The Antarctic seasonal sea ice cycle is characterised by growth to maximum extent in September, followed by retreat to a minimum in February. Between the minimum and maximum sea-ice extents lies the Marginal Ice Zone (MIZ), characterised by sea-ice concentrations between 15 % and 80 %. In the northeastern Weddell Sea, the median sea-ice edge (15 % sea-ice concentration) retreats rapidly with spring melt. In October/November, when the snow petrels start to return to their nesting sites, the sea-ice edge is located north of 60° S. Rapid sea-ice retreat through December brings the sea-ice edge to ∼ 70° S by early January, spanning the snow petrel incubation phase, then retreats further westward until the minimum in February (Fig. ). The Weddell Sea gyre advects sea ice clockwise across the basin where it outflows in the northwest (Gupta et al., 2025). Sea ice tends to accumulate to the greatest extent in the south of the Weddell Sea due to this gyre (Hutchings et al., 2012).

The preferred foraging habitat of snow petrels has been observed to track the MIZ, where the presence of open waters within the sea ice facilitates access to prey (Wakefield et al., 2025). This foraging behaviour can see snow petrels foraging over and beyond the continental shelf, reaching ∼ 700–1400 km from the nest site depending on the time of the breeding cycle and sea-ice extent (Honan et al., 2025). Based on regional tracking studies to the east, in Dronning Maud Land (Honan et al., 2025; Wakefield et al., 2025) three foraging options are available for snow petrels at Heimefrontfjella: (1) they head north to the MIZ lying to the south of the outer ice edge; (2) they head north east to the Maud Rise, where an intermittent open-ocean polynya and earlier sea ice melt provides foraging habitat; (3) they forage in the MIZ closer to the continental shelf, which can be associated with coastal polynyas. Tracking studies show that the most likely scenario is that early in the breeding season snow petrels forage in association with the northern edge of the MIZ as the sea ice retreats (Wakefield et al., 2025). As the sea-ice edge retreats so does the foraging range, so that later in the season snow petrels are more likely to be found in coastal waters (Honan et al., 2025; Wakefield et al., 2025).

Stomach-oil deposit 3012MUM2 was collected in season 2014–2015 from the Boysennuten nunatak in the Heimefrontfjella Range (74°34.14^′ S, 11°15.02^′ W) (Fig. ). The ∼ 32 cm × 23 cm deposit with a maximum thickness of 19 cm had an irregular, mamillated outer surface (McClymont et al., 2022). It was located immediately beneath a sheltered rock crevice, a typical habitat for nesting snow petrels, at an elevation of 1336 m above sea level (Fig. S1a and b in the Supplement). It was kept in the dark and frozen at -20 °C throughout the transportation processes to Durham University where sampling was carried out from 2021. The deposit was sliced using a circular saw while still frozen to preserve the internal millimetre-scale laminae which, when oriented, spanned a depth of ∼ 19 cm along the cutting axis (Fig. S1c). Sub sampling was carried out at 2.5 mm resolution with 3.0 mm biopsy punches for organic geochemistry, isotopes and radiocarbon analyses. We selected a sampling line which represented the maximum accumulation rate and a continuous sequence through the stratigraphy; noting some heterogeneity either side of this line (Figs. c and S1c). A 19 cm long slice from the opposite face of the cut was mounted in plastic trunking for high-resolution X-ray fluorescence (XRF) analysis of elemental composition.

An age-depth model for 3012MUM2 was constructed from twelve 14C ages (Table ). The top and bottom ages were sampled at 0 and 18.5 cm, immediately above and below the first geochemical samples (0.5 and 18.25 cm, respectively). Most of the radiocarbon ages were carried out by Beta Analytic (Miami, USA) using 14C-AMS via graphitization on untreated samples, which were oxidised to CO2 by combustion in O2 and converted to graphite with Co powder as a catalyst. To assess for the effects of acid removal on 14C ages samples at 5.0 and 10.5 cm were repeat-sampled and analysed at SUERC (Scottish Universities Environmental Research Centre) Environmental Radiocarbon Laboratory by digestion in 1 M HCL (hydrochloric acid) (at 80 °C, 2 h), washed free of mineral acid with deionised water, dried and homogenised. Carbon was recovered from the residue as CO2 by heating with CuO in a sealed tube and converted to graphite by Fe/Zn reduction. Samples were then analysed by AMS at the Keck Carbon Cycle AMS Facility, University of California, Irvine, USA. For calibration of radiocarbon ages to calendar ages (Table ) the MARINE20 radiocarbon age calibration (Heaton et al., 2020) was used, taking into account the regional marine reservoir of ΔR = 670 ± 50 years, measured at Hope Bay in the western Weddell Sea (Björck et al., 1991). This approach has previously been applied to snow petrel stomach-oil deposits (McClymont et al., 2022). A Bayesian age-depth model was then built in OxCal v4.4 (Bronk Ramsey, 2009) using the default settings (applied with a general outlier model, except for paired dates where we used the SSimple outlier model). Our choice of a Bayesian approach to age-depth modelling means that the age uncertainties of all dates are considered for the entire age model.

