Journal list menu
Environmental filtering of avian communities along a rural-to-urban gradient in Greater Washington, D.C., USA
Abstract
The composition of avian communities in human-dominated habitats is thought to be determined by the interaction between species-specific traits and environmental characteristics. Traits such as dietary habits and habitat specialization influence the vulnerability of species to land use change. As species are excluded from anthropogenic environments, local species pools are differentially sorted from the regional species pool. This sorting process, environmental filtering, is characterized by a decline in the functional diversity of local biotic communities and may result in a loss of regional biodiversity as landscapes are urbanized. Environmental filtering due to urbanization is hypothesized due to an ecosystem stress gradient, which describes a decrease in species richness or abundance with increasing urban intensity. Conflicting patterns of species richness and species abundances have limited our ability to determine whether urban environments filter avian communities. To evaluate the hypothesis that environmental filtering is occurring, we analyzed avian point count data collected along a rural-to-urban gradient in metropolitan Washington, D.C. We examined predictions that species richness, functional diversity, and the total and relative abundances of some life history guilds exhibit the pattern expected under the ecosystem stress-gradient hypothesis. Species richness and functional diversity declined monotonically with increasing impervious surface. Life history guilds, representing species’ diet, foraging, nest, and migration habits, exhibited differential rates of decline across the rural-to-urban gradient, resulting in marked shifts in the composition of communities. Our results support the hypothesis that urbanization filters bird communities as a function of avian traits and provide further evidence of trait-level responses to urban environments.
Introduction
The impact of urbanization on biological communities has become a subject of key conservation concern in the 21st century. Globally, the proportion of people living in urban environments increased from <30 to over 70% between the years of 1950 and 2014, with the proportion of urban land area projected to increase by 185% by the end of this century (Seto et al. 2012). Under current rates of urban expansion, developed land in the continental United States is projected to increase by 63% from 2001 to 2051 (Lawler et al. 2014). Urbanization has detrimental impacts on populations and ecosystems (Czech and Krausman 1997). At current rates of expansion, urbanization is expected to become one of the most important drivers of species extinction over this century (Marzluff 2001) and cause a loss of critical habitat for one-third of the bird species currently at risk of extinction (Lawler et al. 2014). Mitigating projected losses in global biodiversity is in part dependent on understanding how urbanization structures biological communities and the development of wildlife management strategies that incorporate urban ecosystems (Miller and Hobbs 2002).
As landscapes become urbanized, avian communities are impacted by modification to the structure and composition of habitats within the suburban and urban matrices (Beissinger and Osborne 1982, Marzluff et al. 1998). The process of urbanization generates characteristic changes to bird habitats (McKinney 2002) that often include an increase in non-native plant species cover (White et al. 2005), decline in the structural diversity of vegetation (Gavareski 1976, Evans et al. 2009), and a decrease in habitat availability for native species (Marzluff 2001). Because species are known to have a differential response to human-built environments, communities that occupy urban environments are often distinct from those that inhabit areas of lower human impact and tend to be dominated by higher densities of a few species able to persist in these habitats (Chace and Walsh 2006, McKinney 2006, Kark et al. 2007, Conole and Kirkpatrick 2011).
Patterns in avian community structure along the rural-to-urban gradient are thought to be determined largely by an interaction between land use and life history traits of associated species that determine whether a given species will be a winner (urban-adapted or urban-exploiter species) or loser in human-dominated landscapes (urban-avoiding species; McKinney and Lockwood 1999, Blair 2001). Species with specialist niche requirements are expected to be especially sensitive to human-induced habitat modification and may therefore experience high rates of local extinction across the urban habitat matrix (Devictor et al. 2007). For example, obligate insectivores (Lancaster and Rees 1979) and cavity-nesting species (Blewett and Marzluff 2005) may be considerably impacted by even minor modifications of the urban landscape, while species with omnivorous or generalist nesting habits are expected to be positively affected by, or even thrive, in urban environments (reviewed in McKinney and Lockwood 1999).
Urban bird communities are therefore hypothesized to exhibit biotic homogenization by which urban communities across biogeographic regions will be more taxonomically and functionally similar than their rural counterparts due to differential resource opportunities across life history traits (McKinney and Lockwood 1999, Devictor et al. 2007, Luck and Smallbone 2011). The effects of urbanization on local and regional habitat structure and function are therefore expected to act as an environmental filter on avian communities by excluding species with specialist traits that are less adapted to urban habitats while providing additional habitat for species with traits, such as omnivorous diets, that may facilitate their persistence in these environments (Croci et al. 2008, Jokimaki et al. 2014, Aronson et al. 2016). As the strength of environmental filtering increases across the urbanization gradient, both species richness and functional diversity—the variation in life history traits associated with local communities—are expected to decline with increasing urbanization (Ortega-Álvarez and MacGregor-Fors 2009, Sol et al. 2014). While such declines are consistent with environmental filtering, species richness might vary along spatial gradients for other reasons (e.g., resource availability). Because any trend in richness could generate a parallel trend in functional diversity through a simple sampling effect, environmental filtering can only be inferred when using a metric of functional diversity that takes species richness into account (Petchey et al. 2007, Flynn et al. 2009).
