Long-term climate forcing by atmospheric oxygen concentrations

See allHide authors and affiliations

Science  12 Jun 2015:
Vol. 348, Issue 6240, pp. 1238-1241
DOI: 10.1126/science.1260670

Change was in the air

The atmospheric fraction of molecular oxygen gas, O2, currently at 21%, is thought to have varied between around 35 and 15% over the past 500 million years. Because O2 is not a greenhouse gas, often this variability has not been considered in studies of climate change. Poulson and Wright show that indirect effects of oxygen abundance, caused by contributions to atmospheric pressure and mean molecular weight, can affect precipitation and atmospheric humidity (see the Perspective by Peppe and Royer). These effects may thus have produced significant changes in the strength of greenhouse forcing by water vapor, surface air temperatures, and the hydrological cycle in the geological past.

Science, this issue p. 1238; see also p. 1210


The percentage of oxygen in Earth’s atmosphere varied between 10% and 35% throughout the Phanerozoic. These changes have been linked to the evolution, radiation, and size of animals but have not been considered to affect climate. We conducted simulations showing that modulation of the partial pressure of oxygen (pO2), as a result of its contribution to atmospheric mass and density, influences the optical depth of the atmosphere. Under low pO2 and a reduced-density atmosphere, shortwave scattering by air molecules and clouds is less frequent, leading to a substantial increase in surface shortwave forcing. Through feedbacks involving latent heat fluxes to the atmosphere and marine stratus clouds, surface shortwave forcing drives increases in atmospheric water vapor and global precipitation, enhances greenhouse forcing, and raises global surface temperature. Our results implicate pO2 as an important factor in climate forcing throughout geologic time.

On geological time scales, Earth’s atmospheric oxygen concentration is regulated by burial of organic carbon and pyrite, which adds O2 to the atmosphere, and oxidative weathering, which removes it (1). Long-term imbalances in these processes have caused pO2 to vary on the order of 1% on million-year time scales. Charcoal evidence of fire deposited in sediments suggests that O2 abundances have remained within the fire window, approximately 13 to 35%, since at least the Silurian (~420 million years ago) (2). Estimates of the evolution of pO2 through the Phanerozoic have been made using geochemical models of carbon and sulfur cycles (1, 37) and proxy methods based on abundances of charcoal in wetland environments (8) and stable carbon isotopes of plant resins and ambers (9). These reconstructions differ substantially, with some indicating that pO2 was essentially at or above modern levels for most of the past 300 million years (5, 6, 8) and others indicating substantial variations, including a long-term rise from levels as low as 10 to 11% in the Mesozoic to modern values (1, 3, 4, 7, 9).

In the current atmosphere, oxygen (O2) constitutes approximately 20.95% of air by volume, with nitrogen (N2) and argon (Ar) contributing 78.09% and 0.93%, respectively. As a result of this composition, dry air has a mean molecular weight of 28.97 g/mol, a typical density at sea level of ~1.2 kg/m3, and a mean sea-level pressure of 101.3 kPa. Changes to atmospheric pO2 of several percent, assuming no change in the amounts of other constituents, have a substantial effect on atmospheric density, molecular weight, and pressure (Table 1).

Table 1 Dry atmospheric conditions under variable O2 levels.

Atmospheric pN2 and pAr are fixed at modern values; Mv/Ma is the ratio of the molecular weight of water to the molecular weight of dry air.

