Extinction probabilities of small structured populations: adequate short-term model predictions in Folsomia candida

Population management requires predictions of extinction risk based on a general understanding of these risks and on system-specific modelling. Life tables, available for numerous populations and species, permit calculating population growth and the construction of multi-type branching process models which predict population survivorship and ultimate extinction probabilities. We exemplify this approach and tailor it to an experimental model to predict extinction probabilities per unit of time. In age-structured populations, founders from different age classes lead to different predicted extinction probabilities. Age effects interact with environmental effects such as culling levels, which influence population growth rates. We assess the accuracy of predictions based on an age-structured matrix model, in an extinction experiment over an eight-week period on the springtail Folsomia candida, with crossed founder age and culling level treatments. Using parameter estimates from an accessory experiment, the fit of model predictions to observed extinction probabilities was generally good. A modified branching process model which allowed culling events between and at observations reduced prediction error. However, additionally maximizing the likelihood of observed extinction probabilities based on survival and fecundity parameters, or on a parameter which concentrated fecundity within a subinterval, did not significantly reduce prediction error according to the AICc. Our study shows that satisfactory predictions of establishment probabilities and of the initial persistence of small populations can be made using multi-type branching processes and available parameter estimates. Predictions can be improved by integrating knowledge of when events occur within intervals. This can be done without additional parameter estimation.