Range inference
To infer the distribution range of the species, we followed a two-step procedure. Firstly, we included all the geographical cells that presented the same environmental scores as the sampled populations (predicted range #1). This first approach provides a baseline prediction with no genetic information that can be used to quantify the improvement in predictive ability supplied by GEAM. Secondly, we extrapolated such GEAM to all possible combinations of alleles for the outlier loci significantly associated with environment (i.e., all possible genotypes under selection), so that we could predict all environmental values that were suitable for at least one genotype. By fulfilling all grid cells with those environmental values, we could extrapolate our prediction to the overall distribution range of the species. Finally, we removed from the inferred range a few disconnected patches (in central France, coastal Italy, and the Mediterranean islands) that were too far from the main distribution range of P. algirus (>8 km from the nearest inferred distribution limit, which was the largest distance between adjacent but disconnected patches within the inferred range), whose low dispersal rate (Santos et al. 2009) is supported by the fact that genetic differentiation can be detected even among forest fragments separated by 350 m of unsuitable arable land (Pérez-Tris et al. 2019). This produced our second (and final) inferred distribution range (predicted range #2). The extent of overlap between real and predicted distribution ranges was estimated using QGIS v2.18.16 (QGIS Development Team 2018).
The delineation of a real distribution range that could be compared with our inferences #1 and #2 was based on two previously published atlases: Bons and Geniez (1996) for north-west Africa, and Pleguezuelos (1997) for the Iberian Peninsula. To make such delineation as reproducible as possible, we digitalized these previous maps to get a grid of 10x10 km geographical cells, which was the original scale in the atlases. We subsequently added all the areas devoid of lizards, but embedded within the previous range, that were tagged with an artificial land use in the Corine database (European Environment Agency under Copernicus program, 2018). This was done to correct for the fact that absences from such areas should be attributed to land use (unsuitable for P. algirus ) and not to the environmental variables included in our model. On the other hand, our inferred ranges were based on environmental variation measured across a grid of 1x1 km geographical cells, because a scale of 10x10 km cells would have been too rough to properly model the distribution range of the species. Thus, and for the sole purpose of estimating the accuracy of our inferences using a criterion as similar as possible to the one used in the atlases, we lowered their resolution by including all 10x10 km grid cells with at least 25% of its surface corresponding to inferred presences.