The Carboniferous: The Construct Changed Its Environment, and the Environment Turned Back to Kill It

1 Coal Is Not Evenly Spread Through Time

Earth's coal is not distributed evenly across geological history. An extraordinary amount is crowded into one stretch of it. During the hundred million years from roughly 350 million to 250 million years ago, swamps laid down a very large share of what we now dig out of the ground and burn. Much of the coal mined in China's Shanxi Province is what became of those swamps.

The first part of that stretch is called the Carboniferous. Two British geologists gave it the name in 1822: it means “coal-bearing.” Britain had been mining coal for centuries, and most of it, including much of the fuel that fed the steam engines of the Industrial Revolution, came from this succession of rocks. The Chinese name is just as literal: its first two characters are an old Chinese word for coal. The second part of the stretch is the Permian.

Plants had already been living on land for more than a hundred million years before this interval began. They went on living for more than two hundred million years after it ended. Why should this particular stretch of time have buried so much carbon underground instead of returning it to the atmosphere?

We can put a number to “so much.” Estimates that bring together the rock record, organic carbon, and isotopes identify the Carboniferous as the fastest period of global carbon burial in roughly the past five hundred million years. The modeled peak is about 78 million metric tons of carbon a year, with uncertainty that could take it roughly 50 percent higher or lower.

That is a calculated, long-term global average, not a quantity measured in a particular coal mine. Nor did all the buried carbon gather into coal seams. Much of it was scattered through mud on the floors of seas and lakes, mud that later became black shale. Coal is simply the most conspicuous and concentrated part of that larger burial.

For comparison, humanity now releases roughly a hundred times that peak quantity of carbon into the atmosphere every year. The Carboniferous did not lay down so much coal by working fast. It had time, and a succession of conditions that happened to fall into place together.

Before going further, we need to be clear about how coal forms. Everything in the argument that follows depends on it. Coal begins as peat: dead plant material that has accumulated without rotting away completely. The soft, dark-brown peat sold at garden centers is the same sort of stuff. For it to accumulate, one very particular condition has to be met. There must be water.

Where oxygen is available, microorganisms make short work of dead plants. Where the ground stays flooded, oxygen has trouble getting in, and decomposition slows. New plants keep growing, dying, and piling onto the surface while the plants underneath have yet to finish rotting. That is a peat swamp.

Today's tropical peatlands and northern sphagnum bogs are still doing the same thing. The waterlogged peat-swamp forests of Southeast Asia are often used as the closest living counterparts of Carboniferous coal swamps. Bury the peat, compact it, and heat it; drive out the water and volatile substances; concentrate the carbon that remains. What you have then is coal.

Coal formation therefore requires more than “lots of plants.” Three conditions have to coincide. First, plant production must outpace decomposition. Second, the root zone must remain water-saturated and short of oxygen for long periods. Third, the basin must subside at roughly the rate at which the peat grows upward.

The third condition is the easiest to overlook, and one of the most important. If the basin sinks too slowly, the peat surface spends too long above the water. Oxidation, fires, and erosion by rivers can strip away carbon that has already accumulated.

If the basin sinks too fast, or relative sea level rises too quickly, deep water and sediment overwhelm the swamp. The requirement is a window, not a direction: more is not always better. Coal does not call for several conditions to become extreme. It calls for several conditions to remain, together, within a fairly narrow range.

How fast is the right rate of subsidence? Roughly fast enough to keep pace with the upward growth of the peat. In Southeast Asia today, peat can grow by one or two millimeters a year. Burial then compresses it severely and squeezes out its water: it takes on the order of ten meters of peat to make one meter of coal. Expressed as eventual coal thickness, that annual growth is only about 0.1 to 0.2 millimeters. Behind a coal seam one meter thick there are often several thousand to more than ten thousand years of swamp.

As the basin sinks little by little, it keeps making fresh room for the peat to fill, layer after layer. Geologists call that room “accommodation space.” We will need the term repeatedly.

The coal-forming regions were not scattered at random, either. The most characteristic giant coal swamps of the Late Carboniferous lay along the equatorial belt of their time. The continents we know today were then coming together into one great landmass, Pangaea.

As that assembly proceeded, the regions now occupied by eastern North America and western to central Europe straddled the ancient equator. The North China and South China blocks, among others, also occupied tropical or low-latitude positions.

Meanwhile, most of the high-latitude terrain of Gondwana was under glaciation. This southern landmass still held together what are now South America, Africa, Antarctica, Australia, and India. Latitude and tectonic setting lined up closely: Euramerica, the joined European and North American landmass, lay in front of a rising mountain belt. The mountains' weight bent down the crust ahead of them, producing immense basins. Continued subsidence, fast enough but not too fast, kept creating new accommodation space.

The existence of coal swamps in North China and elsewhere in East Asia will become crucial to an entire section later on. When the lycopsid wetlands of Euramerica declined sharply, similar wetland plants in the east carried on into the Permian.

Finally, there are the plants that actually made the coal. Their identity bears directly on the next section. In the Late Carboniferous coal swamps of Euramerica, the most important contributors were tree-sized lycopsids, relatives of today's clubmosses.

In many seams, these plants accounted for 60 to 80 percent of the plant biomass. The clubmosses still alive today are small plants that hug the forest floor, most of them only a few centimeters to a few tens of centimeters tall.

