We analysed a set of 41 surface sediment samples (0–1 cm sub-bottom depth) from different areas of the Southern Ocean (Fig. 1) retrieved by multicorers and giant box corers during RV Polarstern expeditions over the past 15 years. Sixteen surface sediment samples from the Amundsen Sea continental shelf were collected during expeditions PS69 in 2006 (Gohl, 2007) and PS104 in 2017 (Gohl, 2017). Twenty-five surface sediment samples from the southeastern and southwestern Weddell Sea continental shelf were collected during expeditions PS111 in 2018 (Schröder, 2018) and PS118 in 2019 (Dorschel, 2019). This new data set was complemented by data from 26 surface sediment samples collected in Bransfield Strait/WAP, which were previously published by Vorrath et al. (2019).

The sediment material was freeze-dried and homogenized with an agate mortar and stored in glass vials at -20 ∘C before and after the above-mentioned initial preparation steps in order to avoid degradation of targeted molecular components. Total organic carbon (TOC) contents were measured on 0.1 g of sediment after removing inorganic carbon (total inorganic carbon, carbonates) with 500 µL 12 N hydrochloric acid. TOC contents were determined with a carbon–sulfur analyser (CS 2000; Eltra) with standards for calibration being routinely measured before sample analysis and after every 10th sample (error ±0.02 %).

Lipid biomarkers were extracted from the sediments (4 g for PS69 and PS104, and 6 g for PS111 and PS118) by ultrasonication (3×15 min) using DCM:MeOH (3×6 mL for PS69 and PS104, and 3×8 mL for PS111 and PS118; 2:1 v/v) as solvent. Prior to this step, the internal standards 7-hexylnonadecane (7-HND; 0.038 µg per sample for PS69 and PS104, 0.057 µg per sample for PS111 and PS118), 5α-androstan-3-ol (1.04 µg per sample) and C46 (0.98 µg per sample) were added to the sample for quantification of HBIs, sterols and GDGTs respectively. Via open-column chromatography, with SiO2 as stationary phase, fractionation of the extract was achieved by eluting the apolar fraction (HBIs) and the polar fraction (sterols and GDGTs) with 5 mL n-hexane and 5 mL DCM:MeOH 1:1 respectively. The polar fraction was subsequently split into two fractions (sterols and GDGTs) for further processing. The sterol fraction was silylated with 300 µL bis-(trimethylsilyl)-trifluoroacetamide (BSTFA; 2 h at 60 ∘C). Compound analyses of HBIs and sterols were carried out on an Agilent Technologies 7890B gas chromatograph (GC; fitted with a 30 m DB 1MS column; 0.25 mm diameter and 0.25 µm film thickness) coupled to an Agilent Technologies 5977B mass selective detector (MSD; with 70 eV constant ionization potential, ion source temperature of 230 ∘C). The GC oven was set to 60 ∘C (3 min), 150 ∘C (rate: 15 ∘C min-1), 320 ∘C (rate: 10 ∘C min-1) and 320 ∘C (15 min isothermal) for the analysis of hydrocarbons and to 60 ∘C (2 min), 150 ∘C (rate: 15 ∘C min-1), 320 ∘C (rate: 3∘C min-1) and 320 ∘C (20 min isothermal) for the analysis of sterols. Helium was used as the carrier gas. The HBI and sterol compounds were identified by their GC retention times and mass spectra (Belt et al., 2018, 2000; Boon et al., 1979). Lipids were quantified by setting the individual, manually integrated, gas chromatograph–mass spectrometer peak area in relation to the peak area of the respective internal standard and normalization to the amount of extracted sediment. IPSO25 and HBI-trienes were quantified by relating their molecular ions (IPSO25, m/z 348, and HBI-trienes, m/z 346) to the fragment ion m/z 266 of the internal standard 7-HND (Belt, 2018). Sterols were quantified by comparing the molecular ion of the individual sterol with the molecular ion m/z 348 of the internal standard 5α-androstan-3-ol. Instrumental response factors for the target lipids were considered as recommended by Belt et al. (2014) and Fahl and Stein (2012). All biomarker concentrations were subsequently normalized to the TOC content of each sample to account for different depositional settings within the different study areas.

For calculating the phytoplankton IPSO25 (PIPSO25) index, we used the equation introduced by Vorrath et al. (2019): 1PIPSO25=IPSO25/(IPSO25+(phytoplankton marker×c)), where c (c = mean IPSO25/mean phytoplankton marker) is applied as a concentration balance factor to account for high concentration offsets between IPSO25 and the phytoplankton biomarker (see Table 1 in the Supplement for c factors of individual PIPSO25 calculations).

Following the approach by Müller and Stein (2014) and Lamping et al. (2020), a PIPSO25 value of 1 was assigned to samples with exceptionally low (at the detection limit) concentrations of both biomarkers (see chap. 4.1.2). This comprises the five Weddell Sea samples PS111/13-2, /15-1, /16-3, /29-3 and /40-2 (marked as an orange x in Fig. 1).

