). Although the predominantly intermediate SpSIC seen in our two archives is consistent with a warm period, our results nonetheless show that the northern Barents Sea was never entirely free from seasonal sea ice at this time.
Across the Barents Sea-Svalbard region, marine geological archives also suggest markedly warmer conditions beginning~11 cal ka BP. This is considered to be tied to the concomitant northward inflow of relative warm and saline Atlantic waters from the south 17-20 following deglaciation and the YD, even if model simulations vary in the exact timing of the recovery of the AMOC (Atlantic meridional overturning circulation) 21,22 . The warm signal is especially prominent along the western margin of Spitsbergen, where benthic foraminiferal transfer functions suggest some of the highest bottom water temperatures of the Holocene at 11.5-8.2 cal ka BP 23 . Additionally, molluscs collected from glacio-isostatically uplifted shorelines show that warm water faunas including Mytilus edulis (blue mussel) inhabited the northwestern coast of Svalbard during the HTM 24 . An overall warmer early Holocene climate is also revealed by terrestrial data across the Svalbard archipelago, suggesting smaller glacier extents and higher air temperatures 25 (Fig. 3), as well as minimal sea ice covering the fjords 26 .
Many other marine archives across the Barents Sea-Svalbard region imply drastically reduced sea ice, if not completely open water conditions, during the early Holocene 23,27-30 . For example, biomarker profiles from Fram Strait, west of Svalbard, mimic the individual HBI patterns seen in our two archives (Fig. 2), with high concentrations of IP 25 during the YD, before declining in the early Holocene, when elevated brassicasterol levels point to high phytoplankton productivity 27 . Nevertheless, considerable regional variability exists, not least in the northern Barents Sea where existing records are complicated by scarce or low-resolution Holocene sediments 31,32 and chronological issues 33 . The only other HBI-based record in the vicinity of our two sites commences at 9.5 cal ka BP 29 (Fig. 1), thus overlapping with our archives by 500 cal a. Even so, the lower part of that archive 29 shows IP 25 and HBI III concentrations on a par with our individual profiles of these two biomarkers, as well as confirming the early Holocene peak in HBI III (Fig. 2). Other relevant studies south of our sites (Fig. 1) show differing reconstructions depending on the proxy method used. Dinoflagellate-cyst-based Modern Analogue Technique transfer functions suggest 8-10 months of sea-ice cover per year at~11.3-8.7 cal ka BP 34 while benthic foraminifera in the same and a nearby core (S2825 and S2519, respectively) show pronounced peaks in taxa indicative of bioproductivity and Atlantic Water inflow 33 . Conversely, biogenic proxies (benthic foraminifera assemblages, planktic foraminifera and foraminiferal stable isotopes) in a nearby marine archive (NP05-71GC) suggest reduced sea ice, warm surface water conditions, and a strong inflow of Atlantic Water at 11.7-8.6 cal ka BP 20 .
Regional variability is also evident in our reconstructed records ( Figs. 2 and 3). The northern site (11GC) exhibits lower SpSIC compared to the southern site (06GC), with the latter occasionally showing extensive (>50%) SpSIC. This may be a result of 06GC being affected by fast ice anchored to the nearby Svalbard coast and/or meltwater off Svalbard promoting sea-ice formation, even if meltwater influence would have been considerably diminished by the early Holocene compared to deglaciation in the late Pleistocene 35 . However, the differences in SpSIC may also be a function of the variable influence of Atlantic Water at the two core sites, with changes in direction and magnitude of subsurface inflow during this time. Benthic foraminiferal oxygen stable isotopes (δ 18 O) in 06GC are high during the early Holocene (ca. 5‰; Fig. 3), whereas the more northerly core 11GC shows slightly lower values (ca. 4.8‰) and a 2-step increasing trend, which may explain the difference in SpSIC between the two sites ( Fig. 3). Since the δ 18 O signal (of calcifiers) in the modern Barents Sea is primarily driven by temperature rather than salinity 36,37 , bottom waters may have been slightly warmer at the more northerly site during the HTM. At the same interval in core 06GC we also observe the benthic foraminiferal species Cassidulina neoteretis (K. Husum, personal observation 2020), an indicator of Atlantic Water 38,39 often found in stratified settings 40,41 . Collectively with the stable isotopes, this suggests that Atlantic-derived bottom water reached both sites during the early Holocene, similar to south of Kvitøya where a contemporaneous record 20 shows comparable benthic δ 18 O values. A strong Atlantic Water influence (below the pycnocline at~100 m water depth) is also exerted on the west Svalbard shelf during this time 23 . The presence of sea ice throughout the HTM at both our sites, as indicated by biomarker data presented herein, argues against Atlantic Water occupying the whole water column, however (as suggested on the western Svalbard margin after 9.6 cal ka BP 23 ), because its heat tends to hinder sea-ice formation 37 . It is therefore likely that the bottom Atlantic Water was overlain by an intermediate, Arctic-derived layer and sea-ice-related halocline, as it is today 42 , with limited vertical mixing occurring 43 .
