Data from: Clay larvae do not accurately measure biogeographic patterns in predation
Data files
Jan 29, 2025 version files 91.60 KB
Abstract
Aim: Spatial variation in predation can shape geographic patterns in ecology and evolution, but testing how predation varies across ecosystems is challenging as differing species compositions and defensive adaptations can mask underlying patterns. Recently, biogeography has borrowed a tool from ecology: clay prey models. But clay models have not been adequately tested for geographic comparisons, and a well-known problem –that clay prey only appeal to a subset of potential predators– could bias detected geographic patterns whenever the relative importance of predator guilds varies among sites. Here, we test whether clay larvae accurately capture geographic differences in predation on real larvae.
Location: 90° of latitude and >2000 m elevation across the Americas.
Taxon: vertebrate and invertebrate predation on ‘superworms’ (Zophobas larvae).
Methods: Across six sites that vary dramatically in latitude, elevation, and biome, we quantified predation on live, dead, and clay larvae. We physically excluded vertebrate predators from some larvae to distinguish total predation and invertebrate-only predation.
Results: Predation on live superworms almost doubled from our high-elevation high-latitude site to our low-elevation tropical site. Geographic patterns were consistent among live and dead larvae, but clay larvae missed extremely high predation at some sites and therefore mis-measured true geographic patterns. Clay larvae did a particularly bad job at capturing geographic patterns in predation by invertebrates, although sample sizes for invertebrate predation were small.
Main conclusions: Clay larvae are inappropriate for comparing predation rates across sites. They should be abandoned for biogeographic studies and reserved for comparisons within, rather than across, predator communities.
README: Clay larvae do not accurately measure biogeographic patterns in predation
https://doi.org/10.5061/dryad.dz08kps4r
We tested whether clay models of larvae accurately detect geographic differences in predation experienced by real larvae, by pairing real and alive, real and dead, and clay model larvae. Attack rates were recorded after 24 hours. Data were collected from 6 sites in Canada, Panama, Colombia, and Argentina. Live larvae were not used in Argentina.
Description of the data and file structure
Data are compiled in a single csv file, which contains raw data from all experimental runs at all sites. Missing values (i.e. we could not find larvae) are already removed.
Data are analyzed in a single R script.
Variables are described in the R code and below:
Geographic variables: country, region, lat = latitude (decimal degrees), elev = elevation (m above sea level), siteID = short hand for each of the 6 sites in our study
Date variables refer to start of the 24 h experiment: year, month (numeric), date = year.month.day
Experiment set up variables:
plot = 1 to 30 (1 larvae per larvae-type per plot);
cage.treat = cage treatment, either caged (CG) to exclude vertebrates and measure predation by invertebrates only, or uncaged control (CT) to measure predation by vertebrates and invertebrates (CT)
prey.type = type of superworm (Zophobas) larvae: alive, dead, or clay model
Predator marks: we attempted to assign predator marks to predator guild as commonly done in clay bait studies, but ultimately were not convinced that this was reliable. These columns are not used in analyses other than to calculate whether the larvae was attacked or not. 1 = yes (marks present) 0 = no marks
Predation variables: damaged = was the larvae damaged? 0 = not damaged, 1 = damaged (ie attacked). lost = '1' means larvae was gone but the plot and larvae marker were refound (ie larvae eaten)
Code/Software
The R script is formatted to be read in R Studio with the outline option visible (this provides a structured table of contents to help navigate). The R script includes all analyses and code to make the data panels for all figures, though final figures were compiled in PowerPoint.
Libraries needed are loaded at the beginning of the script.
Statistical results are from lme4 (for binomial mixed models, v 1.1.27.1), car (for likelihood ratio tests to compare models, v 3.0.12) and emmeans (for post-hoc comparisons within significant main effects, v 1.5.1).
Visual inspection of data and results uses lattice and visreg.