Discussion

With 6 years of continuous sampling with passive traps we collected over 26,000 bees and 144 species. The leveling off of the species accumulation curve suggests we captured most, but not all, of bee biodiversity in the system (Figure 2A). The inability to fully document biodiversity (i.e., reach an asymptote in the species accumulation curve) is typical for other extensive bee monitoring efforts (Wilson et al., 2008, Russo et al., 2015) and species-rich invertebrate communities more generally (Gotelli and Colwell, 2001).
We found that all measures of community biodiversity varied dramatically within years and across years (Figure 3, Figure 4). Abundance, richness, and diversity all peaked in summer, though diversity to a lesser extent. By contrast, phylogenetic structure was most even in spring (May) and became most clustered in summer. Community composition also varied greatly within years with communities changing quickly early then becoming much more stable between July and September (Figure 4). This high-level of variation in species-level phenological patterns (Figure 6) resulted in the dramatic turnover of communities between seasons.
All measures of biodiversity also changed across years (Figure 3, Figure 4). The magnitude of changes within years was about twice as large for abundance, richness, and community composition than it was for changes across years. However, for diversity and phylogenetic structure, the changes within and across years were of similar magnitude. After 2016, all measures of biodiversity declined. Community composition also shifted in that time, but not dramatically (Figure 4). These changes in diversity metrics and composition were the result of, in part, 13 species that declined in abundance over time, which were dispersed across the bee evolutionary tree (Figure 5).

Insights from species-level changes in abundance

Species across the bee evolutionary tree showed a wide variety of phenological patterns (changes in abundance within years). Among the 40 species for which we had sufficient data for, we observed three general patterns, which could be called “phenological syndromes” illustrated in Figure 6. First, Andrena and Osmia species emerged early in the year and had narrow breadth. Second, species in the tribe Eucerini, (Melissodes and Eucera ), and other sister clades had narrow breadth but, in the summer, rather than in spring. The third group was composed of species with wide phenological breadth including the social and multivoltine species in the genera Bombus, Apis ,Xylocopa and Ceratina , and nearly all the sweat bees (family Halictidae). Monitoring of species that represent these different phenological syndromes is important because they provide unique ecological functions (Ogilvie and Forrest, 2017). For example, many of the early emerging bee species are of critical importance for early flowering plants such as spring ephemeral wildflowers, and these interactions may be particularly sensitive to disruptions from climate change (Kudo and Ida, 2013). And many crops such as apples and blueberries rely on pollination by early emerging wild bees (Isaacs and Kirk, 2010, Biddinger et al., 2018, Grab et al., 2019, Reilly et al., 2020).
We found that 33% of species had at least some evidence of declines while only 3% increased, and 65% percent showed no changes over time. For comparison, a study using museum records of 187 bee species in eastern North America found significant decreases in the relative number of samples in collections for 29% of species and increases for 27% of species (Bartomeus et al., 2013). Similarly, 38% of non-parasitic bumble-bee species in the UK show clear signs of decline (Williams and Osborne, 2009). While it is possible that we could have had more power to detect changes in rare species with more thorough sampling, we found significant changes among species with a wide variety of abundance (min = 37, max = 3774, mean = 894) and there was no correlation between species’ total abundance and amount of predicted change (r = 0.18, P = 0.26). Therefore, our finding of 65% of species being stable is robust and comparable to studies in North America and Europe. We did not find that bee family was a significant predictor of which bees are stable or declining. But, there were some clades that were more prone to declines than others, notably bumble bees (Bombus ) and sweat bees (Halictidae). In another case, 2 closely related longhorn bees (genusMelissodes ) showed large changes in abundance in opposite directions. More generally, this suggests that phylogenetic relationships are not a good predictor of species changes over time. Understanding which adaptations or life history traits are associated with population increases or decreases over time is likely to be a better approach (Williams et al., 2009). The great variation in species’ changes in abundance is also aligned with van Klink et al. (2022) who found that, on average, different insect species’ population trends are only weakly correlated.

