Classification of forest communities
The 135 plots harboured a total of 332 vascular plant species, with species richness values ranging from two to 71. We classified the different forest communities according to their plant species composition. We log transformed and scaled the relevé data via the Hellinger method prior to data analysis (Legendre and Gallagher 2001). We calculated dissimilarities in species composition using the Bray-Curtis Index and created hierarchical cluster dendrograms using option “ward.D2” for Ward clustering within the R package Vegan (Murtagh and Legendre 2014). To find the optimal numbers of clusters, we used the clustering method of the R package NbClust (Charrad et al. 2014); we chose “NULL” for distance, “ward.D2” as method, and “kl” as index.
This resulted in four clusters of forest plant communities that clearly differ in species composition, due to environmental differences, such as soil chemistry, altitude and slope, as well as tree species richness (Table S1, Fig. S2). Therefore, we tested our hypotheses with both the entire dataset and these four forest types separately.