Atmospheric Carbon Dioxide Concentration Across the Mid-Pleistocene Transition

See allHide authors and affiliations

Science  19 Jun 2009:
Vol. 324, Issue 5934, pp. 1551-1554
DOI: 10.1126/science.1171477


The dominant period of Pleistocene glacial cycles changed during the mid-Pleistocene from 40,000 years to 100,000 years, for as yet unknown reasons. Here we present a 2.1-million-year record of sea surface partial pressure of CO2 (Pco2), based on boron isotopes in planktic foraminifer shells, which suggests that the atmospheric partial pressure of CO2 (pco2) was relatively stable before the mid-Pleistocene climate transition. Glacial Pco2 was ~31 microatmospheres higher before the transition (more than 1 million years ago), but interglacial Pco2 was similar to that of late Pleistocene interglacial cycles (<450,000 years ago). These estimates are consistent with a close linkage between atmospheric CO2 concentration and global climate, but the lack of a gradual decrease in interglacial Pco2 does not support the suggestion that a long-term drawdown of atmospheric CO2 was the main cause of the climate transition.

The mid-Pleistocene transition (MPT) is the period around 1250 to 700 thousand years ago (ka), when global climate variability changed from the dominant 40-thousand-year (ky) orbital period of the Pliocene/early Pleistocene to the 100-ky ice-age cycles of the past 700 ky (13). Orbital variation does exert some forcing on the 100-ky time scale, but it is relatively weak and seems a feeble explanation for the 100-ky ice ages. The change in periodicity was accompanied by a gradual increase in value and amplitude of the oxygen isotopic composition of benthic foraminifer shells, suggesting that total ice volume increased and/or deep-water temperatures probably decreased over the MPT (3, 4). It has been suggested the MPT was caused by global cooling, possibly due to a long-term decrease in atmospheric CO2 concentrations (5), but the evidence is inconclusive. Sea surface temperature (SST) estimates from eastern basin upwelling areas (69) are consistent with substantial cooling, but estimates from the western Pacific warm pool (WPWP) indicate relatively stable temperatures across the transition (10, 11). Because the WPWP is an area particularly sensitive to changes in radiative forcing, that temperature stability has been used to argue that a secular decrease in atmospheric partial pressure of CO2 (pco2) did not occur (10). In contrast, another study observed higher glacial SSTs before the MPT and ascribed these to changes in greenhouse forcing (11). Thus, there currently is no direct evidence substantiating any long-tem trend in pco2.

The most accurate archive for atmospheric pco2 comes from ancient air trapped in polar ice. Ice core records reveal that pco2 varied between 180 and 300 parts per million by volume (ppmv) during the last four glacial cycles (12) and between 172 and 260 ppmv for the period from 800 to 450 ka (13, 14). Contrary to the suggestion that pco2 decreased toward the late Pleistocene, the earlier pco2 amplitude and average were lower than in the more recent past. However, existing ice core records are limited to the past 800 ky, and no ice core data are available for the full duration of the MPT.

Because CO2 is well mixed in the atmosphere over the time scale of a few years, and because CO2 is exchanged rapidly between the surface ocean and atmosphere, marine proxy records of past sea surface carbonate chemistry can place constraints on past atmospheric pco2. The boron isotopic composition of planktic foraminifer shells is a proxy for past seawater pH. This proxy is based on the equilibrium reaction between the two dominant species of dissolved boron in seawater and the isotope fractionation between the two species [supporting online material (SOM)]. Atmospheric pco2 can be estimated from the boron isotopic composition of those shells if (i) aqueous partial pressure of CO2 (Pco2) at the core site is in equilibrium with atmospheric pco2, and (ii) another carbon parameter of the water in which the foraminifers grew is known. Using reasonable assumptions about seawater alkalinity, quantitative replication of select intervals of the Vostok pco2 record from boron isotopes in the planktic foraminifer Globigerinoides sacculifer (15) has demonstrated the validity of this method.

Here, we extend the existing 400-ky boron isotope record from Ocean Drilling Program (ODP) site 668B beyond the MPT to 2.1 million years ago. ODP site 668B is located on the Sierra Leone Rise in the eastern equatorial Atlantic (4°46′N, 20°55′W) at a water depth of 2693 m. Nearby oceanographic data (from World Ocean Circulation Experiment cruise A15, station 34) indicate that in the modern ocean, aqueous Pco2 and atmospheric pco2 are in equilibrium. Reconstruction of the local marine carbonate chemistry thus allows us to estimate Pco2 and infer pco2.

