Changes in atmospheric oxygen concentration over Earth history are commonly related to the evolution of animals and plants. But there is no direct geochemical proxy for O2 levels, meaning that estimations rely heavily on modeling approaches. The results of such studies differ greatly, to the extent that today’s atmospheric mixing ratio of 21% might be either the highest or lowest level during the past 200 m.y. Long term oxygen sources, such as the burial in sediments of reduced carbon and sulfur species, are calculated in models by representation of nutrient cycling and estimation of productivity, or by isotope mass balance (IMB)—a technique in which burial rates are inferred in order to match known isotope records. Studies utilizing these different techniques produce conflicting estimates for paleoatmospheric O2, with nutrient-weathering models estimating concentrations close to, or above, that of the present day, and IMB models estimating low O2, especially during the Mesozoic. Here we reassess the IMB technique using the COPSE biogeochemical model. IMB modelling is confirmed to be highly sensitive to assumed carbonate δ13C, and when this input is defined following recent compilations, predicted O2 is significantly higher and in reasonable agreement with that of non-IMB techniques. We conclude that there is no model-based support for low atmospheric oxygen concentrations during the past 200 m.y. High Mesozoic O2 is consistent with wildfire records and the development of plant fire adaptions, but links between O2 and mammal evolution appear more tenuous.
Oxygen fuels the chemical reactions that take place in the mitochondria of eukaryotic cells, and pO2 therefore places limits on the performance and survival of animals. Thus, changes in atmospheric O2 concentration over Earth history are commonly seen as triggers for animal, and later for mammal, evolution (Lyons et al., 2014; Falkowski et al., 2005). Ratios of O2:CO2 determine the efficiency of photosynthesis, and variations in pO2 dramatically influence wildfire dynamics, leading to strong potential links between pO2 and plant evolution (He et al., 2012). But long-term variations in oxygen are difficult to estimate: the continuous presence of fossilized charcoal in sediments younger than 420 Ma indicates sufficient oxygen to sustain combustion (O2 > 15% of the atmosphere; Belcher and McElwain, 2008), and the severity of fires in hyperoxic environments suggests that O2 has remained below ∼30% during this period (Jones and Chaloner, 1991; Belcher et al., 2010).
Between these limits, calculating variations in atmospheric oxygen relies on “forward” biogeochemical models of long-term O2 source-and-sink processes or the interpretation of geochemical proxies (Fig. 1). Forward models can be divided into two groups, depending on how they estimate the burial rate of reduced carbon and sulfur, which are the principal sources of O2 over geological time scales. These organically mediated fluxes can be either estimated from the input of material and nutrients via weathering (Arvidson et al., 2013; Bergman et al., 2004; Hansen and Wallman, 2003; shown in green in Fig. 1) or inferred by comparing geological carbon and sulfur isotope records to the isotopic composition of modeled sediments (isotope mass balance [IMB]; Berner, 2009; Falkowski et al., 2005; red in Fig. 1).
Estimates of O2 differ greatly between different forward models. Nutrient and weathering models typically predict higher values, while IMB models predict low pO2 during the Mesozoic, potentially in conflict with the evidence for widespread fires (Belcher and McElwain, 2008). All models show general agreement for a gradual rise in pO2 during the Cretaceous, although they disagree on whether this was a rise from low O2 toward present values, or a rise from present to superambient levels, followed by a decline over the Cenozoic.
“Proxy inversion” methods estimate atmospheric oxygen by reference to geochemical data. Glasspool and Scott (2010) assumed a correlation between the abundance of charcoal in mires and atmospheric oxygen, scaling to the present-day value and an assumed a Permian-Carboniferous O2 maximum of 30%. Algeo and Ingall (2007) related Corg:P ratios in organic-rich sediments to benthic redox conditions, and therefore to global atmospheric O2 levels, scaling to the fire window. The “rock abundance” method of Berner and Canfield (1989) utilizes the carbon and sulfur contents of ancient sediments as well as sedimentation rate to infer oxygen production rates, linking this to pO2. Tappert et al. (2013) inferred pO2 from measured plant resin δ13C and CO2 proxies, reasoning that plant δ13C reflects the CO2:O2 ratio of the growth environment. This technique produces very low estimates and is subject to high uncertainty in quantifying the plant δ13C response to global O2 and CO2 variations.
The level of disagreement in current O2 reconstructions is extreme and is masked to some degree by the scaling of many results to the fire window. This makes it difficult to assess the role of oxygen in the evolution of plants and animals during the Mesozoic and Cenozoic. Moreover, the forward models discussed here are commonly applied in studies of Paleozoic and Precambrian oxygen shifts (Clapham and Karr, 2012; Lenton and Watson, 2004), with important implications for the evolution of animals and land plants.
In this paper, we focus on the question of whether atmospheric oxygen concentration over the past 200 m.y. has been generally below or above the present-day value. We address this by exploring the isotope mass balance technique, which currently produces the most reliable and widely cited evidence for low Mesozoic oxygen.
