Resource availability
To quantify the resources available to bees, we performed monthly field surveys at each site, except for the July survey at site 1 that was missed. We established five 100 x 2 meter transects along the edges of experimental crop plantings or in fallow fields within 500 meters of the hives at each site, for a total of 15 transects per survey period over the three sites. One exception to the 500m distance was one transect at site three which, due to measurement error, was placed 600 meters from the hives. Each transect was subsampled using ten 1-meter diameter circular quadrats. In each quadrat, we counted the number of individuals of each plant species. To estimate floral abundance, we counted either the number of flowers in a quadrat for plant species with non-clustered flowers (e.g. Caryophyllaceae) or the number of racemes (e.g. Fabaceae) or heads (e.g. Asteraceae) for plant species with clustered flowers. Because the absence of a plant species in a quadrat resulted in many zero counts, we estimated plant richness and diversity at the transect rather than the quadrat level. Plant richness and diversity were calculated at the levels of plant species and family, along with morphotype to align with pollen identification. Diversity was calculated using the Simpson D Index as described above for bee-collected pollen. The same categories established for pollen morphotypes in pollen pellets were used for plant resource morphotypes based on their pollen morphology (Table 2). We determined the impact of site and survey period on plant richness and diversity using a linear mixed model (proc mixed, SAS v. 9.4) with site and survey period as fixed effects, and their interaction as a random effect. A transect was the replicate in the model, and the fixed effects were tested against the two-way interaction error term.