The GDGT fraction was dried under N2, redissolved with 120 µL hexane : isopropanol (v/v 99:1) and then filtered using a polytetrafluoroethylene (PTFE) filter with a 0.45 µm pore size membrane. GDGTs were measured using a high-performance liquid chromatograph (HPLC; Agilent 1200 series HPLC system) coupled to an Agilent 6120 mass spectrometer (MS), operating with atmospheric pressure chemical ionization (APCI). The injection volume was 20 µL. For separating the GDGTs, a Prevail Cyano 3 µm column (Grace, 150 mm × 2.1 mm) was kept at 30 ∘C. Each sample was eluted isocratically for 5 min with solvent A (hexane/2-propanol/chloroform; 98:1:1) at a flow rate of 0.2 mL min-1, and the volume of solvent B (hexane/2-propanol/chloroform; 89:10:1) was increased linearly to 10 % within 20 min and then to 100 % within 10 min. The column was back-flushed (5 min, flow rate of 0.6 mL min-1) after 7 min and after each sample and was re-equilibrated with solvent A (10 min, flow rate of 0.2 mL min-1). The APCI was set to the following: N2 drying gas flow at 5 L min-1 and temperature to 350 ∘C, nebulizer pressure to 50 psi, vaporizer gas temperature to 350 ∘C, capillary voltage to 4 kV and corona current to +5 µA. Detection of GDGTs was achieved by means of selective ion monitoring (SIM) of [M + H]+ ions (dwell time 76 ms). GDGT-0 (m/z 1302), GDGT-1 (m/z 1300), GDGT-2 (m/z 1298), GDGT-3 (m/z 1296) and crenarchaeol (m/z 1292) as well as brGDGT-III (m/z 1050), brGDGT-II (m/z 1036) and brGDGT-I (m/z 1022) were quantified by relating their molecular ions to the molecular ion m/z 744 of the internal standard C46-GDGT. The late eluting hydroxylated GDGTs (OH-GDGT-0, OH-GDGT-1 and OH-GDGT-2 with m/z 1318, 1316 and 1314 respectively) were quantified in the scans (m/z 1300, 1298 and 1296) of their related GDGTs, as described by Fietz et al. (2013).

TEX86L values and their conversion into SOTs were determined following Kim et al. (2012): 2TEX86L=LOG[GDGT-2]GDGT-1+GDGT-2+GDGT-3,3SOTTEX[∘C]=50.8×TEX86L+36.1. Temperature calculations based on OH-GDGTs were carried out according to Lü et al. (2015): 4RI-OH'=OH-GDGT-1+2×[OH-GDGT-2]OH-GDGT-0+OH-GDGT-1+[OH-GDGT-2],5SSTOH∘C=RI-OH′-0.1/0.0382. To determine the relative influence of terrestrial organic matter input, the branched isoprenoid tetraether (BIT) index was calculated following Hopmans et al. (2004): 6BIT=brGDGT-I+brGDGT-II+[brGDGT-III]Chrenarchaeol+brGDGT-I+brGDGT-II+[brGDGT-III].

Satellite sea-ice data were derived from Nimbus-7 Scanning Multichannel Microwave Radiometer (SMMR) and Defence Meteorological Satellite Program (DMSP) Special Sensor Microwave Imager/Sounder (SSM/I-SSMIS) passive microwave data and downloaded from the National Snow and Ice Data Center (NSIDC; Cavalieri et al., 1996). The sea-ice data represent mean monthly SIC values, which are expressed to range from 0 % to 100 % and are averaged over a period from the beginning of satellite observations in 1978 CE to the individual year of sample collection. The monthly mean SIC values were then split into different seasons: winter (JJA), spring (SON) and summer (DJF) (Fig. 2a, b, c), and these data are considered to represent the recent mean state of sea-ice coverage. All satellite data are provided in Table S3.

Modern annual mean SSTs and SOTs were derived from the World Ocean Atlas 2013 and represent averaged values for the years 1955–2012 CE (WOA13; Locarnini et al., 2013).

TOC contents in marine sediments are often viewed as an indicator for primary productivity in surface waters (Meyers, 1997). However, we are aware that additional factors, such as different water depths and depositional regimes, may exert control on sedimentary TOC as well. The TOC contents of the investigated surface samples are lowest in Drake Passage with values of around 0.12 %–0.54 % and increase from northwest to southeast into Bransfield Strait, where they range from 0.59 % to 1.06 % (WAP, Fig. 3a). Along the EAP, higher TOC contents (0.57 %–0.86 %) prevail around the former Larsen A Ice Shelf and north of James Ross Island, but they decrease towards Powell Basin (0.22 %–0.37 %) and then increase to 0.50 % around the South Orkney Islands, which may point to elevated productivity or an enhanced supply of reworked terrigenous organic matter in this area (EAP, Fig. 3a).

