Introduction

Southern high-latitude sea surface temperature (SST) records from the Oligocene (33.9–23.0 Ma)1,2,3,4 show unexpectedly warm-temperate conditions, despite evidence for the coeval presence of a large Antarctic ice sheet5, which extended to the margins of the continent6,7,8. This apparent contradiction requires reconciliation9. A warm Oligocene Southern Ocean could be the result of generally high atmospheric pCO2 concentrations (300–700 parts per million (ppm); https://www.paleo-co2.org10), but higher pCO2 would also typically be associated with reduced ice volume11. Furthermore, since enhanced Antarctic ice volume should have cooled marginal seas12 the mystery of warm high-latitude SSTs and greater ice volume grows. One alternative hypothesis is that marine ice sheet terminations were restricted to the southernmost parts of the Antarctic margin, facilitated by a higher-than-modern Antarctic paleotopography13,14, while elsewhere the ice sheets were mostly terrestrial (e.g., refs. 15,16), limiting the Antarctic glacial cooling effect on proximal SSTs17. Or as a final hypothesis, critical Southern Ocean gateways (Tasmanian Gateway and Drake Passage) may have played a role in ocean heat redistribution18. Closed ocean gateways, could reduce the meridional temperature gradients by enhancing ocean poleward heat transport or by increasing local radiative heating through albedo feedbacks, but might also have enhanced the atmospheric moisture transport2,19,20, thereby potentially maintaining terrestrial ice sheets. Each of these factors (radiative forcing, ice sheet configuration and tectonic changes) would have had a unique spatial fingerprint of Southern Ocean SST changes relative to those at the Antarctic continental margin and unravelling the complexities is challenging. Atmospheric pCO2-associated radiative forcing would be expected to cause globally synchronous temperature trends on both long- and orbital-time scales, although with a degree of polar amplification. Ice sheet growth increase poleward heat transport21, but at the same time induces local cooling at the Antarctic margin12. The different response to Antarctic glaciation in different model experiments (Supplementary Table 1) can be attributed to subtle changes in paleogeography and model set-up22. In any case, ice volume change has the most effect close to the Antarctic continent, and is further evident in benthic foraminiferal δ18O23 and deep-sea cooling24. Finally, opening of gateways would result in profound cooling of Antarctic proximal waters25,26 while leaving the rest of the world’s sea surface temperatures largely unaffected12,19. Although the Antarctic proximal cooling could result in deep-sea cooling27, which would redistribute globally, the question remains how strong the resulting global cooling would be. Thus, as opposed to ice volume changes, gateway opening cause a stepwise, unidirectional change in temperature: changes in SSTs could then be stratigraphically linked to phases of gateway opening25,28. Currently, Oligocene SST records from the subtropics are lacking, which hinders establishing the latitudinal SST gradients needed to provide context for ice-proximal SST changes, and to evaluate the possible factors that drove the development of Oligocene Southern Ocean surface conditions.

Here, we present TEX86-based SST estimates from Late Eocene–Early Miocene ODP Site 1168 sediments, 339–765 meters below sea floor (mbsf), west of Tasmania (red dot in Fig. 1). To explore the development of the Oligocene SST gradient across the Tasmanian Gateway region, we compared our data with TEX86 based SSTs from east of Tasmania (ODP Site 1172)29, west of the Campbell Plateau (DSDP Site 277)4, north of the Ross Sea (DSDP Site 274)2, offshore Wilkes Land (IODP Site U13561 and DSDP Site 26930) and inorganic chemical weathering indices recording terrestrial temperature from the Cape Roberts Project (CRP) in the Ross Sea31 (Fig. 1). As most of these records cannot be resolved at orbital, glacial-interglacial variability, we predominately discuss long-term relative SST trends and the amplitude of variability. These SST records are compared to the temperature distribution in a coarse-resolution (3° horizontal), fully coupled general circulation model (GCM, following Kennedy‐Asser, et al.26 HadCM3L – Hadley Centre Coupled Model), which simulates equilibrium temperature response to CO2 forcing, and the role of ice volume and geographic boundary conditions of the Early- and Late Oligocene (33.9–28.1, 28.1–23 Ma). Details of ocean heat transport and consequences of local bathymetry are subsequently investigated by comparing SST results to high horizontal resolution (0.25°) ocean-only model simulations25. Our results show an increase in the SST gradient across the Southern Ocean starting at 26 Ma, when Antarctic-proximal SSTs cooled. This is in contrast to a synchronous decrease in global benthic foraminiferal δ18O indicating ice mass loss/deep sea warming. Considering the potential drivers of such cooling, we conclude that the Late Oligocene Antarctic-proximal SST cooling is not primarily driven by changes in pCO2 and ice sheet configuration, but by paleogeographic configurations.

