LETTERSPUBLISHED ONLINE: 1 AUGUST 2010 | DOI: 10.1038/NGEO923

Phanerozoic concentrations of atmosphericoxygen reconstructed from sedimentary charcoalIan J. Glasspool1* and Andrew C. Scott2

Variations of the Earth’s atmospheric oxygen concentration(pO2) are thought to be closely tied to the evolution oflife, with strong feedbacks between uni- and multicellular lifeand oxygen1,2. On the geologic timescale, pO2 is regulatedby the burial of organic carbon and sulphur, as well as byweathering3. Reconstructions of atmospheric O2 for the past400 million years have therefore been based on geochemicalmodels of carbon and sulphur cycling4–6. However, thesereconstructions vary widely4–10, particularly for the Mesozoicand early Cenozoic eras. Here we show that the abundanceof charcoal in mire settings is controlled by pO2, and usethis proxy to reconstruct the concentration of atmosphericoxygen for the past 400 million years. We estimate that pO2

was continuously above 26% during the Carboniferous andPermian periods, and that it declined abruptly around the timeof the Permian–Triassic mass extinction. During the Triassicand Jurassic periods, pO2 fluctuated cyclically, with amplitudesup to 10% and a frequency of 20–30 million years. Atmosphericoxygen concentrations have declined steadily from the middleof the Cretaceous period to present-day values of about 21%.We conclude, however, that variation in pO2 was not the maindriver of the loss of faunal diversity during the Permo–Triassicand Triassic–Jurassic mass extinction events.

Fire is an exothermic oxidation reaction dependent on the rapidcombination of fuel and oxygen in the presence of heat11. Charcoalis a by-product of wildfire and is first documented in the latestSilurian12,13, and has subsequently been recorded in all geologicalperiods from a range of sedimentary settings14. Calculation of fuelflammability at varying oxygen concentrations enables past pO2 tobe constrained within the range 15–35% (‘fire window’) whenevercharcoal is recovered from the fossil record11,15,16. To extrapolatefurther and to predict a pO2 curve from the geological abundanceof charcoal, both sources of heat and fuel must be decoupledfrom this relationship.

Lightning is the pre-eminent source of heat for the ignitionof fossil wildfires17. Owing to its global frequency (44 ± 5strikes/second18) and proven ‘fossil’ record as fulgurites19, it isunlikely that a lack of lightning could ever have limited fireinitiation. However, through deep time, fuel availability willhave affected the frequency of ignition. All terrestrial vegetationis potential fuel. As recorded by the occurrence of charcoalthroughout deep time, a source of fuel has been present for at least419million years12. However, in the fossil record the distributionof biomass has varied both spatially and temporally. Peat-formingenvironments are by definition regions of biomass accumulation. Inthis environment, an absence of fire ignition cannot be attributed toan absence of vegetation.

However, the presence of vegetation need not equate to thepresence of fuel, as vegetation may not be combustible under the

1Department of Geology, Field Museum of Natural History, 1400 S. Lake Shore Drive, Chicago, Illinois 60605, USA, 2Department of Earth Sciences, RoyalHolloway University of London, Egham TW20 0EX, UK. *e-mail: iglasspool@fieldmuseum.org.

