A Pleistocene ice core record of atmospheric O2 concentrations

See allHide authors and affiliations

Science  23 Sep 2016:
Vol. 353, Issue 6306, pp. 1427-1430
DOI: 10.1126/science.aaf5445


The history of atmospheric O2 partial pressures (Po2) is inextricably linked to the coevolution of life and Earth’s biogeochemical cycles. Reconstructions of past Po2 rely on models and proxies but often markedly disagree. We present a record of Po2 reconstructed using O2/N2 ratios from ancient air trapped in ice. This record indicates that Po2 declined by 7 per mil (0.7%) over the past 800,000 years, requiring that O2 sinks were ~2% larger than sources. This decline is consistent with changes in burial and weathering fluxes of organic carbon and pyrite driven by either Neogene cooling or increasing Pleistocene erosion rates. The 800,000-year record of steady average carbon dioxide partial pressures (Pco2) but declining Po2 provides distinctive evidence that a silicate weathering feedback stabilizes Pco2 on million-year time scales.

The importance of O2 to biological and geochemical processes has led to a long-standing interest in reconstructing past atmospheric O2 partial pressures (Po2, reported at standard temperature and pressure) (112). However, there is no consensus on the history of Phanerozoic Po2, with reconstructions disagreeing by as much as 0.2 atm, the present-day pressure of O2 in the atmosphere (e.g., 7, 10). Even over the past million years, it is not known whether atmospheric O2 concentrations varied or whether the O2 cycle was in steady state (Fig. 1A). Knowledge of Po2 over the past million years could provide new insights into the O2 cycle on geologic time scales and serve as a test for models and proxies of past Po2. Here we present a primary record of Po2 over the past 800,000 years, reconstructed using measured O2/N2 ratios of ancient air trapped in polar ice.

Fig. 1 δO2/N2 and PO2 values versus age from ice cores and from model and proxy predictions.

(A) Comparison of the ice core data with model and proxy predictions (112). (B) δO2/N2 versus ice age from ice cores. δO2/N2 decreases by 8.4‰ per million years (±0.2, 1σ). Gray bands are 95% confidence intervals. Data are corrected for gravitational, interlaboratory, and bubble close-off fractionations (19). kyr, thousand years; myr, million years.

O2/N2 ratios of this kind have been extensively used to date ice cores on the basis of the correlation between O2/N2 and local summertime insolation (1317). Despite being directly tied to atmospheric compositions, O2/N2 ratios have never before been used to reconstruct past Po2. Landais et al. (16) and Bazin et al. (17), while using O2/N2 ratios for ice core dating, noted a decline in O2/N2 values with time (i.e., toward the present). They suggested that this decline could be due to secular changes in air entrapment processes, gas loss during core storage, or changes in atmospheric O2/N2, but they did not evaluate these hypotheses. Given the potential for O2/N2 ratios to directly constrain Pleistocene Po2, we present compiled O2/N2 measurements from multiple ice core records and evaluate their geochemical implications.

We compiled published O2/N2 ice core records from Greenland [Greenland Ice Sheet Project 2 (GISP2) (18)] and Antarctica [Vostok (13), Dome F (14), and Dome C (17); table S1], along with previously unpublished Antarctic Ar/N2 records [Vostok and Dome C; table S2]. The data were treated as follows [see (19) for more details]. (i) Measured ratios were corrected for gravitational fractionations and are reported using δ notationEmbedded Image (1)Embedded Image(2)where brackets denote concentrations. A decrease in δO2/N2 of 1 per mil (‰) equates to a 0.1% decrease in Po2 relative to the preanthropogenic atmosphere (i.e., the modern atmosphere corrected for fossil fuel combustion). We define the preanthropogenic atmosphere as having δO2/N2 = 0‰ and δAr/N2 = 0‰. (ii) Only analyses of bubble-free ice with clathrates were considered. (iii) The portions of the δO2/N2 and δAr/N2 signals linked to insolation (1317) were removed (figs. S1 and S2). (iv) We corrected for differences in bubble close-off fractionations between ice cores and interlaboratory offsets by assuming that, in the absence of such effects, trapped gases of a given age share identical atmospheric O2/N2 and Ar/N2 values (figs. S3 and S4).

The fully corrected data are plotted versus ice age in Figs. 1B (δO2/N2) and 2A (δAr/N2). δΟ2/N2 values decrease by 8.4‰ per million years (±0.2, 1σ), consistent with the observations of Landais et al. (16) and Bazin et al. (17). δAr/N2 values increase by 1.6‰ per million years (±0.2, 1σ), which is discussed below.

