Degree of adaptive response in urban tolerant birds shows influence of habitat-of-origin

Study area and data handling

The study area is metropolitan Melbourne; capital city of the State of Victoria in coastal southeastern Australia, within a 50 km radius of its Central Business District (Fig. S1) (37°49′S and 144°58′E).

Approximately 220,000 records of birds were extracted from the Birds Australia ‘New Atlas of Australian Birds’ database (Barrett et al., 2003), and intersected with a 1 × 1 km grid (Hahs & McDonnell, 2006) to produce a matrix of grid cells by species presence/absence. Species and sites were filtered out according to criteria for representativeness (see Supplemental Text) to arrive at a final list of 141 species and 390 cells (Conole & Kirkpatrick, 2011).

Environmental and demographic indices

Spatial data on the degree of urbanisation of the study area employed in this study were developed at ARCUE and are discussed in detail by Hahs & McDonnell (2006); a brief summary of the two selected factors follows.

Frequency Greenspace (hereafter greenspace) is the reciprocal of the average amount of impervious surface calculated at the sub-pixel level from the impervious surface fraction image created during the spectral mixture analysis of the 2000 Landsat ETM + image (Hahs & McDonnell, 2006).

Combined index (Index_Combined) is the average value of Index_Image and Index_Census; where Index_Image is calculated from fraction images produced by the spectral mixture analysis of the 2000 Landsat ETM + image, and Index_Census = the total number of people multiplied by the proportion of males employed in non-agricultural work, as enumerated in the 2001 Australian census (Hahs & McDonnell, 2006).

Other environmental factors considered in analyses included PC_ URB (percent cover of urban landform), People per square kilometre (People/km²—the total number of people in census collection districts) and Dwellings per square kilometre (Dwellings/km²—the total number of houses in census collection districts) (Hahs & McDonnell, 2006; Conole & Kirkpatrick, 2011).

Data analysis

Statistical analyses were performed in r (R Core Team, 2013) using base r functions and procedures from the r-packages ‘vegan’ (Oksanen et al., 2013) and ‘bayespref’ (Fordyce et al., 2011). Figures were drawn using r base graphics, r-packages ‘vegan’ and ‘ggplot2’ (Wickham, 2009; Oksanen et al., 2013), and QGIS (QGIS Development Team, 2013).

An earlier assemblage analysis (Conole & Kirkpatrick, 2011) was the basis for partitioning the total bird datasets for this study; detailed methodology is described therein. Adapter and exploiter species were further partitioned into two new matrices for this study, and separate non-metric multidimensional scaling (NMDS) ordinations performed for each (see Supplemental r Script #1). Only factors for which p ≤ 0.01 were considered further in analyses, and where a choice between the overlapping PC_ URB and Index_Combined factors was required, the recommendation of Hahs & McDonnell (2006) for Index_Combined was adopted.

Boxplots of species richness of the two urban tolerant assemblages were made, binned by an index of urbanisation intensity (Index_Combined—hereafter urbanisation index) and cover of vegetation (greenspace) (see Supplemental r Script #2).

Species richness of exploiter and adapter species was enumerated for each of 390 grid cells (Conole & Kirkpatrick, 2011), along with the index of urbanisation intensity and cover of vegetation. Data were then modelled as hierarchical Bayesian models using r-package ‘bayespref’ (Fordyce et al., 2011) to test the preferences of exploiters and adapters for partitioned urban habitats. Model parameters were estimated using a Markov Chain Monte Carlo (MCMC) approach, with 10,000 MCMC steps following a burn-in of 1,000 generations. The parameters estimated in this way are intended to directly address the hypothesis (Fordyce et al., 2011), namely that adapter and exploiter bird assemblages show preferences for urban habitat characterised by differing levels of urbanisation intensity or vegetation cover. The hierarchical Bayesian approach has the advantage of directly estimating the parameter of interest (in this case preference for levels of urbanisation or green space by urban tolerant bird assemblages), and models the uncertainty around those parameters as well as allowing comparisons between a priori identified groups, in contrast to methods such as ANOVA or t-tests, which assess whether the mean difference is different from zero (Fordyce et al., 2011). The estimates are population-level preferences (Fordyce et al., 2011).

