The Role of Carbon Dioxide During the Onset of Antarctic Glaciation

See allHide authors and affiliations

Science  02 Dec 2011:
Vol. 334, Issue 6060, pp. 1261-1264
DOI: 10.1126/science.1203909


Earth’s modern climate, characterized by polar ice sheets and large equator-to-pole temperature gradients, is rooted in environmental changes that promoted Antarctic glaciation ~33.7 million years ago. Onset of Antarctic glaciation reflects a critical tipping point for Earth’s climate and provides a framework for investigating the role of atmospheric carbon dioxide (CO2) during major climatic change. Previously published records of alkenone-based CO2 from high- and low-latitude ocean localities suggested that CO2 increased during glaciation, in contradiction to theory. Here, we further investigate alkenone records and demonstrate that Antarctic and subantarctic data overestimate atmospheric CO2 levels, biasing long-term trends. Our results show that CO2 declined before and during Antarctic glaciation and support a substantial CO2 decrease as the primary agent forcing Antarctic glaciation, consistent with model-derived CO2 thresholds.

The onset of Antarctic glaciation ~33.7 million years ago is associated with an ~+1.5 per mil (‰) shift in deep-sea benthic foraminiferal oxygen isotopic (δ18O) values (1), an abrupt appearance of glaciomarine sediments around the Antarctic margin (2, 3), a shift in clay mineralogy at Southern Ocean sites marking an increase in physical over chemical weathering (4, 5), a permanent deepening of the calcite compensation depth (1), and substantial high-latitude and global cooling (6, 7). Proposed causes for the abrupt onset of glaciation near the Eocene-Oligocene (E-O) boundary include a CO2 decrease across a critical CO2-induced temperature threshold (8), decreasing CO2 combined with an orbital configuration that reduced polar seasonality (1, 8), and changes in regional ocean circulation and heat transport to the southern high latitudes resulting from the progressive tectonic separation of Australia and South America from Antarctica (9).

Published CO2 reconstructions provide contradictory evidence for the role of CO2 during high-latitude cooling and Antarctic ice sheet expansion, with alkenone-based CO2 estimates showing a large CO2 decline occurring 2 to 3 million years after maximum ice sheet expansion (10). Curiously, alkenone-based CO2 estimates show an increase just before and across the E-O transition (10). In contrast, CO2 records based on boron isotope–pH estimates from one ocean locality provide evidence for declining CO2 levels coeval with global cooling, followed by a rebound and subsequent CO2 decline immediately after the E-O climate event (11). Thus, existing alkenone-based CO2 records appear to conflict with recently published boron-based CO2 estimates, as well as our theoretical understanding of the relation between climate change and greenhouse gas forcing.

The published Cenozoic alkenone-based CO2 record is a composite of alkenone carbon-isotopic values from both oligotrophic and eutrophic environments in the modern ocean (10). Alkenone stable carbon-isotope records from individual localities spanning the E-O climate transition are lacking and do not sample the E-O climate event in detail. Instead, published CO2 trends for the E-O interval were inferred from analyses from Deep Sea Drilling Project (DSDP) sites 511 and 612 for the interval preceding the E-O climate transition and from DSDP sites 511, 513, and 516 and Ocean Drilling Program (ODP) site 803 for the interval after the event [supporting online material (SOM)]. Three data points from site 513, two values from site 511, and one data point from site 516 immediately follow the E-O climate transition (10). Earliest Oligocene CO2 estimates responsible for the apparent rise in CO2 across the E-O climate event derive from sites 511 and 513, which are presently located in poorly stratified, nutrient-rich subantarctic waters, raising the possibility that these sites do not reflect air-sea CO2 equilibria (SOM).

For this study, we evaluated regional differences in CO2 estimates and trends across the E-O climate transition by using the carbon-isotopic (δ13C) values of alkenones from six globally dispersed ocean localities, including DSDP or ODP sites 277, 336, 511, 925, 929, and 1090 (Fig. 1), representing a range of oceanographic conditions and algal-growth environments. Chronologies and sea surface temperatures (SSTs) for each site have been previously established (6) or updated for this study (SOM).

Fig. 1

DSDP and ODP sites used in this study overlain on model-derived surface ocean phosphate concentration anomaly. Sites are rotated back to their late Eocene paleopositions. (Left) Contours indicate anomalies of the UVic climate–ecosystem–carbon cycle model-predicted surface ocean phosphate concentration (late Eocene simulation minus modern simulation; SOM). Modern continents are indicated by stippling, and Eocene continents are white with black outline. (Right) Zonal mean phosphate anomalies. Late Eocene continental configuration is associated with greatly reduced Southern Ocean phosphate and increased tropical and Northern Hemisphere phosphate.