FORWARD MODELING OF PALEOATMOSPHERIC OXYGEN CONCENTRATION
Forward models are based on the long-term carbon and sulfur cycles (e.g., Kump and Garrels, 1986), as shown in Figure 2. These systems consider atmospheric and oceanic carbon (A) and sulfur (S) and the much larger sedimentary reservoirs of oxidized and reduced species. The crustal reservoirs can be split into young (y) and ancient (a) sediments, with the assumption that the young reservoirs are smaller and constitute the majority of interaction with the surface system. This “rapid recycling” permits the isotopic signature of the young reservoirs to change more quickly and has a buffering effect when burial rates are calculated via isotope mass balance (Berner, 1987, 2009).
Oxygen sources are the burial (B) of photosynthetically derived carbon and of pyrite sulfur (blue arrows in Fig. 2). Burial of these reduced species results in oxygenation of the surface environment. The buried species are eventually uplifted and weathered (W) or are returned to the surface via metamorphism and degassing (D), which represent oxygen sinks (red in Fig. 2). The source-sink balance for O2 is:where G is organic carbon and PYR is buried pyrite.
Models calculate these fluxes, informed by internal parameters such as temperature, rates of erosion and degassing, rock exposure, and biological processes (Berner, 2006; Bergman et al., 2004). Burial, weathering, and degassing of the oxidized forms of carbon and sulfur (black arrows in Fig. 2) do not impact oxygen concentration directly but do affect the size and isotopic composition of the surface reservoirs (A, S), so cannot be ignored.
Isotope Mass Balance
Organic carbon and pyrite sulfur are isotopically lighter than the CO2 and SO4 they are derived from, due to kinetic selection during photosynthesis and sulfate reduction. The canonical isotope ratios for the present-day system (Hayes et al., 1999) are shown in Figure 2, alongside the fractionation effects ΔC (carbon) and ΔS (sulfur). These isotopic compositions and fractionation effects have changed over Earth history. For example, increasing the burial rate of organic, isotopically depleted carbon would act to increase the isotope ratio δ13C in the parent surface reservoir A.
Assuming that buried carbonates and sulfates reflect ancient oceanic isotopic composition, the geological δ13C and δ34S records can be used to back-calculate the required rate of burial of organic carbon and pyrite sulfur and therefore the rate of oxygen production (Berner, 1987). This requires knowledge of the input fluxes via weathering and degassing (W, D), the isotopic composition of the crustal reservoirs, and the fractionation effects ΔC and ΔS. The isotope mass balance equations consider isotopic inputs and outputs and are rearranged to calculate burial rates. The mass balance for the carbon system is shown below, and the sulfur system follows the same structure. See Berner (1987) and Berner (2001) for detailed descriptions. Here δ(X) is the isotopic composition of reservoir X:
The GEOCARBSULF model (Berner, 2006, 2009; Fig. 1A) combines the isotope mass balance technique with calculations for biogeochemical carbon and sulfur fluxes, and is generally considered the current “best-guess” atmospheric O2 prediction. Error analysis of the GEOCARBSULF model (Royer et al., 2014) plots model predictions for variation in all input parameters and robustly predicts low Mesozoic O2. However, this study is hampered by high model failure rate (the model crashes when some inputs are changed significantly from their default values), allowing only minimal variation of the δ13C input [δ(A) in Equation 2], far from the uncertainty in global records.
The IMB-COPSE Model
We re-evaluate the oxygen predictions via isotope mass balance using the revised COPSE model (Mills et al., 2014). COPSE is a derivative of the GEOCARB models and uses many of the same calculations, but it differs from GEOCARBSULF in several ways that make it potentially more useful for evaluating O2 predictions: the model is solved numerically using an implicit variable order method (Shampine and Reichelt, 1997), which greatly reduces model failure rate and allows testing of different δ13C inputs. The model also integrates recent research on the global rate of CO2 degassing and the weathering of volcanic rocks (Van Der Meer et al., 2014; Mills et al., 2014), which has not previously been applied to oxygen calculation.
The standard COPSE model includes nutrient cycles in order to estimate productivity and calculate the fluxes of organic carbon and pyrite sulfur burial. In this exploration (IMB-COPSE), the nutrient cycles are removed and the productivity and burial calculations are replaced with the standard isotope mass balance equations (Berner, 2001; Equation 2), following their incorporation into GEOCARBSULF (Berner, 2006). This includes the addition of rapid recycling. See the GSA Data Repository1 for full model description.
MODEL INPUTS AND RESULTS
Initially the IMB-COPSE model is run using the GEOCARBSULF δ13C and δ34S inputs (red dashed line in Fig. 3A). Despite the differences in model weathering and degassing processes, the IMB-COPSE model produces O2 predictions that are strikingly similar to those of GEOCARBSULF (black and red lines in Fig. 3B). This includes a prolonged period of low atmospheric O2 during the Jurassic and Early Cretaceous.