Early Holocene characterized by high biological productivity and Atlantic Water influence. Individual profiles of the pelagic diatom-derived biomarkers HBI III and HBI IV in our two northern Barents Sea archives reveal that open-water biological productivity likely reached a maximum during the early Holocene (Fig. 2), following a period of much reduced productivity during the YD, probably as a result of the contrasting sea-ice conditions. This early Holocene phytoplankton productivity, while elevated, may have rarely reached the particularly high levels seen in the modern Barents Sea MIZ 44 , however. The HBI T 25 index, a ratio between HBI III and HBI IV (see 'Methods'), has recently been proposed as a potential proxy for the spring diatom bloom, on the basis of its strong association with spring chlorophyll-a 44 . In our two archives, HBI T 25 values are mostly below the threshold of phytoplankton blooming at the MIZ (defined 44 as spring chlorophyll-a >1.5 mg m −3 ), indicating the absence of spring blooms in 11GC (all HBI T 25 < 1), with sporadic blooms (i.e., HBI T 25 > 1) at the 06GC site during the YD and around 9.3 cal ka BP (Fig. 2). Nevertheless, our average HBI III concentrations within the early Holocene at 11.7-9.1 cal ka BP (06GC: 9.3 ng g −1 ; 11GC: 11.1 ng g −1 ) are similar to the mean modern value from the Barents Sea MIZ based on surface sediments 45 (12.9 ng g −1 ). A productive early Holocene sea-ice edge with intermediate SpSIC is further supported by the relatively low endobenthic foraminiferal δ 13 C values in both records (Fig. 3). Lower δ 13 C values at 06GC compared to 11GC are suggestive of more favourable conditions farther south, which could be due to increased climatic amelioration and/or proximity to the northern Svalbard coasts bringing additional nutrients and mixing to the water column.
High pelagic productivity during the early Holocene has been observed elsewhere across the Barents Sea 28 and adjacent regions 46 , including off northern Norway 47 and in eastern Fram Strait 27 . Warm-water (Atlantic-affiliated) benthic foraminifera suggest heightened biological productivity at the YD/Holocene transition off north Nordaustlandet 31 and north off Kvitøya Trough, where palaeoproductivity at the time was estimated at 120 g cm 2 a −1 , approximately twice the present-day regional values 48 . This regional palaeoproductivity increase was likely promoted by high insolation 27 . The biologically productive MIZ was probably proximal to our study sites, though alternatively leads within the seasonal sea ice pack may have provided a suitable environment for open water algal growth 49 .
From~9.1 cal ka BP onwards, both our records show lower concentrations of individual biomarkers, including those produced by sea-ice diatoms and pelagic algae (Fig. 2), whereas ratiobased spring sea-ice concentrations increase and HBI T 25 remains low (Fig. 2). By~8.3 cal ka BP, both sites are characterized by predominantly extensive SpSIC (consistently > 50%). The seemingly low concentrations of individual biomarkers after~9.1 cal ka BP are likely a result of a masking effect by their very high earlier concentrations during and after the YD (Fig. 2), in addition to individual HBIs suffering from variability in factors such as production, export, and preservation compared to ratiobased approaches 50,51 . In fact, ratio-based methods, such as SpSIC and HBI T 25 , show much better semi-quantitative calibrations (i.e., to measured spring sea-ice concentrations and spring chlorophyll-a values, respectively) than individual biomarkers 44,51 . Our pattern of the early Holocene reduction of individual biomarkers following high YD and post-YD concentrations is in keeping with other regional records, including those in eastern Fram Strait 27 and the western Barents Sea 28 . The concomitant increase in ratio-based SpSIC seen in our records after~9.1 cal ka BP is furthermore in keeping with a regional trend towards more polar conditions influenced by Arctic surface waters that is observed across the region (Fig. 3), including subsurface and bottom water cooling from~9 cal ka BP onwards 23,52 . This signifies the end of the HTM on the northern Svalbard margin as both the insolation maximum and Atlantic Water inflow diminished. To the south of our sites, however, the end of the HTM may have been delayed, as indicated by increasing seasonal sea ice in Olga Basin after 5.9 cal ka BP 29 and the expansion of glaciers across Svalbard after~6 cal ka BP 25 .
