Influence of new residential construction varying in housing density on bird species, human tolerance guilds, and communities

Abstract

Human population growth and changing settlement patterns fuel the development of urban fringe lands worldwide, with implications for biodiversity. We conducted a 12-year study of birds in the fast-developing urban fringe lands of the central Puget Sound region, Washington, USA, to examine the effect of development configuration on birds. We hypothesized that lower-intensity conservation developments, compared to higher-intensity planned community developments, would benefit the overall bird community, as well as native forest birds (avoiders of human development) and avian generalist species (adapters to human development), but that higher-intensity planned community developments would benefit synanthropic species (exploiters of human development). We fit single-species and multi-species occupancy models to test these hypotheses. Consistent with our hypotheses, we found that a greater proportion of the overall community, avoiders, and adapters occupied lower-intensity conservation developments compared to higher-intensity planned community developments. However, we did not detect an effect of development type on the exploiter guild, and we found that species in the exploiter guild are variable in their response to the configuration of suburban developments. We also hypothesized that human tolerance guilds would be a useful predictor of individual species responses to development type. This hypothesis was somewhat supported: we found that, for avoiders, 87% of species in the guild had the same response to development type as the overall guild; for adapters, 63% had the same response as the overall guild, and for exploiters, only 44% had the same response as the overall guild. Our results indicate that the configuration of suburban developments can have a meaningful impact on bird communities, particularly on those species that are most sensitive to any level of development.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-54885-3.

Introduction

Human population growth and changing settlement patterns fuel global urbanization, leading metropolitan areas to swell through densification and expansion^1–4. The resulting modifications in land use and land cover restructure plant and animal communities^5–15. Exurban¹⁶ and rural lands at the fringe of metropolitan boundaries may experience dramatic landscape alteration as native habitats and agricultural areas are subdivided into residential housing and associated infrastructure^17–25. This ongoing peri-urbanization indicates the need for greater understanding of impacts on local and regional ecological processes critical for conservation^26,27. Paramount is the direct assessment of residential development effects on birds and bird habitats^{12,21,22,28–39}.

Understanding avian responses to urbanization requires approaches for evaluating responses across large and diverse species assemblages. Aggregating species responses by functional guilds has proven valuable in many community ecology contexts^40,41. Birds inhabiting urbanizing landscapes are regularly grouped by their tolerance of human development: including avoiders (native habitat specialists), adapters (early successional species utilizing newly created or existing edge habitat), and exploiters (habitat generalist synanthropic species)^29,42–44. Species’ responses to the development of urban fringe lands are likely to vary across these guilds and may further vary depending on the configuration of suburban development. All suburban developments are not created alike: certain suburban developments have been classified as conservation developments (CDs)^45–56, with generally lower housing density, and where housing is more clustered together, resulting in the retention of more native vegetation. Developments of these kinds have been contrasted with (master-) planned community developments (PCDs)^57–63, which have higher housing density, and sometimes preserve portions of residual habitat nearby.

We studied birds in actively developing urban fringe lands of the central Puget Sound region, Washington, USA. Between 1986 and 2007, this region, home to the city of Seattle, increased urban land cover by nearly 140%⁶⁴. This region provides an ideal location to investigate how dynamic suburbanization may influence bird communities. Building on past research on birds^65–69, landscape change^70,71, and human-tolerance guild associations^68,72, we developed several hypotheses related to responses of the overall bird community, human tolerance guilds, and individual species to new development configurations in urban fringe lands.

First, we hypothesized that greater bird diversity is supported by CDs compared to PCDs, as they retain more, though equally fragmented⁷⁰, natural habitat than PCDs. Therefore, we predicted that occurrence of the bird community overall would be higher in CD sites than in PCD sites. Second, we hypothesized that the lower-density housing in CD sites relative to PCD sites benefits avoiders and adapters, while exploiters benefit more from the abundance of anthropogenic features in PCD sites. Consequently, we predicted that occurrence of avoiders and adapters would be higher in CD sites while occurrence of exploiters would be higher in PCD sites. Finally, we hypothesized that individual species’ responses to the development of urban fringe lands are consistent with overall guild responses. Local species belong to human tolerance guilds, based on their natural history⁷², which if uniformly responsive to development style can be useful for urban and conservation planning^73,74. We predicted that most species members within each guild would respond similarly to development type as the guild overall. Our long-term study provides a valuable complement to the “space-for-time” experimental design dominating urban ecology and further supports the idea that suburban development can be executed in a manner more compatible with the maintenance of avian biodiversity.

