Habitat structure drives the evolution of aerial displays in birds

Abstract

Physical properties of the environment may shape signalling traits by determining 11 how effective the signals are in affecting the behaviour of other individuals. 12 Evidence abounds of signalling environment driving the evolution of colours and 13 sounds, but little is known about its influence upon gestural displays. Here, we 14 performed a continent-wide phylogenetic comparative analysis to test the 15 hypothesis that habitat structure drives the evolution of aerial sexual displays in 16 passerine birds. We found that aerial displays are seven times more likely to evolve 17 in open habitats than in forests, likely as a result of physical properties that allow 18 aerial displays to transmit more broadly in open habitats. Our results provide an 19 emblematic example of how environmental factors may help predict the direction of 20 evolution of otherwise unpredictable sexual traits. The broader range of aerial 21 displays in open habitats may also mean that females can sample more males, 22 potentially leading to more intense sexual selection over open-habitat, aerial-23 displaying males. 24 25

Habitat structure drives the evolution of aerial displays in birds

“…We were able to find data about sexual display and habitat structure for 469 (19.2%) species of New World passerines belonging to 41 families (Menezes & Santos, ). The remaining species (for which data were not available) were excluded from analyses.…”

“…Overall, we were able to determine whether 470–499 passerine species (depending on how strict we were in treating a display as sexual) from 41 families exhibited aerial displays or not, which represents 19.2%–20.4% of the initial sample set (Menezes & Santos, ). The remaining species were classified as NA and thus dropped from the analyses (we applied this same decision to the habitat variable as well).…”

“…In each analysis, we pruned trees to match the corresponding species for which we had complete data. Of the 469 species that were included in the main analyses, 415 (88.5%) have had their phylogenetic position determined based on genetic data by Jetz et al (; see Menezes & Santos, ). We incorporated phylogenetic uncertainty by iteratively running each Bayesian analysis with 50 trees (Supporting ML Analyses were run with 1,000 trees).…”

“…We provide all the R codes used for handling data, running analyses and designing figures in Menezes and Santos (). Refer to them for further details in any procedure below.…”

“…These arguments result in 1,000 sampled estimates per model per tree. As hyperprior, we chose an exponential distribution with the mean seeded from a uniform distribution on the interval between zero and two times the maximum ML estimate for any parameter in that model across 1,000 trees [see Supporting ML Analyses (Appendix S2) for the maximum‐likelihood estimates, and codes in Menezes and Santos () for the hyperpriors]. We assessed chain mixing by visually inspecting parameter traceplots, likelihood traceplots, autocorrelation plots and acceptance rates using the R package coda (version 0.19.1; Plummer, Best, Cowles, & Vines, ).…”