Paleobotanical Evidence for Near Present-Day Levels of Atmospheric CO2 During Part of the Tertiary

See allHide authors and affiliations

Science  22 Jun 2001:
Vol. 292, Issue 5525, pp. 2310-2313
DOI: 10.1126/science.292.5525.2310


Understanding the link between the greenhouse gas carbon dioxide (CO2) and Earth's temperature underpins much of paleoclimatology and our predictions of future global warming. Here, we use the inverse relationship between leaf stomatal indices and the partial pressure of CO2 in modern Ginkgo bilobaand Metasequoia glyptostroboides to develop a CO2 reconstruction based on fossil Ginkgo andMetasequoia cuticles for the middle Paleocene to early Eocene and middle Miocene. Our reconstruction indicates that CO2 remained between 300 and 450 parts per million by volume for these intervals with the exception of a single high estimate near the Paleocene/Eocene boundary. These results suggest that factors in addition to CO2 are required to explain these past intervals of global warmth.

Atmospheric CO2 concentration and temperature have been tightly correlated for the past four Pleistocene glacial-interglacial cycles (1). Various paleo-CO2 proxy data (2, 3) and long-term geochemical carbon cycle models (4–6) also suggest that CO2-temperature coupling has, in general, been maintained for the entire Phanerozoic (7). Recent CO2 proxy data, however, indicate low CO2 values during the mid-Miocene thermal maximum (8,9), and results for the middle Paleocene to early Eocene, another interval of known global warmth relative to today, are not consistent, ranging from ∼300 to 3000 parts per million by volume (ppmv) (2, 9). Here, we address this problem by developing and applying an alternative CO2 proxy based on the inverse correlation between the partial pressure of atmospheric CO2and leaf stomatal index (SI), with the aim of reconstructing CO2 for both intervals to determine its role in regulating global climate.

Most modern vascular C3 plants show an inverse relationship between the partial pressure of atmospheric CO2 and SI (10–12), a likely response for maximizing water-use efficiency (10). SI is calculated as: SI (%) = [SD/(SD + ED)] × 100, where a stoma is defined as the stomatal pore and two flanking guard cells, SD = stomatal density, andED = non-stomatal epidermal cell density. Since SI normalizes for leaf expansion, it is largely independent of plant water stress, and is primarily a function of CO2 (10,12). This plant-atmosphere response therefore provides a reliable paleobotanical approach for estimating paleo-CO2 levels from SI measurements on Quaternary (13) and pre-Quaternary fossil leaves (14). Because stomatal responses to CO2 are generally species-specific (12), one is limited in paleo-reconstructions to species that are present both in the fossil record and living today. Fossils morphologically similar to living Ginkgo biloba and Metasequoia glyptostroboides extend back to the Early and Late Cretaceous, respectively, and many workers consider the living and fossil forms conspecific (15, 16). In this study, we use G. adiantoides and M. occidentalis, the forms most closely resembling G. biloba and M. glyptostroboides, and also G. gardneri, which has more prominent papillae and less sinuous upper epidermal cells than G. biloba(16).

Measurements of SI made on fossil Ginkgo andMetasequoia were calibrated with historical collections ofG. biloba and M. glyptostroboides leaves from sites that developed during the anthropogenically driven CO2 increase of the past 145 years and with saplings ofG. biloba and M. glyptostroboides grown in CO2-controlled greenhouses (17). These data show a strong linear reduction in SI for both species between 288 and 369 ppmv CO2 and a nonlinear response at CO2 concentrations above 370 ppmv (Fig. 1). Because SI responds to partial pressure, not concentration (11), the effects of elevation must be considered. All of the leaves measured for the training set grew at elevations <250 m where concentration ≅ partial pressure, so a correction is not needed. Both nonlinear regressions are highly significant (Fig. 1); however, a discontinuity exists for Ginkgo between the experimental results above 350 ppmv and the rest of the calibration set. Many species require more than one growing season for SI to adapt to high CO2 (12), and so these experimental results likely represent maxima for a given CO2 level. Nevertheless, due to this discontinuity as well as the small sample size and decreased sensitivity at high CO2 for both Ginkgo and Metasequoia, paleo-CO2 estimates >400 ppmv should be considered semi-quantitative.

