Fig. 2

Workflow followed to construct SDMs and test for significance of climate signal. Null data were generated with similar spatial patterns to those present in observed data (dashed arrows; see Additional file 2). Observed and null presence data were split 1000 times into training and testing data (60:40 %; double line arrows) and used to construct observed and null models. Multiple post-hoc pairwise comparisons of median AUC values from observed and null models were undertaken using critical values from the t distribution to test for a significant climate signal. The workflow was followed once for Maxent and once for Mahalanobis Distance SDMs for each species, producing four independent outcomes per species; tests of observed Maxent model against null climate (1) and null presence (2) models, and tests of observed MD model against null climate (1) and null presence (2) models. All four outcomes were considered in evaluating whether there is a significant effect of climate on a species’ distribution