Within ‘bayespref’ a facility for assessing model convergence (indicated by MCMChain mixing) by plotting MCMC steps against population level preferences is available (Fordyce et al., 2011). A well-mixed chain is one characterised by a broad scatter of data points in the scatterplot without obvious clumping (Figure SR4 in Supplemental r Script #3), whereas clumping of data points indicates poorly-mixed chains. Although a subjective visual measure, it is sufficient to identify satisfactory MCMChain mixing, and this method was used here to determine when satisfactory model convergence had been achieved.

Proposal distance in the MCMC is set by the ‘bayespref’ switch ‘dirvar’; usually at the default setting of 2. Runs of ‘bayespref’ with a ‘dirvar’ value of 2, 5, 10 and 20 were executed, to determine whether optimal mixing of the MCMChains influenced the overall trends in habitat preference (see Supplemental r Script #3), but the gross trends were unchanged. Nonetheless, results cited in this paper use the highest tested proposal distance (‘dirvar’ = 20) to ensure thoroughly mixed MCMC chains.

Outputs from the ‘bayespref’ analysis were plotted, with base r functions, as binned median preference with 95% confidence intervals (see Supplemental r Script #4).

Adapter and exploiter species’ habitats-of-origin were determined by reference to the literature (Marchant & Higgins, 1993; Higgins & Davies, 1996; Higgins, 1999; Schodde & Mason, 1999; Higgins, Peter & Steele, 2001; Higgins & Peter, 2002; Higgins, Peter & Cowling, 2006), and shown in Table 1. Habitat-of-origin is used here to mean the primary natural (pre-urbanisation) habitats that species are known to have occupied. The data for cluster analysis consisted of a standard array, with species as rows and habitat-of-origin as columns (forest, woodland, heath, scrub, urban, farm, air). A Bray-Curtis distance matrix was prepared, and groups of species were formed by hierarchical agglomerative clustering using Ward’s algorithm performed on the distance matrix, using core r-function ‘hclust’ (R Core Team, 2013) (see Supplemental r Script #5).

Results

In an earlier ordination of all bird species from the Melbourne study, urban exploiters and adapters are shown as overlapping but distinct clusters in ordination space (Fig. S2) (Conole & Kirkpatrick, 2011). When the exploiters and adapters were partitioned from the avoiders and run as separate ordinations, different pictures of response to urban environmental factors became apparent (Figs. 1 and 2).

For exploiters the observed species richness vector (S_obs) was orthogonal with both greenspace and the urbanisation index (Fig. 2). The equivalent vector for adapters (Fig. 1) was orthogonal with the urbanisation index, but almost aligned with that for greenspace (Fig. 2). Greenspace and the urbanisation index were chosen as representative of structural and demographic aspects of urbanisation intensity even though other parameters were included in the initial analyses, and further analyses were limited to these two factors.

The same data plotted as binned boxplots showed that adapter species richness was positively associated with increasing greenspace, but exploiter species richness was flat across the range (Fig. 3). Whilst broadly similar trends were evident for both groups as binned boxplots plotted against the urbanisation index (Fig. 4), adapters trended to zero species richness at the highest levels, whilst 10–15 species of exploiters persisted at the same level. Peak species diversity of urban adapter birds occurred in the middle of the range of urbanisation intensity (Fig. 4). Adapter richness peaked at approximately 0.8 frequency green-space; exploiters at around 0.55 (Fig. 3).

The hierarchical Bayesian models for greenspace showed a relatively flat preference by urban exploiters across the range; though increasing preference by urban adapters for higher levels of greenspace (median = 0.46; credible intervals 0.424–0.494) almost match exploiter preference (0.54; 0.506–0.576) in the highest bin (Fig. 5; Table S1). Even allowing for high levels of variance in the lower bins where data points were more scarce, the preferences of urban exploiters and adapters did not overlap in any of the greenspace bins.

