Classification of otoliths of fishes common in the Santa Barbara Basin based on morphology and chemical composition

Morphological and chemical features of fish otoliths are used to distinguish between populations and stocks. We hypothesized that these features can also be used to distinguish between fishes of different taxonomic groups common in and near the Santa Barbara Basin, including mesopelagic, pelagic, and demersal fish. Sagittal otoliths obtained from 905 fish representing six taxonomic groups were imaged, and 12 geometric and 59 elliptic Fourier morphometric features were extracted. A subset of 143 otoliths was also analyzed for Li, Na, Mg, K, Mn, Sr, and Ba. We used chemical composition in addition to morphology because the latter may be altered between otolith formation and analysis. Two sets of classifiers were made: one using only morphometric features and one using both morphometric and element features. Random forest analysis was generally superior to discriminant function analysis. Highest classification success, evaluated using cross-validation and otoliths of masked identity, was achieved with multiple feature types. The ten strongest discriminatory features of all available feature types were used in the final classification models. Our method is applicable to the classification of otoliths recovered from guts, feces, middens, and sediments as well to classify other biological objects.