Functional attributes change but functional richness is unchanged after fragmentation of Brazilian Atlantic forests

Study Area

This study was based in the state of Espírito Santo, in South-east Brazil. Within the region, we focused on the municipalities of Sooretama, Linhares and Jaguaré (18^o54′–19^o15′S and 39^o54–40^o15′W, 28–65 m.a.s.l) (Fig. 1), which contain a landscape matrix composed mainly of grasslands, and Eucalyptus spp., coffee, and papaya plantations (Rolim et al. 2005). The climate is tropical wet (Köppen classification), with an annual precipitation of 1403 mm and a distinct dry season from May to September, when precipitation is just 33 mm month⁻¹ (Peixoto & Gentry 1990). The predominant soil in the study region is Yellow Podzolic (IBGE 1987).

This region is part of the phytogeographic domain Atlantic Forest and is officially classified as Lowland Rain Forest (IBGE 1987) or Tertiary Tablelands Forest (Peixoto & Silva 1997). The study area is of high conservation importance due to the presence of two forest fragments larger than 20 000 hectares, which contain a high diversity of plant and animal species (Peixoto & Silva 1997; Chiarello 1999; Masden et al. 2001).

Tree Sampling

Fieldwork was conducted from January 2011 to January 2012. We created permanent plots along transects within 9 fragments ranging from 13.2 to 1318.3 ha in area (mean = 333.9 ha) and within two control forest blocks larger than 20 000 ha in Reserva Natural Vale (RNV) and Reserva Biológica de Sooretama (REBIO) (Table S1). Our sampling unit was a transect that contained ten 10 × 10 m plots at 20-m intervals, and we sampled each fragment with one edge and one interior transect. A major review of 33 abiotic and biotic edge parameters showed that 10 parameters penetrated less that 25 m from forest edge and that all but one parameter penetrated <300 m (Laurance et al. 2002). We thus stationed edge transects ~5 m inside the fragment and parallel to the forest edge to capture all edge parameters, and interior transects with distance ≥300 m from the nearest edge (noting that the smallest fragments did not permit a longer minimum distance from edge). We also allocated six transects within control areas: one edge and one interior transect in RNV, and two edges and two interior transects in REBIO, with a mean distance of 17.1 ± 10.4 km between transects. All plots were on the same soil type (Yellow Podzolic).

Within each plot, we sampled every living tree with a diameter at breast height (DBH) ≥4.8 cm at 1.3 m height and collected samples from each individual. We identified this material with reference to collections at the CVRD Herbarium of the Vale and the VIES Herbarium of the Federal University of Espírito Santo, and with aid from taxonomic experts in plant species identification in specific families (e.g. Myrtaceae and Sapotaceae). Botanical material collected in the fertile stage was deposited in the collection of Vale Herbarium of the Reserva Natural Vale in Linhares, ES.

Functional Trait Matrix

We use functional traits that are relevant to the morphological and physical adaptations of trees in their role as food resources, their dispersal and in carbon storage and forest structure (Tabarelli & Peres 2002; Bolmgren & Eriksson 2005; Laurance et al. 2006a,b; Bongers et al. 2009; Moran & Catterall 2010; Tabarelli et al. 2010). Within these three broad types of functional role, we had five categorical traits and one continuous trait, classified as: (i) fruit size, (ii) seed size, (iii) fruit type, (iv) fruit dispersal syndrome; (v) successional group and (vi) wood density (continuous variable).

Food Resources: Fruit and seed sizes for each of the species identified were classified into four categories according to Tabarelli & Peres (2002): small (<0.6 cm in length), medium (0.6–1.5 cm), large (1.6–3.0 cm) and extremely large (>3.0 cm). We used fruit and seed sizes as categorical variables because both metrics are poorly explored within the Brazilian Atlantic forest, such that many species lack sufficient precision to use as a continuous variable. To place species into categories, we used a combination of reliable measurements for the subset of fruits within the CVRD herbarium and the literature, supplemented by observational records (e.g. of seed scar size in fruits), descriptions from the literature and botanical knowledge (e.g. consistent family traits and from botanists at the CVRD herbarium). We also categorized the fruits into two types: (i) fleshy fruits (i.e. the pericarp can accumulate water and many organic compounds; see Coombe (1976)) and non-fleshy fruits.

