Were Earth’s Five Mass Extinctions Really Sudden? New Fossil Research Challenges the Timeline

Key finding: An October 2026 report in Nature describes OneStrata, a machine-learning reconstruction built using information from nearly four million fossils. Its developers say that three of the traditional “Big Five” mass-extinction events look more like prolonged biodiversity declines than single abrupt crashes in their global timeline. This is a provocative result, not a final verdict that three catastrophes never happened. Independent palaeontologists point out that detailed regional records already reveal complicated sequences of extinction pulses. Read the Nature report.

The difference matters because a catastrophe can have an abrupt trigger and a long biological aftermath. Fossils record when organisms lived and died imperfectly; they do not timestamp every death. This guide explains the five events, what the model measures, why time resolution changes the apparent shape of extinction, and which conclusions readers should not draw from a headline.

What are the Big Five mass extinctions?

The Big Five are unusually large losses of biodiversity identified across the fossil record. They are commonly named for the geological intervals they ended: Ordovician, Devonian, Permian, Triassic and Cretaceous. Most quantitative reconstructions rely heavily on marine animals because shells and other hard parts are preserved more consistently than organisms living on land. The US National Park Service offers a useful introduction: Mass Extinctions Through Geologic Time.

These are approximate ages, not a claim that each event lasted exactly one day or one year. “Late Devonian” is especially important: the name covers a complicated series of biological crises. Different authors also use different taxonomic units—species, genera or families—which can produce different-looking graphs even when analysing broadly similar evidence.

What did the October 2026 OneStrata report actually say?

On 20 October 2026, Nature reported on a machine-learning model called OneStrata, developed from the OneStratigraphy fossil database. The researchers assembled information relating to nearly four million fossils and attempted to infer a coherent timeline of biodiversity changes. According to the report, the resulting global reconstruction makes three of the Big Five look more gradual than some earlier global summaries. The report followed a presentation at the Geological Society of America meeting in Denver on 12 October 2026.

Crucially, the Nature piece is a news report about research presented at a meeting, not itself a peer-reviewed methods paper validating every inference. Until readers can examine a complete accessible study, model diagnostics and independent replications, the safest description is “the researchers’ reconstruction suggests,” not “science has proved.” The article did not establish that every locality experienced identical timing or that extinction rates were always low.

Why train a model on fossils?

Rock sequences from different continents rarely line up like consecutive pages in a book. One location may preserve a thick sedimentary interval while another contains a gap caused by erosion or non-deposition. The same species can appear at different depths because sediment accumulates at different rates. Researchers use fossils, rock chemistry, radiometric ages and correlations between strata to assemble a global chronology. Combining thousands of partially overlapping sequences is computationally difficult.

Earlier work from the same research programme reported a marine biodiversity reconstruction with roughly 26,000-year temporal resolution for thousands of species. The newer machine-learning approach aims to scale inference to much larger data collections without the prohibitive computational cost of repeatedly solving every stratigraphic correlation from scratch. More fossils are helpful, but a large dataset is not a substitute for testing how preservation, sampling and model assumptions influence the result. Nature describes the model and its caveats.

Why “sudden” can mean different things

In ordinary language, sudden means seconds, hours or perhaps days. In geological research, an interval of 100,000 years may be considered relatively rapid when compared with a 500-million-year history. Conversely, an asteroid impact can occur in minutes, while extinction of organisms across oceans and continents may be recorded over much longer intervals. The trigger, environmental disruption, biological deaths and later recovery need not share the same duration.

Illustrative calculation, not a study result: Suppose two populations disappear 20,000 and 180,000 years after a disturbance. In a graph using one-million-year bins, both disappear within the same broad interval and may appear simultaneous. At 20,000-year resolution, the events may look separate. Reversing the problem, a series of losses over 200,000 years can appear as one sharp boundary when the data are coarsely grouped. This is why bin size must be reported alongside dramatic claims about speed.

What makes the fossil record incomplete?

Preservation bias

An animal without a durable skeleton is much less likely to fossilise than a shell-bearing organism. Deep-ocean mud, shallow reefs, forests and deserts preserve life differently. Environmental shifts can also change which sediments are deposited, making a group seem to disappear from a locality when its preferred habitat simply moved.

Sampling and geographic bias

Some geological formations have been studied for generations; others remain inaccessible or buried. Fossil collections are shaped by roads, funding, museum practices and the interests of individual researchers. An apparent global trend can partly reflect how many sites and specimens were available from each period. A strong reconstruction should show sensitivity tests that reduce the influence of heavily sampled regions.

The Signor–Lipps effect

The youngest fossil found for a species almost never represents the actual death of its final member. Rare species may have disappeared after their last preserved occurrence. This creates uncertainty around extinction dates even in otherwise excellent sequences. When many species are affected, the uncertainty can make a truly sharp biological event look spread out in time. It can also interact with models in ways that deserve careful testing.

The five events, viewed through a more careful lens

1. End-Ordovician: cooling and a changing sea

Late Ordovician ecosystems were predominantly marine. Ice growth, falling sea level and later environmental changes transformed habitats. The event is often discussed as more than one extinction pulse. That matters for the new debate: a reconstruction showing an extended decline may capture real biological phases rather than merely a flaw in dating. The exact balance between glaciation, ocean chemistry and recovery remains an active research question.

