Introduction

Temperature contrasts between the equator and high latitudes are mitigated through poleward atmospheric and ocean heat transport1,2. Variability in the latitudinal sea surface temperature (SST) gradient is mostly a function of polar temperatures, which are much more variable than those at low latitudes because of polar amplification3. In turn, polar SSTs, especially offshore Antarctica, vary with prevailing cryosphere conditions, including sea ice extent4,5. The steepest part of the latitudinal SST gradient is at mid-latitudes, at the boundary between subtropical gyres and subpolar waters. In the Southern Hemisphere, this is the subtropical front (STF): the northern limit of the Southern Ocean and the Antarctic Circumpolar Current (ACC), and the centre of ocean carbon uptake6 (Fig. 1). The ACC and associated oceanographic fronts, driven by westerlies and steered by bathymetry7, regulate deep ocean ventilation8,9,10 and heat exchange between low and high latitudes11,12. In turn, the latitudinal position of westerlies is influenced by the extent of sea ice around Antarctica13,14. Oceanographic conditions around the ocean fronts thus play a central role in the latitudinal distribution of heat in the Southern Hemisphere, including the heat source that causes basal melt and instability of marine-terminating Antarctic ice sheets6. Future projections of polar climate change, and the consequences for cryosphere melt and sea level are highly uncertain15, because changes in and interactions between Antarctic ice sheets, sea ice and oceanography bear numerous poorly constrained, non-linear feedbacks6,16. Important constraints on the functioning of this system in a warming world might come from reconstructions of geologic episodes during which the partial pressure of atmospheric CO2 (pCO2) was as high as projected for the future.

Fig. 1: Study area and sites locations.
figure 1

a Global ocean bottom (>2500 m) water temperature, in blue diamonds the sites from which Miocene clumped isotope data has been generated36,37,38, based on refs. 116,117. Grey line indicates the area of Fig. 1b. b Map of the Southern Ocean Sites with modern sea surface temperature117 and frontal systems positions118. STF subtropical front; SAF subantarctic front; PF polar front. Attribution: https://cp.copernicus.org/articles/19/787/2023/ under https://creativecommons.org/licenses/by/4.0/; Illustrations of SAF and PF are added.

Full size image

Throughout the Neogene (23–2.58 Ma), pCO2 declined from 800 to 300 parts per million (ppm)17, global temperatures dropped18,19, latitudinal SST gradients increased20 and global ice volume19,21,22,23 and sea ice expanded24. The current paradigm assigns pCO2 decline as the primary driver, which, through polar amplification of cooling, stimulates ice growth and cooling in the regions of deep-water formation18,25,26. Yet, recent data have challenged this view. A recent study found that Neogene SST gradients increased in the subtropical gyre but decreased from the subtropical front to polar waters27. With relatively stable equatorial28,29 and polar SSTs30, this indicates that the mid-latitudes, rather than the high-latitudes27, cooled most profoundly in the Neogene. Antarctic-proximal records suggest a retreated Antarctic ice sheet and warm Antarctic-proximal conditions24,30,31 during the mid-Miocene Climatic Optimum (MCO) and profound seaward ice sheet advance during subsequent cooling termed the Middle Miocene Climatic Transition (MMCT)21,23,32,33,34, in line with pCO2 estimates35. Along with a rise in deep ocean benthic foraminifera oxygen isotope ratios (δ18Obf), this suggests a strong increase in global ice volume. Yet, the existing series of clumped isotope measurements (Δ47, which deconvolves temperature and ice volume components in δ18Obf records) on Miocene benthic foraminifera36,37,38 suggest higher-than-previously-estimated Bottom Water Temperatures (BWTs) during the MCO, and as a result, large global ice volume. These records also indicate strong BWT cooling during the MMCT, explaining most if not all of the δ18Obf rise, and therefore little to no ice volume buildup. How this generally connects to far-field sea level changes is still poorly reconciled. However, the uncertainties in clumped isotope data and the limited resolution and temporal range of the records leave ambiguity on the true amount of BWT drop and ice volume buildup during the mid-Miocene.