prevailing environmental conditions. Vegetation is heterogeneousin composition, where the most important heterogeneity affectingpotential flammability is moisture content20. For ignition, fuel mustbe heated sufficiently to drive-off moisture and to liberate volatilesthat can be oxidized to generate a self-supporting exothermicpyrolytic reaction (that is, fire). The greater themoisture content theless flammable fuel becomes2,20. Peat-forming environments wereselected for this study as their formation requires that ‘groundwatermust remain throughout the whole year, above or close to theground surface’21. Therefore, these are high-moisture environmentsthat best minimize the potential role of climatic variation onthe accumulation of charcoal deposits through deep time. Thisapproach is confirmed by data from a range of modern peatsrepresenting divergent ecological settings and vegetation types(Supplementary Table S1). In these peats (modern–Pleistocene, 21seams/81 samples), which include environments such as the FloridaEverglades and Kalimantan, wheremodern fires are frequently doc-umented, charcoal abundance is consistently low with no samplecontaining more than 11.1% (average 4.3%). Even extrapolating tothe past 50million years, average charcoal abundance is still lowcompared with earlier deposits. Of 139 seams (759 samples) fromthis interval, only one contains >20% charcoal (26.5%: average4.8%), 13 contain none and 23 contain <1.0%. Throughout thisinterval, global climate and weather have fluctuated markedly,whereas charcoal abundance demonstrably has not. Therefore,it can be argued that, by examining charcoal abundance inpeat-forming environments, the impact of moisture content on fuelignition, although not eliminated, is greatly restricted.

In the fossil record, peats are preserved as lignites or coals21 andbecause of their economic importance are routinely studied by op-tical reflectance microscopy. Using this technique, the organic con-stituents are described in terms of macerals21. One maceral group(inertinite) is almost exclusively considered the by-product of wild-fires and is synonymous with charcoal22. The percentage (by vol-ume) of inertinite is commonly reported and provides an extensiverecord of charcoal abundance23. A database of percentage inertiniteabundance in coal (Inert%), calculated on a mineral-matter-freebasis of all ages and geographic distribution, has been compiledfrom the literature (Fig. 1a, Supplementary Table S2). Data arebinned to 10-million-year intervals (Supplementary Table S3) tomatch existing geochemical models for pO2 (refs 5–7). The conceptof comparing inertinite with pO2 is not novel, plots of fusinite andsemifusinite, two inertinite group macerals, having been made24.However, this study did not include data on other, often abundant,inertinitemacerals andmade no attempt to predict pO2.

To calculate pO2 from Inert%, the generation of a calibrationcurve is required (Fig. 1b). From experiments, wildfire is unsustain-able at levels of pO2≤ 15% (ref. 16), and therefore charcoal will notbe generated (Inert%= 0%). However, as zero values preclude the

NATURE GEOSCIENCE | VOL 3 | SEPTEMBER 2010 | www.nature.com/naturegeoscience 627© 2010 Macmillan Publishers Limited. All rights reserved.

LETTERS NATURE GEOSCIENCE DOI: 10.1038/NGEO923

d

Iner

tinite

%

a

c

456

789

10

(1)(2)

(5)

(3)

(4)

b

1/1

5/8

3/26 7/

93/

108/

18

383/

1,196

30/3

6

27/1

3564

/14

411

8/54

74

7/91

45/

8725

/85

1/1

31/3

122

/48

5/5

25/2

99/

3512

/12

41/

90

34/9

152

/24

4

9/4

7

4/6

817

/74

1/4

9/77

70/2

46

18/5

312

/53

62/2

16

19/2

017

/220

42/

280

31/1

33

19/8

5

300 200 100

Age (Myr)

400 0

100

50

0

30

25

20

pO2

(%)

35

1510 20 30 605040

Inert%

0 70

30

20

10

pO2

(%)

40

0300 200 100

Age (Myr)400 0

300 200 100Age (Myr)

400 0

30

20

10

pO2

(%)

40

0

Figure 1 | Phanerozoic inertinite distribution and predictions of pO2. a, Inertinite abundance. Line; bin mean. Error bars 1 s.d. from mean. Lower axis;seams/samples per bin. b, Power-law regressions for conversion of Inert% to pO2. Labels refer to equations in main text. c, Biogeochemical predictions ofPhanerozoic pO2 published to 2009 (refs 4–10). Lines: publication references in key. d, Prediction of pO2 from Inert%. Line; best estimate based on latePalaeozoic pO2 maxima of 30%. Error bars 1 s.d. from mean. Shaded area; estimate of maximum error assuming Phanerozoic pO2 maxima of 35%+ 1 s.d.(upper margin) and 25%− 1 s.d. (lower margin).

