Demographic response to transplantation explained by traits
To investigate if traits explained demographic responses to transplantation, we quantified species’ position in multivariate trait space by performing a Principal Component Analysis (“PCA”). We included all 27 traits measured at 2000 m (“+0°C”) and six plasticity indices for those traits that showed significant differences between sites on average (WUEinst, GSnight, Enight, δ18O enrich, δ13C and depth, Figures S4-S7). To test if species’ locations in the trait space related to their demographic responses to transplantation to lower elevation, we built linear mixed effects models (LMMs) with either 2020 cover (cm2) or frequency (number of quadrants where species was present) as the response variable. We used both species cover and frequency, because we expected these to correspond to different demographic processes: while changes in frequency depend on species’ survival and recruitment, changes in cover are also influenced by vegetative growth (and, in contrast to frequency, potentially by errors in cover estimations). The response was modeled as a function of the initial cover or frequency of the species per site in 2017, the mean summer temperature at each site, the first four PC trait axes and the two-way interactions between the trait axes and temperature. The significance of fixed effects was assessed by dropping one explanatory variable at a time, starting from a full model. The change in the model fit was assessed using likelihood ratio tests (Crawley 2007). For responses showing a significant trait axis × treatment interaction, we evaluated the amount of variation explained by calculating pseudo-R2 (Nakagawa & Schielzeth 2013). To estimate the usefulness of traits as predictors, we compared the amount of variation explained by traits to that explained by species identity. To do so, we compared variation explained by models with or without species or traits as fixed effects (see Table S3 for details).
To further investigate which traits were associated with demographic response to climate, we tested if warming resulted in communities of species possessing particular trait values. We calculated community-weighted means for each trait per each replicate turf across the four sites, based on the 2020 percent cover of all species and their 2000 m (+0°C) trait value. The only exceptions were δ18O and δ13C, for which we used both the site-specific trait values, and the values as measured at 2000 m (+0°C). For these two traits, we evaluated the degree to which changes in community-weighted trait means were due to within-species trait change versus species turnover. Differences in community-weighted mean trait values between sites were tested with linear models.