Carbon and nitrogen stable isotope analysis was performed using a Costech ECS400 elemental analyser coupled to a Thermo Scientific Delta V Advantage isotope ratio mass spectrometer in the Stable Isotope Biogeochemistry Laboratory (SIBL) at Durham University. The method used is described in McClymont et al. (2022). Results are reported in standard delta (δ) notation in per mil (‰) relative to Vienna Pee Dee Belemnite (VPDB) and atmospheric nitrogen (AIR), respectively. The linear range for δ13C was between -46 ‰ and +3 ‰ and for δ15N between -4.5 ‰ and +20.4 ‰, based on daily analysis of international (e.g. IAEA-600, IAEA-CH-3, IAEA-CH-6, IAEA-N-1, IAEA-N-2, NBS 19, USGS24, USGS40) and in-house standards, enabling a 2-standard-deviation analytical uncertainty of ± 0.1 ‰ for international standards (replicated) and < ± 0.2 ‰ on replicated samples. Total carbon (wt%C) and nitrogen (wt%N) were obtained simultaneously using an internal standard of glutamic acid (40.8 wt%C; 9.5 wt%N).

The biomarker sub-samples (0.02–0.05 g) were extracted in 4 mL dichloromethane (DCM) : hexane (3:1) after addition of internal standards (nonadecane, heptadecanoic acid, 5α-androstane, 5α-androstanol) and then sonicated for 15 min. Extracts were decanted and the procedure was repeated three more times. Extracts were combined and taken to dryness using rotary evaporation and N2. The entire sample was then saponified using 1 mL KOH (8 %) in methanol (95 %), heated for 1 h at 70 °C and left overnight. The neutral fraction was extracted with 3 × 3 mL hexane. The remaining sample was acidified to pH < 3 using drops of 2 M HCL, followed by extraction of fatty acids with hexane and evaporation to dryness with N2. Fatty acid methyl esters (FAMEs) were generated by methylating the fatty acid fraction using 3 mL methanol:HCL (95:5) heated for 12 h and left to cool. Samples were rinsed with 4 M DCM-cleaned H2O and then FAMEs were extracted with at least three DCM:hexane rinses (4:1), before evaporation to dryness under N2. Neutral fractions were separated in up to four fractions using 4 cm deactivated silica (heated 140 °C for 16 h) columns in glass pipettes plugged with extracted cotton wool (silica pore size 60 Å, 220–240 mesh particle size; 35–75 µm particle size; Sigma-Aldrich 60738-1KG). Hexane was used to condition the columns (x3 column volumes) followed by injection of the sample dissolved in 500 µL DCM directly into the column. Elution order into separate fractions used three column volumes each of hexane, DCM, DCM:methanol (1:1) and methanol. All fractions were decanted and transferred into GC vials and evaporated to dryness using N2. Fraction F3 (DCM:methanol (1:1), which recovered n-alcohols, phytol and sterols, was then further derivatized to trimethylsilyl esters prior to analysis using 50 µL DCM and 50 µL BSTFA (N,O-Bis(trimethylsilyl)trifluoroacetamide) (with 1 % TMCS (chlorotrimethylsilane)) heated for 1 h at 70 °C and left overnight prior to analysis. Samples were evaporated to dryness and dissolved in hexane prior to analysis.