Like the modern, changes in the Southern Ocean, notably regarding fronts and currents, were likely vital for heat transport towards the ice sheet in the Neogene. A relatively weak ACC, initiated during the Eocene39, intensified in the late Oligocene ~26 Ma40 but modern-like strengths only developed in the late Neogene41. The development and latitudinal position of the fronts associated to the ACC are, however, still poorly constrained. Meanwhile, a long-term trend of BWT change and how the oceanic processes are coupled to Antarctic ice dynamic is still unclear. To shed light on the links between (Antarctic) ice volume and dynamics, Southern Ocean oceanography and latitudinal SST gradients, we present a detailed reconstruction of Neogene STF migration history and surface and bottom water temperature offshore Tasmania, and pair these with estimates of Antarctic ice volume change from the MCO across the MMCT. We use dinoflagellate cyst (dinocyst) biogeography42 to reconstruct the position of Southern Ocean currents and fronts and combine these with published SST reconstructions27. Finally, from benthic foraminiferal Δ47, we assess deep-water temperature changes at Ocean Drilling Program (ODP) Site 1168, as well as sea water δ18O (δ18Osw) as a proxy for Antarctic ice change.

In this study, we demonstrate that there is a strengthening and equatorward migration of the STF from ~53° to ~42° between ~14 Ma and 7 Ma, concurrent with progressive sea surface and bottom water cooling. The deep ocean cooling can completely explain benthic foraminifer δ18O evolution, implying stable global ice volume. After 7 Ma, the northward shift of the STF is limited by the Australian continent, even though the SSTs continue to decrease. To reconcile expansion of subpolar ocean conditions and progressive Neogene Southern Ocean cooling with stable ice volume and compelling evidence of ice advance, we argue that the Miocene Antarctic ice sheet progressively lowered in height while expanding seawards during the mid-Miocene. We present idealised ice sheet model simulations that physically constrain this hypothesis. This changed geometry induced strong regional oceanographic responses with expansion of sea ice, cooling of the region of bottom-water formation and northwards migration of ocean fronts.

Results

Dinoflagellate-based surface oceanographic reconstruction of the subtropical front

The vast majority of the dinocysts encountered in the Neogene sediments from ODP Site 1168 are extant species of the modern Southern Ocean. The use of inferences from modern biogeographic distributions and affinities of dinocyst assemblage clusters42 (see “Methods” section) hence allows reliable reconstructions of paleoceanographic conditions.

In early Miocene sediments at Site 1168, dinocyst assemblages are dominated by warm/temperate Spiniferites spp. (Fig. 2c). This assemblage resembles the Spin-cluster of ref. 42 (Fig. 2b), which now mainly thrives along the northwest coast of Australia and in low latitudes in the eastern Indian Ocean42. This cluster is associated with a modern SST of ~29 ± 0.5 °C, a temperature in line with that derived from biomarkers (Fig. 2a). Early Miocene SSTs at Site 1168 were ~13 ± 6 °C (calibration error) warmer than today based on biomarkers, in spite of a ~ 10 ° more poleward position of the site43. Given these SSTs and dinocyst assemblages, we infer a strong influence of the (proto-) Leeuwin Current, delivering heat and sustaining low-latitude dinoflagellate assemblages from western Australia towards the site. It implies that the STF was located to the south of the site. Gradual increases of Operculodinium spp. in this interval suggests gradually cooler-water influence, with an approaching STF from the south. We find occasional northward migrations of the STF (e.g., at ~22 Ma) in sporadic abundance of N. labyrinthus, concomitant to SST cooling (Fig. 2c).

Fig. 2: Dinoflagellate cyst assemblage results with published sea surface temperature (SST) records from Ocean Drilling Program Site 1168.
figure 2

a SST of Site 1168 based on TetraEther indeX of 86 carbons (TEX86), alkenone unsaturation ratio (Uk’37)27, and dinocyst assemblages (this study). TEX86, Uk’37 use bayspar119,120 and bayspline121 calibrations, respectively. 95% confidence interval is indicated in the panel. Dinocyst-based median SST estimates are based on their environmental affinities42 in 4 time intervals (see “Methods” section) with 25–75% quantiles. b Dinocyst clusters based on ref. 42. c Grouped dinocyst assemblages by their ecological affinities based on ref. 42 (see Table S1). Dinocysts are ordered from their known occurrence in latitude from north (top) to south, with uncertain groups and heterotrophic species to the bottom. MCO Mid-Miocene Climatic Optimum, MMCT mid-Miocene Climatic Transition.

