Introduction
The fundamental role of CO2 as a driver of past warm climates in the Mesozoic and early Cenozoic periods remains obscured for a number of reasons. Different proxies (fossil leaf stomatal indices and palaeosols) yield widely disparate CO2 estimates for this critical phase in Earth's history3; in general, palaeosols5 suggest CO2 levels thousands of p.p.m.v. higher than fossil leaves12. Furthermore, it is unclear whether mismatches between oxygen-isotope-based (
18O) palaeoclimate records and ancient CO2 estimates1, either from proxies or numerical carbon-cycle models2, 3, 4, genuinely reflect decoupling between CO2 and climate or are an artefact associated with the difficulties of accurately reconstructing these critical features of Earth's history13. However, evidence for brief (typically<1 Myr) cool pulses punctuating generally warm Jurassic and Cretaceous climates is challenging the prevailing high CO2 Mesozoic 'greenhouse world' paradigm3 indicated by palaeosols5 and geochemical carbon-cycle models4.
Here, we report a series of CO2 estimates for the ancient atmosphere reaching back over 200 Myr, by using a new method based on the strong dependency of carbon-isotope fractionation (
13C) in terrestrial bryophytes on atmospheric CO2 during photosynthetic carbon uptake6, 7. The approach is the terrestrial analogue of the marine phytoplankton CO2 proxy14, successfully used to reconstruct glacial–interglacial CO2 changes from Pleistocene peat cores15. In functional terms, 13C approaches the maximum fractionation value (
30
) expressed by the primary carboxylating enzyme ribulose-1,5-carboxylase/oxygenase (RuBisCo), as the CO2 concentration at the site of fixation (Ci) approaches the external CO2 concentration (Ca) (refs 6,7). Because Ci is set by the combined influence of net photosynthetic CO2 uptake (A) and resistance to the inward diffusion of CO2 (r), 13C is proportional to Ci/Ca, where Ca-Ci=A
r. In bryophytes, r is not subject to stomatal control, as in the leaves of vascular plants, and 13C varies with Ca and the photosynthetic demand for CO2 by RuBisCo (refs 6,7).
We estimated ancient atmospheric CO2 concentrations on the basis of 93 13C measurements on 61 liverwort gametophyte compression fossils (Fig. 1a–d). The fossils spanned 150 Myr and were sampled from 12 localities on 5 continents (Table 1). Conversion of 13C measurements to palaeoatmospheric CO2 concentrations is achieved after calculating 13C to account for variations in the 13C of atmospheric CO2 (Table 1), and inverting a well-validated mechanistic mathematical model (BRYOCARB) describing the CO2 dependency of 13C in bryophytes7. All calculated fossil 13C values lie within the range sensitive to variations in CO2, except those for the Palaeocene (Ypresian) specimens (Fig. 1e, inset). BRYOCARB models 13C as a function of Ci/Ca, which is determined by the balance between photosynthetic CO2 draw-down and diffusional CO2 supply to the tissue from the atmosphere. Photosynthesis is represented with established biochemical theory for CO2 assimilation and explicitly accounts for the interactive effects of irradiance, O2 and temperature7 (see the Supplementary Information for derivation of values). The supply of CO2 from the atmosphere to the site of photosynthesis is modelled by analogy with Ohm's law7. BRYOCARBP calibrates 13C change in liverworts with pores, where r is regulated by varying pore density and morphology. However, some groups of liverworts lack pores and are unable to regulate r in this way. Because these are not always distinguishable as fossils, we used a second version of the model, BRYOCARBNP, to calibrate 13C changes in liverworts without pores, with a prescribed resistance and a lower maximum rate of RuBisCo-limited carboxylation (Vcmax; see the Supplementary Information)6. Uncertainties for each CO2 estimate are characterized by deriving probability density functions (PDFs) using large (25,000) ensemble Monte Carlo simulations to integrate uncertainties in inferred and measured input variables for both versions of BRYOCARB (see the Methods section).
Figure 1: Atmospheric CO2 concentrations from fossil liverworts.
a–d, Example liverwort compression fossils. All scale bars are 5 mm. e,f, Example PDFs of CO2 concentrations estimated from fossil liverwort 13C calibrated with BRYOCARBP (e) and BRYOCARBNP (f). Inset: Modelled response of 13C to CO2 for fossils a,c and d. g,h, Reconstructed CO2 histories using fossil liverwort 13C and BRYOCARBP (g) and BRYOCARBNP (h). The CO2 histories predicted by a geochemical model with formulations for standard18 (GEOSW), moderate20, 21 (GEOVW) and enhanced19(GEOEVW) volcanic weathering are also shown (with the latter two assuming very low rates of basalt seawater reaction). Squares indicate means, error bars and boxes, the 10–90% and 25–75% uncertainty ranges respectively. Tr: Triassic, J: Jurassic, K: Cretaceous, Pal: Palaeocene, N: Neogene.