Fruit dispersal syndrome: Fruits were classified as zoochoric or non-zoochoric following Van der Pijl (1982)A zoochoric tree produces diaspores surrounded by fleshy pulp, an aryl or other features that are typically associated with dispersal by animals, and a non-zoochoric tree has characteristics that indicate dispersal by abiotic means, such as winged seeds, feathers or a lack of features that indicate dispersal via methods other than downfall or explosive indehiscence.

Carbon storage and forest structure: We classified species into the successional groups defined by Bongers et al. (2009). We considered as pioneers those trees that develop in conditions of high light and generally do not occur in the understorey, as early secondary those trees that develop in intermediate shading conditions and as late secondary those trees that develop exclusively in permanently shaded understorey. Species were classified using the data base by Jesus & Rolim (2005) from the Reserva Natural da Vale. Data for wood density in dry weight (g cm⁻³) were obtained from The Global Wood Density (GWD) data base in the subsection Tropical South America (http://datadryad.org/repo/handle/10255/dryad.235, Chave et al. 2009; Zanne et al. 2009). We made two adjustments (following Flores & Coomes 2011; Hawes et al. 2012): for morphospecies only identified to the family or genus level, we used the average wood density of the taxonomic group; and for species not in the GWD data base, we used the average wood density for the species' genus.

Species identified at morphospecies level represented only 1.1% of species richness and 0.2% of total abundance. These species were not treated in any of the functional traits described above, being considered only in the analysis of species richness and community structure.

Data Analysis

We used non-metric multidimensional scaling (NMS) ordination analysis in the PC-ORD 6 package (McCune & Mefford 2011) to identify changes in community structure between different-sized fragments, and between edge and interior habitats. We used the raw species abundance data from each plot and the Sørensen (Bray–Curtis) distance. We considered the NMS results arising from tree species abundance data as a measure of community structure (Barlow et al. 2010), with the number of individuals representing abundance.

To analyse functional diversity, we used three indices proposed by Villéger, Mason & Mouillot (2008): functional richness (FRic), functional evenness (FEve) and functional divergence (FDiv). FRric represents the volume of space of a functional convex hull occupied by the community, FEve the regularity of the distribution in abundance on this volume and FDiv the divergence in the distribution of species characteristics within the volume occupied by each functional trait. To calculate these three indices, we used methods and scripts from Villéger, Mason & Mouillot (2008), in R, version 2.15.1 (R Development Core Team. 2012.

To investigate fragmentation effects on community structure and functionality of tree species, we considered two explanatory variables: (i) fragment size as a continuous variable and (ii) habitat type with two levels (interior and edge). We also applied an interaction between size and habitat type to investigate possible differences in the relationship with fragment size at edges and interiors. We used the raw species abundance data from each transect, and generalized linear mixed models were carried out using the glmmadmb function from the glmmadmb package. We used negative binomial error distributions for count data, since our data showed significant overdispersion, and a Gaussian error distribution for the remaining data. The sites (each fragment) were codified as a random variable in all analyses (Bolker et al. 2009). We used the dredge function from the mumin package to test all possible combinations of the variables included in the global model. To determine which of these explanatory variables were the most decisive in possible changes in species richness, community structure, functional diversity and functional trait attributes, we used an information theoretical approach based on the Akaike Information Criterion of Second Order (AICc), which is indicated for small sample sizes, and the best model was indicated by the AICc lower value (Burnham, Anderson & Huyvaert 2011). All analyses were performed in the R, version 2.15.1 (R Development Core Team. 2012).

