Quantification of environmental variation
To quantify environmental variation all over the potential range of the species, we selected an area that included its actual distribution range (based on Bons and Geniez 1996 [for north-west Africa], Pleguezuelos 1997 [for the Iberian Peninsula] and IUCN web [for the rest of north Africa]) plus a 450-850 km wide perimeter belt around it (width variation depended on the distance from the edges of a non-quadrilateral range to the rectangle that was used to set the subsequent inference models). Within this area, we used data from the Bioclim 2.0 dataset (cell resolution = 1x1 km; Booth et al. 2014) to compute the score of each cell on a principal component analysis that combined all Bioclim environmental variables using R core (R Core Team, 2013). This PCA yielded a single principal axis that opposed hot areas with low precipitation to temperate ones with high precipitation (Fig. 1B).