Combining geometric morphometrics with pattern recognition for the investigation of species complexes

Abstract

The parasitoid Hymenoptera contain a large number of species complexes that are as yet unresolved by traditional taxonomic studies. Evolutionary studies as well as biological control programmes often require further investigations which cannot rely on simple qualitative morphological characters; in many cases, particularly with dried specimens housed in museums, molecular approaches cannot be used. Recent developments in geometric morphometrics and statistical exploratory approaches open new perspectives for the objective evaluation of morphological characters in this taxonomic context. In this study, geometric morphometrics and pattern recognition approaches were applied to the wing shape and venation of two closely related braconid species considered to differ by subtle qualitative morphological head characters. Exploratory analyses such as kernel density estimates and Gaussian mixture analyses were used to explore the structure of the data in the multivariate morphometric space. Discrimination techniques (linear discriminant functions and neural networks combined with cross-validations) were used to estimate the taxonomic value of qualitative characters. Gaussian mixtures highlighted the existence of two non-overlapping groups. A good congruence was found between one of the two groups and the a priori defined Bassus tumidulus. The misclassification rate was higher for B. tegularis specimens, which also appeared morphometrically heterogeneous. Discrimination between the two a priori defined species was incomplete with misclassification rates higher than, or equal to, 6%. In most cases, the lack of congruence between species and morphometrically defined subgroups could be related to specimens that exhibited ambiguous qualitative character states. In summary, if two entities are present, they still need to be defined morphologically, while B. tegularis heterogeneity calls for further investigation of specimens of known origin and hosts. © 2003 The Linnean Society of London, Biological Journal of the Linnean Society , 2003, 80 , 89‐98. ADDITIONAL KEYWORDS: cross-validation ‐ Gaussian mixture analysis ‐ kernel density estimates ‐ linear discrimination ‐ neural networks ‐ Procrustes superimposition ‐ shape ‐ size ‐ wing venation.