A fossil leaf can keep the shape of a fan, the fork of a vein and the ragged place where its margin tore. None of those features is the atmospheric record in this story. The record sits deeper, in a skin thin enough to be overlooked: microscopic pores bordered by pairs of guard cells.
Those pores are stomata. In life, they admitted carbon dioxide for photosynthesis while giving water vapour a route out. A plant could open and close each pore, but it also made a longer-term developmental choice about how many stomata to build, how large to make them and where to place them. In many plant lineages, leaves formed under lower carbon-dioxide levels develop more stomata, while leaves formed under higher levels develop fewer. When the cuticle survives fossilization, part of that developmental response can survive with it.[1][2]
This is often shortened to a seductive claim: count fossil pores, read ancient air. The real method is better and more demanding. It does not recover a bubble of atmosphere from a leaf. It converts preserved anatomy into a proxy through modern calibration, taxonomic comparison, laboratory preparation, repeated sampling and a model whose assumptions have to travel with the result.
The atmosphere survives here not as captured gas, but as a ratio between cells.
The useful fossil is the leaf's outer skin
A compression fossil may preserve little more than a dark silhouette, or it may retain portions of the original carbonaceous cuticle—the waxy envelope that protected the leaf from drying, ultraviolet radiation and abrasion. If cell boundaries remain legible, a paleobotanist can isolate the cuticle, distinguish the upper and lower surfaces, and examine the epidermis under transmitted light, fluorescence or a scanning electron microscope. Preservation quality determines whether the specimen offers a recognizable leaf, countable cells, measurable pore geometry or some combination of the three.[1][4]
The cover photograph makes that hierarchy visible by hiding its most important scale. It shows Smithsonian specimen USNM P 37192 A, identified as Ginkgo adiantoides from the middle Miocene Latah Formation near Spokane, Washington. The fan-shaped impression and branching veins are clear. Individual stomata are not. Nor does the collection page say that this particular specimen retains cuticle suitable for carbon-dioxide work. It is an honest macro view of the kind of fossil organ that may carry a microscopic archive, not evidence that every handsome leaf can become a barometer.[7]
Two measurements are easy to confuse. Stomatal density is the number of stomata in a unit of leaf area. Stomatal index is the percentage of epidermal cells that are stomata: the stomatal count divided by the combined count of stomata and ordinary epidermal cells, multiplied by 100.[1][2]
Density can change even when a plant makes the same number of stomata, because the rest of the leaf may expand more or less after those pores have formed. The index partly controls for that expansion by keeping stomata in relation to their cellular neighbours. It is not a universal correction for shade, water stress, development, taxonomy or preservation. It simply asks a sharper question: what fraction of this sampled epidermis was allocated to gas-exchange pores?[1][2]
That denominator is why the proxy is anatomical rather than decorative. A fan-shaped fossil may identify a ginkgo-like leaf from across a room. A carbon-dioxide estimate requires enough cellular structure to decide which small shapes are functional stomata, which are ordinary epidermal cells and which are damage, mineral film or arrested development.