A decline in species richness, and often an increase in abundance, across the rural-to-urban gradient has been a widely observed phenomenon across biogeographic regions (Jokimaki and Suhonen 1993, Clergeau et al. 1998, McKinney 2006). Whether urbanization acts as an environmental filter on avian communities, however, remains largely unclear. One line of evidence for environmental filtering is provided by the ecosystem stress-gradient hypothesis (Rapport et al. 1985), which describes a decrease in species richness or abundance with increased stress. A study by Lepczyk et al. (2008) found support for the stress-gradient hypothesis, and thus environmental filtering, along a rural-to-urban gradient by looking at avian species richness and abundance of species along a gradient of anthropogenic land cover. A competing hypothesis, the intermediate disturbance hypothesis (Connell 1978), can be applied to the rural-to-urban gradient. Under this hypothesis, peak species richness, and often peak species abundance, can be observed at intermediate levels of urbanization. It is thought that this pattern, which has been observed in several studies (Blair 1996, Lepczyk et al. 2008), arises from conflicting pressures from competition at the rural end of the gradient and physical factors at the urban end (Chace and Walsh 2006).
Although numerous studies have addressed avian community composition at varying intensities of urbanization (Lancaster and Rees 1979, Blair 1996, Crooks et al. 2004), comparably fewer studies have addressed the response of guilds to urbanization across the continuous rural-to-urban gradient and the urban end of the gradient tends to be relatively undersampled (but see Crooks et al. 2004, Blair and Johnson 2008, Garaffa et al. 2009, Vignoli et al. 2013). Moreover, the associations between life history guilds and habitat characteristics may vary widely among biogeographic regions (Hansen and Urban 1992); thus, guild-specific measures of environmental filtering in response to urbanization have often produced conflicting results (Evans et al. 2011). Attributing the response of individual guilds to environmental conditions is also confounded by joint membership of species across guilds. Because of this, the response of a guild to urbanization necessitates accounting for the correlation structure between guilds across multiple niche axes (Croci et al. 2008). Finally, an association between guild abundance and urban land cover provides only incomplete evidence for environmental filtering—for example, while Lepczyk et al. (2008) found that the majority of species with a significant association with urban intensity fit the stress-gradient hypothesis, life history traits did not predict positive or negative correlations with anthropogenic land cover overall.
While a decline in species richness and reduced abundances of some life history guilds with increasing impervious surface provide clues that environmental filtering is occurring, neither measure provides direct evidence for the filtering effect of urbanization. Functional diversity provides a metric of environmental filtering (e.g., among plant communities, see Coyle et al. 2014) that is insensitive to the response of individual guilds to urbanization and can therefore be broadly applied across regions despite conflicting responses among life history guilds. While previous research has suggested that urban bird communities are filtered as a function of avian life history traits (notably Croci et al. 2008), no previous study to our knowledge has yet examined environmental filtering by assessing the functional diversity of bird communities on local gradients of urbanization.
Here, we evaluate the composition of bird communities at sampling locations spanning the rural-to-urban gradient in the Greater Washington, D.C., area. We hypothesize that urban habitats act as an environmental filter on the avian communities within our study region. We use three distinct lines of evidence to test this hypothesis and explore the nature of environmental filtering within our study region, testing the following predictions: (1) Species richness declines along the rural-to-urban gradient (Lepczyk et al. 2008); (2) functional diversity, as expressed by the diversity of life history guilds, declines along the rural-to-urban gradient, controlling for species richness (Schütz and Schulze 2015); and (3) species with life history traits associated with specialist dietary habits and foraging behaviors (e.g., insectivores and foliage-gleaners, Beissinger and Osborne 1982), nesting habits (e.g., cavity- and ground-nesting species; Blewett and Marzluff 2005, Vignoli et al. 2013), and long-distance migration are more sensitive to urbanization than other species within the regional species pool (e.g., Kark et al. 2007, Fig. 1 for guild-level predictions).
Methods
We examine the response of individual guilds to suburban and urban development by assessing variation in the estimated abundance (i.e., abundance accounting for detection) and relative abundance. These metrics, when used in tandem, provide a measure of both how a guild responds relative to other guilds for a given trait (relative abundance) and whether urban habitats positively or negatively influence avian abundances for that guild. Furthermore, because the attribution of a guild response is likely confounded by co-membership between guilds, we assess the degree of correlation between guilds. The use of three distinct lines of evidence and addressing the abundance, relative abundance, and collinearity between guilds provide a robust and novel analysis of both the extent and nature of environmental filtering in urban habitats.
Avian count data were collected as a part of the Neighborhood Nestwatch Program (NN), a project run by the Smithsonian Migratory Bird Center. NN has established a network of sampling sites within the Greater Washington, D.C., metropolitan area with sampling predominantly located at the homes of project participants. For the purpose of this study, a site is defined as a 100 m radius area surrounding location of the point count. Project participants are accepted into the study based on a wide range of criteria, including level of interest, expected degree of engagement, and position along the rural-to-urban gradient, as assessed by the proportion of impervious surface relative to the impervious surface within our study region and that of existing sites. Habitats represented by study sites range from rural open and forested areas to suburban and urban environments. In order to more accurately characterize the rural-to-urban gradient within our study region, supplemental survey data were collected from an additional 30 randomly sampled sites (total sites = 203) from forested and core urban (>50% impervious surface) habitats, habitat classes which were under-represented within the NN study (Evans et al. 2015; Appendix S1). Random sampling points were jittered to the nearest accessible location.