Hierarchical Bayesian models for the combined index showed a joint preference by urban adapters and exploiters in the middle of the range of the urbanisation index (20.0–29.9). Areas of low (0–19.9) and high (30.0–50.0) urbanisation index were strongly preferred by urban exploiters but not adapters (Fig. 6; Table S2).

The cluster analysis of adapters and exploiters by habitat of origin returned a dendrogram showing two clear major clusters. All of the adapters clustered together in a woody vegetation habitat group, along with a group of exploiters; five indigenous nectarivores (Red Wattlebird Anthochaera carunculata (Shaw, 1790), Little Wattlebird A. chrysoptera (Latham, 1802), White-plumed Honeyeater Lichenostomus penicillatus (Gould, 1837), Musk Lorikeet Glossopsitta concinna (Shaw, 1791), Little Lorikeet G. pusilla (Shaw, 1970)), two indigenous avivorous raptors (Australian Hobby Falco longipennis Swainson, 1837, Brown Goshawk Accipiter fasciatus (Vigors and Horsfield, 1827)) and two exotic species which are not exclusively synanthropic (Common Blackbird Turdus merula, Linnaeus, 1758, Common Starling Sturnus vulgaris Linnaeus, 1758) (Conole, 2011). The cluster of exclusively exploiter species were characterised by those originating from open grassy or urban habitats.

The boxplots (Figs. 3 and 4) and the hierarchical Bayesian models (Figs. 5 and 6) showed clear but distinct trends of urban habitat preference by urban exploiter and adapter bird assemblages against these two representative urban habitat measures. The landscape scale preferences of urban adapters and urban exploiters for levels of greenspace never overlap, though they come close to each other at the highest values as exploiter preference declines and adapter preference increases. In contrast, landscape preferences for urbanisation intensity measured by the urbanisation index overlap strongly in the middle of the range but are strongly divergent at the lowest and highest values.

Discussion

The diversity of urban adapters on the gradient of urban intensity follows a humped distribution (Fig. 4); the trend even more strongly humped when viewed as landscape scale preference (Fig. 6). This is consistent with the trend seen for urban tolerant birds in other studies (Marzluff & Rodewald, 2008; Tratalos et al., 2007), and for bird species richness in response to several environmental factors at a landscape scale (Bar-Massada et al., 2012). The inverted, humped curve for exploiters is not consistent with the trends for urban tolerant bird species richness seen in other studies (Tratalos et al., 2007; Luck & Smallbone, 2010; Shanahan et al., 2014), and this marks a strong divergence in response by exploiters and adapters to urbanisation intensity.

This quadratic trend in diversity also resembles that described by the Intermediate Disturbance Hypothesis (IDH), where diversity peaks at a midpoint along a gradient of disturbance (Catford et al., 2012; Fox, 2013). The urban–rural gradient is, however, not a true analogue of a disturbance gradient. Suburban areas are more stable habitats than either the developing fringe or the intensely re-shaped core of the city, and so disturbance itself shows a quadratic distribution along the urban–rural gradient. Also implicit within IDH is a notion of competition/colonisation trade-off amongst species more or less adapted to disturbed environments, and at least for urban adapted birds it has been suggested that competition is not important (Mikami & Nagata, 2013) except for specific cases such as the ‘despotic’ Noisy Miner (Manorina melanocephala) (Kath, Maron & Dunn, 2009; Maron et al., 2013; Robertson et al., 2013).

The zone of overlap in habitat preference along the human demographic gradient accords broadly with the inner ring of suburbs in Melbourne; long established and heavily vegetated (Hahs & McDonnell, 2006). At the extremes of this gradient lie the new suburbs/exurbia at the fringe, and the central business districts (‘down town’) at various central locations—either lightly vegetated or with largely treeless vegetation (lawns and pasture) (Hahs & McDonnell, 2006). The overlap represents depressed preference by exploiters coincident with greatest preference shown by adapters.

