Semiautomated identification of a large number of hoverfly (Diptera: Syrphidae) species based on wing measurements

Abstract

Automated species identification is a promising new method of solving the problem of identification of insects. However, in most previous studies based on this method, the number of identified species was relatively low. In this study, we tested the possibility of the identification of more than 200 species of Syrphidae. The identification was based on wing measurements involving the indication of 13 landmarks. Either single-level linear discriminant analysis (LDA) or hierarchical LDA was used for classification. Identification success varied widely between species. In the case of a single level LDA, 40% of species were identified without error. Overall species classification success of a single level LDA was 89%. In the case of hierarchical LDA, overall classification success was lower; however, identification success at the lowest level of hierarchical classification was higher. Identification based on wing measurements proved to be helpful for the identification of a large number of hoverfly species. Even if the identification of some species was not effective, it allowed the identification of genera or groups of species.