Environmental species sorting dominates forest-bird community assembly across scales


  • Korhan Özkan,

    Corresponding author
    1. Freshwater Ecology Group, Department of Bioscience, Aarhus University, Silkeborg, Denmark
    2. Ecoinformatics and Biodiversity Group, Department of Bioscience, Aarhus University, Aarhus, Denmark
    Search for more papers by this author
  • Jens-Christian Svenning,

    1. Ecoinformatics and Biodiversity Group, Department of Bioscience, Aarhus University, Aarhus, Denmark
    Search for more papers by this author
  • Erik Jeppesen

    1. Freshwater Ecology Group, Department of Bioscience, Aarhus University, Silkeborg, Denmark
    2. Greenland Climate Research Centre, Greenland Institute of Natural Resources, Nuuk, Greenland
    3. Sino-Danish Centre for Education and Research, Beijing, China
    Search for more papers by this author


  1. Environmental species sorting and dispersal are seen as key factors in community assembly, but their relative importance and scale dependence remain uncertain, as the extent to which communities are consistently assembled throughout their biomes.
  2. To address these issues, we analysed bird metacommunity structure in a 1200-km2 forested landscape (Istranca Forests) in Turkish Thrace at the margin of the Western Palaearctic (WP) temperate-forest biome. First, we used spatial regressions and Mantel tests to assess the relative importance of environmental and spatial factors as drivers of local species richness and composition within the metacommunity. Second, we analysed species' abundance–occupancy relationship across the metacommunity and used null models to assess whether occupancy is determined by species' environmental niches. Third, we used generalized linear models to test for links between species' metacommunity-wide occupancy and their broader WP regional populations and assessed whether these links are consistent with environmental species sorting.
  3. There was strong environmental control on local species richness and composition patterns within the metacommunity, but non-environmental spatial factors had also an important joint role.
  4. Null model analyses on randomized communities showed that species' occupancy across the metacommunity was strongly determined by species' environmental niches, with occupancy being related to niche position marginality.
  5. Species' metacommunity-wide occupancy correlated with their local abundance as well as with their range size and total abundance for the whole WP, suggesting that the same assembly mechanisms act consistently across local to regional scales. A species specialization index that was estimated by bird species' habitat use across France, incorporating both niche position and breadth, was significantly related to species' occupancy and abundance at both metacommunity and WP regional scales. Hence, the same niche-related assembly mechanisms appear to act consistently across the WP region.
  6. Overall, our results suggest that the structure of the Istranca Forest' bird metacommunity was predominantly controlled by environmental species sorting in a manner consistent with the broader WP region. However, variability in local community structure was also linked to purely spatial factors, albeit more weakly.


Niche processes have been recognized as important for community assembly for nearly a century, with emphasis on both abiotic and biotic factors (Grinnell 1917; Hutchinson 1959; Gotelli, Graves & Rahbek 2010). However, the potential importance of dispersal has received increasing recognition since MacArthur & Wilson's (1967) seminal work on island biogeography. Controversially, Hubbell's (2001) neutral theory emphasizes dispersal as a key control of community structure, with species' niche differences being unimportant. Recently, niche and dispersal processes have been integrated in a framework, in which local communities are interconnected by dispersal as parts of a wider metacommunity (Leibold et al. 2004). Under this perspective, the relative importance of niche and dispersal processes may vary from pure dominance by niche assembly (environmental species sorting) over intermediate situations to complete dominance by dispersal assembly (neutral model). With greater importance of dispersal, speciation, extinction and colonization across large geographical regions may also influence metacommunity structure (Ricklefs 1987; Zobel 1997; Griffiths 1999; White & Hurlbert 2010). Ricklefs (2011) argued that local communities are open assemblages made up of the overlapping regional populations of their constituent species. Regional effects on communities may alternatively reflect niche-driven processes across large ecologically coherent regions under this perspective (also cf. Belmaker & Jetz 2012).

