Trait diversity metrics can perform well with highly incomplete datasets

O_LICharacterizing changes in trait diversity at large spatial scales provides insight into the impact of human activity on ecosystem structure and function. However, the approach is often based on trait datasets that are incomplete and unrepresentative, with uncertain impacts on trait diversity estimates. C_LIO_LITo address this knowledge gap, we simulated random and biased removal of data from a near complete avian trait dataset (9579 species) and assessed whether trait diversity metrics were robust to data incompleteness with and without using imputation to fill data gaps. Specifically, we compared two commonly used metrics each calculated with two methods: trait richness (calculated with convex hulls and trait probabilities densities) and trait divergence (calculated with distance-based Rao and trait probability densities). C_LIO_LIWithout imputation, estimates of global avian trait diversity (richness and divergence) were robust when 30-70% of species had missing data for four out of 11 continuous traits, depending on severity of bias and the method used. However, when missing traits were imputed based on present morphological trait data and phylogeny, trait diversity metrics consistently remained representative of the true value, even when 70% of species were missing data for four out of 11 traits and data were not missing at random (biased with respect to body mass). Trait probability densities and distance-based Rao were particularly robust to missingness and bias when combined with imputation, with convex hull-based trait richness being less reliable. C_LIO_LIExpanding global morphometric datasets to represent more taxa and traits, and to quantify intraspecific variation, remains a priority. In the meantime, our results show that widely used methods can successfully quantify large-scale trait diversity even when data are missing for two-thirds of species, so long as missing traits are estimated using imputation. C_LI