Evolution of South Atlantic density and chemical stratification across the last deglaciation

Significance

The cause of the rise in atmospheric pCO₂ over the last deglaciation has been a puzzle since its discovery in the early 1980s. It is widely believed to be related to changes in carbon storage in the deep ocean, but the exact mechanisms responsible for releasing CO₂ from the deep-ocean reservoir, including the role of ocean density stratification, remains an open question. Here we reconstruct changes in the intermediate-deep density gradient in the South Atlantic across the last deglaciation and find evidence of an early deglacial chemical destratification and a late deglacial density destratification These results suggest that other mechanisms, besides deep-ocean density destratification, were responsible for the ocean–atmosphere transfer of carbon over the deglacial period.

Abstract

Explanations of the glacial–interglacial variations in atmospheric pCO₂ invoke a significant role for the deep ocean in the storage of CO₂. Deep-ocean density stratification has been proposed as a mechanism to promote the storage of CO₂ in the deep ocean during glacial times. A wealth of proxy data supports the presence of a “chemical divide” between intermediate and deep water in the glacial Atlantic Ocean, which indirectly points to an increase in deep-ocean density stratification. However, direct observational evidence of changes in the primary controls of ocean density stratification, i.e., temperature and salinity, remain scarce. Here, we use Mg/Ca-derived seawater temperature and salinity estimates determined from temperature-corrected δ¹⁸O measurements on the benthic foraminifer Uvigerina spp. from deep and intermediate water-depth marine sediment cores to reconstruct the changes in density of sub-Antarctic South Atlantic water masses over the last deglaciation (i.e., 22–2 ka before present). We find that a major breakdown in the physical density stratification significantly lags the breakdown of the deep-intermediate chemical divide, as indicated by the chemical tracers of benthic foraminifer δ¹³C and foraminifer/coral ¹⁴C. Our results indicate that chemical destratification likely resulted in the first rise in atmospheric pCO₂, whereas the density destratification of the deep South Atlantic lags the second rise in atmospheric pCO₂ during the late deglacial period. Our findings emphasize that the physical and chemical destratification of the ocean are not as tightly coupled as generally assumed.

The last glacial termination was accompanied by an 80-ppm rise in atmospheric pCO₂ (1, 2), and it is widely believed that this increase in pCO₂ was driven by processes occurring within the Southern Ocean (3–5). These Southern Ocean processes are proposed to have released CO₂ from the deep ocean through a combination of decreased nutrient utilization (6), increased vertical mixing (7), and increased air–sea gas exchange (8). Geochemical records show evidence for an “old” (9) respired dissolved inorganic carbon pool in the glacial Southern Ocean below 2,500 m (10, 11) which became better ventilated over the course of the deglaciation (9, 12), supporting the idea that the deep ocean was isolated from the atmosphere during glacials. Over the deglacial period this chemical stratification between the deep ocean and the overlying intermediate ocean decreased, e.g., ref. 11, implying a change in circulation or ventilation within the Southern Ocean which enabled CO₂ to be upwelled and outgassed to the atmosphere (7). The chemical destratification of the ocean has been attributed either to (i) an increase in air–sea gas exchange, through a decline in the extent of sea ice (8) and/or a decrease in surface ocean stratification (13); or (ii) a breakdown in the density stratification between the poorly ventilated deep ocean and the better-ventilated water masses above (14). Evidence supporting either scenario remains elusive.

Pore-water profiles from deep-ocean sediments have provided the first estimates of the density of the deep ocean during the Last Glacial Maximum (LGM) (15). These studies found that the glacial deep ocean was highly saline [∼37 practical salinity units (psu)] and had an in situ density that was 2 kg/m³ denser than modern deep water. These studies lend support to the hypothesis that CO₂ storage within a highly stratified glacial ocean played a significant role in driving lower glacial atmospheric pCO₂. However, pore-water profiles only provide a “snapshot” of the physical properties of the deep ocean at the LGM, and do not provide information about the time-dependent changes in the density of deep water over the deglaciation. Thus, from these studies alone, it is impossible to assess whether the destratification of the deep-ocean density gradients drove the atmospheric pCO₂ increase over the deglacial period.

Isotope-enabled intermediate complexity models have been used to suggest a mechanistic link between the physical (density) and chemical (δ¹³C) properties of the ocean over glacial–interglacial timescales (16, 17). These models suggest that deep-ocean stratification, generated by the formation of dense brines during sea ice growth, is required to reconcile the spatial distribution of seawater δ¹³C. This result implies that a decrease in Antarctic sea ice, and therefore reduced brine formation, over the deglacial period will affect both the density of the deep ocean and its chemical properties synchronously. Testing this hypothesis of a mechanistic link between the physical and chemical properties of the ocean requires observational evidence of the density structure evolution of the Southern Ocean over the entire deglacial period.

Here, we determine the deglacial evolution of the intermediate-deep density gradient in the high-latitude South Atlantic Ocean by generating temperature and salinity proxy records over the last 20 ka at the intermediate depth site of sediment core GC528 (598 m; 58° 02.43′W, 53° 00.78′S) and the deep site of core MD07-3076Q (3,770 m; 14° 13.7′W, 44° 09.2′S) in the sub-Antarctic Atlantic (Fig. 1). We make the assumption that geochemical changes at a given depth occur synchronously within the South Atlantic (see the Supporting Information). Combined Mg/Ca and δ¹⁸O measurements on the benthic foraminifer Uvigerina spp. are used to estimate benthic seawater temperature (18) and to calculate the δ¹⁸O of deep and intermediate water masses (hereafter referred to as δ_w). Temperature and δ_w (closely related to seawater salinity) are combined to produce a continuous record of the evolution of the density gradient in the South Atlantic over the last deglaciation (Methods). We compare the evolution of the density gradient with benthic δ¹³C and ¹⁴C records from the two sites to assess the hypothesis of a causal link between the physical and chemical properties of the deglacial ocean.