Insights from multiple measures of community biodiversity

Our thorough collections of bees throughout the seasons, and measurements of communities using a variety of metrics, highlighted the unique biodiversity of bee communities in the spring. Measures of abundance and richness suggested that biodiversity in the spring is low. However, diversity was similar in April and May as it was in July and August despite huge differences in richness. This relative elevation of diversity in the spring was a consequence of greater evenness, or more equal abundances among species. The total amount of spring species captured across all sites and years was also high despite the low abundances. Using rarefaction to standardize the number of individuals collected, we detected 58 species per 900 individuals in April compared to 40 in July. The month of May was also the time with the most phylogenetically even (overdispersed) communities. A non-mechanistic interpretation is that in May, spring bees (largely from Andrenidae and Megachilidae) and some summer bees (mostly in Apidae and Halictidae; Figure 5) were both active resulting in long branch lengths between pairs of species. This parallels results by Ramirez et al (2015) who found that orchid bee communities in Colombia were much more phylogenetically even in the transition period between wet and dry seasons. Composition of bees also shows great uniqueness of spring bee communities and the fast turnover communities resulting in totally unique communities in April, May, and June. These unique aspects of spring biodiversity would be completely missed by looking at only abundance and richness measures and not diversity, phylogenetic structure, and composition. This suggests that studies seeking to understand the phenological changes of bee communities and the impacts of environmental change on spring bees, need to have robust sampling and multiple measure bee biodiversity.
Repeated measures of bee communities across years suggested a loss of community biodiversity over time, though the patterns of declines depend on which metric you look at (Figure 3). Total abundance showed a linear decline over time which mirrors the patterns we observed for many individual species (Figure 3E, Figure 5). The reasons for these changes over time are not clear. While habitat loss, land-use changes, and pesticide use all likely impacts bee communities in this system, these were all relatively unchanging over the course of this study (Biddinger et al., 2018). Changes in the floral resources of the flower strips where we sampled could have been a factor since they likely experience a reduction of plant diversity over time, as is typical in restored grasslands (Sluis et al., 2002). Climate could also be a driver of population declines, though longer-term data would be needed to test the effect of climate on bee abundance declines (Ogilvie et al., 2017, Kammerer et al., 2021). Other biodiversity metrics showed similar, but more nuanced patterns than abundance. Richness, diversity, and phylogenetic structure were steady or increasing for the first three years, and then declined for the last three (Figure 3). Similarly, community composition also shifted but primarily in the last three years (Figure 4B). Longer-term monitoring is needed to see if these declines are part of an ongoing trend or a result of year-to-year fluctuations.
From a bee monitoring and conservation perspective, changes in abundance, richness, and diversity are easy to interpret. In most cases, decreases in these metrics are problematic and suggest some environmental degradation is causing losses of biodiversity. However, metrics like composition, and phylogenetic structure are harder to interpret without a reference point but can reveal changes not seen in simpler measures (Tucker et al., 2017, Nerlekar & Veldman, 2020). For example, Tonietto et al. (2017) found that old fields, restored prairies, and remnant prairies all had the same abundance and diversity of bees but community compositions were different. And similarly, Frishkoff et al. (2014) found that one type of agricultural land-use did not change bird richness, but it did lead to more phylogenetically clustered communities compared to forest reserves. Going forward, more long-term bee monitoring studies are needed to determine if biodiversity measures like composition and phylogenetic structure provide unique and useful information for conservation efforts.

Implications for bee monitoring

There are a variety of bee monitoring approaches that range from standardized and repeated collections of bees with detailed taxonomic identification to visual observations of broad taxonomic groups that involve participation from the public. There are pros and cons to studies using methods on both ends of this spectrum (Woodard et al., 2020). Our approach involved continuous collecting using passive Blue Vane traps and species-level identification of all bees. The sampling throughout the year gave us the ability to quantify seasonal changes in biodiversity with fine resolution. The huge number of bees collected means we had 40 species with sufficient data to characterize phenological patterns. And the standardized sampling over many years allowed us to quantify changes in abundance over time, something many most studies have limitations with (Portman et al., 2020). However, it is important to highlight that studies using passive trapping need to be interpreted with caution as the data do not reflect true population sizes (Portman et al., 2020, Briggs et al., 2022). This is because some species are more attracted to traps than others and because trapping results are impacted by context (Kuhlman et al., 2021). While our data may not reflect the absolute abundance of species in the wild, it does show that standardized passive trapping is effective at measuring relative changes within and across years. Overall, the intensive type of monitoring of our study is a good approach to answer questions about community biodiversity change and the unique population dynamics across many co-occurring species.
Our sampling approach and experimental design have several constraints for its implementation on large-scale monitoring projects that aim to detect bee declines. First, collecting, processing, and databasing large numbers of bees is labor-intensive and taxonomic identification requires specialized skills and expertise. This makes specimen curation and identification untenable and impractical for large-scale projects. Second, tens of thousands of bees were killed in the sampling. Concerns have been raised that sampling many bees with Blue Vane traps could cause declines in some species (Gibbs et al., 2017). While we did not estimate how our collections impacted populations, the lack of correlation between the number of individuals captured and that species change over time (r = 0.18, P = 0.26) provides at least some evidence that this was not the case in our study. Third, implementing passive traps exclusively have inherent biases in the species they collect and these biases impact biodiversity metrics. While collections with other techniques would have resulted in different biodiversity measures, we know from our system that Blue Vane traps provide the most thorough sample of the whole bee community (Joshi et al., 2015). And finally, our collections are only from one relatively small area (Figure 1). Given the local nature of our dataset, the observed changes within and across years could be unique to our study area. However, similar phenological patterns and declines have been found in other studies (Bartomeus et al., 2013, Leong et al., 2016, Graham et al., 2021, Kammerer et al., 2021).