Native pollinators and their pollination services are critical in supporting ecosystems worldwide (Klein et al., 2007; Winfree et al., 2007, 2008; Ollerton et al., 2011; Kennedy et al., 2013; Rodger et al., 2023). Evidence that many pollinator species are experiencing population declines (Biesmeijer et al., 2006; Cameron et al., 2011; Bartomeus et al., 2013; Burkle et al., 2013; Koh et al., 2016) has raised alarm among conservationists and the public. In response, there has been a rapid increase in research aimed at understanding basic habitat associations at community and species levels to inform pollinator conservation efforts (Williams et al., 2010; Tonietto and Larkin, 2017; Neumüller et al., 2020; Milam et al., 2022). To date, research within the temperate zone suggests that native bees tend to be more abundant and diverse within open and early-successional communities (Taki et al., 2013; Hanula et al., 2015; Roberts et al., 2017; Wagner et al., 2019; Milam et al., 2022; Ulyshen et al., 2022). As a result, the bulk of native bee research in the interest of pollinator conservation has focused on early-successional plant communities over other natural communities such as closed-canopy forest (herein referred to as ‘forests’; Ulyshen et al., 2023).

Limited research addressing native bee communities within forests has primarily focused on the understory (e.g., Rodríguez and Kouki, 2015; Roberts et al., 2017; Smith et al., 2019; Rivers and Betts, 2021; Eckerter et al., 2022; reviewed in Ulyshen et al., 2023). This sampling strategy, although accessible and convenient, is potentially problematic since the understory contains only a fraction of the vertically distributed resources available to bees (Saunders, 2018; Smith et al., 2019; Requier and Leonhardt, 2020; Urban-Mead et al., 2021). Indeed, field experiments that sampled at heights reaching beyond the understory have firmly established that bees are well distributed throughout temperate forest canopies (Ulyshen et al., 2010, 2020; Campbell et al., 2018; Cunningham-Minnick and Crist, 2020; Simon et al., 2021; Urban-Mead et al., 2021; Allen and Davies, 2022; Milam et al., 2022; Cunningham-Minnick et al., 2023), sometimes at significantly greater abundances than in the understory (Ulyshen et al., 2010; Campbell et al., 2018; Urban-Mead et al., 2021). Thus, sampling approaches that rely on understory sampling alone (e.g., Roberts et al., 2017; Harrison et al., 2018; Wagner et al., 2019) are informative but may not sufficiently characterize the full breadth of the forest bee community, although the magnitude and meaningfulness of this bias – from both a statistical and conservation perspective – remains uncertain. Further research in forests from a broader geographic range are necessary to gauge the extent to which vertical sampling affects the characterization of forest bee communities.

Milam et al. (2022) addressed the concern that only sampling forest understories may bias results and influence conclusions when comparing forest and early-successional communities in restored barrens. This study demonstrated how sampling the forest canopy does not change the conclusion drawn when comparing bee communities of the forest understory to the restored barrens: bee communities in barrens were more abundant and diverse. Though not the focus of their study, the authors also found significant differences in bee species composition between the forest understory (1.0 m) and canopy (8.6 m) in stands ~20 m tall, yet did not compare abundance or species richness between strata. Similarly, Ulyshen et al. (2010) sampled the forest understory at 0.5 m and canopy at 18.8 m near trees with an average height of 31.7 m and found a unique canopy assemblage with greater abundance, species richness, and diversity than the understory. Urban-Mead et al. (2021) sampled at 1.0 m and 22.5 m in a stand with an average maximum canopy height of 25 m and also found that the bee community was more diverse in the canopy but only more abundant for one of the three sampling years. Though regional differences (e.g., landscape composition or configuration; bee or plant species) may explain variation in vertical stratification of species among studies, the interaction between trap positioning and vegetation height may play an important role in the distribution of native bee communities (Geroff et al., 2014; Roberts et al., 2017; Ulyshen et al., 2023). Thus, the variation in results among these studies could also be due to differences in the maximum canopy height of these forests, which may affect the accessibility of canopy resources, or differences in the relative position of bee traps relative to the canopy, which could affect the proximity of the traps to different resources (e.g., canopy flowers, dead wood, alternative food sources; Cunningham-Minnick and Crist, 2020; Urban-Mead et al., 2021; Ulyshen et al., 2023). Of course, the maturity (size and flowering) and life histories of the tree species that comprise the forest community are likely to determine the availability of preferred resources throughout the canopy and thus the presence of particular bee species at different strata (Cunningham-Minnick and Crist, 2020; Urban-Mead et al., 2023). These points raise the question of whether the vertical stratification of bee communities can be better predicted from height above the ground or in relation to the height of the canopy. The inconsistency in findings across studies highlights the issue of how authors define each stratum, which complicates comparisons among studies and is exacerbated by studies that sampled at three (Ulyshen et al., 2020), four (Cunningham-Minnick et al., 2023), or more (Cunningham-Minnick and Crist, 2020) heights. Ulyshen et al. (2020) reconciled this issue with their recommendation of sampling at 5 m above the forest floor following findings that the bee assemblages at 1 and 15 m above the ground were distinct in composition while the assemblage at 5 m was not statistically different from either. Still, the variation in canopy height among forest types and forest composition based on geographic locations suggests the need for a more standardized approach that can account for these differences.