Methods

Study design and data collection: the effects of residential development on birds and their habitats

We conducted our study over 12 years (1999–2010) at sites actively converting from forest (largely mixed coniferous-hardwood second growth⁷⁵ to variable-density residential housing within the lowlands of central Puget Sound, Washington State (Fig. 1)^70,71. This longitudinal approach allowed us to track the spatial evolution of landscape development and characterize local bird communities across study sites. Data collection either preceded new development deforestation and construction at study sites or was initiated soon after deforestation (but prior to construction), due to variations in logistics and landowner access.

Fig. 1.

Predicted occupancy of 19 bird species in the adapter human tolerance guild per sampling point for years 1999 (year 1 on plot) to 2010. Observations were made in the central Puget Sound region, Washington, USA across two suburban development types: conservation developments (CD) and planned community developments (PCD). Predictions were made under the top-ranked occupancy model for each species (see text for details on model selection), which either included an effect for development type (CD/PCD) or did not include this effect (Combined). The number in parentheses after the species name is the top-ranked model for the species. Model selection results are given in Table 1 and estimated model coefficients are given in Table SM2. Shading in each figure is the 95% Bayesian credible interval.

We separated subdivisions into conservation developments (CDs) and planned community developments (PCDs), initially based on pre-construction available density plans, and later confirmed through landscape classification analysis^70,71. By the end of our study, CD sites retained approximately 30% more forest, 10% more grass/bare-ground, and 20% less built cover than PCD sites. Both development types had similar amounts of total edge (Fig. 2). Shrub and open water cover types were negligible and undifferentiated between development types (Fig. 2). We ultimately selected three high-density (PCD) and four low-density CDs for analysis (Fig. 1). This represented the population of PCD sites in the region and a substantial proportion of CD sites at the start of our study. Elsewhere, we have contrasted these sites with preexisting developments and forested reserves^66–68.

Fig. 2.

Predicted occupancy of 9 bird species in the exploiter human tolerance guild per sampling point for years 1999 (year 1 on plot) to 2010. Observations were made in the central Puget Sound region, Washington, USA across two suburban development types: conservation developments (CD) and planned community developments (PCD). Predictions were made under the top-ranked occupancy model for each species (see text for details on model selection), which either included an effect for development type (CD/PCD) or did not include this effect (Combined). The number in parentheses after the species name is the top-ranked model for the species. Model selection results are given in Table 1 and estimated model coefficients are given in Table SM2. Shading in each figure is the 95% Bayesian credible interval.

We employed 50-m fixed-radius point-count surveys. We randomly placed eight sampling points in each site, except for one CD site that had only five sampling points due to its smaller size. At sites where development was already underway, we stratified sampling points to include areas of forest and matrix. All points were > 150 m apart, except for a few forest points where we maximized separation within the only forest fragment at the site⁷⁶. At each point, we surveyed avian communities for 10 minutes⁷⁷ during 3–4 visits per breeding season from 1999 to 2010 (though one CD site was not sampled in 1999 nor in 2009 and one PCD site was not sampled in 2009)^68,78. We reduced variation in detection across space and time by limiting the number of observers, re-training survey techniques prior to commencing data collection each year, and sampling from the start through the end of the breeding season. To account for remaining variation in detection, we fit occupancy models to these data (see Data Analysis). All research was approved by the IACUC committee at the University of Washington (protocol 3077-01).

Data analysis

We defined guilds based on avian tolerance to human presence or activity, categorizing birds that were observed in the point counts as either human avoiders (18 species), adapters (26 species), or exploiters (9 species)⁴⁴. Individual species assignment was informed by published natural history characteristics⁷⁹ and past research in this study system⁷²(Table SM1). For 10 of these species, we had fewer than 30 observations across the entire dataset, so we excluded these species from analyses (Table SM1).