application ofmost nonlinear trendlines, 0.2% inertinite is adopted,based on typical coal petrographic counts of ≥500 points, formaceral analyses21 (that is one charcoal particle= 0.2%). ModernpO2 is 21% and, as discussed above, based on a range of ecolog-ically, climatically and geographically differing peats, is calculatedto equate to 4.3% inertinite. These two points can be assignedwith some confidence. For calibration, Inert% data at the upperlimit of Phanerozoic pO2 are required, but prediction of valuesat above modern levels must be subjective. However, combustion

research suggests theoretical upper levels of pO2 from 25% (ref. 2)to 35% (refs 2,11). Previous research indicates that in the latePalaeozoic pO2 exceeded 25% (ref. 15), but owing to increasedplant flammability was less than 35% (refs 2,11). Inert% predictionsuse the mean of these values for the late Palaeozoic (pO2 = 30%).Although a significant assumption, this value is in accord with otherpredictions of the theoretical Phanerozoicmaximum25.

Independent of fire, mass balance, biogeochemical and carbonisotopic fractionation models predict maximal Phanerozoic pO2

628 NATURE GEOSCIENCE | VOL 3 | SEPTEMBER 2010 | www.nature.com/naturegeoscience

© 2010 Macmillan Publishers Limited. All rights reserved.

NATURE GEOSCIENCE DOI: 10.1038/NGEO923 LETTERSduring the Permian at ∼30–35% (refs 5,7,26; Fig. 1c). The timingof these maximal pO2 data corresponds with the timing of maximalinertinite abundance (that is Inert% = 44.4% (n = 547) in theEarly Permian (280Myr)). On this basis we suggest an Inert% of44.4% can be correlated to 30% pO2 (derived from theoreticalcalculations25). This datum is used as the uppermost value necessaryto enable calibration of inertinite abundance against pO2.

From these estimations, calibration of pO2 from Inert% is basedon: 15% pO2 = 0.2% inertinite, 21% pO2 = 4.3% inertinite and30% pO2 = 44.4% inertinite. From these data, fire and pO2 arebest described by a power relationship, this regression yieldingthe highest R2 values (30% pO2 curve: 0.991). However, by usingnonlinear regression a diminished ability to predict variations inpO2 above 21%exists. The equation for the calibration curve is:

pO2= 18.113x(Inert%0.1273) or y = c∗(x∧b) (1)

where:

c = EXP(INDEX(LINEST(LN(y),LN(x),,),1,2))

b= INDEX(LINEST(LN(y),LN(x),,),1)

The error in the mean inertinite data is plotted to one standarddeviation (s.d.) (Fig. 1a). This error is fed-back into the calibrationto generate two further error curves for the 30% pO2 assessment(Fig. 1b, Supplementary Fig. S1).

pO2= 19.293x(Inert%0.1376) or y = c∗(x∧b) (2)

pO2= 17.723x(Inert%0.1185) or y = c∗(x∧b) (3)

Error bars in Fig. 1d (Supplementary Table S4) represent theimpact of these curves on pO2 estimation. In addition, to providea crude assessment of the maximum variation in pO2 predictions,assuming late Palaeozoic pO2 highs had fallen in the range 25% or35%, two further calibration curves are plotted (Fig. 1b, shading inFig. 1d). The first is based on assumptions of a late Palaeozoic pO2high of 25% calibrated against Inert%− 1 s.d.

pO2= 18.633x(Inert%0.0985) or y = c∗(x∧b) (4)

where: late Palaeozoic pO2= 25%.The second is based on a late Palaeozoic pO2 high of 35%

calibrated against Inert%+ 1 s.d.

pO2= 17.998x(Inert%0.1427) or y = c∗(x∧b) (5)

where: late Palaeozoic pO2= 35%.As with previous predictions5,9 (Fig. 1c), our data indicate a

marked rise in pO2 within a 30-million-year interval following thegrowth of the Middle–Late Devonian forests. Following the EarlyMississippian, pO2 remained at or above 26% until the Mesozoic,data that are in accord with predictions based on floral carbonisotopic fractionation26. The apparent Pennsylvanian decline isconsidered an artefact resulting from the uniquemorphology of thedominant arborescent lycopsid coal-forming vegetation and theirecology. These lycopsids were morphologically distinct from plantsthat subsequently dominated this setting and were both highlyfire resistant27 and grew predominantly in rheotrophic settings(that is, mires with flowing water) without significant litter layers.Therefore, fires were probably constrained to the canopy andproduced comparatively little charcoal.