Considering bird community assembly, vegetation structure and habitat heterogeneity have emerged as important environmental factors (Thiollay 1992; Lampila, Mönkkönen & Desrochers 2005; Meynard & Quinn 2008). In contrast, less evidence exists for dispersal effects in bird community assembly (Meynard & Quinn 2008; Driscoll & Lindenmayer 2009). An analysis of bird communities across France provided support for the joint influence of environmental factors and dispersal (Meynard et al. 2011), while a study of bird communities across North America provided evidence for a moderate regional pool enrichment effect, consistent with the influence of dispersal, even though local environment was the dominant assembly factor (White & Hurlbert 2010). In addition, there is evidence that biotic interactions – even beyond those driven by environmental factors – may also affect bird community assembly, such as interactions within and among bird species (Gotelli, Graves & Rahbek 2010) as well as trophic and non-trophic interactions with other organism groups, particularly plants (Kissling, Rahbek & Böhning-Gaese 2007; Qian & Kissling 2010). Presently, however, we still lack a comprehensive understanding of the factors driving bird community assembly and how these link from local assemblages to regional communities across whole biomes (Ricklefs 2011).

Here, we investigated forest-bird metacommunity structure and its underlying drivers across a 1200-km2 forested landscape at the geographical margin of the Western Palaearctic (WP, Europe, including Caucasus up to west Urals) temperate-forest biome. We assessed community structure and its potential drivers at three spatial scales, namely local communities, the metacommunity as a whole and the whole WP region. First, to understand drivers of local community assembly within the metacommunity, we analysed local species richness and composition. Second, to assess metacommunity scale assembly, we assessed whether species' occupancy across the metacommunity is linked to their niche breadth and/or niche position, as expected under niche assembly (Brown 1984; Hanski, Kouki & Halkka 1993). Third, we assessed whether occupancy across the metacommunity is related to species' range and abundance at the WP regional scale, which would indicate that consistent regional processes drive community assembly and tested whether these patterns are consistent with the environmental species sorting.

Materials and methods

Study Area

The study was conducted in Istranca (Yıldız) Mountains (Fig. 1) and spanned c. 1200 km2 including the highest point (1031 m) in Turkish Thrace. Average annual precipitation is 542 mm and average annual temperature is 13 °C (Turkish State Meteorological Service, unpubl. data). The vegetation is primarily temperate deciduous forest, dominated by Quercus and Fagus. The forest is managed for timber production and most stands are homogeneous in age and species composition. Tree cover is fragmented by small openings, agricultural areas and villages.

Figure 1.

Location of the survey area (Istranca Forests, Turkish Thrace) overlaid with the bird survey sites included in the present analysis. Hill-shade effect represents the terrain.

Breeding Bird Surveys

Bird counts were conducted between 30 April and 6 July 2009 by four surveyors in two teams. Survey locations were selected using a stratified random design with at least 200-m distance from the habitat edge and with necessary logistic modifications. At each locality, sampling was conducted at three consecutive sites along a predetermined direction with an average inter-site distance of 308 m (Fig. 1), using circular plots modified after Bibby (2004). All birds identified by sight or sound during the 10-min survey period were recorded in four distance bands (≤10, 11–20, 21–100, >100 m). Morning counts were conducted within the first 6 h after dawn, and sites were also visited after 10 p.m. for night counts of nocturnal bird species. Only sampling sites in forests (natural forests, including riparian forests, as well as plantations) were included in the present analyses (317 sites, 272 of which were also visited for night counts). Global average abundances for night birds were used for sites missing night counts. Birds flying above the canopy or exhibiting migratory behaviour, as well as Cuculus canorus, which migrate early in the breeding season, were excluded from the analysis.

Environmental Variables

Environmental variables were analysed as two groups: (i) on-site habitat characteristics and (ii) landscape characteristics surrounding each site. On-site habitat characteristics (dominant tree genera, tree genus richness, stand age, canopy cover and undergrowth characteristics) were recorded by visual observations at each site within c. 100 m. Three canopy cover (10–40–70–100% cover) and five stand age (0–8–16–25–50->50 cm, diameter at breast height) classes were applied. If stands held mixed age groups of two consecutive age classes, a composite category was used (nine classes in total). Undergrowth was characterized by total cover and by the presence of logs-and-snags and Rhododendron ponticum. Pairwise correlations were small among on-site environmental variables (r < 0·34).

