Taxonomic classification of seabird long bones using 3D shape

Abstract

Identifications from isolated bones using molecular methods are not always possible (e.g. fossils and older sub-fossils), and specialist taxonomic expertise is not always available. Shape-based, geometric morphometric methods for identifying taxa from a single bone can provide a potential solution for faunal studies requiring taxonomic information about isolated bones. We present a generalisable method for taxonomic classification using 3D shape analyses combined with partial least squares discriminant analysis (PLSDA). Our classification protocol uses both well-established landmark-based techniques and novel landmark-free (i.e. pseudolandmark-based) shape analysis methods. Landmark-free methods can improve the automation of the classification method. We demonstrate with three case studies that 3D shape-based PLSDA can accurately predict taxonomic identities for seabird long bones to at least family-level. Landmark-based PLSDA assigned femora from penguins and tubenosed birds to the correct order and family with 100% accuracy, sensitivity and specificity. Pseudolandmark-based PLSDA assigned humeri from penguins and tubenosed birds to the correct family with 100% accuracy, sensitivity and specificity. Our 3D shape-based classification method could reliably contribute to the identification of bone remains when only isolated or typically nondiagnostic bones (i.e. elements with comparatively little variation in discrete character states or measurement proportions when comparing between closely-related taxa) are available, and when molecular methods are not possible or feasible. We provide a framework for applying this method to 1) much larger datasets which may allow finer-level classifications (genus or species), and 2) groups beyond seabirds. Show more