Data and code from: Roadside pollinator populations and the negative impact of intensive roadside vegetation maintenance
Data files
Sep 28, 2026 version files 435.34 KB
-
README.md
6.55 KB
-
ROW_abundance.csv
35.06 KB
-
ROW_beta_all2023_2024_PRECLEAN.csv
12.97 KB
-
ROW_beta_all2023_2024.csv
13.30 KB
-
ROW_beta_all2023.csv
9.67 KB
-
ROW_beta_all2024.csv
13.26 KB
-
ROW_beta_bee2023_2024.csv
8.60 KB
-
ROW_beta_bee2023.csv
6.21 KB
-
ROW_beta_bee2024.csv
8.58 KB
-
ROW_beta_ff2023_2024_PRECLEAN.csv
4.61 KB
-
ROW_beta_ff2023_2024.csv
4.94 KB
-
ROW_beta_ff2023.csv
3.69 KB
-
ROW_beta_ff2024.csv
4.92 KB
-
ROW_event.csv
5.82 KB
-
ROW_site.csv
1.91 KB
-
ROW_spp.csv
180.66 KB
-
SM-DRYAD-AnalysesROW_POSTREVIEW.R
114.62 KB
Abstract
Insect population declines are increasingly well documented, and pollinators are amongst these threatened taxa. However, modified roadside habitats may be suitable for supporting pollinator populations.
This study used a landscape-scale natural experiment to test how different roadside maintenance practices affect bees (Hymenoptera: Anthophila) and flower flies (Diptera: Syrphidae). Treatments studied were Low Intensity (≤2 mowing events per year and spot-spraying/hand-pulling of invasive plants) and Business As Usual (several mowing events per year and heavy use of herbicide). Pollinators were collected at 31 sites, 24 times each, over 2 years (372 site-visits/year).
We found that Business As Usual practices significantly reduced abundance and species richness of pollinators. However, community composition was similar across both treatments, indicating that roadsides support a fairly homogeneous pollinator community, regardless of management.
To increase pollinator abundance in roadsides, we recommend that managers adopt practices similar to the Low Intensity management of our study.
Dataset DOI: 10.5061/dryad.wpzgmsc29
Description of the data and file structure
Sites with different treatments were sampled for bees and flower flies, which were later identified and counted in the lab. Each site, or cluster of sites, was paired with another site/cluster with a different treatment. Treatments were: Low Intensity (≤2 mowing events per year and spot-spraying/hand-pulling of invasivs), Business As Usual (several mowing events per year and heavy use of herbicide), and Hydroseeding (as in Low Intensity but site previously prepared with a mix of forbs and grasses, but these sites were removed from the final analyses (see script)). Specimens were collected at 35 sites (31 without HS sites), 24 times each, over 2 years (420 visits/year); each collecting event had the start time recorded. Some variables have the same name between datafiles to aid on data wrangling.
Files and variables
File: SM-DRYAD-AnalysesROW_POSTREVIEW.R
Description: Annotated script for analyses
File: ROW_abundance.csv
Description: Abundance data
Variables
- site: site codes were specimens were collected
- event: unique event identifier using date and period of the day when the specimen was collected
- ff_ab: abundance of flower flies
- bee_ab: abundance of bees
- start_time: start time of collecting
- start_time_mod: modified start time to decimals
- notes_ab: notes on the data from this row
File: ROW_beta_all2023.csv
Description: data file for beta analyses; bees+flower flies for 2023. First column is for site names, remaining columns are for numbers of that species collected at each site.
File: ROW_beta_all2023_2024.csv
Description: data file for beta analyses; bees+flower flies for 2023+2024. First column is for site names, remaining columns are for numbers of that species collected at each site.
File: ROW_beta_all2023_2024_PRECLEAN.csv
Description: data file for beta analyses; bees+flower flies for 2023+2024, with some aff. and cf. Syrphidae identifications corrected to their final identification and also removing p_sp. First column is for site names, remaining columns are for numbers of that species collected at each site.
File: ROW_beta_all2024.csv
Description: data file for beta analyses; bees+flower flies for 2024. First column is for site names, remaining columns are for numbers of that species collected at each site.
File: ROW_beta_bee2023.csv
Description: data file for beta analyses; bees for 2023. First column is for site names, remaining columns are for numbers of that species collected at each site.
File: ROW_beta_bee2023_2024.csv
Description: data file for beta analyses; bees for 2023+2024. First column is for site names, remaining columns are for numbers of that species collected at each site.
File: ROW_beta_bee2024.csv
Description: data file for beta analyses; bees for 2024. First column is for site names, remaining columns are for numbers of that species collected at each site.
File: ROW_beta_ff2023_2024.csv
Description: data file for beta analyses; flower flies for 2023+2024. First column is for site names, remaining columns are for numbers of that species collected at each site.
File: ROW_beta_ff2023.csv
Description: data file for beta analyses; flower flies for 2023. First column is for site names, remaining columns are for numbers of that species collected at each site.
