Introduction

Attempts to better understand and mitigate the negative effects of urbanization on wildlife have grown in response to rapidly increasing human populations (Steffen et al. 2015), the continuing expansion of megacities (Seto et al. 2012; Kennedy et al. 2014), and recognition that urbanization represents a major extinction threat (McDonald et al. 2008; Seto et al. 2012). As cities expand beyond present boundaries into surrounding semi-natural habitats, the potential, and need, to incorporate reserves for native species will both grow. Studies of birds have been at the forefront of urban ecology (Marzluff et al. 2001), but our understanding of what structures avian systems is still wanting because of the many potential sources of variation in urban settings, and regional differences in their importance (Lerman et al. 2021).

Answering the question of what drives variation in species richness and abundance among urban avian communities is complicated because it requires consideration of the geometry (i.e., size and shape) of the park itself and other factors operating from the higher landscape to the lower local level (Nielsen et al. 2014). For instance, species richness of birds regularly increases with park area (Crooks et al. 2001; Husté et al. 2006; Ikin et al. 2013; Myczko et al. 2014; Huang et al. 2015; Kang et al. 2015; Maseko et al. 2020; de Groot et al. 2021; but see Cooper and Walters 2002, Martensen et al. 2008), while elongated fragments that contain relatively high amounts of edge has been associated with low species richness (Natuhara and Imai 1999) and low nest success and/or survival of fledglings (Batáry and Báldi 2004; Shipley et al. 2013). Regardless of geometry, parks are not true islands and land use in a park’s surrounding landscape (i.e., the “matrix”) can either ameliorate or exacerbate negative effects of shrinking park area, habitat fragmentation, and isolation (Taylor et al. 2016, Amaya-Espinal et al. 2019, Spake et al. 2020). The negative effects of highly developed matrices may arise from several sources, including inhibition of dispersal movements among parks (Martensen et al. 2008; Tremblay and St. Clair 2011), support of synanthropic species that may also use parks and compete with native habitat specialists (Donnelly and Marzluff 2004; Rodewald and Bakermans 2006; Oliver et al. 2011; Canedoli et al. 2018; Amaya-Espinel et al. 2019), or increase in depredation of birds and/or their nests by mesopredator species that increase in response to the absence of large predators in urban environments (Fischer et al. 2012; but see Rodewald et al. 2011). On a more local level, species richness of parks often responds strongly to variation in structural vegetation complexity (Myczko et al. 2014; Rush et al. 2014; Kang et al. 2015), habitat diversity (Donnelly and Marzluff 2004; Cornelis and Hermy 2004), or floristic composition (Husté et al. 2006; Shwartz et al. 2008; Paker et al. 2014; Huang et al. 2015; Taylor et al. 2016).

However, because birds vary greatly in nesting, feeding, and migratory ecology we should not expect all species to respond equally to urbanization (Concepción et al. 2015; Callaghan et al. 2020). For example, urbanization adversely affects nesting site specialists (Callaghan et al. 2020) and the ground nesting guild typically suffers in urban settings presumably because of elevated rates of nest predation from mesopredators and/or domestic/feral animals (Chace and Walsh 2006). Noninsectivorous passerines and raptors often fare well in cities, in part, because of elevated food supplies resulting from either direct (Robb et al. 2008; Plummer et al. 2019) or indirect (Chace and Walsh 2006) supplementation by humans, respectively. Migratory behavior represents a fundamental difference in life history among birds, but our understanding of how different migratory guilds respond to urbanization is limited and lacks consistency. Park and Lee (2000) reported that, relative to migrants, richness of resident bird species increased more quickly with park area and that larger parks supported twice the number of residents as migrant species. Husté and Boulinier (2007) found equal rates of increase in richness of residents and migrants with area, but that residents were again more diverse. By contrast, Taylor et al. (2016) reported that resident bird species richness responded to vegetation composition within parks but was independent of park area, whereas area was the best predictor of migrant species richness. In general, negative responses to urbanization appear stronger in long-distance migrants than other species (e.g., Friesen et al. 1995, Hennings and Edge 2003, Rodewald and Bakermans 2006), possibly because the small population size of the former may lead to more frequent local extirpation (Husté and Boulinier 2007). Given the poor state of the world’s birds (Lees et al. 2022), declining populations of many migrant species in Europe (Sanderson et al. 2006; Both et al. 2010) and North America (Rosenberg et al. 2019), and that urban growth is inevitable, urban greenspaces must be better planned to conserve migrant birds (see also Rodewald et al. 2013).

