The idea that ecological communities are unsaturated is central to many explanations for regional gradients in species diversity.
We describe a test for differing degrees of saturation across a regional diversity gradient, based on within-species geographical variation in ecological attributes. If communities in species-poor regions are less saturated than communities in species-rich regions, species that straddle both regions should have broader niches in species-poor regions, exploiting resources that are consumed by other species in species-rich regions.
We studied 10 species of Old World leaf warblers that range across the Himalayas. Elevational range and feeding method showed niche contractions in the species-poor north-west Himalayas with respect to the species-rich south-east Himalayas, whereas prey size did not vary geographically. Niche contractions are contrary to the expectation of character release in depauperate environments, as has been shown, for example in mainland-island comparisons.
We show that arthropod abundances are likely a limiting resource, and that niche contractions are consistent with measurements of a narrowing of resource availability.
Results suggest that north-western warbler communities are at least as saturated as the south-east and that lower resource diversity drives reduced species numbers.
Across the terrestrial environment, warm, wet, less seasonal areas contain more species than cold, dry, more seasonal areas (Hawkins, Porter & Diniz-Filho 2003; Davies et al. 2007). This is manifest most clearly in the latitudinal diversity gradient, whereby most groups contain more species in tropical than temperate regions (Hillebrand 2004). One explanation for the pattern is that species are close to a carrying capacity and a site with many species has a higher species carrying capacity than one with few species (MacArthur 1965; Rabosky 2009; Hurlbert & Jetz 2010; Cornell 2013). The alternative is non-equilibrial and species continue to accumulate (Ricklefs 2006; Mittelbach et al. 2007). In this case, regional differences are attributed to differences in the historical processes of the time a clade has been present in the region, its diversification rate and directions of species dispersal between regions (MacArthur 1965; Jablonski, Roy & Valentine 2006; Wiens 2011). However, because differences in regional diversity must result from these processes (Wiens 2011), simply documenting how the pattern has been generated does not in itself address the question of how far diversity is from equilibrium (Ricklefs 2006). For example, in marine molluscs, Jablonski, Roy & Valentine (2006) found higher tropical diversity to be associated with higher tropical diversification rates, but these authors presented an equilibrium explanation for the latitudinal gradient: ‘the average higher-latitude species is ecologically more generalized and requires a larger share of available resources than the average lower-latitude species', with the result that ‘fewer species can be supported in trophically variable environments' (Valentine et al. 2008, p.173).
If the latitudinal gradient is driven primarily by variation in carrying capacity, local communities along the gradient should be close to saturation, that is, resistant to invasion by more species. The term saturation has been used in several different ways, associated with different scales across both time and space, and which should be considered the potential source pool of species (Terborgh & Faaborg 1980; Chase & Myers 2011). Here, we define saturation with respect to a clade: a saturated local community implies no more clade members can enter that community, because of both biotic and abiotic factors that affect local species carrying capacity, assuming current conditions persist into the future. Community resistance occurs at two levels. First, already formed species outside the local community may be unable to become established, and second, new species may form from extant clade members, but once formed they are unable to enter the community.
In this study, we address the question of whether local communities across a regional diversity gradient are saturated, and in particular, whether communities in species-poor areas are less saturated than those in species-rich areas. We introduce a new way to test this, based on geographical variation in the degree of specialization within those species whose range spans regions of low and high diversity. Assuming populations are limited by resource availability, if communities in regions of low diversity are less saturated than those in regions of high diversity, a species found in both regions should show niche expansions in the low-diversity region (compare Fig. 1a with Fig. 1b). This is a strong directional prediction: the unsaturated model is rejected if species are more specialized in regions of low diversity. The prediction is not simply abstract theory but has empirical support from islands. Islands are generally depauperate, attributed to a low rate of successful colonization (MacArthur & Wilson 1967; Lack 1976), with communities unsaturated when compared to the mainland (Terborgh & Faaborg 1980; Ricklefs 2012). As expected, island communities regularly show niche expansions (Lack 1976; Terborgh & Faaborg 1980; Scott et al. 2003). For example, Lack (1976) showed that of 31 bird species present in both Honduras and Jamaica, 15 clearly occupied a greater range of habitats on Jamaica, whereas only one occupied a greater range of habitats on Honduras.
