MECHANISTIC MODELS FOR TERRESTRIAL EXTINCTION EVENTS

Patterns of survivorship in clades and guilds across mass extinctions have varied wildly in terrestrial communities. For instance, freshwater vertebrates suffered high extinction levels in the end-Tr but had high survivorship in the end-K. Although each of these extinctions is thought to be comparable in terms of proximate mechanism (primary productivity shutdown), the ultimate causes are debated (e.g., volcanism vs bolide impact). The different patterns have led, however, to discussions focused on the dynamics in each extinction as a distinct event, and the seeming unpredictability of extinction potential has been typified by differential responses among clades to mass extinctions. However, even if two organisms were identical in all ecologically relevant characteristics, and their ecosystems were perturbed in identical ways, differing ecosystem structures would cause variance in extinction potential. “Identical” large carnivores may, for instance, have different survival probabilities in ecosystems with broader prey bases than in ecosystems with only a few hyper-abundant prey species. Essentially, examining the patterns of extinctions within individual clades, and trying to extrapolate to overall extinction processes, is impossible without considering the broader ecological milieu in which those clades existed. The concept that an individual species (or clade) will respond differently to the same perturbation based on the larger network it exists in has a long history in ecology. By using preliminary data on terrestrial taxa in individual basins that span three mass extinctions (Karoo Basin in South Africa, Chinle and Newark Groups in USA, Hell Creek Fm. in USA), I tested a mechanistic model of extinction based on trophic interactions in three different ways (topological, Bayesian Network, and CEG) and compared their predictive accuracy (% extinct taxa predicted) to those of phenomenological models (body size, diet category and habitat type each having their own extinction potential). Overall, the food web models performed as well as the phenomenological models for both background and the Tr-J and K-Pg extinction events (measured via likelihood), and all out-performed null models in most cases, although the high extinction levels at the P-Tr render the models indistinguishable from one another.