Portland, Oregon (USA), is the 26th largest city in the US (~ 666,500 in 2022) http://worldpopulationreview.com/us-cities/), is growing rapidly (600,000 + more people are expected by 2035; METRO 2015), but also supports an extensive park system (4740 ha, 13.7% of city land area). Most of the system (3206 ha [68%]) is maintained as natural areas or otherwise undeveloped parks (no development beyond trails), and individual patches range from ~ 1 to 2064 ha. We surveyed breeding bird communities in 48 sites to evaluate affects of park geometry and connectivity, landscape composition, and local (within-park) habitat structure and floristic composition on richness and abundance of bird species associated with upland forest habitats. Based on past literature (see above) we hypothesized that (1) species richness and density of bird populations would increase with park area and eclipse all other factors as the primary source of variation in bird community structure for all migratory guilds (defined below), and that (2) given Portland’s high tree cover (~ 30%), high structural habitat/floristic diversity within parks would have a greater positive influence on birds than high park connectivity or favorable landscape composition. Furthermore, because conservation planning requires estimates of species-specific minimum area requirements, (3) for those species that are most dependent on forested parks, we attempt to identify minimum area needs (sensu Robbins et al. 1989).

Methods

Study area, site selection, and site properties

We sampled birds and vegetation in 48 second growth parks located in the northern Willamette, Sandy, and lower Columbia river watersheds in the greater Portland metropolitan area (city center: 45.52 N, -122.68 W). The region is characterized geologically by broad alluvial flats punctuated by scattered volcanic buttes that rise to low hills in the west and south of the Willamette and Columbia rivers, respectively. Elevation ranges from 20 to 250 m. Climate is characterized by cool, wet winters (December: 4.7o C and 13.9 cm precipitation) and warm, dry summers (August: 20.8o C and 1.7 cm precipitation). Later seral-stage vegetation is dominated by shade tolerant conifers including western hemlock (Tsuga heterophylla), grand fir (Aibes grandis), and western red cedar (Thuja plicata), but earlier seral-stage Douglas fir (Pseudotsuga menziesii) and big-leaf maple (Acer macrophyllum) now dominate the mainly second growth stands throughout the study area.

Parks were selected by stratified random sampling to represent the full range of possible areas (< 1 to 2,000 ha). All sites contained multistory forest with ≥ 50% canopy closure and limited development. All but one (Oxbow Regional Park) fell within Portland’s urban growth boundary, and 42 of 48 were city parks. The other six were privately owned parcels. Sites were delimited by roads and adjacent urban development (i.e., the presence of buildings, impervious surfaces, or intensively maintained vegetated habitats), rather than ownership or parcel boundaries (for details see Supplementary Information [SI]). Park area, shape and connectivity (Table 1) were extracted using ArcGIS and FRAGSTATS (McGarigal and Marks 1995) from data provided by Portland’s METRO Regional Services’ RLIS database (Metro 2004). All analyses utilized digitized 2001 aerial photographs of the greater Portland metropolitan region with a pixel resolution of 3.05 m. Details are provided in the SI.

Table 1 Description of the variables used in the analysis of among park variation in species richness and density of birds from 2003 in Portland, OR (USA). Description of each principal component (PC) axis (left to right) reflects negative to positive scores. Percentages refer to the portion of total variation accounted for by each PC axis. Details of principal component analyses in Electronic Supplementary Materials
Full size table

Landscape composition: quantification and analysis

We quantified landscape composition (undeveloped forest, developed forest and total forest [= undeveloped + developed], light urban development, heavy urban development, total urban development [light + heavy], and open spaces without trees), human population density, and street density in the 500 m buffer surrounding each park (see SI). We then subjected these data to a principal component analysis (PCA) to identify the main gradients of landscape variation across Portland for the purpose of testing for avian responses to differences in landscape structure. Principal component 1 (LandPC1) accounted for 54% of the variation in landscape structure and described a gradient in which more urbanized landscapes (negative scores) that were characterized by high residential development and high population density were replaced by landscapes with increasing dominance of trees, low residential development, and low human population density in the 500 m buffer surrounding each park (positive scores; Table 1). Full details of PCA are provided in the SI.

Quantification and analysis of vegetation

We sampled vegetation in 2003 at 2 to 16 plots per park (total = 279), with larger parks having a greater number of plots to capture potentially greater vegetation diversity. Plots were randomly located (ArcGIS and Garmin 12XL GPS units; ± 10 m) within parks and associated with either vegetation sampling alone, bird point count stations, small mammal traplines, or amphibian pitfall arrays. In some cases, the plots associated with amphibian surveys were moved up to 10 m to accommodate pitfall installation. At all but the smallest parks, points were located at least 100 m away from forest edge and 200 m from one another.

Vegetation structure and composition at each plot were measured in 10 m radius circles. We identified most plants to species, and measured diameter breast height (DBH) of all trees with DBH > 2.5 cm. We also counted the number and estimated the volume (cross-sectional area x length) of all snags (standing dead tree) and logs (fallen dead tree) > 10 cm diameter. Subsamples of vegetation of the shrub and herbaceous layers, and forest floor were taken from two perpendicular transects that crossed through the center of the circle. Canopy closure was also estimated visually along the same two transects (see SI for full details).