Fig. 1: Paleogeographic map of the Oligocene locations of study sites and prevailing ocean currents.
figure 1

The approximate paleogeography at 27 Ma is reconstructed through G-plates (http://www.gplates.org), based on the global geodynamic rotation model from Müller, et al.70. Black represents the outline of modern coastlines. The grey outline corresponds to the modern 2000 m water depth contour. PLC Proto-Leeuwin Current, STF Subtropical front, TC Tasman Current, proto-ACC proto-Antarctic Circumpolar Current and ACountC=Antarctic counter current (after Houben, et al.3).

Results

Our SST record is based on 148 samples from ODP Site 1168 which were processed for TEX86 paleothermometry. Twenty-two showed potential for non-thermal overprints, thereby considered unreliable, and discarded from the dataset (see Supplementary Information). Results indicate Late Eocene–Early Miocene (35–20 Ma; red line Fig. 2b; Supplementary Data32) SSTs of 20–29 °C ( ± 4 °C standard calibration error). The amplitude of SST variability was high (5–7 °C) around 28 Ma and from 25 Ma onwards, and low (~3 °C) between 32–29 Ma and 27–25 Ma. Our record indicates 4 °C cooling (from 27 to 23 °C) across the Eocene–Oligocene transition (ca. 34 Ma) and then a return to high temperatures of ~29 °C at 33.2 Ma. Temperatures then gradually cooled until ~28 Ma. A transient warming of 6 °C occurred from 27.8–24.3 Ma, followed by a gradual cooling from 24.3–22.2 Ma. The TEX86-based SSTs are generally warm and in line with Uk'37-based SSTs of 19 °C to 29 °C derived from the same records for the 29.8–16.7 Ma interval33 (Supplementary Fig. 7). The Uk'37-based SST record shows a more prominent Late Oligocene warming, although Uk'37-based SSTs remains within the variability of the TEX86-based SST record (stippled line Fig. 2b). The discrepancy between TEX86 and UK’37 could be explained by difference in the season or in the water depth where the proxies are produced (e.g. Herbert et al.,34). Yet, a prominent input of deep-dwelling archaeal GDGTs at this site is unlikely, as the GDGT2/3 ratios are low35,36 (<5; Supplementary Fig. 3f). Moreover, SST derived by TEX86 proxy are warmer than those from UK37. With a substantial contribution from deep-dwelling archaea, a cold-bias in TEX86 would be expected because the photic zone requirement of alkenone producers makes UK'37 a proxy for SST, while GDGT producers occur also below the photic zone. This lends confidence that the TEX86 proxy reflects SST at this site. Further support for the warm-temperate SSTs comes from dinoflagellate cyst (dinocyst) assemblages analyzed on the same samples37.