1
) stable-carbon-isotope composition (
13C) and fractionation (
13C). Specimen numbers, species lists and provenances are given in ![Table 1 - Details of fossil liverworts and their mean (|[plusmn]|1|[sigma]|) stable-carbon-isotope composition (|[delta]|13C) and fractionation (|[Delta]|13C). Specimen numbers, species lists and provenances are given in Supplementary Information.](/common/images/table_thumb.gif)
Our new atmospheric CO2 reconstruction, spanning approximately one-third of the Phanerozoic eon (the past 540 Myr), exhibits coherent trends throughout the Mesozoic and early Cenozoic (Fig. 1g–h). These trends primarily reflect CO2-driven systematic shifts of up to 5 in fossil bryophyte 13C; other environmental inputs play a secondary role in determining the pattern (see the Supplementary Information). CO2 estimates derived from fossil bryophyte 13C calibrated using either BRYOCARBP or BRYOCARBNP both rise from comparatively low concentrations in the Triassic and Early Jurassic to a peak of 1,130 p.p.m.v. in the Middle Cretaceous, and then decline towards 680 p.p.m.v. in the early Cenozoic (Fig. 1g–h). Calibrated 13C changes using BRYOCARBNP lead to CO2 estimates approximately 60 p.p.m.v. higher than those using BRYOCARBP, as the increased resistance to inward CO2 diffusion meant that a higher Ca is required to reproduce the Ci/Ca inferred from fossil 13C. The only exception to the coherent trend is a singular high CO2 estimate during the Ypresian (48.6–55.8 Myr ago, 2,300–4,740 p.p.m.v.), which may reflect the reduced sensitivity of the bryophyte proxy at high CO2 and should therefore be regarded as less certain than the others. Although we recognize that it coincides with peak Cenozoic warmth, until better understood by investigation of further fossil materials, we have omitted this from our subsequent time-series analyses.
We used the bryophyte CO2 records to evaluate the role of this greenhouse gas in the evolution of Mesozoic and early Cenozoic warm climates by calculating CO2-forced temperatures for comparison with independent palaeoclimate records. Cross-correlation coefficients were computed between the resulting changes in mean global surface temperature (T) and a pH-adjusted tropical surface ocean temperature series based on the 18O of marine calcium carbonate fossils1, 16, after accounting for uncertainties in T due to imperfect knowledge of vital effects1, post-depositional alteration9 and the 18O of ocean water17 (see the Methods section). Other evidence supports the major pattern revealed by the pH-adjusted 18O record showing cooler Middle Jurassic (160–167 Myr ago)8 and warmer Cretaceous climates9, 10, suggesting that it is a useful metric for evaluating greenhouse forcing by calculated CO2 trends. Agreement between calculated and observed changes in T was quantified using a hierarchical curve reconstruction technique accommodating the irregular spacing of CO2/T determinations and uncertainties in their estimation and dating (see the Methods section). The same approach was also used to evaluate CO2 as a driver of climate change using an extensive compilation of CO2 estimates derived from studies of fossil leaves and palaeosols3.
Results indicate that the bryophyte and stomatal CO2 histories reproduce the average Mesozoic climate, yielding average global temperatures within 0.7 °C of that estimated from the pH-adjusted 18O climate record (Fig. 2a,b). In contrast, calculated temperatures with palaeosol CO2 histories were 2.0–2.5 °C higher than the marine record (Fig. 2b). Cross-correlations revealed that only T patterns calculated from the CO2 histories of fossil bryophytes were positively correlated with T from the marine 18O record (median coefficients=0.2–0.5; Fig. 2c,d). Neither the stomatal nor the palaeosol proxies produced positive correlations with independent records of climate (Fig. 2d). The capacity of the bryophyte proxy to describe both overall temperatures and the magnitude of climate change may be evidence for CO2 forcing of major climatic shifts, but this suggestion requires assessment of the reliability of each of the three proxies used to reconstruct ancient atmospheric CO2 concentrations.
Figure 2: Evaluation of CO2 forcing with reconstructed changes in temperature (
T).