Their Carboniferous relatives grew into trees tens of meters high. Diamond-shaped leaf scars covered their trunks in a repeated pattern like fish scales; one of the best-known genera, Lepidodendron, takes its name from that scaly appearance. Early studies proposed that some could shoot up to those heights in little more than a decade, and that some species died after producing spores just once.

More recent estimates suggest that they generally lived for several hundred years. A tree tens of meters tall, alive for a dozen or so years or for centuries: even that basic difference remains unsettled.

A Carboniferous coal swamp would have looked unlike any forest standing today. A Lepidodendron trunk rose straight, with scarcely a branch, until it reached the top. Only then did it fork in two, then four, opening out into a crown crowded with long, slender leaves.

Below the canopy stood ferns grown to tree size and tree-sized horsetails, their jointed stems like enormous asparagus stalks. There were no flowers, no grasses, and no birds. Water covered the ground year-round. The roots spread sideways through the mud instead of plunging downward into it.

These trees did not hold themselves up with the thick columns of wood found in modern conifers. Much of their support came from massive outer tissues and cortex. True wood could make up only a small fraction of the stem.

Think of a thick-walled, hollow column rather than a solid wooden post. The chemistry of those outer tissues differed from that of wood, and so did their resistance to decomposition. The intuitive equation “lots of coal means vast quantities of hard-to-decay wood were buried” does not hold.

A great many dead trees do not, by themselves, make coal. Water level, subsidence, and production all have to find their places within the same narrow window.

2 The Debate Over Coal Formation

One explanation has traveled far and wide. Plants invented lignin first; fungi took tens of millions of years to invent a way of eating wood. During the gap, dead plants could not rot, so they piled up and became coal.

Lignin is the substance that makes wood hard. At the end of Episode Twelve, we saw why it resists breakdown: its structure is irregular, without a fixed repeating unit, whereas most enzymes are fitted to regular structures. The story therefore sounds eminently plausible. The idea from which it developed was considerably more cautious.

In 1990, a researcher proposed that the chemical makeup of land-plant tissues, together with the evolution of their decomposers, could alter the terrestrial organic-carbon cycle. She argued that a scarcity of lignin decomposers in the Paleozoic, or limits to what they could do, might have increased the proportion of land-derived organic carbon that survived.

Her central proposal was not a “fungal switch” that flipped in one precisely dated million-year interval. It was a long-timescale hypothesis about biological evolution changing the speed at which organic carbon turns over. The stronger version gained currency mainly in the era of molecular phylogenetics. A molecular study published in 2012 offered it an exceptionally tempting match of dates.

First, a word about white rot. White-rot fungi are named for the wood they leave behind: once the lignin has been dismantled, the remaining fibers look pale and tear away in stringy strips. Brown-rot fungi take the opposite approach. They consume cellulose and leave the lignin behind, turning the wood brown; as it dries, it breaks into little blocks.

The researchers analyzed 31 fungal genomes, 12 of them newly sequenced for the study. They concentrated on the class of enzymes that white-rot fungi use to oxidize lignin.

Their reconstruction showed an expansion of those enzymes along the lineage leading to the ancestor of a major fungal group. They reconstructed that ancestor as capable of white-rot-style decomposition. Most of the familiar mushrooms in today's forests belong to this large group.

For the dates, they used a molecular clock calibrated with fungal fossils. The central estimate for the group's origin was about 290 million years ago, shortly after the end of the Carboniferous. This put the enormous coal formation of the Late Carboniferous first, with the expansion of a complete white-rot decomposition system following roughly after the coal peak.

But “290 million years ago” is not a pin that can be driven into one exact spot in the rock record. Fungal molecular clocks are sensitive to sparse fossil calibrations, the branching arrangement of the evolutionary tree, and the models used. The uncertainty around that ancestral node is far wider than the stratigraphic dating error around the end of the Carboniferous itself.

The hypothesis has three evident strengths. It identifies a distinct biological innovation that can be tested: find the date at which the white-rot enzyme system originated. It also brings coal, organic-carbon burial, rising atmospheric oxygen, and falling carbon dioxide together in a single causal chain.

And the genome studies really do show that the different feeding strategies of white-rot and brown-rot fungi were accompanied by expansions and repeated losses in the relevant gene families. There is nothing wrong with the proposition that the capacity for decomposition evolves.

The disputed claim is the stronger one. Was the late appearance of this enzyme system a necessary condition, or even the principal cause, of the late Paleozoic coal peak? That is a different proposition from saying that fungal lignin decomposition underwent important evolutionary changes. A 2016 paper made the distinction with a blunt title: a delay in fungal evolution did not cause the Paleozoic peak in coal formation.

Its objections can be gathered under at least six headings. First, many of the chief coal-forming plants were not “heaps of lignin” in the first place. As we have just seen, lycopsids could contribute roughly 60 to 80 percent of the biomass, and their massive outer tissues were not the highly lignified wood of modern trees.

The premise that the coal peak resulted from a buildup of lignin therefore fails precisely where it most needs to work: in the plants that supplied most of the coal.

Second, peat can accumulate without lignin becoming wholly immune to decay. Modern peatlands build up large quantities of organic matter despite having a full complement of fungal and microbial decomposers. Net accumulation is the difference between production and decomposition. Persistent waterlogging, low oxygen, acidity, and nutrient limitation are quite capable of slowing the second term.

As long as plant inputs continue to exceed losses through decomposition, there is no need for a special interval in which “nothing alive knows how to eat lignin.” Today's peatlands do this with white-rot fungi already well established. In some Southeast Asian peat swamps, peat builds to depths of well over ten meters, even twenty.