Fig. 2 Evidence that the observed decline in δO2N2 with time does not originate from either secular changes in bubble close-off fractionations or ice core storage.

(A) δAr/N2 and δO2/N2 versus ice age. Bubble close-off processes and gas loss would cause δAr/N2 and δO2/N2 to covary with slopes of 0.3 to 0.6. The observed δAr/N2 trend does not overlap with these expected trends (orange wedge), indicating that such processes did not cause the decline in δO2/N2. (B) Dome C δO2/N2 versus ice age and (C) versus depth. Dotted lines were fit to ice >400,000 years old or >2600 m deep and extrapolated to younger ages or shallower depths. Extrapolations of the fits pass through the younger data (B) but miss the deeper data [beyond 4σ (C)], indicating depth-dependent glacial properties did not cause the decline in δO2/N2. Gray bands are 95% confidence intervals. Data are corrected for gravitational, interlaboratory, and bubble close-off fractionations (19).

The decline in δO2/N2 with time could result from temporal changes in bubble entrapment processes, effects of ice core storage, a decline in Po2, or an increase in the partial pressure of atmospheric N2 (Pn2). We now evaluate these possibilities in the context of the δO2/N2 record.

δO2/N2 values of gas extracted from ice are ~5 to 10‰ lower than those of ambient air (1318, 20, 21). Additionally, δAr/N2 covaries with δO2/N2 along slopes of 0.3 to 0.6 (fig. S5) (19, 21, 22). These depletions and covariations have been attributed to fractionations created during bubble close-off on the basis of measurements and models of firn air (20, 22) and the covariation of δO2/N2 and δAr/N2 with local insolation (figs. S1 and S2) (1317, 19). If secular changes in bubble close-off fractionations caused the decline in δO2/N2, then δAr/N2 values should covary with δO2/N2 along slopes of 0.3 to 0.6 and thus decline by 2.5 to 5.0‰ per million years. Instead, δAr/N2 increases with time by 1.6‰ per million years (±0.2, 1σ; Fig. 2A). The increasing trend is largely due to a subset of Vostok data from 330,000- to 370,000-year-old ice that is lower in δAr/N2 by ~1‰ compared with younger data. Exclusion of this subset yields an increase in δAr/N2 with time of only 0.35‰ (±0.20, 1σ), within the 2σ error range of no change. Regardless, whichever way the δAr/N2 are analyzed, they are inconsistent with the decline in δO2/N2 being caused by bubble close-off processes (Fig. 2A).

Ice core storage, under some conditions, causes the δO2/N2 values of trapped gases to decline (1417). Thus, the second possibility that we consider is that ice core storage lowered the δO2/N2 values so that the slope observed in Fig. 1 is an artifact. For example, a change in δO2/N2 correlated with ice age but unrelated to atmospheric compositions could result if the retention of O2 versus N2 during storage is a function of pre-coring properties controlled by original ice depths (e.g., in situ temperature, pressure, or clathrate size). We evaluate this possibility by using three approaches. (i) Gas loss during core storage causes δAr/N2 to decline at half the rate of δO2/N2 (21, 23, 24). However, as discussed above, the δAr/N2 values are not consistent with such a change (Fig. 2A). (ii) Because some ice properties (e.g., temperature and pressure) can vary linearly with ice depth, we tested whether the Dome C δO2/N2 data are better fit by a linear relationship when plotted against ice age or depth. (We note that only the Dome C ice core’s age-depth relationship is sufficiently curvilinear for this test to be useful. We linearly regressed both age and depth against δO2/N2 for ice older than ~400,000 years (i.e.,deeper than 2600 m) and extrapolated the fits to younger ages and shallower depths. The extrapolation for age (Fig. 2B) passes through the younger data, whereas the extrapolation for depth (Fig. 2C) misses the shallower data (by >4σ). (iii) Repeat δO2/N2 measurements of Vostok ice from the same age interval (150,000 to 450,000 years ago) made 10 years apart (13, 15) differ on average by 6‰, with longer storage leading to lower δO2/N2. Despite this, regressing δO2/N2 against time yields statistically identical (within 1σ) slopes of δO2/N2 versus age for both data sets (fig. S6).

