The residence time of Southern Ocean surface waters and the 100,000-year ice age cycle

See allHide authors and affiliations

Science  08 Mar 2019:
Vol. 363, Issue 6431, pp. 1080-1084
DOI: 10.1126/science.aat7067

Resetting the glacial timer

The periodicity of glacial cycles changed from 100,000 to 41,000 years during the middle of the Pleistocene epoch. Why? Hasenfratz et al. measured the oxygen isotope composition and magnesium/calcium ratio in benthic and planktonic foraminifera from the Antarctic in order to reconstruct changes in the rate of transfer of ocean water from the depths to the surface over the past 1.5 million years (see the Perspective by Menviel). The emergence of the 100,000-year cycle coincided with a reduction in deep-water supply and a freshening of the surface ocean. This slowing may have caused more prolonged ice ages by making the Antarctic less responsive to orbitally paced drivers of carbon dioxide release.

Science, this issue p. 1080; see also p. 1040


From 1.25 million to 700,000 years ago, the ice age cycle deepened and lengthened from 41,000- to 100,000-year periodicity, a transition that remains unexplained. Using surface- and bottom-dwelling foraminifera from the Antarctic Zone of the Southern Ocean to reconstruct the deep-to-surface supply of water during the ice ages of the past 1.5 million years, we found that a reduction in deep water supply and a concomitant freshening of the surface ocean coincided with the emergence of the high-amplitude 100,000-year glacial cycle. We propose that this slowing of deep-to-surface circulation (i.e., a longer residence time for Antarctic surface waters) prolonged ice ages by allowing the Antarctic halocline to strengthen, which increased the resistance of the Antarctic upper water column to orbitally paced drivers of carbon dioxide release.

The mid-Pleistocene transition (MPT) occurred in the absence of any discernible changes in the orbital parameters that control the seasonality and meridional distribution of incoming solar radiation (1), driving decades of research into the possible contributors to the change (27). Evidence is growing that the Southern Ocean was critical to the global cooling of the late Pleistocene ice ages, enhancing ocean carbon storage, lowering atmospheric CO2 (pCO2) concentrations, and thus weakening the global greenhouse effect (810). The available data are consistent with an increase in the amplitude of glacial/interglacial changes in pCO2 since the MPT, largely through a deepening of the CO2 minima reached during ice ages (1012). Thus, the evolution of glacial/interglacial cycles since 1.25 million years ago may have reflected the progress of glacial-stage conditions of the Southern Ocean toward those observed in the glacial maxima of the late Pleistocene.

In the modern Southern Ocean, nutrient- and CO2-rich deep waters ascend to the surface, where the scarcity of light and iron impedes the consumption of major nutrients by phytoplankton, allowing for the evasion of previously sequestered carbon. In the Antarctic Zone (AZ) south of the Antarctic Polar Front, the evasion of CO2 could have been inhibited during ice ages by increased sea ice cover (13) or by a reduction in the exchange of water between the Antarctic surface and the deep ocean (1416). The relative importance of these two alternative causes of CO2 decline have not been resolved for the late Pleistocene ice ages, let alone the MPT. Data on the evolution of AZ conditions across the MPT may offer new insight.

In this study, through measurements on the sediments from Ocean Drilling Project (ODP) Site 1094 (53.2°S, 05.1°E, 2807 m water depth) (fig. S1), we sought to determine the history of the relationship between surface and deep waters across the ice ages of the MPT. ODP 1094 is currently bathed by Lower Circumpolar Deep Water (LCDW). Above, Upper Circumpolar Deep Water (UCDW) upwells to the surface, and vertical mixing between UCDW and surface waters also occurs. However, the surface waters are lower in salinity, and the resulting gradient in salinity (the halocline), and thus density, moderates vertical exchange (17). Biogenic opal and barium accumulation rates indicate that export production at the site reached minima in each of the glacial stages back to pre-MPT (9, 10), indicating that ODP 1094 remained within the AZ throughout the ice ages of the past 1.5 million years. Consistent with this notion, sea surface temperatures (SSTs) reconstructed for the glacials throughout the past 1.5 million years were rarely >1.5°C higher than that of the Holocene and the other interglacials (see below), when the Antarctic Polar Front was north of the study site—a conclusion further supported by diatom assemblages (18).

