2.3.1 Network structure
We used foraging frequency to construct the interaction network, and divided the whole year into four seasons, spring (March, April, and May), summer (June, July, and August), autumn (September, October, and November), and winter (December, January, and February), to analyze the differences in plant-frugivore network characteristics between different seasons. We characterized the structure of weighted interaction network using the following nine statistics through the function “network-level” in the “bipartite” package (R Core Team, 2022): (1) number of bird species (b); (2) number of plant species (p); (3) network size (b×p); (4) number of links (n); (5) connectance (C), the proportion of links that are realized among the pool of all possible links (n/b×p) (Cruz et al., 2013); (6) nestedness (nestedness), which quantifies the degree to which species with few interactions are connected to highly connected species and has been proposed to be associated with network stability (Ramos-Robles et al., 2016); (7) specialization (H2´ ), which quantifies the overall specialization within a network, that is, whether species in a network tend to separate or share their interaction partners (Blüthgen et al., 2006); (8) interaction diversity (H2 ), a Shannon index based measure of diversity estimated from interaction frequencies, which reflect whether the links are strong (high interaction frequencies) or weak (low interaction frequencies) (Zhang et al., 2022); (9) interaction evenness (E2 ), which depicts heterogeneity in the distribution of interactions across species in the network, with high values indicating more even distribution (Sakai et al., 2016).
We used the function “null model” to randomize plant-frugivore interactions, and compared the differences of structure between the observed network and null model (1000 iterations). Randomizations can determine which nodes (species) interact with one another and how strong the interactions are under a simple null hypothesis and determine whether interaction frequencies between consumers and resources are a consequence of the relative abundances of the potential resources (Vaughan et al., 2018). The null model can reshuffle interactions while maintaining the observed matrix dimensions and connectance to reduce the influence of sampling effects on the network interpretation (Pigot et al., 2016).