The directional prediction of the unsaturated model is made assuming resource distributions (and hence carrying capacities) are similar across regions, but predictions from the saturated model explicitly depend on resources (MacArthur 1965). If resources in different locations have the same distribution but differ in abundance, as, for example, may happen in association with variation in productivity, then a species may generalize in regions of low diversity (MacArthur 1965; Fig. 1c). Generalization may also be expected if resource distributions vary seasonally, with one resource of prime importance in one season and another in the other (Valentine et al. 2008). However, specialization, or no geographical variation at all, is expected if the resource distribution narrows sufficiently (MacArthur 1965; Fig. 1d). Thus, predictions from the saturated model can only be made if the resource base is measured. The importance of considering resource distributions in the context of diversity gradients is widely recognized (Ricklefs 2012) but resources do not seem to have been previously measured, at least concordantly with measurements of resource utilization.
The question of generalists and specialists and what it means for the regional variation in species numbers diversity is receiving renewed attention (Vázquez & Stevens 2004; Belmaker, Sekercioglu & Jetz 2011; Salisbury et al. 2012). Two ecological dimensions have been addressed: diet and elevational range. With respect to bird diets, across the globe, Belmaker, Sekercioglu & Jetz (2011) found that regions with more species had a higher proportion of specialists than regions with fewer species, even after accounting for associations of species richness with productivity (seasonality was not considered). Belmaker, Sekercioglu & Jetz (2011) suggested that specialization may increase diversification rates, an explicitly unsaturated model. Jocque et al. (2010) and Salisbury et al. (2012) presented a mechanism for the way in which specialization might drive speciation, arguing that it leads to reduced dispersal and hence reduced gene flow between populations, a critical early step in the speciation process. Vázquez & Stevens (2004) also assumed an unsaturated model, but reversed causality, suggesting that high species diversity leads to specialization. In contrast, as noted above, Valentine et al. (2008) invoked a saturated model, in which generalists in seasonal environments out-compete specialists. For the most part, authors have not explicitly compared the alternatives.
With respect to elevational range, some studies have found elevation ranges to be generally broader in the temperate than the tropics (Ghalambor et al. 2006; McCain 2009), but others have not (Kozak & Wiens 2007; Cadena et al. 2012). Unlike diet diversity and island elevational range expansions, recent explanations for large elevational ranges in temperate continental regions have generally assumed that they are a response to local conditions and hence apply to saturated communities. The best known is that climatic variability drives a broader physiological tolerance in temperate regions, resulting in ability to expand range across climatic regimes, generating larger elevational ranges (Kozak & Wiens 2007; McCain 2009). This idea dates back to Janzen (1967) who briefly noted that species in tropical environments seem to be especially faithful to specific habitats along mountainsides (reviewed in Ghalambor et al. 2006). Wider elevational ranges in temperate regions are thus predicted both in cases where communities are saturated and ranges are responding to climatic variability, as well as in unsaturated models, where range expansions are primarily a result of competitive release, with the temperate region further from equilibrium than the tropical region. These examples from diet and elevation illustrate more generally that, because cross-species comparisons often find increased generalization in species-poor regions (e.g. Belmaker, Sekercioglu & Jetz 2011), they are consistent with saturated and unsaturated models.