In benthic and planktonic foraminiferal tests from ODP 1094, we have measured the calcite δ18Ο (δ18ΟC) and the Mg/Ca ratio, the latter serving as a proxy for seawater temperature (5, 19). Seawater δ18Ο (δ18ΟW) was determined by removing the temperature effect on δ18ΟC using Mg/Ca-derived temperature estimates (5). Although δ18Ο is a traditional tool for reconstructing regional temperature and salinity as well as global ice volume, we focus here on its utility as a passive tracer in the upper water column of the AZ in order to infer circulation rates. The AZ upper water column δ18ΟW is lowered by the excess of precipitation relative to evaporation, with an additional reduction due to melting of Antarctic land ice, ice shelves, and/or icebergs. The amplitude of this δ18Ο lowering depends not only on the rates of these processes but also on the residence time of water in the upper water column of the AZ. Enhanced input of deep water into the AZ surface, in the absence of other changes, would lower the residence time and thus raise the δ18ΟW of the AZ surface, which would be reflected in the δ18Ο of planktonic foraminifera. Conversely, reduced input of deep waters would increase the residence time, allowing the AZ surface and deep δ18ΟW to diverge. Following this logic, in the absence of major changes in precipitation or glacial melt rate, coupled planktonic and benthic δ18ΟC measurements in the AZ, corrected to yield changes in δ18ΟW, can record past changes in the residence time of water in the AZ surface and thus the rate of deep-to-surface circulation in the region. As a result of the multi-year residence time of water in the surface AZ and the lack of major effects from the formation and melting of sea ice (20), modern δ18ΟW varies little across the open AZ [by <0.3 per mil (‰)] (21, 22). This mitigates concerns regarding the use of a single site to infer changes in the AZ as a whole.

The planktonic [Neogloboquadrina pachyderma (sinistral)] and benthic (Cibicidoides spp., Melonis pompilioides) δ18ΟC records are highly correlated (Fig. 1A), which is consistent with a substantial contribution from whole-ocean δ18ΟW change and/or the possibility that deep waters are responsive to climate evolution at high southern latitudes (5). To constrain the temperature component in the foraminiferal δ18ΟC records, we systematically corrected the Mg/Ca records of both N. pachyderma (s.) and M. pompilioides for Mn contamination (23) and converted these to temperatures (figs. S2 to S4). [Cibicidoides was not used for Mg/Ca paleothermometry because of a carbonate saturation effect on this taxon (24).] Our new bottom-water records, although discontinuous in the late Pleistocene (due to a scarcity of shells), provide a constraint on the Pleistocene evolution of temperature and δ18ΟW of the deep Southern Ocean south of the Antarctic Polar Front.

Fig. 1 Antarctic Southern Ocean surface and bottom water records.

Surface- and bottom-water data are shown in red and black, respectively. (A) Foraminiferal δ18ΟC of planktonic N. pachyderma (s.) [(45) and this study] and of benthic Cibicidoides spp. and M. pompilioides (composite record) from ODP Site 1094. (B) ODP 1094 SST derived from N. pachyderma (s.) Mg/Ca compared to Antarctic temperature (25) (blue). (C) ODP 1094 BWT derived from M. pompilioides Mg/Ca compared to ODP Site 1123 (5). (D) δ18ΟW of surface and bottom waters at ODP Site 1094. The uncertainty envelopes include the uncertainties from Mn correction and calibration (1σ SD). The stars represent the modern summer temperature (46) and the δ18ΟW at ODP 1094 (47, 48) (fig. S10) and ODP 1123 (21) for the surface and deep water, respectively. VPDB, Vienna PeeDee belemnite; SMOW, standard mean ocean water.