Fig. 1.

Location of intermediate (GC528) and deep (MD07-3076Q) sites. Site locations overlain on a schematic map of ocean circulation for (Top) modern ocean and (Bottom) LGM, adapted from Ferrari et al. (41). The grayscale colors indicate the flow path of major water masses. Background colors indicate the relative salinity of water masses (blue, relatively fresh; red, relatively saline).

Deglacial Changes in the Physical Properties of the South Atlantic

Although the deglacial decrease in Uvigerina spp. δ¹⁸O is similar in both sites (Fig. 2C), our Mg/Ca-derived temperature reconstructions (Fig. 2B) suggest that the intermediate waters (GC528) were in fact cooler than the deep water (MD07-3076Q) for the majority of the glacial termination. This temperature inversion between the intermediate and deep sub-Antarctic reverses during the Early Holocene, from 9 ka onward. It is physically impossible for cold intermediate water to overlie warmer deep water (even accounting for the effect of adiabatic decompression) and remain dynamically stable unless these differences are compensated by salinity.

Fig. 2.

Deglacial records from intermediate (GC528) and deep (MD07-3076Q) water. (A) Atmospheric CO₂ [West Antarctic Ice Sheet (WAIS) Divide (50), black; EPICA Dome C (EDC) (51), gray]. (B) Mg/Ca-derived benthic temperatures for the intermediate site (GC528; green, open symbols) and the deep site (MD07-3076Q; brown, closed symbols). Thick bars at the start of each timeseries show modern temperature range. (C) Foraminiferal δ¹⁸O for the same sites as in B. (D) δ¹⁸O offset from contemporaneous global mean δ¹⁸O (δ_w-ice) for the same sites as in B. Large circles at the start of each timeseries show modern salinity at each core site. (E) In situ density of the intermediate site (GC528) and the deep site (MD07-3076Q). Modern in situ density shown by large circle at the start of each timeseries. The 1.5-ky spline and 1σ confidence interval for each plot shown as solid line and polygon, respectively.

We use the Mg/Ca-derived benthic temperatures and global sea-level records (19) to deconvolve δ¹⁸O of foraminiferal calcite (Fig. 2C) into its principal components (20), i.e. temperature and δ_w (Methods). To extract the salinity component from δ_w (21), we express δ_w as an offset from the contemporaneous “global mean” δ_w, hereafter referred to as δ_w-ice (Fig. 2D), by subtracting the isotopic effect of melting continental ice using global sea-level records (19). The isotopic effect of melting ice will have spatial and temporal variations which are masked in our subtraction of a global mean δ_w; however, without transient tracer models to prescribe a regional “ice volume effect,” the offset from the contemporaneous global mean δ_w remains the best approximation of ice sheet melt-derived δ¹⁸O changes in the South Atlantic. Salinity is calculated from δ_w-ice assuming that the modern relationship between salinity and δ_w for the Southern Ocean (21) holds over the deglacial period. Recent isotope-enabled fully coupled general circulation model (GCM) experiments have suggested that the salinity–δ_w relationship in the South Atlantic is more temporally constant than in other ocean basins (22), ruling out significant biases in our inferred salinity estimates due to potential variations in the salinity–δ_w relationship. However, brine rejection during the formation of sea ice increases salinity without fractionating oxygen isotopes; therefore, we suggest that salinity values derived from δ_w-ice are minimum estimates.

The difference in intermediate water salinity between the LGM and the Holocene is small (Fig. 2D) as most of the deglacial δ¹⁸O Uvigerina spp. variation can be accounted for by warming. By contrast, the deep-water site (MD07-3076Q) was relatively saline (36–37 psu) during the glacial period and salinity decreased by 3–4 psu during the late deglacial (12–9 ka).

Using the Mg/Ca-derived benthic temperatures and our estimates of minimum salinity, we calculate the in situ density of the intermediate and deep site according to the equation of state (23) (Fig. 2E). We find good agreement between our density estimates for the Holocene and modern South Atlantic density measurements (Fig. 2E). The deglacial density reconstructions indicate the presence of a strong density gradient between intermediate and deep water for the glacial and much of the deglacial period. This density gradient decreases dramatically from 12 ka, and by 10 ka the deep ocean is 2 kg/m³ less dense than at the LGM. Our Mg/Ca-δ¹⁸O–derived deep-ocean LGM salinities are broadly consistent with previous pore-water–based estimates (15) (Fig. 3). Although both reconstructions carry substantial uncertainty (ref. 24 and Propagation of Errors in the Calculation of Potential Density), the convergence of two independent methods strengthens our confidence in both approaches. Before 12 ka the high salinity of the deep-water site implies a strong density gradient, despite the temperature inversion of cool intermediate waters overlying warmer deep water. After 10 ka the salinity stratification has broken down, and a weaker density gradient is maintained by the temperature difference between intermediate and deep water. Provided the assumption of zonal seawater density continuity across the sub-Antarctic Atlantic holds true, our findings signal a significant mode switch in the primary physical parameters that govern ocean density stratification in the southern high latitudes through the last deglaciation.

Fig. 3.

Temperature and δ¹⁸O offset from contemporaneous global mean δ¹⁸O (δ_w-ice) evolution of Southern Ocean Water masses over the last deglaciation. Composite plot of intermediate water (GC528; green, open symbols) and deep water (MD07-3076Q; brown, closed symbols) evolution, ages in ka shown in circles. In situ density isopycnals of GC528 (green dashed lines) and MD07-3076Q (brown dotted lines) shown in background. Gray dots represent modern seawater T-δ_w-ice (30). CDW, Circumpolar Deep Water.

