https://doi.org/10.5061/dryad.ngf1vhj55

Description of the data and file structure

Data and analysis for Habitat associations of wild bees: on the importance of phenology and the idiosyncrasy of scale by Dylan T Simpson, Colleen Smith, and Rachael Winfree.

The data needed to run the analysis are in Data_and_Code.zip: ~/Data/In, and the scripts are in ~/Code/. The file directories used in the scripts assume your working folder is this parent folder (i.e., the location of this readme).

For convenience, we have also included the intermediate products and outputs so one can start anywhere without having to run all the analyses (which are rather time and memory-intensive).

Details on the datasets are included in the supplement, but in short:

• Datasets are bee abundance data from surveys conducted by pan traps and, in some cases, vane traps.
• Data are compiled from 5 studies. These are described in the supplement, and most have been published elsewhere. Use of these data should cite the primary citations where possible (citations available in the supplement and main text).
• Collectively, data were collected at 165 sites.
• We focus on 84 species that were represented by ≥ 30 individuals at ≥ 10 sites. These 84 species were represented by ~30 K specimens. 
•  Additionally, the data folder includes species trait data. Specifically, body size, which was measured on specimens in the Winfree Lab; phenology, which was estimated from data collected by the Winfree Lab and from the American Museum of Natural History, which was curated by Ignasi Bartomeus (Bartomeus et al 2013, PNAS); and sociality, which is based on natural history information from a variety of sources, e.g., Michener, Discover Life, and personal communications with natural historians like John Ascher and Jason Gibbs

Details on the data tables themselves and the purpose of each R script are found in the two metadata files.

Files and folders

The .zip file included is meant to act as a working directory to recreate the analyses from the paper.

The following describes the contents of the files and folders. Additionally, there is another metadata file included in the folder's main directory, "metadata_FilesAndFolders.rtf", that includes descriptions of individual columns from each data table.

-Dylan Simpson, December 2024

Code: folder containing R scripts to compile data and run the analysis

• 0.1_ModelData.R: R script to combine survey data with species info (phenology, body size, sociality info) and landcover data, and then break apart data into the form used as model inputs
• 1.1_GeneratingNullExpectations.R: R script to run a null model version of the main analysis, in which the abundance of 12 example species is repeatedly modeled as a function of randomly generated data that have the same structure as our habitat data. The performance of these “null models” is used to define a null expectation of model performance given the scale-optimization procedure and the cross-scale correlation structure of our landcover data.
• 1.2_DefiningNullDistribution.R: An R script that processes the output of the previous script and saves as an output a table in which each row is an iteration of the randomization-based null model. The column of log-likelihood ratios is the numerical estimation of the null distribution.
• 2.1_ScaleSelectionAnalysis.R: This script runs the analysis. Within a set of nested map( ) functions, this script determines the appropriate null model for each species (i.e., which covariates should be included and in what form), models the abundance of each species as a function of each habitat at each scale, and measures the performance of every model against the null (i.e., covariate only) model.
• 2.2_SelectingSignificantModels.R: A processing script that determines which habitat associations were significant. The script pulls the scale-optimized models from the previous step, calculates p-values by comparing to the simulated null distribution, determines the critical p-value that maintains a false discovery rate of 5%, and then filters the results to only those models with a p ≤ the critical p.
• 3_Results.R: A script that summarizes and visualizes the results from the significant models. All figures are generated within this script, and all additional statistical tests (e.g., tests of the mobility hypothesis) are also done in this script.
• Functions: a folder of custom R functions that are called by earlier scripts
  • FDR_pCalculator.R: A function to determine the critical p-value to maintain a pre-determined false discovery rate. The algorithm to identify the critical p-value is taken from Benjamini et al 2006 (Biometrika), as summarized by Pike 2011 (Methods in Ecology & Evolution)
  • Plotting_functions.R: functions to make two plots that get called by the Results script. The first is the ‘scale selection bar plot,’ which plots the AIC weight of habitat associations at each scale. The scale of the model with the highest weight is taken as the scale of effect for that species-habitat association. These are the plots in panel 3 of Figure 1 and Figures S4 and S9. The other is a function to make the tile plot of species-habitat associations. This is figure 2 in the main text, though the final figure was modified in Adobe Illustrator to make it more readable.

Data / In: Folder with all input data. If run from the beginning, all R scripts and analysis should work with only the contents of this folder.

• Bee_bodySize.csv: table containing intertegular distance (ITD) measurements for each species. Although it is worth noting, this only contains info for 75 of our 84 species. Some species had been measured for a previous project by the lab group. Of the remaining species, we only measured those that had significant habitat associations and were thus included in the mobility hypothesis test.

• Bee_phenology.csv: table containing phenology data for each species.
• Bee_sociality.csv: table identifying the social strategy of species
• Habitat_proportions.csv: table with the proportion of each focal habitat/landcover around each, measured at each of 30 spatial scales (or, technically, spatial extents), and measured using each NLCD release between 2001 and 2021.

Data / Intermediate: folder to contain intermediate data products. If running the R scripts from the beginning, this will become populated with the following files. For convenience, I include here copies of these products produced by my runs of the code, so that a user can pick up anywhere they like in the workflow.

•  Model_Data: a folder containing .rds files of model input data for each species. Each .rds file is a list of survey data; each element is the same survey data paired with habitat/landcover data measured at a different spatial scale.

• Null_LLR_distribution.csv: a table in which each row is the result of one iteration of the randomization-based null analysis (described in the supplement and run by code/1.1_GeneratingNullExpectations.R and code/1.2_DefiningNullDistribution.R.
• Species_summary.csv: A table containing sample size and observed site occupancy for each species. Occupancy here is the number of sites at which a species was observed. The total possible is 165. Important note, however: Some sites were only measured in one season or another (e.g., spring but not summer, or vice versa). As a result, not every species was available to be sampled at every site. A robust estimate of occupancy would first filter to sites that were sampled during that species’  flight window (as we did in the main analysis for our models of abundance).
• Species_to_test_null.csv: a subset of the previous table, containing the species that were used to generate the null distribution of LLR values. Ten of these were randomly selected. We additionally added Augochlora pura and Ceratina calcerata.

Data / Out: folder containing analytical outputs. As above, this would become populated by running the R scripts in Code/, but I included my copies here for convenience.

• Model_fits: a folder containing individual .csv files with results for each species. Each of these .csvs contains summary info for the model of bee abundance as a function of each habitat at each scale.
•  nullResults.rds: an R list containing the results of the null analysis for each species. Each element is a dataframe for one species, in which each row is an iteration of the null analysis for that species. This is a post-processed file; for each iteration, the best model has been pulled out and the performance of that model has been measured.
• nullRuns: a folder containing the raw outputs from the null analysis for each species. Each .rds file contains a list of 1000 elements. Each element is the scale selection table for that run of the null.
• scaleSelectionTable.csv: a table with the raw output of the main analysis. It is the concatenated scale selection tables for every species and habitat. For each species and habitat, the table contains the model results for how that habitat, at each scale, predicts the abundance of that species.
• significantModels.csv: A subset of the table above that contains only the final, significant models. This was generated by first selecting the best model for each species-habitat combination (each model represents a scale, so this is the scale optimization step), then determining the significance of that scale-optimized habitat model.

Figures: an empty folder to contain figures created by the Results script.

Access information

Much of this data was collected for studies that have been previously published (see Related Works and the Data Description in this readme), but the data are not currently available elsewhere in this form.