TESTING THE BOREAL-TETHYAN SHIFT OF NERINEOID GASTROPODS USING CONVOLUTIONAL NEURAL NETWORKS

Abstract The southward biogeographic shift of shallow warm-water fauna from the Boreal to the Tethyan Realm as a result of cooling in higher latitudes has been suggested for several groups, including nerineoid gastropods. Yet, this southward ‘Boreal spread’ was not quantitatively evaluated. The Nerineoidea, a group of extinct gastropods, was one of the most abundant and diverse groups in Jurassic-Cretaceous shallow marine environments. Internal spiral folds that are viewed in polished axial sections characterize the group and form the basis of their classification. However, conflicting taxonomic classifications hindered large-scale analyses based on species occurrences. To address this, we compiled a dataset of nerineoid images from published literature and our own data, to train a convolutional neural network (CNN) to classify images into three predefined groups: Ceritellidae, Ptygmatididae, and a combined group of Nerinellidae+Eunerineidae+Nerineidae. By customizing a pre-trained model, we achieved 89% success, where nerineoid whorl outlines are highly compatible for automation. The analysis of the nerineoid image-based dataset displays a southward expansion from the Early Jurassic Boreal Realm to the Middle Jurassic Tethyan Realm, consistent with biogeographic shifts in other macrofaunal groups. Variations in the relative frequency of occurrence of the nerineoid groups may help to shed new light on their adaptation to different paleoecological settings, possibly owing to their different fold structures. Our automated approach makes nerineoid studies more accessible and facilitates dataset expansion easily.