Before analyzing the effects of height and other positional variables on functional traits of epiphytes, something should be considered. First, the six functional traits measured in this study (including LT, SLA, LDMC, LWC_area, Chl_area and Chl_mass) may be correlated with each other. The relationships between them should be analyzed in advance in order to better interpret the relationships between traits and positional variables. Second, species using different strategies may have different functional trait syndromes (combination of six functional traits), and may be improper to analyze together.
For example, functional traits may differ between simple-leaved species and compound-leaved species, and the effect of leaf type may shade the effects of positional variables on traits. Therefore, I calculated mean values of each functional trait for those epiphyte species having three or more measured individuals (25 species in total), and then conducted a PCA on species-trait data (trait values were standardized in advance). This could show not only the relationships between traits but also the difference of trait
doi: 10.6342/NTU201900489
49
syndromes between species. The result indicates that LT and LWC_area were highly positive correlated, while they were negatively correlated with Chl_mass and LDMC (Figure 20). SLA was negatively correlated with Chl_area, but these two traits were almost independent to other four traits. Besides, it seems that chlorophyll content calculated based on area (Chl_area) or based on mass (Chl_mass) had different meaning in an ecological perspective, because they were nearly independent to each other on PCA space. Moreover, the result of PCA also indicates that trait syndrome of simple-leaved species and compound-leaved species were rather different (Figure 20). Compound-leaved species usually had leaves with higher SLA and Chl_mass but lower LT and LWC_area compared to simple-leaved species. This difference were considered in further analyses.
doi: 10.6342/NTU201900489
50
Figure 20. Result of the PCA on mean trait values of 25 epiphyte species
Functional traits were labeled with red text and their directions were shown using arrows.
Leaf thickness (LT) and leaf water content per unit area (LWC_area) were highly positively correlated, while they were negatively correlated with leaf dry matter content (LDMC) and chlorophyll content per unit mass (Chl_mass). Specific leaf area (SLA) was negatively correlated with chlorophyll content per unit area (Chl_area), but these two were almost independent to the others. The differences between trait syndromes (combination of six traits) of species are also shown in this figure. Trait syndromes of compound-leaved species (labeled with blue text) are rather different from those of simple-leaved species (labeled with grey text). Complete scientific names of all the species abbreviations are listed in Appendix 2.
doi: 10.6342/NTU201900489
51
After analyzing the relationships between traits, two different approaches were used to analyze the effects of height and other positional variables on functional traits of all epiphytes in the study site. The first one was individual-based linear model. This approach directly used the individual-based data. Functional trait values (including LT, SLA, LDMC, LWC_area, Chl_area and Chl_mass) of all individuals were response variables, while positional variables were explanatory variables (Figure 21a). Trait values were transformed using square root or natural logarithm in advance to make their distribution meet the assumptions of linear model. The positional variables used in linear model were the same as those used in CCA, including height, inclination angle, angle difference for sunlight (ADS) and angle difference for water (ADW). A forward selection procedure using partial F-test was conducted to find optimal linear model for each functional trait and determine what explanatory variables had significant influences. To include some marginal significant results, the threshold of p-value was set as 0.1 when conducting forward selection. As mentioned in previous paragraph, trait syndromes of simple-leaved species and compound-leaved species were rather different, so these two types of species were analyzed separately. For simple-leaved species, there were 221 individuals belonging to 28 species analyzed, while there were 68 individuals belonging to 7 species with compound leaves.
doi: 10.6342/NTU201900489
52
The second approach was plot-based community weighted mean (CWM) analysis.
Species abundance of plots, positional variables of plots and mean trait values of species were all needed in this approach. Mean trait values of species were calculated in previous work when analyzing relationships between traits, but in this approach only the mean trait values of simple-leaved species were used (20 species in total). Afterwards, community weighted mean (CWM) trait values was calculated for each plot by averaging the mean trait values of each species within the plot and using species abundance (transformed with natural logarithm in advance) as weights. Then the relationships between CWM trait values and positional variables of plots were examined using linear model and the significance was tested using pmax permutation test (Dray and Legendre, 2008; ter Braak et al., 2012) (Figure 21b). This testing method actually included two permutation tests.
One permutes the rows (plots) of plot-species abundance matrix (number of permutations
= 999), calculation the p-value of F-test, that was the proportion of F statistics from permutations larger than F statistics from real data, and the other permutes the columns (species) and then also calculated p-value. The larger p-value of this two permutation tests (pmax) is a reliable index to test the relationships between functional traits and positional variables and determine the significance.
Both of these two approaches were used to analyze the effects of height and other positional variables on functional traits. I conducted both of them because they both have
doi: 10.6342/NTU201900489
53
some advantages and disadvantages. Individual-based linear model is straightforward and easy to understand, while the result largely depends on the number of individuals measured for each species and for each vertical zone, and may be bias because those individuals were not collected in a completely random way. Plot-based CWM analysis suffers less from sampling bias, while intraspecific trait variations are not considered because it uses mean trait values of each species. In addition, the number of species that can be used in CWM analysis is fewer (only 20 species). As a result, I conducted both approaches and compared their results to make conclusions. All the statistical analyses in these two approaches were done using R 3.5.0 (R Core Team, 2018) and the weimea (v0.1.10; Zelený, 2018) package.
doi: 10.6342/NTU201900489
54
Figure 21. Two approaches used to analyze relationships between functional traits and positional variables
(a) First approach was individual-based linear model, which directly linked functional traits and positional variables by individual-based data. (b) Second approach was plot-based community weighted mean (CWM) analysis. Trait values of individuals were first used to calculate mean trait values of species, and then mean trait values of different species were averaged using species abundance as weights to calculate CWM trait values for each plot. Finally, CWM trait values of plots were linked to positional variables of plots using linear model and the significance was tested by pmax permutation test.
doi: 10.6342/NTU201900489
55