Implications for ongoing climate warming. Atmosphere-ocean coupled general circulation models (CGCM/AOGCM) have demonstrated the possibility of perennial sea ice in the Arctic Ocean north of our core sites shifting to a seasonal sea-ice regime during the HTM 15 . Although driven by different factors during the HTM (heightened insolation then 16 vs. greenhouse gas emissions today 53 ), such a scenario may be a possibility for a future Arctic Ocean 53 , with marginal regions such as the Barents Sea shifting to open-water conditions year-round 6 . However, considerable uncertainties exist around model projections 8 , in part because the short period of sea-ice observations does not capture the internal variability of the system 4 . These factors hinder a sole reliance on model simulations to assess how sea ice will respond to continuing climate warming. This knowledge gap can, however, be bridged by geological data like those presented here, which use robust sea-ice proxies to elucidate warmer-thanpresent time periods at high temporal resolution.
Our results demonstrate that the northern Barents Sea margin was characterized by intermediate spring sea-ice concentrations at 11.7-9.1 cal ka BP during the HTM, when records in eastern Fram Strait 27 , the western and north-central Svalbard Shelf 23 , and the western Barents Sea 28 generally indicate substantially reduced or even absent sea ice, while northern areas like the Yermak Plateau were characterized by contemporaneous perennial sea ice 32 . The presence of open-water (phytoplankton) biomarkers, alongside those synthesized by sea-ice diatoms, in both cores, implies that the productive MIZ was nearby, located likely to the west of our study area. The situation to the south and east is less clear, however. Benthic foraminifera (diversities and assemblage composition) along with the presence of planktic foraminifera point to a relatively productive contemporaneous setting marked by reduced sea ice, heightened inflow of Atlantic-source water, and warm surface water conditions south of Kvitøya 20 . Northwest of Kvitøya, Modern Analogue Technique transfer functions on dinoflagellate cysts imply 'dense' sea-ice cover most of the year (8-10 months per year) and reduced primary productivity during this time 34 , while benthic foraminifera suggest enhanced primary productivity under Atlantic Water inflow, apparently denoting polynya development 33 . These contrasts, together with the fact that the only other available HBI record from the area 29 covers the period from 9.5 cal ka BP onwards, make it at present impossible to determine if and how far south and east seasonal sea ice extended in the Barents Sea during the early Holocene.
The recent and ongoing decline in seasonal sea ice in the Barents Sea has been linked primarily to the increased inflow of oceanic heat via the northernmost extension of the Norwegian Atlantic Current (the West Spitsbergen Current, Fig. 1), which regulates sea-ice extent and particularly affects winter sea-ice cover 42,54 . Such 'Atlantification' has also been implied during the HTM by palaeo-records in the Barents Sea, which show increased abundances of Atlantic-affiliated foraminiferal taxa along the Svalbard margin, especially in the west 23 . An Atlantic Water signal is also apparent in our cores on the northwestern Svalbard shelf, which suggest that subsurface waters of Atlantic origin reached this high-latitude area during the early Holocene. Even so, this northward Atlantic inflow, in tandem with  16 for 80°N in terms of: e annual, and f monthly (May, June) means (in W m −2 ). g Annual mean sea ice thicknesses for the Arctic Ocean from model simulations using dynamic surface albedo parameterization (normal simulation) 15 . h Percentage Svalbard glacier cover 25 . the warm HTM prevailing at the time, were apparently insufficient to result in fully open water conditions during spring. While oceanic heat has been recognized as an important factor in preventing the formation of sea ice in the Barents Sea at present 42 , its driving force may diminish in future, as changes in atmospheric circulation are likely to decrease Atlantic Water inflow into the Barents Sea 55 . Nonetheless, increasing atmospheric temperatures driven by greenhouse gas emissions may still prevent sea-ice formation or cause substantial thinning of ice in the following season, given continuing ice-albedo feedbacks, especially if a threshold of oceanic absorption of summer solar radiation has already been crossed 56 . It is also noteworthy that Northern Hemisphere sea ice appears to have persisted along the East Greenland coast (but not, noticeably, on the Yermak Plateau) during the Early Pliocene under warm global temperatures and high atmospheric CO 2 levels (380-400 ppm), although it is unclear whether it was produced in situ or transported into the area and trapped along the coast 57 . Our results showing that sea ice persisted at 11.7-9.1 cal ka BP through the HTM are in keeping with model projections placing the sea-ice edge around the northern Barents Sea from the 2070s onwards; this is despite continuing Atlantic Water inflows and the feedbacks for sea-ice formation in subsequent seasons 55 . One reason for the continuing presence of sea ice in the future may be water column stratification 37 , an oceanographic setting also implied in our reconstructions. This suggests that if the warm Atlantic Water layer is sufficiently separated from the surface by a cold (and fresher) intermediate layer, such as Arctic Water flowing in from the north and east 7 , sea ice may continue to persist at high Arctic latitudes (>80°N) even in periods of warm climate. A continued freshwater supply, such as import of sea ice from the Arctic Ocean, may however be required to maintain this intermediate buffer and prevent vertical mixing 42,43 . It is unclear whether and how fast the current deterioration of stratification observed in the northern Barents Sea, a result of heat input to the intermediate layer and potentially signifying the shift from an Arctic towards an Atlantic domain 42 , would reverse under a future scenario of lessened Atlantic influence 55 .