Between the years of 2009 and 2012, Smithsonian technicians visited sites once annually during avian breeding season (May–August) and conducted a ten-minute, fixed radius point count (50 m; Petit et al. 1995) between the hours of 07:00 and 10:00 local time. The distances between the observer and individual birds were estimated in 10 m distance segments (e.g., 0–10 m). To ensure that counts adequately reflect organisms utilizing the habitat of the point count area, flyovers were not recorded and count data were restricted to species with known breeding ranges within our study region. Because the number of years a given site was visited was variable (between one and three years), one point count was sampled randomly from a given site for analysis.
We approximated the degree of urbanization of a given site (i.e., point count location) as the proportion of impervious surface (30 m resolution; Xian et al. 2011) within the 100 m radius site area (this variable is referred to hereafter as simply impervious surface). These data were evaluated using the raster package in R (Hijmans 2015). Impervious surface has been found to be predictive of avian demographic response to urbanization (Ryder et al. 2010, Evans et al. 2015) and adequately reflects the variation in habitat distribution and quality across the rural-to-urban gradient (reviewed in McKinney 2002). Though the scale at which birds respond to the urban environment undoubtedly varies by species, a distance of 100 m was chosen because we expect that this distance is roughly representative of the overlap between the typical breeding territory size of the observed individuals in this study and the point count radius. Additionally, we examined community composition, functional diversity, and species richness at multiple radii, from 30 to 1000 m, and this scale of analysis was shown to have the greatest predictive capacity (Morelli et al. 2013, Jackson and Fahrig 2015).
To test our prediction that species richness will decline along the rural-to-urban gradient, we estimated site-level species richness using Chao's richness estimator (1984) and evaluated richness patterns across the rural-to-urban gradient in a generalized linear model framework in R. Chao's richness estimator accounts for imperfect detection by using singletons and doubletons to predict the number of undetected species in a sample. In doing so, Chao's estimator is expected to provide a robust method of comparison between sites with potential differences in detectability and has been shown to exhibit low bias and high stability relative to other richness and diversity metrics (Walther and Martin 2001). Observer, Julian day, and year, as well as an interaction term between Julian day and year, were included as predictor variables for richness. As these variables are considered nuisance terms within this context, the effects of these variables were not directly explored. AICc was used to compare models that included a null of only detection covariates and models that included linear and quadratic impervious surface terms (Burnham and Anderson 1998). These terms have been used throughout as an indication of whether variation in species richness or abundance patterns (below) provides supportive evidence for environmental filtering as expected under the stress-gradient hypothesis (Lepczyk et al. 2008). To ensure that the use of Chao's richness estimator did not bias our results, we compared raw and estimated species richness (r = 0.67)—while estimated richness was greater than observed richness, the overall trend in richness about the rural-to-urban gradient was not shown to be influenced by the use of Chao's estimator.
For each species observed, we evaluated guild representation across dietary, foraging, migratory, and nesting life history traits, obtained from the avian trait database compiled by Wilman et al. (2014). When necessary, we supplemented trait data with avian life history information from the Birds of North America (Rodewald 2015). See Appendix S1: Tables S1, S2 for species observed during point counts and associated life history traits. To estimate total abundance for a given guild and trait, we summed counts for a guild across species and adjusted count data for detectability using the R package unmarked (Fiske and Chandler 2011). We used the generalized distance sampling model of Chandler et al. (2011) to fit yearly and site covariates to adjust abundance estimates by detectability. As birds in open habitats are more readily observed, we calculated the proportional canopy cover within 50 m of the point count location and used this value as a continuous covariate for detection (Homer et al. 2015). Likewise, because bird activity and habitat associations vary considerably at different stages of the breeding cycle, we included the linear and quadratic terms of Julian day of the point count as a covariate for detectability (McClure and Hill 2012). Because the probability of detection is also dependent on the skill level of the observer, the technician who conducted the count was included as a detection covariate (Sauer et al. 1994). We calculated abundance for each guild and across guilds and used Akaike's information criteria, AICc, adjusted by the overdispersion parameter c-hat, for model comparison.
We tested whether there is evidence for environmental filtering within our study region using the community classification tree method of Petchey and Gaston (2002). We constructed a trait matrix using life history traits expected to be predictive of avian response to urbanization (Appendix S1: Tables S1, S2). Categorical traits (i.e., life history guilds) were reclassified as binary variables and weighted by the number of variables required to describe the trait (Villéger et al. 2008, Laliberte and Legendre 2010). Species by trait distance matrices were calculated in the R package FD (Laliberte et al. 2014) using Gower distances between species, a distance calculation that allows for the combined use of continuous and categorical variables (i.e., guild membership coded as asymmetric binary variables; Rodewald and Bakermans 2006). We constructed a community classification tree for our regional species pool using hierarchical clustering in the vegan R package (Unweighted Pair Group Method with Arithmetic Mean; Legendre and Legendre 1997, Oksanen et al. 2018) and calculated the functional diversity (FD) for each community as the sum of branch lengths across species observed at a given site (Petchey and Gaston 2002, Fig. 1A). Because FD is positively correlated with species richness (Mouchet et al. 2010), we compared observed FD with a null distribution in which tip labels were randomly assigned (n = 104). This method maintains species richness while randomizing the species assigned to a given tip (Petchey et al. 2007). We then calculated the standardized effect size (SES) for each site as the difference in summed branch lengths between observed and sampled communities divided by the standard deviation of sampled communities. Negative and positive SES values represent communities that are less or more functionally diverse than expected given their species richness. We evaluated the mean and 95% confidence intervals of SES values across sites, with values of less than zero providing supportive evidence for the influence of environmental filtering (Mason et al. 2013). To test our prediction that the urbanization filters species as a function of their life history traits, we evaluated SES values across the impervious surface gradient, with the expectation that values will decline with increasing urbanization (Flynn et al. 2009). We used AICc, as above, to compare generalized linear models of SES.