View this table:

Unlike greenhouse gases, O2 is inactive in the infrared band, contributes weakly to the radiative forcing of the atmosphere, and is not considered an important factor in climate forcing (10). However, through its influence on atmospheric density, O2 directly influences solar optical depth (τα), a measure of atmospheric opacity, by absorption and scattering of solar radiation. In its simplest form, the solar optical depth is expressed asEmbedded Image (1)where k (m2/kg) represents a mass absorption coefficient, ρ (kg/m3) is the density of the absorber, and Δz (m) is the thickness of the atmosphere. Following Beer’s law, the fraction of solar radiation (I) that passes through the atmosphere is related to the top-of-atmosphere (TOA) intensity I0 and optical depth byEmbedded Image (2)In the modern atmosphere, ~23% of the total incoming radiation (~79 W m−2) is scattered to space by air molecules and water droplets in clouds (11). In an atmosphere with less O2, collisions between photons and air molecules are less common, reducing scattering. Using Eqs. 1 and 2 and assuming a 22.5% decrease in total air density with a decline in O2 from 21% to 10% (Table 1), total scattering is estimated to decrease by ~16 W m−2.

Here, we quantified the effect of changing pO2 on global climate with the use of an atmospheric global climate model (GCM) to account for changes in atmospheric density, mass, and molecular weights (Table 1 and supplementary materials). In doing so, we maintained the partial pressures of other atmospheric constituents (N2, Ar, CH4, N2O, and CO2, except as indicated) at constant values. We focused on simulations of the Cenomanian, a mid-Cretaceous stage characterized by high pCO2 [e.g., (12)] and the warmest conditions of the past ~100 million years (13). Paleo-pO2 estimates inferred from stable-isotope carbon compositions indicate that the atmospheric percentage of O2 was as low as 10 to 11% during this interval (7, 9), with one biogeochemical model indicating levels as high as 32.5% (6). We developed a series of simulations with O2 abundance varying between 5% and 35%, which captures the full range of both Cretaceous and Phanerozoic pO2 values estimated by geochemical modeling and proxy techniques, and pCO2 varying between preindustrial (280 ppmv) and 8 times preindustrial levels (2240 ppmv).

Varying pO2 in our simulations alters both TOA and surface radiative forcing. In the simulation with 1120 ppmv pCO2, approximately the median value of Cenomanian pCO2 proxy estimates [e.g., (12)], the global TOA shortwave forcing increases by 3.4 W m−2 with a decrease in O2 from 21 to 10% and is balanced by an increase in outgoing TOA longwave radiation. The equilibrium TOA shortwave forcing response is consistent with a reduction in outgoing shortwave radiation by Rayleigh scattering but is less than anticipated from our scaling estimate because of climate feedbacks. To demonstrate this, we ran a Cenomanian simulation with 10% O2 branched from an equilibrium 21% O2 simulation (see supplementary materials). The initial 1-day difference in the total atmospheric scattering, 23.5 W m−2, is similar in magnitude to our scaling estimate, with increases in clear-sky and cloud surface shortwave forcing of 4.5 W m−2 and 19 W m−2, respectively. The increase in cloud shortwave forcing is greater than anticipated from a reduction in scattering alone because the cloud fraction initially decreases by 10.5% as a result of the instantaneous decrease in atmospheric pressure and relative humidity. The large, instantaneous increase in shortwave forcing with a reduction in O2 drives an increase in surface latent heat flux of a similar magnitude, leading to increases in atmospheric humidity and a rebound in cloud fraction (fig. S1). As atmospheric humidity and cloud fractions rise, total shortwave scattering increases and surface and TOA shortwave forcing falls.

Under equilibrium conditions, the global net (shortwave + longwave) absorbed surface radiation is greater by 8.4 W m−2 under 10% O2 than under 21% O2. Of this total, an increase in shortwave radiative forcing accounts for 4.8 W m−2 and consists of an increase in total cloud shortwave forcing of 5.7 W m−2 and a decrease in clear-sky shortwave forcing of 0.85 W m−2. The increase in cloud forcing arises from a ~7% reduction in cloud liquid water content, causing clouds to be optically thinner and more transparent to incoming shortwave radiation, and a ~9% reduction in low-level marine stratus clouds associated with increased instability and convective activity due to enhanced surface shortwave forcing. The reduction in clear-sky shortwave forcing results from an increase in total atmospheric moisture of ~25%, which increases clear-sky scattering.