Alkenones are long-chained (C37 to C39) unsaturated ethyl and methyl ketones exclusively produced by haptophyte algae (12). Alkenone-based CO2 reconstructions derive from the stable carbon-isotope composition of the di-unsaturated C37 methyl ketone (δ13C37:2) and the total carbon-isotope fractionation that occurred during algal growth (εp37:2) (SOM). Nutrient-limited chemostat cultures show that εp37:2 values vary with the concentration of aqueous CO2 ([CO2aq]), specific growth rate, and cell geometry (13). Dilute batch cultures with nutrient-replete conditions produce substantially lower εp37:2 values, a different relationship for εp versus μ/CO2aq, a relatively minor response to [CO2aq] (14), and an irradiance effect (15). Concentrations of reactive soluble phosphate ([PO43–]) are necessary to estimate [CO2aq] from εp37:2 values (10, 16) (SOM). This relationship with [PO43–] reflects differences in rates of carbon fixation, but it is unlikely that [PO43–] by itself is responsible for the variability in carbon fixation rates. Instead, it is generally assumed that the availability of other biolimiting micronutrients that display distributions similar to phosphate also affects the growth characteristics of these populations, whereas [PO43–] acts as a proxy for these other variables. Despite these complexities, a study of Pacific Ocean sediments demonstrates that the alkenone methodology resolves relatively small differences in water-column [CO2aq] when SSTs and [PO43–] are reasonably constrained (16). For this work, we calculated [CO2aq] at each site assuming maximum haptophyte production occurred between 30 and 50 m depths and applied the average [PO43–] determined from modern mean-annual phosphate depth profiles and paleo-estimates from model simulations at each paleo-location (SOM). SST was determined by using temperature proxies Embedded Image (alkenone unsaturation index) and TEX86 (tetraether index) (6) (SOM). A sensitivity analysis for proxy temperatures was performed by using climate model–derived SSTs as an alternative to proxy interpretations.

Our δ13C37:2 analyses provide evidence for the persistence of two distinct, isotopic groupings (Fig. 2A). Higher δ13C37:2 values characterize Northern Hemisphere sites, the equatorial region, and southern subtropics; whereas lower δ13C37:2 values are associated with sites located in subantarctic and Antarctic waters. εp37:2 values, which eliminate the isotopic influence of inorganic aqueous carbon used during photosynthesis, reduce differences between Antarctic/subantarctic sites and more northerly sites, but distinct regional isotopic offsets persist (Fig. 2B).

Fig. 2

Compilation of isotopic compositions of di-unsaturated alkenones and εp37:2 values. (A) Carbon-isotopic compositions of di-unsaturated alkenones. Analytical accuracy of carbon isotope measurements is ±0.2‰ (root mean square error) on the basis of an n-C20 alkane standard injected daily. Standard error, represented by the error bars, was ±0.4‰ or better. (B) εp37:2 values calculated by using foraminiferal carbonate δ13C values and temperature estimates (SOM) from both Embedded Image and TEX86. Previous published data from (10). Site 336 lacks carbonates necessary for the calculation of εp37:2. Ma indicates million years ago.

Reconstructed CO2 estimates using proxy temperatures and modern nutrient distributions (10) show stark differences among sites (Fig. 3A). Specifically, all Antarctic and subantarctic sites are characterized by substantially higher CO2 estimates as well as CO2 offsets in comparison to low-latitude sites 925 and 929. These offsets that are implausible either by analogy to modern CO2 disequilibrium distributions or by those generated by our Eocene modeling that indicate Southern Ocean sites maintained CO2 values between 1 and 200 parts per million (ppm) lower than atmospheric values (SOM). Most sites show evidence of CO2 decline across the E-O climate transition (SOM). However, a compilation of CO2 records with Antarctic and subantarctic data included renders an indeterminable evaluation of atmospheric CO2 concentration and its relationship to the E-O climate shift (Fig. 3A).

Fig. 3

Calculated pco2 from εp37:2 values. Error bars indicate associated with each point reflect the uncertainty in CO2 because of the temperature uncertainties for each temperature proxy. (A) pco2 calculated by using Embedded Image and TEX86 values, assuming εf = 25‰ and using modern [PO43–] averaged between 30 and 50 m for each paleolocation (SOM). (B) Same as (A) but using an εf = 28‰. (C) Same as (A) but using modeled temperatures averaged between 30 and 50 m for each paleolocation (SOM). (D) Same as (C) but using an εf = 28‰. (E) Same as (A) but using modeled [PO43–] averaged between 30 and 50 m for each paleolocation (SOM). (F) Same as (E) but using an εf = 28‰.