Fig. 2: Late Eocene–Early Miocene Australian-Antarctic Gulf temperature records, with their paleolatitude, compared with global benthic δ18O and pCO2 records and GCM models.
figure 2

a Paleolatitude evolution of sites65. Colors refer to sites in Fig. 1. b TEX86- SST reconstructions. Thick lines represent smoothed long-term trends, with a local weighted polynomial regression (LOESS; span of 0.35). For the age model of Site 1168 see Supplementary Data32. c Mean annual air temperature (MAT) from Site CRP-331. All ages in a–c are converted to GTS201251. d Benthic foraminiferal δ18O splice, smoothed by a locally weighted function over 20 kyr (thin black curve; CENOGRID23). Thick black curve is the LOESS smoothed (span = 0.1). e paleo-CO2 compilation from https://www.paleo-co2.org10. The vertical error bars indicate the 95% confidence interval for each pCO2 estimate. Blue curve is the LOESS smoothed (span = 0.1). f Shows the GCM model temperature gradient (corresponding to the scale in b) between Site 1168 and Site U1356 in the Early- and Late Oligocene.

Discussion

Temperature gradient across the Australian-Antarctic Gulf

We focus the discussion on the SST gradient across the Australian-Antarctic Gulf (AAG), between Sites 1168 and U1356 in the proxy data compilation (arrows, Fig. 2), due to their high temporal resolution, while Sites 1172, 277, 274 and 269 offer a broader regional context. We note a persistent SST gradient (5–10 °C) between the Antarctic-proximal (Site U1356) and the subtropical (Site 1168) sites, albeit smaller than at present-day (14 °C)38. Still, this implies that (polar) frontal systems separated water masses latitudinally in the Oligocene AAG. The Early Oligocene latitudinal separation of water masses is further corroborated by the strong latitudinal separation in dinocyst assemblages from ~30 Ma onwards between the Australian (Site 11723 and 116837, relatively oligotrophic) and Antarctic (Site U135615, 26930 and 2742, eutrophic, upwelling) margins of the AAG.

The reconstructed SST gradient is in line with the output of two Oligocene GCM simulations (Fig. 3b), albeit absolute SSTs are in general lower throughout the region in the model simulations (10–20 °C; Fig. 2, Fig. 3a,b) than in the TEX86-based SST records (15–29 °C). This could be the result of a warm bias in the TEX86 proxy (e.g., Hartman et al.1), as suggested by the slightly cooler Uk'37-based SSTs33 (stippled line Fig. 2b) and/or to too low climate sensitivity in the GCMs. We acknowledge that a cold bias is known to be a persistent issue with the Had3CML GCM as with nearly all GCMs for past warm climate intervals, including in the Oligocene (9,39). The two different ice sheet sizes in the Early- and Late Oligocene simulations can be used to evaluate the effects of variability in ice sheet size as well as the effect of long-term changes in geographic boundary conditions on the simulated latitudinal SST gradient. We do acknowledge that this interpretation derives from a low-resolution model, which poorly represents eddy flow, as no high-resolution ocean models that simulate the oceanographic effect of Antarctic ice sheet size are available. Interestingly, while SST difference derived from sediments deposited during glacial or interglacial intervals from Site U1356 is high1,15,40 (indicated by separate symbols in Fig. 2b), two different ice sheet sizes in the Early- and Late Oligocene GCM simulations, which also includes a widening of the AAG, are of little (~1 °C) impact to the simulated latitudinal SST gradient (Fig. 3b). We also note that the difference in amplitude of SST change between offshore Australia and Antarctica is too large to be caused only by greenhouse gas induced radiative forcing with a factor of polar amplification. Therefore, we ascribe the high amplitude temperature signal to shifts in ocean frontal systems. The small effect of the northward tectonic drift of Site 1168 on regional SSTs, indicates that the subtropical front (STF) likely migrated northward along with the Australian landmass, as has been suggested from microfossil data37. This effect is further muted at Site 1168, because Australia hinders northward migration of the STF, which explains the smaller temperature swings offshore Australia.