Full size image

Dinocysts are poorly preserved in MCO sediments (Fig. 2c), and Glycerol Dialkyl Glycerol Tetraethers concentrations are low27, pointing to enhanced sediment oxidation44. The available palynological data for the MCO shows that the Spin cluster was replaced by Impagidinium paradoxum and I. patulum, which in the modern are restricted to temperate to equatorial open ocean regions between sub-tropical and subpolar systems45. Although it is unclear how this dinocyst assemblage differs from the Spin cluster in terms of ocean temperature, the biomarker-based SSTs indicate continued warmth during the MCO at the site (Fig. 2). In any case, Site 1168 remained north of the STF.

The mid Miocene Climatic Transition (MMCT, ~14.5 to 12 Ma) marks the first interval of prevailing N. labyrinthus (Fig. 2b, c). This species (Nlab cluster in ref. 42) is found most abundant in sediments south of the STF, in the modern subantarctic zone. We interpret the proliferation of Nlab and a progressive cooling towards subantarctic zone-like conditions (Fig. 2) as a northward migration of the STF. At MMCT, the STF reached a similar position relative to that of Australia as during the last glacial maximum46,47,48 (Fig. 3). Subsequent high-amplitude, short-term fluctuations of dinocyst assemblages between the Nlab- and Spin-cluster, and, albeit less pronounced, biomarker-based SSTs (Fig. 2a), indicate strong (SSTs between 29 °C and 11 °C; Fig. 2c) variability of the latitudinal position of the STF until 7 Ma (Fig. 2).

Fig. 3: Subtropical Front (STF) migration history in the southeast Indian and southwest Pacific Ocean in 5 time slices from the start of the Neogene.
figure 3

a early Miocene 20–17 Ma, (b) Miocene Climatic Optimum, 17–14.5 Ma (c) 9–7 Ma (d) 5–2 Ma (e) modern, 0 Ma. Average dinocyst assemblages for these time slices at Site 1168 are presented along with those from Site U135624, Site 269A122, Site 27440 and ANDRILL-2A78. The present-day dinocyst distribution is based on Thöle et al.42. Red arrow indicates the (proto-) Leeuwin Current. Average sea surface temperature for time slices at Site 1168 are presented along with those from U135624, Site 1172123, Site 112520, Site 59420 and Ross Sea sites30. Solid black line indicates the STF, whereby the thickness of the line denotes the relative strength of the STF. Positions of the STF are hypothetically drawn based on this study and refs. 81,118. Paleogeographic position of the continents and sites are generated with the software GPlates94,124. Dark brown areas indicate present-day landmass, dark grey indicates continental crust.

Full size image

From ~7 Ma, N. labyrinthus started to decrease in abundance. We interpret a southward migration of the frontal systems relative to Australia from the decline in Nlab and the return of north-of-STF clusters. Apparently, this continued cooling is not directly related to continued shifts of ocean frontal systems, but a cooling of the STF itself and tectonic drift. The Nlab-Cluster and the slightly warmer high-Ocen-Cluster alternate in the Pliocene (4.5–2.5 Ma) on orbital timescales (Fig. 2b, c), which is close to the modern assemblage, and bracket the modern STF42 (SSTs between 25 °C and 10 °C; Fig. 2a). The Pliocene stands out as a warmer interval than the end of the Miocene, from both the biomarker-based SSTs (~15 °C)27 and the dinocysts assemblages. In addition, the Pliocene yields much more abundant O.centrocarpum and I. aculeatum than the rest of the record, which means that the STF was much closer to ODP Site 1168 in the Pliocene than in the early Miocene, when Spiniferites dominated.

Overall, we deduce long-term cooling from the dinocyst assemblages, despite the ~8° northward tectonic movement of the site during the Neogene. There was strong variability over glacial-interglacial climate fluctuations. The STF moved gradually northwards during 22–7 Ma from ~53°S to 42°S (Fig. 3a–c). We infer a concomitant strengthening of the STF from steepened latitudinal SST gradient among mid latitudes27, and from the fact that the STF was progressively pushed towards the southern margin of the Australian continent. From 7 to 2.5 Ma, the STF moved south from the site again, likely because of Australia’s continued northward drift. This allowed for the return of influence of the warm (proto-) Leeuwin Current at the Site (Fig. 3c, d).