Collectively, the data and tests presented above provide no support for the observed decrease in δO2/N2 over time being an artifact of either bubble close-off processes as they are currently understood or ice core storage. Consequently, we hypothesize and proceed with the interpretation that the observed decline in δO2/N2 reflects changes in Po2 or Pn2. Because N2 has a billion-year atmospheric lifetime (25), we link the decline in δO2/N2 with time exclusively to a decline in Po2. Our hypothesis is further supported by the observation that data from all four ice cores individually exhibit the same general trends and magnitudes of decreasing δO2/N2 with time (table S3), even though each was drilled, stored, and analyzed differently.

The question raised by this record is why Po2 has decreased by ~7‰ over the past 800,000 years. Changes in Po2 require imbalances between O2 sources [dominantly modern sedimentary organic carbon (Corg) and pyrite burial] and sinks (dominantly ancient sedimentary Corg and pyrite oxidation) (26). Thus, a higher rate of oxidative weathering relative to Corg and/or pyrite burial over the past million years could have caused the observed Po2 decline. The ~2-million-year (+1.5/–0.5 million years) (26) geological residence time of O2, combined with the decline in δO2/N2 of 8.4‰ per million years, indicates that O2 sinks were 1.7% larger than sources over the past 800,000 years (27). We now explore possible causes for this drawdown, examining first the impact of changing erosion rates and second the impact of global cooling on Po2.

Global erosion rates influence the amount of rock weathered (consuming O2) and sediment buried (releasing O2). These rates have been suggested to have increased up to 100% in the Pleistocene relative to the Pliocene (28) [though this is debated (29)]. Thus, the possibility exists that increased Pleistocene sedimentary erosion and burial rates affected Po2 levels. Indeed, Torres et al. (30) modeled that increasing erosion rates over the past 15 million years enhanced oxidation of sedimentary pyrite relative to burial so that Po2 declined on average by 9 to 25‰ per million years. This is similar to the decline given by the ice core record (8.4‰ per million years). We note that whether increasing erosion rates cause Po2 to decline (instead of increase) is unknown (31).

Large increases (e.g., 100%) in Pleistocene erosion rates, if they did occur, likely would have required processes that keep O2 sources and sinks balanced within ~2% (the observed imbalance). Such processes could include the proposed Po2-dependent control of Corg burial fluxes on sedimentary phosphorus burial rates (32). Alternatively, sedimentary mineral surface area is known to positively correlate with total sedimentary Corg and pyrite content (33). Hedges and Kiel (33) proposed that the total eroded and total buried mineral surface areas today are about equal. If this was true in the past, the conservation of eroded versus newly generated mineral surface area may have acted to balance Corg and pyrite weathering and burial fluxes (and thus O2 fluxes), regardless of global erosion rates (33).

Alternatively, on the basis of 13C/12C and 18O/16O records from sedimentary carbonates, Shackleton (2) proposed that Po2 declined over the Neogene as a result of oceanic cooling. He suggested the following feedback loop: Cooling increases O2 solubility. This raises dissolved O2 concentrations, which increases the volume of ocean sediment exposed to dissolved O2 and thus also increases global aerobic Corg remineralization rates (33). On million-year time scales, Corg burial rates and, therefore, Po2 and O2 concentrations decline until seawater O2 concentrations return to their initial (precooling) levels. At this new steady state, Corg burial rates have returned to their original values, but Po2 is stabilized at a lower value.

Shackleton’s hypothesis can be evaluated to first order in the context of the δO2/N2 data by using records of past ocean temperature. Specifically, temperatures in the deep (>1000 m depth) ocean were roughly constant from 24 to 14 million years ago (34, 35). Assuming an O2 residence time of ~2 million years and the hypothesis that changes in ocean temperature modulate Po2, then O2 sources and sinks would have been in balance by 14 million years ago. The oceans have cooled on average by 0.3°C per million years over the past 14 million years and 0.5° to 1.1°C per million years over the past 5 million years (34, 35). Cooling of 0.3° to 1.1°C per million years increases O2 solubility by ~7 to 25‰ per million years (36). If dissolved O2 concentrations remained constant (as this hypothesis requires), such changes in O2 solubility necessitate a decline in Po2 of ~7 to 25‰ per million years. These rates bracket the rate of decline given by the ice core record (8.4‰ per million years; Fig. 1A). We note that deep ocean cooling rates track average marine cooling rates, but not precisely, because modern deep waters form in and thus reflect the temperatures of high latitudes. Regardless, the critical point is that this simple calculation is consistent with the ice core–derived δO2/N2 record and supports the hypothesis that global temperature stabilizes Po2 on geological time scales through feedbacks associated with Corg burial rates.