Of the six objections, this is the most direct. The condition required by the hypothesis does not hold on Earth today, yet peat goes on accumulating.

Third, there is already evidence of decay in the Carboniferous. Late Paleozoic plant tissues show widespread signs of biological breakdown, erosion, and fungal activity. In some fossil sections preserved down to the cellular level, fungal threads can be seen penetrating cells. There are also tunnels chewed by small arthropods, along with the fecal pellets they left behind.

Fourth, the molecular clock is not precise enough to establish that “coal stopped when white rot arrived.” Its central date lies close to the Carboniferous–Permian transition, but the uncertainty interval is wide. Nor are the “ancestral white-rot node,” the “enzyme-expansion node,” and the capacity for complete lignin mineralization in the modern sense exactly the same biological event.

Fifth, giant coalfields kept forming after white-rot fungi appeared. There were important coal-forming intervals in the Permian, from the Jurassic into the Cretaceous, and in the Cenozoic.

Look at the ages of the world's coal resources and there is no final, once-and-for-all “end of coal” at roughly 290 million years ago. The coal of Shendong in China's Ordos Basin, one of the country's largest-producing coalfields today, is Jurassic. White-rot fungi had long been around by then.

Sixth, the coal peak fits climate and basin conditions better. The late Paleozoic anomaly arose from a chance overlap: a persistently wet equatorial climate coincided with immense sedimentary basins created as Pangaea assembled. Plants grew vigorously. High water tables restricted decomposition. At the same time, basins subsided at a suitable pace, continually making new space in which the accumulated material could be preserved.

The two sides, then, are not actually arguing about the same proposition. The 2012 study showed that the enzymes used to decompose wood have an evolutionary history we can reconstruct. The 2016 paper rejected the claim that their late arrival caused the global coal peak. Both statements can be true at once.

For now, the better answer lies with climate, tectonics, and basins. That explanation works across different latitudes and different basins, and it accommodates the repeated episodes of major coal formation after white rot had appeared. Decomposers are not out of the picture. They may still alter the fraction of organic matter that survives.

The two sets of factors carry different weight. Decomposers determine how much of a given heap of dead plants is left. Climate and basins determine how large an area keeps producing such heaps, and for how long. What made the Carboniferous truly unusual lay chiefly in the second set.

The fungal hypothesis does have a particular strength: its prediction is sharp. Find a white-rot fossil older than the coal peak, or push the molecular-clock node back to an earlier date, and the hypothesis loses its footing. The climate-and-basin explanation has no single point quite so sharp, but it can draw on far more independent lines of evidence: ancient latitude, basin-subsidence history, coal-seam thickness, the advance and retreat of the sea, moisture recorded in ancient soils, and glacial cycles. Each can be checked on its own.

3 Oxygen Nobody Measured

The mechanism linking large-scale organic-carbon burial to rising oxygen is clear. Photosynthesis reduces the carbon in carbon dioxide to organic matter and releases oxygen in the process. If that organic matter is subsequently consumed in respiration, rots, or burns, the net effect is largely canceled out.

A long-term net source of oxygen remains only when some of the reduced carbon escapes this cycle of oxidation and stays buried in sedimentary rock. It is an easy point to slide past. Most of the oxygen released by photosynthesis is consumed again when the organic matter is reoxidized.

A forest can hold an immense quantity of biomass and still be close to neutral in the long-term oxygen budget, provided its carbon eventually returns to the atmosphere. What matters for a lasting oxygen source is the escape fraction, not the amount produced. Think of a piggy bank. However much money comes in, if nearly as much goes out again, the only savings are the small amount you did not spend.

This brings us straight back to the argument in the previous section. Decomposer capacity affects precisely that escape fraction. Climate and basin conditions affect how large an area operates with a high escape fraction, and how long it keeps doing so. How high, then, did oxygen actually rise?

Several other questions hang on that number. Did it let giant insects grow? How often did wildfires burn? How much coal-swamp carbon was burned straight back into the air? Before using the figures, we need to know where they come from.

Nobody can measure the oxygen content of the air three hundred million years ago. Instead, researchers build a model. They write equations for the movements of carbon and sulfur through Earth's surface system: how much volcanoes release, how much weathering consumes, how much organic matter is buried, and how much is oxidized again. Isotope records constrain those quantities. Solving the equations yields a history of oxygen through time.

The number is therefore calculated, not measured. How well the model performs depends on how good the numbers fed into it are.

Oxygen makes up about 21 percent of the air today. Early models put the Permian–Carboniferous peak as high as roughly 35 percent. A 2006 model produced a broad late Paleozoic peak, with a Permian maximum of about 30 percent.

In 2014, a study systematically varied all 68 inputs to this kind of model together, within plausible ranges, for the first time. Imagine a panel with 68 knobs. Give each knob a turn within its reasonable range and run the calculation. Turn them again and run it again. Do that ten thousand times, then look at where the answers fall.

The result mattered. A late Paleozoic high-oxygen peak still appeared in the median prediction, and the median curve rose above 30 percent.

Its absolute height, however, was much less secure than popular-science diagrams tend to suggest. Reading the graph near the peak, the middle 95 percent of those ten thousand runs spanned roughly 22 to 45 percent oxygen. The researchers also found that no one parameter accounted for the uncertainty. It was distributed among several carbon reservoirs, sulfur reservoirs, isotope values, and weathering parameters.

In other words, there was no single critical parameter for which one could say, “Measure this accurately and the uncertainty band will narrow.”

