Seasonal dynamics of flower visitors
Because hymenopteran and dipteran insects occupied > 90% of all flower visitors (see Results), factors affecting the visitation frequencies of bees and flies were separately analyzed by generalized linear models (GLMs). Visitation frequencies of bumble bees, syrphid flies, and non-syrphid flies in each census were used as responding variables, i.e., three GLMs were constructed. Other insect groups were excluded from the analysis due to low visitation frequencies (see Results). Explanatory variables were observation date (mean day of year in each census term), ambient temperature, relative moisture, and year. Interactions between observation date (both linear and quadratic terms) and year (2017 and 2018) were included in the model. For the GLMs, a zero-inflate Poisson distribution model (Brooks et al., 2002) was conducted because there were many zero values in some insect groups. Of 208 census data, three data for bumble bees and two data for non-syrphid flies were excluded from the analyses because of unusually excess counts of insects probably due to miscounts of insects in the field.
The network structure between flower-visiting insects and foraging plant species across seasons and years was visualized using packagebipartite in R (Dormann et al., 2022). For obtained network structure, binary connectance (frequency-based connectance), niche overlap among insect visitors (based on Horn’s index ranging from 0 with no common use to 1 with perfect niche overlap), mean number of shared plant species, and Shannon-Weaver’s H’ diversity index were calculated. In the network analysis, we constructed network structure of order-level comparison (Hymenoptera, Diptera, Coleoptera, Lepidoptera) and major taxonomic group comparison (bumble bees, syrphid flies, and non-syrphid flies). Furthermore, similarity of foraging flowers between major insect groups and Shannon-Weaver’s H’ diversity index of each insect group were calculated using the Vegan package.
Finally, the relationship between visitor frequency and the number of plant species at flowering was analysed for bee-visited plants and fly-visited plants, respectively. GLM postulating a Poisson error distribution was conducted in which the number of flowering species observed in each term (for bee-visited or fly-visited plants) was set as a responding variable and the average number of insects (for hymenopteran or dipteran insects) observed in each term as an explanatory variable.