A global analysis of bird plumage patterns reveals no association between habitat and camouflage

Data collection

We focused on the most prominent aspect of patterning: the four different types of spatial arrangement of pigmentation (Fig. 1). To collect plumage pattern information we referred to field guides covering all major landmasses: North and Central America (Sibley, 2000; Van Perlo, 2006), South America and Antarctica (De La Pena & Rumboll, 1998; Restall, Rodner & Lentino, 2007), sub-Saharan Africa and Madagascar (Langrand, 1990; Sinclair & Ryan, 2003), Europe, North Africa and Central Asia (Heinzel, Fitter & Parslow, 1995; Grimmett, Inskipp & Inskipp, 1999; Arlott, 2007; Arlott, 2009; Brazil, 2009), East and South-East Asia (MacKinnon & Phillips, 1993; Coates & Bishop, 1997; Robson, 2005), and Oceania (Beehler, Pratt & Zimmerman, 1986; Pratt, Bruner & Berrett, 1987; Simpson & Day, 2004; Robertson & Heather, 2005). Where multiple subspecies were present, we collected information on the nominate subspecies. Overall, we examined 8006 avian species (80% of class Aves) to categorize their patterning. Juvenile plumages are less frequently drawn in field guides so we were only able to categorise juvenile plumage information in 2,603 species (26% of class Aves). Species that were unpatterned and/or without habitat information were excluded from further statistical analyses, giving an overall sample size of 2,756 avian species (34% of the species considered), providing information for 2,593 adult and 2,104 juvenile avian species (see the list of species in Appendix S1).

We scored the plumage of both sexes as well as juveniles of each species for: an absence of patterns, mottled, scaled, barred and spotted patterns, both on the ventral and dorsal surface, separately (Fig. 1). Irregular mottled patterns were defined as pigmentation that did not have a clearly discernible motif or the putative motif had irregular borders, as frequently found on the dorsum of sparrows (in contrast with the stripes of species from the genus Gavia that have clear and neat boundaries in the stripes on the neck). Regular patterns had a motif with regular borders and were repeated across a patch of patterning (Fig. 1). Some species have multiple patterns on either the ventral or dorsal surface (e.g., the male zebra finch, Taeniopygia guttata, has barred patterns on the breast and spotted flanks): on the ventral surface, 62 and 64 species for males during the non-breeding and breeding seasons respectively, 70 and 72 species for females during the non-breeding and breeding seasons respectively, and 63 species for juveniles; and on the dorsal surface, 229 and 222 species for males during the non-breeding and breeding seasons respectively, 231 and 233 species for females during the non-breeding and breeding seasons respectively, and 102 species for juveniles. In these types of species, all of the different types of plumage patterns on each surface were scored and each pattern type was analysed separately. Where species exhibited variable patterns between moults, we collected the breeding and non-breeding plumage given that there may be variation in selection pressure on the different types of patterns exhibited. Squares, triangles and stripes also occur within the plumage of birds, but are comparatively rare (e.g., out of these three rare patterns stripes are the most common: 43 species). Stripes are comprised of regular longitudinal lines (unlike bars, which are pigmented transversal to the feathers axis). These rare types of patterns were excluded from the analysis due to low statistical power.

We examined all of the combinations of gender/age (male/female/juvenile), season (breeding/non-breeding), and body part (ventral/dorsal), individually. Herein we refer to each as biological combinations.

Species geographical distributions

Global geographical distributions of all avian species were obtained from Birdlife International & NatureServe (2012) for all species in the plumage dataset and treated as described in Somveille et al. (2013). Briefly, the polygons in the dataset represent the global distribution of species and we included only the polygons for which species presence was coded as extant or probably extant and the species origin was coded as native or reintroduced, and we excluded the polygons corresponding to species in passage (i.e., during migration). In addition, breeding distributions (defined as polygons corresponding to the areas where they are present only during the breeding season) were analysed separately from non-breeding distributions (defined as polygons where they are present only during the non-breeding season). We focused solely on land birds, as the geographical distribution of marine species is not well known and represents less than 3% of avian species.

We statistically examined our data for an association between habitat and plumage patterns for each biological combination. For adult breeding and non-breeding plumages, we used the breeding and non-breeding geographical distributions of bird species, respectively. For juvenile plumages, we used the breeding geographical distributions as this life stage occurs predominantly during the adult breeding season. Note that for resident species (i.e., non-migrants; 85% of avian species), the breeding and non-breeding geographical distributions are the same. Irregular patterning is found in 24% of the 8,006 avian species analysed (1,887 species) whereas regular patterning is found in 23% (1,858 species). Of the regular patterns (scaled, barred and spotted), scaled and spotted patterns are prevalent in 25% (470 species) and 30% (549 species) of avian species (across biological combinations) respectively, whereas barring is observed in 66% (1,220 species) of avian species (across biological combinations)–percentages are relative to the number of species with regular plumage patterns in at least one biological combination.

