Habitat loss often reduces the number of species as well as functional diversity. Dramatic effects to species composition have also been shown, but changes to functional composition have so far been poorly documented, partly owing to a lack of appropriate indices. We here develop three new community indices (i.e. functional integrity, community integrity of ecological groups and community specialization) to investigate how habitat loss affects the diversity and composition of functional traits and species. We used data from more than 5000 individuals of 137 bird species captured in 57 sites in the Brazilian Atlantic Forest, a highly endangered biodiversity hotspot. Results indicate that habitat loss leads to a decrease in functional integrity while measures of functional diversity remain unchanged or are even positively affected. Changes to functional integrity were caused by (i) a decrease in the provisioning of some functions, and an increase in others; (ii) strong within-guild species turnover; and (iii) a replacement of specialists by generalists. Hence, communities from more deforested sites seem to provide different but not fewer functions. We show the importance of investigating changes to both diversity and composition of functional traits and species, as the effects of habitat loss on ecosystem functioning may be more complex than previously thought. Crucially, when only functional diversity is assessed, important changes to ecological functions may remain undetected and negative effects of habitat loss underestimated, thereby imperiling the application of effective conservation actions.
The realization that biodiversity affects ecosystem functioning has led to a major shift in research goals within conservation biology, from preserving species towards maintaining ecosystem properties and functions . While measuring ecosystem functions directly is often very difficult, changes to the provisioning of ecological functions are often investigated through an assessment of functional trait diversity of particular communities, assuming that communities characterized by lower trait diversity provide fewer ecological functions [2,3]. Significant losses of functional trait diversity in direct response to habitat changes have already been observed in many taxa, including mammals, birds, anurans, fish and trees, and in a variety of terrestrial and aquatic ecosystems (e.g. [4–7]). Current knowledge, however, does not point to a simple pattern, as some studies have found that functional diversity remained unchanged or was even positively affected (e.g. [5,8–11]). Existing evidence thus suggests that the consequences of environmental changes for ecological functions are complex and far from being well understood.
It is not surprising, however, that ecological functions are affected by environmental change in various ways, as communities also show complex responses [12–15]. For instance, habitat loss often results in a loss of species but recent findings show that changes in species composition without systematic declines in species diversity are common and widespread [12–15]. Because changes in ecological functions occur via species turnover (assuming no changes in traits occur, i.e. no intraspecific trait variability ), it is likely that modifications in functional trait composition can also take place without concomitant changes in functional trait diversity. Thus, studying changes in functional trait composition may help to understand why a variety of functional diversity responses have been observed, and may also improve our knowledge on the functional consequences of environmental change.
Existing studies regarding functional composition do not capture the complexity of anthropogenic changes to ecological functions. For instance, community-weighted mean trait values are frequently used as a measure of functional trait composition, but only provide information on a single trait at a time [17,18]. Additionally, our knowledge on how alterations in species composition in a community lead to changes in the ecological functions performed is still limited. For instance, specialists are believed to execute roles in the ecosystem that are more complementary to each other and to be more efficient in executing particular functions than generalists, yet are usually more adversely affected by anthropogenic disturbances [19,20]. However, most studies use measures of habitat specialization (e.g. [21,22]) for which a direct link with ecological functions performed is less obvious. Also, many studies investigated to what extent habitat changes result in species turnover (e.g. [12,13]) but did not explore these changes in community composition in the light of ecological functions. A possible reason why functional consequences of environmental change are still poorly examined is a lack of appropriate indices.
In this paper, we introduce three new community indices, i.e. community integrity of ecological groups, community specialization and functional integrity (see glossary) to assess how environmental changes affect the ecological functions performed in a community, and use them to study the impact of habitat loss in the Brazilian Atlantic Forest on birds. In particular, we assess how the diversity and composition of functional traits and species change along a gradient of habitat loss. We use a dataset of 137 bird species captured in 57 sites located in the Brazilian Atlantic Forest biome, one of the most threatened biodiversity hotspots in the world . These sites encompass a large range of forest cover (18–100%), including continuously forested reference sites, and it has already been demonstrated that the bird species composition in these sites is strongly affected by habitat loss, while species diversity is barely influenced [12,24]. Birds execute a variety of ecological functions, which can be deduced from their ecological traits with relative ease [25,26]. We selected 13 traits related to diet, foraging substrate and foraging stratum relevant to avian ecological functions. We analysed whether the amount of forest cover surrounding the sampling site affects two complementary metrics of functional trait composition (functional integrity and community-weighted mean trait values), functional trait diversity (functional richness and functional evenness ) and species composition (community integrity of ecological groups and community specialization).