In addition to the increases in surface shortwave forcing, the downward longwave radiation at the surface increases by 3.6 W m−2. Clear-sky longwave forcing increases by 9.2 W m−2 as a result of infrared absorption and emission by water vapor, and cloudy-sky longwave forcing decreases by 5.6 W m−2 primarily as a result of a decrease in stratus cloud fraction. The net increase in surface radiation is balanced by an increase in surface latent heat flux by 9.8 W m−2 and a small reduction in the surface sensible heat flux.

Single-column radiative transfer model experiments substantiate our GCM results: In the absence of increases in atmospheric humidity, surface clear-sky radiation increases under low O2 conditions. Enhanced atmospheric humidity reduces the net clear-sky shortwave forcing and raises surface longwave radiation. And finally, changes in the cloud optical depth and cloud fraction increase surface shortwave forcing and decrease longwave forcing (fig. S2).

The increases in surface radiative forcing and compensation by the latent heat flux with decreasing pO2 lead to substantial increases in global surface precipitation rates and temperature (Fig. 1). Under 1120 ppmv pCO2, global mean annual precipitation rates (P) range from 1.11 m/year (35% O2) to 1.27 m/year (21% O2) and 1.40 m/year (10% O2). Precipitation rates generally increase, except in subtropical regions (~30° latitude), with a decrease in pO2 (Fig. 2, B and C). These precipitation anomalies reflect greater surface evaporation rates and vapor amounts, as well as enhanced vapor export from subtropical regions under reduced pO2. Enhanced latent heat fluxes raise global-average precipitable water amounts (q) from 4.9 kg/m2 (35% O2) to 6.1 kg/m2 (21% O2) and 7.6 kg/m2 (10% O2). Although atmospheric humidity increases, the total mass flux of water, approximated as M ≈ P/q (14), decreases by ~19% as O2 decreases from 35% to 10%, consistent with a simulated decrease in mean vertical velocities (fig. S3) and with previous studies that describe slowdown of atmospheric circulation with global warming [e.g., (15)].

Fig. 1 Simulated global annual surface temperature (°C) versus precipitation rate (m/year) under variable pCO2 and pO2.

Solid circles represent 20-year means from each run averaged over model years 41 to 60. CO2 (280, 560, 1120, and 2240 ppm) is indicated by the color of the solid circles, with higher values represented by darker shades. O2 ranges sequentially from 5 to 35% (as shown) for each set of pCO2 experiments.

Fig. 2 Simulated Cenomanian annual precipitation rates with different O2 concentrations.

(A) Annual precipitation rate with 21% O2 and 1120 ppm CO2. (B and C) Annual precipitation change from conditions in (A) with O2 decreased to 10% (B) and increased to 35% (C).

Global surface temperatures also increase with decreasing pO2 (Fig. 1) and, under 1120 ppmv CO2, range from 18.2°C (35% O2) to 22.5°C (10% O2), yielding differences of –2.3°C and +2.1°C relative to those under current pO2. Surface temperature differences are regionally amplified by feedbacks (Fig. 3, B and C). Under 10% O2, reductions in Northern Hemisphere snow and sea ice decrease surface albedo by up to 0.25, causing polar warming. Over subtropical continents, enhanced radiative forcing over arid surfaces with low soil moisture enhances sensible heating of the surface. In addition, reductions by up to 15% in low-level marine stratus clouds along the equator reduce reflected shortwave radiation at the surface, leading to additional warming.

Fig. 3 Simulated Cenomanian annual surface temperatures with different O2 concentrations.

(A) Annual surface temperature with 21% O2 and 1120 ppm CO2. (B and C) Annual surface temperature difference from conditions in (A) with O2 decreased to 10% (B) and increased to 35% (C).