Fragmentation and Community Organization

Across all fragments, we sampled a total of 4140 tree individuals belonging to 443 species (Table S2). Considering only the best model selected on the basis of their AICc values, species richness at the transect level was significantly influenced by the interaction between fragment size and edge versus interior habitat (GLZ: Z = 2.52; P = 0.03), where increasing fragment size had a significant negative effect on species richness in interiors (F = 6.95: P = 0.02; Fig. 2a) but no significant effects near edges (F = 1.89: P = 0.1982; Fig. 2b). Species richness was significantly higher within fragment interiors than at fragment edges (GLZ: Z = 2.79; P = 0.02, Fig. 2b).

NMS analysis of species composition and abundance (community structure) parameters indicated significant changes in community structure for the axes 1, 2 and 3 (P = 0.02; Fig. S1). However, GLZ models revealed that only axes 1 and 2 of tree community structure were significantly influenced by fragments' size and by edges. The best model selected by AICc for axis 1 indicated that changes in community structure were strongly influenced by the creation of an edge habitat (GLZ: Z = 3.84; P < 0.01; Fig. 2c). For axis 2, the best model included the interaction between fragment size and habitat (GLZ: Z = −4.05; P < 0.01; Fig. 2d), with changes significantly related to fragment size in the interior (F = 95.95: P = 0.01; there was no significant effect of edge versus interior habitat, F = 7.76: P = 0.37).

Fragmentation and Functional Diversity

We found no significant effects of edge versus interior habitat (GLZ: Z = −0.46; P = 0.66), fragment size (GLZ: Z = −0.78; P = 0.45) or the interaction between habitat type and fragment size (GLZ: Z = 0.38; P = 0.71) on functional richness, indicating that forest fragmentation causes no overall loss in functional richness in this landscape. Functional evenness was, however, significantly negatively related to fragment size (GLZ: Z = −2.23; P < 0.05; Fig. 3a), indicating that the evenness of traits is less heterogeneous in larger than smaller fragments. Functional evenness was also significantly higher in fragment interiors versus edges (GLZ: Z = 4.8; P < 0.001, Fig. 3b). Functional divergence showed a significant negative relationship with fragment size (GLZ: Z = 2.1; P < 0.04; Fig. 3c), demonstrating that smaller patches are more divergent (i.e. have less functional redundancy) than larger fragments in the landscape. There was no effect of edge versus interior on functional divergence (GLZ: Z = −0.87; P = 0.4). Changes in functional evenness and functional divergence thus indicate that, despite finding no significant impact of fragmentation on functional richness, there are shifts in the provision of different functional trait attributes following fragmentation.

We investigate this variation in our fifteen functional trait attributes using two metrics: the richness of species exhibiting a particular functional trait attribute and the abundance of individuals exhibiting a particular functional trait attribute. For the richness of species exhibiting particular functional trait attributes, we find that edge versus interior habitats had the largest effect, being a significant predictor in 6 of 15 of the best models proposed by AICc (Table 1). Proximity to forest edges had a significant negative influence on the richness of tree species with zoochoric dispersion, fleshy fruits and later secondary species (i.e. these traits had higher species richness in interiors), and a significant positive influence on the richness of non-zoochoric dispersed pioneer and initial secondary species (Table 1). Fragment size, and the interaction between size and habitat, had a significant negative effect on the richness of tree species with very large fruits (Table 1 and Fig. S2), with the other 14 functional trait attributes not selected in size and size*habitat best models.

Edge versus interior was also prominent for the abundance of individuals exhibiting a functional trait attribute (significant predictor in 9 of the 15 best models), negatively influencing the abundance of zoochoric dispersion, fleshy fruits, very large, medium and small fruits, medium seeds, and late secondary species, and positively influencing the abundance of non-zoochoric dispersers and pioneers. Fragment size (significant predictor in six of the 15 best models) showed a negative influence on the abundance of very large-fruited trees species and a positive influence on zoochoric dispersion, fleshy fruits, medium fruits, and large and medium seeds trees species. The interaction between fragment size and edge versus interior habitat had a significant, positive influence on the abundance of very large fruits and initial secondary species (Table 1, Fig. S3), with the remaining 13 functional trait attributes not selected. Wood density did not respond significantly to fragment size, edge versus interior habitats or the interaction between size and habitat (Table 1).