Processes Controlling the Physical Properties of Water Masses over the Deglaciation

To determine the factors affecting the variability in the physical properties of deep and intermediate water masses over the deglaciation, we compare glacial water mass temperature and δ_w-ice properties with those of water masses present in the Southern Ocean today (Fig. 3). Comparison of intermediate LGM temperature-δ_w-ice with modern measurements of seawater temperature and δ_w suggests that LGM intermediate water was analogous to modern Antarctic Surface Water (AASW), which is characterized by near-freezing temperatures (Fig. 3). Such low temperatures suggest that the source location of intermediate water was strongly influenced by sea ice (which forms from seawater with a temperature of −2 °C). Glacial sea ice reconstructions using diatom transfer functions (25, 26) predict that winter sea ice could have extended as far north as 57°S at this longitude. We therefore suggest that glacial intermediate water masses at GC528 reflect the cold and fresh signature of surface waters close to the winter sea ice edge that is influenced by seasonal sea ice meltwater. This is in stark contrast to modern intermediate water at GC528, which is strongly influenced by a modified component of upwelled Circumpolar Deep Water (27) (Fig. 3).

The processes controlling the benthic temperature of the deep site are more complex. Deep water formed close to the sea ice margin can only gain heat in two ways: (i) through mixing with other, warmer water masses, and (ii) through the accumulation of geothermal heat in the deep sea. The most probable warm water masses that can mix with sinking southern-sourced waters during the LGM are Drake Passage through flow waters from the Pacific. Although it has been shown that Pacific deep water was not significantly warmer than freezing (18, 28), it remains to be determined whether Pacific intermediate waters were significantly warmer and had an influence on southern-sourced deep water in the Atlantic. Alternatively, the source of warmth in the deep South Atlantic may be derived from the accumulation of geothermal heat, assuming that the deep ocean is stagnant and cannot lose heat to the surface (29). It has been shown by a conceptual model that it would take ten thousand years (ky) to heat 2 km of seawater by 2 °C, based on a heat flux of 500 mW/m² (29). Both of these processes may have potentially accounted for the observed warmth of the salty LGM deep water in the South Atlantic. However, an analysis of whether a completely stagnant and isolated deep-water pool is physically possible or whether it was significantly influenced by Pacific inflow waters goes beyond the scope of this study.

The deglacial evolution of deep-water temperature and δ_w-ice (Fig. 3) can be understood by invoking a combination of (i) changes in the northern versus southern mixing ratio and (ii) changes in the end-member δ¹⁸O of Antarctic Bottom Water (AABW). In the modern Atlantic Ocean, saline North Atlantic Deep Water (NADW) overlies fresh AABW (Fig. 1). The high δ¹⁸O signature of NADW reflects high rates of evaporation at the surface, whereas the δ¹⁸O of AABW is comparatively more negative (30). The warming and increase in δ_w-ice over much of the deglaciation (16–12 ka) may indicate a greater proportion of northern-sourced water at MD07-3076Q. This is supported by benthic foraminifer ε_Nd data that suggest an increasing contribution of northern-sourced water in the South Atlantic throughout the deglaciation (31). The late deglacial change in deep-water δ_w-ice to isotopically lighter values after 12 ka is likely driven by a change in the mode of AABW formation. The two different modes of southern-sourced deep-water formation, i.e., brine rejection during sea ice formation and supercooling of Ice Shelf Water (ISW) beneath the Antarctic ice shelves (32), impart very different signals on δ_w-ice. Brines have a δ¹⁸O signature close to surface water values (∼0‰ in the Southern Ocean), whereas ISW has a negative δ¹⁸O signature reflecting incorporation of overlying ice shelf meltwater. We would argue that the marked decrease in the δ_w-ice of deep water at 12 ka is related to an increasing contribution of ISW constituting AABW. This change may be associated with a retreat in the grounding line of Weddell Sea ice shelves and an intrusion of relatively warm modified northern-sourced water under the ice shelves, further melting the marine-terminating ice sheets around Antarctica (33). In summary, the modes of intermediate and deep water formation have a profound impact on the density structure of the sub-Antarctic ocean.

Link Between the Physical and Chemical Properties of the South Atlantic and Atmospheric pCO₂

To assess the hypothesis of a causal link between the physical and chemical properties of the ocean, we compare the timing of changes in the breakdown of the density gradient with changes in benthic foraminifer δ¹³C and foraminifer/coral benthic-atmospheric ¹⁴C ages at both core locations (Fig. 4). Benthic foraminifer δ¹³C at the intermediate water site (GC528) was more positive than the δ¹³C of the deep site [MD07-3076Q (34)] at the LGM (Fig. 4B). This large δ¹³C gradient breaks down over the deglaciation (15–10 ka) resulting in relatively homogeneous δ¹³C values at both sites during the Holocene. Foraminifer/coral benthic-atmospheric ¹⁴C age offsets between intermediate [Burdwood Bank (12), Chile Margin (35)] and deep water [MD07-3076Q (9)] are broadly consistent with the δ¹³C record (Fig. 4C). These two records provide strong support for glacial chemical stratification, which subsequently breaks down relatively early in the deglaciation.

Fig. 4.

Comparison between the intermediate-deep density gradient and the chemical gradient. (A) Atmospheric CO₂ [WAIS Divide (50), black; EDC (51), gray]. (B) Chemical properties, δ¹³C of foraminiferal calcite, for intermediate (GC528; green, open symbols) and deep water (MD07-3076Q; brown, closed symbols). (C) Apparent water mass ventilation age (B-Atm), intermediate water [Burdwood Bank; pale green, open symbols (12); Chile Margin; dark green, open symbols (35)] and deep water [MD07-3076Q; brown, closed symbols (9)]. (D) In situ density of the intermediate site (GC528) and the deep site (MD07-3076Q).

