2.4 Statistical analyses
The abundance of the plant community was determined by the total number of individuals in each 10 × 10 m grid cell. The species diversity of each 10 × 10 m grid cell was calculated using the Shannon-Weiner index (, where is the number of individuals of the th species, is the total number of individuals of all species) (Magurran 2004). The availability and heterogeneity of each variable for the 10 × 10 m grid cell were calculated using four measurements from the 2 × 2 m quadrat of each grid cell, the mean of four measurements was used to express resource availability, and the coefficient of variation (CV) was employed as the measure of resource heterogeneity (Marchand and Houle 2006; Reynolds et al. 2007; Shirima 2016; Ulrich et al. 2018).
Prior to further statistical analysis, all data were checked for normality using the Shapiro-Wilk test and homogeneity of variance using Levene’s test. Any data that did not meet normality and homogeneity of variance were transformed using an appropriate method. All above-mentioned analyses were performed using SPSS 13.0 (SPSS Inc. IL, USA). The influence of elevation on the plant community can be neglected in this study because the elevation of different plant communities have a small change with 17m in shrubland and 177m in woodland (Table 2). Therefore, in the following data analyses, elevation was not included among the variables.
To test the effects of resource availability and heterogeneity on species diversity and abundance in different vegetation types, we first quantified the main axes of availability and heterogeneity of overall variables using principal component analysis (PCA). The first axis scores of the PCA were used as a multivariate proxy for availability and heterogeneity of overall variables. Then we tested the relationship between species diversity and abundance with the first PCA axis scores of resource availability and heterogeneity by linear regression. The PCA was performed using the “princomp” function in the R stats package (R Core Team, 2014). Linear regression and Pearson’s correlation analysis were performed using Origin 8.5 (Origin, Northampton, MA, USA).
The variation partitioning approach was used to analysis the relative influence of resource availability and heterogeneity on community species composition. The variation of community species composition was partitioned into the components explained by resource availability and heterogeneity together, as well as the components explained by each of these independently. Before variation partitioning, forward selection was applied within each of the two sets of explanatory variables to identify the significant environmental factors (Blanchet et al. 2008). The variation in species composition was decomposed in R 3.5.3 using the “varpart” function of the vegan package (Oksanen et al. 2012), and forward selection was performed in R 3.5.3 using the “forward.sel” function of the adespatial package (Dray et al. 2017).
Finally, we used a structural equation model (SEM) to explore the direct and indirect relationships between the resource availability and species diversity and abundance for different vegetation types. SEM was used because it enables the testing of direct and indirect hypothesized relationships among the variables and provides more insights into complex systems than univariate analyses (Kubota et al. 2004). Initially, all plausible interaction paths among all variables were considered in a full model. Then, some modified models were developed by removing direct and indirect pathways with low and nonsignificant path coefficients until an adequate fit was obtained (Grace et al. 2010). All direct and indirect path coefficients (λ) were standardized regression coefficients, and the pathways of SEMs were significant if P< 0.05 of the standardized regression coefficient (Désilets and Houle 2005; Van der Sande et al. 2017; Kumar et al. 2018). The relative importance of causal factors for species diversity and abundance was compared using the total effect from direct and indirect effects. The goodness of fit for the working model was determined by the maximum likelihood Chi-squared statistic (χ2), the comparative fit index (CFI) and the standardized root mean square residual (SRMR). The model was judged as a reasonable fit ifP >0.05, which indicates that fitting covariance matrices are not significantly different from observed covariance matrices (Grace et al. 2010). CFI> 0.95 suggests a very good fit, which is little affected by sample size compared to the Chi-square test (Rosseel et al. 2012). SRMR ≤ 0.05 indicates a very close fit between a model and the observed data (Browne and Cudeck 1993). All models were implemented in R 3.5.3 using the “sem” function of the lavaan package (Rosseel et al. 2019).
Results
3.1 Effects of resource availability and heterogeneity onspecies diversity and abundance
The first PCA axis (PC1) accounted for 40% of the resource availability variation for shrubland, 55% of that for woodland and 30% of the resource heterogeneity variation for both shrubland and woodland; moreover, the PC1 eigenvectors of both resource availability and heterogeneity were greater than 2.00; therefore PC1 explained the majority of the variability in the resource availability and heterogeneity in this study (Fig. A1, Table A1, Supplementary material).
With increasing resource availability, species diversity and abundance significantly increased in both shrubland and woodland except species diversity in the shrub layer of woodland (Fig. 2a, b, c d). Species diversity and abundance were both significantly negatively related to resource heterogeneity in shrubland (Fig. 3a, b). However, community abundance significantly increased with increasing resource heterogeneity in woodland (Fig. 3c). There was no significant relationship between species diversity and resource heterogeneity in woodland (Fig. 3d).