There is another route: find something in the rocks that gives an indirect reading of oxygen. Geologists call such evidence a proxy. Charcoal is one example. The dark little streaks in coal, faintly silky in luster and liable to shed black dust when touched, look like charcoal because that is what they are: charcoal left by fires three hundred million years ago. The underlying logic is simple. A wildfire needs enough oxygen both to start and to keep burning.

Researchers have ignited real plant fuels in combustion chambers where oxygen concentration can be adjusted. For fuels resembling land vegetation, the lower oxygen threshold for sustained burning falls around 16 to 17 percent. The air today is only four or five percentage points above that line.

Below it, getting a flame to catch does not mean that the fire can spread by itself. Widespread wildfire charcoal preserved where it formed therefore supplies an independent constraint for any geological interval: atmospheric oxygen is unlikely to have remained below that range for long.

The argument works in the other direction, too. It was once thought that an atmosphere with 35 percent oxygen would make forests almost impossible anywhere on Earth. Yet experiments at high oxygen levels found that real forest fuels could still burn in physically plausible ways at around 35 percent oxygen. Endless global conflagration was not an inevitable result.

Wildfire therefore constrains a range; it does not impose a simple, absolute upper limit. At the broadest level, models and proxies agree: the Late Carboniferous and Early Permian were not an age of low oxygen.

At the next level of detail, agreement is much weaker. Was the peak around 26 percent, 30 percent, or above 35 percent? We cannot claim to have settled that precisely. The sharpest tension lies between some newer geochemical estimates of oxygen and the classic models.

Some data have even yielded values near 15 percent for parts of the late Paleozoic. That is close to, or below, the roughly 16 percent lower threshold for fire spread discussed above, making it hard to reconcile with the widespread charcoal record. Contradictions of this kind show that a “direct proxy” can also be affected by whether a sample remained chemically closed, by later exchange, or by the way the proxy was calibrated.

A proxy is not inherently more reliable than a model. The safer conclusion is that oxygen was probably higher than it is today, with most models placing it somewhere around 30 percent. Whether the peak was 26, 30, or above 35 percent remains unresolved. The much-repeated figure of “35 percent” is simply one estimate toward the high end.

4 Wings Longer Than a School Ruler

The most famous inhabitants of the Carboniferous were arthropods of almost preposterous size. The air also acquired flying animals for the first time: the earliest unambiguous fossils of winged insects come from the Carboniferous.

Take Meganeura monyi, a distant relative of dragonflies. The first specimen discovered preserved substantial portions of all four wings. One wing was about 32 centimeters long, longer than a standard school ruler.

Allow for a pair of wings and the body between them, and the estimated wingspan reaches 65 to 70 centimeters, about the width of a pigeon with its wings spread. The largest dragonfly relatives alive today manage only something in the teens to around 20 centimeters. Meganeura was a hunter. Rows of stiff spines lined its legs; brought together, those legs formed a basket for scooping other flying insects out of the air.

The creatures crawling below could be larger still. Arthropleurids were millipede-like, many-legged animals. In 2018, a block of sandstone fell from a cliff on the British coast and split open, revealing a section of one animal's exoskeleton. The study of it was published in 2021.

From its width and the proportions of its segments, researchers estimated an entire animal about 1.9 to 2.6 meters long and roughly 50 kilograms in weight. It would have been longer than a person lying down. What they found was only a section; the whole animal was reconstructed by inference.

Most giant fossil insects, too, are represented only by their wings. When a fossil database tells us how big “the biggest insect” was, the measurement is generally a wing, not a body mass.

What allowed them to grow so large? Oxygen is the most widely circulated explanation. To understand it, we first need to understand how an insect breathes.

An insect has no lungs, and blood is not its principal oxygen-delivery system. Air enters through a row of small openings in the body surface and passes into a network of tubes. The tubes branch again and again, becoming finer until they run directly alongside the tissues. Oxygen reaches those tissues by diffusing along the tubes, rather than being carried there in blood. Larger airways can also be actively ventilated by body movements. The pure-diffusion constraint discussed here chiefly concerns the final distance to the tissues.

Diffusion, however, has a physical limitation: the farther it has to go, the longer it takes. More specifically, the time required to cover a distance by diffusion alone increases approximately with the square of that distance.

Double the distance and the time becomes four times as long. In an insect twice as thick, if the tissues are twice as far from the air tubes, oxygen takes four times as long to diffuse across the gap.

As an animal thickens and tissues lie farther from the tracheae, its respiratory network therefore needs a larger cross-sectional area, more branches, or more powerful ventilation. Higher oxygen levels increase the oxygen partial-pressure gradient between air and tissue. Other things being equal, that lets the same set of tubes deliver more oxygen.

This hypothesis makes testable predictions. Low oxygen should impose a limit on large size sooner. Larger insects should have to devote an ever-greater proportion of their internal space to tracheae. Three-dimensional studies of beetles have indeed found that investment in the tracheal system rises faster than body size across larger beetle species, particularly at the narrow passage where a leg joins the body.

The size of the effect in experiments is the crucial point here. In 2009, researchers artificially selected fruit flies for large body size over eleven generations, under oxygen partial pressures corresponding to roughly 10, 21, and 40 percent oxygen.

In the normal-oxygen and high-oxygen groups, average mass increased by about 11 to 17 percent, while the upper quartile of the size distribution could increase by roughly 25 to 32 percent. The decisive result was that the response to selection did not differ significantly between 21 and 40 percent oxygen. What stood out was the way low oxygen suppressed selection for larger size.