Comparison of the deglacial density records (Fig. 4D) and the chemical δ¹³C and ¹⁴C records (Fig. 4 B and C), shows a marked difference in the timing of the breakdown of density and chemical gradients. In the deep site, both the δ¹³C and ¹⁴C records begin to change early in the deglaciation (17–15 ka), and in the case of the foraminiferal ¹⁴C record, these early decreases in the benthic-atmospheric ¹⁴C age have been linked to a synchronous rise in atmospheric CO₂ (9). However, there is no concomitant change in the intermediate-deep density gradient at this time (Fig. 4D). The onset of the physical destratification occurs during the Early Holocene (∼10 ka), and appears to lag the late deglacial rise in atmospheric pCO₂. Our proxy records suggest that (i) changes in ocean chemistry in the South Atlantic occur without large-scale reorganization of the ocean’s density structure, challenging the propositions of a close coupling between physical and chemical ocean stratification as suggested by intermediate complexity models (17), and (ii) the density destratification of the South Atlantic could only have impacted atmospheric pCO₂ during the late deglaciation.

CO₂ stored in the deep ocean can be impeded from being released back to the atmosphere by two physical processes: (i) via an increased residence time of deep-water masses in the ocean interior, through an increase in deep-ocean stratification acting as a lid to deep carbon (7), or (ii) reduced efficiency of air–sea gas exchange in the regions of deep mixing and upwelling (8). We suggest that during the LGM, both of these processes will have contributed to lower atmospheric pCO₂, owing to increased density stratification in the Southern Ocean and because permanent sea ice (8) and/or shallow stratification (13, 36) acted as a barrier preventing CO₂ from escaping the surface of the Southern Ocean. Indeed, it is plausible that the retreat of sea ice during the early deglaciation (26, 37) effectively removed a barrier to air–sea exchange, and thus contributed to the increase in atmospheric pCO₂ through enhanced ventilation of the deep overturning cell (9, 31). The hypothesis of an early retreat in the extent of Antarctic sea ice is also supported by the warming trend observed in GC528 at 17–15ka (Fig. 2B). However, the retreat of sea ice cover in the South Atlantic might not have had a significant effect on the density difference between the intermediate and deep overturning cells because the mode of deep-water formation did not change, which is controlled by the position of the grounded ice sheet relative to the continental shelf break. Although the geological evidence for the position of the grounding line in the Weddell Sea is inconsistent (38), there is indication that, at least around the Antarctic Peninsula, the grounding line retreat occurred late in the deglaciation (39, 40). Thus, the glacial brine-dominated mode of southern-sourced deep-water formation may have persisted until as late as 10 ka. In summary, we argue that changes in the extent of permanent sea ice may occur earlier than changes in the maximum position of the grounded icesheets, thus it is possible to ventilate the deep ocean without decreasing its density.

Over the deglacial period (17–11 ka), increasing deep-ocean temperatures (Fig. 2B) coupled with a 0.6‰ increase in benthic foraminifer δ¹³C (34) (Fig. 4B) is indicative of a greater proportion of northern-sourced water in the deep South Atlantic. The impinging of warm modified northern-sourced water on the Antarctic continental shelf has been suggested as a possible mechanism (33) which could melt back the grounded ice sheets in the Weddell Sea, freeing shelf space for the formation of ISW, resulting in the observed decrease in deep-ocean density stratification at 10 ka. Whereas a decrease in the density of deep water has been invoked to explain the deglacial rise in atmospheric pCO₂, via associated changes in the rate of diapycnal mixing and the vertical position of the isopycnal separating the two overturning branches of circulation (16, 17, 41), we suggest that its impact is relatively minor compared with the impact of changes in the rate of air–sea gas exchange in the Southern Ocean. However, it should be noted that vertical shifts in the position of the main pycnocline (41) cannot be resolved in this study, thus it remains to be determined whether these changes play a more important role in regulating atmospheric CO₂.

Although density destratification of the South Atlantic does not appear to play a leading role in regulating atmospheric pCO₂, we propose that density destratification may have been instead important in “locking in” the incipient transition to an interglacial climate state. Before the destratification event, the ocean was able to return to its “glacial regime” following a transient perturbation, but not afterward. We suggest that the density destratification of the South Atlantic, initiated by a change in mode of formation of deep water, acted as a “flip switch,” eliminating the ocean’s ability to restock its CO₂ inventory at the expense of the atmosphere, and thus forcing climate to switch to an interglacial state. Longer records of the density gradient within the Southern Ocean are required to test this hypothesis.

Conclusion

This study provides, to our knowledge, the first deglacial record of density changes in the deep and intermediate South Atlantic spanning the last deglaciation. We find that the intermediate ocean was significantly colder than deep waters at the LGM, and this temperature inversion requires that ocean stability is maintained by salinity gradients. We suggest that the physical properties of the glacial South Atlantic were regulated by an increase in Antarctic sea ice extent, which resulted in colder surface waters in the sub-Antarctic, but also led to deep waters being primarily formed through the creation of brines as opposed to supercooling of ISW. Over the deglaciation, intermediate water warmed in response to a retreat in the Antarctic sea ice margin, and the deep South Atlantic started to reflect both a greater proportion of northern-sourced water but also isotopically lighter AABW from the incorporation of Antarctic ice sheet meltwater. A rapid change in the dominant mode of deep-water formation at the onset of the Holocene, from brines to supercooled ISW, likely resulted in the density destratification of the intermediate-deep ocean during the late deglacial period.