At the WAP, concentrations of the sea-ice biomarker IPSO25 increase from northwest to southeast. IPSO25 is absent in samples from the permanently ice-free Drake Passage and increases towards the continental slope and the seasonally ice-covered shelf (0.37–17.81 µg g OC-1; Fig. 3b; Vorrath et al., 2019). The highest IPSO25 concentrations are observed in samples of the northern Bransfield Strait. Here, the inflow of waters from the Weddell Sea transports sea ice into Bransfield Strait (Vorrath et al., 2019). Elevated IPSO25 concentrations are also observed at the seasonally sea-ice-covered EAP, where relatively high concentrations of the sea-ice biomarker prevail in samples located in the area of the former Larsen A Ice Shelf and north of James Ross Island (12.59–17.74 µg g OC-1; Fig. 3b). Because these locations are influenced by the northward drift of sea ice within the Weddell Gyre (Fig. 1), the elevated IPSO25 concentrations could also result from sea ice advected from the southern Weddell Sea. We suggest that the decrease in IPSO25 concentrations towards the Powell Basin and the South Orkney Islands (0.59–5.36 µg g OC-1; Fig. 3b) is connected to warmer ocean temperatures in the north and reduced sea-ice cover during spring.

Concentrations of the phytoplankton biomarker HBI Z-triene around the Antarctic Peninsula are highest in eastern Drake Passage and along the WAP continental slope (where IPSO25 is absent) and decrease in Bransfield Strait (0.33–26.86 µg g OC-1; Fig. 3c; Vorrath et al., 2019). Elevated HBI Z-triene concentrations have, so far, been detected in surface waters along the sea-ice edge (Smik et al., 2016) and, hence, were suggested to be a proxy for marginal ice zone conditions (Belt et al., 2015; Collins et al., 2013; Schmidt et al., 2018). Vorrath et al. (2019), however, relate the high concentrations of HBI Z-triene at the northernmost stations in the permanently ice-free eastern Drake Passage to their proximity to the Antarctic Polar Front. Here, productivity of the source diatoms of HBI-trienes (e.g. Rhizosolenia spp.; Belt et al., 2017) may be enhanced by meander-induced upwelling leading to increased nutrient flux to surface waters (Moore and Abbott, 2002). As Cárdenas et al. (2019) document only minor abundances of Rhizosolenia spp. in seafloor surface sediments from this area, we assume that HBI-trienes might also be biosynthesized by other diatom taxa. Moderate concentrations along the continental slope of the WAP and in Bransfield Strait were associated with elevated inflow of warm BSW which leads to a retreating sea-ice margin during spring and summer (for more details, see Vorrath et al., 2019, 2020). Samples from the EAP shelf and Powell Basin are characterized by relatively low HBI Z-triene concentrations (0.1–2.37 µg g OC-1; Fig. 3c) that decrease from southwest to northeast, whereas the northernmost sample closest to the South Orkney Islands is characterized by an elevated HBI Z-triene concentration of ∼ 8.49 µg g OC-1 (EAP, Fig. 3c). This relatively high concentration may be related to an “island mass effect”, coined by Doty and Oguri (1956), which refers to increased primary production around oceanic islands in comparison to surrounding waters. Nolting et al. (1991) found extraordinarily high dissolved iron levels (as high as 50–60 nM) on the South Orkney shelf, while Nielsdóttir et al. (2012) observed enhanced iron and Chl-a concentrations in the vicinity of the South Orkney Islands. These authors explain the increased dissolved iron levels with input from seasonally retreating sea ice, which is recorded by satellites (Fig. 2a, b, c) and probably leads to substantial annual phytoplankton blooms, which may also cause the elevated TOC content in the corresponding seafloor sediment sample (Fig. 3a). Alternatively, remobilization of shelf sediments or vertical mixing of iron-rich deep waters, leading to high iron contents in surface waters, may stimulate primary productivity (Blain et al., 2007; de Jong et al., 2012). However, it remains unclear why the brassicasterol concentration is distinctly low in this sample, and we assume that different environmental preferences of the source organisms may account for this. In Drake Passage and along the EAP, brassicasterol displays a similar pattern to HBI Z-triene, with relatively high concentrations (more than 2 orders of magnitudes) ranging from 1.86 to 5017.44 µg g OC-1 (Fig. 3d).

In the Weddell Sea, TOC contents are generally low (< 0.4 %), with slightly elevated values in the west (up to 0.50 %) and right in front of the Filchner Ice Shelf (up to 0.52 %; Fig. 3a). The Amundsen Sea is characterized by slightly higher TOC contents, with concentrations of up to 0.91 % in the west and lower values in the east (0.33 %; AS, Fig. 3a).