Regionally higher surface-water [CO2aq] at Antarctic/subantarctic sites compared to other sites is qualitatively implied by relatively higher Antarctic/subantarctic εp37:2 values. Another approach to qualitatively evaluate regional CO2 differences is through an analysis of cell-size differences of ancient alkenone producers. Previous work interpreted substantial reductions in cell-size dimensions of alkenone-producing coccolithophores across the E-O climate transition as a response to CO2 limitation during carbon fixation (17). Our new cell-size measurements indicate that alkenone-producing algae from Antarctic/subantarctic sites were larger compared to more northerly sites (SOM). Smallest mean cell diameters are found in the equatorial Atlantic, where small and medium-sized (3 to 7 μm) reticulofenestrids and Cyclicargolithus dominated. By contrast, the abundance of large reticulofenestrids (including Reticulofenestra umbilica and R. bisecta) was significantly higher in the Antarctic/subantarctic (SOM). Theory argues that cell geometry differences should directly affect the magnitude of εp37:2 values (13). However, if differences in cell geometry were primarily responsible for the expression of carbon-isotopic compositions, relatively lower εp37:2 values would be predicted from Antarctic/subantarctic sites (13, 17), in contrast to what is observed (Fig. 2B). Instead, it is more likely that differences in cell geometry reflect higher [CO2aq], lower algal growth rates (perhaps because of low nutrient flux), or a combination of both at southern sites (17) (SOM).

In addition to cell size, CO2 estimates determined from εp37:2 rely on various assumptions including the value for the carbon isotope fractionation associated with carbon fixation (εf), SST, and [PO43–]. An εf value of 29‰ is widely accepted for the in vitro fractionation due to RuBisCO (ribulose-1,5-bisphosphate carboxylase- oxygenase) (18, 19). However, smaller in situ fractionations ranging from 25 to 28‰ are attributed to the effects of other enzymes, such as the β-carboxylase (phosphoenolpyruvate-carboxylase) (13, 15, 20, 21). Considering that β-carboxylation is responsible for bicarbonate uptake, it is possible that its expression was greatly reduced by higher ambient CO2 levels, leading to higher εf values. Application of higher εf values substantially reduces calculated regional CO2 offsets (Fig. 3B) (SOM). In comparison with other variables, the value of εf constitutes the most important lever on the magnitude of CO2 estimates, particularly as εp37:2 approaches the value of εf applied in the calculation of CO2 (Fig. 3, B, D, and F).

Proxy SST values (Embedded Image and TEX86) used in this study are subject to uncertainty, given difficulties in determining diagenetic alterations and modifications to modern empirical calibrations because of evolutionary, ecological, and physical oceanographic changes with time. For example, Eocene temperature reconstructions from the southwest Pacific (22) imply meridional temperature gradients and high-latitude zonal temperature gradients (23) that are difficult to explain by known ocean-atmosphere physics or other published records (23, 24). It is noteworthy that model simulations of Eocene climates indicate substantially lower temperatures for the southwest Pacific high-latitude site evaluated in this work (SOM), although they are generally in agreement at most other high-latitude ocean localities (6), as well as with terrestrial paleoclimate records (23). If we assume model SSTs are valid, calculated CO2 offsets between sites are also substantially reduced (Fig. 3C). Indeed, spurious proxy temperatures could prove to be the critical impediment in advancing our understanding of ancient climate systems, given their importance in our evaluation of all other aspects of ocean and atmospheric dynamics in addition to CO2 estimates.

High [PO43–] applied to Antarctic/subantarctic sites derives from modern water-column nutrient distributions (SOM), assuming that Eocene-Oligocene circulation was similar to today. This is likely a poor assumption, because the modern upwelling of deep, nutrient-rich Southern Ocean water relies on the presence of a deep channel at the latitudes of the Drake Passage. In the absence of a deep circum-Antarctic circulation during the Eocene (25), upwelling must have come from shallower depths (26, 27) and limiting nutrient concentrations were lower. To demonstrate that this is the case, we ran a present-day and an Eocene configuration of the UVic Earth system model of intermediate complexity, including a marine ecosystem and full carbon cycle (SOM). Nutrient cycles are free to evolve, and present-day simulations agree favorably with modern observations (SOM). In accordance with ocean circulation theory, our Eocene modeling results show that upwelling from the abyss was weaker in the Southern Ocean, forcing older, high-phosphate waters to upwell elsewhere and leading to lower mixed-layer nutrient concentrations in the Southern Ocean (Fig. 1). Therefore, the application of present-day [PO43–] (and the implicit assumption of high haptophyte growth rates) is likely a poor choice on the basis of known changes in paleobathymetry and the well-established physical oceanographic and nutrient flux responses they engender.