The SST record compares well with ice core–derived Antarctic temperatures (25) during the recent glacial cycles (Fig. 1B). Prior to ~600,000 years ago, however, the two records display distinct patterns, with the glacial SSTs gradually decreasing from ~2°C to near-freezing while Antarctic temperatures remained relatively stable. For almost all glacials after the MPT as well as some that preceded the MPT, reconstructed bottom-water temperatures (BWTs) are close to the freezing point of seawater (Fig. 1C). Deep-water temperatures in the Southwest Pacific (5) are similar but somewhat warmer during the past few hundred thousand years, consistent with the modern BWT difference between the two sites.

In contrast to the glacial maxima, the Mg/Ca records within the deglacials and interglacials bear larger uncertainties owing to corrections required for Mg associated with Mn phases coated on foraminifera (fig. S2). Hence, this study focuses on the peak glacials. To aid comparison between the glacial evolution of surface and deep waters, we averaged all the peak glacials on the basis of the LR04 benthic δ18ΟC stack (23, 26) (Fig. 2 and figs. S5 to S8). Starting at MIS 16 (~700,000 years ago) or perhaps slightly earlier, the glacial temperature and δ18ΟW data indicate contraction of the vertical temperature gradient and an expansion of the vertical δ18ΟW gradient (Fig. 3, C and D).

Fig. 2 Glacial Antarctic Southern Ocean surface and bottom water records from ODP Site 1094.

(A) Planktonic and benthic foraminiferal δ18Ο. (B) Surface- and bottom-water temperature. (C) Surface and bottom water δ18Ο. Bottom-water δ18Ο for the LGM (barren of benthic foraminifera) was estimated using the δ18Ο of Cibicidoides spp. and assuming BWT of –1.5° ± 0.5°C. The green symbol represents the pore water δ18Ο at the LGM at ODP Site 1093 (49) (fig. S1). Error bars are 1σ SD.

Fig. 3 Paleoceanographic changes in the Southern Ocean across the MPT.

(A) Atmospheric CO2 records (1012, 41). The dashed line around the CO2 data from (12) indicates the age uncertainty. (B) Benthic δ18ΟC stack (26) and normalized power spectral density of the 100,000-year cycle in the data calculated using 500,000-year windows and ½ lags. (C and D) Surface-to-deep vertical gradients in glacial temperature and seawater δ18Ο at ODP Site 1094 (see Fig. 2, B and C). (E) Iron flux to Subantarctic ODP Site 1090 (8) (fig. S1).

Under globally colder conditions, most climate models simulate a decline in precipitation in the AZ (27). Similarly, with reduced snowfall on Antarctica under a colder climate (28), coastal discharge of meltwater and icebergs into the AZ may have declined. Because these reductions in the freshwater input would act to increase the δ18ΟW of the AZ surface, the observed decline in δ18ΟW in the surface relative to the deep water must result from an increase in the residence time of surface waters during the ice ages since the MPT. Given the dominance of the underlying deep ocean as a source of new water to the AZ surface, the δ18ΟW decline thus indicates a reduction in the rate of deep-water inflow into the AZ surface mixed layer. This reduction might have been caused by a reduction in wind-driven upwelling and/or turbulent or convective mixing across the base of the AZ surface mixed layer. Assuming an ice age decrease of 10% in the precipitation excess and meltwater discharge to the AZ surface ocean south of 50°S (27) and accounting for the expected temperature-related δ18Ο decline of precipitation over the AZ and the meltwater discharged from Antarctica during the glacials, the reconstructed vertical δ18ΟW gradient since the MPT indicates more than a halving of the supply of the water from the subsurface (200 to 500 m) relative to modern rates (Fig. 4). Thus, our results provide proxy evidence for a substantial reduction in the supply of subsurface water to the AZ surface during the late Pleistocene ice ages, and also indicate that this condition arose only at the end of the MPT. Today, the input of subsurface water warms the AZ surface. Thus, the cooling of glacial-age AZ surface waters over the MPT (Figs. 1B and 2B) may have been a consequence of the reconstructed decline in this input (29).