In the samples from the Amundsen and Weddell seas, which are both dominated by strong winter sea-ice cover lasting until spring (Fig. 2a, b, c), all three biomarkers are present in low concentrations only. An exception are the samples located in front of the Filchner Ice Shelf with significantly higher concentrations of IPSO25 (7.09–73.87 µg g OC-1; WS, Fig. 3b). Concentrations of IPSO25 on the Amundsen Sea shelf are relatively low (0.04–3.3 µg g OC-1), with slightly higher values observed in the northeast (AS, Fig. 3b). HBI Z-triene concentrations are also very low but are slightly higher in Filchner Trough (0.04–1 µg g OC-1) and at more distal locations on the northeastern Amundsen Sea shelf (0.01–1.88 µg g OC-1; Fig. 3c). Brassicasterol generally shows a similar pattern to HBI Z-triene, with concentrations varying between 1.86 and 220.54 µg g OC-1 (Fig. 3d; for HBI E-triene and dinosterol distribution, see Fig. S1).

The PIPSO25 index combines the relative concentrations of IPSO25 and a selected phytoplankton biomarker, such as HBI-trienes and sterols, as an indicator for an open-ocean environment (Vorrath et al., 2019). The combination of both endmembers (sea ice vs. open ocean) prevents misleading interpretations regarding the absence of IPSO25 in the sediments, which can be the result of two entirely different scenarios. Under heavy/perennial sea-ice coverage, the thickness of sea ice hinders light penetration, thereby limiting the productivity of algae living in basal sea ice (Hancke et al., 2018). This scenario can cause the absence of both phytoplankton and sea-ice biomarkers in the sediment. The other scenario depicts a permanently open ocean, where the sea-ice biomarker is absent as well, but here the phytoplankton biomarkers are present in variable concentrations (Müller et al., 2011). The presence of both biomarkers in the sediment is indicative of seasonal sea-ice coverage and/or the occurrence of stable sea-ice margin conditions, promoting biosynthesis of both biomarkers (Müller et al., 2011). Here, we distinguish between PZIPSO25 and PBIPSO25 using HBI Z-triene and brassicasterol as a phytoplankton biomarker respectively (Fig. 4a, b; for PIPSO25 values based on HBI E-triene and dinosterol, see Table S1 and Fig. S2).

Both PIPSO25 indices are 0 in the predominantly ice-free Drake Passage and increase towards southeast to intermediate values on the WAP slope and around the South Shetland Islands, reflecting increased influence of marginal sea-ice cover towards the coast (0.02–0.70; Vorrath et al., 2019). At the seasonally sea-ice-covered EAP, PZIPSO25 values reach 0.84, while lower values of around 0.25 are observed close to the South Orkney Islands, which is caused by the elevated HBI Z-triene concentrations at the stations there (EAP, Fig. 3c). The PBIPSO25 index exhibits even higher values of up to 0.98 at the EAP/northwestern Weddell Sea. These elevated PIPSO25 indices align well with the significant northward sea-ice drift within the Weddell Gyre, which leads to prolonged sea-ice cover along the EAP.

In samples from the southern Weddell Sea, both PIPSO25 indices show a similar pattern with high values of up to 0.9, and slightly lower values are observed in front of the Brunt Ice Shelf (0.6; Fig. 4a, b). Very low concentrations (close to the detection limit) of both biomarkers in samples from the continental shelf off Dronning Maud Land (Fig. 1) result in low PIPSO25 values, strongly underestimating the sea-ice cover in this area, where satellite-derived sea-ice data document severe seasonal sea-ice cover (Fig. 2). As previously mentioned, we followed the approach by Müller and Stein (2014) and Lamping et al. (2020) by assigning a maximum PIPSO25 value of 1 to these samples to circumvent misleading interpretations and to aid visualization.

The intermediate PIPSO25 value (∼ 0.51) derived for one sample collected in front of the Brunt Ice Shelf points to less severe sea-ice cover in that area. A possible explanation for the relatively low PIPSO25 value is the presence of a coastal polynya that has been reported by Anderson (1993) and which is further supported by Paul et al. (2015). These authors note that the sea-ice area around the Brunt Ice Shelf is the most active in the southern Weddell Sea, with an annual average polynya area of 3516 ± 1420 km2. Interestingly, the reduced SIC here is also captured by our model (see Sect. 4.3).

PIPSO25 values in the Amundsen Sea point to different scenarios. The PZIPSO25 index varies around 0.9, with only the easterly, more distal samples having lower values between 0.3 and 0.6 (Fig. 4a). The PBIPSO25 index generally has lower values, ranging from 0.6 in the coastal area to 0.2 in the more distal samples (Fig. 4b). This difference between PZIPSO25 and PBIPSO25 may be explained by the different source organisms biosynthesizing the individual phytoplankton biomarkers. While the main origin of HBI-trienes seems to be restricted to diatoms (Belt et al., 2017), brassicasterol is known to be produced by several algal groups that are adapted to a wider range of sea surface conditions (Volkman, 2006; see Sect. 5.2).

