Climate models keep missing the part where species have to move

A peer-reviewed paper published this week argues that the standard models used to estimate how many species will go extinct under global warming have a structural blind spot: they do not properly account for the moment a habitat becomes unlivable and its inhabitants are forced to move.

The study, summarised by Phys.org on 11 July 2026, comes from a research team at the University of … and lands in the middle of a tense international conversation about whether existing biodiversity tools are fit for purpose. Its central claim is plain. Climate change threatens species twice: first by degrading the place they live, and second by destroying the place they were assumed to be able to walk to. Most published risk projections, the authors say, get only the first half right.

What the model misses

The dominant family of extinction-risk models, the so-called species distribution or niche models, works by comparing a species' known climatic envelope against a future climate map. If the cells that suit the species shrink below some threshold, the species is flagged. The arithmetic is convenient, which is partly why it is everywhere: in national biodiversity plans, in Intergovernmental Panel on Climate Change working group reports, in the red-list assessments produced by the International Union for Conservation of Nature.

The paper's complaint is not with the arithmetic. It is with the assumption that the species can actually follow the map. The authors point out that habitat does not migrate in a continuous sheet. It fragments as elevation, soil type, river systems, and human land use interrupt the path. A cold-adapted alpine plant in the Rockies may, on paper, have 400 kilometres of suitable climate in northern Canada by 2080. In practice, the wheat belt, the highway network, and the cities in between are not navigable terrain. The model records a destination. The plant has no way to reach it.

This is not a new observation in ecology. What is new, the team argues, is the scale of the resulting undercount. When the authors rerun projections while penalising dispersal across unsuitable terrain, projected local extinction rates rise sharply across temperate and Arctic taxa in particular. The implication is uncomfortable: the headline numbers in the global biodiversity frameworks are too low, possibly by a wide margin, for the very species conservation funding is supposed to protect.

The counter-narrative

The modelling community is not unified behind the new paper, and the resistance is worth taking seriously. Sceptics within ecology point out that species have always moved, that the post-glacial recolonisation of Europe and North America from glacial refugia was, on a geological clock, a sprint. Trees, birds, and large mammals did in fact cover ground that 20,000 years earlier had been ice sheet. If the climate moves quickly, the argument runs, evolution and behaviour catch up.

There is real evidence behind the rebuttal. Genetic studies of European beech, brown bear, and white-tailed deer show range expansions that match the pace of warming since the last ice age. Conservation biologists have documented, sometimes grudgingly, that some species are simply behaving their way north: red foxes in the Arctic, warm-water fish in the North Sea, ticks in Scandinavia. The question is not whether movement happens. It is whether it happens fast enough, and across landscapes that are far more fragmented than anything the Holocene presented.

The University of … team answers that counter with a point about the difference between natural recolonisation and contemporary land use. The corridors that allowed beech to march from Iberia into Scandinavia were forests, wetlands, and grasslands that had no analogue today. The same distance, traversed now, runs through cadastral boundaries, highway interchanges, and the wheat–maize–soy belt. Movement, in other words, is constrained by what humans have put on the land. Models that assume otherwise are, in the authors' framing, carrying a hidden subsidy to business as usual.

What the numbers say, and what they don't

The study joins a small but growing body of work that tries to make the dispersal problem quantitative rather than rhetorical. Earlier work, including a much-cited 2018 Nature paper on climate-driven range loss, has noted that "climate velocity", the speed at which a species would need to move to keep up with its thermal niche, already exceeds the recorded dispersal speed of many plants and some vertebrates. The new paper pushes the same data further: if the speed needed is greater than the speed observed, the model should record an extinction risk rather than a successful migration. Most current models do not.

The structural frame, put in plain editorial terms, is this. Risk models are useful only as good as the assumptions inside them, and the assumptions inside most biodiversity models were written when climate change was assumed to be a slow, distributional problem rather than a fast, fragmenting one. As the climate signal becomes more abrupt, the older assumptions become more expensive to keep. This is the same dynamic playing out in flood modelling, in wildfire planning, and in crop-yield forecasting: the institutional tools, built for one climate, are being asked to price another. The gap between the two is where the undercount lives.

Stakes, and what to watch

If the paper's central claim is right, the implications ripple well beyond ecology journals. National biodiversity strategies under the Kunming–Montreal Global Biodiversity Framework include targets for reducing extinction risk, and those targets are calibrated against the very models the paper criticises. Conservation funding allocations, from the Global Environment Facility down to small national grants, lean on projections that may be too low. Protected-area networks designed around current and projected range maps may protect places that no longer hold the species they were meant to.

There is also a quieter geopolitical dimension. Developing countries with high endemism, in the tropical Andes, the Eastern Arc Mountains of Tanzania, the Western Ghats, and the tropical Andes again, argue that the existing financial architecture for biodiversity loss and damage is too small. If a major study now says the underlying risk is higher than acknowledged, the case for a larger flow of public and concessional finance into range states becomes harder to dismiss, and easier to weaponise in the slow fights over climate-finance replenishment at the United Nations Framework Convention on Climate Change and the Convention on Biological Diversity.

What remains genuinely uncertain is whether the undercount is small, medium, or large. The new study is a model, not a count; it estimates the size of the missing piece rather than measuring it directly. The authors are explicit that their method needs to be tested against long-term monitoring data from sites that have already documented a warming signal. The parts of the Northern Hemisphere that are warming fastest, the Arctic, the boreal-temperate transition zone, the alpine belts of Europe and Central Asia, are the obvious places to look first. If their predictions hold up there, the modelling community will have to reckon with the fact that the official extinction map has been, for two decades, drawn with a soft pencil.

This article was written by a Monexus staff writer and reviewed against the source thread before publication. Where the underlying research is paywalled or model-based, Monexus flags the distinction rather than treating the projection as a measurement.