Predation pressure
To quantify bird predator presence, we used observation data from checklists submitted to eBird for species that are known to prey on lizards. All potential lizard predators included species that opportunistically feed on anoles, and were determined by considering the diet of the bird species listed by the Cornell Lab of Ornithology (https://birdsoftheworld.org/bow/home), or National Audubon Society (https://www.audubon.org/bird-guide). Miguel Landestoy also provided insight into the importance of certain Dominican Republic species as predator of anoles, based on personal observation.
We looked at all eBird checklists for Jamaica and the Dominican Republic from August 2010 – August 2020 and filtered to include both stationary and travelling checklists (omitting incidental observations), with a maximum of 5 hours duration, 10 observers, and 10 km distance. All data were accessed in September of 2020. These checklists were treated as presence-absence data (i.e., we assumed that all species not listed were not observed).
We then used a general additive model using a binomial error distribution to model the probability of each bird species being present throughout Jamaica and the Dominican Republic, granting us the ability to predict presence-probability in the vicinity of each lizard plot location. Specifically, modeled bird presence was based on mean annual temperature (bio1), temperature seasonality (bio4), annual precipitation (bio12) and precipitation seasonality (bio15) obtained from WorldClim (Fick and Hijmans 2017), and also considered checklist duration, location, number of observers, distance travelled, day of year the checklist was recorded on, and time of day at which observations began. All continuous variables were implemented with thin-plate splines, allowing the model to flexibly estimate the degree of curvature in the relationship between the variable and bird occurrence. In addition, we incorporated a two-dimensional thin plate spline based on the latitude and longitude of the checklist to both account for spatial autocorrelation into predictions, and to incorporate any unmeasured environmental factors that correlate with space. Predictions of probability of occurrence were then made for each species at the plot locations where lizards were surveyed, at a resolution of approximately 1 km2 grid cells. To obtain standardized estimates of predation pressure we considered a hypothetical eBird transect conducted by a single observer, lasting for 60 minutes, and spanning 1 km. Because bird occurrence varies between time of day and day of year, we made an ensemble of predictions for each plot location evaluating hypothetical transects every hour on the hour between 4 am and 3 pm, for all days of the year. To summarize per species predation pressure at each plot location we considered the maximum probability of presence across all hours and across the entire year, under the theory that this value best represents true occurrence probability (whereas lower values may be primarily influenced by detection probability from non-ideal times of day for sighting the species in question). In addition to using maximum predator occurrence we also considered mean occurrence, but results were qualitatively similar, and so we present only the maximum.
We used three complimentary sets of bird species to analyze predation pressure. First, we considered all potential lizard predators, second all cuckoo species, and third all pigeon species. The two taxonomically restricted sets were chosen to represent specialized lizard predators, and definitive non-predators respectively. Cuckoos in particular are important predators of anoles, and both islands have a specialized species of lizard-cuckoo. This led us to hypothesize that they might represent a group for which predation signals on community structure may be especially clear. In contrast pigeons are not predators (or competitors) with anoles, and thus serve as a neutral “control” group, for which we would not expect to observe any signal from either top-down or bottom-up forces. Analysis of pigeons thus allows us to more accurately interpret whether any observed effects are due to predator species per se , or just correlation with general bird presence. To obtain total predation pressure we summed the individual presence values for all bird species in each category (i.e. , all predatory birds, all cuckoos, and all pigeons [non-predatory control]). These probabilities of predatory bird presence provided us with quantifiable estimates of predation at a comparable spatial and temporal scale as the lizard community data.
To evaluate whether greater predator occurrence is associated with increased or decreased lizard community metrics we ran a linear mixed-effects model for anole species richness and abundance, the latter of which was square-root transformed to fulfill model assumptions of residual normality. Predictor variables included our metric of predator presence, habitat type (i.e., forest or human-modified), and island identity, with site (where 3 to 6 plots were grouped) as a random effect. We included two interaction terms. One was a nuisance term to account for potential interactions between habitat type and island, thereby allowing the effect of deforestation on anole community structure to differ between Hispaniola and Jamaica. The second was a hypothesis-driven interaction effect examining differences in the way predation pressure affected anole communities on the two islands. We also included mean annual temperature and mean annual precipitation to control for climate. We conducted backward model selection, dropping each term until only significant ones remained. All modeling was done in R 1.4.1103.