Statistical analyses
Morphological attributes were all log-transformed and centralized before calculating trait-based community structure. Principal component analysis (PCA) was applied for size and shape related morphological attributes, and the first two components were used to calculate morphological dispersion. Standardized phylogenetic and trait-based dispersion (SES.MPD and SES.PW) were determined with two-tailed t-test: if community structure is significantly higher than those in null communities (SES >1.97), community will be defined phylogenetically or functionally dispersed; if community structure is significantly lower than expected (SES < -1.97), community is characterized as phylogenetically or functionally clustered. If community structure is non-significant different from that in null communities (-1.975< SES < 1.975), community is characterized as random (Kembel et al. 2010; Webb et al.2008; Webb et al. 2002; Webb 2000). Underlying process (environmental filtering, interspecific exclusion and stochastic process) and the functional role of traits were estimated according to our revised empirical framework. The elevational patterns of phylogenetic and trait dispersion were determined using polynomial regressions. Best predictive model were selected according to Akaike information criterion (AIC). Besides, we also quantified the elevational pattern of species richness, which is reasonable in estimating environmental fitness and habitat heterogeneity (Brown 2001).
    As ecological process and the functional role displayed by trait are habitat-specified, hence, the approach of linear regression model or correlation, to some extent, is more convenient but probably underestimate the dependence between phylogenetic and trait-based community structure. In contrast, pairwise comparison of this kind approach would be more accurate in predicting assembly process and the functional role of a certain trait. In order to obtain comparable results, we applied these two approaches (i.e., correlation analysis and pairwise comparison) in inferring the relationship between phylogenetic and morphological structure. In comparison analysis, if both phylogenetic and trait dispersion showed same status (clustered, random or dispersed), we called it phylogenetic-trait consistence or congruence; or else, we named it phylogenetic-trait inconsistence or incongruence. Pairwise comparison and correlation analysis have been repeatedly conducted between phylogenetic dispersion and morphological dispersion of size and shape related traits.
    With the aim to estimate the environmental dependence for phylogenetic and trait dispersion, we conducted forward selection procedure to choose the best climatic predictor(s). Structure equation models (SEM) were applied to estimate the direct and indirect effects of climatic factors on phylogenetic and trait dispersion. Environmental variables were all log-transformed and centralized in best predictor selection and SEMs. All of these calculations were accomplished under R environment (ver. 3.5.1) (Team 2013). phylogenetic signal detection was carried out with the package ‘phytools’ (Revell 2012); PCA analysis, correlation analysis and polynomial regressions were all accomplished with default packages in R (Team 2013); Following the approach in Swenson (2014), phylogenetic and trait-based structure was calculated with package ‘picante’ (Kembel 2009) and package ‘vegan’ (Oksanen et al. 2007; Dixon 2003); Forward selection in climatic factors was finished with ‘leaps’ (Miller 2002) and ‘vegan’ (Oksanen et al. 2007; Dixon 2003) packages; SEMs were performed with package ‘lavaan’ (Rosseel 2012).