An old leaf needs a living calibration
The inverse tendency between stomatal frequency and carbon dioxide is a biological response, not a physical law imposed uniformly on every plant. Species differ in its direction and magnitude. Some regulate gas exchange mainly by changing pore aperture; others make stronger developmental adjustments in stomatal number or size. Light, water availability, altitude, leaf position and genome history can modify the pattern. A calibration therefore has an inference space: the taxa and environmental range in which its relationship was actually tested.[1][2]
Living Ginkgo biloba has become especially important because historical herbarium sheets place one lineage against a measured atmospheric rise. Richard Barclay and Scott Wing used dated herbarium leaves collected from 1877 onward and modern leaves grown outdoors to rebuild the relationship between Ginkgo stomatal index and carbon dioxide. Their calibration covered roughly 290 to 430 parts per million. Chamber-grown leaves were examined separately but excluded from the response curve because many had malformed stomata and the measurements were highly variable. Fossil leaves assigned to Ginkgo adiantoides are similar enough in morphology and ecology to make comparison scientifically useful, but the comparison remains an assumption about conserved physiology, not proof that an extinct tree responded exactly like its living relative.[3]
The calibration range matters as much as the resemblance. At high carbon-dioxide levels, a response curve can flatten: once a leaf has reduced stomatal investment substantially, another increase in carbon dioxide may produce only a small anatomical change. Extrapolating a modern curve far beyond the observations that shaped it can then turn very different ancient atmospheres into similar-looking indices. Barclay and Wing explicitly judged Ginkgo response above 430 ppm poorly constrained. A formula does not become better calibrated because a fossil is older than the data.[2][3]
Even within the measured range, counting practice leaves fingerprints. In the Ginkgo recalibration, six historical herbarium specimens previously studied by another team were sampled again from different leaves; the new counts produced consistently lower stomatal indices. That did not make either dataset fraudulent. It showed that leaves sharing one specimen record are not necessarily interchangeable, and that preparation and counting protocols can leave measurable fingerprints. In a separate test within individual leaves, however, counts showed no significant difference among five sampled regions. Replication makes both disagreement and null effects visible.[3]
Preparation can improve one measurement and damage another
Before anyone counts a cell, the fossil has to survive the route from rock to microscope slide. The 2025 best-practices synthesis by Xiaoqing Zhang and colleagues treats that route as a sequence of choices, not one universal recipe. Mineral matrix may need to be dissolved. Coalified mesophyll may need to be cleared so light can pass through. The upper and lower cuticles may need to be teased apart with a fine needle or brush. Acids, oxidants and bases that work on one preservation type can shred, over-thin or obscure another.[4]
This creates a less obvious conflict. A treatment that produces beautifully clean cell outlines for stomatal counting can alter the cuticle's chemistry. Yet carbon-isotope composition is an input to some mechanistic carbon-dioxide models. If the whole sample is chemically prepared first, an excellent image may arrive at the price of a compromised isotope measurement. The best-practices review therefore emphasizes testing weak treatments, monitoring continuously and ordering analyses so that a less destructive observation or chemical measurement happens before an irreversible preparation when both are needed.[4]
Sampling continues inside the slide. Stomata are not distributed like dots printed by a machine. Counts can vary from base to tip, between veins, between sun and shade leaves, among leaves on one tree and among individuals. A defensible study records where fields were placed, counts multiple fields on multiple leaves, keeps biological replicates distinct and asks whether additional sampling still changes the mean. One immaculate micrograph cannot represent a forest simply because its cell walls are crisp.[3][4]
Fossilization adds another filter. A cuticle may wrinkle, shrink, split, overlap its counterpart or retain only patches that were unusually resistant. Researchers have to show that counted fields are comparable and that burial did not preferentially erase one class of cell. The photographed outline is fossil evidence. The sampled cuticle is a smaller subset. The atmospheric estimate is an inference built from that subset.
More elaborate models do not escape calibration
There are two broad ways to turn stomata into carbon dioxide. Empirical approaches fit a relationship between a measured feature—often stomatal index—and known carbon-dioxide levels in living or recent material, then apply that transfer function to a fossil counterpart. Their strength is direct calibration. Their weakness is dependence on a conserved response and the temptation to extrapolate beyond the tested range.[1][2][3]