All extracts were analysed using a Thermo Trace 1310 gas chromatograph coupled to an ISQ LT single quadrupole mass spectrometer (GC-MS). FAMEs extracts were separated using a Restek FAMEWAX (crossbond polyethylene glycol) column (30 m × 0.25 mm × 0.25 µm), similar to McClymont et al. (2022) but with some modifications. Briefly, samples were injected (0.8 µL) into a programmable temperature vapouring (PTV) injector in CT Splitless mode with inlet temperature at 250 °C, carrier gas in constant flow and with helium carrier gas set to 1.5 mLmin-1 (split flow 15.0 mLmin-1; splitless time 1.5 min; purge flow 5.0 mLmin-1). GC oven temperature was set to 100 °C for 3 min followed by 2 °Cmin-1 to 230 °C; hold of 12 min. Prep-run timeout was 10 min and equilibration time 0.5 min. MS settings included: transfer line, 230 °C; ion source temperature, 230 °C, mass range 38 to 600 m/z (every 0.5 s). Compounds were identified from their respective mass spectra and retention times, with quantities calculated relative to the peak area of the internal standard heptadecanoic acid and an assumption of a 1:1 response (validated by comparison with Supelco 37 component FAME mix; CRM47885, Merck).

Fraction F3 (previously eluted in DCM:methanol (1:1)) was separated using a Restek Rxi-5ms (crossbond 5 % diphenyl/95 % dimethyl polysiloxane) column (60 m × 0.25 mm × 0.25 µm). Similarly, samples were injected (0.8 µL) into a programmable temperature vapouring (PTV) injector in CT split-less mode but at 280 °C inlet temperature, with helium carrier gas set to 2.3 mLmin-1 set in constant flow mode (split flow set to 23 mLmin-1; splitless time 1 min; septum purge flow 5.0 mLmin-1). GC oven temperature was set to 50 °C hold for 2 min followed by 10 °Cmin-1 to 200 °C; followed by a slower ramp of 3 °Cmin-1 to 300 °C and a hold of 20 min (prep run timeout was 10 min, equilibration time 0.5 min). MS settings included: transfer line, 310 °C; ion volume, 300 °C, mass range 50 to 550 m/z (every 0.5 s). Compounds were identified from their respective mass spectra and retention times, with quantities of trimethylsilyl esters (TMS) calculated relative to the peak area of the internal standard 5α-androstanol and an assumption of a 1:1 response (validated by identically derivatized standard mix which included cholesterol-TMS).

XRF (X-ray fluorescence) analysis was carried out at Durham University, Department of Geography using a GEOTEK XRF Core Workstation (MSCL-XYZ) equipped with a rhodium source X-ray generator with a 10 mm cross-core slit width and a 1 mm downcore window. During operation the XRF scanner was set to analyse four different beam conditions, with a counting time of 10 s per beam. Beam conditions applied to the generator included: (1) no filter, 10 kV; (2) 25 µm silver filter, 20 kV; (3) 125 µm silver filter, 30 kV; (4) 625 µm copper filter, 50 kV. Detector measurement of photons ranged 2–35 keV. To pre-screen complex data, including removal of missing values and selecting elemental compositions from the most appropriate beam, the R 3.6.0 package “tidyverse” (Wickham et al., 2019) was used to produce a master XRF dataset.

To determine local bedrock chemistry, a sample of gneissic granite rock (consistent with local geology; Juckes, 1972) attached to deposit 3012MUM2 was soaked in ethanol to remove stomach-oil residue (repeated 3–4 times). It was then crushed using a fly press, freeze dried for 48 h, and ground to a fine, homogenous powder using a ball mill. Organic matter was removed by adding 4 mL of 30 % hydrogen peroxide to a ∼ 0.5 g aliquot of rock. The sample was then digested for 4 h in 16 mL of Aqua Regia using a DigiPREP digestion block, and subsequently diluted to 50 mL with deionised water and filtered at 0.45 µm. Elemental composition was determined using an Agilent Technologies 5100 Inductively Coupled Plasma Optical Emissions Spectrometer (ICP-OES).

Cluster analysis was carried out to highlight changes in geochemistry between neighbouring units in both inorganic (XRF) and organic (stable isotopes, TOC, C/N ratio and biomarkers) parameters with depth. For each dataset (organic (Org A–C) and inorganic (XRF A–C), separately) a constrained hierarchical cluster analysis based on sample order was performed using the rioja package in R (Juggins, 2020). A broken stick analysis (Bennett, 1996) was used to identify the maximum number of statistically significant clusters.