To avoid an underestimation or overestimation of ecological signal for regular patterns, we categorized regular patterns in two ways for our analyses: (1) all regular patterns grouped together as a single category, and (2) just barred plumage patterns due to these being the most frequent type of regular pattern occurring in bird plumage patterns (Gluckman & Cardoso, 2009; Gluckman & Cardoso, 2010). The results were qualitatively the same for the analyses of regular patterns together and barred plumage patterns alone, and we present the results for the category of all regular patterns together (a comparison of mottled versus barred patterns for all analyses are available in the Supplemental Information 1).

Phylogenetic comparative analysis

To examine the habitat in which species live, we analysed the habitat types found within their geographical distributions. For each species with irregular or regular plumage patterns, we quantified how open or closed is the habitat in which they live, a measure herein called habitat coverage. Information on habitat was obtained from the global MODIS land cover dataset (Channan, Collins & Emanuel, 2014). We coded forests pixels as closed habitat with a value of 1; savannas, grasslands, wetlands, croplands, snow and ice and barren or sparsely vegetated pixels as open habitat with a value of 0; and shrublands, woody savannas and cropland/natural vegetation mosaic pixels as partially closed habitat with a value of 0.5. For each species, we then quantified habitat coverage as the average habitat value across the pixels found within the species’ geographical distribution. This allowed us to obtain a continuous measure of habitat for each species during both its breeding and non-breeding seasons–which can be very different for migrants, representing ∼16% of avian species (Somveille et al., 2013). We then removed species for which no habitat information were available and no plumage patterns were present.

Plumage patterns may be phylogenetically conserved, resulting in similar plumage patterns occurring in closely related species due to phylogenetic inertia (Felsenstein, 1985; Revell, Harmon & Collar, 2008) rather than caused by habitats selection. In addition, the tendency of species to resemble related species more so than randomly selected species from a tree breaks the assumption of independence of ordinary least square models (Revell, 2010). We therefore examined the association between plumage patterns and habitat by performing comparative analyses that take phylogenetic relatedness into account.

For each biological combination, we computed a phylogenetic logistic regression model (Ives & Garland Jr, 2010) with plumage pattern type (regular vs irregular; regular pattern = 0 and irregular pattern = 1) as the dependent variable and habitat coverage as the explanatory variable (R code: phyloglm(plumage pattern type ∼ habitat coverage, phy = tree, method = “logistic_IG10”)). For each biological combination, the analysis was performed only using species with plumage patterns (regular or irregular). As no phylogenetic signal was detected in the models, we also employed standard generalized linear models (GLMs) for comparison (R code: glm(plumage pattern type ∼ habitat coverage, family = “binomial”)). At large sample sizes, trivially small effects can be detected. Therefore, to examine the statistical significance of the relationship for the GLMs, we evaluated this significance by assessing the goodness-of-fit using the McFadden’s pseudo-R² (McFadden, 1974). All statistical analyses were performed in R (R Development Core Team, 2012); phylogenetic logistic regressions were performed using the phylolm R package.

Phylogenetic information for bird species was obtained from http://BirdTree.org (Jetz et al., 2012). This phylogeny randomly resolves species with missing data by assuming monophyletic genera. To account for phylogenetic uncertainties, we performed our statistical analyses (i.e., phylogenetic logistic regression models) on 100 randomly selected trees and reported the observed variation in the parameter estimates. Branch lengths were computed using the Grafen method (Grafen, 1989; R function: compute.brlen). In phylogenetic logistic regressions, a captures the phylogenetic signal in the data and can vary between −4 (no phylogenetic signal) and 4 (strong phylogenetic signal; Ives & Garland Jr, 2010).

Eco-region avian assemblage analysis

To investigate whether species with regular and irregular plumage patterns were distributed randomly across the world, we examined avian assemblages at the scale of the eco-region (Olson et al., 2001), using a global map of the world’s terrestrial eco-regions made available by The Nature Conservancy (2009) (Fig. S1B). The physical component that defines habitat type are the individual units of land (eco-regions) containing habitat type specific plant communities and species, nested within biomes (Olson et al., 2001). We categorized habitat into two main types: closed and open. Closed habitat was defined as broadleaf and coniferous forests as well as Mediterranean forests and the Taiga; and open habitat comprised the various types of grasslands, deserts and tundra as well as the scarce habitats of Inland water, Rock and ice. Mangrove habitat was not considered in our analysis, as it is not clearly composed of solely closed or open habitat.

A bird species was classed as occurring in a given eco-region if its mapped range overlapped with any part of the eco-region. Although this is a coarse species geographical distribution, it represents a good approximation of occurrence given the spatial resolution of the eco-regions (Hurlbert & Jetz, 2007). Species richness was measured as the number of species occurring in a given eco-region. After removing eco-regions in which no bird species occur, 791 eco-regions remained, of which 293 (37%) were composed of open habitat and 498 (63%) of closed habitat.

