A detailed biodiversity map is not necessarily a reliable local prediction
A new global validation study finds that models of biodiversity pressure and response lose predictive skill when moved into unfamiliar ecological settings.
WHAT THE EVIDENCE SAYS
Researchers used the PREDICTS biodiversity compilation, including observations from 101 countries, to evaluate alpha- and beta-diversity models against standard and cross-study validation. Models organized by biome, realm and taxonomic group improved interpolation compared with study-random-effect structures, but none transferred reliably to held-out ecological contexts. Human land-use signals remained visible, while site-level predictive accuracy was low.
WHY IT MATTERS / OUR READ
The familiar fine-grained biodiversity maps are frequently finer than the ecological evidence used to train them. Landscape-scale screening is still useful for prioritizing fieldwork; parcel or corridor decisions need local data and validation, not confidence borrowed from map pixel size. This is a direct challenge to treating high-resolution global biodiversity indicators as measured local condition.
CAVEATS / UNCERTAINTY
This evaluates particular model families and PREDICTS-derived historical records, not every available indicator. Poor local extrapolation does not invalidate broad direction-of-effect inferences. Stronger national data may improve predictions.
SOURCE
Nyström et al., Nature Ecology & Evolution (10 Nov 2026)
View original evidence ↗