2. Material and methods
The data were collected in three study areas situated across a 150 km-wide region in the Atlantic Plateau in the State of São Paulo, Brazil (see the electronic supplementary material, S1). Each study area was chosen to contain a fragmented and a continuously forested landscape of 10 000 ha each. The fragmented landscapes varied in their total proportion of forest cover (approx. 10, 30 and 50% of cover), and the continuously forested landscapes varied in their history of human disturbance (clear-cut, selectively logged and no known history of timber extraction). Forest fragments consisted of secondary vegetation at different successional stages and the continuously forested landscapes were mainly composed of late secondary to old-growth native vegetation [28,29]. A total of 53 sites were sampled within the three fragmented landscapes (between 17 and 19 sites per landscape), and four sites were sampled in each continuously forested landscape. In this study, we focused on comparing indices of diversity and composition between sites that differ in habitat amount and not between sites that differ in history of human disturbance. Therefore, we only analysed the data obtained from the three fragmented landscapes and the continuously forested landscape without timber extraction; the latter serving as a reference condition (i.e. no forest loss or degradation). The average (±s.d.) distance between sampling sites was 4.4 km (±2.2) in the fragmented landscapes and 2.2 km (±0.8) in the continuously forested landscape. A total of 5311 individuals from 137 understory bird species were captured from more than 36 600 h of mist-netting (mean net-hours per site (±s.d.) = 643 ± 71) between March 2001 and March 2007 using 10 mist nets per site with a standardized protocol. To account for small differences in sampling effort among sites, species abundance has hereafter been divided by the number of net-hours per site and multiplied by the median number of net-hours over all sites. See Banks-Leite et al.  and Martensen et al.  for detailed information about the study area and sampling protocol.
(b) Landscape metrics
The proportion of forest cover within 300 m around the sampling site was calculated for each site. This variable combines information about patch size and isolation . We chose a radius of 300 m because preliminary analyses (electronic supplementary material, S2) indicated that bird abundance was most influenced by this spatial scale. The proportion of forest cover within 300 m around the sampling site was not evenly distributed among the 10 000 ha landscapes. However, each of the fragmented landscapes comprised a similar range of forest cover values at the site level, so our results are not likely to be biased. Furthermore, a similar variation in community dissimilarity was observed within each of the 10 000 ha landscapes , showing that our results are unaffected by changes in forest cover at the landscape (10 000 ha) scale.
(c) Species traits
We compiled data on 13 species traits pertaining to three trait categories: diet (invertebrates, vertebrates, fruits, nectar and seeds), foraging substrate (living foliage, dead foliage, bark and air) and foraging stratum (ground, understory, midstory and canopy). As results of functional analyses can vary widely depending on the traits included, selecting traits that are ecologically relevant, and justifying their selection, are crucial . We chose these traits because a bird's function in the community mostly results from its diet and foraging behaviour, and because of similar traits often used for functional classification of birds (e.g. [5,25,32]). Trait data were extracted from del Hoyo et al. , and supplemented with data from Sick , Develey & Endrigo  and Stotz et al. . The foraging substrate of two species (Schiffornis virescens and Thraupis cyanoptera) was derived from their congeners (S. turdina and T. sayaca, respectively). Integers ranging between 0 and 3 were assigned to all traits, with larger values corresponding to a larger importance of a trait for that particular species. To ensure repeatability of trait collation, we searched in the species description for specific keywords (see the electronic supplementary material, table S1) that reflect the importance of a certain trait. For instance, a ‘1’ was assigned when keywords such as ‘occasionally’ and ‘rarely’ were encountered, a ‘2’ when terms such as ‘also important’ and ‘also much’ were present and a ‘3’ was allocated when words referring to a high frequency (e.g. ‘mainly’ and ‘usually’) were found (see the electronic supplementary material, S3, for more details). Note that despite our focus on understory bird species (as birds were captured with mist nets), foraging stratum is considered a relevant trait category in this study, because understory species also use the other foraging strata to some extent. For example, thrushes forage in the canopy for insects, but they are not typical canopy species (e.g. toucans). See the electronic supplementary material, table S2, for a detailed description of all species traits.