Fragmentation and Community Organization

Tree species richness in the interiors of smaller fragments was greater than that in the interior of larger fragments of Brazilian Atlantic forests in our study, and a similar pattern was apparent for the richness of tree genera in forest fragments in Amazonia (Michalski, Nishi & Peres 2007). This contrasts with other studies that showed that while larger fragments typically retain the highest overall species richness (e.g. Laurance et al. 2002, 2011), there was no difference between fragment size and species richness per unit of sample area (Laurance et al. 2006b). One possibility is that there is a lot of spillover of species into the interiors of small fragments, both from the non-forest matrix (Cook et al. 2002) and from large fragments or contiguous forest sources (Brudvig et al. 2009), the latter spillover determined in part by landscape configuration and connectivity (Brudvig et al. 2009). In our study, the small- and medium-sized fragments were often close to large fragments and the continuous forest blocks, and there was high connectivity among them (Fig. 1). Another reason for high species richness in small fragments was that they contained species typical of both forest interiors and matrix areas (e.g. equivalent abundances of zoochoric species and non-zoochoric dispersion species), whereas larger fragments were dominated by zoochoric species.

Across all fragment sizes, we found lower species richness at fragment edges than interiors, a pattern found in other fragments of Brazilian Atlantic forest (Oliveira, Grillo & Tabarelli 2004; Lopes et al. 2009) and in other tropical regions (Saunders, Hobbs & Margules 1991). It is likely that changes in abiotic conditions at fragment edges, such as increased wind speed and air temperature, and decreased humidity (Laurance et al. 2002, 2011), explain these biological edge effects.

We observed significant shifts in community structure near fragment edges and in fragments of varying sizes. These shifts are likely related to the abundance of pioneer and early secondary species, which can lead to a rapid change in forest structure, species composition and ecological functionality (Laurance et al. 2006a,b; Bongers et al. 2009; Tabarelli et al. 2010), and are apparently driven by proximity to forest edge rather than by fragment size per se (Table 1). For instance, Amazonian forest fragments tend to have a greater number and abundance of successional species near edges than in interiors of even small fragments (Laurance et al. 2006a,b; Michalski, Nishi & Peres 2007), and these influence the structure and dynamics of forest patches (e.g. Laurance et al. 2006a,b; Tabarelli et al. 2010).

Fragmentation and Functional Diversity

Among the three indices that describe functional diversity, those accounting for the abundance of species (functional evenness) and the dominance of a functional group over other functional groups (functional divergence) were more sensitive to fragmentation effects than was functional richness, which is more influenced by species richness (see Villéger, Mason & Mouillot 2008; Mouchet et al. 2010). In our study, functional richness remained constant with fragment size and at edges versus interiors; thus, fragmentation does not appear to alter the volume of functional space occupied by species within different communities. Small fragments of forest in the Australian subtropics also retained similar levels of functional richness to large fragments (Kooyman et al. 2013). However, in our study, the retention of functional richness with fragmentation is underpinned by species that exhibit different functional trait attributes (Table 1, see below), indicating shifts in ecological functionality that can modify the diversity of interactions between plants and animals (Moran & Catterall 2010).

A high intensity and frequency of disturbances at fragment edges (Murcia 1995) may explain reductions in functional evenness compared with fragment interiors. Declines in functional evenness indicate that some parts of the functional space within edges decline or disappear (e.g. Mouchet et al. 2010). At fragment edges, the reduction in functional evenness was related to the loss of important functional groups, such as zoochoric dispersion, fleshy fruits and later secondary species, balanced to some degree by an increase in non-zoochoric and pioneer species. Such shifts are likely to make edges functionally more homogeneous, leading to declines in food resources for frugivorous birds and mammals, and reducing forest regeneration potential via dispersal (Bolmgren & Eriksson 2005; Moran & Catterall 2010). In turn, a loss of ecological functionality is likely to have secondary consequences for faunal and floral richness (Fig. 1; Tabarelli & Peres 2002; Laurance et al. 2002; Oliveira, Grillo & Tabarelli 2004; Laurance et al. 2006b).

