Main

Ice-core and surface-ocean proxy records provide a wealth of evidence supporting a central role of the Southern Ocean in modulating the air–sea partitioning of carbon during ice ages4,5. The Southern Ocean could exert a substantial control on the partial pressure of CO2 (), owing to its leverage on the efficiency of the global soft-tissue pump2 (STP) by which the photosynthetic production, sinking and remineralization of organic matter store dissolved inorganic carbon (DIC) in the ocean interior. At present, vertical exchange of water causes deeply sequestered CO2 and nutrients to be brought rapidly to the Southern Ocean surface, while iron (Fe) limitation of phytoplankton prevents the exposed nutrients from being entirely fixed back to organic matter before the water sinks again, allowing a net release of CO2 to the atmosphere. Either decreasing the rate of vertical exchange or enhancing export production would raise DIC concentrations in deep waters, lowering atmospheric CO2. Higher DIC, in turn, enhances deep water corrosivity, dissolving carbonate (CaCO3) minerals and thereby increasing ocean alkalinity, which produces a further CO2 drawdown2.

Phytoplankton productivity in the Subantarctic Zone of the Southern Ocean surface was stimulated during glacial periods by an enhanced supply of Fe-bearing dust, which would have reduced the leakage of CO2 to the atmosphere, strengthening the overall STP6,7. Meanwhile, radiocarbon measurements on deep-sea corals and benthic foraminifera have been interpreted as showing weakened exchange between the surface and the deep Southern Ocean during the Last Glacial Maximum (LGM)3,8; if this is correct, such a weakened exchange would have complemented the effect of dust9. From the LGM to the Holocene, the combination of dwindling dust inputs and accelerated vertical exchange in the Southern Ocean would have released carbon from the deep sea5, contributing to the rise of atmospheric CO2. However, as yet, very little evidence has been reported for STP-driven CO2 storage in the deep Southern Ocean that extends before the LGM.

Here we report new geochemical observations of changes in the oxygenation of the deep Southern Ocean, chronicling STP carbon storage over the last 80,000 years (80 kyr). DIC storage by the STP, as defined here, is equal to zero in the ocean’s surface layer, even though surface waters can contain large amounts of residual carbon inherited from deep waters owing to the slow air–sea exchange of CO2. This non-equilibrium component in surface waters is referred to as ‘disequilibrium DIC’ (ref. 10) and would be affected by sea ice, whereas STP DIC storage is independent of sea ice cover. Vertical stable carbon isotope (δ13C) gradients in the Southern Ocean show large changes on glacial–interglacial and millennial (that is, 1–5 kyr) timescales11. Although these gradients are influenced by storage of respiratory CO2 in the deep ocean, they are also susceptible to large disequilibrium effects that may have decoupled δ13C from DIC storage, since the air–sea equilibration of carbon isotopes is still an order of magnitude slower than for DIC itself12. Because dissolved oxygen (O2) is consumed stoichiometrically during the respiration of sinking organic matter, while being replenished during air–sea exchange an order of magnitude faster than CO2, it can provide a robust constraint on STP-induced carbon storage.

New measurements were made on two sediment cores raised from 2,800 m and 3,600 m depth at sites just to the north of the Weddell Sea (Methods, Fig. 1), where the bulk of Antarctic Bottom Water (AABW) forms through shelf processes and deep convection13. Modern observations show the presence of man-made chlorofluorocarbon-11 (CFC-11) near the core site, unambiguously identifying waters recently ventilated at the Southern Ocean surface (Fig. 1a). Idealized simulations with a fully coupled ocean–atmosphere–biogeochemistry model show that when deep convection in the Weddell Sea is weakened, waters at the core locations become depleted in oxygen, even in a simulation for which North Atlantic Deep Water is somewhat enhanced (Extended Data Fig. 1). At the same time, any change in export production and remineralization rates, including Fe fertilization by abundant glacial dust mobilized from South America during ice ages, would have altered the sinking flux of organic matter to the deep Southern Ocean, thereby changing the consumption rate of oxygen. Prior reconstructions have shown that glacial export fluxes were higher only in the Subantarctic Zone of the Southern Ocean14, so that the Antarctic Zone sites explored here would have been affected by the southward mixing of oxygen-depleted waters from lower latitudes. Thus, the core sites lie at a critical nexus of the Southern Ocean, where the impact of both deep southern ventilation and Fe fertilization on the STP can be recorded by the sedimentary redox state.