Mechanistic approaches begin closer to leaf function. The model introduced by Peter Franks and colleagues combines stomatal density and pore dimensions with carbon-isotope discrimination and parameters for photosynthesis and gas conductance. Instead of asking only how one modern index correlates with carbon dioxide, it estimates how the leaf's anatomical capacity and carbon economy could have operated together.[1][5]
That added biology is valuable, especially for extinct plants without an identical living species. It is not assumption-free. Photosynthetic rate, the fraction of maximum conductance normally used by the leaf, boundary-layer behaviour and other variables cannot all be observed in a fossil. Some must still come from a living relative or equivalent. More parameters can represent more mechanism and also create more places for an unsuitable modern value to enter.[1][3][5]
The Ginkgo test is a useful warning against ranking models by complexity alone. Across its modern calibration set, Barclay and Wing found that the mechanistic model's mean prediction error was about five times that of the stomatal-index regressions; most estimates were too high. That result does not invalidate mechanistic modelling across the fossil record. It says that a model earns authority by reproducing known conditions for the taxon at hand, not by having the longest equation.[3][5]
Agreement is strongest when methods can fail differently
A proxy becomes more persuasive when it converges with evidence that does not share all its vulnerabilities. A study of late Pennsylvanian and early Permian glacial–interglacial cycles in the Illinois Basin paired fossil-plant cuticles with carbonates formed in ancient soils, sampled within the same stratigraphic framework. Empirical and mechanistic stomatal estimates made from extinct medullosan seed-plant leaves showed comparable variation, while the paleosol proxy, sea-level record and modelled ice volume supplied independent context. The agreement supported a changing atmospheric signal rather than one lucky calibration.[1][6]
The phrase independent still needs care. Samples can share an uncertain age model; vegetation and soil formation can respond to the same local hydrology; different calculations may reuse a common assumption. Multiproxy work is not a vote in which three estimates automatically defeat one. Its advantage is diagnostic: where records agree, confidence rises; where they separate, the pattern can identify a taxonomic calibration problem, a preservation bias, a stratigraphic mismatch or a model parameter that needs attention.[2][6]
This is the right scale at which to read any headline about carbon dioxide from fossil leaves. Ask which taxon supplied the cuticle, which leaf surface was preserved and counted, whether its cell boundaries were legible, how many fields and leaves were counted, what modern material built the calibration, whether the ancient estimate lies inside that range, which model translated anatomy into air, and whether another proxy follows the same change. An uncertainty interval should carry sampling, calibration and model error rather than only the precision of a microscope count.[2][4]
A fossil stoma directly preserves a piece of leaf anatomy. Its identity, density, dimensions and cellular neighbours are observations. The plant's gas-exchange strategy is a physiological interpretation. Atmospheric carbon dioxide is the modelled result. Keeping those levels separate does not weaken the proxy. It reveals why a microscopic feature can reach beyond one leaf without pretending to contain the sky.
The mount in the photograph holds a dark fan. Its atmosphere is not visible. If its cuticle survived, it waits in the disciplined comparison between pore and cell, fossil and herbarium sheet, preparation and chemistry, one model and another.
Sources
- Jennifer C. McElwain and Margret Steinthorsdottir, “Paleoecology, Ploidy, Paleoatmospheric Composition, and Developmental Biology: A Review of the Multiple Uses of Fossil Stomata,” Plant Physiology 174 (2017)—stomatal anatomy, empirical and mechanistic carbon-dioxide proxies, taxonomic response and cross-calibration.
- Gregory J. Jordan, “A critical framework for the assessment of biological palaeoproxies: predicting past climate and levels of atmospheric CO2 from fossil leaves,” New Phytologist 192 (2011)—inference space, calibration, extrapolation and evolutionary uncertainty.
- Richard S. Barclay and Scott L. Wing, “Improving the Ginkgo CO2 barometer: Implications for the early Cenozoic atmosphere,” Earth and Planetary Science Letters 439 (2016)—historical-herbarium calibration, counting variance, model testing and limits above the observed carbon-dioxide range.
- Xiaoqing Zhang et al., “Fossil leaf cuticle: Best practices for preparation and paleo-CO2 analysis,” Earth-Science Reviews 264 (2025)—cuticle isolation, cleaning, microscopy, sampling design and analysis-order tradeoffs.
- Peter J. Franks et al., “New constraints on atmospheric CO2 concentration for the Phanerozoic,” Geophysical Research Letters 41 (2014)—the leaf gas-exchange model combining stomatal geometry, carbon isotopes and physiological parameters.
- Isabel P. Montañez et al., “Climate, pCO2 and terrestrial carbon cycle linkages during late Palaeozoic glacial–interglacial cycles,” Nature Geoscience 9 (2016)—paired fossil-cuticle and paleosol records from Illinois Basin cyclothems.
- Cynthia Walden and Smithsonian Institution, “Ginkgo adiantoides USNM 37192 img2,” Wikimedia Commons (2021)—source and collection record for the CC0 fossil-leaf photograph, extracted from the full specimen image.