Benthic foraminiferal stable isotope ratios, Δ47 and sea water δ18O

The δ18Obf and δ13Cbf records generally follow trends recorded at other Southern Ocean sites36,37 (Fig. 4), including a 1‰ negative offset in δ18O compared to the CENOGRID compilation18. At ~10 Ma, δ18Obf gradually increases from 1.5‰ (MCO) to 2.5‰, followed by a further rise to ~2.8‰ at the end of the Miocene (~5.3 Ma). Remarkably, we do not record pronounced steps across the MMCT as seen in other records49. The pronounced δ13Cbf maxima (from 17 Ma) likely reflects the Monterey carbon isotope excursion50,51,52 and values are in line with those in other records.

Fig. 4: Benthic foraminiferal oxygen and carbon stable isotopes of Site 1168.
figure 4

a oxygen isotopes and (b) carbon isotopes of Site 1168 (black dots) together with data from Site 117279 (blue dots), Site 74737 (green triangles), Site 76136 (orange diamonds) and the CENOGRID stack18 (grey dots). MCO Mid-Miocene Climatic Optimum (pink shadow), MMCT mid-Miocene Climatic Transition (blue shadow).

Full size image

The benthic foraminiferal clumped isotope data from Site 1168 fill critical mid- and late-Miocene gaps in existing BWT compilations38 and thus Antarctic ice dynamics (Fig. 5a). BWT, based on Δ47 data in ~1 Myr bins, at Site 1168 decreased gradually from 9.9 ± 4.0 °C (95% confidence interval) in the MCO (17–14.5 Ma) to 5.0 ± 2.5 °C around 10–9 Ma (Supplementary Data 1 and Fig. 5a). While the decreasing trend in mid-Miocene BWT is evident, the confidence intervals on the individual data points leave ambiguity on the significance of the point-to-point cooling. A Student’s t-test on the bins, however, proves a significant difference in Δ47 between the MCO (17–14.5 Ma) and late Miocene (10–9 Ma; p = 0.02; Table S1). Hence, the BWT cooling from the MCO to 9 Ma is significant. The ~8 °C data point at ~8 Ma has only 23 replicates and the longest binned time interval, and because of the resulting high uncertainty we leave this data point out of our interpretations (Fig. S1). By the end of the Miocene (5 Ma), BWTs were slightly elevated (5–6 ± 3 °C) compared to the mid-late Miocene.

Fig. 5: Compilation of records for the Neogene.
figure 5

a Clumped isotope-based bottom water temperature (BWT) and (b) bottom water δ18O (δ18Osw) of Site 1168 (supplementary data 1) along with data from Site 747 (cyan triangles)37, Site 761 (orange diamonds)36 and CENOTREND (grey square)38. Horizontal error bars indicate the time interval of each bin. Vertical error bars indicate 95% confidence interval. Violet lines indicate the BWT and δ18Osw based on Rohling et al.19. c, d Ice-rafted debris record from Site 116587 and U135624, units are weight percentage and counts respectively. e Qualitative geological record of Antarctic land- (light blue), marine- (blue) and sea ice extent (ink)21,24,33,34,125,126. f pCO2 reconstructions based on boron isotopes (green triangles) and alkenones δ13C (green dots)35,82. Vertical error bars indicate 95% confidence interval. Solid lines indicate the pCO2 thresholds of glaciation based on DeConto et al.127.

Full size image

Previous studies have pointed out the unexpected warmth of mid-Miocene BWTs in their reconstructions and discussed potential but undiscernible biases on Δ47-based BWT from recrystallisation and pH36,37,38,53. Since benthic foraminifera at Site 1168 are well preserved (Figs. S2 and S3), and seawater chemistry, dissolution and recrystallisation have very limited influence on benthic foraminifera Δ47 composition54,55, we consider our BWT reconstructions reliable and confirm from Site 1168 the previous inferences of much warmer BWTs in the Miocene than present. By applying a new calibration56, the BWT shifts by around ~2.4 °C towards lower values (Fig. S4) and results in a smaller global ice volume, but the amplitude of changes remains the same for both BWT and δ18Osw (see “Methods” section).