In contrast to the many field studies comparing within-species geographical variation between islands and continents (summarized in Scott et al. 2003), there has been nothing comparable across continental gradients, perhaps because many of these gradients are associated with large turnover in species and often quite distantly related clades (Belmaker, Sekercioglu & Jetz 2011). Thus, regional biodiversity gradients are rarely traversed by the same species. The Himalayan bird diversity gradient is an exception, with >150 species extending >1000 km from south-east to north-west, straddling a decline in total bird species numbers of c. 40%, in association with a climate that varies from more tropical to more temperate (Price et al. 2011). A non-equilibrium, unsaturated, explanation for the reduction in species diversity from south-east to north-west along the Himalayas is reasonable: the north-western Himalayas were apparently treeless at multiple periods during the Pleistocene (Ray & Adams 2001), during which time forest species are expected to have withdrawn to the south-east. The implication is that colonization of the north-west has been relatively recent, and plausibly, many species that have failed to become established over the last 21 000 years since the last glacial maximum could colonize given more time.
We focus on geographical variation in the foraging behaviour, prey size and elevational range of 10 species of warblers in the family Phylloscopidae (Old World leaf warblers) with large geographical distributions. These three ecological dimensions are strongly correlated with morphological attributes in both the north-west (Price 1991; see Fig. S1 in Appendix S1, Supporting information) and south-east (Fig. S1, Tables S1–S4, Supporting information), suggesting their importance in structuring communities. We set out to test the key prediction of an unsaturated, non-equilibrium model (Fig. 1b), which is that conspecific populations are more generalized in their elevational range and foraging ecology in the north-west than in the south-east. In this model, generalization in species-poor regions is a result of competitive release (Fig. 1b), and predictions can be refined based on the ecological characteristics of species present in the species-rich region that are missing from the species-poor region. For example, if two related species replace each other elevationally in a species-rich region and one is missing from the species-poor region, the other should expand elevational range in the direction of the missing species, as has been shown for true islands in comparison with continents (Lack 1976), as well as isolated mountains on continents (Terborgh & Weske 1975). In the results section, we first present characteristics of species in the south-east that are missing from the north-west and use these characteristics to ask how species should generalize across the gradient under an unsaturated model. We then test these predictions for the 10 species that have ranges extending across the Himalayas.
Materials and methods
The Phylloscopidae are small (5–11·5 g), entirely insectivorous warblers. Nineteen species occur in the Indian Himalayas (Johansson et al. 2007) and form dominant components of forest communities along the entire elevational gradient, except at the lowest elevations in the west (Price et al. 2011). Most species are migratory and retreat to low-altitude foothill forests and the tropical regions of peninsular India and South-East Asia in the non-breeding season. We compared breeding communities in the south-east and north-west Himalayas, a minimum of 900 km separated (Fig. S2, Supporting information). Fourteen species breed along single elevational gradients in the south-east and up to 10 species in the north-west. This 30% decline in species numbers from south-east to north-west is commensurate with the decline in total bird species numbers from the east of Nepal to the west of Nepal (Price et al. 2011). In both the south-east and the north-west, the Phylloscopidae are predominantly temperate and form a monotonically increasing proportion of all small insectivorous species with elevation (Fig. S3, Supporting information).
We define the south-east as all study sites east of Nepal and the north-west as all study sites west of Nepal (Fig. S2, Supporting information). In general, the south-east is wetter and less seasonal than the north-west, particularly in the lower elevations (Price et al. 2011; Fig. S2, Supporting information). We combined data from five elevational transects in the south-east and four transects in the north-west (Table S5, Supporting information). Transects were divided into 500-m bands. Each 500-m band was represented by at least one site in each region, with seven sites in the north-west (1000–4000 m) and nine sites in the south-east (100–4000 m), where arthropods were sampled. Some important characteristics of each transect are summarized in Table S5 (Supporting information), and more detailed descriptions of vegetation associations are provided in Ghosh-Harihar (2013). We measured both resources and warbler ecology at multiple locations along the diversity gradient over 5 years (2007–2011).