A drop in Po2 over the past 800,000 years due solely to changes in Corg burial versus oxidation rates (regardless of the cause) requires positive CO2 fluxes (~3 × 1011 moles C per year) into the ocean and atmosphere (19). However, ice core records of past carbon dioxide partial pressures (Pco2) show no obvious change in the mean over the past 800,000 years (3739) (Fig. 3). To understand how changes in Po2 influence Pco2, we developed a simple model of the carbon cycle that allows for changes in weathering and burial rates of carbonates, Corg, and silicates (19). In the absence of any Pco2-dependent feedbacks, a constant decline in δO2/N2 of 8.4‰ over the past million years from a net imbalance in Corg fluxes causes Pco2 to rise by ~140 parts per million over the same time frame. Such a rise is inconsistent with the Pco2 record (Fig. 3). A Pco2-dependent silicate weathering feedback (40) can account for the higher CO2 flux if silicate weathering is enhanced by ~6% relative to volcanic outgassing. For example, response times for silicate weathering of 200,000 to 500,000 years (41) stabilize Pco2 levels within ~1 million years (Fig. 3).

Changes in Cenozoic climate began millions of years before the start of our ice core–based δO2/N2 record 800,000 years ago (e.g., 2, 30, 34, 35). Thus, we suggest that modest enhancements in silicate weathering would already have stabilized the portion of the Pco2 ice core record that is controlled by differences in Corg and pyrite burial and oxidation. Thus, the combination of changing Po2 and constant average Pco2 provides distinctive evidence for feedbacks that regulate Pco2 on geologic time scales (37). Lastly, a 2% imbalance in O2 fluxes results in only a ~0.1‰ shift in the 13C/12C ratio of buried carbon (19).

Fig. 3 Comparison of calculated and measured Pco2 values due to declining Po2 with and without a Pco2-dependent silicate weathering feedback.

Inclusion of a silicate weathering feedback with geologically reasonable response times [200,000 to 500,000 years (41)] stabilizes Pco2 within ~1 million years. Thus, increased silicate weathering rates could have compensated for enhanced CO2 fluxes from increased net Corg oxidation more than 800,000 years ago. The Pco2 records are continuous only from 800,000 years to the present. The model used to calculate PCO2 values is described in (19); the measured PCO2 values are from (38) and (39). ppm, parts per million.

Our results provide a primary record of declining Po2 over the past 800,000 years sustained by a ~2% imbalance between O2 sources and sinks. Critically, this decline is consistent with previously proposed and relatively simple models that invoke either the effects of increased Pleistocene erosion rates or decreased ocean temperature to explain feedbacks in the global cycles of carbon, sulfur, and O2—and the effects of both could have contributed to the observed decline in Po2. Regardless, creating primary records of past Po2 is the necessary first step in identifying the fundamental processes that regulate Po2 on geological time scales. Given evidence that both global erosion rates and temperature have changed markedly over the Cenozoic (42), the ideas presented here may have implications for the history of Po2 beyond the Pleistocene.


Materials and Methods

Figs. S1 to S6

Tables S1 to S3

References (4378)


  1. Materials and methods are available as supplementary materials on Science Online.
  2. The imbalance is calculated as follows: The total imbalance (moles per million years) for O2 is 0.0084 × no2, where no2 is the total number of moles of O2 in the atmosphere. The O2 flux is no2 divided by its residence time. The residence time of O2 is about 2 million years. The percent imbalance is the total imbalance divided by the total flux, or 0.0084no2/(no2/2) = 0.017 (1.7%).
Acknowledgments: D.A.S. acknowledges funding from a National Oceanic and Atmospheric Administration Climate & Global Change postdoctoral fellowship. J.A.H. and M.L.B. acknowledge support from National Science Foundation grant ANT-1443263. All data presented are available in the supplementary materials. We thank W. Fischer, I. Halevy, N. Planavsky, J. Severinghaus, and D. Sigman for helpful discussions and three anonymous reviewers for helpful comments on the manuscript. D.A.S., J.A.H., and M.L.B. conceived the study and wrote the manuscript. D.A.S., J.A.H., M.L.B., and Y.Y. analyzed the data. G.B.D. measured the Dome C δAr/N2 data. The views expressed in this article are those of the authors and do not necessarily represent the views of the Department of Energy or the U.S. Government.

Stay Connected to Science

Navigate This Article