The main ice algae bloom in the Southern Ocean occurs during spring, when solar insolation and air temperatures/SSTs increase and sea ice starts to melt, which results in the release of nutrients and stratification of the water column, stimulating the productivity of photosynthesizing organisms (Arrigo, 2017; Belt, 2018). Hence, the sea-ice biomarker IPSO25 is commonly interpreted as a spring sea-ice indicator, which is why, in the following, we compare the biomarker-based sea-ice reconstructions to satellite-derived and modelled spring SIC. IPSO25 concentrations in the surface sediments around the Antarctic Peninsula exhibit similar trends to the satellite-derived and modelled SIC (Figs. 3, 4), whereas they differ significantly in the Amundsen and Weddell seas, where high SIC values are recorded by satellites and the model but IPSO25 is present in low concentrations. The low IPSO25 concentrations in these areas highlight the uncertainty, when considering IPSO25 as a sea-ice proxy alone, as such low concentrations are not only observed under open-water conditions, but also under severe sea-ice cover. In the Amundsen and Weddell seas, the low IPSO25 concentrations are the result of the latter, where limited light availability hinders ice algae growth, leading to an underestimation of sea-ice cover. Accordingly, we note a weak correlation between IPSO25 data and satellite SIC (R2=0.19; Fig. 5a). As stated above, the combination of IPSO25 and a phytoplankton marker may prevent this ambiguity. The higher sea-ice concentrations in the Amundsen and Weddell seas are better reflected by maximum PZIPSO25 values than by IPSO25 alone. However, we note that the PZIPSO25 index apparently does not resolve SIC values higher than 50 % (see Fig. S3), which may indicate a threshold (here ∼ 50 % SIC) where the growth of the HBI-triene- and IPSO25-producing algae is limited.

In general, however, the PZIPSO25 values correlate much better with satellite and modelled SIC (R2=0.78 and R2=0.76 respectively; Fig. 5b, d) than IPSO25 concentrations. Correlations of satellite and model data with PIPSO25 calculated using HBI E-triene, brassicasterol and dinosterol respectively, are also positive but less significant (Fig. S4); hence, we focus the discussion on PZIPSO25. The AWI-ESM2-derived spring SIC values correctly display the permanently ice-free Drake Passage and the northwest–southeast increase in sea-ice cover from the WAP continental slope towards Bransfield Strait (Fig. 4d). The model, however, significantly underestimates the elevated sea-ice concentrations (up to 70 %) in front of the former Larsen Ice Shelf A and east of James Ross Island at the EAP observed in satellite data. In the Amundsen and Weddell seas, the model predicts heavy sea-ice cover (∼ 90 %), only slightly underestimating the sea-ice cover at the near-coastal sites in front of Pine Island Glacier and Ronne Ice Shelf. Interestingly, modelled SIC in front of Brunt Ice Shelf is as low as ∼ 45 % (Fig. 4d, e), corresponding well to the reduced PZIPSO25 value of ∼ 0.51. This may reflect the polynya conditions in that region documented by Anderson (1993) and Paul et al. (2015). Overall, we note that modelled modern SIC values correlate well with satellite data (R2=0.73; Fig. 5c) and PZIPSO25 values (R2=0.76; Fig. 5d), whereas we observe weaker correlations between modelled palaeo-SIC values and PZIPSO25 values (Fig. S5; see Sect. 5.1).

For a critical appraisal of the applicability and reliability of GDGT indices as temperature proxies at polar latitudes, we here focus on the TEX86L proxy by Kim et al. (2012), which potentially reflects SOTs, and the RI-OH′ proxy by Lü et al. (2015), which is assumed to reflect SSTs. The reconstructions are believed to represent annual mean ocean temperatures (for correlations of TEX86L-derived SOTs with WOA spring and winter SOTs, see Fig. S6). In all samples, the BIT index (Eq. 6) is <0.3, indicating no significant impact of terrestrial input of organic material on the distribution of GDGTs and, hence, their reliability as temperature proxy. RI-OH′-derived temperatures and TEX86L-derived SOTs both show a similar pattern, but different temperature ranges between -2.62 to +4.67 ∘C and -2.38 to +8.75 ∘C respectively (Fig. 6a, b). At the WAP, RI-OH′- and TEX86L-derived temperatures increase northwestwards across the Antarctic continental slope and into the permanently ice-free Drake Passage, which are influenced by the ACC and relatively warm CDW (Orsi et al., 1995; Rintoul et al., 2001). Temperatures decrease towards Bransfield Strait and the EAP, which are influenced by seasonal sea-ice cover and relatively cold water from the Weddell Sea that branches off the Weddell Gyre (Collares et al., 2018; Thompson et al., 2009). At the EAP, a southwestward decrease is observed, with relatively low temperatures at the former Larsen A Ice Shelf and higher temperatures recorded in Powell Basin and around the South Orkney Islands (Fig. 6a, b).

In the Amundsen and Weddell seas further south, reconstructed temperatures are generally lower than around the Antarctic Peninsula. Samples from the Weddell Sea display a temperature decrease from east to west, which may reflect the route of eddies in the northeastern Weddell Gyre. These eddies carry relatively warm, salty CDW westward along the southern limb of the Weddell Gyre, where it becomes WDW (Vernet et al., 2019). The coldest TEX86L and RI-OH′ temperatures (<0 ∘C) at sites along the Filchner–Ronne Ice Shelf front may be further linked to the presence of cold precursor water masses for WSBW.