Figure 1: Modern oceanographic context.
figure 1

a, b, Present-day dissolved oxygen concentration26 is shown on a meridional transect of the South Atlantic, averaged between 25° W and 10° E (a) and on a horizontal surface at a water depth of 2,800 m (b). Contours in a show the concentration of chlorofluorocarbon-11 (CFC-11), in picomoles per kilogram (ref. 27). Both core site locations are bathed by well oxygenated, recently ventilated subsurface water masses and observations suggest that the two sedimentary archives are sensitive recorders with which to monitor changes in deep ocean oxygenation.

PowerPoint slide

Full size image

The fundamental control on sedimentary redox conditions is the balance between the rate of supply of oxygen from overlying bottom waters and the rate of pore-water oxygen removal associated with respiration, which is regulated by the supply of organic matter. Comparing the sedimentary distribution of redox-sensitive trace metals with biogenic fluxes to the sea floor thus permits these processes to be disentangled, so that relative changes in oxygen concentrations at the water–sediment interface can be inferred. Manganese (Mn) enrichments are typically preserved in well oxygenated sediments, whereas dissolved uranium (U) typically diffuses across the water–sediment interface and subsequently precipitates from pore waters under more reducing conditions (Methods). These mineral precipitates are referred to as authigenic phases. Monitoring the sedimentary distribution of these two metals thus provides sensitivity across the full range of oxygen concentrations typically encountered in the open ocean.

At site TN057-13PC, authigenic U concentrations are highest during peak glacial conditions (LGM in Fig. 2c) and decrease rapidly thereafter, coinciding with the increase in opal flux (Fig. 2e), a proxy for the rain of organic material to the seabed12 (Extended Data Fig. 2). Higher authigenic U during intervals of lower organic carbon supply to the sea floor can only reasonably be the result of decreased bottom-water oxygen concentrations15,16, as also observed in prior low-resolution measurements (Extended Data Fig. 3 and Extended Data Table 1), indicating increased storage of remineralized carbon in the deep Southern Ocean during the LGM. Given that the solubility of oxygen in seawater is greater in cold waters, the observation that the deep Atlantic was less well oxygenated during the LGM indicates that the solubility effect on oxygenation was overwhelmed by larger changes in the accumulated utilization of oxygen in the ocean interior10.

Figure 2: Palaeoclimatic reconstructions covering the last glacial termination at site TN057-13PC compared with ice core records.
figure 2

a, b, (a; ref. 28) and atmospheric δ13C (b; ref. 4, error bars represent the standard deviation of replicate measurements) from Antarctic ice cores. c, Sedimentary authigenic U concentrations (reversed scale). d, Sedimentary Mn/Al. e, 230Th-normalized biogenic opal flux29. Shaded intervals highlight cold periods in the Southern Hemisphere. YD, Younger Dryas; ACR, Antarctic Cold Reversal; HS1, Heinrich Stadial 1. The stippled line represents the bulk continental crust (BCC) Mn/Al background value (0.0167; ref. 30). The error (1σ) related to the determination of authigenic U is smaller than the symbol size. p.p.m., parts per million; p.p.m.v., parts per million by volume; VPDB, Vienna Pee-Dee belemnite standard.