Our density records also enable us to address the question of whether deglacial changes in ocean chemistry are driven by a breakdown in the deep-ocean density stratification. We find that the greatest intermediate-deep change in benthic foraminifer δ¹³C and foraminifer/coral ¹⁴C occurs before the density destratification. We suggest that this chemical destratification was driven by an increase in air–sea gas exchange which ventilates the deep overturning cell without affecting its density. The late deglacial breakdown in the density gradient of the South Atlantic occurs at the onset of the Holocene, suggesting deep-ocean density destratification did not play a leading role in driving the deglacial rise in atmospheric pCO₂. The difference in the timing of the breakdown of the intermediate-deep chemical gradient compared with the breakdown of the intermediate-deep physical density gradient suggests that chemical and physical stratification is not as tightly coupled as previously inferred. This also raises the interesting possibility that the density destratification of the South Atlantic, induced by a change in the mode of deep-water formation, could act as the flip switch resulting in the transition to a full interglacial state.

Methods

Materials.

Core GC528 [53.01°S, 58.04°W, 598mbsl] was collected on the cruise JR244 of the RRS James Clark Ross. Located on the Falkland Plateau, this core is situated close to the main inflow of Antarctic Intermediate Water (AAIW) into the Atlantic basin.

Core MD07-3076Q [44° 09.2′S 14° 13.7’W, 3,770mbsl] was retrieved from the eastern flank of the midocean ridge. The age model is based on reservoir-age corrected radiocarbon measurements on monospecific planktonic foraminifera and is described fully in ref. 36.

Age model (GC528).

The age model for GC528 (Supporting Information) was generated using 25 radiocarbon dates of monospecific samples of Uvigerina bifurcata (>125-μm size fraction, 2–6 mg), which were graphitized in the Godwin Laboratory for Paleoclimate Research, University of Cambridge (hereafter, GLPR) using the hydrogen and iron catalyst method (42) and subsequently analyzed at the ¹⁴Chrono Centre at the University of Belfast by accelerator mass spectrometry. Five of the 25 samples were graphitized and analyzed by BetaAnalytic; no interlab offset between samples was found. Carbon-14 ages were calibrated using Bacon age-modeling software (43) with the Marine13 dataset (44). Reservoir age constraints were taken from paired U-Th/¹⁴C ages in corals (12) after 16 ka; before 16 ka a constant reservoir age of 1.36 ± 0.4 ky [the age of the oldest U-Th/¹⁴C dated coral analyzed by Burke and Robinson (12)] was applied downcore.

Sample Preparation.

In GC528, samples of Uvigerina spp. (212–31-µm size fraction) were hand-picked, cleaned using the methodology of ref. 45 and split, with ∼100 μg used for oxygen and carbon isotope and ∼400 μg for trace element geochemistry, and analyzed at the GLPR.

In MD07-3076Q, samples of Uvigerina spp. (212–315-µm size fraction) were hand-picked and ∼3 whole specimens were used for stable isotope analysis, and 10–15 whole specimens were cleaned for Mg/Ca analysis using the methodology of ref. 45.

Oxygen and stable carbon isotopes.

Stable isotopes from GC528 were analyzed using a Multicarb preparation system coupled to a VG SIRA Mass Spectrometer in the GLPR. Measurements of δ¹⁸O and δ¹³C were determined relative to the Vienna Peedee Belemnite standard with an analytical precision of ±0.08‰ for δ¹⁸O and ±0.06‰ for δ¹³C. δ¹⁸O was measured on Uvigerina spp. and δ¹³C was measured on Oridorsalis umbonatus and corrected to equilibrium calcite by +1.0‰ (46). Although O. umbonatus is a shallow infaunal species, the δ¹³C O. umbonatus correlation with coral ¹⁴C trends from intermediate water (12, 35) suggests that there has been no bias in the overprinting of bottom water δ¹³C_DIC by pore waters.

Stable isotopes from MD07-3076Q were analyzed at the Laboratoire des Sciences du Climat et de l'Environnement, Gif sur Yvette, France. The mean external reproducibility of the carbonate standard is ±0.05‰ for δ¹⁸O and ±0.03‰ for δ¹³C. δ¹⁸O was measured on Uvigerina spp. and δ¹³C was measured on Cibicides kullenbergi (34).

Trace metal analysis.

Mg/Ca elemental ratios were determined by inductively coupled plasma–optical emission spectroscopy (47). Long-term instrumental precision of element ratio data, determined by replicate analyses of a standard solution, was ± 0.46%, translating into an uncertainty of 0.06 °C.

Bottom-water carbonate ion concentration has been previously suggested (48) to affect Mg/Ca values in some benthic foraminifera species. Deglacial variation in carbonate ion concentration may exert an effect on Mg/Ca which is unrelated to changes in bottom-water temperature, particularly at the deep site (MD07-3076Q) where carbonate ion concentrations are lower. However, recent studies (e.g., ref. 18) show that Uvigerina spp. Mg/Ca (in particular) is very insensitive to changes in the carbonate ion concentration.

Mg/Ca-Derived Benthic Temperatures.

Mg/Ca values were converted into benthic temperatures using the new calibration curve (Supporting Information):

$$\mathrm{M}\mathrm{g}/\mathrm{C}\mathrm{a}=\left(0.0915\pm 0.005\right)\ast \mathrm{T}+\left(0.843\pm 0.035\right){\mathrm{R}}^2=\mathrm{0.92.}$$

Here, 1σ uncertainty in the temperature estimate of each sample is ±0.7 °C (Supporting Information).

Seawater δ¹⁸O Offset from the Contemporaneous Global Mean Seawater δ¹⁸O (δ_w-ice).

Seawater δ¹⁸O (δ_w) is calculated from the Mg/Ca-derived benthic temperature and the δ¹⁸O of foraminiferal calcite using the linear form of the paleotemperature equation of ref. 20:

$$\mathrm{T}=16.9\mathrm{-}4.0\ast \left({\mathrm{\delta }}_{\mathrm{c}}\mathrm{-}{\mathrm{\delta }}_{\mathrm{w}}\right).$$

This calibration produces good agreement between modern measurements of δ_w and the calculated core top δ_w.