We collapsed the data to a binary response (detected/not detected) for each species at each point and visit because we found that the vast majority of individual species counts (> 90%) at this scale were either 0 or 1. We fit occupancy models⁸⁰ to these data, separately for each species. Occupancy models use repeated sampling within a period when the population is assumed to be closed to decompose the observation and state processes. We treated rounds within sampling points as replicates, such that we predicted a probability of occupancy for each species at each of the sampling points, within the seven sites, in each year. Thus, we modeled two parameters for each species. The first of these was , the probability of detecting a given species at point j in year y at replicate n, conditional on that species being present at point j in year y. We modeled for each species with a single intercept term, a random effect of site with standard deviation , and a random effect of year with standard deviation .

The second parameter we modeled was , the probability of a given species occupying site j in year y. We built six models for . These models were composed of the following fixed effects: (1) a null model, containing only an intercept term, (2) development type (i.e., CD or PCD), (3) development type + year, (4) development type + year + year², (5) year, (6) year + year². This allowed us to estimate a single overall site-type effect for each species, while accounting for potential linear or quadratic changes in occupancy over the years of the study, as development was unfolding.

The single-species occupancy models allowed us to isolate individual species for evaluation of their responses to development. To examine responses of the overall community (all species) and each guild (avoiders, adapters, exploiters), we built multi-species occupancy models^81,82, with each of these four groups. While multi-species occupancy models can be used to effectively account for rare, and even unobserved, species, we wanted to be able to fit single-species occupancy models as well as multi-species occupancy models to all species in the dataset, in order to test our hypothesis about guild membership as a predictor of responses to development type. Thus, we only included those species in our multi-species occupancy models that occurred in the dataset at a high enough frequency (i.e., > 30 total observations) to allow for individual modeling. In the multi-species occupancy models, we added a species dimension, such that we modeled , the probability of detection for species k at point j in year y and replicate n, and , the probability of occupancy for species k at point j in year y. We added two additional pieces to these guild models relative to the single-species models. Specifically, we added random effects of species to , with standard deviation = , and to , with standard deviation = . In these models, we estimated community- or guild-wide fixed effects on , such that we estimated a single, community- or guild-wide effect of development type.

We used Bayesian methods to fit our occupancy models⁸³. We fit the models in NIMBLE (version 1.3.0)⁸⁴ through R (version 4.4.2)⁸⁶. Posterior distributions for individual species were based on three chains, with 125,000 samples per chain after discarding 25,000 as burn-in. Posterior distributions for the community and guilds were based on three chains, with 35,000 samples per chain after discarding 5,000 as burn in. These sampling parameters ensured that the Gelman-Rubin statistics⁸⁷, , were < 1.03 for all monitored parameters. Data and code for our analysis are available at https://github.com/Quantitative-Conservation-Lab/Delap_etal_2026_ScientificReports.

We used the indicator variable method for model selection^88–90. We associated a Bernoulli-distributed indicator variable, w, with each fixed effect in the models for or , save the intercept, which appears in all models. Sampling these indicator variables using Markov chain Monte Carlo (MCMC) sampling allows for variables to be “turned on” or “turned off” across iterations in the MCMC chains, thus resulting in sampling across the set of six candidate models. The proportion of samples in which a given model is selected then serves as a weight in favor of that model. We chose priors on the w parameters to ensure that the six candidate models had an equal prior probability of 1/6, and that there was a dependency between the year and year² parameters, such that year² only appeared in models that also contained year (specifically, ~ Bernoull(1/2), ~ Bernoulli(2/3), | = 1) ~ Bernoulli (1/2); see Smith et al. 2011 for details). We also followed Link and Barker (2006) in setting the priors on all parameters appearing in the model for dynamically to ensure that the total prior variance of the linear predictor was constant, regardless of the model being sampled. We set the priors on parameters to normal(0, ), with the variance parameter set to , with  K equal to the total number of parameters in the model (i.e., 1 + + + , where 1 accounts for the intercept). We then gave the , the total prior variance on the linear predictor for , an inverse gamma distribution (rate = 3.29, shape = 7.80), which, Link and Barker (2006) note, results in an approximately uniform prior distribution of the linear predictor on the probability scale (Link and Barker 2006). We set priors on intercept parameters to normal (0, 1). We set priors on all standard deviations for random effects – , , , and – to uniform(0,1).