![Figure 2 : Evaluation of CO2 forcing with reconstructed changes in temperature (|[Delta]|T).](/ngeo/journal/v1/n1/thumbs/ngeo.2007.29-f2.jpg)
a,b, Absolute mean difference in T between proxies and 18O records over the Mesozoic. Points represent the mean, the box the 25–75% range, and the horizontal lines across the box plot the median. Error bars represent the 10–90% range and short horizontal lines the extreme values. c,d, PDFs of the correlation between normalized T calculated from the proxy CO2 records, and 18O records. Squares indicate means, error bars and boxes, the 10–90% and 25–75% uncertainty ranges respectively.
We therefore benchmarked proxy performance against the predictions of numerical carbon-cycle models to identify those most consistent with known variations in the long-term sources and sinks for CO2. Comparisons were made with a series of three atmospheric CO2 histories simulated using a coupled carbon–oxygen–sulphur cycle geochemical model18 that incorporated different representations of basalt weathering to describe the range of field observations19, 20. Treatment of all other CO2 source and sink terms remained unchanged in each of the following three simulations: (1) standard weathering18 (GEOSW), (2) moderate weathering linked to variations in the production and exposure of volcanic rocks21 assuming unusually low values for basalt–seawater (BSW) reaction rate (GEOVW), and (3) as (2) but with enhanced weathering of basalts19 and very low BSW rates or with moderate weathering and reasonable BSW rates (GEOEVW) (see the Methods section). Proxy performance was defined with summary statistics for absolute differences in proxy–model CO2 values integrated over the Mesozoic and early Cenozoic, and correlation coefficients between proxy–model CO2 time series (see the Methods section).
These comparisons indicate that fossil bryophytes provide a more coherent record of fluctuations in Earth's atmospheric CO2 content over the Mesozoic and early Cenozoic than either of the other two existing proxies covering the same interval. Fossil bryophytes and the stomatal proxy estimated CO2 concentrations lower than all three model simulations, but this underestimation was progressively reduced from -750 p.p.m.v. to only -90 p.p.m.v. with increasing emphasis on the weathering of volcanic silicates (Fig. 3a–c). In contrast, the palaeosol proxy considerably overestimated CO2 values, increasing from +590 p.p.m.v. to +1,370 p.p.m.v. across the same series of simulations (Fig. 3d). The closest agreement between two out of three proxies with GEOEVW strengthens the case for variations in the production and exposure of basalts as playing a prominent role in regulating CO2 levels on a multimillion-year timescale19, 21. However, cross-correlation analyses indicate that only the bryophyte proxy was both a good fit and exhibited a positive correlation in the trends in CO2 simulated with GEOVW and GEOEVW (Fig. 3e–h). This suggests that the bryophyte proxy has accuracy and captures some patterns of the CO2 histories simulated by these two versions of the geochemical model. Neither of the other two proxies fulfilled both criteria simultaneously. The stomatal CO2 proxy record correlated best with GEOSW CO2 predictions, but this simulation gave the largest CO2 difference (Fig. 3c,g). The palaeosol proxy CO2 record was not strongly correlated with any of the three model simulations, and had CO2 concentrations different from model results (Fig. 3d,h).
Figure 3: Comparison of proxy and geochemical model CO2 estimates.
a–d, Difference in average CO2 concentration between models and proxies. e–h, PDFs of the correlation between normalized model and proxy CO2 records. Box plot and horizontal errors as for Fig. 2.
Full size image (75 KB)We conclude that CO2 forcing played an important role in Mesozoic and early Cenozoic climate change, and that earlier claims for a decoupling in the CO2–climate relationship during this critical phase in Earth's history1 are premature. Changes in the CO2 greenhouse effect, driven by geochemical controls on the long-term carbon cycle4, 21, thus contribute an explanation for Jurassic and Cretaceous climates without the need to invoke the influence of cosmic rays on cloud cover and planetary albedo22. Furthermore, our reconstructed CO2 concentrations (500–1,300 p.p.m.v., Fig. 1) coincide with the threshold for the initiation of glaciations determined by global climate modelling calibrated for mid-Cenozoic conditions11 (560–1,120 p.p.m.v.). Although the exact range is likely to be dependent on the particular climate model, CO2 histories derived from palaeosols5 and earlier geochemical carbon-cycle models4 exceed this threshold by several thousand p.p.m.v. Our new CO2 reconstructions therefore offer a resolution for the high CO2 'greenhouse world' paradox by better explaining the apparent susceptibility of the Earth system to experience brief discrete cool events during the Mesozoic3.
