Range inference
Predicted range #1 allowed to forecast 40.09% of the species’ range, while 15.12% of such predicted range fell outside real range limits (Fig. 3 and Fig. 4). ) Predicted range #2 (i.e. the range inferred by extrapolating GEAM to include all grid cells suitable for any possible combination of alleles at the four loci that significantly correlated with environmental variables) was similar to the real species’ range. All grid cells in predicted range #1 were included in predicted range #2, accounting for 50.45% of its total amount. In turn, predicted range #2 captured 77.46% of the actual distribution of P. algirus , with 17.27% of inferred presences beyond real range limits. The 22.54% fraction of the real distribution range unpredicted by GEAM mostly corresponded to the northwest corner of the range, as well as to a large number of small predicted gaps within it. Nevertheless, predicted range #2 accurately reflected not only the northern and southern edges of the real distribution range, but also many of the gaps within it, both in northwest Africa (i.e. far from the sampled populations) and in many Iberian mountain ranges (Fig. 3 and Fig. 4).
Because complete randomizations provided the most reliable null hypothesis for EAAs, we used them to infer the species range following the same procedure as described above. We ran a different EAA per fully randomized dataset because each of them included a different set of randomly generated ‘outliers’ (i.e. SNPs putatively under selection). The mean percentage of inferred range was 41.39% (SD = 4.799, range = 40.09 - 81.27), roughly equivalent to predicted range #1 but smaller than predicted range #2. Of the 500 complete randomization EAAs, only one (P = 0.002, Fig. 5) was able to infer a larger proportion of the species range (80.91%) than predicted range #2 (77.46%), but this was at the cost of a large proportion of inferred presences beyond real range limits (65.02%, vs. 17.27% for predicted range #2).