Landscape characteristics (altitudinal range, habitat richness, dominant tree genus richness, total forest cover, dominant stand age, average stand age and stand age class richness) were analysed in four different distance bands (250–500–1000–3000 m) around each site to avoid an arbitrary range selection. Landscape characteristics were computed in ArcMap (ESRI 2008) using National Forestry Registry Data base (Ministry of Environment and Forest, unpubl. data) and a global digital elevation model (GDEM, ASTER GDEM is a product of METI and NASA). Landscape characteristics were reduced to orthogonal factors by an exploratory factor analysis with varimax rotation (Legendre & Legendre 1998). Six factors were selected by a heuristic approach to achieve the best grouping of related environmental predictors. On-site altitude was correlated with altitudinal range and included in the Factor Analysis. Interpretation of the factor analysis is shown in Table 1, with further details in Appendix S2. The correlations among on-site and landscape characteristics were relatively low (r = 0·59 for on-site stand age and factor for stand age in the landscape; r < 0·34 for the remaining).

Table 1. Interpretation of the factor analysis of the landscape characteristics around each sampling site in Istranca Forests, Turkey. For details see Appendix S2
Stand agePositively correlated with stand age
Habitat diversityPositively correlated with habitat diversity and negatively correlated with forest cover
Tree genus richnessPositively correlated with dominant tree genus richness
Stand age diversityPositively correlated with stand age class diversity
AltitudePositively correlated with altitude and range of altitude
Canopy cover diversity

Positively correlated with canopy cover


Statistical Analyses

To assess community structure and its potential drivers at the three study scales, we first analysed local species richness and composition within the Istranca Forests' metacommunity by using spatial regressions with variance partitioning and Mantel tests, respectively. We then assessed species' occupancy–abundance relationships for the metacommunity as a whole using generalized linear models (GLM) and used null models to analyse whether occupancy is related to species' environmental niches. Niches were quantified as niche breadth (Brown 1984), representing environmental tolerance, and niche position marginality, representing how marginal a species' habitat use is in relation to habitat availability (Hanski, Kouki & Halkka 1993). Third, we analysed whether species' occupancy is related to their WP regional range and population size, which would indicate consistent regional processes driving community assembly. Finally, we used the species specialization index (SSI) from the French breeding bird survey (Julliard et al. 2006) to test whether this measure of species specialization from a distant part of the WP biome explains species occupancy across Istranca Forests as well as species' population and range sizes across the WP. A strong relationship would provide evidence for environmental species sorting acting consistently across the whole region.

Variability In Local Community Structure

We analysed local species richness in relation to on-site and landscape characteristics using ordinary least squares (OLS) regression to assess how much variation is accounted for by the environmental factors. OLS models were simplified by stepwise Akaike Information Criterion (AIC) selection (Venables & Ripley 2002). Subsequently, we refit the final OLS models using a simultaneous autoregressive (SAR) model to assess the importance of spatial structure. SAR model accounts for spatial autocorrelation as an error term (Anselin 1988) and has performed well in simulation studies (Dormann et al. 2007; Kissling & Carl 2008). We also fit a null SAR model excluding all environmental variables to assess the effect of the spatial structure alone. We compared the OLS and SAR model fits to evaluate the relative roles of environmental and spatial factors. Species richness was square-root transformed because of slight deviation from normality. All predictors were standardized to zero mean and unit standard deviation (Gelman 2008). Dominant tree genus as a categorical predictor was coded into binary variables with a contrast to Quercus, the dominant tree genus. Tree genus richness and canopy cover diversity in the landscape, as well as tree genus richness and all undergrowth-related variables on-site, were excluded by AIC selection. SAR model was constructed with a 3-km neighbourhood matrix and its fit was assessed with pseudo-R2 (Nagelkerke 1991). All regressions were examined with diagnostics plots and standard methods; no violations of assumptions were found. Variance inflation factor was ≤1·41, while the maximum difference in betas and fits were 0·76 and 1·1, respectively.