To make a comparison with modern seawater values, the contemporaneous global mean seawater δ¹⁸O, sometimes referred to as “ice volume effect,” was removed assuming a linear relationship (21) between sea level (19) and δ¹⁸O of seawater. Site-specific deviations from the global mean that would result in a synchronous breakdown of the physical and chemical gradients in the South Atlantic require changes in δ_w in MD07-3076Q 5ky earlier than the global mean, which is not supported by regional δ¹⁸O stacks (49). The 1σ uncertainty in δ_w-ice of each sample is ±0.35‰.

In Situ Density of Seawater (σ_θ).

The modern linear relationship between salinity and δ_w-ice for the Southern Ocean (21) is assumed to hold across the deglacial period (Supporting Information). In situ density was calculated from salinity and benthic temperature, using the equation of state (23).

Assumption of Zonal Continuity Across the Sub-Antarctic South Atlantic

Developing an understanding of changes in the physical properties of seawater with depth typically requires depth transect of sediment cores that are located within a confined region, where surface signals should be common to all sites. However, due to a lack of available cores from intermediate water depths in the Eastern Atlantic/Cape Basin combined with low sedimentation rates at those few sites (e.g., refs. 11, 52, 53) and the dearth of deep sites in the Western sub-Antarctic Atlantic, it was necessary to study two rather distal sites (Fig. S1). Despite the geographic separation of these cores, they underlie roughly equivalent surface circumpolar ocean regimes (Sub-Antarctic Zone). The modern sub-Antarctic Atlantic is zonally continuous, in terms of potential density (Fig. S1). In this study we make the assumption that deglacial changes occurring at either site would occur everywhere synchronously in water at the same depth in the sub-Antarctic South Atlantic. We acknowledge that for our findings from the South Atlantic to have global implications, further studies, particularly from the Pacific sector, are required.

Fig. S1.

Density transect of Southern Ocean along the Sub-Antarctic Front (WOA09_Annual). There is minimal zonal difference in density between the eastern and western Atlantic basins. In this study we assume that this zonal continuity remains over the deglaciation, and use the cores to reconstruct a depth transect of the Southern Ocean.

Age Model (GC528)

The age model for GC528 was generated using 25 radiocarbon dates of monospecific samples of U. bifurcata (Methods). The age markers of GC528 are tabulated in Table S1.

Table S1.

Age control points of GC528

Core depth, cm	14C age, y before present	± Error, y	Reservoir age, y	± Error, y	Laboratory
0	2,540	30	967	110	BetaAnalytic
5.5	2,090	30	967	110	BetaAnalytic
30	4,972	45	967	110	Godwin/UB
41	6,178	39	967	110	Godwin/UB
49	7,150	30	967	110	BetaAnalytic
58	9,383	48	967	110	Godwin/UB
61	7,876	51	967	110	Godwin/UB
61	9,026	48	967	110	Godwin/UB
66	11,925	48	987	110	Godwin/UB
72	12,562	55	1,031	110	Godwin/UB
78	12,652	53	1,040	110	Godwin/UB
82	12,295	53	1,077	110	Godwin/UB
88	12,611	47	1,077	110	Godwin/UB
98	13,251	50	1,084	110	Godwin/UB
109.5	12,234	65	1,150	110	Godwin/UB
111.5	13,079	63	1,175	110	Godwin/UB
119	13,670	70	1,298	110	Godwin/UB
124	14,410	81	1,325	110	Godwin/UB
139.5	14,651	80	1,330	110	Godwin/UB
149	16,380	90	1,364	110	BetaAnalytic
199	17,340	101	1,364	110	Godwin/UB
229	17,220	70	1,364	110	BetaAnalytic
289	18,260	108	1,364	110	Godwin/UB
350	20,628	140	1,364	110	Godwin/UB
416	29,475	372	1,364	110	Godwin/UB

Carbon-14 ages were calibrated using Bacon age-modeling software (43) with the Marine13 dataset (44) (Table S1 and Fig. S2). The age model suggests highly variable sedimentation rates in the core across the last deglaciation from ∼10 cm/ky in the Holocene to ∼40 cm/ky at the LGM. The high mean sedimentation rates during the glacial period can be explained by a decrease in sea level which acts to bring the coastline, the source of terrigenous material, closer to the core site.

Fig. S2.

GC528 age model based on 25 radiocarbon dates on the benthic foraminifera U. bifurcata. Output age model generated using the Bayesian age-depth modeling software Bacon (43).

Uvigerina spp. Cleaning

In addition to the Mg/Ca data generated in this study, further Mg/Ca data from MD07-3076Q was used which had been cleaned using a different approach. This additional Uvigerina spp. Mg/Ca data was cleaned using the clay removal and silicate removal steps (45), but without a full oxidative cleaning. Comparison of samples which had been both oxidatively cleaned and nonoxidatively cleaned show an average offset of 0.045 mmol/mol (Fig. S3). We corrected the nonoxidatively cleaned samples by −0.045 mmol/mol; however, this does not significantly change the overall temperature trend (Fig. S3).

Fig. S3.

Comparison of oxidatively cleaned samples versus nonoxidatively cleaned samples. (A) Difference in Mg/Ca ratio between samples from the same depth interval cleaned nonoxidatively and samples cleaned oxidatively. (B) Barplot summarizing the distribution in A. (C) Timeseries showing the difference in the Mg/Ca values of oxidatively cleaned samples (black) compared with nonoxidative samples (blue) and nonoxidative samples corrected by −0.045 mmol/mol (red).