Modified importance values (IV) were calculated for all tree species at all 279 points. Total coniferous and angiosperm IVs, along with canopy closure, number of trees in four size classes (< 10 cm, 10.1–30 cm, 30.1–60 cm, and > 60 cm), total volume of logs and snags (= coarse woody debris), tree species richness (3 classes: native, exotic, and total), and IV of the four most dominant trees in the data (2 conifers: Douglas-fir, western red cedar; 2 angiosperms: big-leaf maple, red alder [Alnus rubra]) were used in a PCA to describe forest tree structure. Eigenvalues of the first four PCs exceeded 1.0, and in total, they accounted for 59.4% of the variation in the data. The gradients in forest structure that they represent are described in Table 1. Similarly, we used proportional ground cover (i.e., bare ground, moss, leaf litter, small woody debris, herbs), litter depth, shrub density (3 layers: 0–1 m, 1.1-2 m, and 2.1-3 m), and proportional contribution to the shrub layer of eight common species in a PCA to describe the structure of the forest floor and understory. Eigenvectors of the first four PCs also exceeded 1.0 and accounted for 51.1% of variation. Descriptions of the gradients described by the shrub to forest floor PCA variables are provided in Table 1 (see SI for full detail). Plot-level PCA scores were averaged within each park to calculate park-level averages for all eight PCA axes.

Avian surveys: richness and abundance

Avian migratory activity drops sharply by late May in Portland and therefore surveys were conducted between 15 May and 18 July, 2003, to avoid sampling transiting migrants; our surveys thus represent samples of the breeding bird communities. All surveys were conducted by one person (DCB) between sunrise and 1100 h on days without rain and little to no wind. Three counts were made at all survey points over the course of the season. Point count locations (1–6 per park) were randomly selected, and all points within a park were surveyed on a single day, but survey times at any given point were changed between survey dates to minimize potential effects of time of day on results. All points were located ≥ 50 m from the park edge and ≥ 150 m from one another. A few small sites were not wide enough to accommodate points ≥ 50 m from a site edge, so points were located as far from edges as possible. Species accumulation curves for all 48 sites individually (Fig. 1a), and for the system in its entirety (Fig. 1b), indicated that 3 surveys were sufficient to describe avian community composition.

Fig. 1
figure 1

Absolute (a) and standardized (b) species accumulation curves in which total number of new species detected are plotted against survey number for all 48 parks. Standardized plots were based on Z-scores calculated for each park based on its mean and standard deviation of number of new species. The lines in (a) and (b) are the predicted number of new species obtained from asymptotic regression of cumulative number of new species against survey number

Full size image

We used the variable circular plot method (Bibby et al. 2000) to record all birds heard or seen out to 50 m. Ten-minute point counts were made following a 1-min period of quiet to allow birds to return to normal activity; flyovers were not included. To minimize double-counting, determination of multiple conspecifics was based on detection of counter-singing or simultaneous visual and/or aural detection along with mapping detections on data sheets in concentric circles at survey points (0–10 m, 10–25 m, 25–50 m). To avoid double-counting birds that fell between adjacent points that were relatively close to each other, mapped locations were compared in the field to assign individuals that fell between points to the nearer station.

Spot mapping is the most accurate estimator of avian abundance (Notes 1970). Hamel’s (1984) comparison of spot mapping estimates of abundance to counts made using the variable circular plot method showed that biasing abundance estimates from the variable circular plot method upward better approximated spot mapping estimates of abundance. Therefore, at each point count location our index of abundance of each species was the average of the two highest of the three counts made, which we averaged across all survey locations within each park to obtain an average abundance for each species. Absence of a species from a park was treated as an abundance of zero.

Analysis of avian species richness and density

Abundance measured within a specified area represents density and henceforth we use density to describe number of individuals detected by point counts. We omitted from our analyses all aerial insectivorous birds and rare species. Aerial insectivores (swifts and swallows) are wide ranging, forage above the canopy, and cannot be associated with a particular point in the forest. Rare species were those whose 95% confidence interval (CI) for average density, computed across all 48 parks, included zero (Appendix 1). Species classified as rare were raptors (Red-tailed Hawk and Cooper’s Hawk; all scientific names in Appendix 1), species that do not breed in the habitats in which we conducted our work, are irruptive, or were never detected at more than two sites (and at low density). For purposes of analysis, we classified species as belonging to 1 of 3 migratory guilds (Appendix 1). Year-round residents are present throughout the year, long-distance migrants overwinter primarily south of the border of the United States (i.e., Nearctic-Neotropical migrants), while short-distance/partial migrants exhibit intermediate behavior. We relied on Birds of the World accounts (Rodewald 2022) and personal experience in this system to classify all species to migratory category.

Park area and other variables that deviated from a Gaussian distribution were transformed (log10) prior to analysis. The quadratic of park area was included in all analyses as preliminary inspection of data indicated frequent nonlinearity in the relationship between park area and the response variables. All predictor variables (Table 1) were z-transformed to enable direct comparisons of coefficients (Schielzeth 2010), and the quadratic of park area was taken after standardization of area (Schielzeth 2010). Variance inflation factors in analyses were almost always below 2.0, rarely exceeded 3.0, and never approached 10 (Quinn and Keough 2002).