To examine whether the composition of plumage patterns in avian assemblages across the world’s eco-regions was random or affected by habitat type, we quantified the ratio of regular to irregular patterns in each eco-region and compared it to a null expectation based on random distribution of the species. The ratio was calculated as the number of species with regular patterns divided by the number of species with irregular patterns (herein referred to as ratio regular-irregular). For the analysis comparing barred and mottled patterns, the ratio was calculated as the number of barred species divided by the number of mottled species (results are presented in Supplemental Information 1). To avoid a skew in the distribution, we log transformed this ratio for all of the analyses. Eco-regions without any species with an irregular plumage pattern (2.6% of the eco-regions on average across the biological combinations) were excluded from this analysis because the ratio could not be calculated.

The null distribution of plumage pattern ratios across the world’s eco-regions was generated by drawing species without replacement 100 times from the hemispherical species pool (i.e., treating Western and Eastern Hemispheres separately). In each eco-region, we drew a number of species corresponding to the observed species richness of the local assemblage, and identified eco-regions with significantly high or low values to the null expectation.

To take spatial autocorrelation into account –whereby species occurring in the local pool of species (i.e., in the neighbouring eco-regions) are more likely to occur in the focal eco-region –we weighted the probability of sampling species based on their degree of occurrence in the local pool. In each avian assemblage (i.e., eco-region), for each species in the hemispherical pool, we determined the proportion of neighbouring eco-regions in which it occurs. Then, for each discrete proportion (i.e., because each eco-region has a discrete number of direct neighbours), we calculated the proportion of species occurring in the focal eco-region. For example, for a given focal eco-region, among all of the species occurring in 50% of the adjoining neighbours, 30% of these species also occur in the focal eco-region. We then fitted a non-linear model (using the nls function in R) to estimate the parameters of this relationship across all the eco-regions (treating the western and eastern hemispheres separately). The fitted curve was then used to determine the probability of a species to occur in a focal eco-region given its occurrence in the neighbouring eco-regions (i.e., the local pool).

We examined the species of the Western hemisphere (the Americas; defined as longitude <−30°W) separately from the Eastern hemisphere (the Old World; defined as longitude >−30°W) as they have distinctive avian species pools. For example, the total ratio regular–irregular (as well as the total ratio barred–mottled) is different between these two hemispheres (see Table S2 and Fig. S3). The global species pool from which species were drawn was therefore separated into two species pools representing each hemisphere.

Following the camouflage hypothesis, we expect the ratio regular–irregular to be higher in open habitats than the expected null values (e.g., regular patterns dominate), and to be smaller than the null expectation in closed habitats (e.g., irregular patterns dominate). To test this hypothesis, we computed a p-value by comparing the observed value to the distribution of null values. If the eco-region was composed of open habitat, we used 1 minus the cumulative probability because we expect the ratio regular-irregular to be higher than the null expectation. To account for multiple testing (i.e., one test for each eco-region) we controlled the false discovery rate using the method developed by Benjamini & Hochberg (1995).

Phylogenetic comparative analysis

Figure 2 shows very little difference in habitat coverage between species with regular versus irregular plumage patterns for all biological combinations, which was confirmed by the statistical models (Table 1). Phylogenetic uncertainty due to randomly resolving species with missing data by assuming monophyletic genera did not meaningfully affect the results (Table 1). For all biological combinations of plumage patterns, a was very nearly −4 indicating that almost no phylogenetic signal could be detected in the data for plumage patterns. Due to the lack of phylogenetic signal, estimates of intercepts and slopes were almost identical between phylogenetic logistic regressions and GLMs (Table 1). Although most slopes were significantly different from the null hypothesis for both phylogenetic logistic regressions and GLMs, the association between plumage pattern type and habitat coverage using GLMs yielded extremely low R² values (0.001–0.039) with a median R² of 0.0065 (Table 1). In addition to the low explanatory power, all the models showed a negative relationship between plumage patterns and habitat coverage, indicating that regular patterns are more associated with closed environments relatively to irregular patterns that are more associated with open environments, contrary to the camouflage hypothesis.

Eco-region avian assemblage analysis

Across the world’s eco-regions, and age, sex and breeding combinations, avian assemblages had similar proportions of species with regular and irregular plumage patterns, regardless of habitat type (Fig. 3; Fig. S2). In addition, all eco-regions contained qualitatively the same ratio of species with and without plumage patterns for each biological combination, regardless of habitat type (Fig. 3; Fig. S3).

The observed ratio regular–irregular in the eco-regions of both open and closed habitats rarely differed from the null expectation (Table 2; similar results were obtained for the ratio barred-mottled; Table S4). Less than 1% of eco-regions with species of birds with patterns had a ratio regular–irregular significantly lower than the null expectation for any biological combination (Table 2), and most of these eco-regions were located on islands.