Statistical analysis
To quantify the main axes of wood trait variation across species, a principal component analysis (PCA) was performed. The first axis (PC1), accounting for 55.9% of variance in litter quality, was strongly related to the contents of wood nutrients (nitrogen, phosphorus), cellulose and lignin, and wood density. We used the PC1 scores for the respective tree species to represent their position along the WES positions in the subsequent analyses. The second axis (PC2) was related to wood water content, accounting for 17.1% of variance (Fig. S1).
We also used PCA to quantify the community-level tree functional trait variability for forest plots in TT and PT. The PC1 of PT and TT accounted for 50.3% and 40.9% of trait variance, respectively and were strongly related to leaf resource economic traits (specific leaf area, nitrogen, phosphorus, mean leaf area) and wood density (Fig. S4). We used the community abundance-weighted mean (CWM) of WES, specific leaf area and wood density to compare differences of community functional identity between PT and TT sites by using Student’s t-tests. To derive CWM of WES we multiplied the PC1 scores of each species with its relative abundance for a given community.
We used ANCOVA to determine the dependence of wood mass loss on specific independent variables. Using the wood (cumulative or period) mass loss % as the dependent variable, harvest time and termite presence/absence as the independent variables, and the WES value as covariate, separate ANCOVAs were performed for PT and TT respectively. To evaluate the relationships between (cumulative) mass loss % (for termite access and exclusion treatments) and position along the WES separately for the different incubation periods, linear regression and non-linear regressions were used to find the best-fit relationship between mass loss % and the WES. To test the relationship between (cumulative or period) mass loss % in the termite access treatment and termite abundance at each harvest time, linear and non-linear regressions were used to find the best-fit relationship between mass loss % and termites abundance. We used Student’s t-test to test the differences in cumulative mass loss % of termite access and exclusion treatment, termite abundance and the contribution of termites to wood mass loss between the sites.
To evaluate the relationship between termite abundance and WES in the two sites at each harvest time, linear and non-linear regressions were used to find the best-fit relationship. To evaluate the effect size of the termites along the WES of the two sites at each harvest time, linear and non-linear regressions were used to find the best-fit relationship. For mass loss data, we used Levene’s test to examine the homogeneity of variance and Shapiro-Wilk test for normality. Wood mass loss was log-transformed as to best meet the assumptions of normality and variance homogeneity. All statistical analyses were performed in R language version 3.5.1.