Fig. 3: Data-model experiment comparisons.
figure 3

a Paleogeographic map with SST data shown in colored dots from the respective drill sites. b Modeled SSTs from fully coupled HadCM3L simulations of the Early-(33.9–28.4 Ma) and Late Oligocene (28.4–23.0 Ma)26. The modeled simulations are TDLUQ and TDLUP respectively (see Supplementary Table 2). TDLUP SSTs are differenced from TDLUQ SSTs in the third panel. c High-resolution ocean model25. The Southern Ocean paleogeography is reconstructed for Late Eocene (38 Ma) for all three panels, red circles indicate the studied sites.

Late Oligocene paleogeographic, ocean temperature, atmospheric pCO2 and ice volume changes

In the Late Oligocene we note the substantially increasing temperature gradient across the AAG latitudinal transect from 6 °C prior to 26 Ma to >9 °C by 23.5 Ma (Fig. 2). This is mostly due to unidirectional progressive cooling of the Antarctic-proximal SST record at Site U1356 and stepwise air temperature cooling at CRP31 starting at around 26 Ma, opposite to slightly increasing SSTs (Uk'37-based SST, increasing the most33) at Site 1168 and the benthic foraminiferal δ18O splice (CENOGRID23, Fig. 2d) showing Late Oligocene warming and ice mass loss. We note that the CENOGRID benthic foraminiferal δ18O records are derived from lower latitudes (<30°) in the Atlantic and Pacific, and might therefore not fully be representative for “global” deep-sea temperatures/ice mass. The Antarctic-proximal cooling continues into the Miocene, where the δ18O record also show deep ocean cooling. Meanwhile the subtropical SSTs at Site 1168 show a Late Oligocene warming coincident with trends in δ18O record (Fig. 2d). The slightly cooler Late Oligocene SSTs at Site 269, northeast of Site U1356, have been attributed to its proximal location to upwelling30, while the cooler SSTs at Site 274 is attributed to its higher latitude and proximity to the colder Ross Sea2, as also inferred from the air temperature record31 (Fig. 2c) and fossil record41 at CRP.

We break down the complex interplay of forcings and feedbacks that kept the Early Oligocene Southern Ocean warm and caused cooling of the Wilkes Land Antarctic Margin at 26 Ma – changes in atmospheric pCO2 levels, ice volume or paleogeography (Table 1). Late Oligocene atmospheric pCO2 gradually decline (700–300 ppm10; Fig. 2e), but, the latitudinally contrasting paleotemperature trends, with Antarctic-proximal SSTs cooling, subtropics remaining warm and equatorial areas warming9, are difficult to attribute primarily to pCO2 as shown in the climate model results (Supplementary Table 2, Supplementary Fig. 8c and f, Supplementary Fig. 9c, f and i). In Community Earth System Model simulations by Goldner et al.12, the expansion of Antarctic ice sheets generates cooling of 6 °C at the Antarctic margin, while in atmosphere - ocean GCM model simulations by Knorr and Lohmann21 an expanded ice sheet would cause regional warming at the Antarctic margin (Supplementary Table 1). Nonetheless, an expanding Late Oligocene ice volume is unlikely given the decreasing benthic δ18O indicating loss of Antarctic ice volume with deep-sea warming23,24, which has been ascribed to local tectonism on Antarctica13,17. Duncan et al.42 recorded the same enigmatic Late Oligocene (25–24 Ma) trend of cooling SSTs at the Ross Sea and Wilkes Land margins and ice volume loss, which they attributed to the tectonic subsidence of West Antarctica which favored the influx of the still relatively warm ocean waters in the Ross Sea region precluding the advance of a marine-based ice sheet. Although this provides a plausible explanation for the SST and ice volume trends in the Ross Sea, it doesn’t explain why Antarctic proximal SSTs cool around the east Antarctic Margin.

Table 1 Scenarios to explain Late Oligocene trends in SST and deep sea δ18O.

Moreover, the Early- and Late Oligocene GCM model results show little effect of AAG widening and ice volume changes on the Southern Ocean SST gradient (Fig. 3b). The Late Oligocene breakdown of the relationship between SST, deep ocean temperature, atmospheric pCO2 and ice volume9 suggests that Antarctic proximal SST cooling was not limited to pCO2 changes or a glaciation-induced negative feedback12, but probably also affected by paleogeographic configurations.

