ECOLOGICAL NICHE MODELING OF BIRDS THROUGH TIME: A COMPARISON OF MAXENT AGAINST OTHER ALGORITHMS

Ecological niche modeling (ENM) is a technique used to quantify habitat suitability for species to address various ecological and evolutionary questions using both modern and fossil data. Various algorithms can be used to identify a relationship between species’ distributions and environmental parameters. Previous research has made significant strides forward in determining how model choice and parameterization affects model outcomes. However, the effects of these choices on model outcomes remain poorly constrained when considering fossil systems.

Here, we examine the effects of varying model parameterization for five algorithms: maximum entropy, generalized linear, generalized boosting, random forest, and surface range envelope. More specifically, we compare the performances of these single algorithms run in biomod2 to ensembles run across multiple algorithms. We also compare two pseudoabsence generation strategies to test their influence on model predictability. Pseudoabsences, in lieu of observed absences, can be used to calibrate ENMs and provide contrast to environmental conditions associated with where a species is located. Analyses conducted used present-day distributional data for ten bird clades to predict past distributional data for each clade in the Eocene and Oligocene.