We explored the influence of urbanization on avian guild associations to determine which traits best explain avian community composition across the rural-to-urban gradient. We calculated the relative abundances of guilds for a given life history trait (e.g., nesting guilds) at each site. We used binomial regression (logit-link) to model variation in the relative abundances of each guild along the rural-to-urban gradient. AICc was used for model comparison. Because inference is limited by joint membership between guilds and among guilds and niche breadth axes, we explored the correlation structure across both sets of variables.
Results
A total of 44 native bird species and 1,474 individual birds were observed across all sites and years. Site-level Chao species richness estimates ranged from 5 to 61 species and averaged 18.8 species per site (±0.051). Richness estimates peaked at sites with low impervious surface and were their lowest at the most urban sites (Fig. 1B, Table 1). There was considerable model support for a monotonic decline in functional diversity with increasing impervious surface (Table 1, Fig. 1C).
| Model | ΔAICc | w | βIMP (95% CI) | (95% CI) | Pseudo-R2 |
|---|---|---|---|---|---|
| Species richness | |||||
| IMP + IMP2 | 0.00 | 1.00 | 4.6 E−3 (−1.6 E−3, 0.011) | −2.7 E−4 (−3.8E−4, −1.6E−4) | 0.31 |
| IMP | 24.5 | 0.00 | −0.01 (−0.013, −7.8 E−3) | –* | 0.28 |
| Null | 241 | 0.00 | – | – | 0.05 |
| Functional diversity | |||||
| IMP | 0.00 | 0.71 | −0.029 (−0.038, −0.019) | – | 0.15 |
| IMP + IMP2 | 1.77 | 0.29 | −0.022 (−0.047, 3.1E−3) | −1.1E−4 (−5.1E−4, 2.8E−4) | 0.15 |
| Null | 30.3 | 0.00 | – | – | 0.00 |
- * Impervious surface was not included as a covariate within the model.
Across guilds, total abundance exhibited declined with increasing impervious surface (β = −0.0074, ΔAICc relative to the null = 34.7). There was no evidence for an increase in abundance across the rural-to-urban gradient for any of the life history guilds (Table 2, Fig. 2). Insectivorous, foliage and aerial foragers, cavity-nesting species, and residents each exhibited steep declines in estimated abundance with increasing impervious surface. The estimated total abundances of shrub- and tree-nesting species, as well as short-distance and Neotropical migrants, exhibited moderate declines across the rural-to-urban gradient, but there was no evidence for variation in the estimated abundance of bark-foraging species with increasing impervious surface.
| Model, Abundance | ΔAICc | w | βIMP (95% CI) | (95% CI) |
|---|---|---|---|---|
| Diet, Omnivore | ||||
| IMP + IMP2 | 0.00 | 1.00 | 0.35 (0.20, 0.50) | −0.46 (−0.61, −0.31) |
| IMP | 35.1 | 0.00 | −0.07 (−0.13, −0.016) | – |
| Null | 39.3 | 0.00 | –* | – |
| Diet, Granivore | ||||
| IMP | 0.00 | 0.40 | 0.10 (−0.03, 0.22) | – |
| Null | 0.07 | 0.39 | – | – |
| IMP + IMP2 | 1.29 | 0.21 | 0.25 (−0.11, 0.61) | −0.15 (−0.48, 0.18) |
| Diet, Insectivore | ||||
| IMP + IMP2 | 0.00 | 0.59 | −0.46 (−0.74, −0.18) | −0.35 (−0.79, 0.081) |
| IMP | 0.76 | 0.41 | −0.67 (−0.81, −0.54) | – |
| Null | 112 | 0.00 | – | – |
| Nest, Cavity | ||||
| IMP + IMP2 | 0.00 | 0.98 | −0.22 (−0.52, 0.072) | −0.66 (−1.1, −0.19) |
| IMP | 7.42 | 0.02 | −0.62 (−0.75, −0.49) | – |
| Null | 107 | 0.00 | – | – |
| Nest, Shrub | ||||
| IMP + IMP2 | 0.00 | 1.00 | 0.35 (0.15, 0.55) | −0.45 (−0.65, −0.25) |
| IMP | 18.7 | 0.00 | – | – |
| NULL | 19.4 | 0.00 | −0.06 (−0.14, 0.011) | – |
| Nest, Tree (cup) | ||||
| IMP + IMP2 | 0.00 | 0.80 | 0.21 (0.034, 0.41) | −0.25 (−0.44, −0.07) |
| Null | 3.49 | 0.14 | – | – |
| IMP | 5.29 | 0.06 | −0.20 (−0.093, 0.053) | – |
| Nest, Ground | ||||
| IMP | 0.00 | 0.76 | −3.1 (−5.4, −0.79) | – |
| IMP + IMP2 | 2.43 | 0.23 | −3.0 (−7.9, 1.9) | 0.52 (−16, 17) |
| Null | 8.51 | 0.01 | – | – |
| Foraging, Aerial | ||||
| IMP | 0.00 | 0.65 | −3.6 (−4.9, −2.3) | – |
| IMP + IMP2 | 1.20 | 0.35 | −4.3 (−6.0, −2.6) | 2.6 (−0.43, 5.7) |
| Null | 62.3 | 0.00 | – | – |
| Foraging, Ground | ||||
| IMP + IMP2 | 0.00 | 1.00 | 0.25 (0.12, 0.38) | −0.34 (−0.48, −0.21) |
| IMP | 24.7 | 0.00 | −0.07 (−0.12, −0.018) | – |
| Null | 1.07 E3 | 0.00 | – | – |