Tectonic deepening of Drake Passage caused cooling along Antarctic Margin

Low-resolution coupled climate models have suggested that opening of Southern Ocean gateways had little effect on poleward ocean heat transport and polar climate12,43. However, recently, the importance of the high horizontal resolution of ocean model simulations in such experiments has been underlined19,26. Eddy-permitting model simulations25 show that deepening of the second of two Southern Ocean gateways (Drake Passage and Tasmanian Gateway) below 300 m drives surface water cooling at the Antarctic margin (up to 5 °C), while leaving the rest of the Southern Ocean with little relative temperature changes (Fig. 3c). These models are uncoupled so it is currently impossible to know if the same result would obtain if the atmosphere and sea ice could respond and feedback on to this cooling. Although it would be desirable to utilize fully coupled high-resolution models, it is too computationally expensive to run them long enough to verify the deep-ocean response. Despite this modeling limitation, it is compelling that this is exactly what can be seen in the SST compilation at 26 Ma: The SST at the STF remains relatively stable despite northward migration and the Antarctic proximal Site U1356 cools (Fig. 2), while the benthic δ18O records shows apparent global warming/ice loss. The gradual northward migration of Australia could have progressively invited a larger volume of east-flowing STF water to follow the southward route around Australia18, without changing the absolute temperature in the STF region. This southward route would also progressively line up better with the westerly wind belt, strengthening the proto-Antarctic Circumpolar Current44,45. A strengthened proto-Antarctic Circumpolar Current would deflect the warm poleward extension of the subpolar gyres, including the proto-Leeuwin Current away from Antarctica, and reduce heat transport towards Wilkes Land Antarctic Margin, increasing polar isolation25. Crucially, the timing of this observed gradient increase coincides with evidence from kinematic reconstructions of Drake Passage28 showing a first deep ocean connection around 26 Ma. Also, sediments from the South Pacific indicate the formation of the proto-Antarctic Circumpolar Current in the Late Oligocene (ca. 25–23 Ma)46. Thus, we deduce that despite proximity to the Tasmanian Gateway, the deep opening of Drake Passage in the Late Oligocene induced strong increase in the Southern Ocean SST gradient and cooling of Antarctic surface waters, also in the Tasmanian Gateway area. We further conclude that despite the cooling induced by widening Southern Ocean gateways, global ice volume apparently declined in this time interval, likely due to regional Antarctic tectonic changes affecting Antarctic topography and ice mass balance42, unless there is a low latitude sampling bias in the global benthic foraminiferal δ18O splice. Although radiative forcing (CO2, orbital variations) is (likely) the primary driver of the Cenozoic climatic evolution (e.g.43,47,48), we here support an important role for paleogeography on Southern Ocean and Antarctic climate.

Conclusion

Our TEX86-based SST record from the Oligocene Tasmanian Margin (ODP Site 1168), representing the development of SSTs around the STF, in comparison with the benthic foraminiferal δ18O splice, pCO2 estimates, regional SST records, and model simulations, show the following:

  • The latitudinal SST gradient across the widening AAG was ~6–8 °C in the Late Eocene–Early Oligocene, and increased from 26 Ma, when Antarctic proximal cooling started.

  • SST data derived from sediments deposited during glacial or interglacial intervals from the Antarctic proximal site show that the latitudinal SST gradient is larger during glacials. This can be ascribed to the latitudinal migrations of ocean frontal systems, which are limited at the northern boundary of the Southern Ocean by the position of Australia.

  • Long term trends in Antarctic ice volume and polar amplification due to decreasing pCO2 cannot alone explain the Antarctic proximal cooling starting at 26 Ma. We correlate this cooling to the first deep opening of Drake Passage which decreased the strength of subpolar ocean gyres and southward heat transport, enhancing Antarctic thermal isolation and circumpolar flow.