We used generalized linear models with a Poisson distribution and log link function to identify the drivers of variation in species richness (i.e., cumulative number of different species). Number of survey points per park was included as an offset to account for differences in sampling intensity. All variables were added to the model, and then removed by backward elimination by removing the variable with the highest P-value at each step until a final set of competitive models (ΔAICc ≤ 2) was obtained. Models within 2 AICc units that added a parameter without improving model deviance were not included in the final set to ensure that ΔAICc values were not the result of uninformative parameters (Arnold 2010). If the number of competitive models exceeded two we report model averaged parameter estimates for all variables retained by the models, but report original model output if only 1 or 2 competitive models existed. Model averaged coefficients were considered “significant” if their 85% confidence intervals did not include zero (Arnold 2010). Poisson regression does not generate a formal R2 but a pseudo-R2 was calculated as the difference in deviance between the null model (i.e., no predictor variables) and the fitted model divided by the null model’s deviance. We also used best subsets regression analysis to examine variation in density of the three migrant guilds in relation to the same set of predictor variables (Table 1). As for the Poisson regressions, we used an information theoretic approach to identify top models. Explained variation is reported as an adjusted R2 (R2adj).

Although we do not report them here, we also conducted individual analyses of density of all species using best subsets regression analysis. From these analyses we identified a set of forest-dependent species that we define as species with positive coefficients (at P ≤ 0.10) between their density and either park area or area2. For these species we performed separate best subsets regression analyses on the summed density of forest-dependent species belonging to the three migratory guilds using the predictors described in Table 1. Our objective was to remove area effects and identify other important determinants of density in the ecologically sensitive forest-dependent species that are most at risk in urban landscapes. For similar reasons we also calculated minimum area requirements (MAR) for the forest-dependent species using generalized linear models with a binomial distribution and logit link function to model presence/absence of each species in relation to park area. For species that reached an asymptote (100% probability of occurrence) within our range of park areas we used park area at 50% of the area at the asymptote to be a conservative estimate of minimum area needed to sustain populations (Robbins et al. 1989). For species for which probability of occurrence continued to increase within the bounds of our study (~ 2000 ha), we used park area at 50% of the probability of occurrence exhibited at 2000 ha as our estimate of MAR (Robbins et al. 1989). We then compared MAR in relation to body mass and migratory behavior to discern if MAR varied predictably with either variable.

We used STATISTIX version 9 (Analytical Software, Tallahassee, Florida, USA) for basic summary statistics, least squares regressions, PCAs, and best subsets regression analyses and JMP Pro 12 (SAS Institute Inc., Cary, North Carolina, USA) for generalized linear models of species richness and calculation of MARs. Statistics are reported as mean ± SE, with statistical significance accepted at P ≤ 0.05.

Results

Excluding aerial foragers, we recorded 41 regularly occurring and 17 rare species. Of the regularly occurring species, most were residents (49%), followed by long-distance migrants (32.0%), and short-distance/partial migrants (19%). Rare species included 2 raptors, and 3, 4, and 6 resident, long-distance migrant, and short-distance/partial migrant native species, respectively, and two invasive species (European Starling and House Sparrow; see Appendix 1 for a full list of species).

Determinants of richness. – Analysis of resident species richness yielded two competitive models (Table 2). TreePC2 appeared in both and showed that richness increased with greater abundance of small (DBH < 10 cm), mostly native, trees (Fig. 2a). The presence of HbShPC4 in the top model suggested resident richness possibly increased as the understory shifted from medium/high to low shrub cover where English ivy (Hedera helix) was abundant, but HbShPC4’s 95% CI included zero (Table 2). Among long-distance migrants, park area was the main contributor to variation in species richness followed by TreePC2 (Table 2). The negative quadratic for park area reflected a decelerating increase in species richness with park area (Fig. 2b). The top models for residents and long-distance migrants reduced the deviance for variation in richness by 34% and 55%, respectively. By contrast, our analysis accounted for little if any of the among park variation in species richness of short-distance/partial migrants (Table 2).

Table 2 Top models (ΔAICc ≤ 2) accounting for variation in species richness of birds in 48 parks from Portland, OR, in 2003. Data analyzed separately for resident species, long-distance migrants, and short-distance/partial migrants
Full size table
Fig. 2
figure 2

Species richness of (a) resident species in relation to TreePC2 and (b) long-distance migrants as a function of park area in 48 parks in Portland, Oregon, in 2003. Richness in both (a) and (b) is the residual after accounting for the offset (number of sample points per park), and for long-distance migrants it also reflects the statistical removal of positive influences of TreePC2 on richness. TreePC2 and park area are both standardized to a mean of zero and standard deviation of 1.0. Positive scores on TreePC2 represent an increase in the abundance of small, mostly native, trees