| Foraging, Foliage | ||||
| IMP + IMP2 | 0.00 | 0.53 | −0.42 (−0.80, −0.051) | −0.42 (−0.99, 0.15) |
| IMP | 0.25 | 0.47 | −0.68 (−0.85, −0.51) | – |
| Null | 71.0 | 0.00 | – | – |
| Foraging, Bark | ||||
| IMP + IMP2 | 0.00 | 0.62 | 0.74 (0.05, 1.4) | −1.11 (−2.2, −0.019) |
| Null | 1.58 | 0.28 | – | – |
| IMP | 3.61 | 0.10 | 0.042 (−0.26, 0.34) | – |
| Migration, Neotropical | ||||
| IMP + IMP2 | 0.00 | 0.46 | 0.14 (−0.13, 0.42) | −0.26 (−0.54, 0.031) |
| IMP | 1.07 | 0.27 | −0.083 (−0.19, 0.029) | – |
| Null | 1.10 | 0.27 | – | – |
| Migration, Short-distance | ||||
| IMP + IMP2 | 0.00 | 1.00 | 0.27 (0.092, 0.44) | −0.34 (−0.52, −0.17) |
| IMP | 13.9 | 0.00 | −0.060 (−0.13, 0.0060) | – |
| Null | 15.1 | 0.00 | – | – |
| Migration, Resident | ||||
| IMP + IMP2 | 0.00 | 0.81 | −0.13(−0.34, 0.075) | −0.28 (−0.053, −0.027) |
| IMP | 2.91 | 0.19 | −0.35 (−0.44, −0.25) | – |
| Null | 57.6 | 0.00 | – | – |
- * Impervious surface was not included as a covariate within the model.
The proportional composition of life history guilds varied considerably across the rural-to-urban gradient (Table 3, Fig. 3). Omnivores, which made up the largest proportion of diet guilds throughout our study region, increased in relative abundance with increasing impervious surface at the rural end of the urbanization gradient but leveled off as urban intensity increased. While granivores made up <10% of the avian community at rural sites, the relative abundance of this guild increased sharply with increasing impervious surface and was similar to that of omnivore abundance at the most urban sites. Conversely, despite making up nearly 30% of the avian community at rural sites, insectivores declined to <5% of the community at the urban end of the impervious surface gradient. Among nesting guilds, shrub- and tree-nesting species increased in relative abundance while the relative abundance of cavity-nesting species exhibited a steep decline with increasing impervious surface, despite each guild making up an equivalent portion of the bird community at rural sites. Among foraging guilds, only ground-foraging species increased in relative abundance with increasing impervious surface, with substantial model support of a monotonic decline in each of the remaining foraging guilds. Resident and short-distance migrant species made up equivalent portions of bird communities as the rural end of the urbanization gradient, but exhibited an inverse relationship with impervious surface. Resident species declined in relative abundance with increasing urbanization while short-distance migrants increased. There was no evidence of a relationship between Neotropical migrant relative abundance and impervious surface, with the null model receiving the greatest model support for this guild.
| Model, Relative abundance | ΔQAICc | w | βIMP (95% CI) | (95% CI) | Pseudo-R2 |
|---|---|---|---|---|---|
| Diet, Omnivore | |||||
| IMP + IMP2 | 0.00 | 0.76 | 0.28 (0.10, 0.45) | −0.21 (−0.41, −0.034) | 0.18 |
| IMP | 2.34 | 0.24 | 0.11 (0.038, 0.17) | – | 0.12 |
| Null | 9.58 | 0.01 | –* | – | – |
| Diet, Granivore | |||||
| IMP | 0.00 | 0.73 | 0.36 (0.24, 0.48) | – | 0.10 |
| IMP + IMP2 | 2.02 | 0.27 | 0.31 (−0.046,0.0.67) | 0.053 (−0.31, 0.40) | 0.10 |
| Null | 22.4 | 0.00 | – | – | – |
| Diet, Insectivore | |||||
| IMP | 0.00 | 0.67 | −0.63 (−0.77, −0.50) | – | 0.33 |
| IMP + IMP2 | 1.41 | 0.33 | −0.52 (−0.82, −0.21) | −0.19 (−0.72, 0.26) | 0.33 |
| Null | 101 | 0.00 | – | – | – |
| Nest, Cavity | |||||
| IMP + IMP2 | 0.00 | 0.78 | −0.24 (−0.55, 0.078) | −0.53 (−1.1, −0.041) | 0.29 |
| IMP | 2.51 | 0.22 | −0.54 (−0.68, −0.41) | – | 0.27 |
| Null | 79.8 | 0.00 | – | – | – |
| Nest, Shrub | |||||
| IMP + IMP2 | 0.00 | 0.58 | 0.29 (0.079, 0.50) | −0.20 (−0.44, 0.035) | 0.06 |
| IMP | 0.71 | 0.41 | 0.12 (0.043, 0.20) | – | 0.05 |
| Null | 7.61 | 0.01 | – | – | – |
| Nest, Tree (cup) | |||||
| IMP | 0.00 | 0.73 | 0.13 (0.054, 0.21) | – | 0.05 |
| IMP + IMP2 | 2.05 | 0.26 | 0.12 (−0.086, 0.33) | 0.014 (−0.21, 0.24) | 0.05 |
| Null | 8.86 | 0.01 | – | – | – |
| Nest, Ground | |||||
| IMP | 0.00 | 0.70 | −2.7 (−5.6, −0.94) | – | 0.17 |
| Null | 6.85 | 0.02 | – | – | – |