In this fruit-fly experiment, low oxygen held back the response to selection for larger bodies; high oxygen brought no greater increase than ordinary air. That supports “too little oxygen can impose a limit” more strongly than “extra oxygen automatically makes giants.” But eleven generations of fruit flies cannot settle what dragonflies could do over millions of years. Putting a modern dragonfly in oxygen-rich air is not the same as recreating the Carboniferous.

Later work raised dragonfly nymphs under oxygen conditions of roughly 12, 21, and 31 percent. Dragonflies reared under high oxygen emerged as adults with wingspans up to about 15 percent larger. Different insect groups did not all respond in the same way.

That is a difference between insects of the same generation growing up in different air, not the outcome of tens of millions of years of evolution. Even so, take the full 15 percent increase and it is nowhere near enough to give the largest living dragonfly a wingspan of sixty or seventy centimeters.

A direct physiological study in 2026 then struck at the mechanism at the center of this causal chain. Its conclusion was that oxygen delivery through the tracheole–muscle system does not limit insect gigantism.

Using electron microscopy, the researchers counted the finest branches of the air-tube system inside the flight muscles of many insect species. In most species, those tiny tubes occupied only about 1 percent of the flight muscle, or even less.

Even if oxygen became scarce, there was plenty of room left in the muscle for additional tubes. The limit anticipated by the classic model, a system packed so tightly that nothing more could be fitted in, was not there.

High oxygen can therefore no longer be treated as a sufficient mechanical explanation for late Paleozoic insect gigantism. That does not prove oxygen had no effect. The main target is one specific mechanism: passive diffusion at the ends of the tracheal system imposing a hard upper limit on body size.

There are other explanations to explore. The authors of that physiological study suggested looking elsewhere, including increasing pressure from natural enemies and the physical limits of the exoskeleton itself. Another study assembled more than 10,500 records of fossil insect wing length. From the Carboniferous to the end of the Jurassic, maximum wing length tracked the oxygen models. In the Early Cretaceous, however, the relationship changed. Even when oxygen conditions became favorable again, new biological factors constrained the largest wings.

The authors regarded the arrival of flying vertebrates, especially birds, as an important source of pressure. Oxygen limitation in aquatic larvae offers another line of explanation. Young dragonflies live in water, which contains far less oxygen than air. Higher late Paleozoic atmospheric oxygen may therefore have raised dissolved oxygen in water and eased the oxygen constraint during the larval stage.

That is a different mechanism from a diffusion limit in an adult insect's tracheal system. It also helps explain why certain groups with aquatic young might be especially sensitive to high oxygen, rather than every kind of land-dwelling arthropod growing larger in step.

Finally, there is the problem of sampling. Maximum fossil insect size is usually represented by the longest wing known from each time window. Such an extreme value is unusually sensitive to the number of specimens sampled. One exceptionally rich insect deposit in a particular rock interval can abruptly raise the “maximum body size” recorded for that time.

More importantly, gigantism did not occur in every arthropod group, and it did not persist throughout the entire high-oxygen interval. The relationship through time between oxygen and maximum size is therefore a statistical correlation, not a one-for-one switch.

5 A Feedback Loop That Changes Sign Within a Single Cycle

Now lay out the chain and mark the strength of the evidence beside each link. Highly productive wetlands, combined with a suitable rate of basin subsidence, increase the net burial of peat and organic carbon.

Strong. Reduced carbon is removed from the surface cycle for the long term; atmospheric carbon dioxide falls, and the net oxygen source grows. The mechanism is strongly established. The absolute fluxes carry a moderate degree of uncertainty.

High-resolution reconstructions for the Late Carboniferous and Early Permian show large swings in carbon dioxide. In the very latest Carboniferous, it could fluctuate with orbital cycles between roughly 150 and 700 parts per million. “Orbital cycles” are the slow changes in the shape of Earth's orbit around the Sun, and in the tilt and direction of its axis, with periods ranging from around twenty thousand to around four hundred thousand years. Ice sheets wax and wane in response, and carbon dioxide rises and falls with them. Some Early Permian records yield values as low as roughly 100 parts per million.

For comparison, atmospheric carbon dioxide today is above 400 parts per million; even at the coldest point of the last glacial period, it was around 180. Proxies such as plant stomata and ancient soils have large uncertainties, so the figure of 100 should not be treated as a precise global average. The broader conclusion, however, is well supported: late Paleozoic carbon dioxide reached one of the lows of the Phanerozoic.

Some climate models go further. Under the weaker solar radiation of the late Paleozoic, carbon dioxide below roughly 40 parts per million sends the modeled planet into a globally frozen state.

The real Earth did not settle into a stable snowball state. But the calculation shows that Late Carboniferous to Early Permian carbon dioxide levels were low enough to bring the climate system very close to a region where powerful ice feedbacks could take over.

Low carbon dioxide, Gondwana's high-latitude position, and weathering associated with mountain building together promoted the expansion of late Paleozoic ice sheets. Rain washed over newly raised mountains; as the rock weathered, it drew carbon dioxide from the air, little by little. That is the role of mountain-belt weathering. The correlations are strong and the modeling support is strong, but this is not a single-cause relationship.

Glacial deposits from that ice age are now scattered across South America, Africa, India, Antarctica, and Australia. More than a century ago, Wegener used them as one line of evidence for continental drift. Put those continents back together as Gondwana, and the scattered glacial traces join up.

