DEEPDIVER: DEEP LEARNING ESTIMATION OF PALAEODIVERSITY IN R AND PYTHON

Variation in the spatial, temporal, and taxonomic scope of the fossil record inherently biases attempts to estimate biodiversity through geological time even after standard correction methods have been applied, especially at the largest spatial scales. Due to this restricted view, the past diversity dynamics of many clades remain uncertain or unknown. Here we present new software, named DeepDiveR, that implements simulation-based training of deep learning models to estimate biodiversity through time while accounting for the incompleteness of the fossil record in R and Python. We find that the method reduces error compared to state-of-the-art methods in the field and can be customized to suit a wide range of clades. Additional features available in the analytical framework such as conditioning the estimates of diversity patterns of extant clades on modern diversity contribute to further increasing the accuracy of the predictions. We demonstrate DeepDiveR in analysis of the carnivoran fossil record and find a peak of diversity in the Miocene with more than 200 genera followed by loss of over 45% of genera between the end of the Miocene and today.