Data analyses
The meteorological data was obtained fromhttp://cdc.cma.gov.cn/home.do, where the mean value of 30 years (1979-2009) was used for analysis. The piecewise SEM (Lefcheck 2016) were employed to evaluate the direct and indirect effects of predictor variables (including geographic and climatic factors and leaf functional traits) on trichome density. A backward stepwise selection procedure was used to simplifying the model until the AIC value began to increase and P value was above 0.05. The sampled sites were considered as a random effect with linear mixed models (LMMs) in our analyses (Lefcheck 2016, Ali et al 2020). For all endogenous variables, the conditional R2c (all factors, including the random effect) and marginal R2m values were estimated their variations (Lefcheck 2016). In the piecewise SEM, the effect of each predictor on the endogenous variables was accounted for through their standardized path coefficients. The piecewise SEM and LMM models were separately implemented with the piecewise SEM package (Lefcheck 2016) and the nlme R package. To avoid the model complexity, all geographic variables (latitude, longitude and altitude) and climatic variables (MAT, MAP, aridity index (AI), potential evapotranspiration (PET), and MMSR) were clustered by principal component analysis (PCA). All variables, used in piecewise SEM, were log-transformed and then standardized. Moreover, one-way ANOVA was employed to test the differences in trichome density between the field samples and common garden population samples. To determine the direction of trait variation in response to environmental changes, linear regressions were performed to correlate environmental factors with trichome density. The above analyses were carried out with R version 3.6.1 (R Development Core Team, available from www.r-project.org/, accessed 2019) and SigmaPlot 10.0 (Systat Software, Inc., R.ichmod, CA, USA).