From this sediment core, we constructed a high-resolution oxygen isotope (δ18O) record from shells of the surface-dwelling G. ruber. The record allows us to establish an age model, which is simultaneously tied (16) to the stack of 57 globally distributed benthic δ18O records [the LR04 stack (4)] and the planktic δ18O record of ODP site 677 (17). Large G. sacculifer shells were selected from extreme glacial and interglacial samples and transitional periods for boron isotope analysis (fig. S1 and SOM). Boron isotopes were measured by negative thermal ionization mass spectrometry and complemented by Mg/Ca analyses on small shells of G. ruber for temperature reconstruction (see SOM for further details). Boron isotope data were then converted into pH estimates, using the empirical calibration for G. sacculifer (18) and following the procedure outlined in (15). Mg/Ca-based SSTs were estimated according to the method of (19). The salinity effect on Mg/Ca temperature estimates discovered by (20) has been considered but found to be of negligible importance for Pco2 estimates (SOM).

In order to translate the pH estimates into Pco2, a second parameter of the carbonate system is required, such as [CO3=] or alkalinity. An evaluation of methods to estimate this second parameter is given in the SOM. We followed a similar procedure to that outlined in (15), bracketing the potential variations in whole-ocean inventory of alkalinity between two possibilities: (i) Alkalinity remained constant in the past (constant alkalinity scenario); and (ii) alkalinity varied as a function of past terrestrial weathering, ocean CaCO3 production, and sediment dissolution rates [varying alkalinity scenario, after (3)]. The degree to which local salinity and alkalinity were increased during glacial periods, when sea level was lower, was approximated with the global sea-level estimate determined by (21) relative to average ocean depth (see the SOM for further details and additional validation of this approach).

The pH, temperature, and Pco2 estimates are shown in Fig. 1. The pre-MPT ocean chemistry revealed from δ11B is characterized by less basic glacials (pH 8.24 ± 0.03, 1 SD) relative to the post-MPT (pH 8.29 ± 0.02), whereas interglacial pH is comparable between the two intervals (pH 8.14 ± 0.02 and 8.14 ± 0.03, respectively). Several uncertainties were considered for the Pco2 calculation (SOM) but were found to be of minor importance. The only significant Pco2 difference stems from the choice of alkalinity, where a total difference in alkalinity of up to 221 μmol kg−1 between scenarios results in a maximum Pco2 difference of 27 microatmospheres (μatm) and an average difference of 17.7 μatm (fig. S1 and SOM). The Pco2 estimates presented here are based on the varying alkalinity scenario [after (3)], in which weathering of the Canadian Shield modulates 4% of global weathering rates. Although all Pco2 estimates are very similar, this one yields the best match with the measured atmospheric pco2 over the past 800 ky (fig. S1 and SOM). According to this scenario, our proxy estimates translate into pre-MPT interglacial Pco2 similar to that of the most recent interglacials (283 μatm) but relatively higher pre-MPT glacial Pco2 (213 μatm). In comparison, the constant alkalinity scenario yields Pco2 ~ 10 to 20 μatm higher for the entire record (fig. S1 and SOM). The large uncertainty in alkalinity thus results in only a small difference in the Pco2 estimate and indicates that seawater pH is a a sensitive parameter for calculating Pco2. Because changes in ocean alkalinity were probably accompanied by similar changes in total dissolved inorganic carbon, they appear to exert only a small impact.

Fig. 1

2.1-million-year estimation of atmospheric pco2 from marine proxies recorded at ODP site 668B in the eastern equatorial Atlantic. (A) The LR04 benthic oxygen isotope stack (4) reflects the change from the dominant 40-ky periodicity of glacial cycles before 1200 ka to the 100-ky ice-age cycles of the past 700 ky. The MPT is indicated by the horizontal gray bar and describes the transition period. (B) Planktic δ11B data reflect extreme glacial and interglacial times and few transitional periods, selected from the δ18O record, and are complemented by (C) Mg/Ca SST estimates and (E) local salinity estimates computed from (D) modeled global sea level (21). (F) Surface seawater pH on the seawater scale (SWS) was calculated as a function of δ11B, SST, and salinity. (G) Alkalinity estimates are based on a modeled global ocean estimate (3), adjusted to modern local alkalinity at the core site and varying sea level. (H) Surface ocean aqueous Pco2 was then calculated as a function of pH, alkalinity, SST, and salinity. Comparison with the ice core record of atmospheric pco2 [dark red line in (H)] reveals a remarkable match for the period from 800 ka to the present. Average glacial and interglacial Pco2 is indicated by solid horizontal black lines for different intervals in (H). Dashed lines indicate G/I averages. Error bars indicate the propagated error of the individual pH, SST, salinity, and alkalinity uncertainties on the Pco2 estimate (see SOM for details).