Full size image

Determinants of density. – Two competitive models emerged from our analysis of variation in resident density. A negative quadratic of park area and TreePC2 appeared in both, indicating that the density of residents was greater in small parks and where smaller, mainly native, trees, were abundant (Table 3; Fig. 3a and c). The positive correlation of density with TreePC3 in the top model suggested residents were more abundant where coarse woody debris was common, but TreePC3’s 95% CI overlapped zero (Table 3). Density in both competitive models of long-distance migrant density indicated that density increased steadily with park area (Table 3; Fig. 3b), and was greater in parks with abundant medium and high shrub cover where little English ivy was present (HbShPC4; Fig. 3d; Table 3), and where the ground surface was comprised of more moss or bare soil than plant litter (HbShPC1; Table 3). The appearance of TreePC2 in the top model again suggested that density tended to be higher in parks where smaller, mainly native trees were abundant, but TreePC2’s 95% CI included zero (Table 3). For short-distance/partial migrants, our single competitive model indicated that density was likely greater in parks embedded within a landscape comprised of fewer trees and more residential development (LandPC1; Table 3). However, LandPC1’s 95% CI included zero.

Table 3 Top models (ΔAIC ≤ 2) accounting for variation in average density of birds per point count in 48 parks from Portland, OR, in 2003. Results of best subsets regression analysis reported separately for resident species, long-distance migrants, and short-distance/partial migrants. Models ranked by ΔAICc, with explained variation reported after adjusting for number of variables in the model (R2adj)
Full size table
Fig. 3
figure 3

Variation in density (mean number of individuals of all species/point count/park) for resident species (a and c) and long-distance migrants (b and d). All predictor variables are standardized to a mean of zero and standard deviation of 1.0. Density declined with the quadratic of area for residents (a) but increased with area in long-distance migrants (b). Resident density was also greater in parks where small, mostly native trees (TreePC2) were abundant, while long-distance migrants were denser in parks with abundant medium to high shrub cover (HbShPC4). Statistically significant least squares regression line describing the relationship between density and the predictor variables plotted

Full size image

Forest-dependent species: habitat and minimum area requirements. – The proportion of species that were forest-dependent (i.e., density increased with increasing park area or area2) was roughly twice as great for long-distance migrants (0.54 [7 of 13]) as for residents (0.25 [5 of 20]; X2 = 2.83, df = 1, P = 0.093); short-distance/partial migrants were intermediate (0.38 [3 of 8]).

Multiple species exhibited abrupt increases in density when park area reached 10 ha (Fig. 4), and separate analyses of migratory guilds revealed that the strength of the relationship between density and area differed among them. Although weaker in resident (R2 = 0.512, P < 0.001) than long-distance migrants (R2 = 0.721, P < 0.001), second-order polynomial regression with a positive coefficient for the quadratic term provided the best fit for both, indicating that density tended to exhibit an accelerating increase with area. Density of short-distance/partial migrants, on the other hand, exhibited only a weak increase with the square of park area (r2 = 0.066, P = 0.077).

Fig. 4
figure 4

Representative plots of density (mean number of individuals/point count/park) for (a) the Pacific Wren, a resident species, and three species of long-distance migrants, the (b) Pacific-slope Flycatcher, (c) Swainson’s Thrush, and (d) Wilson’s Warbler. Data were collected in 2003 from 48 parks in Portland, Oregon. Park area is standardized to a mean of zero and standard deviation of 1.0. The dashed vertical line in all four plots is located at an area equal to 10 ha. Statistically significant second-order polynomial plots of abundance versus park area plotted for all species

Full size image

Best subsets regression analysis of density yielded 3, 7, and 6 competitive models for residents, long-distance migrants, and short-distance/partial migrants, respectively (all model results in the Supplementary Materials). Model averaged coefficients for forest-dependent residents confirmed the area affects, but showed also that density varied with all four variables describing tree community structure (Table 4). The mean explained variation (R2adj) of the three models increased to 0.657 (range = 0.644 to 0.672) from 0.512 for area affects alone, suggesting an important contribution of local habitat structure to richness of residents. Density was greater in parks in which large conifers (mainly Douglas-fir) were dominant (TreePC1), small (< 10 cm DBH) and mostly native tree species (TreePC2) and coarse woody debris (TreePC3) were abundant, and where red alder and trees < 30 cm DBH were common (TreePC4; Table 4).

Table 4 Model averaged standardized coefficients (standard error in parentheses) and 85% confidence intervals from analyses of variation in density of forest-dependent breeding birds in 48 parks in Portland, OR, belonging to 1 of 3 migratory guilds. Predictor variables included park area, shape, and connectivity, and axes that describe variation in landscape structure surrounding parks, and variation in the structure and composition of the tree canopy and understory within parks (see Table 1). Separate analyses are reported for residents (n = 5 species), long-distance migrants (n = 7 species), and short-distance/partial migrants (n = 3 species). Predictor variables with 85% CI (see Arnold 2010) that excluded zero are highlighted in bold font. Number in parentheses following migratory guild name indicates the number of competitive models (i.e. ΔAICc ≤ 2.0) resulting from the analysis of each guild, while frequency refers to the number competitive models in which each variable appeared
Full size table

