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 ↗