The goal of this study was to understand the extent to which sampling at different forest strata influences how forest bee communities might be characterized, and in turn, its implications for understanding broader environmental relationships. We vertically sampled a total of 47 forest sites for one year (17 sites in 2020; 30 sites in 2021) at 5-m increments from the understory into the upper reaches of the canopy to compare the effects of four potential sampling strategies on the characterization of the vertically stratified bee community: (a) understory sampling only, (b) understory and midstory sampling, (c) understory and canopy sampling, and (d) understory, midstory, and canopy sampling. Specific objectives were to: (1) determine if vertical stratification is best described in terms of absolute height above the ground or in relation to the height of the canopy, (2) compare abundance and diversity among these sampling strategies, and (3) examine whether sampling strategy has the potential to influence conclusions about relationships with the environment (e.g., microhabitat).

Traps sampled a total of 4,765 bees of 109 species, including 258 specimens that could not be identified to species due to body damage and unresolved taxonomy; for instance, 131 unidentified specimens were of the subgenus Lasioglossum (Dialictus), a taxonomically difficult group. The specific composition of this community and close examination of its ecological relationships is outside the scope of this study and will be published separately.

Generalized linear mixed effects models demonstrated that variation in the vertical distribution of bee abundance and species richness was better explained by the height of the trap in relation to the mean canopy height than relative height in relation to the maximum canopy height and the absolute height from the forest floor ( Table 1 ; Figure 2 ). Subsequent model selection of strategies with the relative mean canopy height covariate showed that bee abundance and species richness were highest when all strata were considered followed by the strategy that included only understory and midstory strata, while abundance and species richness were lowest when only the understory was sampled ( Figure 3 ). This pattern was also observed when comparing species richness estimates among strategies that considered sample coverage ( Table 2 ) and collectively demonstrates that unique species occurred within each stratum.

The predictor of herbaceous cover was included within the best model for all strategies of the abundance response, as well as strategies modeling the richness response ( Table 3 ; Supplementary Table A2 ). Herbaceous cover was the only environmental predictor in best models for abundance not associated with mature trees, though coarse woody debris on the ground was included species richness models. The explanatory power of herbaceous cover for variation in bee abundance data was highest for the strategy only including understory bees (0.147) and lowest when bees in all strata were pooled (0.046), demonstrating a bee in the midstory or canopy is much less affected (< 33%) by resources offered by herbaceous plants than a typical bee found in the understory ( Table 3 ). This pattern was also apparent when considering only variation explained by the model (understory: 53%, all strata: 23%), affirming this difference in relationships with herbaceous cover among strategies is due to biological responses of bees at different strata and not a modeling artifact (e.g., differences in covariates among models, unmeasured covariates; Table 3 ). The same pattern was found when considering both tree- and non-tree-associated predictors ( Table 3 ). Modeled strategies explaining species richness also demonstrated a decrease in explanatory power of the dataset and model with the addition of bees in higher strata, meaning that variables in the model were explaining the occurrence of bees in the canopy poorer than bees in the understory. However, the strategy including all strata explained more variation in the data when all understory predictors were considered compared to the strategy of understory plus canopy. Further, sampling all strata showed that non-tree-associated factors (coarse woody debris on the ground and herbaceous plants) played a larger role in explaining the bee community than strategies including the understory and just one higher stratum ( Table 3 ). Slopes of linear relationships between response variables and predictors did not vary among strategies with the exception that the understory and midstory sampling strategy in early spring in response to tree foliage was not significant, while the other strategies responded positively ( Figures 4 , 5 ; Supplementary Table A3 ).