File: ROW_beta_ff2023_2024_PRECLEAN.csv
Description: data file for beta analyses; flower flies for 2023+2024, with some aff. and cf. Syrphidae identifications corrected to their final identification and also removing p_sp. First column is for site names, remaining columns are for numbers of that species collected at each site.
File: ROW_beta_ff2024.csv
Description: data file for beta analyses; flower flies for 2024. First column is for site names, remaining columns are for numbers of that species collected at each site.
File: ROW_event.csv
Description: File for establishing the unique event identifier codes
Variables
- event: unique identifier event name
- date: date of collection
- date_mod: date of collection modified for day of the year (1 = first, 365 = last)
- period: morning or afternoon
- year: year of the event
- season: season in which is collected (astronomical seasons)
- round: column to account for which round of collecting the event was associated to (more for collector's control during data gathering)
- notes_ev: notes on the data from this row
File: ROW_site.csv
Description: Information on the sites
Variables
- site: site unique name
- treatment: treatment applied to site
- code: abbreviation of the treatment applied
- county: county or municipality to which the site belongs to
- lat: latitude (in decimal degrees)
- long: longitude (in decimal degrees)
- pair: pair/cluster the site belonged to
- notes_site: notes on the data from this row
File: ROW_spp.csv
Description: Species collected at each site
Variables
- code: abbreviation of the treatment applied
- site: site unique name
- high_taxon: higher taxon to which the species belongs to
- genus: genus where the species belongs to
- sp: species name (abbreviated for analysis)
- sex: sex of the specimen
- event: unique identifier event name
- ab: number of specimens collected
- barcode: if specimen was barcoded or not
- notes_sp: notes on the data from this row
Code/software
Analyses were conducted in R Studio 2025.05.0, build 496, with R v4.3.2 (R Core Studio, 2024), using the packages ‘dplyr’ (Wickham et al., 2023), ‘tibble’ (Müller & Wickham, 2024), ‘tidytext’ (Silge & Robinson, 2016) and ‘tidyverse’ (Wickham et al., 2019) for data manipulation and plotting, ‘glmmTMB’ (Brooks et al., 2017), ‘lme4’ (Bates et al., 2015) and ‘lmerTest’ (Kuznetsova et al., 2017) for linear mixed modelling, ‘DHARMa’ (Hartig, 2022) for testing model fits, ‘multcomp’ for abundance and richness graphs (Hothorn et al. 2008), and ‘emmeans’ (Lenth et al., 2025) for estimated marginal means, confidence intervals, and pairwise comparisons.
I kept notes in the script to explain the reasoning behind some choices in the data wrangling and analyses.
Access information
n/a
Management effects on pollinator species abundance and richness
All sampling events were aggregated annually for each site to obtain a single value of abundance and species richness per site per year. We also calculated total species richness per site as the total number of unique species observed over the two years of sampling. Flower fly and bee data were analyzed both separately and together (as ‘pollinators’). Analyses were also repeated without the most abundant species from each taxon (i.e. the flower flies Sphaerophoria philanthus (Meigen) and Toxomerus marginatus (Say), and the bees Apis mellifera Linnaeus and Bombus impatiens Cresson (Hymenoptera: Anthophila: Apidae)) to evaluate their influence on results. Effects of the different ROW management treatments on pollinator abundance and species richness were tested using three separate generalized linear mixed models (one for abundance, one for annual species richness, and one for total species richness), accounting for year (where applicable), treatment, and their interaction as fixed factors, and site identity nested within block as random effects. Models were tested with different distributions to identify the best fit, i.e., Gaussian, Poisson, zero-inflated Poisson, and negative binomial (see Tables S2 and S4 and S7-S9 for the selected distributions). Model goodness-of-fit was tested through simulation of residuals and fitted models were checked for dispersion, within-group deviation from uniformity, and homogeneity of variance. We used the ‘summary’ function of the base package of R to evaluate the significance (p < 0.05) of fixed effects. When the year x treatment interaction was non-significant (p > 0.05), we re-ran models without the interaction.
Management effects on community composition
For analyses of community composition and beta diversity we used the package ‘vegan’ (Oksanen et al., 2025). To compare community composition of flower flies and bees between management types, we combined data from both years and included common species in the dataset. We used Nonmetric Multidimensional Scaling (NMDS) to plot and evaluate effects of ROW management on pollinator community composition. Species abundances were standardized using a Hellinger transformation, ordinations (up to 5 axes) used Bray-Curtis distances, and ordinations with stress values <0.2 were considered acceptable. We also tested for differences in community composition using the same dataset converted to presence–absence format and using Jaccard distances. In both cases, the lowest-dimensionality acceptable ordination was tested with ‘envfit’ to evaluate whether treatment was significantly associated with the position of sites in ordination space. Beta diversity was compared among management types using the ‘betadisper’ function, which calculates the average distance of each site from the multivariate mean (centroid) of each management type (higher distance values = more heterogeneous community in the treatment). Significance of the dispersion differences between management types was assessed using ‘permutest’. In addition, we tested for indicator species associated with either management type using the ‘indicspecies’ package (De Cáceres & Legendre, 2009).