The relative performance of geometric morphometrics and linear-based methods in the taxonomic resolution of a mammalian species complex.

Morphology-based taxonomic research frequently applies linear morphometrics (LMM) in skulls to quantify species distinctions. The choice of which measurements to collect generally relies on the expertise of the investigators or a set of standard measurements, but this practice may ignore less obvious or common discriminatory characteristics. In addition, taxonomic analyses often ignore the potential for subgroups of an otherwise cohesive population to differ in shape purely due to size differences (or allometry). Geometric morphometrics (GMM) is more complicated as an acquisition technique but can offer a more holistic characterization of shape and provides a rigorous toolkit for accounting for allometry. In this study, we used linear discriminant analysis (LDA) to assess the discriminatory performance of four published LMM protocols and a 3D GMM dataset for three clades of antechinus known to differ subtly in shape. We assessed discrimination of raw data (which are frequently used by taxonomists); data with isometry (i.e., overall size) removed; and data after allometric correction (i.e., with nonuniform effects of size removed). When we visualized the principal component analysis (PCA) plots, we found that group discrimination among raw data was high for LMM. However, LMM datasets may inflate PC variance accounted in the first two PCs, relative to GMM. GMM discriminated groups better after isometry and allometry were removed in both PCA and LDA. Although LMM can be a powerful tool to discriminate taxonomic groups, we show that there is substantial risk that this discrimination comes from variation in size, rather than shape. This suggests that taxonomic measurement protocols might benefit from GMM-based pilot studies, because this offers the option of differentiating allometric and nonallometric shape differences between species, which can then inform on the development of the easier-to-apply LMM protocols.