Considering only extreme glacial/interglacial (G/I) samples, as determined by the oxygen isotope stratigraphy, the values and amplitudes of our reconstructed Pco2 cycles show good agreement with ice core measurements (Table 1). The ice core pco2 data for the intervals 0 to 418 ka (180 to 300 ppmv range, 239 ppmv average) and 540 to 800 ka (172 to 260 ppmv range, 221 ppmv average) (1214) are remarkably consistent with our marine proxy estimates for those intervals (184 to 297 μatm range, 241 μatm average; and 181 to 252 atm range, 217 μatm average, respectively). In addition, the exceptionally low pco2 (172 ppmv) measured in ice cores during marine isotope stage (MIS) 16 is reflected in a similarly low boron isotope estimate of 167 ± 13 μatm.

Table 1

Comparison of glacial and interglacial pco2 extremes as measured from ice cores (1214) and estimated from boron isotopes in planktic foraminifers (see fig. S1 for the selection and number of data included in each average). Cumulative uncertainties have been calculated for the G/I boron isotope Pco2 estimates, assuming that the propagated errors of individual data are normally distributed.

View this table:

Our reconstructed pH and Pco2 changes also agree well with the climate signal recorded in the LR04 stack (4): Extreme interglacial benthic δ18O was relatively constant over the course of our 2.1-million-year record [3.2 to 3.5 per mil (‰)], but extreme glacial benthic δ18O increased from 4.3‰ at 2.1 Ma to 5.1‰ for the post-MPT glacials (Fig. 1). The relatively less extreme glacials of the pre-MPT are thus reflected in a smaller land ice extent and/or warmer deep-sea temperatures and correlate well with higher glacial pco2. Our data are also consistent with SST reconstructions from the WPWP indicating warmer glacial SSTs pre-MPT as compared to post-MPT, but similar interglacial SSTs (11). Strong correlations also exist between sub-Antarctic alkenone SST and ice core change in temperature and between the abundance of alkenones in ODP site 1090 and pco2 measured in ice cores (22). The record covers only 1.1 million years but agrees well throughout with our boron isotope reconstruction. In particular, the alkenone record shows exceptionally high SST and exceptionally low alkenone abundance for MIS 25 (950 ka), suggesting high pco2 of ≥300 ppm. This observation agrees well with the warmest interglacial SSTs observed in the WPWP for MIS 25 (11) and is consistent with our boron isotope estimate, which indicates higher than average Pco2 (308 μatm) for MIS 25. Comparison with independent climate records of polar ice extent and deep ocean temperature (4), sub-Antarctic SST (22), and WPWP SSTs (11) thus corroborates the validity of our estimates and supports the notion that interglacial atmospheric pco2 before the MPT was similar to that in the preindustrial period, but that pre-MPT glacial pco2 was ~31 μatm higher than during post-MPT glacials.

In order to estimate climate sensitivity from the observed CO2 changes, we focus on the pre-MPT glacials, because pre- and post-MPT interglacial Pco2 are statistically indistinguishable (Table 1). We use a logarithmic climate sensitivity equation (23) with an average equilibrium temperature change of 5 K for doubling CO2. This temperature change is at the high end of CO2 sensitivity estimates on short time scales but is more appropriate for the longer time scales considered here, which include slow feedbacks such as ice-sheet albedo effects (24). Based on this sensitivity, the global average surface air temperature during pre-MPT glacials was 1.06 K warmer than during post-MPT glacials. In comparison, the mean global temperature change from the Last Glacial Maximum to the preindustrial period is 3.3 to 5.1 K [as estimated by the Paleoclimate Model Intercomparison Project (25)]. Consequently, the average pre-MPT G/I global temperature change was ~30% smaller than during the past 400 ky. However, the smaller temperature range does not imply a gradual decline in greenhouse forcing over the MPT. Although the average G/I pco2 was ~7 μatm higher before the transition (Table 1), this difference is entirely a result of higher glacial pco2. The higher glacial pco2 is consistent with the warmer glacial SSTs, ice extent, and/or deep-sea temperatures, but interglacial pco2 was similar before and after the transition. We therefore conclude that CO2 was unlikely to have been the main driver of the MPT. We also conclude that present-day atmospheric pco2 is the highest it has been for the past 2.1 million years, requiring a search for an analog for present-day conditions, possibly during the time before the acceleration of Northern Hemisphere glaciation before 2.7 million years ago.

Supporting Online Material

Materials and Methods

Figs. S1 and S2

Table S1


References and Notes

  1. We thank ODP for sediment samples and NSF (grant OCE 06-23621) for financial support. The idea for this study was seeded through conversations with S. Hemming. L. Leon is gratefully acknowledged for laboratory assistance, P. deMenocal for Mg/Ca analyses at the Lamont-Doherty Earth Observatory (LDEO), and M. Segl for isotope analyses at Bremen University. We also thank C. Pelejero and two anonymous reviewers for valuable comments. Discussions with P. Köhler and R. Bintanja improved the manuscript. The oxygen isotope data are available in the National Oceanic and Atmospheric Administration Paleoclimatology database ( This is LDEO contribution number 7261.
View Abstract

Navigate This Article