Methods

Site description, depositional setting and age model

We reconstruct the changes in sea surface temperature from the subtropical front region by studying marine sediments (766–339 mbsf) from Ocean Drilling Program (ODP) Site 1168 (42°38′40′′ S, 144°25′30′′ E, present water depth: 2463 m). The site is situated 70 km off the west Tasmanian coast, north of the oceanographic subtropical front, where relatively carbon-rich siliciclastic sediments have filled the graben basin between two basement highs (25 km length) since the Late Eocene until Early Oligocene and continental slope sedimentation of calcium carbonate-rich sediments thereafter37,49. A detailed description of the site location, depositional setting and oceanographic setting has been given in Hoem, et al.37. For the age model, we updated ages of tie points interpolated (cf. Stickley, et al.50; updated to GTS2012 ages51 in Hoem, et al.37) with the exemption of one last occurrence datum of foraminifera Subbotina angiporoides. We interpolated a loess smooth (span of 0.1) through age tie points to obtain the ages of our samples (Supplementary Fig. 1).

TEX86 paleothermometry

In order to reconstruct sea surface temperature (SST), we applied the TEX86 (TetraEther indeX of 86 carbon atoms) proxy52, which is based on the temperature-dependent cyclisation of isoprenoidal glycerol dialkyl glycerol tetraethers (GDGTs) produced by thaumarchaeotal membrane lipids. A total of 148 samples spanning the period between 35 and 20 Ma (766–339 mbsf) were processed for analysis of GDGTs (Supplementary Methods; Supplementary Data32). GDGTs were extracted from powdered and freeze-dried sediments using a Milestone Ethos X microwave or accelerated solvent extractor system. Lipid extracts were separated into an apolar, ketone and polar fraction by Silica gel column chromatography. GDGT standard was added to the polar fraction and filtered over a 0.45 μm polytetrafluoroethylene filter. The dissolved polar fractions were injected and analysed by high-performance liquid chromatography–mass spectrometry (HPLC–MS) at Utrecht University, using double-column separation53. GDGT peaks in the HPLC chromatograms were integrated using ChemStation software. A more detailed description can be found in the Supplementary Methods.

TEX86 was calculated as defined by Schouten, et al.52:

$${{\,}^{\prime} {{{{{\rm{TEX}}}}}}}_{86}{\,}^{\prime} =\frac{{GDGT}-2+{GDGT}-3+{{Cren}}^{{\prime} }}{{GDGT}-1+{GDGT}-2+{GDGT}-3+{{Cren}}^{{\prime} }}$$

TEX86 results were examined for non-thermal overprints (described in detail in Supplementary Methods) to verify the reliability of their SST signal (Supplementary Data32; Supplementary Fig. 26) and compared to an alkenone-derived (Uk’37) SST record from the same site33 (Supplementary Fig. 7). For systematic calculation of GDGT ratios, data analysis, visualization, and evaluation of isoGDGT, brGDGT and/or brGMGT data, we utilized the R script of Bijl, et al.29: https://github.com/bijlpeter83/RGDGT.git. We uploaded the measured peak areas (Supplementary Data32) of GDGTs in the R script and calculated and plotted fractional abundances, overprinting indices and paleotemperature time series (Supplementary Figs. 24).

In order to translate the TEX86 values into SSTs, we used the regionally varying, Bayesian calibration; BAYSPAR SST (prior mean of 30 °C, prior standard deviation of 20) of Tierney and Tingley54 (Supplementary Fig. 5). The BAYSPAR method compares measured TEX86 values with similar values in the modern SST observations, obtained from surface sediment samples, to derive linear regression parameters: BAYSPAR propagates uncertainties in the surface sediment data into resulting temperature predictions54. SST estimates obtained with the exponential transfer function BAYSPAR are very similar to the TEX86-based SSTs produced by the exponential function from Kim, et al.55 and linear function by O’Brien, et al.56 (Supplementary Fig. 6), varying, at most, by 2 °C.

