Skeletal traits for thousands of bird species v1.0

The dataset spans 2,057 species of birds (Aves: Passeriformes) and includes linear measurements of 12 skeletal elements from 14,419 individuals. In addition to the trait values directly measured from photographs, we leverage the multi-dimensional nature of our dataset and known phylogenetic relationships of the species to impute missing data under an evolutionary model. The traits included in the dataset are: the lengths of the tibiotarsus, humerus, tarsometatarsus, ulna, radius, keel, carpometacarpus, 2nd digit 1st phalanx, furcula, and femur; the maximum outer diameter of the sclerotic ring, and the length from the back of the skull to the tip of the bill (treating the rhamphotheca as part of the bill when it remains present on the specimen). These data are presented in three ways: 1) a dataset that only includes trait estimates for elements that were confidently identified and measured, 2) a complete specimen-level dataset that includes imputed trait values for all missing data, and 3) a species-mean dataset based on a model of trait evolution that includes estimates of the species mean, species-level standard error, variance, and 95% confidence interval of the trait estimates.

Methods
These data were collected from museum skeletal specimens. To measure traits, images were taken of skeletal specimens and then Skelevision, a computer vision method, was used to segment out the bones in the images, identify them, and measure them; this method is described in detail in Weeks et al. (2023). In addition to presenting the data that were generated using Skelevision, we generated a 100% complete dataset by imputing all missing values in the dataset using Rphylopars (Goolsby et al. 2017), which is a method for fitting multivariate phylogenetic models and estimating missing values in comparative data.  We also present species-level means along with associated estimates of uncertainty derived from the Rphylopars model. We validated the Skelevision estimates by comparing them to handmade measurements, and we assessed the trait imputation accuracy by withholding data and imputing the withheld values. The validation procedure and results are outlined in detail in Weeks et al. (2024).

References: 

Goolsby, E.W., J. Bruggeman, and C. Ané. 2017. Rphylopars: fast multivariate phylogenetic comparative methods for missing data and within-species variation. Methods in Ecology and Evolution 8(1): 22-27.

Weeks, B.C., Z. Zhou, B.K. O'Brien, R. Darling, M. Dean, T. Dias, G. Hassena, M. Zhang, and D.F. Fouhey. 2023. A deep neural network for high-throughput measurement of functional traits on museum skeletal specimens. Methods in Ecology and Evolution. 14(2): 347-359.

Weeks, B.C., Z. Zhou, C.M. Probst, J.S. Berve, B.K. O'Brien, B.W. Benz, H.R. Skeen, M. Ziebell, L. Bodt, and D.F. Fouhey. 2024. Skeletal trait measurements for thousands of bird species. bioRxiv. https://doi.org/10.1101/2024.12.19.629481