Principal components analysis (PCA) was carried out in Canoco V.5.51 (Ter Braak and Smilauer, 2002) on log⁡10-transformed and centred data: the inorganic (XRF) and organic (bulk organic geochemistry and biomarkers) parameters were treated separately. Input data for the organic PCA included bulk organics and biomarker concentrations, rather than fluxes (which included TOC normalized data and ratios). Input data for the inorganic PCA included XRF units in counts per second, rather than fluxes. As most XRF parameters had samples with counts < 500 we chose to retain all parameters that had passed the pre-screening process (see Sect.  XRF analyses). The lists of elements and compounds used in inorganic and organic PCA are shown in Tables S1 and S2 in the Supplement.

The stomach-oil deposit 3012MUM2 spans 1260 (1060–1420) cal.yrBP at 0 cm to 6410 (6250–6610) cal.yrBP at 18.5 cm (Table , Fig. a). When taking biomarker and isotope samples we avoided the margins of the deposit, which were easily deformable. As a result, the oldest and youngest biomarker and isotope samples lie at 0.5 cm (1830 (1170–2530) cal.yrBP) and 18.25 cm (6390 (6210–6590) cal.yrBP), respectively in the Bayesian model, or ∼ 1800 to 6400 cal.yrBP when rounded to closest 100 years. The accumulation rate based on the median Bayesian modelled age between radiocarbon dating depths varied between 8.6 and 125.0 mmkyr-1 (Fig. b).

The 3012MUM2 samples were high in organic C (36.6 %–68.3 %; mean 49.8 %) and total N (2.9 %–16.8 % (mean 9.7 %) (Fig. ). The C:Natomic ratio varied between 3.4 and 20.8 (mean 7.5). Bulk δ13C had a narrow range from -31.0 ‰ to -29.5 ‰, with a mean of -30.3 ‰ and a very small standard deviation (SD) of 0.3 ‰ (Fig. ). In contrast bulk δ15N had a wide range between 9.3 ‰ and 19.1 ‰ (mean 12.2 ‰, SD 1.6 ‰) (Fig. ).

Here and in the discussion, we present the majority of biomarkers as fluxes (Fig. ). Given similar patterns with biomarker concentrations (Figs. S2–S4) we consider the trends we have identified to be robust. The 3012MUM2 deposit samples likely comprised wax esters consistent with existing stomach oil studies (Imber, 1976; Lewis, 1966, 1969; Warham et al., 1976; Watts and Warham, 1976), and contributions from free lipids. Once saponified and derivatised, the extracts were dominated by fatty acids (FA) and alcohols (FAlc) (Fig. S3). In 3012MUM2, C16:0 was the most abundant fatty acid (mean 38.9 % ± 8.8 % SD), followed by C18:1(n-9) (FA) (mean 23.6 % ± 10.9 % SD), C14:0 (FA) (mean 22.1 % ± 5.2 % SD), C18:0 (FA) (mean 8.4 % ± 2.1 % SD) and C16:1 (FA) (mean 7.1 % ± 3.5 % SD).

By flux, FA were most abundant (C14:0, C16:0, C16:1, C18:0, C18:1(n-9)) (total mean 182.25 µgg-1cmyr-1, SD 116.69 µgg-1cmyr-1) followed by n-alkanols (C14:0, C16:0, C18:0, C20:0 and C22:0) (FAlc) (total mean 47.95 µgg-1cmyr-1, SD 44.33 µgg-1cmyr-1). The other compounds were less abundant, with cholesterol the highest (mean 9.95 µgg-1cmyr-1, SD 7.47 µgg-1cmyr-1) and phytol in lower concentrations (mean 0.31 µgg-1cmyr-1; 0.22 µgg-1cmyr-1). Phytol is formed from the ester-linked side chain of chlorophyll a and can therefore primarily be considered a biomarker for phytoplankton (Rontani and Volkman, 2003). Cholesterol is a ubiquitous marker but in this context could be considered a krill marker (both Antarctic and ice krill), since it can account for more than 76 % of total sterols in krill (Ju and Harvey, 2004) and is typically lower in concentration in fish (e.g. cholesterol in Dissostichus mawsoni ranges 4.7 %–14.3 % of total lipids; Clarke, 1984). We also assessed the distributions of sterols (e.g. 22-dehydrocholesterol), stanols (e.g. coprostanol) and cholesterol derivatives (e.g. cholesterol α/β-epoxide); cholesterol was the most abundant contributor (Fig. S4).