PowerPoint slide

Source data

Full size image

Subsequently, during Heinrich Stadial 1 (HS1, 17.5–14.7 kyr ago), authigenic U accumulation decreased (Fig. 2c) while Mn enrichments rose above the detrital background (Fig. 2d), indicating improved oxygenation of bottom water. This is in striking contrast to the change observed in the deep North Atlantic, where oxygen concentrations fell to their lowest values during HS1, reflecting the reduced input of oxygenated North Atlantic Deep Water17. Increased oxygenation during HS1 must have included the effect of decreasing organic matter flux in the Subantarctic Zone, given the rapidly dwindling Fe supply to the surface ocean6, but the opposing behaviour with the North Atlantic is also consistent with a north–south ventilation seesaw18. Importantly, a strong Mn enrichment recurred during the Younger Dryas (12.8–11.5 kyr), when the changes in Fe supply were too small to have altered deep water oxygenation much7 (Extended Data Fig. 4). Thus, greater oxygenation of deep water at this location during the Younger Dryas event provides evidence for an increased proportion of oxygen-rich AABW relative to oxygen-depleted circumpolar deep water. A northward shift of the circumpolar fronts may have contributed to this increased oxygenation, but if spatial gradients of dissolved oxygen were similar to those that exist today (Fig. 1) then more rapid formation of AABW must have occurred as well, to induce the inferred increase in oxygen. Together, the records imply two phases of STP weakening in the Southern Ocean: the first (HS1) included combined effects of dust supply and a north–south seesaw of ventilation, and probably reflected most of the net glacial–interglacial change10, while the second (Younger Dryas) was dominated by a transient pulse of ventilation alone. These deglacial observations are consistent with the stable carbon isotopic composition of atmospheric CO2 (δ13Catm) (ref. 4) (Fig. 2b) in suggesting that carbon was removed from the deep Southern Ocean during two episodes of improved deep ocean oxygenation that occurred during HS1 and the Younger Dryas.

Prior to the LGM, atmospheric CO2 concentrations rose repeatedly by 10–20 parts per million during periods of intense cold conditions in the North Atlantic, while Antarctica warmed. These intervals correspond to perturbations of the Atlantic Meridional Overturning and advances of sea ice in the North Atlantic, similar to HS1 and the Younger Dryas19, when deep North Atlantic O2 concentrations plummeted17. Our new data show that each of these intervals was also marked by decreasing authigenic U concentrations (Fig. 3b), concurrent with a greater rain of organic detritus to the sea floor (Fig. 3c). Thus, similar to the millennial-timescale changes of the last glacial termination, the oxygenation proxies are consistent with the release of deeply sequestered remineralized carbon to the atmosphere during climate oscillations between 35 kyr and 80 kyr ago. Despite uncertainties in age models and the potential for nonlinear behaviour in authigenic U accumulation, the Pearson’s correlation coefficient r2 between the TN057-14PC authigenic U and the ice-core CO2 is 0.63 (P = 10−19, n = 89) over the period 35–80 kyr ago. We note less consistency between 25 kyr and 35 kyr ago, when authigenic U was lower than expected from the level; we were unable to find any sedimentological artefact that might have introduced a bias over this interval (Extended Data Fig. 5), and therefore suggest that at this time, during which there was a major expansion of the Northern Hemisphere ice sheets20, was lowered by a mechanism unrelated to the Southern Ocean’s control on the STP, although this result should be further tested.

Figure 3: Palaeoclimatic reconstructions covering the interval MIS2 to MIS5a at site TN057-14PC and Ocean Drilling Program (ODP) site 1090 (42.9° S, 8.9° E, 3,700 m) compared with ice core records.
figure 3

a, Relative abundance of diatom Eucampia antarctica (blue)—a sea surface temperature proxy—in TN057-14PC (ref. 29) compared to the Antarctic ice core local temperature reconstruction (light grey)31 on the new AICC2012 age scale32. b, Ice-core atmospheric reconstructions33,34 (dark grey) and sedimentary authigenic U concentrations at site TN057-14PC (black, reversed scale). c, 230Th-normalized biogenic opal flux at site TN057-14PC (ref. 29). d, 230Th-normalized Fe (brown) and alkenone (green) fluxes at ODP site 1090 (ref. 6). Triangles indicate the tiepoints used to generate the new version of the TN057-14PC age scale. Shaded intervals highlight cold intervals in the Southern Hemisphere, whereas unshaded intervals correspond to North Atlantic cold periods (Antarctic Isotope Maximum (AIM) 1–4 in the Southern Hemisphere). The error (1σ) related to the determination of authigenic U is smaller than the symbol size. SMOW, Vienna mean ocean water standard.