With respect to ongoing discussions regarding whether GDGT-based temperature reconstructions represent SSTs or SOTs (Kalanetra et al., 2009; Kim et al., 2012; Park et al., 2019), we compare our RI-OH′ and TEX86L-derived temperatures with surface and subsurface temperature data obtained by in situ measurements and modelling (Fig. 6c, d, e, f). Comparison of GDGT-derived temperatures with WOA13 temperatures from different water depths reveals the most significant correlation for a water depth of 410 m (for respective correlations, see Fig. S7). Hence, when discussing instrumental and modelled SOTs, we refer to 410 m water depth.

While the correlation between TEX86L-derived SOTs and instrumental SOTs is reasonably good (R2=0.66; Fig. 7a), also supporting a subsurface origin for the TEX86L proxy, we note a significant overestimation of SOTs by up to 6 ∘C in Drake Passage (Fig. S8). This warm-biased TEX86L signal is a known caveat and is, among others, assumed to be connected to GDGTs produced by deep-dwelling Euryarchaeota (Park et al., 2019), which have been reported in CDW (Alonso-Sáez et al., 2011) and in deep waters at the Antarctic Polar Front (López-García et al., 2001). Maximum TEX86L-based SOTs of 5–8 ∘C in the central Drake Passage (Fig. 6b), however, distinctly exceed the common temperature range of CDW (0–2 ∘C). Interestingly, TEX86L-derived SOTs in the colder regions of the Amundsen and Weddell seas relate reasonably well to instrumental temperatures and are only slightly warm-biased (Fig. S8). Correlations between RI-OH′-derived temperatures and instrumental SSTs are weak (R2=0.43; Fig. 7b). Recently, in their study on surface sediments from Prydz Bay (East Antarctica), Liu et al. (2020) concluded that the RI-OH′ index also holds promise as a tool to reconstruct SOTs rather than SSTs. When correlating our RI-OH′-derived temperatures with instrumental SOTs, we similarly find a high correlation (R2=0.73; Fig. 7c), supporting the above-mentioned hypothesis. We further note that the RI-OH′ temperature range is much more realistic than the TEX86L range. This suggests that the addition of OH-isoGDGTs in the temperature index is a promising step towards reliable high-latitude temperature reconstructions and may improve our understanding of the temperature responses of archaeal membranes in Southern Ocean waters (Fietz et al., 2020; Park et al., 2019). Clearly, more data – ideally obtained from sediment traps, seafloor surface sediment samples and longer sediment cores – and calibration studies will help to further elucidate the applicability of the RI-OH′ and TEX86L temperature reconstructions.

Similar to the model-derived sea-ice data, we also evaluate the model's performance in depicting ocean temperatures (Fig. 6e, f). Modelled annual mean SSTs and SOTs are highest (with up to 5 and 3 ∘C respectively) in the permanently ice-free Drake Passage, which is influenced by the relatively warm ACC. Lower SSTs are predicted for the Antarctic Peninsula continental slope and Bransfield Strait (∼ 0.5 to 1 ∘C), coinciding with the increase in the duration of seasonal sea-ice cover in that area. At the EAP/northwestern Weddell Sea, modelled SSTs as well as SOTs increase from southwest to northeast towards Powell Basin. In the Amundsen and Weddell seas, annual mean SSTs are negative, with temperatures ranging from -1 to -0.5 ∘C, whereas SOTs are positive in the Amundsen Sea and negative in the Weddell Sea. Overall, we note that modelled SOTs reflect instrumental SOTs reasonably well (R2=0.76; Fig. 7d). Interestingly, while RI-OH′-derived SOTs relate better to instrumental SOTs (than TEX86L-based SOTs), a better correlation between TEX86L-derived SOTs and modelled SOTs (R2=0.64; Fig. 7e) and a weaker correlation with RI-OH′-derived temperatures (R2=0.55; Fig. 7f) is found.