Discussion

The calculated δ18Osw values from Site 1168 BWTs are 0.3 ± 0.5‰ throughout the MCO and MMCT until 9 Ma (Fig. 5b). The increase in δ18Obf (~1‰) from 16 to 9 Ma could in principle be all reconciled with the ~5 °C BWT drop we infer from the clumped isotope data (Fig. 5a, b). Thus, our Δ47-based record is different from previous far-field sea-level and deep-sea temperature syntheses based on global δ18O stack19,57, one of which recently deconvolved the mid-Miocene δ18Obf decline into in 2.5 °C deep-sea cooling and 25 m of concurrent global average sea-level drop19. The discrepancy in δ18Osw between the study by Rohling et al. 19. and the clumped isotope data is driven by the difference in absolute BWTs and the magnitude of BWT decline. The uncertainty of the Δ47-based BWT (from 9.9 ± 4 °C to 5 ± 2.5 °C) may allow for some ice volume change (0.3 ± 0.5‰). Given the very similar MCO BWTs derived from multiple sites globally36,37 we deem the average MCO BWT of 9.9 °C reliable. By binning our MCO data with that of other sites, the uncertainty in that interval can be further reduced to ±1 °C. The 5 ± 2.5 °C BWT in the 10–9 Ma interval is based on most replicates, and thus has the smallest uncertainty. Only when the 10–9 Ma BWT is at the high end of its 95% confidence interval, can the global 25 m RSL (Relative Sea Level) ice volume build-up of Rohling et al.19 be replicated with our Δ47 data. Apart from δ18O-based reconstructions, local eustatic reconstructions of the mid and late Miocene are relatively crude and scarce58,59, and inevitably hindered by local tectonic activities60,61. Nevertheless, the inconsistence between Δ47-based absolute δ18Osw (therefore ice volume) and far-field sea level reconstructions is a general arising question dependent on calibrations56,62 and requires further exploration38,63,64. Given the low probability of that scenario, we conclude that the clumped isotope data imply a stronger cooling and thus less ice volume build-up during MMCT than in the model of Rohling et al.19.

Although we have confidence in our Δ47-based BWT reconstructions, the higher-than-modern δ18Osw for the mid-Miocene (and thus a larger than modern global ice volume) seems difficult to reconcile with evidence for Antarctic-proximal sea surface warmth24,30, Mg/Ca-based deep-sea warmth65 and high pCO235,66,67 during the MCO. The relatively stable long-term δ18Osw trend (Fig. 5b) also seems hard to reconcile with major episodes of seaward Antarctic ice expansion across the MMCT, e.g., as suggested by ice-rafted debris23,24,68. The only scenario that reconciles all these observations is one whereby a thick AIS was situated inland at the MCO, without marine terminations69. Such a high, inland ice sheet would also lead to relatively low oxygen isotope ratios of Antarctic ice70,71, because the higher-altitude ice sheet would receive less precipitation, and with a lower δ18O72,73. Thus, smaller ice volume would be needed for the mass balance if the δ18O of mid-Miocene land ice was lower than previously assumed. The question is whether such a geometric change in the ice sheet with stable ice volume is dynamically plausible, under realistic boundary conditions. Understanding the detailed interactions between the ocean, climate and ice sheet involved in this situation requires extensive modelling. Here, as a first step, we test the basic viability of a significant change in the volume-to-area ratio of the Miocene Antarctic ice sheet using a stand-alone ice sheet model60, applying a prescribed precipitation anomaly in conjunction with increased ocean heat (“Methods” section and Fig. 6a, b). This leads to large-scale glaciation at a ~ 100 ppm higher CO2 level than in the standard setup, yielding a thickened ice sheet interior while the build-up of ice shelves is prohibited and thereby ice area growth impeded (Fig. 6c). Furthermore, from an ice-dynamical perspective, the volume-to-area ratio of the Antarctic ice sheet waxing and waning on orbital timescales is also affected by the forcing amplitude and frequency, because the ice sheet area generally responds faster than volume to climate changes74. This implies that a decreased frequency or amplitude around the same mean of forcing variability could lead to an ice sheet that is less extended towards the margins but thicker in the interior, and hence equally voluminous74.