PowerPoint slide

Source data

Full size image

Our results show that carbon storage in the deep Southern Ocean was generally amplified when Antarctica was cold, consistent with arguments based on the similarity of Antarctic temperature and atmospheric CO2 over glacial cycles. However, the results also highlight a fundamental difference between the glacial–interglacial transition, during which the O2 changes in the north and south deep Atlantic correlate, and the millennial-timescale changes, for which O2 changes are antiphased between north and south. This difference can be explained by a contrast in the dynamics of ocean circulation, which were dominated by the bipolar ventilation seesaw on millennial timescales21 rather than the steady-state adjustment to different CO2 levels, orbital forcing and ice sheet configuration that determined the glacial–interglacial change10. Iron fertilization, caused by enhanced dust input during Antarctic cold phases6, would have contributed some degree of the observed Southern Ocean STP strengthening during both LGM and millennial-timescale changes. In this light, it may seem remarkable that changes in dust fluxes and deep ventilation were coordinated (Extended Data Table 2), reinforcing their synergistic impacts on the STP.

We suggest that the coordination between ventilation and Fe supply on millennial timescales occurred through a mechanistic link between the two. Model simulations have suggested that a weakening of the Atlantic Meridional Overturning Circulation can lead to enhanced Southern Ocean ventilation via direct ocean circulation changes21 as well as by forcing a migration in the Southern westerly winds22. In addition, it has been shown that a southward shift in the precipitation-bearing Southern Westerlies could initiate the retreat of Patagonian glaciers23. We suggest that, owing to this coupling of ocean and atmospheric circulation, millennial-timescale periods of enhanced deep Southern Ocean ventilation generally co-occurred with southward shifts of the Southern Westerlies, which simultaneously caused Patagonian glaciers to retreat, reducing both the production of fine-grained lithogenic material by glacial erosion24 and the supply of dust by trapping glacial flour in proglacial lakes25. Thus, shifts in atmospheric circulation, coupled with changes in the Atlantic Meridional Overturning Circulation, would have caused the intensity of Fe limitation and upwelling to co-vary on millennial timescales, while the intensity of deep convection was altered by both atmospheric and oceanic drivers. These coupled processes, coordinating the input flux of carbon to the deep sea as well as its release by deep convection and upwelling, can explain the remarkable power of the Southern Ocean in modifying the efficiency of carbon storage over millennial timescales, and thereby atmospheric CO2 levels.

Methods

Material and sediment composition

TN057-13PC (53.2° S, 5.1° E, 2,848 m) and TN057-14PC (52.0° S, 4.5° E, 3,648 m) were retrieved from the Atlantic sector of the Southern Ocean, south of the present-day position of the Antarctic Polar Front (Fig. 1). The cores contain diatomaceous ooze sequences with discontinuous carbonate-bearing intervals and occasional ice-rafted debris layers. The deglacial sequence in core TN057-14PC is disturbed and thus was not further considered (Extended Data Fig. 6).

Age model

The TN057-13PC age model is taken, unchanged, from ref. 29. The stratigraphy of sediment core TN057-14PC has been updated on the basis of graphical correlation between the relative abundance of diatom E. antarctica29, a proxy for sea-surface temperature, and EDC deuterium (δD) (ref. 31) expressed on the new AICC2012 age model32, assuming an in-phase relationship (Fig. 3). Sedimentation rates are high, typically in the range <15–60 cm kyr−1, for glacial and peak interglacial intervals, respectively.

Analytical methods

230Th-normalized fluxes were evaluated as described in ref. 35. Concentrations of U and Th isotopes were determined by isotope dilution inductively coupled plasma mass spectrometry (ICP-MS)36. The authigenic U content of each sample was evaluated using the measured 238U and 232Th concentrations using an activity ratio of 0.5 for the detritic endmember37. Measured 230Th concentrations were corrected for the detrital contribution using 232Th and for ingrowth produced by authigenic U, and then decay-corrected to the time of deposition to derive 230Th-normalized fluxes. Sedimentary aluminium (Al), manganese (Mn) and barium (Ba) concentrations were measured by inductively coupled plasma optical emission spectroscopy (ICP-OES; Varian Vista Pro) by ALS Minerals, Canada. Precision was better than 2% for replicate measurements. The biogenic fraction of barium—a useful proxy for integrated export production38—was determined using a detritic Ba/Al ratio of 0.0067 (ref. 30). The calculations of normative Ba/Al and Mn/Al—the excess Ba and Mn fractions relative to the detrital background, respectively—are based on the assumption that the composition of the Al-bearing phases of the terrigenous material remained constant in space and time. CaCO3 was quantified by coulometric CO2 determinations assuming no other carbonate-bearing phase was present. Precisions are ±3% of the reported values.

