Trait-based species distribution models (trait-SDMs) enable prediction
to new species and situations based on traits. However, predictive
transferability is unknown. We fit trait-SDMs with specific leaf area
(SLA), maximum height and seed mass as species level predictors in
generalised linear mixed models with four environmental predictors for
20 species of eucalypt trees in an outlying reference region.
Trait-environment interactions included heavy-seeded species increasing
in rugged areas and high-SLA species increasing in areas receiving
runoff. We predicted occurrences using traits for 82 species across 18
target regions over >100,000 km2 in
south-eastern Australia. Median predictive performance for new species
in target regions was 0.65 (area under the receiver operating curve) and
1.24 times that of random (area under the precision recall curve).
Prediction in target regions did not worsen across geographic,
environmental or compositional space. This work provides a path for
first-order models of species distribution using traits.