Fig. 4 Reconstructed decreases in the subsurface water supply to, and the salinity of, the Antarctic Zone surface during the ice ages since the mid-Pleistocene transition.

(A) Rates of net precipitation (Fp), meltwater discharge from the continent (Fm), and the supply of subsurface (>200 m depth) water to the AZ surface (Fsubsurf.), as well as their δ18Ο and salinity (S), for the modern ice age (MIS 6). The subscripts p and m refer to precipitation and meltwater. (B) Same as (A) for the post-MPT ice age (MIS 16). The lateral and downward fluxes out of the AZ surface are indicated by dashed arrows and must together match the supply of subsurface water, meltwater discharge, and precipitation to the surface. Regular typeface indicates observations, either for the modern or for glacials based on our data; bold typeface indicates calculated values using a simple mass balance approach (23). This approach neglects changes in meridional transport and mixing, and does not take into account changes in continental ice volume because they affect planktonic and benthic δ18Ο equally (23). δ18Οp was estimated using modern and glacial air temperature estimates at our study region (23). The asterisks in (B) indicate that δ18ΟW,subsurf. and Ssubsurf. are artificially held at their modern values. δ18ΟW,surface and Ssurface are calculated from these modern values and the observed surface-to-deep differences. In this way, the only properties indicated to change between (A) and (B) are those involving the surface AZ. The δ18ΟW,surface value in (B) is based on the average δ18ΟW surface-to-bottom water gradient observed for the post-MPT glacials (–0.48‰), which is then corrected for the modern δ18ΟW difference between the shallow subsurface ocean (200 to 500 m depth) and the δ18ΟW of the bottom water at the 2807-m-deep core site (a δ18ΟW difference of 0.17‰). The subsurface-to-deep δ18ΟW gradient is assumed to be constant, which is judged to be the greatest weakness in this calculation (50). The modern δ18ΟW values are from measurements obtained close to ODP 1094 (47, 48) (fig. S10), the modern δ18Οm value is based on δ18Ο measurements of surface snow close to the coast of the Antarctic continent (23, 48), the modern salinity values are from (51), and Fp and Fm are from (31). For the glacials, the net precipitation and the discharge of meltwater are decreased by 10%, reflecting the results of model simulations (27).

In glacial stages prior to MIS 16, no δ18ΟW lowering of the surface relative to the subsurface is discernible, and the opposite is observed for several glacials, with a slightly lower δ18ΟW reconstructed for deep water (Fig. 3D). This is in contrast to the expected signal from glacial-stage reductions in surface/deep water exchange inferred from export production records (9, 10). During glacials prior to MIS 16, warmer AZ SST may have quickly melted calving icebergs near the coast, lowering the δ18ΟW of coastal surface waters and the deep water that they form. This would have lowered the δ18Οw of deep water forming near the Antarctic margin more than it lowered the δ18Οw of open AZ surface waters. In the Holocene, melting is also concentrated along the coasts, and it does influence the δ18ΟW of newly formed deep water (30). However, southern-sourced deep water has dominated the deep Antarctic only during the glacials of the past 900,000 years (7), such that the effects of these surface AZ processes on bottom-water δ18ΟW at ODP 1094 would have been stronger during the glacials than during the interglacials.

δ18ΟW is mechanistically related to salinity (21): As with meltwater from the Antarctic ice sheet, an excess in precipitation relative to evaporation over the AZ lowers both δ18ΟW and salinity, while the upwelling subsurface water increases both quantities. Therefore, the observed surface δ18ΟW decline at ODP 1094 during the past 700,000 years likely also signals a salinity decline. Although Antarctic ice has a very low δ18ΟW, the rate of glacial meltwater addition is currently only 10 to 20% that of excess precipitation (23, 31) and was likely even lower during glacial intervals (28). Neglecting it in our calculation, the decreased δ18ΟW in the surface ocean during the ice ages of the late Pleistocene would have corresponded to a salinity decrease of around 0.6 practical salinity units (psu) relative to the underlying subsurface ocean, suggesting a doubling of the halocline strength compared to today (23) (Fig. 4).

