Within the study we used averaged o

Within the study we used averaged output from the three GCMs while the supplementary materials show results of model predictions from single GCMs and RCPs. When the species distribution models that we developed were applied to the projected climate change scenarios and GCMs and confronted with the observed distribution for each species, there were four possible scenarios for each 2.5 raster cell: (1) positive, and (2) negative, overlap of current and future projected ranges, (3) potential range expansion and (4) potential range contraction. Positive overlap of current and future projected ranges (true positive) indicates that under changing climate the species in the raster cell under consideration still will be located in its climatic niche in 2061-2080. Negative overlap of current and future projected ranges indicates lack of a given species currently and in 2061-2080. Potential range expansion, i.e. when a species does not occur currently and it has been predicted that it will occur in 2061- 2080, indicates potentially suitable future habitat. However, due to dispersal limitation its occurrence is uncertain and cannot be considered without detailed data about dispersal capacity (Meier et al., 2012). Potential range contraction indicates that under the considered climate change scenario, the species currently in the particular raster cell will be outside its climatic optimum in 2061-2080. As trees are long-lived organisms and climate change may increase the occurrence of disturbances and biotic damaging agents, e.g. fungi and insects (Seidl et al., 2014), we assumed this part of the observed distribution was an area where occurrence of the species is threatened. Potential range contraction in current habitat suitability may include both habitat unsuitability and spatial bias of the model. Therefore, we assumed that the proportion of potential range contraction within the observed range (i.e. false negatives obtained by comparing model predicted range under current climatic conditions with the current occurrence data) was a measure of bias in prediction of threatened area, which refers to the type II threat risk measure of Ohlemüller et al. (2006). In this meaning, we assumed prediction bias for each species as the proportion of the number of false negatives compared to the number of all observed presences among raster cells. To address the uncertainty of threat levels we used standard error (SE) to supplement mean threat levels, calculated using projections for each species from the three GCMs included in the study. After threat analysis we divided the species studied into three groups: winners species with less than 50% of current distribution threatened under the pessimistic scenario (RCP8.5), losers species with more than 50%, and alien species geographically alien to Europe.Principal components analysis (PCA) based on presence/absence occurrences from averaged outputs of each species for all raster cells (n=1,235,520) was used to describe differences in distributions among the 12 species for current conditions and the three climate change scenarios (n = 12 x 4 = 48). PCA was chosen, as it gave biologically meaningfuloutput and the algorithm was able to do computations on an extensive matrix with empty rows, in contrast to correspondence analysis and detrended correspondence analysis. Further inspection of PCA results showed that there were no problems with artifacts and horseshoe effects. We also confirmed applicability of both PCA axes using a screeplot we took into account only those components with eigenvalues higher than predicted by a random broken stick model (Jackson, 1993). We used species-scenario scores, i.e. values of principalcomponents for each species in each scenario, for analyses of changes in potential distribution. All analyses were conducted using R software (R Core Team, 2015).