Model evaluation
For BRT models, evaluation of model performance included measures of
residual deviance, k-fold cross validation and the area under the
receiver operator characteristic curve (AUC) (De’ath, 2007; Elith &
Leathwick, 2009). GLM model performance was evaluated with AUC and
adjusted D2 (a measure of the difference between null
deviance and model deviance adjusted for degrees of freedom) and Akaike
information criterion (Guisan et al., 2002; Franklin, 2010). Maxent
models were evaluated by comparative training and test AUC, and omission
on test and training samples against random prediction
(specificity-sensitivity curves) (Hijmans, 2012; Elith & Leathwick,
2017).
For predictions of current habitat suitability, maps showing the
predicted probabilities of presence or absence are presented for the
best model with soil type and without soil type. Plots of the partial
response of C. halei to each of the environmental variables and
the proportional contribution of each variable to model performance are
also presented.