HOW PREDICTABLE IS EXTINCTION? FORECASTING SPECIES SURVIVAL AT MILLION-YEAR TIMESCALES
To address these questions we analyze the fossil record of Cenozoic planktonic microfossil taxa. Our best supported model of species survival includes the change in geographic range and lag of global temperature as covariates, which indicates that the past improves our estimates of the present and future. The improvement in predictive power by including these historical covariates is modest at best and reflects the rarity of extinction events (i.e. class-imbalance) and the extremely stochastic nature of species survival. This result implies that at million-year timescales geographic range trajectories are nearly Markovian, perhaps because the processes driving geographic range changes vary on substantially shorter timescales. The effect of change in geographic range on survival most likely stands for many interacting and unobserved processes which in-turn produce that species' geographic range and its affect on survival.
Finally, we find support for species' extinction risk increases with age, though the strength of this effect varies among taxonomic groups. We also find that the time of an observation is a greater source of variation in survival probability when compared to the effect of species age.
Ultimately, we find that including information on a species' change in geographic range size on average improves our predictions of species survival at million-year timescales. However, the effect of change in geographic range is much smaller than the effect of current geographic range, and highly variable through time.