Information on the actual age of the surface sediment samples is a major requirement determining their suitability to reflect modern sea surface conditions. When comparing sea-ice conditions or ocean temperatures estimated from biomarker data obtained from 0.5–1 cm thick surface sediment samples (easily spanning decades to millennia, depending on sedimentation rates) with satellite-derived sea-ice data or instrumental records (covering only the past ∼ 40 and 65 years respectively), the different time periods reflected in the data sets need to be considered when interpreting the results. To address the issue of lacking age constraints for most of the surface sediments investigated here, we also performed palaeoclimate simulations providing sea-ice concentration data for three time slices (2 ka, 4 ka and 6 ka BP; see Fig. S5) to evaluate if the surface sediments may have recorded significantly older environmental conditions. Correlations of PIPSO25 values with these (palaeo-) sea-ice concentrations are notably weaker (Fig. S5) than the correlations with recent (1951–2014 CE) SIC model output, which points to a young to modern age of the majority of the studied sediments. This is further supported by AMS 14C dating of calcareous microfossils and 210Pb dating of seafloor surface sediments from the Amundsen Sea shelf documenting recent ages for most sites (Hillenbrand et al., 2010, 2013, 2017; Smith et al., 2011, 2014, 2017; Witus et al., 2014) as well as modern 210Pb dates obtained for three multicores collected in Bransfield Strait (PS97/56, PS97/68 and PS97/72; Vorrath et al., 2020). AMS 14C dates obtained for nearby seafloor surface sediments in the vicinity of the South Shetland Islands and the Antarctic Sound revealed ages of 100 and 142 years BP respectively (Vorrath et al., 2019). As both uncorrected ages lie within the range of the modern marine reservoir effect (e.g. Gordon and Harkness, 1992), we still consider these two dates as recent. However, in an area that has been significantly affected by rapid climate warming over the past decades and a regionally variable sea-ice coverage, the age uncertainties for at least 14C-dated samples may easily lead to an over- or underestimation of biomarker-based sea-ice cover and ocean temperatures respectively, which needs to be taken into account for comparisons with instrumental data. The utilization of (palaeo-) model data may alleviate the lack of age control for each seafloor sediment sample to some extent. Nevertheless, we recommend that for a robust calibration of, for example, PIPSO25 values against satellite-derived sea-ice concentrations, only surface sediment samples with a modern age confirmed by 210Pb dating are incorporated.

Biomarkers have the potential to reveal the former occurrence of their producers, which requires knowledge of the source organisms. While there is a general consensus on Thaumarchaeota being the major source of iso-GDGTs (Fietz et al., 2020, and references therein) and diatoms synthesizing HBIs (Volkman, 2006), the main source of brassicasterol, which is not only found in diatoms but also in dinoflagellates and haptophytes (Volkman, 2006), remains unclear. Accordingly, the use of brassicasterol to determine the PIPSO25 index may introduce uncertainties regarding the environmental information recorded by this phytoplankton biomarker. A further aspect concerns the different chemical structures of HBIs and sterols, which raises the risk of a selective degradation (see Belt, 2018, Rontani et al., 2018, and Rontani et al., 2019, for detailed discussion) with potentially considerable effects on the PIPSO25 index. Regarding the different areas investigated in our study, spatially different microbial communities and varying depositional regimes, such as sedimentation rate, redox conditions and water depth, may also lead to different degradation patterns. This means that variations in the biomarker concentrations between different areas may not strictly reflect changes in the production of these compounds (driven by sea surface conditions) but may also relate to different degradation states. In particular, lower sedimentation rates and, thus, extended oxygen exposure times promote chemical alteration and degradation processes (Hedges et al., 1999; Schouten et al., 2013). However, it has been previously reported that the formation of mineral aggregates and fecal pellets often accelerates the transport of organic matter from the sea surface through the water column to the seafloor during the melting season, leading to a more rapid burial and, hence, better preservation of the organic compounds (Bauerfeind et al., 2005; Etourneau et al., 2019; Müller et al., 2011).

Another rather technical drawback concerning the use of the PIPSO25 index occurs when the concentrations of the sea-ice proxy IPSO25 and the phytoplankton marker are similarly low (due to unfavourable conditions for both ice algae and phytoplankton) or similarly high (due to a significant seasonal shift in sea-ice cover and/or stable ice edge conditions). This may lead to similar PIPSO25 values, although the sea-ice conditions are fundamentally different from each other. This scenario is evident for five sampling sites in the Weddell Sea (PS111/13-2, /15-1, /16-3, /29-3 and /40-2; Fig. 3b, c), where IPSO25 and the HBI Z-triene concentrations are close to the detection limit and PZIPSO25 values are very low, suggesting a reduced sea-ice cover. Satellite and model data, however, show that these sample locations are influenced by heavy, nearly year-round sea-ice cover. We conclude that biomarker concentrations of both biomarkers at or close to the detection limit need to be treated with caution. Here, we assigned a maximum PZIPSO25 value of 1 to those samples, and we note that such a practice always needs to be clarified when applying the PIPSO25 approach. Nonetheless, the coupling of IPSO25 with a phytoplankton marker provides more reliable sea-ice reconstructions. Regarding all these ambiguities, we recommend not only calculating the PIPSO25 index but also to carefully considering individual biomarker concentrations and, if possible, to utilizing other sea-ice proxies, such as data from well-preserved diatom assemblages (Lamping et al., 2020; Vorrath et al., 2019, 2020). While the PIPSO25 index is not yet a fully quantitative proxy for (palaeo-) sea-ice concentrations, several calibration iterations have been applied to the GDGT palaeothermometers (Fietz et al., 2020). As noted above, the observation of distinctly warm-biased TEX86L-derived SOTs calls for further efforts such as regional calibration studies and/or investigations of archaean adaptation strategies at different water depths and under different nutrient and temperature conditions.

