Material and Methods
We selected two trees ⁓50 m apart in each of two forest patches on the campus of the University of Massachusetts-Amherst in Amherst, Mass., USA (Fig. A1); each pair of trees consisted of a northern red oak (Quercus rubra L.) and a red maple (Acer rubrum L.). Both sites were in USDA Hardiness Zone 5a and were characterized by an herbaceous stratum of ferns (e.g., Dennstaedtia punctilobula(Michx.) T.Moore, Polystichum acrostichoides (Michx.) Schott), wild sarsaparilla (Aralia nudicaulis L.), white wood aster (Eurybia divaricata (L.) G.L.Nesom), star flower (Lysimachia borealis (Raf.) U.Manns & Anderb), Canada mayflower (Maianthemum canadense Desf.), partridge berry (Mitchella repens L.), and Solomon’s seal (Polygonatum spp .). The understory of these sites consisted of brambles (Rubus spp. ), poison ivy (Toxicodendron radicans (L.) Kuntze), maple-leaf viburnum (Viburnum acerifolium L.), witch hazel (Hamamelis virginiana L.), glossy buckthorn (Rhamnus cathartica L.), and seedlings of the dominant canopy trees (e.g., A. rubrum, A. saccharum Marshall, Betula lenta L., B. papyriferaMarshall, Q. rubra and Q. alba L.). We chose A. rubrum and Q. rubra because they are dominant species in forests of the area and represent different flowering systems and blooming times that span the duration of most available floral resources in these canopies. We chose these forest patches due to their accessibility and general representation of dominant species in forests of the area.
The bee community was sampled using blue vane traps in the understory, midstory, canopy, and above-canopy strata of the forests at each focal tree. Three traps were individually attached to a rope hung over a high branch in the canopy as in Cunningham-Minnick & Crist (2020). Traps were placed 1, 10, 20, and ⁓30 meters above the ground (Table A1) to represent the following strata: understory, midstory, canopy and above canopy (Fig. 1). The trap above the canopy was set one meter above the tallest leaf bearing branch of each tree using a telescoping hanger attached to a vertical limb in the crown of the canopy as described in Cunningham-Minnick et al. (in press ). Traps were deployed on April 2, 2022, and checked every 1–3 weeks until August 21, 2022, for a total of 12 checks. Bees were sorted, pinned, and identified to species by JM using published keys (e.g., Gibbs, 2011; Gibbs et al., 2013; LaBerge, 1987, 1989; Mitchell, 1960, 1962) and the online source Discoverlife.org (Ascher & Pickering, 2020); voucher specimens are in the possession of MC-M. To distinguish differences in microclimate from other conditions among strata, Onset HOBO® Pendant data loggers (Part AU-002-64) were placed directly above each trap to record light intensity and temperature every 10 minutes June 7–21, 2022, to provide data of daily microclimate conditions and hourly from June 22–August 21, 2022, to represent seasonal change.
Data Analysis : To compare bee abundance and species richness across vertical strata throughout the sampling season, we built generalized linear mixed effects models with negative binomial errors and created 95% confidence intervals of pairwise comparisons for each response across strata. Models were made using the glmmTMB function in the glmmTMB package (Brooks et al., 2017) with fixed effects of stratum (understory, midstory, canopy, above canopy), sample (1–12) as a continuous variable, and their interaction. We allowed the model intercept to vary by tree to account for tree-specific differences, and accounted for differences in sampling effort by including an offset term of the log of the trap deployment duration (days). Significance of interaction terms was evaluated by likelihood ratio tests; simulated model residuals through the DHARMa package were used to evaluate overall model fit (Hartig, 2020). Post-hoc comparisons were made using the confint and glht functions in the multcomp package (Hothorn et al., 2008). Differences in bee species composition among strata were visualized with non-metric multidimensional scaling ordinations performed on a species occurrence matrix of Sorensen distances using the metaMDS function in the vegan package (Oksanen et al., 2019); statistics and p-values were derived using the pairwiseAdonis function with a Bonferroni adjustment for multiple comparisons (Arbizu, 2017).