Modern syntheses place the late Paleozoic ice age at roughly 370 million to 260 million years ago, a span of about a hundred million years. Ice was most extensive from the middle to late Carboniferous into the Early Permian. It was the longest ice age since animals appeared on Earth. But it was not one unchanging ice cap sitting there for a hundred million years. It was a succession of expansions, retreats, and intervals when the ice almost vanished.

Euramerica lay on the equator, far from the glaciers. Its rocks recorded the ice's echo rather than the ice itself. When the southern ice sheets grew, seawater was locked away in ice and global sea level fell. When they retreated, sea level rose again. To say that “coal caused the late Paleozoic ice age,” however, is to go too far.

At least four other sets of variables have to stand alongside it: Gondwana's position across the South Pole; the assembly of Pangaea and the changing distribution of land and sea; intensified silicate weathering associated with mountain building; and changes in ocean circulation, volcanic outgassing, and orbital forcing.

Each link holds up when examined separately. Carbon burial lowers carbon dioxide: strong. Low carbon dioxide favors the growth of ice sheets: strong. But string them together into one exclusive chain, “coal caused the ice age,” and the evidence is no longer enough. Every link also has other inputs.

Changes in ice volume produce large rises and falls in global sea level. Strong.

Studies of the Late Carboniferous generally infer ice-driven sea-level swings of tens of meters to around a hundred meters or more; some syntheses put the largest near 120 meters. Sea level, in turn, controls the water table and accommodation space in coastal basins. Peat swamps are repeatedly established, exposed, or drowned. Strong.

When sea level falls, continental shelves and coastal plains emerge, rivers cut downward, and existing peatlands may undergo oxidation and erosion. As the ice melts and the sea rises again, groundwater rises with it. Low-lying plains flood once more, and peat swamps develop during the suitable part of the sequence.

If the sea continues to advance, lagoonal, deltaic, or shallow-marine sediments cover the peat and preserve it as a future coal seam. A typical coal-bearing cycle can therefore show, in succession, an ancient soil, coal, rocks made from transported sediment, and shallow-marine carbonate rock or shale.

Read that stack as a record: this place was dry land, then a swamp, then covered by mud and sand brought in by rivers, and finally submerged beneath the sea. The stack may recur dozens of times. It resembles an old ledger with the same page copied out again and again. Every fresh copy corresponds to another swelling and shrinking of the ice sheets.

These rock layers are not scenery around the Carboniferous story. They are a direct record of the loop itself. Coal, then mud and sand, then marine sediment, then coal again: this is what the metronome looks like when it leaves its marks on paper.

On one occasion, a beat of the metronome came with startling speed. In two coal mines in Illinois, once miners had removed the seam, they could look up at an entire forest more than three hundred million years old. It was impressed into the roof rock where it had stood, spread across roughly ten square kilometers.

Researchers infer that a major earthquake suddenly dropped the swamp below sea level, after which sediment sealed it within months. Lycopsids more than forty meters tall, tree-sized horsetails, and tree ferns remained rooted at the top of the coal seam, preserved in place. This is why neither “the lower the sea, the more coal” nor “the higher the sea, the more coal” will do.

Coal formation requires a window of suitable water level and accommodation space. Glacial sea-level oscillations can repeatedly open and close that window within a single basin. This is the clearest link in the loop, and also the one most easily written backward. It is not a simple, one-way positive feedback.

An ice-driven retreat of the sea can expose peat and reduce its preservation. The ensuing marine advance can raise the water table and bury the peat rapidly. The feedback thus changes sign within one cycle. The ice age turned coal formation from a nearly continuous process into one repeatedly switched on and off by the sea-level metronome.

Fire can change sign in much the same way. With more oxygen, even damp vegetation catches more readily, and charcoal turns up throughout the coal of the time. One fire sends some plant carbon straight back to the atmosphere while turning another portion into charcoal that is exceptionally resistant to further decay. The first effect weakens burial; the second strengthens it. The same fire presses both the brake and the accelerator. Which pedal you see depends on which portion of carbon you ask about.

Over longer timescales, the equatorial belt of Euramerica showed increasingly clear signs of seasonality and drying in the Late Carboniferous. Persistently flooded lycopsid wetlands contracted. Opportunistic ferns and tree ferns became more abundant. Later, seed plants able to tolerate seasonal drought came to dominate larger and larger areas.

The trend was neither globally synchronous nor an instantaneous changeover. The turnover in Euramerican wetland vegetation was striking, while North China and other eastern low-latitude regions retained important lycopsid coal-forming ecosystems into the Permian.

6 The Rainforest Broke Into Islands

Around 305 million years ago, the continuous coal-swamp forests of Euramerica fragmented within a relatively short interval. A 2010 study gave the event a memorable name: the Carboniferous Rainforest Collapse. Botanists had long known that the plants of the Euramerican coal swamps had undergone a major turnover. The new contribution was to tie that change to the fortunes of tetrapods and make a complete ecological story of it.

The tetrapods in those swamps came in many forms. Some resembled salamanders; some had no legs and looked like snakes; others had long bodies like crocodiles. In the proposed story, continuous rainforest broke up into separate, moist “ecological islands.”

Tetrapods were stranded on their respective islands and evolved separately, producing more and more kinds found in only one place. The paper reported a striking number: family-level diversity within a local community could fall from roughly twenty families to seven. A family is a taxonomic rank two levels above species; dogs, wolves, and foxes, for example, all belong to one family.