(d) Functional trait composition
Functional trait composition was quantified with a one-dimensional measure, community-weighted means, and with a multidimensional index developed for this study: functional integrity. Community-weighted mean trait values are considered as the main determinants of ecological function [17,18]. They are calculated for all possible combinations of traits and sites as the sum of all values for a particular trait multiplied by the species' relative biomass (product of relative species abundance and body mass) in a particular site (adapted from [17,18]). We considered biomass instead of abundance in the formula of community-weighted mean since the contribution of individual species to ecological function is biomass-dependent . Functional integrity is computed in a similar way to (taxonomic) community integrity (e.g. [28,38]); however, it is based on the sites-by-traits matrix containing community-weighted mean trait values, instead of on the sites-by-species matrix with species' abundance or occurrence. The first component of an ordination based on Bray–Curtis dissimilarities on the sites-by-traits matrix composed of community-weighted mean trait values was calculated. All values were multiplied by minus one such that the four sites from the continuously forested reference landscape had, on average, higher values than the fragmented sites. Functional integrity was then quantified as the difference between the value of each fragmented site and the average of the reference sites. Finally, we added one to all values to obtain a positive range and to assure that the closer the value approaches one, the more similar the functional integrity of a fragment is to the continuously forested reference landscape .
(e) Functional trait diversity
Functional trait diversity was assessed using two independent metrics: functional richness and evenness, because of the correspondence with common measures of species diversity [39–41]. Functional richness represents the amount of the functional space that is filled by the species in a community, and can, hence, be considered as an index of the number of functions in a community. It represents the potentially used/unused niche space . Functional evenness corresponds to how regularly species abundances are distributed in the occupied functional space, and it can thus be considered as a measure of how similar the abundances of the functions are in a community. It reflects the under- or overutilization of resources . We used the metrics proposed by Villéger et al. , because they are independent from each other (Pearson r = −0.02, p = 0.89), are easy to calculate  and are often used in the literature (e.g. [7,8,11]), which makes our results more comparable to other studies. Higher values for these metrics correspond to higher functional trait diversity. All traits were standardized to a mean of zero and a standard deviation of one before calculating these metrics . The first eight principal coordinates were used (representing 89% of the variation) to calculate functional richness, and functional evenness was weighted by relative species biomass.
We removed the effects of species richness from both metrics to convert them into pure measures of functional trait diversity (; see also the electronic supplementary material, S4). To this end, we used a simulation approach to create a null distribution of expected values for these metrics of functional trait diversity. Holding species richness constant, we generated 1000 random communities per site by swapping (n = 1000 swaps) species abundances within sites with species selected from the regional species pool . We obtained metrics of functional trait diversity that were independent of species richness by including the mean and standard deviation of the expected values in the formula of standardized effect size (SES = (observed – mean expected)/standard deviation of expected ).
(f) Community integrity of ecological groups
We developed an index of community integrity of ecological groups to examine whether habitat loss leads to a turnover of species with specific traits. Similar to other measures of community integrity (e.g. [28,38]), our metric compared species composition in the community against a reference condition (here, sites from the continuously forested landscape without known history of timber extraction), but specifically this metric allowed for the valuation of the integrity of a community with regard to particular functions. For each of the traits, we calculated the Bray–Curtis dissimilarity index on the sites-by-species matrix containing the trait values multiplied by the species' biomass. Data on body mass were extracted from Ramirez et al. , with body mass obtained from adults and averaged across both sexes for sexually dimorphic species. Species absent from a particular site and/or without a particular trait received a zero in this matrix, while large, abundant species with high trait values received the highest values. In two sites, no bird species were recorded eating nectar. As a meaningful dissimilarity index cannot be obtained for such ‘empty’ sites, they were removed from the matrix. Afterwards, the Bray–Curtis dissimilarities were submitted to an ordination. Community integrity was assessed for each ecological trait separately as the difference in the first component of the ordination between each fragmented site and the average of the four sites from the continuously forested reference landscape. If Bray–Curtis values of control sites were negative, all values for that particular trait were multiplied by minus one so control sites would have higher values, i.e. to represent higher integrity . Finally, we added one to all values to obtain a positive range and such that the mean of the control site had a value of one.
(g) Community specialization
First, we calculated a specialization index per species for the categories diet, foraging substrate and foraging stratum separately as the distance from the least specialized species (i.e. all importance values equal ‘2’):where xi is the species' importance value for a specific trait and n the number of traits within the relevant trait category. For example, we used trait values for invertebrates, vertebrates, fruits, nectar and seeds to calculate diet specialization, thus n = 5. For the category habitat, the index is derived from the number n of habitats in which a species lives  as (N − n)/(N − 1), where N is the total number of habitats occupied by at least one species in the community (N = 19). Values obtained with these formulae range between 0 (least specialized) and 1 (most specialized). Second, the resulting values were multiplied by the species' relative biomass for each of the sites and summed per category, obtaining four different values for community specialization: diet, substrate, stratum and habitat specialization. The trait ‘ground’ was not only included to calculate stratum specialization, but also to calculate substrate specialization to avoid computational problems as some species use no other foraging substrate than the ground. Although habitat specialization has no direct link with ecological function, it was calculated for comparison with other studies, which usually base specialization indices on habitat preference (e.g. [21,22]).