Lower functional divergence and functional evenness in the interiors of large versus smaller fragments suggest niche homogenization between species (Mouchet et al. 2010), with most abundant species being functionally similar and exhibiting higher competitiveness, coupled with low prevalence or absence of some parts of functional space. Reductions were accompanied by an increase in the abundance of fleshy fruit, zoochoric dispersed species, and a decrease in abundance of initial secondary species, implying greater resources and interactions with fauna in large fragments (see Bolmgren & Eriksson 2005; Bongers et al. 2009). Indeed, in our study landscape, mammal communities shifted from one dominated by medium- and large-bodied frugivorous species in large fragments to one dominated by herbivorous mammals in small fragments (Chiarello 1999). Low functional divergence also suggests that there is higher functional redundancy in large versus smaller fragments. This means that there is a degree of resilience for key ecosystem functions in large fragments, since there is likely to be a functionally equivalent species in the tree community if another is lost (Laliberté et al. 2010). However, reductions in patch size apparently reduce functional redundancy and thus resilience, with the potential for the long-term erosion of ecosystem functions in small fragments.

Despite the significant increases in the species richness and abundance of early successional trees at forest edges and in the interior of smaller fragments, we found no significant effects of fragmentation on wood density. Thus, the apparent lack of response of wood density to fragmentation suggests that even small fragments or fragment edges can be managed to play an important role in carbon storage. However, since wood densities of species can be affected by environmental changes due to fragmentation (e.g. Laurance et al. 2006a,b; Thomas, Montagu & Conroy 2007; Nock et al. 2009), the average wood densities we used from the literature (Chave et al. 2009; Zanne et al. 2009) would miss any such variation. Our functional diversity analyses could be refined in future analyses using wood densities derived from in situ measurements, and, when herbarium collections permit, fruit and seed sizes as continuous rather than categorical variables.

Conclusions and Implications for Conservation

The changes in functional diversity, functional trait attributes, species richness and community structure at forest edges and in smaller fragments have important implications for conservation in fragmented landscapes. Although our fragments are typically ~40 years old, we cannot rule out the possibility that there remains an extinction debt in smaller fragments and at edges, especially given the longevity of many tree species in tropical rain forests (Chambers, Higuchi & Schimel 1998; Laurance et al. 2004), that could further degrade the functional value of these habitats over far longer time-scales (but see Kooyman et al. 2013). Additionally, tree populations in smaller fragments may be sustained to some extent by seed dispersal from larger blocks of natural habitat, and we highlight understanding the impacts of landscape connectivity on ecosystem functioning in fragmented systems as a key research frontier. Hence, the functional diversity of small fragments and edges may change over time or if other forest fragments were removed from the landscape. Consequently, protecting large forest fragments from conversion and interiors from degradation must remain top conservation priorities.

Nonetheless, our results support previous studies suggesting that even small forest patches can retain high conservation value, especially within highly threatened biomes, such as the Brazilian Atlantic forest (e.g. Arroyo-Rodriguez et al. 2009; Gardner et al. 2009; Santos et al. 2010). First, we found similar communities in the interior of small and large fragments, suggesting that small fragments could represent important reservoirs of forest specialist trees and aid seed dispersal or connectivity across landscapes. Secondly, the retention of functional diversity within small fragments and edges was maintained by non-zoochoric pioneer and initial secondary tree species. These species are excellent dispersers and, in the event of agricultural abandonment or land purchases to reconnect forest fragments (Martínez-Garza & Howe 2003; Cortines & Valcarcel 2009; Simmons, Wu & Whisenant 2011; see also savingspecies.org), they could play important roles as sources of seeds in the recovery of early secondary forests. Thirdly, the apparent absence of change in wood density across the fragmentation gradient suggests that even small fragments can be important carbon stores, with potential cobenefits between carbon market payments, ecological services and biodiversity protection (Díaz, Hector & Wardle 2009; Phelps, Webb & Adams 2012).