The story had winners as well as losers. Amphibians, unable to break their dependence on water, suffered most. Amniotes came through. These are the branch of tetrapods able to lay their eggs on dry land, the branch that includes today's reptiles, birds, and mammals.

Their eggs brought a little pool of water with them for the embryo to grow in, so reproduction no longer required a return to open water. They were better suited to the newly drier conditions. Some began specializing in plants, others in eating other animals; the range of ways to make a meal expanded sharply.

The changes on the plant side are solidly established, though they unfolded more slowly than the word “collapse” suggests. First came a gradual increase in fast-growing ferns, quick to occupy whatever space opened up.

Then, within a short interval, the dominant tree-sized lycopsids declined drastically and tree ferns gained the upper hand. Tree ferns are exactly what the name says: ferns that grow as trees. The scaly tree ferns still found in valleys in southern China are one example.

The spore-and-pollen record is especially important here. A trunk or leaf represents the small patch of forest in which it was buried. Spores and pollen travel far on wind and water, recording vegetation over a much wider region.

Ancient soils provide another, independent clue. Some soils formed around the event indicate shorter periods of waterlogging and improved drainage. These two lines of evidence rely on different modes of preservation, yet they point the same way: toward drier conditions.

There is much less agreement about why it grew drier. One explanation links the abrupt vegetation change to a brief but intense expansion of Gondwanan glaciation. The same body of research also keeps another possibility in play: a longer interval of greenhouse warming could likewise dry the continental interior.

That the plants changed is not in dispute. The disputed half of the story concerns the animals. In 2018, researchers reassembled a global dataset at the species level and corrected it for differences in sampling intensity.

The raw number of Carboniferous tetrapod species turned out to be strongly correlated with how much had been sampled. After correction, diversity first rose in the Late Carboniferous. The more clearly significant decline appeared closer to the Carboniferous–Permian boundary.

More importantly, their biogeographic network analysis did not support a marked increase in local endemism after the rainforest collapse. The Early Permian instead showed stronger signs of widespread distributions: the same animals were reaching more places. The island-fragmentation link could not be found in the data.

Further work in 2023 again pointed to sampling. A substantial part of the apparent changes in early tetrapod diversity and endemism could be explained by how intensively rock intervals and fossil localities had been sampled. Here sampling bias may affect more than a curve: it can also undermine the ecological mechanism we build around that curve.

Quantitative extinction curves for insects are much less secure. The insect record in coal-bearing rocks depends heavily on a small number of exceptional fossil deposits, so no reliable global figure can currently be given.

Nor does this event rank alongside the usual Big Five mass extinctions. The continents did not move in step. As the lycopsid coal swamps of Euramerica shrank dramatically, lycopsid wetlands persisted in North China and other eastern tropical regions.

At Wuda in Inner Mongolia lies a fossil locality sometimes called a “plant Pompeii.” Roughly 298 million years ago, already in the Permian, volcanic ash fell over a matter of days and buried a peat-swamp forest where it stood.

Leaves remain attached to their branches. Some small trees preserve leaves, branches, trunks, and cones together. Rising high into the canopy were great lycopsids such as Sigillaria. When the lycopsid forests of Euramerica had left the stage, those in North China were still standing.

A more fitting description, then, is a large-scale reorganization of low-latitude Euramerican wetlands. “Collapse” is easy to remember. Precisely because it is so memorable, it compresses a staggered, geographically uneven process into a single moment.

7 What the Coal Measures Selected

A coal seam is more than fuel. It is also a preservation machine with pronounced favorites. Sometimes minerals fill the plants lying in a swamp early enough to turn them into hard nodules known as coal balls.

Miners dislike them. They are hard, and they will not burn. Paleobotanists treasure them. Cut one open and grind it thin, and you find roots, stems, leaves, and spore cases preserved down to the level of individual cells.

As a result, we know far more about Carboniferous wetland plants than about plants from the dry uplands of the same age. The Carboniferous in our imagination is almost one immense coal swamp. That image is itself a product of the preservation machine's favoritism. The direction of the bias matters.

This is not a random shortfall of specimens. It systematically magnifies one ecological type. And the plants of that particular ecological type happen to be the main subjects of the debate about coal formation.

We know coal-forming plants in extraordinary detail because they fell into exceptionally good conditions for preservation, the very conditions that produced the coal. We know much less about what the plants that did not form coal looked like, even though they lived at the same time.

That asymmetry has an immediate consequence. Both sides of the coal debate discuss the plants visible in coal: how lignified they were, what proportion consisted of outer tissues, how resistant those tissues were to decay. For the corresponding questions about uplands, there is scarcely a comparable record. What grew there? What was the chemistry of its tissues? Where did its organic matter go?

The debate is therefore being conducted on a systematically biased sample. That does not invalidate either side, but it limits the range of questions either can answer.

Bias changes more than the numbers and shapes we recover. It also changes which questions can be asked at all. We can ask whether the degree of lignification in coal-forming plants determined the amount of coal because coal balls preserve those plants down to their cells.

For upland plants, the same question is much harder to ask in such detail. Fossils do exist; what is scarce is material comparable in resolution with the coal-swamp record.

The Joggins cliffs of Nova Scotia, Canada, now a World Heritage Site, have been called the “Galápagos of the coal age.” They preserve a different kind of favoritism. Petrified lycopsid stumps stand upright in the rock layers. After the trees died, their centers rotted out first, leaving a series of vertical holes.