A segmented modeling approach revealed that relationships between bee abundance and each predictor included negative and positive slopes for most strategies, demonstrating that the relationship was not linear ( Figure 4 ). Furthermore, there were differences in segment slopes within most strategies both in direction and magnitude (mean or confidence interval on the positive or negative side of zero, respectively). Relationships of bee species richness via segmented regression similarly showed variability in direction and magnitude of fitted slopes among strategies and covariate values ( Figure 5 ). Notably in the understory, there was high uncertainty that the slope of all segments across some predictors (i.e., foliage cover, basal area of deciduous trees) was different from zero; meaning that bees sampled in the understory were not responding consistently negatively or positively to environmental variables of the understory ( Figure 5 ). Among strategies and within covariate values, there were many occurrences of non-overlapping confidence intervals, indicating that interpretation of the bee–environment relationship will differ by sampling strategy. Finally, bee community composition differed among all strata by season combinations except the midstory vs canopy in the early spring and summer, as well as the understory vs canopy and midstory vs canopy in the late spring ( Supplementary Table A4 ).

Our findings highlight that study design approaches, including sampling strategies and their analysis, can change how forest-bee communities are characterized. We demonstrate that not sampling higher strata may lead to poor estimates of forest bee community metrics and potentially misrepresent forest bee habitat associations. We discuss these issues as well as how they are likely to bias our understanding of bee communities within forests and the conservation value of closed-canopy forest systems.

The current approach to most forest-bee studies largely limits 1) sampling of the bee community and environmental factors to the understory in the spring before leaf-out and 2) analyzing relationships of bee communities with environmental covariates using linear relationships. We demonstrate that only sampling bees and resources within a single season and stratum, or even two strata, does not represent the vertical stratification and seasonal dynamism of the forest bee community. Therefore, to prevent underestimation of the abundance and richness of forest-bee communities, we recommend that future studies of native bees in forests consider the bees occurring in at least the canopy, midstory, and understory in respect to the mean canopy height throughout a greater part of the year. We further urge researchers to consider analyzing the relationships between bees and local forest conditions at these strata using non-linear methods. Doing so will allow forest scientists and land managers to better evaluate the role of forests in conserving native bee communities. Notably, the implications of these recommendations may complicate the feasibility of land managers to survey or monitor forest-bee communities at this level; however, the sampling strategy we have selected should facilitate the discovery and description of mechanisms shaping the distribution of forest bees and other pollinator taxa, and modifications will likely be needed with additional research characterizing forest-pollinator communities. Following the guidelines presented in this paper will more broadly increase our understanding of the role of closed-canopy forests across forest types in supporting native pollinators, help standardize sampling and monitoring efforts of native bee and other pollinator communities, and further contribute to a deeper understanding of the currently undescribed mechanisms explaining forest pollinator distribution and conservation needs.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The manuscript presents research on animals that do not require ethical approval for their study.

MC: Conceptualization, Data curation, Formal Analysis, Investigation, Methodology, Project administration, Software, Visualization, Writing – original draft, Writing – review & editing. HR: Conceptualization, Data curation, Investigation, Methodology, Project administration, Writing – original draft, Writing – review & editing. JM: Data curation, Investigation, Methodology, Resources, Writing – review & editing. DK: Data curation, Funding acquisition, Investigation, Methodology, Resources, Writing – review & editing.