| Foraging, Aerial | |||||
| IMP | 0.00 | 0.66 | −3.4 (−7.2, −5.1) | – | 0.33 |
| Null | 51.6 | 0.00 | – | – | – |
| Foraging, Ground | |||||
| IMP | 0.00 | 0.62 | 0.10 (0.041, 0.17) | – | 0.18 |
| IMP + IMP2 | 1.07 | 0.37 | 0.18 (0.017, 0.35) | −0.093 (−0.28, 0.090) | 0.20 |
| Null | 8.26 | 0.01 | – | – | – |
| Foraging, Foliage | |||||
| IMP | 0.00 | 0.66 | −0.56 (−0.74, −0.40) | – | 0.14 |
| IMP + IMP2 | 1.35 | 0.34 | −0.39 (−0.76, 0.010) | −0.31 (−1.0, 0.26) | 0.14 |
| Null | 34.6 | 0.00 | – | – | – |
| Foraging, Bark | |||||
| IMP + IMP2 | 0.00 | 0.62 | 0.25 (−0.41, 1.0) | −0.99 (−2.4, 0.033) | 0.05 |
| IMP | 1.21 | 0.34 | −0.33 (−0.61, −0.084) | – | 0.03 |
| Null | 5.77 | 0.03 | – | – | – |
| Migration Neotropical | |||||
| Null | 0.00 | 0.67 | – | – | – |
| IMP | 2.06 | 0.24 | 1.8 E−3 (−0.11, 0.11) | – | 0.00 |
| IMP + IMP2 | 4.06 | 0.09 | −0.038 (−0.31, 0.24) | 0.049 (−0.27, 0.36) | 0.00 |
| Migration, Short-distance | |||||
| IMP | 0.00 | 0.56 | 0.17 (0.10, 0.24) | – | 0.11 |
| IMP + IMP2 | 0.50 | 0.44 | 0.28 (0.090, 0.48) | −0.13 (−0.35, 0.076) | 0.12 |
| Null | 18.1 | 0.00 | – | – | – |
| Migration, Resident | |||||
| IMP | 0.00 | 0.73 | −0.27 (−0.37, −0.18) | – | 0.15 |
| IMP + IMP2 | 2.00 | 0.27 | −0.25 (−0.47, −0.018) | −0.037 (−0.33, 0.24) | 0.15 |
| Null | 30.6 | 0.00 | – | – | – |
- * Impervious surface was not included as a covariate within the model.
Discussion
Consistent with previous research, we observed a decline in species richness and a predictable shift in the structure of avian communities with increasing urbanization (Blair 1996, Clergeau et al. 1998, Croci et al. 2008). The data support the hypothesis that species richness decreases with increasing impervious surface. The observed pattern, a monotonic decline in species richness across the rural-to-urban gradient, suggests that the intermediate disturbance hypothesis (Connell 1978, Marzluff 2001) does not appropriately describe our study system and provides supportive evidence for environmental filtering along an ecosystem stress gradient (Rapport et al. 1985; see Lepczyk et al. 2008). Likewise, the observed decline in functional diversity supports our prediction that the influence of environmental filtering on avian community composition increases with increased urbanization.
Patterns of relative and total abundance across the rural-to-urban gradient showed a decline in the abundance almost every life history guild across the rural-to-urban gradient. These results provide supportive evidence for environmental filtering via an ecosystem stress gradient but often ran counter to our expectations. For example, despite our expectations that omnivores and ground-foraging species are positively associated with urbanization, the estimated total abundances of these guilds declined with increasing impervious surface (Vignoli et al. 2013). While we expected granivores to be positively affiliated with urban environments, the null model that excluded impervious surface received equivalent model support to the model which showed an increase in the estimated abundance with increasing impervious surface for this guild. Likewise, while we predicted that Neotropical migrants would be negatively influenced by urbanization, there was no evidence of a relationship between the relative abundance and impervious surface for this life history guild.
This study is among a small set of studies in its application of functional diversity to assess environmental filtering across the rural-to-urban gradient (Croci et al. 2008). As functional diversity was found to decline with increasing impervious surface even after accounting for a decrease in species richness, our results provide strong supportive evidence that urban habitats filter avian communities as a function of their traits (Petchey et al. 2007). Emerging evidence for evolutionary homogenization of urban bird communities, by which urbanization reduces the evolutionary distinctness of bird communities, suggests that the loss of functional diversity may have evolutionary consequences that lead to unstable phylogenetically similar urban communities (Morelli et al. 2016).