(h) Statistical analyses
General linear regression models were run to assess whether (i) functional trait composition, (ii) functional trait diversity, and (iii) species composition were affected by the proportion of forest cover within 300 m around the sampling site. Models were run separately for each trait (community-weighted mean trait values and community integrity of ecological groups), trait category (community specialization) or metric (functional integrity, richness and evenness). Mean specialization indices, such as our community specialization metrics, cannot test whether specialists and/or generalists effectively decline with habitat loss, potentially leading to erroneous conclusions when used on their own . We therefore ran additional analyses (electronic supplementary material, S5) showing that most species that were significantly affected by habitat loss declined. Because birds were sampled at multiple spatial locations, we also tested for spatial independence. Although models that accounted for spatial correlation were significantly better for some (but not all) of the metrics, parameter estimates and significance levels were very similar for models with and without spatial correlation structure (see the electronic supplementary material, S5). None of the statistical models violated the assumptions of linearity, homoscedasticity and normality of residuals. All statistical analyses were carried out in R v. 3.1.1 (R Development Core Team 2013). R codes for metric calculation and statistical analyses are available in the electronic supplementary material.
3. Results and discussion
We found that habitat loss in the Brazilian Atlantic Forest results in a change in functional trait composition (figures 1 and 2), while measures of functional trait diversity were either not affected, or were positively affected (figure 3). This change is owing to a shift in species composition: particularly, we observed both a turnover of species that exhibit the same functional trait (figure 4) as well as a replacement of specialist by generalists (figure 5). Hence, while our study demonstrates that communities from more deforested sites do not provide fewer functions in comparison with those that suffered less from habitat loss, the change in functional trait composition shows dramatic changes in terms of which functions are provided.
Functional integrity decreased with habitat loss (p = 0.003; figure 1; electronic supplementary material, table S3). In other words, the more deforested the landscape, the more functional trait composition deviates from that of communities found in the continuously forested landscape. Communities subjected to habitat loss thus seem to provide functions that are different from those provided in pristine habitats. Additionally, community integrity of most ecological groups (p < 0.04) and community specialization were lower at sites surrounded by less forest (p < 0.001; figures 4 and 5; electronic supplementary material, table S3). While species turnover and a replacement of specialists by generalists in response to habitat loss and disturbance have been often observed (e.g. [12,13,15,20]), as far as we know, this is the first study to show that such alterations in species composition lead to changes in functional trait composition. Given that alterations in species composition are often observed after habitat loss, such changes in functional trait composition are likely a widespread phenomenon.
Notwithstanding the modifications in functional trait composition, there is no loss of functional trait diversity, despite worldwide evidence that environmental changes, such as habitat loss, often lead to a loss of functional trait diversity in a variety of communities (e.g. [4–7], but see e.g. [8–11] for metrics of functional trait diversity that remained unchanged or were positively affected). More specifically, the number of functions performed in the community (i.e. functional richness) was not related to forest cover (p = 0.53), while the degree of similarity between the abundances of the functions present was even weakly positively influenced by forest loss (i.e. functional evenness; p = 0.03; figure 3; electronic supplementary material, table S3). The explanation is that some of the community-weighted mean trait values decreased while others increased with habitat loss (figure 2; electronic supplementary material, table S3), resulting in no, or even a positive, average effect on functional trait diversity. These results hence show that changes in functional trait composition can occur without a loss of functional trait diversity, similar to results obtained for species diversity versus composition [12–15]. Possibly, in communities where a loss of functional trait diversity has already been demonstrated, the change in functional trait composition is even greater than the one observed in this study.
The fact that community integrity of several ecological groups decreases significantly with habitat loss, sometimes without concomitant changes in community-weighted mean trait values (figures 2 and 4), confirms that studying species or functional trait composition in isolation may create a distorted view of the functional consequences of habitat loss. Nevertheless, modifications were observed in mean values of some of the traits. Average values of three traits (fruits, ground and canopy) increased at sites surrounded by less forest (figure 2; electronic supplementary material, table S3). As birds were sampled in the understory, the fact that more canopy-foragers were observed (p < 0.0001) may illustrate a rise in generalist species with forest loss. Alternatively, it could be the result of decreased foliage stratification and lower tree height in fragmented forest [47,48]. Similarly, the mean value for fruit consumers increased with forest loss (p = 0.004; figure 2; electronic supplementary material, table S3) despite the loss of community integrity within the fruit-consumer community (figure 4; electronic supplementary material, table S3). The rise in generalist species that include, but are not restricted to, fruits in their diet might partially compensate the decline of strict frugivores, which are typically very sensitive to deforestation (e.g. [49,50]). Thus, communities with higher integrity do not necessarily provide more functions, they just provide different functions. Average values of two traits (dead foliage and air), on the other hand, decreased with habitat loss, but this should not necessarily imply a loss of function, because specific ecological functions may no longer be required in a forest with a modified structure [11,48,51]. For instance, dead-foliage foragers decline with habitat loss (p = 0.005; figure 2; electronic supplementary material, table S3; ), but dead trees are often rare in secondary forest patches such as the ones included in this study .