Results

Hypothesis 1: CD sites have greater bird diversity compared to PCD sites

We predicted that occupancy probability would be higher for the bird community overall in conservation development (CD) sites compared to planned community development (PCD) sites. The best model, based on our indicator variable selection approach, supported our prediction (model 4; Table 1). Under this model, the coefficient for development type, which is an estimate of the effect of being in a PCD site relative to a CD site, was negative (β_{development Type (PCD)} = -0.44, 95% Bayesian Credible Interval = -0.55, -0.33; Table SM2). In general, the highest occupancy across all species in the community occurred in the middle years of the study (Fig. 3).

Table 1.

Model selection results for 43 species, classified into three human tolerance guilds (avoiders, adapters, and exploiters), in addition to the overall community and the overall guilds, observed in developing urban fringe lands of the central Puget Sound region, Washington, USA, between 1999 and 2010, based on occupancy models for each species, guild, and the community. Model names include model number used in text. 

Guild or species	Model
	1:Null	2:Development type	3:Development type Year	4:Development type Year Year2	5:Year	6:Year Year2
Community	0.000	0.000	0.000	1.000	0.000	0.000
Avoiders	0.000	0.000	0.800	0.200	0.000	0.000
Brown Creeper	0.015	0.356	0.357	0.261	0.005	0.006
Black-throated Gray Warbler	0.005	0.025	0.704	0.248	0.006	0.012
Chestnut-backed Chickadee	0.000	0.013	0.280	0.708	0.000	0.000
Golden-crowned Kinglet	0.024	0.276	0.439	0.216	0.021	0.024
Hairy Woodpecker	0.223	0.300	0.198	0.093	0.126	0.060
Hutton’s Vireo	0.000	0.001	0.276	0.636	0.004	0.082
Pacific Wren	0.000	0.000	0.412	0.588	0.000	0.000
Pacific-slope Flycatcher	0.000	0.000	0.601	0.399	0.000	0.000
Purple Finch	0.091	0.091	0.199	0.156	0.263	0.200
Red-breasted Nuthatch	0.023	0.584	0.196	0.134	0.023	0.041
Red-breasted Sapsucker	0.010	0.012	0.116	0.379	0.134	0.349
Steller’s Jay	0.037	0.121	0.481	0.241	0.038	0.083
Swainson’s Thrush	0.000	0.178	0.725	0.097	0.000	0.000
Western Tanager	0.015	0.017	0.024	0.402	0.034	0.509
Wilson’s Warbler	0.030	0.175	0.512	0.208	0.047	0.027
Adapters	0.000	0.000	0.000	0.999	0.000	0.001
American Goldfinch	0.000	0.000	0.023	0.310	0.075	0.592
American Robin	0.242	0.053	0.034	0.120	0.139	0.412
Black-capped Chickadee	0.229	0.455	0.148	0.061	0.075	0.032
Bewick’s Wren	0.011	0.514	0.248	0.208	0.012	0.007
Black-headed Grosbeak	0.054	0.182	0.425	0.119	0.166	0.055
Bushtit	0.166	0.257	0.214	0.133	0.141	0.088
Cassin’s Vireo	0.001	0.061	0.020	0.863	0.001	0.055
Cedar Waxwing	0.000	0.013	0.280	0.708	0.000	0.000
Dark-eyed Junco	0.438	0.214	0.075	0.053	0.131	0.089
Northern Flicker	0.065	0.076	0.204	0.341	0.079	0.236
Orange-crowned Warbler	0.005	0.021	0.026	0.653	0.008	0.287
Pine Siskin	0.001	0.001	0.042	0.718	0.004	0.233
Song Sparrow	0.002	0.199	0.079	0.670	0.001	0.048
Spotted Towhee	0.000	0.490	0.429	0.082	0.000	0.000
Violet-green Swallow	0.000	0.000	0.029	0.132	0.232	0.606
Warbling Vireo	0.008	0.026	0.344	0.189	0.266	0.167
White-crowned Sparrow	0.000	0.000	0.300	0.124	0.409	0.167
Willow Flycatcher	0.000	0.003	0.004	0.456	0.005	0.532
Yellow-rumped Warbler	0.064	0.062	0.222	0.184	0.255	0.213
Exploiter	0.000	0.000	0.000	0.227	0.000	0.773
American Crow	0.000	0.000	0.152	0.037	0.667	0.143
Anna’s Hummingbird	0.118	0.091	0.067	0.301	0.075	0.348
Barn Swallow	0.230	0.338	0.143	0.097	0.115	0.077
Brown-headed Cowbird	0.007	0.039	0.161	0.632	0.013	0.148
European Starling	0.000	0.000	0.045	0.361	0.076	0.518
House Finch	0.185	0.163	0.268	0.121	0.171	0.092
House Sparrow	0.000	0.043	0.712	0.244	0.000	0.001
Rock Pigeon	0.000	0.000	0.000	0.285	0.000	0.715
Rufous Hummingbird	0.221	0.238	0.101	0.179	0.090	0.172

Details on models are provided in the text. Model selection results show posterior model weight for each species. The best model for each species is shown in bold.