During the Early Permian pO2 reached a Phanerozoic maximumof ∼29%. Inert% predicts a bimodal pO2 distribution in the

Permian, similar to previous modelling9. Although these datasuggest a Middle Permian decline in pO2 they do not indicatehypoxia as a contributing factor in the end Guadalupian(∼260Myr) mass-extinction event28. Similarly, examination ofChanghsingian (253.8–251Myr) age coals indicates abundantcharcoal within the last 2.8million years of the Permian(average 38.9%; Supplementary Table S2). Therefore, irrespectiveof the Changhsingian onset of the Early Triassic superanoxicevent29, pO2 must have remained elevated until at least the veryend of the Permian. It is noteworthy that in using 10-million-yearbinning, Late Permian data plot at 250Myr immediately followingthe Permo–Triassic mass-extinction.

During theMesozoic, Inert% predicts greater pO2 volatility thanother models5,9 and indicates almost cyclic fluctuations until theLate Cretaceous. As early as the Middle Triassic, pO2 is indicatedto have declined to its lowest level throughout the Phanerozoic(∼18.5%). However, the timing of this event is obscure, as no EarlyTriassic coals exist23, and although the decline in this interval mayhave been even more abrupt and extreme, no data exist to testthis hypothesis either.

Among the predicted fluctuations, it is worth highlightingelevated pO2 through the Triassic–Jurassic mass-extinction, anabrupt decline during the Toarcian (183–175.6Myr) and elevatedlevels from the late Middle Jurassic into the earliest Cretaceous.Analysis of Mist Mountain Formation (150.8–140.2Myr) coals(n= 56) shows those in the basal 250m contain more than doublethe inertinite (45.5%) of those above (23.0%) (SupplementaryTable S5). Inert% data from unequivocal Early Cretaceous (145.5 to∼121Myr) coals are sparse, but those from the Bickford Formation(n= 37, Supplementary Table S6) in particular suggest depressedpO2 immediately following the Jurassic–Cretaceous boundary, withlevels rising rapidly thereafter.

Late Early Cretaceous coals (late Aptian (∼121–112Myr))(Supplementary Table S2) indicate rising pO2, a trend thatcontinued through the Albian–Cenomanian (112–93.5Myr), whenaMesozoic maximum of 29% is predicted. This was the last intervalwhen pO2 was at such elevated levels. It is noteworthy that manycharred flowers have been recovered from this interval, and it hasbeen speculated that fire, as an agent of environmental disturbance,played a role in the early evolution of angiosperms30.

From the earliest Late Cretaceous, a gradual stepped decline inpO2 to ∼24% 50Myr is predicted. This fundamental petrographicdifference between Cenozoic and older coals has been recognizedpreviously21, but no explanation has been given. Over the past40million years pO2 seems to have been relatively constant at∼21–22%, roughly equivalent tomodern levels.

Charcoal abundance in mire settings documents the rise ofpO2 following the diversification of early land plants and itssubsequent fluctuations throughout the Phanerozoic. From thebest estimate curve, these fluctuations are more frequent thanpreviously predicted, particularly during the Triassic and Jurassic,when 20–30million year cyclicity is detected. These data suggest animportant role for fire as a major Earth system process13 but alsoindicate that the governing feedbacks are not yet fully understood.Although large amplitude shifts are calculated, fire activity indicatesthat low pO2 events did not drive catastrophic faunal diversity loss ineither the Permo–Triassic or the Triassic–Jurassic mass-extinctionevents. Instead, charcoal indicates that pO2 remained elevated untilat least the latest Permian, and declined precipitously only in theEarly Triassic. Similarly, earliest Jurassic charcoal data indicatethat pO2 was not low but rather was elevated to levels among thehighest of the Phanerozoic.