The breeding season commences earlier in the south-east as compared to the north-west, given the lower latitude (M. Ghosh-Harihar & T.D. Price, pers. obser.), and we designed our sampling to coincide with the nest-building stage at each site as far as possible, during which the males were most vocal.
We studied both vegetation and arthropod availability. At multiple sites across the Himalayas, TDP established 5 ha grids and sampled tree density (within a 5 m radius of 15 randomly chosen points), tree type (coniferous vs. broadleaf) and foliage density (number of leaves an imaginary pole extending to the sky would intersect, over 30 random points; see Price et al. (2011) for details). MGH sampled arthropods along the elevational transects, as indicated in Table S5 (Supporting information), and detailed methodology is described in Ghosh-Harihar (2013). Arthropods were later sorted into two size classes (‘large’ and ‘small’) based on whether the total body length of the arthropod was greater or less than 4 mm. For some analyses, we estimated arthropod prey abundance for a 500-m elevational band within each region. When more than one site was present in a 500-m band in a region, arthropod numbers for these sites were averaged. In the seasonal environments we study, arthropod abundance should vary by date. This was one reason why we standardized our sampling to coincide as far as possible with the pair-formation, egg-laying stage. The seven sites we sampled in different 2 years (Table S5, Supporting information) showed no significant differences in arthropod abundance between the two sampling events (t-tests based on bags as replicates, all P > 0·05), suggesting we are capturing consistent differences between sites. In our analyses, we combined arthropod samples across years for these sites.
Elevational distributions of the birds
On a given morning from 0600 to 1000, MGH surveyed a particular 500-m elevational band. At every 25 m, as measured with an altimeter, she counted all singing males seen or heard within c. 25 m radius for 5 min. Each interval was covered twice during a breeding season. Price et al. (2003) showed that this census method gave repeatable results for this highly vocal group across years in one breeding location (Manali, Himachal Pradesh). We define an ‘elevational distribution’ as the upper and lower limits of the breeding range; that is, where we recorded at least one male singing (isolated males >100 m above or below other males were excluded). We estimated warbler densities in 500-m elevational bands as the average detections across point count stations and across sites within regions. The magnitude of the elevational range is taken to be a direct measure of generalization on this dimension.
We searched actively for the target species. Once located, we followed a bird until it captured a prey item and then moved on to the next observation. We therefore consider each observation to be independent. Behavioural sampling was carried out opportunistically between 0600 and 1400 hours, during which c. 5 km was searched along forest trails. At each prey capture, we recorded foraging height (m), substrate (leaf, twig, branch, conifer needles and air), tree height (m), foraging manoeuvre and prey size (Tables S3, S4, Supporting information). Following Price (1991), we defined a foraging manoeuvre as (i) standpick if the bird picked a stationary prey without involving any flight; (ii) hoverpick if the bird searched and picked a stationary prey from a leaf or twig while in flight; (iii) flypick if the bird observed a prey resting on a substrate, while perched or hopping, and then captured it from the substrate by flying, and (iv) flycatch if the bird sallied from a perch to catch a flying prey (Table S3, Supporting information). For many of our analyses, these feeding methods were categorized into two, depending on whether the bird used flight in prey capture, that is, standpicks vs. flymoves (flypick + flycatch + hoverpick). A prey item was defined as large if it was seen in the bill and required mandibulation or small if the prey was not visible in the bill. This estimate correlates well with assessments based on measurements of arthropod remains in faeces (Price 1991; Price & Gross 2005). MGH collected all observations. Foraging mode and prey size are measured as binary variables (flying vs. picking and small vs. large, respectively), so generalization is maximized when the proportions are 0·5.
We used ancova and generalized linear model (binomial error) to analyse the effects of region (south-east and north-west), elevation and elevation2 on total arthropod abundance and proportion of large arthropods, respectively. Analyses followed a backwards stepwise procedure whereby all variables and interactions (elevation × region, elevation2 × region) were initially included, and non-significant interactions were sequentially removed from the model.