The Barents Sea is one of the most productive marginal seas of the Arctic Ocean and supports some of the World's most important and lucrative fish stocks 58 . The recently observed physical shifts in the Arctic, including declining sea ice and increased ocean heat transport, have already been linked to biological changes, most notably considerably higher (>30%) annual net primary productivity since the late 1990s 59 , earlier seasonal phytoplankton blooms 60 , and algal blooming under relatively thin first year ice 61 . Our northern Barents Sea records confirm relatively productive surface water conditions at 11.7-9.1 cal ka BP during the HTM, characterized by high phytoplankton biomarker concentrations and low benthic foraminiferal δ 13 C values (Fig. 3). This corroborates the proposition that reduced sea-ice regimes are likely to promote higher biological productivity, especially through increasing light penetration which boosts phytoplankton growth 1 , as long as vertical stratification still allows upwelling of nutrients 62 . Furthermore, if seasonal sea ice persists in the northern Barents Sea during periods of warmer climate, as indicated by our data, primary productivity may also increase within the sea ice (sympagic) ecosystem itself, especially under a thinning ice regime. In turn, this could contribute considerably to overall Arctic net primary productivity, as indicated by modern measurements 63 . Although climatic amelioration and seaice change may boost primary productivity in the short-term, such changes in the overall physical system also have the potential to alter phytoplankton community structure and distribution, inevitably cascading through the foodweb to higher trophic levels both in the water column and in the benthos, with mixed (i.e., interacting positive and negative) effects for most key functional groups 1 . The distribution of the commercially important Arctic cod (Gadus morhua) is closely tied to the inflow of Atlantic Water 64 , and the species has been recently observed in waters northwest and north of Svalbard 65 . However, the persistence of sea ice in the northeastern Barents Sea, as indicated by our reconstructions and by future projections 55 , may curtail the expansion of Arctic cod into the high Arctic, as it does to other taxa such as haddock (Melanogrammus aeglefinus) via thermally limiting spawning capabilities 66 . Lastly, continuing seasonal sea ice in the northern Barents Sea may also hinder marine transport via the Central Arctic Ocean or the so-called Northern Sea Route 67 , as well as the extraction of natural resources such as hydrocarbon reserves 1 .
In light of the present context of anthropogenically driven climate change and cryospheric deterioration, our results provide a vital perspective on the nuanced response of seasonal sea ice to increasingly warmer oceanic conditions by demonstrating that seasonal sea ice persisted in the northern Barents Sea between 11.7 and 9.1 cal ka BP through the HTM, and that the productive MIZ was located nearby. The warmer-than-present spring and summer conditions during this time, along with the inflow of subsurface Atlantic Water, were insufficient to cause sea-ice-free conditions, perhaps due to stratification efficiently buffering a warmer, saline subsurface from colder, fresher surface waters. As such, our reconstructions allow an important long-term perspective on sea ice beyond the short observational time period and provide data for constraining numerical models of future projections in warmer climates.

