FST was stronger and more significant when the adaptive SNPs were used instead of all SNPs (mean FST all loci = 0.09; mean FST outlier = 0.59; p < 0.05), indicating that selection may drive the spatial genetic differentiation between populations. Compared to all loci, the adaptive loci exhibited stronger IBD and IBE patterns (Fig. 3a). Using reciprocal causal modeling (RCM) and a maximum likelihood population mixed-effects model (MLPE), IBE was consistently identified as superior to other competitive models using adaptive SNPs when controlling population structure, indicating that genetic differentiation of the adaptive SNPs was mainly influenced by environmental variation (Fig. 3b-c; Table 1). Furthermore, the divergence of the genetic structure of HC from those of the other clusters was greater when divergence was assessed by adaptive SNPs than by neutral SNPs (Fig. S4a-b). By contrast, IBR based on topographical resistance was selected as the best model by MLPE and RCM using all sites, suggesting that the overall genetic differentiation of the species was mainly influenced by topographical dispersal barriers (Fig. 4a-b; Table 1). The conductance layers generated by the CIRCUITSCAPE algorithm and the estimated effective migration surface (EEMS) analysis also supported topographical resistance across regions, with greater barriers to dispersal at higher elevations (Fig. 4c-d).