Planting native wildflowers improves vacant land as bee habitat in a post-industrial city

As people leave post-industrial cities, abandoned homes are demolished and transformed into vacant lots. These greenspaces have been demonstrated to provide habitat for urban wildlife and supply ecosystem services to communities. In the post-industrial city of Cleveland, Ohio, U.S.A., approximately 37% of the state’s bee fauna has been collected within vacant lots.

Our goal was to determine if planting native wildflowers (“pocket prairies”) on vacant land would improve these sites as bee habitat. We hypothesized that pocket prairies would support a greater proportion of the regional bee species pool, represented by Metropark grassland bee communities in the suburban landscape, compared to unaltered vacant lots. Using pan traps and hand vacuums, we sampled bees in each treatment from June to September 2019.

We collected 1,087 bees representing 24 genera and 81 species. Bees visited over 30 floral species, including native wildflowers and urban spontaneous vegetation. Metropark grasslands supported a higher bee species richness and diversity than urban pocket prairies. Both Metropark grasslands and pocket prairies supported a higher bee abundance, diversity, and species richness than urban vacant lots.

Synthesis and Applications: Despite the substantially smaller extent of the pocket prairies, these habitats supported a similar bee abundance to the Metropark grasslands. Bees foraged on intentionally planted wildflowers and non-native spontaneous vegetation, highlighting the importance of managing both components in urban greenspaces. Our results suggest that greening vacant land can improve post-industrial cities as bee habitat.

Methods
Local bloom assessment  

Bloom assessments occurred monthly in 2019 (June 11, July 10, August 6, September 17). In the center of each pocket prairie and vacant lot, we created a 7 × 15m grid, randomly selected six 1m2 grid quadrats, and placed a 0.5m2 PVC pipe square in the center of each quadrat. We identified every unique flowering species present in a quadrat to calculate bloom richness. Then, we counted all floral units to determine bloom abundance of each flowering species (Turo et al., 2021). Afterwards, we took five random measurements (mm2) of individual floral units for each species. Average bloom size per species was calculated and multiplied by bloom abundance to quantify total bloom area at a site. In each Metropark grassland, we established a 20m transect, randomly selected six 1m2 grid quadrats to sample along the transect and placed a 0.5m2 PVC pipe square in each quadrat’s center wherein the same bloom data were collected.  

Bee sampling

We used pan traps to assess the bee community present throughout the season. We sampled bees once a month in each treatment (June 11, July 10, August 6, August 12, September 17, September 19). Pan traps were plastic souffle cups (3.25 oz; Solo© Dart Container Corporation, Mason, Michigan, U.S.A.) 2/3 full of 1% dish soap solution (Blue Dawn© Proctor and Gamble, Cincinnati, Ohio, U.S.A.) painted fluorescent blue, white, or yellow (© Guerra Paint & Pigment Corp.). Pan traps were deployed at prairie flower height on 1m elevated stands which consisted of a 15 × 20cm corrugated plastic platform secured to an L-bracket and fixed to a step-in fence post. 12 pan traps were deployed for 24 hours at each site and collected the following morning. Insect samples were stored in 70% ethanol solution (© ThermoFisher Scientific, Waltham, Massachusetts, U.S.A.) and processed in the laboratory.

To evaluate patterns of bee foraging, we used hand vacuums (© Bioquip) to collect bees from July to September 2019 (July 15, August 12, September 20). Bees were actively sampled from blooming vegetation on sunny days with clear conditions between 10am and 4pm. During each observation period, blooming floral species were each observed for 4.5 minutes. Each time a bee landed on the cluster of blooms under observation, collection was attempted.  

Using dichotomous keys, we identified bees to species whenever possible (Ascher and Pickering, 2018; Gibbs, 2011; Gibbs et al., 2013). These identifications were verified by Sam Droege at the USGS Native Bee Inventory and Monitoring Lab in Laurel, Maryland, U.S.A. Then, we classified bees by functional traits, including body size, lecty (generalist/specialist), nesting guild, origin (native/alien), and sociality. Community-weighted means were used in functional trait analyses of bee body size (Ascher and Pickering, 2018; Sivakoff et al., 2018; Turo et al. 2021). Nesting guild categories included cavity nesting, colony nesting, pith nesting, soil nesting, wood nesting, and parasitic bees that do not use nests. Bee sociality was categorized as eusocial, parasitic, solitary, or subsocial. 

Statistical methods

Bee community assessment

To assess bee community composition differences by treatment, we conducted a non-metric multidimensional scaling (NMDS) analysis using the ‘vegan’ package (Oksanen et al., 2019) in R (R Core Team, 2022), applying a Bray-Curtis distance matrix (Bray and Curtis, 1957) to our compiled bee abundance data. We pooled bee abundance by site and treatment across the season. Using the ‘pairwiseAdonis’ package (Martinez Arbizu, 2020), we then performed a pairwise permutational multivariate analysis of variation to assess the significance of differences in bee community composition.  

To evaluate the effect of habitat establishment on bee communities, we used generalized linear models with bee abundance, Shannon-Weiner diversity, and species richness of bee samples collected from pan traps as response variables (Bates et al., 2015). We also used pan trap collections to compare body size and functional trait abundance between treatments. In our analyses, we screened three explanatory bloom variables (abundance, area, richness) for normality and variance inflation factor (Peterson, 2020). We also included a fixed effect for treatment and an offset term for the number of viable pan traps. Lastly, we considered mixed effects, including a random intercept for month. We used a stepwise backwards model selection approach to select the best fit model for each response variable (Fox and Weisberg, 2019).

Plant-bee network analysis

To assess to what extent bees used the seeded native prairie plants, we characterized bee foraging by generating plant-bee networks of vacuumed bee samples. Using the ‘bipartite’ package (Dormann et al., 2008), we constructed networks of bee foraging by treatment and site (Metropark Grassland, Pocket Prairie, and Vacant Lot). We used three indices in the ‘grouplevel’ function to assess the structure of these networks: generality (mean number of plant species a bee visits), mean number of shared partners (mean number of plant species any two bee species both interact with), and niche overlap (mean similarity of plant interactions between bee species). We also calculated three indices in the ‘networklevel’ function, including edge betweenness centrality (measure of centrality of a bee or plant species in a network), linkage density (ratio of realized plant-bee interactions to all possible interactions), and H2’ (measure of network specialization). After we calculated these indices, we generated 1,000 null models for each network using the ‘nullmodel’ function and the ‘r2dtable’ method (Dormann et al., 2008), calculating the same indices for each null model. Then, we calculated mean index scores from each null model for all three indices, divided them by the standard deviation of the null model scores, and used the resulting values to correct raw index values as z-scores. This was done to ensure that our descriptions of plant-bee networks were not an artifact of sampling intensity or web dimension (Dormann et al., 2009).