Analysis of forest-dependent, long-distance migrants showed that density was greater in rounder and less elongated parks (i.e. low score for Shape) and where shrubs were abundant 1–3 m above ground and English ivy was an uncommon ground cover (Table 4). The R2adj of the 7 competitive models averaged 0.757 (range = 0.749 to 0.766), an addition of only 0.036 to the variation accounted for by area alone (i.e. 0.721). TreePC2 and HbShPC1 were also found in some of the competitive models of the forest-dependent, long-distance migrants but their 85% confidence intervals included zero (Table 4). For short-distance/partial migrants, adding other predictors to the best subsets regression of density did not substantially raise the explained variation (mean adjusted R2adj = 0.138, range = 0.113 to 0.185) above that due solely to park area (R2adj = 0.066). Greater density was found in parks in which big-leaf maple and trees in the 3rd largest size class (30–60 cm DBH) were abundant, but confidence intervals of two other variables found in competitive models, both measures of shrub density, included zero (Table 4).

MAR calculated using Robbin et al.’s (1989) graphical method could not be determined for the Pine Siskin, but the average area of parks in which they were found was large (35.4 ± 19.23 ha, n = 6). Of the remaining 14 forest-dependent species, MAR averaged 48.9 ± 16.80 ha (95% CI = 12.6 ha to 85.2 ha [Table 5]). MAR did not differ between long-distance migrants and other species (t = 0.70, df = 12, P = 0.499), and increased with body mass (coefficient = 0.785 ± 0.261 SE, r2 = 0.430, df = 12, P = 0.011) in identical manner in long-distance migrants and other species (ANCOVA: Fbody mass = 7.69, P = 0.018, Fmigrant category = 0.00, P = 0.947). The area of the three smallest parks in which each species was detected (Table 5) all correlated with MAR, with the weakest relationship being that with the smallest park (log-log analysis, r = 0.524, df = 12, P = 0.054) and the strongest with the largest of the three (r = 0.830, df = 12, P < 0.001). Thus, while individuals of forest-dependent species occurred in parks below their MAR, the ability of said parks to predict MAR waned steadily as the size of these parks declined.

Table 5 Area of forest cover at which probability of occurrence (P) was maximum and then at 50% of maximum (minimum area requirement) for forest-dependent birds breeding in 2003 in parks and greenspaces in Portland, Oregon. Body mass and area of the three smallest fragments in which each species was detected also given, along with designation as resident (R), long-distance migrant (LD) or short-distance/partial migrant (SDP) as superscript
Full size table

Discussion

The importance of park area

In general, forest park area has a positive influence on species richness of birds in urban settings (Crooks et al. 2001; Husté et al. 2006; Ikin et al. 2013; Myczko et al. 2014; Huang et al. 2015; Kang et al. 2015; Maseko et al. 2020; de Groot et al. 2021). However, and contrary to our expectations (Park and Lee 2000; Husté and Boulinier 2007), area effects were not expressed uniformly across migrant guilds. Species richness of long-distance migrants rose with increasing park area, but then reached an asymptote between 50 and 100 ha, suggesting that parks exceeding 100 ha may yield diminishing returns for increasing richness of long-distance migrants. Richness of neither residents nor short-distance/partial migrants varied with park area. Robbins et al. (1989) documented the same difference between long-distance Nearctic-Neotropical migrants and residents/short-distance migrants breeding in nonurban habitat fragments in forests of eastern North America.

Density of short-distance/partial migrants was also independent of area, while that of resident and long-distance migrants declined and increased with park area, respectively. The latter observation suggests large parks were not only preferred by long-distance migrants, but that quality and quantity of habitat for migrants was greatest in the largest parks. What constitutes high “quality” is not clear, but large parks typically have less edge habitat, which may result in lower rates of nest predation and offspring mortality (Batáry and Báldi 2004; Shipley et al. 2013). Low nest failure rated may be critical for long-distance migrants that have short breeding seasons and limited time to replace failed nests before fall migration (e.g. Mumme et al. 2021). Declining density of residents with increasing park area may be, at least in part, related to preferences for earlier successional habitats and more omnivorous diets that enable them to either directly (i.e. bird feeders) or indirectly exploit supplemental foods provided by humans (Robb et al. 2008; Plummer et al. 2019) in the surrounding landscape.

The absence of an area effect on richness and abundance of short-distance/partial migrant group might be disregarded as a sampling artifact as the number of species involved was small. The same cannot be said for residents given they comprised 20 of the 41 species in the sample. Limiting the number of point counts to a maximum of six per site likely did not lead to an underestimate of richness in large parks because Donnelly (2002) showed that species richness did not increase with more than six points in surveys conducted in urban forest fragments in Seattle, Washington (USA) in habitat similar to Portland. We suggest the failure to detect area effects for residents was partially an artifact of the method of analysis. Number of survey points and park area were highly correlated (r = 0.859, P < 0.001), and inclusion of number of survey points as an offset may have eliminated the possibility of detecting weak to moderate positive influences of area on resident species richness. Indeed, removal of the offset resulted in a positive relationship between species richness and park area (P < 0.001); Donnelly and Marzluff (2004) likewise saw a diminution of area effects when bird species richness of parks was rarefied. However, the fact that the density of residents in our study declined with increasing area suggests that most residents tended to avoid heavily forested habitat, and an area effect on species richness, if it existed, was likely weak. By contrast, that long-distance migrants exhibited greater richness as park area increased (Fig. 1b), despite the offset, emphasizes the importance of park area for migrants that overwinter largely south of the United States.