Fig. 3.

The study area (inset square) in Washington State (top) and enlarged (bottom) indicating the location of seven study sites across two suburban development types, conservation developments (CD, n = 4) and planned community developments (PCD, n = 3) monitored for avian species from 1999–2010. Maps were created using ArcGIS Online (ESRI) software, www.esri.com.

Hypothesis 2: avoiders and adapters benefit from CD sites, exploiters benefit from PCD sites

We predicted that occupancy probability of avoiders and adapters would be higher in CD sites while occupancy probability of exploiters would be higher in PCD sites. Our results partially supported these predictions. The best model for both avoiders (model 3; Table 1) and adapters (model 4; Table 1) included the effect of development type (PCD relative to CD at the mean value for year). This effect was negative for both avoiders (β_{development type (PCD)} = -1.61, 95% BCI = -1.82, -1.40; Table 1) and adapters (β_{development type (PCD)} = -0.35, 95% BCI = -0.50, -0.19; Table 1). However, development type did not appear in the top model for exploiters. In general, occurrence of avoiders declined over the course of the study, while occurrence of adapters and exploiters peaked in the middle years of the study (Fig. 3).

Hypothesis 3: guild membership is a useful predictor of individual species responses

We predicted that individual species within each guild would respond to the configuration of development similarly to the guild overall. For avoiders, our prediction was generally supported; 13/15 = 87% of species in the avoider guild had an effect of development type in their top models, as did the guild overall (Table 1), and in each of these cases, the coefficient was negative (Table SM2, Fig. 4). The two species without an effect of development type in their top model included Purple Finch (Haemorhous purpureus) and Western Tanager (Piranga ludoviciana). For adapters, our prediction was somewhat supported. Twelve of 19 = 63% of species had an effect of development type in their top model, as did the guild overall (Table 1). In all of these cases, the coefficient was negative, as it was for the guild overall (Table SM2, Fig. 5). Thus, for those adapter species in which an effect of development type was discernible, it was the same effect as the guild overall, but the effect was discernible in only just more than half of the species in the guild. Adapter species without an effect of development in their top model included American Goldfinch (Spinus tristis), American Robin (Turdus migratorius), Dark-eyed Junco (Junco hyemalis), Violet-green Swallow (Tachycineta thalassina), White-crowned Sparrow (Zonotrichia leucophrys), Willow Flycatcher (Empidonax traillii), and Yellow-rumped Warbler (Setophaga coronata). For exploiters, our results did not support our predictions. Five of the 9 exploiter species had an effect of development type in their top models, but the guild overall did not (Table 1; Fig. 6). For those 5 species, the estimated effects of development type were variable: negative for 3 and positive for 2 (Table SM2, Fig. 6).

Fig. 4.

Landcover composition and configuration values for percent cover classes of forested, built (impervious surface), bare ground/grass, shrub, open water, and total edge (meters) at conservation development (CD) and planned community development (PCD) sites based on orthoimagery classification from 1998–2010 in central Puget Sound region of Washington State. Orthoimagery was not available for all sites in all years. Scales are adjusted among panels to view temporal trends more easily.

Fig. 5.

Predicted occupancy of the bird community in toto, and human tolerance guild members, per sampling point for years 1999 (year 1 on plot) to 2010. Observations were made in the central Puget Sound region, Washington, USA across two suburban development types: conservation developments (CD) and planned community developments (PCD). Human tolerance guilds include avoiders (n = 15 species), adapters (n = 19 species), and exploiters (n = 9 species). Predictions were made under the top-ranked occupancy model for each guild (see text for details on model selection), which either included an effect for development type (CD/PCD) or did not include this effect (Combined). The number in parentheses after the guild name is the top-ranked model for the guild. Model selection results are given in Table 1 and estimated model coefficients are given in Table SM2. Shading in each figure is the 95% Bayesian credible interval.