Temperature data compilation

We compiled existing Oligocene TEX86 data and applied the BAYSPAR SST calibration54 for consistency, from east of Tasmania: ODP Site 117229, north of the Ross Sea: DSDP Site 2742, offshore Wilkes Land: IODP Site U13561 and DSDP Site 26930. The age model for the terrestrial temperature records from the Ross Sea: Cape Roberts Project (CRP)31 based on Lavelle57, McIntosh58, and Florindo, et al.59, was converted to GTS2012 ages51 for the purpose of this paper.

Fully coupled climate model

For model-data intercomparisons, we utilized a suite of general circulation model (GCM) HadCM3BL‐M2.1aE model experiments, with full atmospheric coupling, carried out and described by Kennedy-Asser, et al.26 and compiled by the Bridge Consortium of the University of Bristol (found at http://www.bridge.bris.ac.uk/resources/simulations). Simulations that were selected for comparison (Fig. 3b) were built using boundary conditions, developed by Getech Group plc, appropriate for Rupelian and Chattian-age paleogeographies with either a closed or open Drake Passage, narrow or wide Tasmanian Gateway, using varying ice sheet constructions (either no ice sheet, East Antarctic Ice Sheet (EAIS) only, or full Antarctic Ice Sheet (AIS)), and varying pCO2 concentrations (either 560 or 1120 ppm) (Supplementary Table 2, Supplementary Figs. 8 and 9). These simulations were used in this study because the model has been extensively applied in the Eocene and Oligocene and extensive sensitivity tests to length of integration, paleogeographic boundary conditions, and pCO2 have been conducted. Although the HadCM3L ocean resolution is low, the influence of an Antarctic ice sheet does not necessitate a high-resolution ocean to a great degree. Furthermore, the lengthy spin-up time of the Kennedy-Asser, et al.26 experiments help minimize any uncertainty in their results and makes them highly robust for this comparison. As with nearly all coupled climate models, this model tends to require higher greenhouse gas concentrations than is typical of reconstructions to achieve sufficiently warm high latitude temperatures. This is likely a byproduct of a too low value of global equilibrium climate sensitivity as has been widely described previously60,61,62. We are here comparing relative changes, so this systematic bias is not likely to importantly impact the interpretation of our results.

High-resolution eddy-resolving ocean model

The presented high-resolution ocean simulations (Fig. 3c) are taken from Sauermilch, et al.25. It uses the Massachusetts Institute of Technology ocean general circulation model (MITgcm)63 with a circum-Antarctic model domain (from 84°S to 25°S). The spatial resolution is 0.25° (3–25 km resolution) and vertical resolution contains 50 layers (ranging from 10 m at the sea surface to 368 m at the sea floor). Southern Ocean paleogeography is reconstructed to the Late Eocene (38 Ma) position using the plate tectonic model of Matthews, et al.64) in a paleomagnetic reference frame65.

The model is ocean-only and atmospheric forcing are taken from a coupled atmosphere-ocean simulation (GFDL CM2.1) run with atmospheric pCO2 concentration of 800 ppm66. A restoring time scale of 10 days is applied. Although not directly coupled to the atmosphere or ice sheet, the resolution of the ocean model is higher than most previous paleo-simulations and permits the formation of ocean eddies which are responsible for the majority of the ocean heat transport67. The high-resolution model is less diffusive, allowing accurate simulation of subsurface velocities and current structure. In addition, detailed paleobathymetry features such as the seafloor roughness, but also small depth changes in the critical gateway regions, Tasman Gateway and Drake Passage, can be accurately resolved. To accommodate this advantage, new high-resolution paleobathymetry grids are used68 which are able to reconstruct detailed seafloor roughness features, such as seamounts and fracture zones, that have a substantial impact on the large-scale ocean circulation (e.g. LaCesca, et al.69). See Sauermilch, et al.25 for further details about the methodology of the high-resolution ocean model and paleobathymetry reconstruction.