Characterisation of sampling sites using spatial, environmental and landscape variables
A number of spatial, environmental and landscape variables were calculated to characterize the sampling sites and quantify the distances among localities, taking into account the high topographic complexity and environmental/landscape heterogeneity of the study area. We calculated pairwise weighted topographic distances (SPATWD) based on a digital elevation model (DEM) at 90 m resolution using the topoDistance R package (Wang, 2020) as described in the Supplemental Information. We also generated a set of high-resolution environmental variables (at 90 m resolution) for Cyprus, by spatial interpolation of temperature and precipitation layers at lower resolution using the aforementioned DEM (see Supplemental Information). Specifically, we interpolated six WorldClim (annual mean temperature, maximum temperature of warmest month, minimum temperature of coldest month, annual precipitation, precipitation of wettest quarter and precipitation of driest quarter; Fick & Hijmans, 2017) and three ENVIREM (climatic moisture index, Thornthwaite aridity index and topographic wetness index; Title & Bemmels, 2018) variables. These variables are known to affect the water-energy dynamics and to explain patterns of diversity in several organismal groups (Hawkins et al., 2003). We extracted values of each interpolated variable along with the elevation for all sampling sites and for 500 randomly distributed points throughout Cyprus, to avoid potential biases resulting from only considering conditions at focal sites. We then applied a principal component analysis (PCA; Figure S1) to reduce the dimensionality of the dataset and eliminate covariance among variables and we retained the two first principal components (PCs). PC1 (accounting for 81.3% of variation) was positively correlated with altitude and precipitation variables and negatively with temperature variables and PC2 (8.0%) was positively correlated with the topographic wetness index (Table S3). The PC1 and PC2 scores for each sampling site were considered as topoclimatic predictors (ENVPC1 and ENVPC2) for downstream analyses. We also calculated a Euclidean distance matrix among sampling points based on the obtained scores of the two retained PCs, which was used as a topoclimatic predictor (ENVPC1-2) for matrix regression analyses.
Finally, we applied a circuit theory approach (McRae, 2006) to quantify habitat connectivity among the Qa sampling sites. We focused specifically on the Qa habitat because it is broadly distributed but highly fragmented across Troodos, in contrast to the CbPn and Jn forests which are very narrowly distributed and the Pb forest which is very extensive with largely continuous distribution (Figure 1). Based on the assumption that dispersal among isolated forest fragments can be impeded by the presumed lower habitat suitability of the surrounding landscape (Brodie & Newmark, 2019), we built an isolation-by-resistance (IBR) scenario of connectivity (FRAIBR) defined by the distribution of Qa forest patches according to the existing cartography (see Supplemental Information). Two alternative IBR scenarios were constructed for comparison: the TRIIBR representing the topographic complexity of the study area as estimated by the terrain roughness index (Title & Bemmels, 2018) and the NULLIBR representing a completely “flat landscape” with a fixed resistance (=1) value assigned to all cells. Resistance distances among all Qa sampling points (n = 11) were calculated under each alternative IBR scenario (FRAIBR, TRIIBR and NULLIBR) in circuitscape v.4.0.5 (McRae & Beier, 2007).