Fig. 6.

Predicted occupancy of 15 bird species in the avoider human tolerance guild per sampling point for years 1999 (year 1 on plot) to 2010. Observations were made in the central Puget Sound region, Washington, USA across two suburban development types: conservation developments (CD) and planned community developments (PCD). Predictions were made under the top-ranked occupancy model for each species (see text for details on model selection), which either included an effect for development type (CD/PCD) or did not include this effect (Combined). The number in parentheses after the species name is the top-ranked model for the species. Model selection results are given in Table 1 and estimated model coefficients are given in Table SM2. Shading in each figure is the 95% Bayesian credible interval.

Discussion

We conducted a long-term study of avian communities inhabiting newly developing residential suburbs occurring in the central Puget Sound region of Washington State. Our results suggest that a less dense, conservation-minded pattern of development (lower housing density and greater retention of pre-development vegetation) has the potential to both boost species diversity in moderately disturbed areas^44,69,71,91 and minimize biotic homogenization compared to the very dense developments in nearby urban areas by better supporting native forest species (i.e., species that are avoiders of human development). In addition, we found that human tolerance guilds were a useful^42,92, though imperfect, predictor of responses for species within the guilds, particularly for avoider species. For adapter species, only slightly more than half exhibited the same (negative) response to conservation developments (CDs) relative to planned community developments (PCDs) as the guild overall, and we were not able to detect an effect of development type in the remaining adapter species. For exploiter species, we found no consistent effect of development type: 4 species showed no effect, 3 showed a negative effect, and 2 showed a positive effect. This mix of species’ responses effectively washed out any effect of development type for the exploiter guild.

Framing bird community change by the response of human development tolerance guilds has been useful in past research^21,29,43. In this study, CDs benefited most avoiders, some adapters, and a very few exploiter species, relative to PCDs. Avoiders likely benefited from greater remnant forest in CDs^28,32. Initial increases in avoider presence may result from these forest-associated species crowding into persisting forest patches in the aftermath of fragmentation^66,78,93. Later declines could result from human activity⁷⁰ and insufficient adaptation to the shifting nest-predator community⁷⁸. Overall, avoider occurrence declined over the course of our study in both development types. While tolerance guilds were useful, they were not perfect predictors of responses across individual species; previous literature considers these in greater detail for this system^{67,68,76,78,94}.

The adapter guild benefited from CDs relative to PCDs, although individual species’ responses were variable. Generally, adapters appeared to increase slightly from the beginning to the end of our study, with the highest occurrence in the middle years of development. This chronology coincides with periods when habitat conversion is dominated by forest removal and establishment of new bare earth/grass landcover types (note the residential development types differed in their residual forest cover, but experienced similar degrees of fragmentation¹¹. Adapters, or early successional species, colonize these emerging habitats for their new nesting and foraging opportunities^44,94,95,96].

The exploiter guild appeared to have no clear response to development type. Exploiters benefit from habitat augmented with built land cover^21,44,71. For instance, artificial nest substrates associated with human structures can lead to high nest success⁷¹, despite likely increases in locally subsidized predators, both native (e.g., corvids [Corvidae spp.], North American opossums [Didelphis virginiana], North American raccoons [Procyon lotor] ), and non-native (introduced rats [Rattus norvegicus, Rattus rattus], domestic cats [Felis catus]^96–98. High nest success coupled with an abundance of anthropogenic food resources (bird feeders, pet food, vegetable gardens, ornamental fruit trees, and shrubs) likely translates into elevated survival rates for at least some juvenile and adult exploiters⁷¹. Our results suggest that exploiters, as a guild, are able to persist regardless of development type.

Past research¹⁰⁰ has revealed increased avian community richness following the conversion of native vegetation to suburban developments, likely responding to the increased heterogeneity of habitat¹⁰¹. However, this may not be true in all development configurations²¹. In our study, the occurrence probability across all species in the community was highest during the middle years of the study, suggesting this may correspond to the highest habitat heterogeneity during the establishment phase of new suburban development. This is consistent with urbanization gradient analyses in this study area⁷¹ and elsewhere^28,44,102. In general, as habitat becomes more heterogeneous, avian species diversity increases^29,71 often accompanied by increased breeding territory densification^29,103 and thus a rise in the total population of birds occupying the landscape. However, configuring high-density developments to maintain nearby tracts of intact primary or secondary forest may contribute to the maintenance of habitat specialist species, such as the avoider species identified in our study^104–106.