The climate sensitivity to modulation of pO2 is distinctly different from that of pCO2. A doubling of pCO2 in the GCM leads to average increases in radiative forcing, surface temperature, and precipitation of 6.2 W m−2, 3.0 K, and 7.6 cm/year, yielding temperature and precipitation sensitivities of 0.48 K per W m−2 and 1.2 cm/year per W m−2, respectively. The precipitation sensitivity is much higher for pO2 than for pCO2 (fig. S4) for two reasons. First, solar forcing provides a larger surface forcing than greenhouse forcing. For each W m−2 in total radiative forcing, surface forcing increases by 2.1 W m−2 and 0.63 W m−2 for changes in pO2 and pCO2, respectively. Second, because CO2 decreases radiative cooling, it is less effective than solar forcing in driving changes in global precipitation [e.g., (16)]. The temperature sensitivity to pO2 is lower than that for pCO2 at low pO2 and greater at higher pO2 (fig. S4). The variability in temperature sensitivity to pO2 is at least partly due to a lapse rate feedback. Under low pO2, enhanced latent heat fluxes and convection moisten the troposphere, leading to warming throughout the column but preferentially in the mid- and upper troposphere (fig. S5). As a consequence of this warming profile, infrared emission to space is enhanced, reducing greenhouse forcing.

Atmospheric CO2 is widely considered to be the primary driver of Phanerozoic climate change; this view is supported by the coincidence of low proxy pCO2 with cold climates and of high proxy pCO2 with warm climates (17). Paleoclimate modeling has demonstrated, however, that proxy levels of pCO2 alone are not sufficient to fully explain past warm climates, particularly those of the mid-Cretaceous and early Paleogene. Paleoclimate models generally predict high-latitude temperatures that are too low, causing too-large equator-to-pole gradients (12, 1821). The low pO2 (~10% O2) estimated for the Mesozoic and early Paleogene (7, 9) would have contributed to the high-latitude warming of these greenhouse periods (Fig. 3B).

Less well recognized but equally problematic is the underestimate by paleoclimate models of precipitation rates during greenhouse periods [e.g., (20)]. A variety of proxies, including lithologic indicators [e.g., (22, 23)], geochemical proxies [e.g., (24, 25)], and leaf analyses of paleoflora (2630), indicate that mid- and high-latitude continents of the mid-Cretaceous and early Paleogene were more humid than at present. Although sparse and associated with high analytical uncertainty (28), mid-Cretaceous proxy precipitation rates based on leaf area analysis of fossil leaves from Alaska and northern Siberia typically exceed >1000 mm/year (26, 30). Low pO2 may help to account for such high values. In our Cenomanian simulations, mean annual precipitation rates in these high-latitude regions increase by up to 45% with a decrease in pO2 from current values to 10% O2 (Fig. 2B). Simulated precipitation rates increase further to values at or approaching 1000 mm/year with 10% O2 and 2240 ppm CO2, values well within the uncertainty range of proxy estimates (fig. S6 and table S3). Finally, in our simulations, the combination of low pO2 and high pCO2 causes temperate forests to expand their range to high latitudes, consistent with paleobotanical evidence (3133) (fig. S7).

Our results for the mid-Cretaceous are broadly applicable throughout Earth history and demonstrate that variations in atmospheric mass, whether through changes in pO2 or other constituents (i.e., pN2), were an important factor in climate forcing on geological time scales that must be included in paleoclimate models. We also note the broad inverse correlation between some pO2 records [e.g., (1, 9)] and global climate change during the Phanerozoic; we speculate that, to the extent that atmospheric pO2 and pCO2 have been linked through photosynthetic productivity and organic carbon burial, O2 may have amplified CO2-driven climate change.

Supplementary Materials

Materials and Methods

Supplementary Text

Figs. S1 to S7

Tables S1 to S3

References (3439)

References and Notes

  1. Acknowledgments: Supported by NSF Sedimentary Geology and Paleobiology Program grant F035175 and NSF Marine Geology and Geophysics grant F033663. We thank R. Feng, R. Fiorella, I. Montañez, and three anonymous reviewers for discussion of and comments on this work. The model data are available at
View Abstract

Stay Connected to Science

Navigate This Article