Mn/Ca was measured to monitor cleaning efficiency and diagenetic effects. There is no relationship between Mn/Ca and Mg/Ca (Fig. S4), implying that diagenetic coatings are not affecting the Mg/Ca ratio. Nonoxidatively cleaned samples typically have a higher Mn/Ca ratio than oxidatively cleaned samples; however, the concentration of Mn/Ca is so small (up to 0.3 mmol/mol) that the Mg contribution of a diagenetic coating would have minimal effect. A Mg/Mn ratio of 0.1 mol/mol within a diagenetic coating, consistent with Mg/Mn ratios found in manganese carbonate in marine sediments (54), would imply a maximum contribution of 10⁻² mmol/mol to Mg/Ca in this record, well within the reproducibility found from duplicate analyses.

Fig. S4.

Cross-plot of Mn/Ca versus Mg/Ca for MD07-3076Q. There is no correlation between Mn/Ca and Mg/Ca for either oxidatively cleaned (black) or nonoxidatively cleaned samples (blue) suggesting that diagenetic coatings do not affect the Mg/Ca measured.

Mg/Ca-Temperature Calibration

Comparison of available Uvigerina spp. Mg/Ca-temperature calibrations in the literature shows there is a considerable range in the regression lines (Fig. S5). We find that the best-fit calibration curve to the core top data from GC528 and MD07-3076Q is provided by the core-top calibration study of Elderfield et al. (18). However, we find that this calibration generates temperature estimates below the freezing point of seawater for GC528 at the LGM, the minimum temperature being −3.8 °C. This may be in part caused by the lack of Mg/Ca core-top constraints for bottom water below 0 °C. We attempt to improve the calibration by adding the constraint that the LGM minimum Mg/Ca value in GC528 measured cannot record temperature below the freezing point of seawater (Fig. S5). The new calibration curve is defined as

$$\text{Mg}/\text{Ca}=\left(0\text{.}0915\pm 0\text{.}005\right)\text{* T}+\left(0\text{.}843\pm 0\text{.}035\right){\text{R}}^2=0\text{.}92\text{.}$$

The 1σ uncertainty in the temperature estimate of each sample is ±0.7 °C. Discussion of the propagation of error in benthic temperatures can be found below.

Fig. S5.

Comparison of published Uvigerina spp. calibration curves. Comparison of various Uvigerina spp. Mg/Ca-temperature calibration curves: Bryan and Marchitto (56), dot-dashed line; Yu and Elderfield (57), solid line; Elderfield et al. (18) core-top calibration, dashed line; Elderfield et al. (18) pore-water calibration, dotted line. The low-temperature core-top data from Elderfield et al. (18) are displayed as black diamonds. The Mg/Ca data from the top of the two cores are plotted as green (GC528) and brown (MD07-3076Q) boxes. Calibration of Mg/Ca-benthic temperatures based on data from Elderfield et al. (18) has been improved with an additional constraint that the minimum Mg/Ca value (green bar on the y axis) cannot generate a benthic temperature below the freezing point of seawater. This amended calibration is shown as a bold gray line.

However, it is important to note that the choice of calibration line does not alter our main observation of a decrease in the density gradient over the deglacial period (Fig. S6).

Fig. S6.

Comparison of in situ density of GC528 and MD07-3076Q based on different calibration curves. Comparison of the in situ density of generated using various Uvigerina spp. Mg/Ca-temperature calibration curves: Bryan and Marchitto (56), dot-dashed line; Yu and Elderfield (57), solid line; Elderfield et al. (18) core-top calibration, dashed line; Elderfield et al. (18) pore-water calibration, dotted line. Mg/Ca-temperature calibration used in this study shown by the green and brown lines.

Subtracting the Global Ice Volume Effect

To calculate salinity, the isotopic effect of melting continental ice must be subtracted from the δ¹⁸O of seawater (as calculated from the paleotemperature equation) to apply modern salinity–δw relationships. The ice volume effect was calculated assuming a linear relationship (21) between sea level (19) and δ¹⁸O of seawater. The isotopic effect of melting ice will have spatial and temporal variations which are masked in our subtraction of a global mean δ_w, but these remain poorly constrained. As transient δ¹⁸O tracer models that provide information on local δ¹⁸O variability and their governing mechanisms are not available, we adhere to the global mean isotopic δ¹⁸O record as the best approximation for ice sheet melt derived changes in ocean δ_w. The effect of subtracting the global ice volume effect from the calculated δ¹⁸O of seawater is shown in Fig. S7.

Fig. S7.

Comparison of seawater δ¹⁸O before (dashed) and after (solid) ice volume correction.

Choice of Regression Line Between Salinity and δ_w-ice

In this study we use the modern Southern Ocean linear relationship between salinity and δ_w-ice (21)and assume that it holds across the deglacial period. We argue that processes such as brine rejection may have the effect of increasing salinity without increasing δ_w-ice, thus the salinity estimates should be considered minimum salinity estimates.

An alternative regression line may be based on the LGM salinity and seawater δ¹⁸O estimates of Adkins et al. (15) from the two Southern Ocean sites, 1123 and 1093. Assuming 1‰ of the LGM seawater δ¹⁸O is related to the ice volume, the regression line between salinity and δ_w-ice based on the pore-water studies is

$${\mathrm{\delta }}_{\text{w}‐\text{ice}}=0\text{.}191\text{* S}-6\text{.}913\text{.}$$

The difference between the salinity estimates produced by this “LGM regression line” and the “modern Southern Ocean regression line” used in the paper is small relative to the magnitude of the change in salinity over the deglaciation (Fig. S8). We have chosen the modern Southern Ocean regression line over the LGM regression line based on the fact that it has more data making up the regression and the data included in the regression are located within the region of MD07-3076Q and GC528.