Methods
Uncertainty analysis of bryophyte CO2 estimates
PDFs characterizing uncertainty in CO2 determinations were estimated from 25,000 Monte Carlo simulations of the derivation based on alternative values of physiological (Vcmax, respiration rate, diffusion resistance and response of resistance to CO2) and environmental (temperature, O2 and irradiance) parameters (see the Supplementary Information) and 13C including errors in measuring/estimating 13Cp and 13Ca (Table 1). Values were sampled from either normal or truncated normal distributions with dispersion parameters obtained from the literature, measurement errors (Fig. 1e,f) and, in the case of diffusion resistance and its response to CO2, nonlinear regression fitting of data from modern specimens. The possibilities of uncertainty in dependence between physiological parameters, irradiance and 13C were allowed for. Variations in water content of field populations of liverworts exert minimal effects on 13C and are excluded from the analyses6, 7. The resulting uncertainty PDF for CO2 is interpreted in the Bayesian sense as a representation of rational belief.
Statistical comparisons of time series
To compare proxy CO2 estimates with model predictions, and to compare calculated and observed T, we used a hierarchical curve reconstruction technique that accommodates the irregular spacing of CO2 determinations and allows for uncertainty in both their estimation and dates. At the first stage, synthetic sets of CO2–age/T–age values were generated from uncertainty distributions derived here or on the basis of reports in the literature. For CO2 estimates based on the bryophyte, stomata and palaeosol proxies, and models, we used log-normal distributions representing uncertainties derived from this letter, from reported errors for replicate leaves and palaeosols3, and an error envelope determined from earlier multiparameter sensitivity analyses4, respectively. Uncertainties in T due to imperfect knowledge of the 18O of ocean water, diagenesis and vital effects, and ages were represented with normal distributions1. Age uncertainties for proxies were represented by uniform distributions based on reported dating error ranges3.
At the second stage, for each resulting synthetic data set a continuous interpolant for CO2 or T consistent with conditional uncertainty (given the discrete data points) about the rate and magnitude of changes between data points was generated from a gaussian stochastic process with a powered exponential covariance kernel. Parameters were adjusted to represent likely fluctuations in the relationship of magnitudes around 250 p.p.m.v. per million years for CO2, consistent with ice-core data23 or equivalent values for T using the same radiative forcing function4. The simulated histories resulting from the combined stages represent samples from the uncertainty distributions of the respective continuous functional relationships24.
A measure of the agreement between two functional relationships over a time interval T is given by the integrated correlation 
, where x(t) and y(t) are the functions representing the relationships standardized to zero mean and the unit integrated square over T; generally |r|
1, and r=1 indicates perfect agreement. The PDF of r was estimated by kernel density estimation25 from the ensembles of curves, and measures the agreement between histories, taking account of their respective uncertainties.
Palaeoclimate estimates
We calculated CO2 forcing using a simple CO2 greenhouse forcing function that accounts for solar evolution and the logarithmic relationship to CO2 concentration, with a moderate equilibrium climate sensitivity4 (2CO2 T=2.3 °C). Tropical surface ocean 18O temperatures1 were converted by dividing by 0.87 to approximate global surface values and facilitate comparisons with the CO2 forcing function26. An important caveat with this exercise is that the marine 18O record includes a pH adjustment based on CO2 from a geochemical model that revises T estimates upwards16. Lower CO2 concentrations, from either bryophyte records or the GEOEVW simulation, would revise temperatures down by 0.5 °C over the Mesozoic.
Geochemical model CO2 calculations
The volcanic-corrected GEOCARBSULF model18, 21 was used to calculate atmospheric CO2 concentrations by varying values of Wv/Wnv and NV. Wv and Wnv are the rates of weathering per unit mass respectively of volcanic and non-volcanic Ca and Mg silicates. Values used here were Wv/Wnv=1 for GEOSW (GEOCARBSULF with no volcanic effect18), Wv/Wnv=2 for GEOVW (moderate volcanic effect20) and Wv/Wnv=10 for GEOEVW (enhanced volcanic effect19). Both volcanic effects assume very low BSW reaction rates. More realistic BSW values give results for Wv/Wnv=2 similar to those for low BSW and Wv/Wnv=10. NV is a parameter that relates the deviation of the global mean 87Sr/86Sr from the present value (0.717) for non-volcanic Ca and Mg silicate weathering. Its value is uncertain and can only be crudely estimated. We use NV = 0 for GEOVW and NV=0.01 for GEOEVW. The use of NV=0.01 and Wv/Wnv=10 for GEOEVW applies only to very low BSW reaction rates and leads to a maximum effect on CO2 of changes in volcanic exposure.
Author contributions
B.J.F. conducted the geochemical and data analyses and drafted the manuscript, S.J.B. conducted data analyses, C.W.A. conceived and designed the uncertainty analyses and time-series comparisons, R.A.B. undertook the geochemical carbon-cycle modelling and D.J.B. planned the project, drafted the manuscript and undertook data analyses.