Our study raises the question of how the functional structure of the Atlantic Forest bird community will change in cases of ongoing habitat loss below the range of forest cover included in this study (18–100%). We believe that communities from more deforested sites are relatively resistant to further environmental changes, having passed through a strong ecological filtering process, resulting in generalist-dominated communities. Generalist species such as the five thrush species recorded in our study areas (Turdus albicollis, T. amaurochalinus, T. flavipes, T. leucomelas and T. rufiventris) thrive in human-modified habitats, which suggests that they may continue to execute several functions in seriously degraded landscapes. Indeed, a theoretical study suggested that in variable environmental conditions, for instance in forest fragments owing to reduced microclimate buffering, generalists may have equal or greater ecological function than specialists [53,54]. Empirical study is needed to verify this suggestion.
In this study, we developed three new community indices (i.e. functional integrity, community integrity of ecological groups and community specialization) to investigate how habitat loss affects functional trait diversity, functional trait composition and species composition; community components that have rarely been investigated concurrently (e.g. ). As a consequence, we were able to perform a more robust assessment of the functional consequences of habitat loss than most other studies. Our approach can be applied to other taxonomic groups and ecosystems as well to improve our understanding of the functional effects of habitat loss and other environmental changes. Our study also provides opportunities for further in-depth investigations by considering combinations of trait categories, such as the consumption of invertebrates in the understory only, and by direct measurement of ecological functions, while the use of experimental set-ups will allow us to rephrase our results in terms of cause and effect relationships.
In conclusion, we show that habitat loss can drive changes in functional trait composition with little influence on functional trait diversity. Communities in the highly endangered Atlantic Forest biome thus not only experience large modification in species composition but also in functional trait composition (this study, [12,24]). Similar to the observation that species diversity remains largely unaffected by habitat loss , a loss of functional trait diversity was not observed (see also ). This study thus shows the importance of investigating changes in the diversity and composition of species and functions, because changes in the ecological functions performed in a community may remain undetected if only functional trait diversity is studied. This may lead to an underestimation of the negative effects of habitat loss and other major threats to biodiversity, and jeopardize the implementation of conservation actions.
All fieldwork complied with Brazilian ethical guidelines for animal welfare. Permission to capture birds was granted from CEMAVE-IBAMA, and permissions to work in the study sites were granted from Instituto Florestal and private landowners.
The datasets supporting this article have been uploaded as part of the electronic supplementary material.
G.D.C. designed the study, carried out the statistical analyses and drafted the manuscript. C.B.-L. and J.P.M. organized field data collection and helped draft the manuscript. All authors gave final approval for publication.
We have no competing interests.
Funding was provided by the São Paulo Research Foundation (FAPESP grant 2012/06866-7 and 2014/14746-7) to G.D.C., NERC grant NE/H016228/1 and a Marie Curie International Incoming Fellowship within the 7th European Community Framework Programme to C.B.-L., and research fellowship 307934/2011-0 from the Brazilian Science Council (Conselho Nacional de Desenvolvimento Científico e Tecnológico) to J.P.M.
- Community integrity of ecological groups
- similarity in community composition between a community of interest and a reference community with the same functional trait. It reflects the turnover of species that have a particular trait.
- Community specialization
- average degree of specialization of species within a community, weighted for the species' relative biomass (i.e. product of relative abundance and body mass).
- Functional integrity
- similarity in functional trait composition between a community of interest and a reference community. It reflects the turnover of functional traits.
We thank Maarten Mariën, Elizabeth Nichols, Luc Lens, Richard Inger and an anonymous referee for very helpful comments that improved the quality of this publication. This article is a contribution to Imperial College's Grand Challenges in Ecosystems and the Environment initiative.
- Received November 19, 2014.
- Accepted May 29, 2015.
- © 2015 The Author(s) Published by the Royal Society. All rights reserved.