Fig. 6: Antarctic ice volumes modelling.
figure 6

a Simulated equilibrated Antarctic ice volumes at different CO2 levels, and (b) the relation between ice volume and ice area, yielded by a 3D thermodynamical ice sheet/shelf moderl (Methods). Results are obtained using the standard climate forcing (solid lines) and applying a fixed precipitation increase and enhanced sub-shelf melt rates (dashed lines). c Equilibrated ice thickness difference between the reference simulation at 392 (red) ppm and the simulation with anomalous forcing at 504 ppm (blue). This transition (from the blue to the red symbols) exemplifies our hypothesised Antarctic ice sheet change across the mid-Miocene Climatic Transition.

Full size image

Following the hypothesis of a dynamic AIS geometry, then, at the MMCT the AIS increased in surface area, advanced seawards, and reduced in height (“Methods” section and Fig. 6, switch from blue to red symbols). When the AIS undergoes spatial expansion, the periphery of the ice sheet receives a greater proportion of precipitation as compared to the central region. As a consequence of precipitation starvation in the hinterland, the overall elevation of the central AIS reduces. Such a change in geometry would have left global ice volume relatively unaltered but would have had large consequences for ice-ocean interactions and regional climate. Marine-terminating ice sheets provide profound regional cooling to have sea ice expanding75. The latitudinal position of westerlies and the sea ice edge determine the position of the STF7, in absence of continental obstructions76. So, in principle, the gradual northwards migration of the STF that we reconstruct is in line with the abundant evidence of seawards land ice expansion across the MMCT21,23,34: this induced more marine-terminating ice sheets, and through that a more extensive sea ice. Also on orbital time scales, it is found that marine-terminating ice sheets were strongly sensitive to local solar insolation changes forced by obliquity34 and so was Southern Ocean paleoceanography77 (Fig. 5e). This local cooling of high latitudes reduced BWT and pushed ocean fronts northward.

The dinocyst assemblages, combined with previously published SST reconstructions27 demonstrate the profound latitudinal changes of the STF. In the mid-Miocene, dinocyst assemblages were surprisingly similar between Site 1168 and the Antarctic margin24,31,78. Yet, perhaps counterintuitive, the latitudinal SST gradient between Australia27 and the Ross Sea30 was largest during MCO. This is because the south Australian Margin was ~10 °C warmer than today, while the inner basins of the Ross Sea remained under local influence of the Antarctic ice sheet and thus relatively cold30. In any case, the strong latitudinal SST gradient testifies to the presence of ocean frontal systems that separated mid-latitude water masses from polar water masses. Our reconstructed STF migration is not without corroborating evidence. The southwards STF migration at <7 Ma is coincident with a rapid drop in radiolarian abundance at the East Tasman Plateau79,80 and decreased K% (Potassium) in southwest Australia81, both interpreted as a southerly shift in the frontal systems and westerlies relative to the Australian continent. At the same time, at the Agulhas Plateau82 and in the South Atlantic83, oceanographic reconstructions suggest an equatorward migration of oceanic fronts, rather than a southward migration as in Australia. This suggests an asymmetric behaviour of oceanic fronts around Antarctica.

With BWTs around 5 °C at 7–5 Ma, we can attribute the progressive rise in δ18Obf of 0.2‰ between ~9 and ~6 Ma to about 20 m RSL-equivalent global ice volume build-up (Fig. 5b). This is concurrent with the early significant ice accumulation in Greenland and South America84,85, and expansion of the west AIS86, along with enhanced ice-rafted debris off east Antarctica87 (Fig. 5c, d). The current clumped isotope data compilation (Fig. 5) points to the latest Miocene as the phase of profound global ice volume build-up, rather than the MMCT.

The long-term Southern Ocean BWT cooling signal reconstructed from Site 1168 reflects a high-latitude surface ocean cooling, notably that of the region of deep-water formation. These surface waters where deep-water formed were arguably impacted by the seaward expansion of the ice sheet through katabatic winds88. This process expanded sea ice, pushing the westerlies and the STF northward (Fig. 4). In this scenario, the cooling of high latitude surface water and spatial extension of ice reduces the ocean-land thermal contrast and strengthens the polar vortex, leading to less moisture and precipitation transported into Antarctica89,90. The progressive cooling of subantarctic waters and increased vertical mixing induced by the northwards-migrated westerlies would have increased the efficiency of the subantarctic ocean carbon sink, the largest single ocean carbon sink system on the planet. As such, the geometric change of the ice sheet could have induced a more efficient ocean carbon storage in the subantarctic zone, which in turn contributed to the lowering of atmospheric pCO291 in the Miocene (Fig. 5f).