Sea ice formation, the loss of brine from the surface to the ocean interior, and the subsequent melting of the sea ice at the AZ surface also lower the salinity of the AZ surface mixed layer, especially in the open ocean (32). Because its effects on δ18ΟW are negligible, the salinity of the AZ surface during glacials would have been even lower than calculated from δ18ΟW. Thus, although the multiple processes at work in the AZ currently preclude a quantitative reconstruction of salinity based on δ18ΟW, the data reported here provide robust evidence for a strengthening of the AZ halocline during the post-MPT ice ages. A strengthening of the AZ halocline would have been promoted by the reconstructed increase in surface water residence time, which itself would have resulted from a decrease in the subsurface-to-surface water supply rate (33). Alternatively, or in addition, the halocline strengthening may have caused the decrease in subsurface water supply through an increase in the vertical density gradient and thus the resistance to surface/subsurface water exchange (34). Indeed, our reconstructions of temperature and salinity together indicate stronger density stratification during the peak glacials of the past 700,000 years (Fig. 3). Thus, a positive feedback between reduced flow of deep water into the AZ surface and its freshening may have been a central dynamic in the glacial ocean.

For at least the two most recent ice ages, diatom- and coral-bound nitrogen isotope data indicate that low subsurface-to-surface supply of nitrate (and thus also water) was associated with more complete nitrate consumption in the surface AZ waters (3537), signaling an increase in the efficiency of the biological carbon pump in the region and thus reduced AZ CO2 leakage (14). Moreover, the reduction in subsurface-to-surface water supply since MIS 16 may have involved a complementary decrease in the flow of AZ surface waters into the deep ocean, which represents an additional mechanism for diminishing AZ CO2 leakage (14, 38, 39). The initial drivers of the reconstructed reduction in subsurface-to-surface flow in the AZ did not necessarily involve a complementary decline in surface-to-subsurface water flow (i.e., in deep ocean ventilation). Nonetheless, the increase in the residence time of AZ surface waters would have contributed to the strengthening of the AZ halocline, which may have had the eventual consequence of slowing deep ocean ventilation (33, 40).

Changes in AZ overturning and their CO2 effects have been proposed to contribute to the orbitally paced glacial cycles throughout the Pleistocene (9, 10). By 800,000 years ago, atmospheric pCO2 had reached an ice age minimum value of ~180 parts per million (ppm) that has been roughly maintained over the late Pleistocene (41) (Fig. 3A). In contrast, prior to the MPT, ice age pCO2 minima are estimated to be consistently greater than 200 ppm (10). An increase in Fe-bearing dust supply to the Subantarctic Ocean represents one possible explanation for this downward shift (810). Much of the intensification of the glacial temperature and δ18ΟW gradients reconstructed here occurred from ~700,000 years ago onward, late in the MPT (Fig. 3, C and D). Thus, at least this portion of the increase in AZ surface residence time occurred without concomitant declines in the ice age minima in pCO2.

However, the increase in AZ surface residence time at MIS 16 coincides with an abrupt increase in benthic δ18ΟC, which has been suggested to reflect an increase in global continental ice volume (3, 26, 42) (Fig. 3B). Given that deep water temperatures were close to the freezing point during the ice ages within the past 1 million years (5, 43) (Fig. 3B), the δ18ΟC increase observed in our record (Fig. 1A) points to a substantial increase in ice volume (3). The latter is corroborated by the first occurrence of Heinrich stadials in the North Atlantic, 640,000 years ago, indicating that the volume of the Laurentide Ice Sheet possibly exceeded a critical threshold (42, 44). This contrasts with the previously proposed notion of relatively stable glacial ice volume after the so-called 900,000-year event (5). The lack of any δ18ΟW increase in the Southwest Pacific record at MIS 16 may have been due to local hydrographic changes (fig. S9).