To compare species attributes, we used standard tests as well as corrections for phylogenetic non-independence (Felsenstein 1985), based on the ultrametric phylogeny published in Johansson et al. (2007). Independent contrasts were calculated using the package ape (Paradis, Claude & Strimmer 2004) in R 2·12·1 (www.r-project.org) and significance assessed via regressions and correlations forced through the origin. Because some of the dependent variables were presence/absence data, we confirmed all results using general estimating equations in ape, specifying normal or binomial errors as appropriate.
A major assumption of the study of ecological generalization as a test for saturation is that resources and competition are important factors limiting species abundances. In breeding season temperate bird assemblages, evidence for food limitation and competition has come from both experimental and correlative studies (Bourski & Forstmeier 2000; Rodenhouse et al. 2003) but nest predation (Martin 1988), parasites (Ricklefs 2011), winter limitation (Johnson & Sherry 2001; Price & Gross 2005), migration costs (Sillett & Holmes 2005) and direct effects of the abiotic environment, such as temperature tolerance (Swanson & Garland 2009), may all reduce the importance of competition in structuring breeding assemblages. In this system, resource limitation is implied by the strong correlation between arthropod and warbler abundances along elevational gradients in both the south-east and the north-west, as well as between regions (Fig. 2).
In the unsaturated model, populations should show increased generalization in the species-poor north-west Himalayas, compared with the south-east. As noted in the introduction, the predictions of this model can be refined based on those species that are present in the south-east but missing from the north-west. In this case, generalization is expected in the direction of the ecological attributes of missing species. We studied the ecology of all species in the south-east, comparing attributes of the 10 species whose range extends to the north-west with the four whose range does not extend north-west. The four species confined to the south-east are found at relatively low elevations in comparison with the geographically widespread species (Fig. 3, right, two-sample t-test, t12 = 2·8, P = 0·02, phylogenetic correction, P = 0·02) and capture prey exceptionally often in flight (Fig. 3, right, Table S3 (Supporting information), t12 = 3·3, P = 0·01, phylogenetic correction, P = 0·02). Thus, competitive release should result in species in the north-west having broader elevational ranges with lower elevational range limits and a higher proportion of prey captured in flight. In contrast, the third niche dimension we studied, that of prey size consumed, does not differ between those species that extend north-west and those that do not (Table S3, Supporting information, t12 = 0·9, P = 0·38, phylogenetic correction, P = 0·15), leading to no expectation of geographical variation within species, based on what confamilials are doing in the species-rich areas.
We evaluated geographical variation in the three niche dimensions given these predictions, using the 10 species that have ranges extending across the Himalayas as replicates in paired tests of south-east vs. north-west attributes. First, elevational ranges are narrower in the north-west, on average by about 60% (Table 2, paired t-test: t9 = 3·4, P = 0·01). Second, species catch prey by flying less in the north-west than in the south-east (Table 2, paired t-test on arcsine square-root transformed data: t9 = 2·8, P = 0·02). Third, the proportion of large prey captures does not vary geographically within species (data in Table S3, Supporting information, paired t-test, t9 = 0·7, P = 0·94). In summary, consistently across species, both feeding method and elevational distribution show niche contractions in the species-poor north-western Himalayas, not niche expansions as expected under an unsaturated model. The predictions of that model are made assuming species carrying capacity is similar in both locations (Fig. 1b). We now turn to ask if resource distributions may result in a lower carrying capacity in the north-west.