Application of simulated [PO43–] also helps to reduce CO2 offsets (Fig. 3E) but, similar to the effect of temperature, still fails to completely eliminate higher CO2 distributions in the high-southern latitudes. Higher [CO2aq] in these regions is unsupported by our model results (SOM) but supported by one interpretation of our cell-size analysis, and so the persistence of an offset is not inconsistent with the totality of proxy data. Alternatively, whereas cell-size trends imply a temporal decrease in [CO2aq], regional cell-size differences could reflect spatial differences in nutrient limitation (17) and very low haptophyte growth rates in the high-southern regions, which promoted higher εp37:2 values, consistent with model results.

Uncertainties in the magnitude of εf, [PO43–], and temperature values compromise the efficacy of the alkenone proxy at southern high-latitude localities in this study, with the potential to greatly overestimate CO2 levels and obfuscate global temporal CO2 variations when included in global compilations. Importantly, other published CO2 results from equivalent localities (28) could reflect similar biases and offsets. In contrast, these effects are substantially reduced across low-latitude sites, where uncertainties in temperature and nutrient input are smaller and the lower magnitude of εp diminishes the range of CO2 uncertainty. Accordingly, we focus on CO2 reconstructions derived from low-latitude sites 925 and 929 and apply modern phosphate concentrations at each paleolocation.

Low-latitude records show a persistent CO2 decline beginning about two million years before the onset of rapid cooling 33.7 million years ago (Fig. 4) that continues just beyond the climate event. Site 925 data suggest CO2 increased somewhat before cooling; however, corroborating results from site 929 are lacking. Site 925 also shows an increase in CO2 during the mid-Oligocene that tracks warming inferred from benthic δ18O values (Fig. 4), a pattern corroborated by εp37:2 trends from site 336 (Fig. 2A). A long-term lowering of CO2 during the Oligocene is consistent with early Miocene CO2 reconstructions from other localities (10), but our sensitivity analysis allows for higher early Miocene CO2 levels relative to previously published results if [PO43–] and temperature estimates were higher than originally assumed.

Fig. 4

Reconstructed pco2 records from low-latitude sites 925 and 929. Calculations apply εf = 25‰ and modern [PO43–] at paleolocations (SOM). Error bars associated with each point reflect the uncertainty in CO2 because of temperature uncertainties for each temperature proxy. Gray line represents a compilation of δ18O values of benthic foraminifera (30).

The decline in the partial pressure of atmospheric carbon dioxide during the E-O climate event was substantial, but absolute CO2 concentrations depend on the value of εf applied. Collectively, CO2 estimates calculated by using Embedded Image and TEX86 SST estimates and a range of εf values indicate that CO2 decreased ~40% from 35.5 to 32.5 million years ago (SOM). Application of reasonable εf values (25 to 28‰) indicates that the partial pressure of atmospheric CO2 fell from 1200 to 1000 ppm to 700 to 600 ppm. Interestingly, the change in CO2 determined from this study, as well as the boron-isotope methodology (11), is consistent with model estimates for a threshold CO2 level required for rapid Antarctic glaciation (8, 29).

We conclude that the available evidence supports a fall in CO2 as a critical condition for global cooling and cryosphere evolution ~34 million years ago. Whether CO2 acted alone to cause the E-O transition or whether a threshold CO2 level in combination with favorable orbital configurations (1) ultimately triggered glaciation cannot be determined from our results. However, during the E-O transition both CO2 decline and enhanced ice albedo account for global temperature changes. Lastly, the long-term permanence of the CO2 decline (10) and the impermanent inorganic carbon isotope shift (1) implicate the role of silicate weathering rates over the influence of short-term organic-carbon burial rates as the primary cause for long-term change in atmospheric carbon dioxide.

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S15

Tables S1 to S6

References (31112)

References and Notes

  1. Acknowledgments: This work was supported by a postdoctoral fellowship provided by Yale University Department of Geology and Geophysics, with funding provided by the NSF under awards OCE-0902882, OCE-0902993, EAR-0628358, ATM-0513402, ATM-0927946, and ANT-034248. This is PCCRC publication number 1115. Natural Environment Research Council grant NE/G003270/1 provided funds for the stable isotope analysis of carbonates. J.H. was funded by the Royal Swedish Academy of Sciences (KAW 2009.0287).

Stay Connected to Science

Navigate This Article