Skip to content
Ecological Intelligence

Ecological evidence.
Independent perspective.
Explore Wild Lands ↗

Wild Lands / Ecological Intelligence

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.

A detailed biodiversity map is not necessarily a reliable local prediction
Clint Wirick/USFWS — illustrative comparison of grazed and ungrazed prairie, not a PREDICTS sampling site or model output. · USFWS Public Domain · Rights: Public domain
Evidence
Peer-reviewed
Geography
Global terrestrial evidence, 101 countries
System
Terrestrial biodiversity; land-use pressure; spatial modeling

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 ↗