Local community species composition was analysed using partial Mantel tests (Legendre & Legendre 1998) to examine the relationship between community dissimilarity, geographical distance and environmental dissimilarity between each pair of sampling sites, where geographical distance represents dispersal or other non-environmental spatial processes (Svenning & Skov 2002). Species-site matrix, geographical coordinates and the environmental variables were converted into pairwise Bray–Curtis dissimilarity, log-transformed Euclidean distance and Gower dissimilarity matrices (Gower 1971), respectively, before Mantel tests. We used forward selection on environmental variables for Mantel Tests. Environmental variables were added into the pairwise-dissimilarity calculations until there was no increase in the correlation or the test became insignificant. Stand age diversity and tree genus richness in landscape as well as dominant tree and presence of R. ponticum and logs-and-snags on-site were excluded.

Metacommunity-Wide Occupancy and Abundance

We assessed whether species' occupancy across the Istranca Forests' metacommunity is related to local abundance. Local abundance was quantified as maximum and mean abundance in occupied plots. Species were classified into five habitat association groups (Appendix S1) following Cramp & Simmons (2006) in order to examine the relationships exclusively for forest-associated birds. We used negative binomial GLM with log-link to model occupancy because of overdispersion. Colonial breeders, post-breeding foraging groups and few species with high leverage (Cook's distance >1) were excluded from the analyses.

We assessed whether species' occupancy across metacommunity is determined by species' niche breadth and niche position marginality. Ten environmental variables selected in richness or community composition analyses were used as species' niche axes. Niche breadth of a species across an environmental variable was quantified as the range of an environmental variable at occupied sites divided by overall range of this environmental variable across all sites, so that each niche breadth was scaled between zero and one (i.e. niche breadth relative to available variation on each environmental axis). Mean range over ten environmental axes was calculated as niche breadth of each species. Niche position marginality was quantified as Euclidean distance in 10-dimensional environmental space between the means of environmental variables over all sites and over the occupied sites, that is, as niche position marginality relative to niche availability. Each niche position marginality axis was scaled by dividing with the maximum possible distance for that environmental variable. Quantification of niche breadth and position on the categorical variable, dominant tree genus, was performed differently. Niche breadth was calculated as the number of different tree genera occupied by each bird species, and niche position was calculated as the Euclidean distance between tree genera composition over all sites and occupied sites for each species. As both niche breadth and position measures are mathematically constrained at lower sample sizes, only species with more than 10% occupancy (>32 sites) were included in the analysis. To assess the significance of the relationships between species niche traits and occupancy, we used null model analysis. We randomized the metacommunity over the landscape 1000 times with replacement, that is, species were randomly assigned to sampling sites, while their occupancy over the landscape was kept constant to test whether the observed relationship between species' niche traits and occupancy differs from random expectations. Niche breadth and positions were recalculated for each randomization.

Link To The Western Palearctic Region

We assessed whether species' occupancy across the Istranca Forests' metacommunity is related to their total breeding population (geometric means of minimum and maximum estimated population) and range sizes across WP (BirdLife International 2004), again using negative binomial GLM with log-link. Colonial breeders, post-breeding foraging groups and few species with high leverage (Cook's distance >1) were excluded from the analyses. We assessed whether any Istranca Forests-WP link reflects environmental species sorting by testing whether SSI explains species' occupancy across the Istranca Forests as well as their population and range sizes across the WP. SSI was calculated as variance of average densities among different habitat classes analysed across France (Julliard et al. 2006) and acts as a composite measure of niche breadth and position marginality. We used negative binomial GLM with log-link to model occupancy and gamma GLM with log-link to model WP population and range sizes. Species missing SSI data and Dendrocopos medius with high leverage (Cook's distance >1) were excluded from the analyses.