Small tetrapods, arthropods, and land snails entered those holes. Some fell in and could not climb out; others may have lived there. Later, rapidly incoming sediment sealed them inside. In 1852, two geologists chipped away at one such fossil stump and uncovered the bones of a tetrapod. From then on, hollow stumps became places to look for small animals. The little creatures living on a forest floor thus received an unusually good chance of preservation.

Insects, meanwhile, are often preserved in lakes near coal swamps, in fine delta mud, or in exceptional fossil deposits. Mazon Creek in Illinois is one such place. More than three hundred million years ago, delta sediments enclosed even soft-bodied animals in iron-rich nodules. Open-cast coal mining brought thousands upon thousands of those nodules to the surface. Break one open and you might find a fern leaf, or a jellyfish. A jellyfish is almost all water; ordinarily, it leaves no fossil at all.

Among the animals there is the Tully monster, now the state fossil of Illinois. It is so peculiar that whether it was a vertebrate remains a matter of debate. The giant Meganeura specimens from France also come from fine-grained sediments associated with coal-bearing deposits, not from a lump of coal itself.

The fossil record therefore tends to amplify wetlands, river and lake margins, and lowland communities, while seriously underrepresenting the communities of seasonally dry uplands. This is where the sampling problem in Section Four begins. Maximum insect wing length is extraordinarily sensitive to exceptional fossil deposits, and those deposits are themselves disproportionately found near wetland environments.

The Joggins stumps have a sequel. One of the oldest secure amniote body fossils was found in a hollow stump there: an animal from a little over 310 million years ago that looked like a small lizard, twenty-odd centimeters long. In 2025, however, researchers reported clawed tetrapod tracks from Australia dating to about 356 million years ago and interpreted them as amniote-type footprints.

If that interpretation holds, the origin of amniotes must be pushed back at least to the earliest part of the Carboniferous. Footprints can tell us that animals with amniote-like feet were already walking on land. They cannot tell us whether those animals laid shelled eggs.

We met this once before in Episode Twelve, in the story of the move onto land: the footprints appeared before the feet. Here it is again. The reason is the same. Places where footprints are made and places where bones are buried are not the same kinds of places. Which appears first in the record, tracks or bones, cannot be read directly as the date when the ability itself evolved.

8 The Environment Turned Back; the Construct Did Not Kill Itself

Now return to the title. The “construct” here is the arrangement life built for itself: a lycopsid swamp together with its way of burying carbon. “The construct changed its environment, and the environment turned back to kill it” describes a machine, not a tale of retribution.

The chain runs like this. Swamps bury carbon underground, and carbon dioxide falls. Lower carbon dioxide, combined with Gondwana's position near the South Pole and accelerated rock weathering along mountain belts, allows the southern ice sheets to expand.

As the ice sheets grow and retreat, global sea level rises and falls. With those sea-level changes, coastal swamps are repeatedly established, exposed, and drowned. Eventually, a drying climate and increasingly fragmented wetlands dismantle the combination that had been burying carbon.

But the machine did not commit suicide. We saw in Section Five that the same sea-level cycle can strip away peat in its first half and bury it again in its second. The machine is closer to a metronome. Nor is there a paper that can lay the entire ice age at coal's door.

This series calls whatever a construct can neither accommodate nor eliminate its “remainder.” In this episode, the remainder lies in the explanations themselves. Think of a criminal case. Not one piece of physical evidence has disappeared. What changes is which pieces are joined together to tell which story.

We have taken apart two causal chains that have found their way into textbooks. One says, “Fungi arrived late, so there was coal.” Pull it apart and there are two statements of quite different strength. The enzyme system has an evolutionary history: that holds up. Its late arrival caused the coal peak: that does not. The popular version arose by running those two statements together as though they were one.

The other says, “Oxygen was high, so insects were large.” Its load-bearing support was the claim that diffusion at the ends of the tracheal system imposes a hard limit on size. The 2026 physiological study pulled out that support. It did not show that oxygen was irrelevant, but the chain had been resting on that mechanism all along.

The two chains came apart in different ways. For the first, we split one sentence into two and reread the original argument. For the second, we went and measured the machine itself.

Neither operation required a single new fossil. This gives the remainder here a property that those in the preceding episodes did not have: it can be repaired. The way to repair it is through a better understanding of mechanisms, rather than by finding more fossils.

And after the explanations have been taken apart, not one of the phenomena has gone away. The coal is still there. The Late Carboniferous and Early Permian still account for an exceptionally large share of today's coal resources.

The evidence for high oxygen is still there: most models and the charcoal record support oxygen above today's level. The giant insects are still there; those thirty-two-centimeter wings have not moved. The drastic decline of lycopsid forests is still visible in the spores and pollen.

Only their places in our causal chains have changed. What has been rewritten is not the world, but the line we draw to connect its phenomena.

You can try the same thing on other scientific stories. When you encounter “because A, therefore B,” split it into two questions. Is A true? And is A the principal cause of B? Examine them separately and see which one stands.

The next time you pick up a lump of coal, read it this way: a lycopsid forest standing in water; a basin sinking at just the right pace; a southern ice sheet swelling and shrinking; a sea that advances and retreats, then does it again. Throw the coal into a furnace, and in a few minutes you run tens of millions of years of that machine's work backward.

For now, climate and basins have the stronger case. Fungi remain in the running. The fungal nodes are still being redated, and the giant insects still await a complete new explanation. The criteria are still sifting.