MethodsFor this study, two coal samples were collected and analysed for maceralcomposition. The L’Anse-à-Brillant coal from Tar Point, Gaspé, Canada is of

NATURE GEOSCIENCE | VOL 3 | SEPTEMBER 2010 | www.nature.com/naturegeoscience 629© 2010 Macmillan Publishers Limited. All rights reserved.

LETTERS NATURE GEOSCIENCE DOI: 10.1038/NGEO923

Emsian age. The South Tancrediakløft coal from Jamesonland, Greenland is ofvery basal Hettangian age. The coals were embedded in polyester resin and countedfollowing established protocols20. Maceral counts were assessed from 500 points ofdata, including mineral matter.

Maceral data from the literature, used to determine Inert%, were includedin this analysis only where the inclusion/exclusion of mineral matter was clear.All data were standardized to a mineral-matter-free basis. With three exceptions,coals whose stratigraphic resolution was greater than 15million years wereexcluded (for example, Taiyuan formation = Kasimovian–Sakmarian). Thethree samples included in the database derive from poorly sampled stratigraphicintervals where they represent the only data: Givetian–Frasnian (Weatherall–HeclaBay–Beverley Inlet formations) and the Anisian-Carnian (Basin Creek andMungaroo formations).

Where not tabulated or stated in the text, data were measured fromgraphics by pasting the image into CorelDRAW and overlaying guidelines toobtain exact measurements of data point positions (data are highlighted inSupplementary Table S2). Preference was given to literature citing named seams.Where multiple references provide data from one seam, these data were averagedand all references cited.

Received 10 February 2010; accepted 1 July 2010; published online1 August 2010

References1. Flück, M. et al. Coping with cyclic oxygen availability: Evolutionary aspects.

Integr. Comp. Biol. 47, 324–331 (2007).2. Watson, A., Lovelock, J. E. & Margulis, L. Methanogenesis, fires and the

regulation of atmospheric oxygen. Biosystems 10, 293–298 (1978).3. Kump, L. R. Terrestrial feedback in atmospheric oxygen regulation by fire and

phosphorous. Nature 335, 152–154 (1988).4. Falkowski, P. G. et al. The rise of oxygen over the past 205 million years and

the evolution of large placental mammals. Science 309, 2202–2204 (2005).5. Berner, R. A. GEOCARBSULF: A combined model for Phanerozoic

atmospheric O2 and CO2. Geochim. Cosmochim. 70, 5653–5664 (2006).6. Berner, R. A. Phanerozoic atmospheric oxygen: New results using the

GEOCARBSULF model. Am. J. Sci. 309, 603–606 (2009).7. Berner, R. A. & Canfield, D. E. A new model for atmospheric oxygen over

Phanerozoic time. Am. J. Sci. 289, 333–361 (1989).8. Hansen, K. W. &Wallmann, K. Cretaceous and Cenozoic evolution of seawater

composition, atmospheric O2 and CO2: A model perspective. Am. J. Sci. 303,94–148 (2003).

9. Bergman, N. M., Lenton, T. M. & Watson, A. J. COPSE: A new model ofbiogeochemical cycling over Phanerozoic time. Am. J. Sci. 304, 397–437 (2004).

10. Arvidson, R. S. & Mackenzie, F. T. MAGic: A Phanerozoic model for thegeochemical cycling of major rock-forming components. Am. J. Sci. 306,155–190 (2006).

11. Jones, T. P. & Chaloner, W. G. Fossil charcoal, its recognition andpaleoatmospheric significance. Palaeogeogr. Palaeoclimatol. Palaeoecol.97, 39–50 (1991).