We explore model sensitivity to assumed δ13C and δ34S records by removing the GEOCARBSULF inputs and replacing them with values from recent literature compilations. Current records for Phanerozoic δ34S (Algeo et al., 2015) do not differ greatly from the GEOCARBSULF inputs, and their substitution has little impact on model predictions (see the Data Repository). Recent compilations of carbonate δ13C, however, show notable differences from the curves used in GEOCARBSULF.
The δ13C compilation of Saltzman and Thomas (2012; denoted GTS2012) is shown in blue in Figure 3A. The solid line shows the moving average, and dashed lines show ±1σ over 5 m.y. bins. The Mesozoic record is at higher resolution than the GEOCARBSULF input but does not show substantial base-level differences. The GTS2012 curve incorporates recent Cenozoic data from benthic foraminifera (Cramer et al., 2009), which agrees with the record of bulk sediment δ13C (Katz et al., 2005) in giving a present-day oceanic δ13C value close to 0‰, whereas the GEOCARBSULF curve has a present-day value closer to 2‰. This value is extremely important in isotope mass balance modeling as it sets the relative state of the system as we explore ancient time periods. During the Jurassic period, the GEOCARBSULF curve assumes a global ocean δ13C signature that is isotopically lighter than at present, potentially indicative of lower organic carbon burial and less oxygen production. The GTS2012 curve, however, shows a generally heavier signal than at present.
These differences in assumed oceanic δ13C translate into large differences in model O2 predictions, which tend to follow this input qualitatively. Under the GTS2012 input, the average predicted O2 concentration (blue in Fig. 3B) remains at or above present values during the past 200 m.y., resolving the conflict with nutrient- and weathering-based models and with the wildfire minimum. Note that the O2 predictions for the ±1σ δ13C inputs cross each other due to the present-day O2 constraint. See the Data Repository for further model uncertainty estimates, including constraints on the sulfur cycle.
Examination of IMB modeling confirms that the predicted rate of burial of organic carbon (the largest source of O2) is heavily dependent on the assumed carbon isotope record and much less dependent on other model processes, meaning that assumptions about the variation in oceanic δ13C are critical in determining pO2.
Compiling the global record of average whole-ocean δ13C is difficult, as differences exist across species, depth, and temperature (Saltzman and Thomas, 2012; Cramer et al., 2009). The paleogeographic source of information at different times adds further uncertainty: in sediments older than the Early Cretaceous the majority of records are sourced from epeiric seas rather than open-ocean margins or the deep ocean. A variety of studies have shown that ancient epeiric-sea water masses could develop isotopic signatures distinct from those of the open ocean for a number of isotope systems including carbon (Coulson et al., 2011; Panchuk et al., 2006; Newton et al., 2011).
These sources of uncertainty and variability lead to significant uncertainty in the overall curve, and crucially, in whether the present-day value is lower or higher than average values over the Mesozoic and Cenozoic. Current GEOCARBSULF model predictions of low Mesozoic pO2 rely on the assumption that modern oceanic δ13C values are higher than those during the Mesozoic, which is not shown in recent records based on either bulk-rock (Katz et al., 2005) or single-organism (Cramer et al., 2009) compilations. We therefore conclude that there is no model-based support for low Mesozoic O2 concentrations.
Taking our results together with the forward modeling approaches that calculate oxygen production via weathering and nutrient systems (Fig. 1), we argue for high O2 during the Mesozoic and Cenozoic, with a rise to above-modern oxygen concentrations during the Cretaceous. This view is compatible with the limits of combustion (Belcher and McElwain, 2008). Low-oxygen predictions are not a necessary consequence of isotope mass balance modeling, while estimations based on the δ13C composition of plant material (Tappert et al., 2013) are extremely difficult to validate due to the high variability of measured values and absence of controlled growth experiments in different CO2:O2 ratios.
Linking variations in oxygen concentration to animal evolution is speculative, and it is difficult to separate ecological and climatic drivers (Smith et al., 2010; Clapham and Karr, 2012). Proxies for O2 based on plant flammability are useful, but must be expanded to consider linkages between pO2 and fire-adapted trait selection (e.g., He et al., 2012, 2015; Lamont and Downes, 2011). Reconstructing atmospheric oxygen via modeling studies depends greatly on the ability to accurately compile average, whole-ocean δ13C for the ancient past, whether this record is used to directly drive the model (IMB) or as a means of comparison to model outputs. It is clear that modelers and paleontologists should seek to work together if we are to better explore the links between oxygen and evolution.
We thank the late R.A. Berner for the GEOCARBSULF model code, and M.R. Saltzman and E. Thomas for isotope data. Mills is funded by a University of Leeds Academic Fellowship; Belcher acknowledges funding from European Research Council Starter Grant ERC-2013-StG-335891-ECOFLAM. Lenton acknowledges funding from the Leverhulme Trust (RPG-2013-106). We thank Noah Planavsky and two anonymous reviewers for assessing this work.
- Received 14 June 2016.
- Revision received 4 October 2016.
- Accepted 4 October 2016.
- © 2016 Geological Society of America