The ice volume increase at MIS 16 coincides with the establishment of the dominant high-amplitude quasi–100,000-year glacial cycles as revealed by spectral analysis of the benthic δ18ΟC stack (3, 26) (Fig. 3B). Thus, the rise in ice volume may have been the result of the lengthening of the glacial intervals, which allowed for longer periods of uninterrupted ice sheet growth. We propose that the lengthening of ice ages, in turn, was promoted by the increased residence time of Antarctic surface waters reconstructed here. With a low glacial baseline rate of subsurface water supply to the AZ surface, which allowed the halocline to strengthen, and the positive feedback between the two, it may have been more difficult than previously for orbitally paced changes to enhance AZ surface-to-deep water exchange to a rate that would have caused CO2 venting in the AZ surface. Consistent with this view, model experiments indicate that, relative to interglacial conditions, as increased nutrient consumption and/or reduced deep ocean ventilation in the AZ together become more extreme, the pCO2 declines become progressively smaller, with the system leveling off to a pCO2 of ~175 ppm (39). Thus, AZ changes at the end of the MPT may have been secondary in lowering pCO2 but central in preventing subsequent pCO2 rises, ensuring that glacial conditions persisted despite orbital changes to the contrary. In this view, the reconstructed change in glacial AZ conditions at the end of the MPT may have allowed Northern Hemisphere ice sheets to survive periods of orbitally paced summer insolation maxima on a more regular basis, thereby increasing the longevity of ice ages.

Supplementary Materials

Materials and Methods

Figs. S1 to S15

References (5282)

References and Notes

  1. See supplementary materials.
Acknowledgments: We thank U. Röhl, V. Lukies, and S. Steffer for technical support during the x-ray fluorescence analysis at MARUM; S. Bishop and M. Jaggi (ETH Zürich) and I. Mather (Cambridge) for technical assistance with the isotopic analysis; C. Archer, S. Westermann (ETH Zürich), and M. Greaves (Cambridge) for technical support with the trace metal analysis; T. Gordijn for helping with the sample preparation; and J. Gottschalk and J. Roberts for helpful discussions. Funding: Supported by ETH Research Grant ETH-04 11-1 (A.P.H.); Swiss National Science Foundation grants PP00P2_144811 and PP00P2_172915 (S.L.J.), PZ00P2_141424 (A.M.-G.), and 175162 (F.A.H.); the Max Planck Society (A.M.-G. and G.H.H.); and NSF grant OPP-1401489 and ExxonMobil through Princeton University’s Andlinger Center for Energy and the Environment (D.M.S.). Modern oxygen isotope sample collection by F.A.H. was supported by the BNP Paribas Foundation and ACE, which was a scientific expedition carried out under the auspices of the Swiss Polar Institute, supported by funding from the ACE Foundation and Ferring Pharmaceuticals. This research used sediment samples provided by the ODP, which is supported by NSF and participating countries under management of Joint Oceanographic Institutions (JOI) Inc. Author contributions: A.P.H., S.L.J., A.M.-G., D.A.H., and G.H.H. designed the study; D.A.H. and H.F.K. provided washed and sieved sediment samples from 0 to 121 meter composite depth and produced planktonic foraminiferal stable isotope data within this depth range; A.P.H. washed and sieved the sediments, prepared the samples, performed most of the analysis, and wrote the first version of the manuscript with assistance from S.L.J. and A.M.-G. throughout the entire study and from D.M.S. during the writing; F.A.H. compiled modern seawater δ18Ο data for the Southern Ocean and helped with their interpretation in context of this study; S.M.B., D.A.H., and D.V. were involved in the isotopic and elemental analysis, respectively; and all authors provided substantial input to the final manuscript. Data and materials availability: Data not previously reported are archived at the Bern Open Repository and Information System (BORIS) (

Stay Connected to Science

Navigate This Article