Data handling and statistical analysis were performed using R statistical software (R Development Core Team 2011) using vegan package (Oksanen et al. 2011) for Gower dissimilarity and Mantel test, MASS package (Venables & Ripley 2002) for GLM, spdep (Bivand 2011) and ncf packages (Bjornstad 2009) for SAR.


A total of 67 breeding bird species were recorded in 317 sites across Istranca Forests (Appendix S1). Median species richness per sampling site was 8, ranging from 2 to 19.

Variability In Local Community Structure

Environmental and non-environmental spatial factors explained 28% of the variation in species richness (Table 2). Variance partitioning revealed that 19% and 4% of the variation in local community richness were uniquely attributable to environmental variables and spatial factors, respectively, and 5% was shared. Environmental effects were mediated by both local habitat and landscape characteristics, with dominant tree genus having the strongest role (Table 2).

Table 2. Final reduced spatial (SAR) and non-spatial (OLS) models for local bird species richness in Istranca Forests. Standardized regression coefficients, model R2 and associated P values are given. R2 for null model with no environmental variables were given in parenthesis for SAR model. Significant values are shown in bold (P < 0·05). F denotes factors from factor analysis (see Table 1) and DT denotes dominant tree genus in stands
 Bird species richness
R 2 0·278 (0·093) 0·240
P <0·001 <0·001
 Coef. P Coef. P
F-stand age−0·0830·276−0·1060·110
F-habitat diversity0·0610·365 0·126 0·044
F-stand age diversity 0·146 0·012 0·145 0·007
F-altitude0·150 0·026 0·154 0·007
Canopy cover on-site0·182 0·001 0·209 <0·001
Stand age on-site 0·209 0·004 0·204 0·004
DT - Alnus0·0470·7030·0620·610
DT - Fraxinus 0·491 0·002 0·495 0·003
DT - Fagus−0·0270·697−0·0270·693
DT - Carpinus−0·1370·318−0·0390·780
DT - Pinus0·521 0·001 0·473 0·002
DT - Populus0·570 0·001 0·534 0·002

Dissimilarity in local community species composition correlated with environmental dissimilarity (r = 0·26, < 0·001) and less strongly with geographical distance (r = 0·08, < 0·001) in partial Mantel tests, after the effect of one was corrected for the other.

Metacommunity-Wide Occupancy and Abundance

The studied metacommunity was dominated by a suite of widespread, locally abundant species. Species' occupancy was positively related to maximum abundance (math formula = 0·36, P < 0·001; Fig. 2), but not significantly related to mean abundance (P = 0·29). When data were restricted to forest-associated birds, the relationship with maximum abundance became stronger (math formula = 0·39, P < 0·001) and the relationship with mean abundance became significant and moderate (math formula = 0·19, P < 0·001). Species' occupancy across the Istranca Forests' bird metacommunity increased with niche breadth (r2 = 0·58) and decreased with niche position marginality (r2 = 0·53; Fig. 3). However, only the relationship of niche position marginality differed significantly from randomized communities (P < 0·001), while of niche breadth did not (P = 0·110).

Figure 2.

Bird species occupancy vs. maximum (a) and mean abundance at occupied sites within the Istranca Forests' metacommunity (b), and their total population (c) and breeding range size (d) across the Western Palaearctic (WP) region. Forest-associated species (Appendix S1) are indicated by crosses and the remaining by circles. Generalized linear model regression lines are given for all species (solid) and separately for forest-associated species (dotted). Models fits as deviance ratio (math formula) are given for the latter.

Figure 3.

Bird species' metacommunity-wide occupancy vs. niche breadth (a) and niche position marginality (b) in Istranca Forests. Linear regression lines and associated r2s are shown. Niche breadth and position were only analysed for species with more than 10% occupancy, and significance of the relationships was assessed by randomization. Crosses indicate mean niche breadth and position marginality in randomized communities, with the shaded area showing confidence intervals for estimated means. Black circles indicate significantly different niche breadth and position values (< 0·05). Significance of the relationships in comparison with that of randomized communities is given.

