*2.4. Data Analysis*

We used *C. chinensis* with DBH ≥10 cm as the center of distance annuli and sampled the number of individuals within each phylogenetic group. We used 20 m, the scale always used as the minimum observation size of forest plots, to test the phylogenetic relationships between *C. chinensis* and other individuals within the distance annulus. To avoid a sampling scale bias, we also reported the results for 28.28 m, 34.64 m, 40 m, 44.72 m, and 48.99 m with the same sampled area in the Supplementary Materials. There were 2061 *C. chinensis* trees in the succession forest, so there were 2061 sampling annulus. There are 161 *C. chinensis* trees in the mature forest, also 161 annulus are used to sample the mature forest. Individuals of each phylogenetic group within each annulus were counted and the mean number of individuals of each phylogenetic group calculated. Finally, we counted the mean individual percent (IP) of each phylogenetic group's total abundance (Figure 3).

Specifically, we used simulations to estimate a null model of "individual percent" in each phylogenetic group by randomizing the label of the species name in the phylogenetic group to control for the effect of phylogenetic relationships. 999 random samples were simulated keeping the number of species of each phylogenetic group the same as reported in Table 2. However, the number of individuals of each phylogenetic group was randomly changed. Occasionally, the most closely-related (Group 1) and most distantly-related species (Group 6) were sampled with nearly the same number of individuals as Group 3. Individual percent was calculated for each simulation, thus resulting in 999 individual percentages. If the observed percentage fell within the 2.5th and 97.5th quartiles, then we failed to reject the null hypothesis of a random distribution of phylogenetic relationships, otherwise we concluded that there was a significant phylogenetic relationship between the dominant species *C*. *chinensis* and neighbor tree species.

In order to test the robustness of our results we perform Mantel test. We measured the correlation between two matrices in each annulus with 999 permutations: the matrix of phylogenetic distance between *C. chinensis* and other trees and the matrix of geographic distance between *C. chinensis* and other trees. The oberservation value of Mantel test is the coefficient of correlation *r*. All analyses were conducted in R2.6.2 platform, which was available from R Foundation for Statistical Computing, Vienna, Austria (ISBN 3-900051-07-0, http://www.R-project.org). And the Mantel test was completed using R package "ADE4" [29].

**Figure 3.** Six species groups quantitative distribution around *Castanopsis chinensis* at six scales in succession forest (**a**) and mature forest (**b**). Horizontal axis is scales number representing outer diameter of annulus around *C. chinensis*: 20 m, 28.28 m, 34.64 m, 40 m, 44.72 m, and 48.99 m. The vertical axis represents percent of neighbors within that group.

In order to test phylogenetic signal of functional traits, the current functional trait data were used to verify whether there is pedigree conservatism in Dinghushan sample plots (Table S1). The descriptive statistic*K* presented in Blomberg et al. was used to measure the phylogenetic signal [10]. The significance of the observed *K* value was determined using a permutation test. To evaluate the significance of the phylogenetic signal *K*, we generated a null expectation of *K* under no phylogenetic signal by randomizing the names of taxa 1000 times in the phylogeny [11], and a probability (*p*-value) that the observed *K* is higher than randomization is calculated. Thus, this probability indicates statistical significance of phylogenetic signal of a functional trait across a phylogeny.