While the seminal work of Croci et al. (2008) found evidence for regional variation in avian traits in response to urban environments, their study did not detect a local-scale filtering effect. A key difference between our studies, however, is that Croci et al. sampled patches of forest within the urban and suburban matrix whereas we sampled from forested patches and from within the matrix itself. Additionally, we evaluated land cover at a finer spatial resolution (100 m) relative to Croci et al. (600 m)—our study is therefore more likely to observe effects at the scale of a bird's territory while Croci et al. is more likely to detect effects of landscape context. This may suggest that bird communities along the rural-to-urban gradient may be structured by land cover distributions at the territory scale rather than the influence of surrounding land cover (Evans et al. 2009). Conversely, differences in urban form, such as maintained green space in urban areas (Alberti 2005), likely generate different diversity patterns—therefore, replicating this study across multiple geographic regions is necessary to determine whether our results are unique to our focal city. This was illustrated by Ferenc et al. (2014), who found a strong latitudinal trend in beta diversity in which urban cities in Northern latitudes of Europe experienced higher levels of biotic homogenization. Finally, a differential avian community response to urbanization may be driven by the length of time that North American and European areas have been settled. Species in these regions may exhibit a delayed response to environmental modification in younger cities or may rebound as plant communities in older cities mature (Aronson et al. 2014, Pidgeon et al. 2014).
Species richness
We observed a decline in species richness across the rural-to-urban gradient, a relationship that has been observed across numerous biogeographic regions (reviewed in Marzluff 2005, Chamberlain et al. 2017). Several studies have documented a peak in species richness at intermediate portions of the rural-to-urban gradient, a pattern expected under the intermediate disturbance hypothesis (e.g., Blair 2001, Blair and Johnson 2008, reviewed in Marzluff 2017). This pattern is expected to be indicative of increased richness along ecotones, as avian assemblages are comprised of both urban sensitive and insensitive species at this portion of the gradient (Crooks et al. 2004). While the model of a quadratic relationship between richness and relative abundance was best supported by the data, we found no evidence that species richness was enhanced in suburban environments (Fig. 1B). These results therefore support the ecosystem stress-gradient hypothesis.
The observed pattern in species richness may have been influenced by the removal of non-native species from our analysis. Urban environments have been found to support high densities of non-native species that are able to exploit urban and suburban habitats, such as the House Sparrow (Passer domesticus), due to life history characteristics (e.g., omnivorous dietary niche) that are expected to allow them to colonize this portion of the gradient (Lancaster and Rees 1979, Sol et al. 2012). Despite our removal of non-native species, however, there was no observable decline in species richness within exurban (5–20% impervious surface) or suburban portions of the urban gradient (30–50% impervious surface; Marzluff 2001) and our findings do support considerable variation in guild structure at these levels of impervious surface. Likewise, addition of non-native species to our data (results not shown) did not change the overall observed trend in species richness, though it did slightly moderate the rate of decline in species richness with increasing impervious surface.
Dietary guilds
Our results provide considerable evidence for species sorting based on avian dietary guilds in Greater Washington, D.C., indicating that food resources may be an important determinant of the persistence of species in urban environments. We observed a decline in the relative abundance of insectivorous birds and an increase in the relative abundances of omnivorous and granivorous dietary guilds across the rural-to-urban gradient. In their review of the effects of urbanization on avian communities, Chace and Walsh (2006) found that urban environments select for omnivorous and granivorous dietary guilds. As the estimated abundance of each of the dietary guilds was negatively associated with impervious surface, the observed variation in relative abundances was largely driven by a steep decline in the estimated abundance of insectivorous birds. Indeed, a negative response of insectivores to urbanization has been a geographically widespread phenomenon. For example, Lim and Sodhi (2004) found a decline in insectivores in metropolitan Singapore and Sengupta et al. (2014) observed a similar loss of insectivores in urban environments in India despite high proportions of this guild within exurban landscape (but see Raupp et al. 2010). The substantial decline in the estimated abundance of insectivores suggests that while urban environments may support higher relative abundances of omnivore and granivores, the dominance of these dietary guilds in previous studies is likely a result of the loss of insectivorous birds at urbanized sites rather than a benefit conferred to these guilds by urban habitats.
Foraging guilds
Across foraging strategies, only the relative abundance of ground-foraging birds, such as the American Robin (Turdus migratorius) and Mourning Dove (Zenaida macroura), increased with impervious surface and the relative abundances of foliage and aerial feeding species exhibited a steep decline with increasing impervious surface. The observed increase in the relative abundance of ground foragers has been supported across a number of studies (Emlen 1974, Lancaster and Rees 1979, Johnston 2001), though we found no evidence that the abundance of ground-foraging species varied across the rural-to-urban gradient. Likewise, while there was some evidence for a negative association between the relative abundance of bark-foraging species and urbanization, there was no evidence for variation in the absolute abundance of this guild with impervious surface.
Nesting guilds
We observed a decline in the relative abundance of ground- and cavity-nesting birds and increase in the relative abundance of tree- and shrub-nesting guilds with increasing impervious surface. Additionally, there was considerable support for a decline in the absolute abundances of cavity-nesting and ground-nesting birds, though no evidence for a relationship between tree- or shrub-nesting birds and impervious surface. The decline in relative and absolute abundances of cavity- and ground-nesting birds suggests that these life history guilds may be filtered along the rural-to-urban gradient. Indeed, Blewett and Marzluff (2005) found that snags and standing dead trees in which cavity-nesting species typically build nests were absent from the human-built environment, leading to an overall reduction in the abundance of this guild in suburban landscapes of Washington State, USA. The negative association between ground-nesting species has been shown across study regions (reviewed in Marzluff 2001) and is expected to be a result of sensitivity to enhanced rates of predation for this nesting guild (Jokimaki and Huhta 2000). The observed decline in cavity-nesting species with increasing urbanization, however, is likely context dependent, as many studies have observed no response or even increased abundances of these species in urban environments (Jokimaki and Huhta 2000, Croci et al. 2008, Jokimaki et al. 2014).