Landscape versus local habitat effects

Landscape structure (LandPC1), while possibly influencing the density of short-distance/partial migrants, had no relationship with species richness or density of residents, long-distance migrants, or forest-dependent species. The absence of landscape effects contradicts many published studies from urban landscapes (Nielsen et al. 2014; Huang et al. 2015; Reider et al. 2018; Amaya-Espinel et al. 2019), and is likely a consequence of, first, the importance of park area in the Portland system, and second, the city’s generally high tree cover (METRO 2015). The absence of an influence of connectivity on either richness or abundance in our study is not uncommon in birds (Crooks et al. 2001; Donnelly and Marzluff 2006; Radford et al. 2005; Husté et al. 2006; Ikin et al. 2013; Kang et al. 2015; Maseko et al. 2020). In large nonurban landscapes, richness of woodland species declined steadily but slowly until a precipitous drop in richness began when landscape tree cover declined to 10% (Radford et al. 2005)Tremblay and St. Clair (2011) also reported that 20–40% canopy cover proved adequate for successful movement of nonurban adapted forest birds in the urban landscape of Calgary, Alberta. At 30% average tree cover (METRO 2015), Portland’s landscape probably does not present strong barriers to avian dispersal, and this is especially true of the portions of Portland where we worked; landscape canopy cover surrounding the 48 sites averaged nearly 40% out to a distance of 1 km (M. T. Murphy, unpubl. data).

As predicted, the influence of habitat structure and floristic features on species richness and density surpassed that of landscape composition, connectivity, and in most cases, park shape. TreePC2 played a particularly prominent role as it was a primary correlate of species richness for residents and long-distance migrants (in the latter case after removal of area effects), and for the density of residents, in particular the forest-dependent species (and possibly also the full guild of long-distance migrants, Table 3). The positive coefficients in all of these relationships indicated that richness and abundance were greater in parks with abundant smaller, and mostly native, tree species. The importance of TreePC2 may relate to the relatively greater invertebrate prey abundance in younger early seral stage forests where light levels, angiosperm leaf area, and herbivory are greater than in mature conifer stands (Shaw et al. 2006; Campbell and Donato 2014).

By contrast, only forest-dependent residents responded to variation in TreePC1 and TreePC3. The increase in density of forest-dependent residents with TreePC1 and TreePC3 indicated that populations were densest in parks dominated by large conifers in which coarse woody debris was abundant. Habitat descriptions (Rodewald 2022) of the forest-dependent residents in our sample (Table 5) repeatedly note their association with large trees (most often conifers), coarse woody materials, and dense, closed canopies. These features describe the oldest and most mature forests in the Portland system, and represents a mid-successional sere for Pacific Northwest forests (Franklin and Dyrness 1988). The importance of these structural habitat features for species that are year-round occupants is thus perhaps not surprising. On the other hand, TreePC4 described a gradient in the abundance of angiosperm trees in which red alder was replaced by larger big-leaf maple. Alder is abundant in earlier stage successional forests and is commonly associated with riparian zones (Franklin and Dyrness 1988), a habitat that enhances diversity and abundance of numerous taxa (e.g. Sabo et al. 2005).

Unlike the forest-dependent residents, their long-distance migrant counterparts exhibited no association with tree community structure. Instead, park area and shape were the primary factors driving their abundance. Indeed, the only habitat feature that varied consistently with the abundance of long-distance migrant was the density of shrubs 1 to 3 m aboveground (i.e. HbShPC4). With only one major exception (sword fern [Polystichum munitum]), all shrubs in the Portland system were angiosperms, and others have detected a positive association between species richness and/or density of birds with angiosperms (Myczko et al. 2014; Taylor et al. 2016). Greater avian richness and density in urban parks with abundant shrub cover is not uncommon (Donnelly and Marzluff 2004; Myczko et al. 2014; Paker et al. 2014), presumably because shrubs provide foraging substrates for leaf-gleaning insectivores, nesting substrates, and/or cover for shrub and ground nesting birds. High shrub cover along HbShPC4 was also associated with low abundance of invasive English ivy as a ground cover. The importance of HbShPC4 may thus stem not only from the addition of an important habitat feature (i.e. shrubs), but also because of the absence of a noxious invasive species. The absence of any strong relationships between short-distance/partial migrant density and our multiple predictors may reflect the ecological and behavioral variability of this group that, for the most part, represents species that do not fit into the far more discrete categories of resident or long-distance migrant.