Our CDs provided increased avian richness and refugia for forest specialist bird species compared to the more intensively built PCDs. Relative to PCDs, CDs remain highly forested with diffuse, low-density residential structures. We expect that bird communities are responding positively to the moderate increase in habitat heterogeneity paired with a smaller human footprint in CDs^107,108. The likelihood of CDs proliferating in the future, as opposed to PCDs, will largely depend on the changing pressures of population growth²¹. Should existing residential areas densify through infill or the creation of new high-density subdivisions, we expect the abundance of avoider bird species to decrease in those areas. We did not document localized species extinction from our high-density development study sites, though occupancy probabilities for species such as Brown Creepers (Certhia americana), Black-throated Gray Warblers (Setophaga nigrescens), Golden-crowned Kinglets (Regulus satrapa), Hutton’s Vireos (Vireo huttoni), Pacific Wrens (Troglodytes pacificus), Pacific-slope Flycatchers (Empidonax difficilis), and Wilson’s Warblers (Cardellina pusilla) all declined discernibly across both development types over the course of our study. Dispersal may sustain specialist species in densely settled areas for a time, though as their reproduction and survival declines so too will their populations^71,78.

Our primary objective was to correlate differences in avian community metrics with habitat changes occurring at variable-density residential subdivisions in real-time (onset of development activity, during development, post-development). While alterations to physical habitat composition and structure are critical considerations, additional anthropogenic factors present alternative mechanisms for any observed changes in bird communities. We did not directly investigate alternative variables of human presence, such as landscape context⁹², which have been shown to stress species and alter foraging and nesting decisions directly. Likewise, the introduction of domestic animals, namely cats and, to a lesser extent, dogs (Canis familiaris), presents existential threats to some birds⁸². Noise pollution can limit or modify avian vocal communication⁸⁸ as well as their detection by observers. Artificial light sources may disrupt songbirds directly⁸⁹ as well as localized nocturnal avian and mammalian nest predator communities⁹⁰. Avian mortality resulting from collisions with anthropogenic features is significant⁹¹. Window strikes take a toll on some species⁹², and the prevalence of windows may increase post-development, making matters worse for birds (though applications reflecting ultra-violet light may mitigate some negative impacts^92–94. The frequency of bird collisions with automobiles⁹⁵ may also shift, gaining in severity when traffic rises or falls with the adoption of alternative transportation means and corridors. While this suite of human presence-related variables may positively correlate with development density, additional research is needed to determine if any are causal mechanisms and which species might be most affected.

While novel in its longitudinal monitoring of bird communities in and around active new residential development sites, our study design was limited to four CD sites and three PCD sites. These were all the known CD sites and PCD sites occurring in the study area at the start of data collection (circa 2000). A few additional CD sites were followed for several seasons but ultimately omitted from analyses due to a lack of long-term data (due to a lack of observer access or other logistical challenges). This small sample size can limit the generalizability of our findings beyond the specific locations if the sites we selected do not represent landscape and bird community changes in the system at large. In addition, a small sample size limits statistical power, which may hinder the detection of meaningful relationships. Although limited, the power of our study was the substantial long-term data collected at each of the study sites, which produced a dataset that was sufficient to detect differences^98,99 between development types. These differences were consistent across replicates of each development type and were consistent with predictions based on natural history. Similarly, here, we found strong effects that were consistent with our predictions, lending credence to our results. While increasing the sample size was not possible within our study system alone, future collaboration with researchers in other geographic locations¹⁰⁰ could test and extend our results.

Author contributions

J.D and J.M. conceived the experiment(s), J.D. and J.M. conducted the experiment(s), S.C. and J.D. analyzed the data. J.D. led the interpretation of results and preparation of the manuscript. All authors reviewed the manuscript.

Data availability

The datasets generated during and/or analyzed during the current study are currently available in the GitHub repository, https://github.com/Quantitative-Conservation-Lab/Delapetal_InReview, and will be released to the Zenodo repository upon acceptance.

Declarations

Competing interests

The authors declare no competing interests.