Fig. S8.

Comparison of salinity estimates based on different salinity–δ_w-ice regression lines. “Modern regression” is based on the modern salinity–δ_w-ice relationship for the Southern Ocean (21). “LGM regression” uses LGM pore-water salinity and δ_w-ice estimates from site 1123 and 1093 (15).

In this study we make the assumption that the relationship between salinity and δ_w-ice does not significantly differ between the Holocene and the LGM. There are a couple of lines of evidence which can be used to justify this assumption. First, we have performed a simple thought experiment whereby we pose the question: What must the salinity–δ_w-ice relationship have to be at the LGM to eradicate the late deglacial decrease in salinity? We find that the required salinity–δ_w-ice relationship plots well outside the range of modern southern hemisphere water masses and well outside the predicted LGM salinity–δ_w-ice relationship from pore-water profiles (15) (Fig. S9), provided the reported pore-water–derived LGM salinity and δ_w estimates are robust (24). We therefore suggest that it is valid to invoke significant seawater salinity variations to explain the observed δ_w-ice change in MD07-3076Q. A second line of evidence supporting the idea that a change in the salinity–δ_w-ice relationship did not occur over the last deglaciation comes from newly published isotope-enabled GCM studies (22). This study found that the South Atlantic exhibits a relatively constant salinity–δ_w-ice relationship across spatial and temporal scales supporting the validity of applying a temporally constant salinity–δ_w-ice relationship to our datasets.

Fig. S9.

Thought experiment investigating the range of deglacial salinity changes possible due to a change in the salinity–δ_w-ice relationship. The modern and LGM data are shown for MD07-3076Q (red and blue filled diamonds) overlain onto the salinity–δ_w-ice relationships for modern water masses (21), AABW, CDW, and NADW. The Holocene-LGM difference in salinity is shown for three experiments, where (i) there is no deglacial change in the salinity–δ_w-ice regression; (ii) there is a shift in the salinity–δ_w-ice regression from a salinity–δ_w-ice regression as estimated by Southern Ocean pore-water profiles (15) to the modern Southern Ocean salinity–δ_w-ice regression (21); and (iii) when all of the salinity change in the deep ocean can be explained by a change in the salinity–δ_w-ice regression.

Propagation of Errors in the Conversion of Mg/Ca to Benthic Temperature

Propagating the replicate error (σ_Mg/Ca = 0.7 °C) and the error in the new calibration curve (σ_calib = 1.1 °C) gives a 1σ uncertainty in the temperature estimate of each sample of ±1.3 °C. Previously published calibrations typically give 1σ uncertainties of 0.5–1.0 °C (e.g., ref. 55). This difference may be due to a less critical assessment of the errors involved in the calibration curve.

However, the errors associated with the calibration curve (σ_calib) are largely due to it being based on old data from different laboratories, with different cleaning procedures. It is likely that improvements in analytical technique have decreased the error associated with the calibration curve. The data used in this study were generated in the same laboratory, using the same method and have been checked for repeatability. We infer that although the absolute values for temperature may be uncertain, the raw Mg/Ca values generated in each core should be directly comparable. Therefore, we assume an error in benthic temperatures of 0.7 °C, equivalent to the replicate error in the Mg/Ca measurement.

Propagation of Errors in the Calculation of Potential Density

To calculate in situ density at each site, salinity had to be calculated, which was derived from the ice-volume corrected δ¹⁸O of seawater (δ_w-ice).

Using the paleotemperature equation (20), the δ¹⁸O of seawater (δ_w) is calculated. The error in the δ¹⁸O of seawater (${\sigma }_{\delta w}$) is a combination of the error in the measurement of δ¹⁸O of Uvigerina spp. (${\sigma }_{\delta c}$ = 0.08‰) and the error in the Mg/Ca-derived benthic temperature (σ_T = 0.18‰): ${{\sigma }_{\delta w}}^2=\hspace{0.17em}\sqrt{{{\sigma }_T}^2+{{\sigma }_{\delta c}}^2}$, consequently ${\sigma }_{\delta w}$ = 0.19‰.

To compare paleoestimates of ${\sigma }_{\delta w}$ with modern data, the so-called “ice volume effect” is removed. Using a compilation of observed sea-level estimates (19), we calculate the isotopic effect from the melting of continental ice and estimate a variance of ${\sigma }_{SL}=$ 0.10‰. This gives a total error in the ice-volume–corrected δ¹⁸O of seawater of ${\sigma }_{\delta w\text{-}ice}=\sqrt{{{\sigma }_{\delta w}}^2+{{\sigma }_{SL}}^2}=$ 0.22‰.

We assume a linear relationship between ${\sigma }_{\delta w\text{-}ice}$ and salinity, based on the Southern Ocean gridded δ¹⁸O seawater data set (21). The variance in salinity ${\sigma }_S$ is calculated using the ${\sigma }_{slope}$ and ${\sigma }_{icept}$ given in ref. 21:

$${\mathrm{\sigma }}_{\text{S}}=\sqrt{{\mathrm{\sigma }}_{\text{δw}}^2{\left(\frac{\partial \text{δw}}{\partial \text{S}}\right)}^2+{\mathrm{\sigma }}_{\text{icept}}^2{\left(\frac{\partial \text{δicept}}{\partial \text{S}}\right)}^2+{\hspace{0.17em}\mathrm{\sigma }}_{\text{slope}}^2{\left(\frac{\partial \text{slope}}{\partial \text{S}}\right)}^2}=2.24\mathrm{psu}\text{.}$$ [S1]

In situ density was calculated using the equation of state expressed in ref. 23. Variance was calculated similarly to Eq. S1, returning a variance in in situ density of ${\sigma }_{\theta }$ = 1.85 kg/m³.