The sympagic, tube-dwelling, diatom B. adeliensis is a common constituent of Antarctic sea ice and preferably flourishes in the relatively open channels of sub-ice platelet layers in near-shore locations covered by fast ice (Medlin, 1990; Riaux-Gobin and Poulin, 2004). Based on investigations of sea-ice samples from the Southern Ocean, Belt et al. (2016) detected this diatom species to be a source of IPSO25, which, according to its habitat, led to the assumption of the sea-ice proxy being a potential indicator for the presence of platelet ice. As stated above, B. adeliensis is not confined to platelet ice and is also observed in basal sea ice and described as well adapted to changes in the texture of sea ice during ice melt (Riaux-Gobin et al., 2013). Platelet ice formation, however, plays an important role in sea-ice generation along some coastal regions of Antarctica (Hoppmann et al., 2015, 2020; Lange et al., 1989; Langhorne et al., 2015). In these regions, CDW and High Saline Shelf Water (HSSW) flow into sub-ice-shelf cavities of ice shelves and cause basal melting and the discharge of cold and less saline water (Fig. 8; Hoppmann et al., 2020, Scambos et al., 2017). The surrounding water is cooled and freshened and is then transported towards the surface. Under the large Filchner–Ronne and Ross ice shelves, the pressure relief can cause this water, called Ice Shelf Water (ISW), to be supercooled (Foldvik and Kvinge, 1974). The temperature of the supercooled ISW is typically below the in situ freezing point, which eventually causes the formation of ice platelets that accumulate under landfast ice attached to adjacent ice shelves (Fig. 8; Holland et al., 2007; Hoppmann et al., 2015, 2020).

In an attempt to elucidate the relationship between IPSO25 and platelet ice, we investigated our data with respect to the locations of observed platelet ice formation. While the maximum IPSO25 concentrations in front of the Filchner Ice Shelf could be directly related to the above-mentioned platelet ice formation in this area, the elevated IPSO25 concentrations north of the Larsen C Ice Shelf at the EAP could be linked to several processes. According to Langhorne et al. (2015), sea-ice cores retrieved from that area did not incorporate platelet ice. Hence, the high IPSO25 concentrations could be explained by either input from drift ice transported with the Weddell Gyre or by basal freeze-on. However, we note that our samples may reflect much longer time periods than the sea-ice samples investigated by Langhorne et al. (2015), and the lack of platelet ice in their investigated sea-ice cores does not rule out the former presence of platelet ice, which may be captured in our investigated sediment samples.

There are several previous studies on IPSO25 that reported a close connection of the proxy with proximal, coastal locations and polynyas in the seasonal ice zone (i.e. Collins et al., 2013; Smik et al., 2016). They do not, however, discuss the relation to adjacent ice shelves as possible “platelet ice factories”. We note that the core locations investigated by Smik et al. (2016) are in the vicinity of the Moscow University Ice Shelf, where Langhorne et al. (2015) did not observe platelet ice within sea-ice cores. Hoppmann et al. (2020), however, report a sea-ice core from that area, which incorporates platelet ice. The different observations by Langhorne et al. (2015) and Hoppmann et al. (2020) highlight the temporal variability in the occurrence of platelet ice in the cold-water regime around the East Antarctic margin.

For the observed IPSO25 minimum in the Amundsen Sea (AS, Fig. 3b), which we tentatively relate to the extended and thick sea-ice coverage, the absence of platelet ice there is an alternative explanation. The Amundsen/Bellingshausen seas and WAP shelves are classified as “warm shelves” (Thompson et al., 2018), where the upwelling of warm CDW (Schmidtko et al., 2014) hinders the formation of ISW, which makes the presence of platelet ice under recent conditions highly unlikely (Hoppmann et al., 2020). This is also supported by Langhorne et al. (2015), who stated that platelet ice formation is not observed in areas where basal ice-shelf melting is considerable, such as on the West Antarctic continental shelf in the eastern Pacific sector of the Southern Ocean (Thompson et al., 2018). Accordingly, if the formation and accumulation of platelet ice – up to a certain degree – indicates sub-ice-shelf melting on “cold shelves” (Hoppmann et al., 2015; Thompson et al., 2018), high IPSO25 concentrations found in marine sediments may serve as an indicator of past ISW formation and associated ice-shelf dynamics. This is, however, probably only true up to a certain threshold, where platelet ice formation decreases or is hampered due to warm oceanic conditions causing overly intense sub-ice-shelf melting (Langhorne et al., 2015).

When using IPSO25 as a sea-ice proxy in Antarctica, it is important to consider regional platelet ice formation processes too, as these may affect the IPSO25 budget. Determining thresholds associated with platelet ice formation is challenging. Therefore, further investigations, such as in situ measurements of IPSO25 concentrations in platelet ice or culture experiments in laboratories, are needed to better understand the connection between IPSO25 and platelet ice formation (and basal ice-shelf melting).