Climate models are quietly underestimating extinction risk, a University of Göttingen team finds
A new study argues that standard species-distribution models miss the mechanics of forced migration. The gap could reshape how extinction debt is calculated across warming continents.
On 11 July 2026 a team of researchers at the University of Göttingen published findings that put a sharp question to one of the most widely used tools in climate biology: the species distribution model. Their answer, in plain terms, is that these models routinely miss the extinction risk that comes not from a habitat disappearing, but from a habitat moving.
The distinction matters. For two decades, conservation planners have leaned on correlative models that ask, of a given species: where does it live today, and what climatic conditions correlate with that range? The future is then sketched by overlaying climate projections on the same envelope. The Göttingen paper argues that the envelope is the wrong object to draw. What kills a population, in many cases, is the journey it must make to keep up with shifting climatic zones, the dispersal gauntlet, the fragmentation, the lag. None of that shows up in a static map.
What the models miss
Species distribution models, in their correlative form, take presence records and associate them with climate variables: mean temperature, precipitation seasonality, snow cover duration. The model learns the envelope and projects it forward. The Göttingen team's critique, as reported in the 11 July 2026 study, is that the technique treats a population's location as a free variable. The animals and plants are assumed to be wherever the climate is right. Reality is messier: a montane plant cannot migrate across a highway interchange. A forest mammal cannot re-establish in soil that has been tilled for a century. The gap between suitable climate and reachable habitat is, in many ecosystems, the actual driver of decline.
The researchers frame this as a process error rather than a data error. Adding more occurrence records does not fix it. What is needed, the paper argues, is mechanistic modelling: explicit treatment of dispersal ability, generation time, landscape permeability, and the demographic stochasticity of small founder populations. None of these inputs are standard in the correlative workflow that dominates applied conservation work today.
The counter-narrative from applied conservation
The modelling mainstream is unlikely to abandon correlative tools quickly, and for defensible reasons. They are computationally cheap. They scale to thousands of species in a single run. They produce the map-shaped outputs that funders, ministries, and intergovernmental assessment panels are set up to consume. The Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services and national red-list authorities both depend on workflows built around the correlative envelope. Replacing the engine mid-flight is not a small ask.
There is also a counter-argument from within ecology: that mechanistic models are only as good as their parameterisation, and that dispersal data for most of the world's described species are sparse to non-existent. A poorly parameterised mechanistic model can produce more confident-looking wrong answers than a transparently limited correlative one. The Göttingen paper does not dispute this. It argues, rather, that the correlative baseline systematically understates risk in a way that has policy consequences, and that even imperfect mechanistic layers should be bolted on where the data permit.
What the structural shift looks like
In practical terms, the finding nudges climate-biology practice toward a hybrid default. The correlative envelope remains useful as a first-pass map of where the climate is heading. On top of it, dispersal and landscape-resistance layers, already standard tools in circuit-theory connectivity analysis and least-cost-path modelling, become mandatory rather than optional. The redrawn workflow treats the species not as a point on a map but as a population with a movement budget that may or may not be sufficient to track its climate.
This matters most in fragmented landscapes: the temperate agricultural belts of Europe, the eastern deciduous forest of North America, the heavily logged zones of Southeast Asia. In a continuous wilderness like the boreal shield of northern Canada or the Amazon basin, dispersal lag is a smaller share of total extinction risk. In a checkerboard of fields, roads, and towns, it is the dominant share. The Göttingen result, if it holds up to replication, implies that the extinction debt already accumulating in temperate agricultural zones is larger than current assessments suggest.
What to watch
Three signals will tell whether the paper's argument is being absorbed. First, the next round of national red-list assessments, due in many European countries within the 2027-2028 cycle, and whether they cite mechanistic dispersal layers alongside correlative envelopes. Second, the response from the major integrated assessment modelling groups, particularly those feeding into the next IPBES global assessment, and whether their species-risk workflows get re-engineered or merely annotated. Third, the demand side: whether conservation funders begin to require mechanistic layers as a condition of grant support, the way they once required climate-envelope projections.
What remains uncertain is the magnitude of the correction. The paper argues that correlative models underestimate risk, but does not, on the basis of the available reporting, give a clean multiplier. A reader who wants to know whether the global extinction estimates published over the last decade are off by ten percent or by a factor of two will have to wait for the follow-up work. The structural point, that the dominant workflow has a known blind spot and that the blind spot is policy-relevant, is now on the record. The calibration of the correction is the next contest.
This piece treats the Göttingen result as a methodological finding with downstream policy consequences rather than a single new extinction number. Monexus frames the story around the modelling workflow, where the durable shift will occur, rather than around the headline statistic, where the dispute will be noisier.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- https://en.wikipedia.org/wiki/Species_distribution_modelling
- https://en.wikipedia.org/wiki/Climate_change_and_biodiversity