12. Glasspool, I. J., Edwards, D. & Axe, L. Charcoal in the Silurian as evidence forthe earliest wildfire. Geology 32, 381–383 (2004).

13. Bowman, D. M. J. S. et al. Fire in the Earth system. Science 324, 481–484 (2009).14. Scott, A. C. The pre-quaternary history of fire.

Palaeogeogr. Palaeoclimatol. Palaeoecol. 164, 281–329 (2000).15. Wildman, R. A. et al. Burning of forest materials under late Paleozoic high

atmospheric oxygen levels. Geology 32, 457–460 (2004).16. Belcher, C. M. & McElwain, J. C. Limits for combustion in low O2 redefine

paleoatmospheric predictions for theMesozoic. Science 321, 1197–1200 (2008).17. Cope, M. J. & Chaloner, W. G. Fossil charcoal as evidence of past atmospheric

composition. Nature 283, 647–649 (1980).18. Christian, H. J. et al. Global frequency and distribution of lightning as

observed from space by the optical transient detector. J. Geophys. Res. 108,4005 (2003).

19. Scott, A. C. & Jones, T. P. Fossil charcoal: A plant-fossil record preserved by fire.Geol. Today 7, 214–216 (1991).

20. Whelan, R. J. The Ecology of Fire (Cambridge Univ. Press, 1995).21. Taylor, G. H. et al. Organic Petrology (Gebruder Borntraeger, 1998).22. Scott, A. C. & Glasspool, I. J. Observations and experiments on the origin and

formation of inertinite group macerals. Int. J. Coal Geol. 70, 53–66 (2007).23. Diessel, C. F. K. The stratigraphic distribution of inertinite. Int. J. Coal Geol.

81, 251–268 (2010).24. Robinson, J. M. Phanerozoic atmospheric reconstructions: A terrestrial

perspective. Palaeogeogr. Palaeoclimatol. Palaeoecol. 97, 51–62 (1991).25. Lenton, T. M. & Watson, A. J. Redfield revisited: 2. What regulates the oxygen

content of the atmosphere. Glob. Biogeochem. Cycles 14, 249–268 (2000).26. Beerling, D. J. et al. Carbon isotope evidence implying high O2/CO2 ratios

in the Permo-Carboniferous atmosphere. Geochim. Cosmochim. Acta 66,3757–3767 (2002).

27. Robinson, J. M. Phanerozoic O2 variation, fire, and terrestrial ecology.Palaeogeogr. Palaeoclimatol. Palaeoecol. 75, 223–240 (1989).

28. Retallack, G. J. et al. Middle-Late Permian mass extinction on land. Geol. Soc.Am. Bull. 118, 1398–1411 (2006).

29. Isozaki, Y. Permo–Triassic boundary superanoxia and stratified superocean:Records from lost deep sea. Science 276, 235–238 (1997).

30. Bond, W. J. The tortoise and the hare: Ecology of angiosperm dominance andgymnosperm persistence. Biol. J. Linn. Soc. 36, 227–249 (1989).

AcknowledgementsWe thank the Grainger Fund at FMNH for financial support, and acknowledge R. A.Berner, W. G. Chaloner and A. J. Watson for helpful discussions. The work of A.C.S. isfinancially supported by private charitable donations.

Author contributionsI.J.G. and A.C.S. designed the research. I.J.G. gathered, compiled and interpreted dataandA.C.S. contributed data. I.J.G. wrote the paper with additions by A.C.S.

Additional informationThe authors declare no competing financial interests. Supplementary informationaccompanies this paper on www.nature.com/naturegeoscience. Reprints and permissionsinformation is available online at http://npg.nature.com/reprintsandpermissions.Correspondence and requests formaterials should be addressed to I.J.G.

630 NATURE GEOSCIENCE | VOL 3 | SEPTEMBER 2010 | www.nature.com/naturegeoscience

© 2010 Macmillan Publishers Limited. All rights reserved.