2. Late Devonian: a long and complicated crisis

The Late Devonian has long been considered a complex interval with multiple pulses rather than one single universally timed catastrophe. Reef ecosystems suffered greatly, and researchers have explored oxygen depletion, changing nutrient cycles, climate shifts and the expansion of land plants as possible contributors. A model depicting prolonged decline here is not automatically a radical reversal of established palaeontology; it may add global detail to an already nuanced picture.

3. End-Permian: the largest recognised crisis

Around 252 million years ago, life experienced the most severe of the Big Five. Massive volcanism in the Siberian Traps is strongly implicated in warming and changes to ocean oxygen and chemistry. A 2026 review in npj Biodiversity notes that some ecological recovery processes were rapid while full rebuilding of stable, complex communities took much longer. The speed of species loss and the speed of ecosystem recovery are separate measurements. Read the 2026 review.

4. End-Triassic: environmental upheaval before dinosaur dominance

The end-Triassic event, roughly 201 million years ago, coincides with immense volcanic activity associated with the Central Atlantic Magmatic Province. Carbon release and climate disruption are central explanations. Several biological changes occurred around the boundary, but different organism groups and locations do not always show identical timing. Dinosaurs survived and subsequently expanded into ecological roles vacated by other animals.

5. End-Cretaceous: a fast impact with cascading consequences

The Chicxulub impact approximately 66 million years ago is supported by the crater, a globally distributed iridium anomaly, shocked minerals and other geological evidence. Non-avian dinosaurs disappeared, while birds survived. The impact itself was abrupt, but darkness, food-web collapse and changes in marine and terrestrial communities played out across different timescales. A model that smooths global losses should not be read as disproving the impact.

How the new model compares with earlier research

In 2025, a paper in Scientific Reports argued that the Big Five show evidence of paired extinction-related episodes with varying gaps between them. The authors’ proposed chronology is itself an interpretation and should not be treated as universal consensus. It nevertheless illustrates why a neat one-spike-per-event graphic can oversimplify an enormous body of geological evidence. Read the 2025 paper.

Other palaeontologists have already used finely resolved local sequences to study extinction rates and pulses. The Nature report quotes an outside researcher welcoming the large global reconstruction while questioning whether its broad conclusions fundamentally overturn what well-studied sites already show. The useful advance may therefore be improved global integration, rather than a declaration that three extinctions were never rapid anywhere.

A practical guide to reading extinction headlines

When a news headline says a fossil study “rewrote history,” ask five questions: What was actually measured? Was it a fossil’s last occurrence, an extinction-rate estimate, or the date of an environmental trigger? What is the time resolution? A million-year bin cannot resolve a decade. How broad is the dataset? Marine invertebrates are not a census of every living thing. Is this peer reviewed? Conference presentations can be exciting but preliminary. Were alternative models tested? A result is more convincing when it survives different assumptions about missing fossils and uneven sampling.

Another useful check is whether a news story describes the loss of species, genera or families. A genus can contain many species, so a percentage loss in one category cannot be silently substituted for another. Likewise, a biodiversity curve showing the number of known taxa is not the same as a direct measurement of the number of individual animals that died.

Does this change what we know about a possible sixth mass extinction?

Not directly. The OneStrata report concerns how scientists reconstruct ancient fossil records. It does not provide a new count of modern extinctions or a forecast of how quickly present-day ecosystems will change. Comparing past events with today requires consistent definitions, accurate current biodiversity monitoring and careful handling of time scales. The distinction between extinction rate and total magnitude is particularly important. Our World in Data explains these concepts with historical context.

What ancient crises do show is that ecosystems can undergo both abrupt shocks and prolonged deterioration, and that recovery of ecological functions is not necessarily quick even after species numbers begin to rise. The Permian–Triassic recovery literature is a clear example of why measuring the return of species is different from measuring a stable food web.

What should researchers test next?

The strongest next steps would be to publish full model documentation, compare OneStrata against independently dated local sections, test alternative preservation assumptions and report uncertainty bands around each inferred extinction curve. Researchers could also check whether the same gradual pattern appears when particular regions or taxonomic groups are excluded. Such checks help distinguish a genuine biological signal from artefacts of global data assembly.

For readers, the lesson is less dramatic but more interesting than a simple reversal of history: a global fossil database can expose patterns that are hard to see at one site, while the most precise local rock records can reveal brief pulses that disappear when averaged across the world. Both scales are necessary.

Bottom line

The Big Five are not being erased. A new machine-learning reconstruction suggests that some of their global biodiversity losses unfolded over longer intervals than familiar textbook diagrams imply. The work could refine the timing of ancient ecological collapse, but its conclusions need scrutiny, accessible methods and comparison with existing high-resolution evidence. The central distinction is between the speed of a physical trigger, the pace of biological losses and the duration of recovery. Those are three different clocks.

Sources and further reading

Nature, 20 October 2026 — Vast new fossil record challenges story of Earth’s big five mass extinctions (news report on OneStrata and expert reactions).

US National Park Service — Mass Extinctions Through Geologic Time (overview and fossil context).

Scientific Reports — Mechanisms of global climate change during the five major mass extinctions (2025 research).

npj Biodiversity — The timing and nature of marine ecosystem recovery following the Permian–Triassic mass extinction (2026 perspective).

Our World in Data — Mass Extinctions (definitions and historical rates).