During saponification of the relevant wax esters, n-alkanols (FAlc) were also formed with saturated even chain lengths; C14:0 (FAlc) to C22:0 (FAlc) n-alcohols were the most abundant (Fig. S3). Amongst the n-alcohols C16:0 (FAlc) was the most abundant (mean 63 % ± 2.9 % SD), followed by C14:0 (FAlc) (mean 22.4 % ± 1.8 % SD) and C18:0 (FAlc) (mean 13.0 % ± 2.1 % SD). C20:0 (FAlc) (mean 1.3 % ± 0.4 % SD) and C22:0 (FAlc) (mean 0.3 % ± 0.1 % SD) n-alcohols were relatively minor contributors.

Broadly, PC axis 1 (30 % variance explained) for the organic geochemistry indicators had high positive loadings in C:N ratio and C18:0/C18:1(n-9) (FA), while PC axis 2 (24 % variance explained) had strong positive loadings in C16:0 (FA), C18:0 (FA), δ15N and C:N ratio (Table S2 and Fig. S9).

Cluster analysis identified three statistically significant organic zones (Org-A, B and C) which are used as a framework to discuss the changes in key biomarkers through time. Although clusters were identified from organic analyses, there were 8 zones of colour visible in 3012MUM2 (Fig. S1c). In terms of lithological units, zone Org-C closely matched with zone 8 (darkest layer, black/brown), with zones 4–7 matching most closely with zone Org-B (lighter, yellowy/orange) and zones 1–3 matching with Org-A (medium-dark, grey brown to black brown).

Cluster analysis identified three major XRF clusters: XRF-C (6390–4570 cal.yrBP); XRF-B (4570–4180 cal.yrBP); XRF-A (4180–1830 cal.yrBP). Zone XRF-C broadly coincides with organic zone Org-C and B, with XRF-B aligning (although slightly offset) with the uppermost part of zone Org-B. Zones XRF-A and zone Org-A are broadly aligned.

Based on XRF mean counts per second (cps) the largest elemental contributors to stomach-oil deposit 3012MUM2 were Cl, Ca, Fe, S, K, Br and P. Key XRF-derived inorganic compositions normalised to accumulation rate are presented in Fig. . Fluxes are characterised by higher variability in XRF-C with peaks in multiple elements at ∼ 6300, ∼ 5300–4900 cal.yrBP and at the top of the zone (∼ 4570 cal.yrBP). In XRF-B values reach a peak at ∼ 4300 cal.yrBP. Elemental fluxes are lower and relatively stable in XRF-A, with a small peak in some elements such as Fe ∼ 2400 cal.yrBP.

Based on the PCA of counts per second data, PC1 reflected 28 % of the variance (Table S1; Fig. S10) and included Si, Ba Al and Fe, which tend to peak around ∼ 6300 cal.yrBP in zone XRF-C (Fig. S5). PC2 reflected 22 % of variance (Table S1; Fig. S10), and included S, Cl, Br, P and Cu, which shift towards lower values and lower variability in XRF-A (Fig. ).

High levels of organic C and N in 3012MUM2 were similar to previously measured snow petrel stomach oil deposits (e.g. McClymont et al., 2022; Berg et al., 2023; Hiller et al., 1988), while bulk δ13C values were within the range of previous measurements of a Holocene deposit (Ainley et al., 2006). The fatty acid (FA) distributions found in the stomach oils are commensurate with previous stomach-oil and source end-member studies which suggest a diet of krill (mainly C14:0 (FA), C16:0 (FA)), squid (C16:0 (FA)) and fish (C16:0 (FA), C18:1(n-9) (FA)) (Cripps et al., 1999; Lewis, 1966) as summarised in McClymont et al. (2022) and further explored in Berg et al. (2023). While some FAs are labile and susceptible to degradation, particularly unsaturated FAs (Stefanova and Disnar, 2000), the similar profiles shown by C14:0 and C18:0 FA with C18:1(n-9) FA suggests that changes in diet, not preservation, is the main driver for the latter FA (Fig. S3).