Figure 1

Training sets for (A) Ginkgo biloba (n = 40) and (B)Metasequoia glyptostroboides (n = 18). Thick black lines represent regressions {Ginkgo:r 2 = 0.91, F(1,38) = 185,P < 0.001, SI = [CO2 – 194.4]/[(0.16784) ×CO2 – 41.6]; Metasequoia:r 2 = 0.85, F(1,16) = 41, P < 0.001, SI = [CO2 – 274.5]/[(0.12373) ×CO2 – 35.3]}. Gray lines represent ±95% prediction intervals. Inset graphs show the linear portions of both response curves in greater detail. Stomatal index determined from herbarium sheets (•), fresh samples from living trees (○), and 6- (▴) and 1-year-old (▵) saplings growing in CO2-controlled greenhouses. Error bars represent standard errors.

To reconstruct atmospheric CO2 changes, we measured the SI of fossil Ginkgo and Metasequoia cuticles from 24 localities in western North America and one from the Isle of Mull (Scotland), and then calibrated these data against the modern training set (Fig. 1) using inverse regression (18). Although not tightly constrained, the paleoelevations for all of the sites were probably <1000 m. This elevation difference could increase our estimates of CO2 concentration by at most 10%, and so our conversion from partial pressure to concentration excludes any correction. Except for a single high CO2 value near the Paleocene/Eocene boundary, all of our reconstructed CO2concentrations lie between 300 and 450 ppmv (Fig. 2 and Table 1). These contrast with two otherGinkgo-based CO2 estimates for the late Paleocene and middle Miocene (19, 20) that are very high (4500 and 2100 ppmv, respectively).

Figure 2

Reconstruction of paleo-CO2 for the (A) middle Paleocene to early Eocene and (B) middle Miocene based on SI measurements from Ginkgo (•) and Metasequoia (▴) fossil cuticles. Errors represent ±95% confidence intervals.

Table 1

Summary of fossil data. n = number of leaves measured for calculation of SI. A = south-central Alberta (Canada), BHB = Bighorn Basin (Wyoming and Montana, United States), M = Isle of Mull (United Kingdom), I = north-central Idaho (United States). Dashes indicate that no analyses were performed BOM = bulk organic matter, δ13Com = δ13C of organic matter. See (39) for δ13C methodology.

View this table:

We have Metasequoia-derived CO2 estimates only from the warm interval of the middle Miocene, but these are similar to coeval estimates derived from Ginkgo cuticles. The convergence of these two independent estimates increases our confidence that both species are reliably recording paleoatmospheric CO2 levels. In addition, the measured SI values from most sites fall well inside the region of high CO2 sensitivity in the training sets (Fig. 1 and Table 1), and the 95% confidence intervals (±50 ppmv or less) are over an order of magnitude lower than the errors associated with early Tertiary CO2estimates from geochemical models (4–6) and other proxies (2, 9). Furthermore, middle Paleocene to early Eocene CO2 reconstructions based on pedogenic carbonate (2, 21) and marine boron isotopes (9) show large changes in CO2 (≥2000 ppmv) over geologically brief periods of time [<1 million years] (Fig. 3) that cannot be readily explained. In contrast, the highly constrained error ranges and consistency among near time-equivalent estimates suggest that our SI-derived CO2 reconstruction is presently the most reliable, particularly for the middle Paleocene to early Eocene.

Figure 3

Estimates of paleo-CO2concentration derived from a variety of methods and their corresponding model-determined temperature departures (ΔT) of global mean surface temperature (GMST) from present day for the (A) middle Paleocene to early Eocene and (B) middle Miocene. Paleo-GMST calculated from paleo-CO2estimates using the CO2-temperature sensitivity study of Kothavala et al. (32). Present-day reference GMST calculated using the pre-industrial CO2 value of 280 ppmv (14.7°C). The error range of GMST predicted from the geochemical modeling-based CO2 predictions of (4) corresponds to the model's sensitivity analysis.