Taking the above together, the available data show little evidence for Miocene ice volume increase forced by CO2-induced global cooling with polar amplification29. First, Neogene surface ocean cooling was not amplified towards the polar regions, as the SST gradient was the largest in the warm MCO and decreasing over the mid-to-late Miocene. Second, the combined STF, BTW and deep ocean δ18O reconstructions suggest that regional temperatures mostly changed due to geometric changes of the Antarctic ice sheet, rather than the other way around. Northwards expansion of sea ice and subpolar conditions occurred because of advancing marine-terminating ice sheets which induced profound regional cooling. Finally, time intervals with progressive pCO2 decline (MMCT) seem to lack global ice volume increase, while time intervals with relatively stable pCO2 (late Miocene) seem to have profound ice volume growth, suggesting a large role for non-linear feedbacks. These fundamental observations put a perspective on the way radiative forcing and complex feedbacks in ocean-ice-atmosphere interactions shaped Neogene ice volume and global climate trends.

Methods

Site description

ODP Site 1168 (42°36.5809′S; 144°24.7620′E; 2463 m modern water depth) (Fig. 1) is located on the continental slope of the west-Tasmanian continental margin, with a modern seafloor temperature of 2.5 °C92. The site sits on the northern edge of the Subtropical Convergence zone, which separates warm, saline subtropical waters from comparably cold and fresh subantarctic water masses93. During the Neogene, the location of Site 1168 tectonically drifted along with Tasmania and Australia from 52°S at 23 Ma to its modern position at 42 °S94. The Neogene bathymetry was lower bathyal/upper abyssal (1000–2500 m), midway on the continental slope92. During this northward tectonic drift, the Southern margin of Australia was continuously bathed by the eastward flowing (proto-) Leeuwin Current40,95. Hence, Site 1168 is well-suited to study the Neogene evolution of the STF. We applied the same age model for the sediments as in Hou et al.27. (Fig. S5).

Palynology

We studied 131 samples for palynological content96. The processing of sedimentary samples for palynological analysis followed standard procedures at the GeoLab of Utrecht University97. Dried sediment samples were crushed and weighed (on average 10 g, standard deviation, SD, of <1 g) before they were dissolved with 30% hydrochloric acid (HCl) and 38% hydrofluoric acid (HF) for carbonate and silicate removal, respectively. The remaining palynological residues were sieved on a 10 μm nylon mesh, using an ultrasonic bath to disintegrate agglutinated organic particles. The palynological residues were mounted on glass slides using glycerine, sealed, and counted (under 200 and 400 magnification) using an Olympus CX41 optical microscope. When possible, at least 200 dinocyst specimens were counted98. Samples containing less than (including) 50 dinocyst specimens were excluded for further analysis and interpretation.

We further applied the model of Thöle et al.42. (Fig. S6) to infer paleoceanographic conditions from dinocyst assemblages. Specifically, we inferred the 25–75% SST ranges of the clusters in Thöle et al.42. that the downcore assemblages compared most to (Fig. 2).

Foraminiferal preparation

Each sediment sample was freeze-dried, washed over a 63 μm sieve, oven-dried at 50 °C and then dry-sieved into different size fractions. We mainly picked tests of Cibicidoides mundulus from the 250–355 μm size fraction for our measurements. We cracked open the picked specimens and ultrasonicated the test fragments in deionized water (3*30 s) to remove adhering sediment, organic lining and nannofossils. The test fragments were dried at room temperature overnight. In order to obtain enough material, other benthic species are also processed. We use Cibicidoides mundulus and Cibicidoides (Planulina) wuellerstorfi for both stable and clumped isotopes analyses. Data from other benthic or infaunal species Pyrgo sp., Gyroidina soldanii, Uvigerina peregerina are only used for clumped isotopes99 (Fig. S2).

