Results

Description of biodiversity

Our final dataset included 26,716 individual bees representing 5 bee families, 30 genera, and 144 species. See [CITATION FOR DRYAD SUBMISSION] for the complete dataset and list of species. We collected 33% of the total number of bee species that have been found in Pennsylvania (Kilpatrick et al., 2020). The species accumulation curve for these data shows a leveling off pattern but did not reach an asymptote (Figure 2A). The abundances among species show a typical rank abundance curve with a small number of very abundant species and many rare species (Figure 2B). Ten species had over 1,000 individuals while over half of the species had 5 or fewer individuals. We captured by far the most individuals and species from the family Apidae (19,870 individuals and 47 species) followed by Halictidae (5,942 and 33), Andrenidae (477 and 28), Megachilidae (383 and 30), and Colletidae (44 and 6).

Biodiversity changes within years

We found very strong evidence for seasonal changes in all measures of biodiversity and on average month explained an average of 74% of the variation in our models (Figure 3, Table 1). Abundance and richness showed a hump-shaped pattern peaking in July. In our models, month explained nearly 90% of variation in abundance and richness. The number of bees captured per site increased by 193% between April and August, which were the lowest and highest values we observed (Figure 3A). Similarly, average richness increased by 89% between April and July (Figure 3B). Diversity also increased and decreased across the seasons, but less sharply than in richness, and peaked in August instead of July (Figure 3C). Diversity increased by 41% between May and August and decreased by 60% from August to October (Figure 3C). Because diversity incorporates both richness and evenness, the weaker pattern in diversity compared to richness is a consequence of evenness having a pattern nearly opposite that seen in richness (P < 0.0001, R2 = 0.68), highest in spring and fall and lowest in July.
Phylogenetic structure also varied between months. Mean pairwise distance dropped (becoming more clustered) by 1.9 standard deviations between May and July and then increased (becoming more even) by 1.1 standard deviations between July and October. The months of April, June, August, and September had intermediate values (Figure 3D). We observed limited variation in phylogenetic structure between sites resulting in our model explaining 87% of the total variation (Table 1). Community composition varied substantially among months with our multivariate model explaining 64% of the variation in bee communities (Figure 4A, Table 1). Spring months (April-June) all had distinct bee communities. July through September had similar compositions which were themselves distinct from spring months and October (Figure 4A).

Biodiversity change across years

We found very strong evidence for biodiversity change over time between 2014 and 2019 with year explaining an average of 42% of variation across all biodiversity metrics (Figure 3, Table 1). Abundance of bees captured declined by 48% between 2014 and 2019 (Figure 3E). Richness declined by 41% between the peak in 2016 and the lowest point in 2019 (Figure 3F). Similar to richness, diversity also declined after 2016, dropping by 59% between 2016 and 2019 (Figure 3G), though the model for diversity explained about half as much variation as the model for richness (Table 1).
Phylogenetic structure increased and decreased over time with the most clustered communities in 2014 and 2019, and the most even communities in 2016 (Figure 3H). Mean pairwise distance increased (became more even) by 1.8 standard deviations between 2014 and 2016, and then decreased (becoming more clustered) by 1.6 standard deviations between 2016 and 2019. Bee communities were quite stable across the first three years though they shifted slightly over time in the last three years (Figure 4B). Year explained about 36% of the variation in community composition.

Species patterns in seasonality, phenological breadth, and change over time

Looking across the 40 species for which we collected 30 or more individuals (Table A1), bee families varied significantly in seasonality (Figure 5, Figure 6, F3,36 = 10.91, P < 0.001). Species in the families Megachilidae and Andrenidae were collected an average of 54 days earlier than species in the families Apidae and Halictidae (Tukey tests, P < 0.006). In the family Megachilidae, Osmia species and Hoplitis pilosifrons were among the earliest emerging species, but Megachile mendica was most active in July and August (Figure 5). In the Andrenidae, allAndrena species are most active in April and May, butCalliopsis andreniformis was most abundant in July. Among species in the family Apidae, Eucera hamata was the only species with peak abundance in May. Species in the genera Anthophora, Ceratina, and Bombus were most active in June, though there is some variation among species within those genera. Other species in the family Apidae, including Ptilothrix , Melitoma, andMelissodes, as well as Eucera (Peponapis )pruinosa, peaked in July and August (Figure 5). Most species in the family Halictidae were most abundant in July and August, though twoAgapostemon species were active earlier (Figure 5, Figure 6).
Phenological breadth varied among families (Figure 5, Figure 6, F3,36 = 3.896 , P = 0.02) with two families that were significantly different. Species in the family Andrenidae had the narrowest breadth at 35 days on average, and species in the Halictidae were active the longest at 89 days on average (Tukey test, P = 0.02). The families Megachilidae (49 days) and Apidae (66 days) showed intermediate breadth. In the Andrenidae, all Andrena species had a breadth of less than 42 days, but Calliopsis andreniformis had a breadth of 79 days. In the Megachilidae, Osmia species’ breadths range from 21 to 58 days while Megachile mendica had a breadth of 71 days. Species in the Halictidae consistently had a wide phenological breadth, greater than 75 days, though Halictus ligatus had a relatively narrow window of 55 days. We split species in the Apidae into two groups, each consisting of related clades, that varied significantly in phenological breadth (t-test, t = -3.7, df = 18.3, P = 0.002). Species in the genera Bombus, Apis, Ceratina,and Xylocopa had an average breadth of 85 days, while species in the genera Eucera, Mellisodes, Ptilothrix, Melitoma, andAnthophora had an average breadth of 44 days (Figure 5, Figure 6).
We observed substantial species-level variation in the changes in abundance across years (Figure 5, Table A1). We detected no little-to no change in abundance for 26 species (P>0.1), 8 species showed moderate to strong evidence for decline (P<0.05), 5 more showed weak evidence for decline (P<0.1), and we found strong evidence for increase in 1 species (P<0.01). While bee family was not a significant predictor of changes in abundance (F3,36 = 1.813, P = 0.16), we observed some patterns among families and genera. Species in the families Megachilidae and Andrenidae were all stable. Other families showed mixed trends. For example, among species in the family Apidae, all carpenter bee species in the genera Xylocopa and Ceratina were stable except forCeratina mikmaqi which declined by 1 standard deviation unit (SD) between 2014 and 2019. Six out of eight species in the Halictidae showed evidence for declines, most species by about 1 SD, but Agapostemon virescens declined by 1.7 SD. Four species in the genus Bombusshowed pronounced declines ranging between 1.2 and 1.8 SD while twoBombus species were stable. Similarly, Melitoma taureadeclined by 1.3 SD. We saw radical variation among species in tribe Eucerini (Apidae): some species stable over time (Eucera and someMelissodes ), Melissodes desponsa had the biggest decline we observed (1.8 SD), and Melissodes bimaculatus increased by 2.2 SD, which was the only significant increase and also the largest magnitude of change we found.