Link to The Western Palearctic Region

Occupancy across the Istranca Forests was related to total WP regional population size (math formula = 0·27, P < 0·001, for all species; math formula = 0·34, P < 0·001, for forest species, Fig. 2) as well as more weakly related to WP regional breeding range size (math formula = 0·15, P < 0·001 for all species; math formula = 0·17, P = 0·005 for forest species). Species' occupancy (math formula = 0·15, P = 0·003; Fig. 4) as well as species' WP population (math formula = 0·22, P < 0·001) and range sizes (math formula = 0·15, P < 0·001) decreased with SSI.

Figure 4.

Relationship between bird species specialization index (SSI) calculated as a composite measure for species' niche breadth and marginality in terms of habitat occupancy across France, and metacommunity-wide occupancy across Istranca Forests in Turkey (a) as well as total population (b) and breeding range size (c) for the Western Palaearctic (WP) region. Generalized linear model regression lines as well as statistical significances and models fits as deviance ratio (math formula) are given.


While community assembly involves complex processes acting at different spatial scales, many ecological studies have focused on a single process or scale. In this study, we assess the roles of environmental species sorting and dispersal at different scales in forest-bird community assembly. Overall, bird assemblages in Istranca Forests constitute a metacommunity dominated by a suite of widespread, locally abundant species in a manner that is consistent with the patterns in the regional bird population across the broader WP region. Furthermore, we found strong environmental control on local community structure, metacommunity-wide occupancy and the link between the latter and the WP regional population and range sizes. Despite this, we also found evidence for a joint, albeit weaker role of non-environmental spatial factors for local community structure.

The Istranca Forests' metacommunity exhibited a strong positive abundance–occupancy relationship, as has been widely documented (Bock & Ricklefs 1983; Hanski, Kouki & Halkka 1993; Gaston et al. 2000). Furthermore, species' occupancy across the Istranca Forests' metacommunity significantly related to species' WP regional population and, to a lesser extent, range size, providing evidence for a close link between local and regional communities, as previously proposed for North American land birds (Bock 1987). This consistency in community structure across local to regional scales suggests that bird community patterns at smaller scales may be influenced by the factors acting on regional populations across large spatial scales (Ricklefs 2011).

Null model analyses showed that species' occupancy across Istranca Forests was clearly driven by species' environmental niche (Grinnell 1917). Although both niche breadth (Brown 1984) and niche position marginality (Hanski, Kouki & Halkka 1993) were correlated with occupancy, only the relationship of niche position marginality deviated from the relationship expected in randomized communities. In agreement with this finding, analyses on British breeding birds (Gregory & Gaston 2000) and invertebrate assemblages of tropical intertidal pools (Azeria & Kolasa 2008), as well as simulations with an individual-based model for the relationships between species ranges, speciation and extinction (Birand, Vose & Gavrilets 2012), have all found stronger support for niche availability than niche breadth hypothesis. Furthermore, SSI as a measure of species specialization from a distant part of the WP biome (France) correlated with species' occupancy across Istranca Forests as well as species' population and range sizes across the whole WP, indicating consistent niche assembly mechanisms acting across the region. Therefore, our results suggest that the link between the local metacommunity and regional population reflects consistent environmental species sorting processes across a large ecologically coherent region. Species' abundance and occupancy may correlate across scales under a random assembly processes (Hartley 1998). However, both null model and SSI analyses clearly suggested non-random assembly linked to species' environmental niches. Such a broad-scale environmental structuring is consistent with the high dispersal capacity of birds but should not necessarily rule out supplementary dispersal-driven regional effects (Ricklefs 1987).

Overall, occupancy was not related to mean abundance, while it increased with maximum abundance. The relationship between mean abundance and occupancy became significant and moderately strong, when only forest-associated birds were analysed. One explanation for these patterns might be the mass effects (Leibold et al. 2004), which could weaken the relationship between occupancy and mean abundance for all birds by allowing non-forest species to sustain high mean abundance in forest stands through dispersal from neighbouring source habitats, like forest openings, agricultural fields or wetlands.