Methods
This study is based on two gravity cores (Fig. 1) 5 Age-depth models for the cores used in this study. a core 06GC (HH15-06GC) and b core 11GC (JR142-11GC). Dark and light blue envelopes outline the 68% and 95% probability density ranges for each age-depth model, corresponding approximately to a one and two age standard deviation, respectively. Light grey curves show probability distributions (likelihood) for single dates in years before present (BP); their marginal modelled posterior distributions, which take into account the constructed age-depth model, are shown as dark grey curves. The median modelled ages of the time markers used are highlighted by black circles. For details on individual dates, refer to Table 1. 152-445 cm). The sediments were subsequently freeze-dried and homogenized with pestle and mortar. About 2.5 g of each sample was processed for HBIs according to a slightly adapted method of the standard procedure 69 . Each batch of sediment samples (n = 10-16) included a procedural blank and an in-house reference sediment sample containing HBIs of interest and with known concentrations. The reference sediment was obtained by homogenizing material from a piston core taken from Barrow Strait in the central Canadian Arctic in 2005 70 . An internal standard (9-octylheptadec-8-ene (9-OHD); 100 ng) was added to each sample prior to extraction, which encompassed the following steps: addition of 6 ml potassium hydroxide (KOH, Fisher Scientific) solution (5 g KOH in 9:1 methanol:milliQ water), heating of sediment-KOH mixture at 70°C for 1 h, addition of hexane (Rathburn Chemicals) in three steps (6 ml total) and centrifugation (2 min at 2500 rpm), collection of hexane by pipetting into new vials, and removal of hexane under gentle N 2 stream (25°C). The sample was then purified by re-dissolving the extract in hexane (6 ml) and running it through conditioned (hexane) silica columns (60-200 µm, ca. 0.5 g, Fisher Scientific), followed by the removal of hexane under gentle N 2 stream (25°C). An additional step of purification was added to the HBI protocol by re-dissolving the dried extract in hexane (3 ml) and running it through silver-ion (Ag-Ion, Discovery ® ; ca. 0.1 g) columns 71 .
Subsequently the extract was concentrated by removing most of the solvent under a gentle stream of N 2 (25°C) and transferred into smaller vials. HBIs were analysed by gas chromatography-mass spectrometry (GC-MS, Agilent 7890 gas chromatograph equipped with the HP5MS fused-silica column (30 m; 0.25 mm film thickness; 0.25 mm internal diameter)) coupled to an Agilent 5975 series mass selective detector), and peaks of IP 25 , IPSO 25 , HBI III and HBI IV were identified and quantified in the software MSD ChemStation (GC retention indices (RIHP5MS) = 2081, 2082, 2044 and 2091 for IP 25 , IPSO 25 , HBI III and HBI IV, respectively). Peak values were translated into HBI concentrations (ng HBI per g of dry sediment) by their relation to peak areas and sediment masses vis-à-vis the internal standard and the reference sediment. HBI concentrations (ng g −1 ) were further used to calculate spring sea-ice concentrations (SpSIC) and the HBI T 25 index (a measure of phytoplankton blooms at the productive MIZ) using the following formulae 44   For foraminiferal stable isotopes, δ 18 O and δ 13 C were measured on 15-20 individuals of pristine Nonionella labradorica per sample with an automatic Finnigan MAT 253 mass spectrometer coupled to an automated Kiel device at the stable isotope laboratory (FARLAB) of the University of Bergen. Sampling intervals focused on the same core sections as biomarkers; due to the low abundance of foraminifera in JR142-11GC, stable isotopes are not available for some of the parts of this core. Results are given relative to the Vienna Pee Dee Belemnite (VPDB) standard, without corrections applied for vital or ice volume effects. Biomarker and stable isotope data were plotted in the C2 software 72 .
Radiocarbon dates were measured on mixed (predominantly Cassidulina reniforme, Elphidium excavatum f. clavata, N. labradorica) or monospecific (N. labradorica) benthic foraminifera; nine dates (including two duplicates) are available for JR142-11GC and 12 dates are available for HH15-06GC. All dates were calibrated in CALIB 8.2 73 using the MARINE20 calibration curve 74 and a regional radiocarbon reservoir offset (ΔR) of 105 ± 24 yr applicable to the Barents Sea (Bondevik & Gulliksen in Mangerud et al. 75 ). The age depth models for the cores (Fig. 5) were developed with Bayesian technique OxCal 4.4 76 using a Poisson deposition model. The estimates of the model parameter k of 0.3 (HH15-06GC) and 0.5 (JR142-11GC)-number of deposition events per unit length, also reflective of fluctuations in the sedimentation rate-are derived by the age model fit optimization procedure 76 . The ages of the core tops for the modelling were set to U [1900,2006] and U [1900,2015] for JR142-11GC and HH15-06GC, respectively, with U standing for the uniform probability distribution function. Since the core sections younger than 6 kyr BP are not considered in the present paper, this specific choice has no influence on the results presented here, as long as it is based on the inferred core-averaged sedimentation rate (Table 1).