Migratory status
The relative abundance of resident species of birds declined markedly with increasing impervious surface while that of short-distance migrants increased and Neotropical migrants showed no response to urbanization. The response of migratory guilds to impervious surface was contrary to our prediction that migrant bird species would decline with increasing impervious surface. Numerous studies have reported a negative association between the presence of Neotropical migrant birds and urbanization (Stratford and Robinson 2005), and several studies (McKinney and Lockwood 1999, Kark et al. 2007, Rodewald and Gehrt 2014) have suggested that resident life histories are one of the key features that defines urban-adapted species (but see Evans et al. 2011). For example, Rodewald and Bakermans (2006) found that resident species increased in abundance with increasing impervious surface surrounding riparian forests while the abundance of Neotropical migrant birds were negatively associated. Our ability to detect the influence of impervious surface may have been masked by high abundances of Gray Catbird (Dumetella carolinensis), an omnivorous species that made up more than 50% of the abundance of Neotropical migrants and show no observable response to impervious surface. Likewise, while short-distance migrants were shown to increase in relative abundance with increasing urbanization, almost all of the individuals that comprise this migratory guild were American Robin, American Crow (Corvus brachyrhynchos), Fish Crow (Corvus ossifragus), and Common Grackle (Quiscalus quiscula) species that also showed no response to, or even an affinity for, urban habitats.
Across guilds, a clear limitation to our analysis was the extent of co-membership represented by life history guilds across traits (Appendix S1: Fig. S1). Our ability to detect the response of migratory species to impervious surface may have been hindered by co-membership across guilds. Indeed, while residents were shown to respond strongly to the urbanization gradient in regard to both absolute and relative abundance, nearly half of the observed resident species were insectivorous and more than half were cavity nesters—both of these guilds were shown to decline across the rural-to-urban gradient. Likewise, all five of the aerial foragers and seven of the eleven foliage foragers are insectivorous. While this does not negate the potential impact of foraging strategy on avian community structure, it is impossible to distinguish the response of the foraging strategy for these guilds separately from the apparent influence of dietary niche as evaluated. While methodologies exist to determine the influence of individual traits on community structure, the paucity of individuals observed at a given site, undoubtedly due to low detection probabilities for a given sample, limited our ability to employ these methods (Legendre et al. 1997, Shipley et al. 2006, Brown et al. 2014, Warton et al. 2015).
The use of the proportion of impervious surface to approximate the urbanization gradient in our study has both strengths and weaknesses. Urbanization is both a spatial and temporal process (Fernández-Juricic 2004), and thus, observed patterns of abundance and species richness may mask delayed local extinctions in modified landscapes (extinction debt, reviewed in Kuussaari et al. 2009). While impervious surface provides a gradient of development intensity (Marzluff 2001), urbanization occurs at multiple spatiotemporal scales through which human socioeconomic and environmental systems are linked (Grimm et al. 2008). Landscape context, for example, has been found to influence avian community assembly—thus, analysis of bird communities at one scale may miss key features driving avian response at others (Hostetler and Holling 2000, Melles et al. 2003). Moreover, the quality of available bird habitat may vary considerably along the urbanization gradient in different areas (Aronson et al. 2014). For example, DeGraaf and Wentworth (1986) found that suburban habitats with mature shrubs and native trees supported higher densities of insectivorous birds, with the type of shrubs and trees present, rather than the proportional cover or even size of tree, the best determinant of insectivorous species abundance. Similarly, White et al. (2005) found significantly higher abundances of insectivores in streetscapes composed of native tree species than exotic streetscapes. Combined, these caveats suggest that landscape context and features within suburban and urban landscapes likely contribute greatly to urban bird communities. To determine how to best conserve biodiversity in urban and suburban environments, further research is necessary to determine which qualities of urbanized landscapes can best buffer the loss in functional diversity.
Conclusions
Several key results in our study suggest that urban environments filter avian communities as a function of their traits. We observed a decline in species richness and decline in functional diversity across the gradient. Likewise, emergent patterns in avian community structure suggest that some life history traits, especially insectivory, were found to be strongly associated with the sensitivity of species to urbanization processes. As such, our results show that development intensity, as measured by impervious surface, is strongly related to avian guild structure, and thus, observed declines in species richness across the rural-to-urban gradient likely result from environmental filtering by life history traits.
Our study provides crucial evidence that urban habitats filter avian communities and is thus a critical early step toward understanding how the process of urbanization influences native avifauna. There remains a clear need to assess the multiple scales of response of bird communities to environmental features, as well as increasing the spatial and temporal extent analyses. As climate change is projected to strongly impact bird distributions in ways that interact with human land use practices (Bateman et al. 2016), the need to understand patterns of community composition across temporal, latitudinal, and rural-to-urban gradients is paramount. Likewise, further analysis of how landscape context, features of local habitat structure and composition, and qualities of the human system explain the variation of birds in response to urbanization may offer key insights into how to manage habitats to mitigate the loss of biodiversity in an urbanizing world.
Acknowledgments
The authors would like to acknowledge the support of the Smithsonian Office of Fellowships in providing funding for this research and Neighborhood Nestwatch participants for providing access to our research sites. The authors would also like to thank Christopher Lepcyk, T. Brandt Ryder, Rebecca Zurlo, and two anonymous reviewers for their helpful suggestions in the preparation of this manuscript.