A period of rapid climatic warming (∼2°C global mean rise within 104 years that lasted 105 years) near the Paleocene/Eocene boundary has been extensively documented (22–25). Although the leading hypothesis for the cause of most of this warming is the rapid release of methane from marine gas hydrates and its subsequent oxidation to CO2 in the atmosphere and ocean (25, 26), all previous attempts to resolve this possible atmospheric CO2 spike have failed (23, 27, 28). Our single high CO2estimate is based on G. gardneri cuticle from Ardtun Head, Isle of Mull. Anomalously low δ13Com values (–30‰), an influx of the marine dinocyst Apectodinium, and a thermophyllic flora (including Caryapollenties veripites and Alnipollenites verus) occur in stratigraphically equivalent sediments elsewhere on Mull. Together, these indicate a possible correlation with a section of a Paris Basin borehole that has been calibrated to this event (29). At Ardtun Head, however, we failed to capture the negative carbon isotope excursion globally associated with this event, which ranges from 2.5‰ in the deep ocean (22, 25) to as much as 6‰ on land (23) (Table 1). Although the precise age of the Ardtun Head site remains uncertain, using a global carbon isotope mass balance model calibrated to Paleocene/Eocene conditions (30), our reconstructed CO2 increase (500 ppmv) is consistent with a release of 2522 Gt of methane-derived carbon, a value close to the estimate (2600 Gt C) calculated to account for the marine carbon isotopic excursion using methane as the carbon source (31).

Carbon dioxide is an important greenhouse gas, and its effect on global mean surface temperature (GMST) can be quantified with general circulation models (GCMs) [e.g., (32)]. Using the model output of Kothavala et al. (32) we predicted GMST from our CO2 results. The GCM used by Kothavala et al. is calibrated to the present day, which allows us to test the effect of CO2 on GMST independent of any paleogeographic or vegetational changes. With the exception of the single value near the Paleocene/Eocene boundary, all predictions lie within 1.5°C of the pre-industrial GMST (Fig. 3). These predictions contrast sharply with most paleoclimatic interpretations for these time intervals. For example, based on a synthesis of global late Paleocene and early Eocene δ18O-derived sea surface temperature data, Huber and Sloan (24) estimated that GMST was 3° to 4°C higher than today at this time, and δ18O-derived temperature estimates for the mid-Miocene thermal maximum [17 to 14.5 million years ago (Ma)] indicate that deep and high-latitude surface ocean temperatures were as much as 6°C warmer than today (33).

As a cross-check on our results, we compared our GMST predictions with those based on a geochemical carbon cycle model and other CO2 proxies for these same time periods. With the exception of the one CO2 estimate by Retallack (19), there is very good agreement among the methods for the middle Miocene (Fig. 3), strongly suggesting that factors in addition to CO2 are required to explain this brief warm period. In contrast, a large disagreement (10°C or greater) exists for the middle Paleocene to early Eocene (Fig. 3). This discrepancy is largely driven by the high CO2 estimates derived from marine boron isotopes (9); however, this proxy is probably less accurate than the other methods (3, 34). Nevertheless, even if the boron-based predictions are discounted, a large range still exists among the remaining three methods. This is striking considering that many of the pedogenic carbonate-derived CO2 estimates are based on the same sediments as our stomatal-based estimates (21); however, these estimates show a large temporal variability (–60 to 2040 ppmv) and are associated with relatively large error ranges (±500 ppmv). If our low SI-based temperature predictions are correct, additional factors such as paleogeography, enhanced meridional heat transport, and high latitude vegetation feedbacks are required to explain this warm period, and new constraints for CO2levels are established for middle Paleocene/early Eocene and middle Miocene GCMs [e.g., (35)]. Understanding the mechanisms of climate change will become increasingly important in the near future as atmospheric CO2 levels climb to levels perhaps unprecedented for the last 60 My.

  • * To whom correspondence should be addressed. E-mail: dana.royer{at}

  • Address as of 1 August 2001 will be Department of Animal and Plant Sciences, University of Sheffield, Sheffield S10 2TN, UK.


View Abstract

Stay Connected to Science

Navigate This Article