In the PCA biplots for the XRF analysis (Table S1 and Fig. S10) negative loadings from elements which are commonly associated with seabirds (e.g. P, Zn, Sr, Ni), including Cu (Shatova et al., 2016; Shatova et al., 2017; Castro et al., 2021; Sparaventi et al., 2021) support our interpretation of dietary sources to the deposit (Duda et al., 2021). In contrast, the positive loading of Fe, Al, Si and Ca on PC1 aligns with the main elements measured in the rock specimen taken from 3012MUM2 (Table S3), which reflects the local gneissic granite bedrock (Juckes, 1972). These elements are interpreted to reflect locally-derived, probably bedrock, erosional contribution. Cu was also present in 3012MUM2, which is known to be a key Antarctic krill marker as a Cu backbone structure is found in hemocyanin (Bridges, 1983) and has been observed in previous snow petrel deposits to be associated with Antarctic krill (McClymont et al., 2022). Broadly elevated Cu in zone Org-B with elevated C14:0 FA and cholesterol (Fig. ) also supports an Antarctic krill source for these components.

A wide range of δ15N values is recorded, which could reflect environmental change (e.g. changing circulation, degradation) or dietary change (e.g. baseline values in phytoplankton, trophic level). Past variations in the δ15N of circumpolar deep water (CDW) which upwells to the sea surface can contribute to high δ15N (Kemeny et al., 2018), but it is so far unknown if this occurred during the timescale of our deposit between 6390 and 1830 cal.yrBP (as opposed to glacial-interglacial timescales). Fragmented coral δ15N records of upper CDW indicate mid- to late Holocene values spanning ∼ 7 ‰–11 ‰ (Chen et al., 2023) but without the resolution to compare shorter-term changes with the variations we observed in 3012MUM2. Diatom-bound records of baseline δ15N show a long-term decline through the Holocene as nutrient availability has increased (e.g. Ai et al., 2020; Studer et al., 2018), which aligns in part with our overall trend. Previous work has cautioned that guano could contribute to stomach-oil deposits (Berg et al., 2019), which would also elevate δ15N (e.g. Wainright et al., 1998; Bokhorst et al., 2007). Finally, weathering would also elevate δ15N values due to microbial biosynthetic pathways causing 14N release (Macko and Estep, 1984). However, we only identified minor contributions from microbial fatty acids in 3012MUM2 to support this interpretation. In our discussion below, we focus on the environmental and dietary information provided by δ15N in combination with other proxy indicators.

We focus mainly on organic compounds for interpretation of palaeoclimate because dietary lipids and organic isotopes reflect the main snow petrel dietary sources which can, in turn, provide secondary information on sea ice distribution (e.g. McClymont et al., 2022). In particular, we focus on C14:0 (FA) as an indicator of past Antarctic krill contributions to the diet, C18:0 (FA) for fish, C18:1(n-9) (FA) for mainly fish, bulk δ15N as primarily an indicator of trophic status and productivity, and phytol as an indicator of past productivity. In terms of past ice configurations, we interpret a diet with more Antarctic krill and high productivity as recording snow petrels feeding in the open ocean (within the MIZ or the open ocean) (e.g. zone Org-C). Zones of higher productivity with reduced Antarctic krill contributions are inferred to represent enhanced snow petrel feeding closer to shore, in neritic zones, either due to a more proximal MIZ or in coastal polynyas along the Antarctic ice sheet margin (e.g. zone Org-B). While zones with lower productivity and accumulation rates reflect sea ice expansion, a less accessible MIZ and a more dense ice pack (e.g. zone Org-A).

By the start of the 3012MUM2 record at ∼ 6400 cal.yrBP, the Filchner and Ronne Ice Shelf had reached its modern position (Nichols et al., 2019; Johnson et al., 2019; Hillenbrand et al., 2017; Grieman et al., 2024). In other parts of Antarctica, cooling of the surface ocean and freshening of shelf surface waters since the middle-Holocene (Ashley et al., 2021) led to increased sea-ice concentrations and sea-ice duration over the continental shelf (Crosta et al., 2008; Mezgec et al., 2017; Johnson et al., 2021). Further offshore, sea ice became less extensive since ∼ 6500 cal.yrBP (Nielsen et al., 2004; Xiao et al., 2016), reflecting shifts in the latitudinal insolation and thermal gradients (Denis et al., 2010) as well as the multi-centennial expression of climate modes, such as ENSO (El Niño-Southern Oscillation) (Crosta et al., 2021). For the wider Southern Ocean, exemplified in East Antarctica, after 4500 cal.yrBP although sea ice and turbulence persisted in locations proximal to the ice shelf, it did not increase further offshore due to northward transport of subpolar surface waters as a response to southern Westerlies reinforcement (Denis et al., 2010).