In summary, our results suggest that landscape composition had no, or at best a weak, influence on community structure of birds breeding in forest fragments in Portland, and that variation in species richness and density were a consequence primarily of differences in park area (especially for long-distance migrants), but also local habitat structure, and floristic composition (especially for residents). Taylor et al. (2016) reported similar findings from a rural-urban landscape in Michigan (USA); migrant richness was most dependent on the area of forest patches while species richness of resident species was unrelated to forested park area and was instead dependent on the proportion of vegetation that was deciduous. Unlike us, however, Taylor et al. (2016) reported that migrant species richness increased with tree cover in the immediate landscape surrounding the focal patch. Resident and migrant birds thus appear to respond differently to urbanization, but more work is needed from additional urban landscapes to establish the generality of this conclusion (see also Lerman et al. 2021).

Minimum area requirements of forest-dependent species

Multiple species, but mostly long-distance migrants, showed striking increases in density beginning at a park area of ~ 10 ha (Fig. 4). Similar patterns have been described previously for urban greenspaces (Chamberlain et al. 2007 and review by Nielsen et al. 2014). Natuhara and Imai (1999), for instance, noted that insectivores were rarely found in greenspaces below 10 ha, while Rodewald and Bakermans (2006) identified insectivory as an important property of area sensitive species in riparian urban forests. Maseko et al. (2020) noted that, although bird species of all diet categories responded positively to park area, insectivorous birds exhibited the strongest relationship. Most forest-dependent species in our study were also strict insectivores, and thus, smaller parks may not support insect communities productive enough to sustain the most insectivorous bird species. Robbins et al.’s (1989) assertion that 10 ha was a threshold below which few forest-dependent species bred in semi-natural forests from eastern North America suggests that 10 ha is a critical threshold not restricted to urban environments.

Despite the apparent importance of 10 ha as a threshold, it fell below the MAR of most forest-dependent species (Table 5). In addition, density increased steadily with area for many of these species (e.g., Fig. 4). MAR represents the area at which a species’ probability of occurrence is at 50% of maximum for that system, and Robbins et al. (1989) viewed it as a conservative estimate of minimum area needed to sustain breeding populations. All forest-dependent species were found in parks below their MAR (Table 5), but it is important to recognize that the presence of a species in a habitat does not demonstrate that it is a self-sustaining population (e.g., Bartos Smith et al. 2016); local extinction is a likely fate of forest-dependent species in these locations (e.g., Husté and Boulinier 2007). MAR of forest-dependent species averaged nearly five times larger than 10 ha, and although MAR increased with body size, even some of the small species had MARs in excess of 10 ha (Table 5). Radford et al. (2005) emphasized that thresholds are “points of instability” below which systems are likely to fail rapidly and therefore minimum thresholds should be avoided. Richmond et al. (2015) further stressed that thresholds may vary with landscape structure. However, the identification of 10 ha as a critical size for forest-dependent species from different locations and habitats (Robbins et al. 1989; Nielsen et al. 2014, present study) suggests a true threshold below which declines of forest-dependent species are inevitable. Given this, species with larger MARs should be the drivers of policy if the goal is to maintain intact communities of birds in forested urban parks.

Summary and recommendations

Avian communities within forest fragments in Portland, Oregon, were dominated by resident species, but among the largest parks, richness of residents and long-distance migrants were roughly equal. If management goals are to restore the landscape’s avifauna, or more realistically, prevent further erosion, we make several recommendations: (1) Expansion of the park system should prioritize acquisition of the largest forest fragments possible. Forest-dependent species, many of which are long-distance migrants, do not have meaningful presence in parks below 10 ha and most forest-dependent species will likely disappear from the landscape unless multiple parks of a minimum size of 30 to 40 ha, and possibly 50 ha, are retained. Donnelly and Marzluff (2004) arrived at a similar figure (42 ha) for nearby (280 km) Seattle, WA. (2) Within-habitat features of parks were important secondary contributors to avian species richness and density and thus parks must remain undeveloped to maintain complex vegetation structure. (3) Floristics were also important as, with the exception of short-distance/partial migrants, richness and/or density of species were higher in forests where there was high diversity of mainly small native tree species (TreePC2). (4) Mature coniferous forests are associated with the Pacific Northwest region (Franklin and Dyrness 1988) and our closest approximation to these forests yielded the highest density of forest-dependent, resident species. Providing for such forests is thus essential, but maintenance of forests at diverse seral stages is desirable to provide for species dependent on earlier successional stages. (5) While park area was the dominant predictor of density of forest-dependent resident and long-distance migrant species, park shape proved important as well for the forest-dependent, long-distance migrants, suggesting that long and thin parks should be avoided. And finally, (6) habitat connectivity and landscape composition had the least influence on density and species richness of birds, but this was likely because Portland’s landscape has abundant tree cover. In this system, maintenance of high landscape tree cover is likely essential for maintenance of high species richness and density.