DISENTANGLING OCCUPANCY FROM SAMPLING BIASES TO INFER THE EXTINCTION DYNAMICS IN PHANEROZOIC MARINE BIVALVES

Neutral evolutionary theory predicts that a taxon’s occupancy, manifesting in its population size, abundance, or geographic range, follows a hump-shaped trajectory: it is low at the taxon’s origination, peaks somewhere during its lifetime, and then declines prior to its extinction. Alternatively, stasis – a prolonged constant occupancy – necessitates a change in biotic interaction strength over the taxon’s lifespan, while a linear rise followed by an abrupt population collapse suggests a rapid change in climate or a catastrophic event. Occupancy dynamics can help us validate alternative hypotheses about the causes and drivers leading to the extinction of lineages and are therefore pivotal to our understanding of macroevolutionary diversification processes. Previously, occupancy history has been approximated from fossil occurrences of selected marine taxa. However, the inevitable incompleteness of the fossil archive, affected by spatial, temporal, and taxonomic biases, means that fossil occurrences cannot be reliably taken at face value. Here we present a novel Bayesian model to analyze the fossil record using an unsupervised neural network and explainable artificial intelligence techniques to infer the dynamics of occupancies throughout the lifespan of lineages while accounting for sampling biases. The model also incorporates the effects of life history traits, biogeography, systematics, major biotic events, and spatiotemporal environmental conditions on sampling rates, allowing us to test hypotheses about the link between these factors and occupancy. After benchmarking the model with simulations, we apply it to a fossil dataset of 2500 marine bivalve genera spanning 450 million years. Our approach advances our ability to infer occupancy in deep time and our understanding of the population dynamics preceding extinction.