The response of urban tolerant birds to increasing Frequency Greenspace is consistent with wider trends in other cities (Chace & Walsh, 2006), and closely mirrors the relationship observed between bird species richness and foliage height diversity observed in a non-urban landscape (Bar-Massada & Wood, 2014). Increasing foliage height diversity is a marker of established suburbs versus the developing fringe in Melbourne (Hahs & McDonnell, 2006). The distinct responses between adapters and exploiters is also less marked with respect to greenspace than urbanisation intensity.

The responses of the two assemblages to two simple measures of urban habitat character were divergent, consistent with the study’s main hypothesis. Though the larger group of urban tolerant bird species may occasionally be treated as one entity, it is clear from this study and others (Croci, Butet & Clergeau, 2008; Catterall, 2009; Conole, 2011; Conole & Kirkpatrick, 2011) that the two groups within it are sufficiently distinct in their responses to urbanisation to caution against using pooled data for urban tolerant species in future studies.

The response of urban adapter species to the urbanisation index is consistent with what we broadly understand them to be; adapted to suburbanisation (Blair & Johnson, 2008). Greenspace typically increases in old suburbs versus the exurban fringe or downtown areas (Hahs & McDonnell, 2006). The strong depression in exploiter preference for mid-range urbanisation intensity (versus the extremes) is less expected. At least with the Melbourne data, there is not a single generalised urban tolerant group of birds. The adapters and exploiters share ecological traits with each other but also with avoiders (Conole & Kirkpatrick, 2011).

In part the contemporary avifauna of an urbanised area is a legacy of the species present in the former landscape, rather than solely being the product of invasion or colonisation (sensu Møller et al., 2012). As urban areas progressively come to resemble woodland, structurally if not floristically (Kirkpatrick, Daniels & Zagorski, 2007), it makes sense that the urban tolerant bird species are likely to include legacy woodland-adapted species. Despite the findings of Blair & Johnson (2008) in North American urban areas, it does not appear that suburban areas within a previously forested landscape in Melbourne are loci for indigenous woodland bird extirpation or exotic bird invasion (Conole & Kirkpatrick, 2011). Instead the reverse seems to be true. They are sites for colonisation and expansion of some indigenous woodland birds (adapters) and places where exotic exploiters are less abundant.

Exploiters are mostly indigenous species derived from open environments such as grassland and grassy open-woodland (Møller et al., 2012), with a small cohort of synanthropic exotic species and indigenous dietary specialists (avivorous raptors, nectarivores) more typical of forest/woodland habitats (Table 1; Fig. 7) (Conole & Kirkpatrick, 2011). Adapters as a group are all indigenous species of forest, woodland and riparian scrub origins (Table 1; Fig. 7), and they have closer affinities with the riparian and bush remnant urban avoiders than the exploiters (Conole & Kirkpatrick, 2011). It is therefore remnants of the former indigenous avifauna of wooded parts of Melbourne that are the source of the emerging group of urban adapted species, though none are yet as successful as the aptly named urban exploiters. The adapters are essentially the vanguard of a group of semi-specialised bird species that utilise particular niches of greater foliage height diversity within urban matrix habitats, but are not yet ubiquitous across the matrix in the way of exploiters.

The responses observed here of each group to both degree of urbanisation and greenspace are largely explained by their ecological histories. The exploiters are able to use disturbed habitats across the matrix analogous to their original habitats, and many of them were established in Melbourne during the early stages of urban expansion and consolidation of the city. As suburban parts of the city became more heavily vegetated and less open, a group of species from analogous riparian/forest habitats became increasingly well established in parts of the city proximate to their source natural habitats. Many parts of the urban matrix are now at or close to the point of saturation with members of the exploiter assemblage due to their ubiquity, but the number of adapter species contributing to bird species richness at points across the matrix is likely to increase on a site by site basis as the process of afforestation of the older suburbs continues. It follows then that the distribution of exploiter species may decline in more established suburban parts of the city over time, though expanding in range and continuing to dominate in developing areas of the city at or near the fringe. Clues to this trend can be found in studies that model the trajectory of abundance for open habitat, ground feeding specialists (such as the Crested Pigeon Ocyphaps lophotes GR Gray, 1842) declining as those habitats become denser with woody vegetation (Kutt & Martin, 2010).