Species Distribution Modeling
As a preliminary estimate of habitat suitability given present-day
climatic conditions, we estimated species distribution models using the
‘maxent’ algorithm in the ‘dismo’ package (Hijmans et al. 2017). We
included the 31 GBS localities from our dataset and the 59 AHE
localities from Pyron et al. (2022c). As an initial set of predictors,
we included 37 variables: WorldClim (Fick and Hijmans 2017), ENVIREM
(Title and Bemmels 2018), and Level IV ecoregions (Omernik and Griffith
2014). These were sampled at 30s resolution and re-projected to North
America Albers Equal Area Conic (ESRI:102008). We used the ‘corSelect’
function in the ‘fuzzySim’ package (Barbosa 2015) to remove
multicollinear predictors based on variance inflation factor (VIF). The
retained set 6 BIOCLIM and 3 ENVIREM variables and the Level IV
Ecoregions; the full list is given in the SI. We optimized a maxent
model using the 90 localities and 1,000 background pseudo-absence points
from a 250km radius. We then projected this model as a three-level
binary prediction under the equate entropy of thresholded and original
distributions, maximum training sensitivity plus specificity, and equal
training sensitivity and specificity to evaluate varying degrees of
potential occupancy across the range (Liu et al. 2015).