Comparative analysis of otolith morphology in icefishes (Channichthyidae) applying different statistical classification methods

The genus Chionodraco includes three morphologically similar species, two of them (C. hamatus and C. myersi) living sympatrically in the Eastern Antarctica, and the third one (C. rastrospinosus) being distributed in the Western Antarctica along the southern Scotia Arc. The few overlapping diagnostic characters are often useless for the right taxonomic identification of three species, complicated by the presence of hybrids between them. In the present study, we tested the discrimination power of sagittal otolith shape as a tool to distinguish among these closely related species. An array of shape indices and elliptic Fourier descriptors were individually obtained from sagittal otoliths and analysed using three different models for species classification: Linear Discriminant Analysis (LDA), Random Forest (RF) and K-nearest neighbour (KNN). According to the characteristics of the adopted metric (precision), none of the methods resulted to be the absolute best approach as the performance largely varied with the species. The overall precision was 63 %, 66 % and 70 % for LDA, RF and KNN respectively, so that KNN can be considered the best classifier model. As expected, C. myersi was the best recognized species by all classifier models, consistently with its morphological and phylogenetic difference from the other two species. Taking into account the observed pattern of otolith morphology of the three species and their own geographical distribution, we were able to evaluate the role of genetic, endogenous or environmental components in influencing otolith size and shape.