With respect to elevation, vegetation is sparser in the north-west than the south-east (Table 1) and arthropod abundance lower (Fig. 4). Thus, if elevation is treated as a resource, a uniform decline is expected to lead to increased generalization, as some species drop out, and others expand resource use (MacArthur 1965; Fig. 1c). However, arthropod abundance is not uniformly distributed across elevations and is highest at intermediate elevations (Fig. 4). Statistics, based on a model with region, elevation and elevation2 included, and from which non-significant interaction terms have been dropped are as follows: elevation: F1,12 = 10·9, P = 0·007, elevation2: F1,12 = 29·2, P = 0·0002, region: F1,12 = 13·1, P = 0·004, interaction between elevation and region: F1,12 = 15·5, P = 0·002 (the interaction arises because abundance peaks at lower elevations in the south-east). The range over which resources are abundant is narrower in the north-west (e.g. see grey dashed line in Fig. 4), so smaller elevational ranges are expected under carrying capacity arguments (Fig. 1d). This is what is observed in the species-poor north-west.
Table 1. Vegetation characteristics from 34 5 ha grids at multiple sites along several elevational gradients in the south-east and the north-west. Tree density refers to number of trees >15 cm circumference at breast height within a 5-m radius of 15 random points; % conifer was calculated from tree density. Foliage density is the total number of leaves estimated that would be touched along 30 vertical lines of sight above 0·2 m from 30 randomly distributed points within the grid
% conifer was not measured at every grid, hence the fewer error degrees of freedom. For this test, generalized linear model with binomial error distribution was used.
<2000 m (154–1932 m)
>2000 m (2340–4024 m)
<2000 m (480–1875 m)
>2000 m (2604–4100 m)
F1,31 = 7·4, P = 0·01
F1,31 = 10·9, P = 0·002
z = −10·8, P = 0·0001
F1,31 = 1·9, P = 0·18
F1,31 = 11·9, P = 0·002
z = −5·39, P = 0·0001
The increased specialization in foraging behaviour is also as expected in response to altered resource distributions. Broadleaf is proportionately rare in the north-west than the south-east (Table 1), and accordingly used less often by north-western populations (paired t-test comparing proportions of observations on broadleaf, t9 = 4·0, P = 0·003, Table 2). Within regions, foraging in broadleaf is correlated with flying movements (Fig. 3, left). To test if regional differences in foraging technique can be explained entirely as a result of substrate, we used a mixed-effects model with the 10 species as a random effect and substrate categorized as broadleaf, needle, twig, branch and air using the lmer package (Bates, Maechler & Bolker 2010). As the procedure does not provide P-values, significance of fixed effects was obtained from 10 000 Markov chain Monte Carlo simulations using the pvals.fnc function in the languageR package (Baayen 2008). We found that use of broadleaved substrates remained significant (P = 0·0001) but the difference across regions disappears (P = 0·48). Thus, altered use of foraging methods between regions, with concurrent increased specialization, is explained as a response to a reduction in broadleaf substrates.
Table 2. Use of flymoves as a feeding method, elevational range sizes and use of broadleaf substrate of the breeding leaf warbler species across two regions (south-east-SE and north-west-NW) of the Himalayas
Elevational range size (m)
Standard errors and sample sizes are in the online tables.
We also considered prey size. The proportion of arthropods that are large is higher at lower rather than higher elevations and is highest at intermediate elevations (Fig. 4, elevation: z =3·66, P = 0·0003, elevation2: z =−5·61, P = 0·0001) but the two regions are similar in this respect (region: z =0·2, P = 0·84), with no interaction terms retained in the final model. Thus, a lack of geographical variation in prey size generalization is expected in response to resources. Because we showed above that a lack of variation is consistent with predictions from missing congeners as well, prey size cannot be used to distinguish the alternative saturated and unsaturated models.
