Outline shape analysis of penguin humeri: a robust approach to taxonomic classification

Abstract

Humeri have been useful bones in taxonomic determinations of extinct penguins. In the context of neontological taxonomic studies, however, their potential remains unsatisfactorily explored. Here, the variation of the overall closed-outline shape of 60 humeri, assignable to five genera of extant penguins, was investigated. A set of normalized outlines was quantified via elliptical Fourier analysis and subjected to linear discriminant analysis on principal component scores extracted from harmonic coefficients. These geometric representations proved to be a source of easily extractable genus-level taxonomic information. The constructed model provided meaningful discrimination between taxa: the first two linear discriminants captured almost 90% of between-group variance. A cross-validation method based on jackknifing yielded 93% correct identifications, and statistically significant differences between group centroids were also detected (multivariate analysis of variance, p < 0.05). Predictions of genus membership for the intentionally noisy test data (20 outlines) were accurate in 80% of cases.