The evolution of Holocene sea ice has been separated into three distinct phases around Antarctica (Early Holocene ∼ 11.5 to ∼ 8 kaBP; mid-Holocene ∼ 7 to ∼ 4–3 kaBP; the late Holocene ∼ 5–3 to 1–0 kaBP), with the phasing of these periods differing depending on regional response to long-term forcing (Crosta et al., 2022). Despite this wider knowledge there is an absence of any paleoenvironmental reconstructions of sea ice for the Holocene within the Weddell Sea region. Here we explore how each of the three statistically significant geochemical zones in the stomach oil deposit align with the timing or direction of the environmental shifts recorded by Crosta et al. (2022) and similar studies elsewhere in Antarctica.

Our records show associations between inference of past sea ice configurations made from our snow petrel stomach oil deposit and regional records of environmental change. For example, there are similarities and differences with records in the wider Weddell Sea region, likely due to the influence of the Weddell Gyre and atmospheric processes that determine sea ice configurations. Our interpretations of high productivity and krill between 6390 and 5960 cal.kyrBP (zone Org-C, Figs.  and 5), correspond with a relatively productive and an open-ocean (pelagic) type sea ice configuration from 7.2 cal.kyrBP at Herbert Sound and Croft Bay, in response to a wider regional mid-Holocene warming (e.g. core NBP0502-Site 2) (Totten et al., 2015). In the Firth of Tay (Fig. ), this mid-Holocene climatic optimum spans 7800–6000 cal.yrBP (core NBP0602A) (Michalchuk et al., 2009) and 7750 and 6000 yrBP (cores NBP0602A-8B and NBP0703-JPC02) (Majewski and Anderson, 2009), and was attributed to seasonally-open marine conditions. Scotia Sea diatom records are also in phase with our results, with seasonally open water assemblages from 8300 to 2400 yrBP (core SS01) (Bak et al., 2007).

Our interpretation of zone Org-A having more sea ice corresponds with HBI records at the Vega Drift in JPC38, when interpreted carefully (Fig. d) (Barbara et al., 2016). Higher HBI diene levels in our deposit zone Org-A (4320–1830 cal.yrBP) correspond to increased sea ice cover. The fact that HBI:2 / HBI:3 ratios vary and are low in zone Org-A are probably due to limitations of the proxy and the issue of normalization, linked with variable pelagic productivity. At Bransfield Strait (core D1–7) persistent sea ice is observed between 5.8 and 3.8 kaBP (core ANT28/D1–7) (Nie et al., 2022).

The neoglacial, sea-ice expansion and cooling phase suggested by zone Org-A is coherent with a sediment record on the Firth of Tay (core NBP0602A-8B and NBP0703-JPC02) which shows cooling from ∼ 3500 yrBP based on foraminifera (Majewski and Anderson, 2009). At Perseverance Drift, the warm interval persists longer than in our deposit, with high abundance of the foraminifera Globocassidulina spp. between 3400 and 1800 yrBP indicating incursions of Weddell Sea Transitional Water and a period of “freshening” consistent with open-marine or seasonally open marine conditions (Kyrmanidou et al., 2018) (cores JKC36 and JPC36), indicative of cooling.

The boundary between zone Org-B and A (4320 cal.yrBP) aligns with the transition to neoglacial conditions (e.g. Crosta et al., 2022). This mid-Holocene neoglacial is consistent with the James Ross Island (JRI) ice core temperature decline (Fig. a) (Mulvaney et al., 2012) and decreasing δ18O signal in 5 Antarctic ice cores (Hodgson and Bentley, 2013; Masson-Delmotte et al., 2011) (Fig. c). We find that the onset of the neoglacial is remarkably in phase with a switch in sea ice biomarkers (both HBI diene and (HBI:2/HBI:3) ratio) indicative of more sea ice in the distant Adélie region in core U1357 (Ashley et al., 2021) (Fig. e). A marked transition is also seen in South Atlantic cores (Hodell et al., 2001) where IRD (as % lithics) increase markedly from ∼ 5 kaBP suggesting cooling waters (together with concomitant changes in diatoms SST index and δ18O on diatoms) which have been linked to the arrival of more sea ice from the Weddell Sea region (core TTN057-13-PC4). Total yearly insolation (e.g. at 65° S) features an ongoing decline during this period (Fig. b) but is unlikely to be the sole driver of the neoglacial (Divine et al., 2010; Renssen et al., 2005).