In many bird groups, including the Phylloscopidae, those species present at higher elevations in the south-east are most likely to have geographical ranges that extend to the north-west (Fig. 4 in Price et al. 2011). Pleistocene glacial maxima were apparently associated with a general drying of the north-western Himalayas (Ray & Adams 2001). In this case, many forest species should have retreated to the south-east, and a reasonable non-equilibrium hypothesis is that some of these have failed to recolonize the north-west. In the historical scenario then, high-elevation forest species have preferentially recolonized the north-west from the south-east. This means that if history accounts for the lack of species in the north-west, high-elevation species are predicted to expand their foraging niches to consume resources of the absent lower elevation species. We found no evidence in support of this hypothesis. Species that occur in both the south-east and north-west do not have broader elevational ranges when they are in the north-west, nor do they use flying manoeuvres more when in the north-west, as would be expected if they were expanding into the niches of the missing south-eastern species. Instead, they have narrower elevational ranges in the north-west and catch prey by flying less. Two Phylloscopus species in our study are restricted to the north–west and not found in the south-east. These two species (P. humei, P. occipitalis) forage with a high proportion of standpicking (Table 2), which is consistent with the general shift in the widespread species towards more standpicking in the north-west and also inconsistent with expanded niches in response to absent south-eastern congeners.
We attribute the pattern of specialization in the north-west to be a direct response to resources. In particular, a lower proportion of broadleaf and greater proportion of conifer in the north-west (Table 1) favour increased standpicking. Further, the generally drier and cooler conditions, with lower foliage density (Table 1) and lower prey abundances (Fig. 4), reduce the elevational range over which any given species persists. Our considerations of prey size and feeding method are very crude, and we have not directly addressed prey types, which would be an important extension, and may contribute to the reduction in flycatching in the north-west. For example, Ghosh-Harihar (2013) split the arthropods into six groups and found that at low elevations, the north-west was especially low in true flies (Diptera), whereas at these elevations, non-volant groups (e.g. spiders) differed less, and it would be of interest to assess the importance of Diptera in the diet of south-eastern populations. Responses in elevational range are quite variable, and two species, P. xanthoschistos and P. chloronotus, have similar ranges in both south-east and north-west (Table 2). These idiosyncratic responses may at least partly reflect a general lowering of resources, which then lead to range expansions under the saturated model (Fig. 1c). Our results point to communities in the north-west being saturated, or at least more saturated than the south-east, despite being relatively species-poor. In Fig. 5, we show that local communities generally have fewer species in the north-west, which we attribute to reduced resource diversity.
Prey size and feeding method are the dimensions we have measured associated with local coexistence. Both these dimensions evolved long ago in the Phylloscopidae (Richman & Price 1992). On the other hand, elevational distributions have evolved more recently and are the prime axis of separation among sister species, especially within the north-western species (Richman & Price 1992). Thus, elevational distribution might be considered a priori more likely to be limited by speciation and dispersal than the other axes and, in an unsaturated model, a particularly likely axis of further niche subdivision. In particular, those species with broad elevational ranges in the north-west (such as P. xanthoschistos and P. chloronotus, Fig. 5) seem especially good candidates to themselves eventually split into two species, which subdivide the elevational range. However, both these species have relatives outside the Himalayas that have failed to expand into sympatry (Johansson et al. 2007). In Fig. 6, we take this further and illustrate the mutual elevational and geographical distributions of the closest relatives in our data set, P. occipitalis and P. reguloides, separated by >2My (Johansson et al. 2007). These species have limited geographical overlap (Fig. 6). They are very similar in feeding method (Table 2), and both take similar quantities of large prey (Table S3, Supporting information), that is, divergence on these niche dimensions is small. However, the species show character displacement with respect to elevational range, which for both, is broader in allopatry than sympatry. Despite this displacement, interactions have not led to divergence sufficient to result in widespread geographical coexistence, implying community resistance. The far north-west has not been colonized by P. reguloides, nor has the south-east been colonized by P. occipitalis (Fig. 6).
