Species and study sites
T. chinensis is a relic species that is endemic to China. This species is designated as “endangered” in the International Union for Conservation of Nature (IUCN) Red List (2013 version), and is ranked as a first priority protected species by the Chinese government (Tomas et al. 2013). The regeneration of the wild population of T. chinensis is not only limited by low pollination rates, seed-predator pressure, and low seedling survival rates, but also limited by different bird species (Li et al. 2015b).
For our study, we selected three field sites in eastern China (Shuangkeng village in Anhui, Chongtou village in Fujian, and Tongkeng village in Zhejiang) that had dense populations of T. chinensis. The Shuangkeng village [30°00′N, 117°18′E; 540 m above sea level (a.s.l.)] is located in the southern part of Anhui Province, and contains 35 individual trees, which are distributed in a human-modified bamboo (Phyllostachys heterocycla) patch (Li et al. 2015a). The Chongtou village (25°15′–25°35′N, 116°45′–116°57′E; 960 m a.s.l.) is situated in the Meihua Mountain National Nature Reserve in Fujian Province, and contains the largest wild population of T. chinensis (approximately 490 adults) in eastern China, including 200 trees that are over 500 years old (Li et al. 2015b). The Tongkeng village (30°00′N, 119°22′E; 553–638 m a.s.l.) contains the largest natural population of T. chinensis in Zhejiang Province (approximately 109 adults), including 34 trees that are more than 500 years old (Li et al. 2015a).
Seed removal by bulbuls and crows across different regions
For studying seed removal by bulbuls and crows across different regions, 10 aggregated mother trees were observed during two fruiting seasons (from late October to early December; Shuangkeng: 2010 and 2011; Chongtou, Tongkeng: 2011 and 2012) at each site. Observations were made with binoculars from a hideout located at least 25 m away from the trees. Each mother tree was observed for an 8-h period once a day during the seasons, with observations being terminated when no more fruits remained on the mother trees.
For each bird that visited the mother tree, we recorded the species and the number of fruits swallowed during visits. If a group of conspecific birds visited the tree, we only focused on the most visible individual (Altmann 1974; Breitbach et al. 2010). After field study, we compared the visiting frequency and foraging amount per visit by each disperser species as a quantity pattern across different regions. Moreover, we used a machine-learning algorithm, random forest model, to plot how bird visiting frequency, foraging amount per visit, bird species, sites, and years influence the numbers of seeds that were removed (R package RandomForest) (Breiman 2001).
Perching by dominated bulbuls and crows across different regions and its contribution to the seedling recruitment pattern
We first used 10 m × 10 m habitat cells to digitize the three study sites. Overall, we set 100 sampling cells in Shuangkeng, Chongtou, and Tongkeng, respectively. We then investigated the 1-year-old seedlings (H ≤ 10 cm) in each cell of each site to determine the suitable recruitment habitat for T. chinensis.
For modeling the contributions of birds to seedling recruitment, we monitored the perching frequencies of two dominant species of bulbuls and crows during the fruiting period, from late October to early December (Breitbach et al. 2010; Li et al. 2015a, 2016). Bird perching was observed from high vantage points on the hills using binoculars. Birds were selected randomly and tracked in sessions that ended once visual contact was lost or if the focal bird could no longer be distinguished from the other conspecifics. During the sessions, the habitat used by birds and their position were recorded every 30 s (Breitbach et al. 2010; Spiegel and Nathan 2012).
We used the function to perform univariate Kriging models to interpolate seedling number and bird perching frequency in the sampled habitat plot. The method used for spatial interpolation is the krige function in the R package gstat. We tested the number of seedlings in the habitat cells by a generalized mixed-effect model, in which the perching frequencies of the bird species and sites were the covariates, while the survey year and their interaction term were the random effects. Because the numbers of seedlings were count data, the function glmer in the R package lme4 was used to run the generalized mixed-effect model with a Poisson link function.