Our analysis indicated that the environmental species sorting and spatial factors jointly drive local community structure within the Istranca Forests' metacommunity, but with the former having the dominant role. In line with our findings, a meta-analysis of metacommunity studies revealed that niche-based processes tend to dominate community assembly across a wide range of organism groups (Cottenie 2005). Environmental species sorting was also the dominant structuring force in Chilean temperate forest-bird assemblages up to 1000 km (Meynard & Quinn 2008). In contrast, an analysis of breeding bird communities across France (Meynard et al. 2011), where environmental gradients are less steep in comparison with the Chilean study, found a joint role of environmental and dispersal processes in community assembly. Moreover, Driscoll & Lindenmayer (2009) similarly found joint roles of environmental and neutral control across Australian bird assemblages in isolated woodland fragments. Istranca Forests span 1000 m in elevation and thus encompass strong environmental gradients within a relatively small geographic extent, probably enhancing the role of environmental control. Overall, dispersal-driven spatial processes might have a more pronounced role for bird communities, when the environmental gradients are not steep and isolation enhances dispersal limitation.

Spatial effects in ecological communities might reflect unmeasured spatially structured environmental variables. However, given our comprehensive environmental data, non-environmental spatial processes such as biotic interactions (Heikkinen et al. 2007; Kissling, Rahbek & Böhning-Gaese 2007; Gotelli, Graves & Rahbek 2010) and dispersal limitation (White & Hurlbert 2010) seem plausible underlying ecological mechanisms. Recent simulation studies have criticized the reliability of variation partitioning in analysing different community assembly processes (Gilbert & Bennett 2010; Smith & Lundholm 2010). However, Tuomisto, Ruokolainen & Ruokolainen (2012) showed that these shortcomings are only important for data sets with large compositional gradients and high dissimilarity saturation (frequency of site pairs that do not share any species). Dissimilarity saturation in the present data set was only 2%.

In conclusion, we found that environmental species sorting played a dominant role in forest-bird community assembly across scales, while non-environmental spatial factors, perhaps dispersal or biotic interactions played a supplementary role in local community assembly. The Istranca Forests species' occupancy patterns were strongly consistent with population and range sizes across the WP region as well as linked to a measure of habitat specialization from a distant part of the biome, suggesting that community assembly is shaped by consistent environmental species sorting processes acting across the whole region as proposed by Finlayson (2011). Nonetheless, we also found a significant role of purely spatial factors in local community assembly. An important avenue for future research would thus be to further elucidate to what extent dispersal (White & Hurlbert 2010) and biotic interactions such as trophic interactions (Kissling, Rahbek & Böhning-Gaese 2007), facilitation (Heikkinen et al. 2007) and conspecific attraction (Gotelli, Graves & Rahbek 2010) interact with environmentally driven niche processes to control bird community assembly across scales.


We thank Murat Bozdoğan, Cemil Gezgin and Ergün Bacak for their participation in bird surveys; Keziban Kaynar and Emre Öztürk for their help in the field; Kerem A. Boyla, Asaf Ertan, Şahika Ertan, Okan Can, Özge K. Can, Ömer Necipoğlu for their suggestions on methodology and local avifauna; Michael Green, Türker Altan, Selim Cesur, Refik Çölaşan, Mustafa İşçioğlu, Volkan Göç and Adil Akyol for their cooperation; Romain Julliard for providing SSI values; and Anne M. Poulsen for linguistic corrections. We also thank Jurek Kolasa, Erol Akcay and three anonymous reviewers for their helpful suggestions. The fieldwork was funded by the Yıldız Mountains Biosphere Project (EuropeAid/125289/D/SER/TR). We acknowledge economic support from the Danish Council for Independent Research – Natural Sciences (grant #272-07-0242 to JCS).