The north-west has a much more temperate climate than the south-east, being cooler, drier and more seasonal (Price et al. 2011). Thus, our results, which show increased specialization in the north-west, conflict with the widespread observation that a higher proportion of breeding bird species are specialists in the tropics than the temperate regions, with respect to both diet (Belmaker, Sekercioglu & Jetz 2011; Salisbury et al. 2012) and elevational range (McCain 2009). We suggest three reasons. The first is that temperate communities are generally unsaturated and the Himalayan Phylloscopidae are atypical. Along the Himalayas, the warblers we studied are migratory and hence likely to have high dispersal distances (Paradis et al. 1998) and cross obstacles relatively easily. Further, they are abundant, reducing chances of extinction after establishment. These two factors contribute to rapid re-colonization of the north-west in this particular group. Even within the Himalayas, other guilds (e.g. genera such as sunbirds and wren babblers, which show steep south-east to north-west declines; Price et al. 2011) may be below saturation, and for example, the purple sunbird, Cinnyris asiaticus, Price et al. 2011) show a large elevational range expansion in the north-west compared to the south-east.
A second reason why increased specialization within the Phylloscopus species may differ from the pattern of increased generalization in temperate regions is that, even if communities are saturated, species respond to seasonal environments by generalizing. This should apply particularly to resident species. For example, finches consume different food items (seeds, arthropods) at different times of the year. It is this kind of seasonal fluctuation that is seen by Valentine et al. (2008) as the explanation for why species in temperate environments are generalists, why they can out-compete specialists and ultimately why fewer species are present in the temperate. A similar argument may apply to elevation. Residents in north temperate regions may respond to large seasonal fluctuations in the temperature they experience by expanding their ranges across elevations (Kozak & Wiens 2007; McCain 2009). This may then lead to saturated communities, but fewer species in total when integrated across the elevational gradient.
A third explanation for the atypical observation of increased specialization in the more temperate region is that our results are a feature peculiar to the Himalayas and are not representative of environments further to the north. In the Himalayas, warbler abundance and food abundance are strongly correlated across sites (Fig. 2), but Irwin (2000) found that along an extended latitudinal gradient into Siberia, the abundance of the greenish warbler, P. trochiloides, decreased even as prey abundance increased. In this case, the decline in P. trochiloides density may be attributed to distance from the wintering grounds in India and associated costs of migration (Forstmeier, Bourski & Leisler 2001). P. trochiloides occupies a wider range of habitats to the north, that is generalizes along this dimension at least (compare Price (1991) with Forstmeier, Bourski & Leisler (2001)). Thus, generalization in Siberia apparently results not so much from the absence of certain species, but rather from reduced numbers of individuals of all species, and Siberian communities do appear to be unsaturated. The strong association of resources and abundances within the Himalayas (Fig. 2) imply that one value of this system is that food resources in the breeding season are more strongly limiting, enabling direct tests of the unsaturated model through tests of a hypothesis of niche expansion.
In summary, our results imply that species richness in this group is lower in the north-west in the main part, not for historical reasons such as the impact of Pleistocene climate, but because resources are less diverse in the contemporary temperate climate. Evidence is consistent with the Old World leaf warbler Himalayan communities being close to saturation, in accord with MacArthur's (1965) view of local New World warbler communities, and as proposed more generally for correlates of productivity with bird species diversity (Hurlbert & Jetz 2010). These conclusions apply to the insectivorous leaf-gleaning guild of birds and need to be assessed for other groups, especially those that are less dispersive. We suggest that studies of within-species geographical variation will be generally a useful approach to the question of saturation, especially when they are coupled with measurements of resources.
We thank Dhananjai Mohan, who collected some of the vegetation data, for much discussion, field assistance, and help with logistics. We also thank Pratap Singh for much help in the field, and J. Belmaker, P. Grant, A. Harihar, A. Hurlbert, A. Phillimore, T. Sherry and G. Rawat for comments on the manuscript. We thank the Director and Dean, Wildlife Institute of India and the Chief Wildlife Wardens of Jammu and Kashmir, Himachal Pradesh, Uttarakhand, Sikkim, West Bengal and Arunachal Pradesh for permits. This study is supported in part by grants from the Wildlife Institute of India (to MG) and the US National Science Foundation (to TDP).