Otolith shape as a tool for species identification of the grenadiers Macrourus caml and M. whitsoni

Accurate species identification of harvested fishes is a central, yet often overlooked, component of fishery monitoring. This study examined the efficacy of using otolith shape to differentiate between the morphologically similar grenadiers Macrourus caml and M. whitsoni and validate species identifications by fishery observers within and adjacent to the Ross Sea region, Antarctica. Otolith shape information was collected from 610 M. caml and 329 M. whitsoni otoliths from research collections, where confidence in species identification was high, and 2558 samples collected by fishery observers from commercial longline catches. Univariate and linear discriminant analyses of research-collected samples revealed consistent differences in otolith shape between the two species, with otoliths of M. caml being larger and more elongate than those of M. whitsoni. No significant effect of collection depth or fishing season on otolith shape was evident for either species, and no significant effect of sampling location on otolith shape was evident for M. caml. Otolith shape of M. whitsoni varied slightly among sampling regions, although the extent of these differences was less than those between species. To validate observer identifications, a random forest (RF) model was trained using otolith shape data of 75% of the research collection samples, validated on the remaining 25%, and used to predict the species of fishery observer-collected samples. The RF model achieved high classification success for individuals from the research collections, with an Out-Of-Bag error rate of 5.97% for the training dataset. Overall classification success of individuals in the validation dataset was 96.2%, with 96.8% of M. caml and 94.8% of M. whitsoni correctly classified. Using this model, 90.6% of M. caml and 85.7% of M. whitsoni sampled by fishery observers were predicted as being correctly identified. Individual observer identification success ranged from 50.5% to 98.2%. The reliable and predictable differences in otolith shape observed between the two species indicates that our approach can be applied to ongoing or archived otolith collections to confirm species identification of fishery-sourced samples to improve the accuracy of fisheries monitoring, facilitate assignment of previously collected material to species level, develop high-confidence datasets for further biological analyses, and to understand and prioritise observer training needs. More broadly, our results highlight the potential for using otolith shape as an effective tool for assessing the accuracy of species identifications in fisheries sampling programs.