Redox-sensitive metals

The sedimentary distribution of redox-sensitive trace metals can be influenced by a number of factors, including (1) the rate of organic matter delivery to the sea floor, (2) changes in sediment accumulation rate, and (3) changes in bottom-water oxygenation. Comparing the abundance of multiple trace metals (in our case, Mn and U) in sediments with biogenic flux to the sea floor reconstructed using 230Th normalization permits these processes to be distinguished. It follows that relative change of oxygen concentrations at the water–sediment interface can be inferred.

The sedimentary geochemistry of Mn is dominated by the redox control of its speciation, with higher oxidation states (Mn(III) and Mn(IV)) occurring as relatively insoluble oxyhydroxides in well oxygenated environments and the lower oxidation state (Mn(II)) being much more soluble in oxygen-depleted settings39. Mn enrichments in the form of Mn-oxide coatings are observed in deep sea sediments where oxic conditions prevail to greater sediment depths as a result of low organic matter respiration rates and well ventilated bottom waters. In sediments that are deposited under reducing conditions, there is no accumulation of Mn oxides and the Mn concentrations of buried deposits are low and entirely controlled by their insoluble detritic fraction. The presence of Mn concentration in excess of what can be expected from the detritic fraction thus suggests that the host sediment must have accumulated under oxygen-replete conditions.

Uranium behaves conservatively in oxygenated seawater and is present as U(VI). One of the primary removal mechanisms for U from seawater is via diffusion across the water–sediment interface of reducing sediments40. It appears that for sedimentary redox conditions near those for conversion of Fe(III) to Fe(II) (that is, suboxic conditions), soluble U(VI) is reduced to insoluble U(IV) (ref. 41). This lowers the pore-water U concentration, and dissolved U from the overlying bottom water can diffuse into the sediment along the declining concentration gradient, leading to an enrichment of authigenic (or excess) U in the sediment.

The emplacement of both Mn and U authigenic mineral phases are uncorrelated to 230Th-normalized opal fluxes (Figs 2 and 3, Extended Data Table 2). As a result, we infer that the variability in organic carbon delivery to the sea floor—and its subsequent respiration below the sediment–water interface—was not the primary factor controlling the sedimentary redox conditions.

The sedimentation rates inferred for both sediment cores are high enough to minimize filtering by bioturbation and to prevent post-depositional remobilization of U upon deepening of the sedimentary oxicline. Manganese, on the other hand, is frequently remobilized to the sedimentary pore fluids under reducing conditions. Dissolved Mn can thus migrate in the sedimentary column and (re)precipitate when oxic conditions are encountered39. As such, large Mn enrichments primarily reflect a shift from oxygen-depleted to oxygen-replete conditions. The sediment accumulation rate increased across the last glacial termination, potentially affecting trace-metal concentrations. However, both authigenic U and Mn/Al downcore records show opposite behaviours across the deglaciation, suggesting that sedimentary dilution was not the primary controlling factor driving the trace-metal distribution (Extended Data Fig. 7). Furthermore, the comparison between sedimentation rates and authigenic U concentrations in core TN057-14PC (Extended Data Fig. 5) clearly shows that the downcore authigenic U pattern was not dominated by changes in sediment accumulation. We thus posit that the observed authigenic enrichments of both U and Mn were sensitive to changing oxygen levels at the sediment–water interface.

Model simulations

The model simulations were carried out with the Geophysical Fluid Dynamics Laboratory (GFDL) coupled ocean–ice–atmosphere model CM2Mc, with the embedded biogeochemical model BLING42. Well equilibrated states of four simulations were used for the plots shown in Extended Data Fig. 1. Extended Data Fig. 1 shows the difference between two simulations with atmospheric CO2 concentrations of 220 parts per million and LGM ice sheet topography, but different obliquity (axial tilt). Strong convection and AABW production occurs under low obliquity (22°), while moderate convection occurs under high obliquity (24.5°). All other boundary conditions and parameters are identical in the two model runs. The plots were made from 100-year averages over the last century of 3,800-year simulations.