Clumped isotope analysis

Clumped isotope measurements were performed using Thermo Scientific MAT 253 and 253 Plus mass spectrometers at the GeoLab of Utrecht University. Both mass spectrometers were coupled to Thermo Fisher Scientific Kiel IV carbonate preparation devices. CO2 gas was extracted from carbonate samples with phosphoric acid at a reaction temperature of 70 °C. A Porapak trap included in each Kiel IV carbonate preparation system was kept at 120 °C to remove organic contaminants from the sample gas. Between each run, the Porapak trap was heated at 120 °C for at least 1 h for cleaning. Every measurement run included a similar number of samples and carbonate standards100. In all, 3 carbonate standards (ETH-1, 2, 3) with different δ13C, δ18O and Δ47 compositions and ordering states were used for monitoring and correction of the results101. Two additional reference standards (IAEA-C2 and Merck) were measured in each run to monitor the long-term reproducibility and stability of the instrument. We achieve the necessary precision by averaging ~30 clumped isotope values measured on small (70–100 μg) carbonate samples101,102,103,104. External reproducibility (1 standard deviation) in Δ47 of IAEA-C2 after correction was 0.033‰. The δ13C and δ18O values (reported relative to the VPDB scale) of IAEA-C2 showed an external reproducibility (1 standard deviation) of 0.18‰ and 0.21‰, respectively105.

Deep sea temperature and δ18Osw calculation

We converted the sample Δ47 values (averages over ~30 separate measurements each) into temperature (T, in °C) using a calibration based on various recent datasets from core-top-derived foraminifera, corrected with the same carbonate standards as used in our study62:

$${{{{{\rm{T}}}}}}=\sqrt{\frac{0.0431\times {10}^{6}}{{\Delta }_{47}-0.1876}}-273.15$$
(1)

Δ47 -based BWTs were used in combination with δ18Obf to calculate δ18Osw (reported relative to VSMOW) with Eq. (9) of Marchitto et al.106:

$$\delta^{18} {{{{{\rm{O}}}}}}_{{{{{\rm{bf}}}}}} ({{{{{\rm{VPDB}}}}}}) - \delta^{18}{{{{{\rm{O}}}}}}_{{{{{{\rm{sw}}}}}}} ({{{{{\rm{VSMOW}}}}}})+0.27=(-0.245 \pm 0.005) \\ \times {{{{{\rm{T}}}}}}+(0.0011 \pm 0.0002) \times {{{{{\rm{T}}}}}}^2+(3.58 \pm 0.02)$$
(2)

For these calculations, δ18Obf values of the genus Cibicidoides were averaged over the same intervals as have been used for Δ47 averaging. Calibration uncertainties and measurement error were addressed by applying error propagation. The Meinicke et al.62,107,108] calibration error was propagated using the R package (clumpedcalib)109 that utilised a bootstrapped York regression slope-intercept pairs to the bootstrapped mean values for each bin. It should be noted that the calibration error is very small compared to the analytical error. The new calibration56 excludes benthic foraminifera-based data and it shifts the BWTs parallelly ~2.4 °C colder. As a result, the magnitude of cooling and relative ice volume changes are unaffected (Fig. S4). Since we are using benthic foraminifera as substrates, we decide to keep using the Meinicke calibration.

Ice sheet modelling

To demonstrate the viability of a precipitation regime change leading to a fundamentally different volume-to-area ratio of the Antarctic ice sheet, we deploy the 3D thermodynamical ice sheet/shelf model IMAU-ICE v1.1.1110,111. In the standard set-up90, climate forcing follows from pre-run warm and cold snapshot climate simulations112. The applied climate forcing is transiently calculated based on the prescribed CO2 concentration and the modelled ice sheet size, through a matrix interpolation method110. Sea ice is included in the climate model forcing, but a dynamical response is not calculated by IMAU-ICE. Equilibrium experiments are performed at various CO2 levels between preindustrial and 3x preindustrial CO2 values, with insolation at present-day levels and initiated from an ice-free Miocene Antarctic topography113. Here, we perform additional sensitivity experiments, in which we apply a fixed precipitation increase and enhanced sub-shelf melt rates. The precipitation anomaly is calculated as 25% of the warm snapshot precipitation fields, sub-shelf melt rates are set to 400 m/yr114,115.

These sensitivity experiments yield large-scale glaciation at a higher CO2 level (Fig. 6a), and an overall increased volume-to-area ratio (Fig. 6b). Notably, simultaneously reducing the CO2 level from 504 to 394 ppm and removing the anomalous forcing, leads to significantly larger ice sheet area, while the interior ice sheet height is severely reduced (Fig. 6c and S7). These idealised experiments exemplify our hypothesised Antarctic ice sheet change at the MMCT.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.