Influence of matrix type on tree community assemblages along tropical dry forest edges†
This research was supported by grants from the Consejo Nacional de Ciencia y Tecnología, Mexico (SEP-CONACYT 83441-R and SEP-CONACYT 174094) to J.B-M and from Posgrado en Ciencias Biológicas, Universidad Nacional Autónoma de México (UNAM) to J.C.G-V. The authors thank the Estación de Biología Chamela for granting permits to carry out our research. We are grateful to S. Araiza-Méndez and A. Verduzco for assistance in the field, and to J. M. Lobato and R. Ahedo for technical support provided.
Abstract
• Premise of the study: Anthropogenic habitat edges have strong negative consequences for the functioning of tropical ecosystems. However, edge effects on tropical dry forest tree communities have been barely documented.
• Methods: In Chamela, Mexico, we investigated the phylogenetic composition and structure of tree assemblages (≥5 cm dbh) along edges abutting different matrices: (1) disturbed vegetation with cattle, (2) pastures with cattle and, (3) pastures without cattle. Additionally, we sampled preserved forest interiors.
• Key results: All edge types exhibited similar tree density, basal area and diversity to interior forests, but differed in species composition. A nonmetric multidimensional scaling ordination showed that the presence of cattle influenced species composition more strongly than the vegetation structure of the matrix; tree assemblages abutting matrices with cattle had lower scores in the ordination. The phylogenetic composition of tree assemblages followed the same pattern. The principal plant families and genera were associated according to disturbance regimes as follows: pastures and disturbed vegetation (1) with cattle and (2) without cattle, and (3) pastures without cattle and interior forests. All habitats showed random phylogenetic structures, suggesting that tree communities are assembled mainly by stochastic processes. Long-lived species persisting after edge creation could have important implications in the phylogenetic structure of tree assemblages.
• Conclusions: Edge creation exerts a stronger influence on TDF vegetation pathways than previously documented, leading to new ecological communities. Phylogenetic analysis may, however, be needed to detect such changes.
Edge effects are the major drivers of change in many tropical fragmented landscapes and represent an inevitable and important consequence of habitat loss and fragmentation (Laurance et al., 2007). Time since edge creation, extreme natural events (e.g., windstorms and fires), and the management, structure and composition of the adjoining vegetation matrix influence the incidence and magnitude of edge effects on tree communities (Ries et al., 2004; Harper et al., 2005; Laurance et al., 2007). As time elapses and edges get older, tree species assemblages, composition, structure and functioning change in relation to the original vegetation and to the interior forest (Laurance et al., 2006).
Most information concerning edge effects on tropical forests, however, comes from the wet tropics, and there is scant information on the impact of edge creation for other tropical systems (e.g., dry forests in Janzen 1983, 1986, 1988; Sampaio and Scariot, 2011). No evidence so far exists regarding the influence of the matrix type on tropical dry forest (TDF) tree communities. Compared to tropical rain forests, TDF show lower height and canopy thickness, smaller diameter of stems and branches, reduced stratification, and high density of trees (Mooney et al., 1995; Murphy and Lugo, 1995). In addition, many TDF trees are physiologically acclimated to heat and desiccation stress, and most species simply drop their leaves during the dry season (Mooney et al., 1995). Consequently, TDF trees seem less susceptible to pervasive edge effects than are trees in the rain forests (Sampaio and Scariot, 2011). In addition, the standard metrics used in community ecology to detect structural and compositional changes along edges (e.g., species richness, diversity, basal area, etc.) are not completely satisfactory for tree assemblages in the dry tropics (Sampaio and Scariot, 2011; Gallardo-Vásquez, 2012). Therefore, the use of other analytical tools such as phylogenetic biology are more appropriate to detect otherwise unnoticeable ecological changes (Webb, 2000; Cavender-Bares et al., 2009).
Fragmentation can drive tree species to local extinction, potentially changing the phylogenetic structure, composition and diversity of the native forest (Santos et al., 2010). Vulnerability to extinction is taxonomically dependent, with nonrandom extinctions being important predominantly in species-rich tree communities with high levels of endemism and species turnover (Vamosi and Wilson, 2008) such as tropical dry forests (Balvanera et al., 2002). Empirical information regarding edge effects on the phylogenetic diversity, structure and composition of tree assemblages in fragmented TDF landscapes is incomplete or missing.
In the current study, we tested whether the type of bordering vegetation matrix (e.g., gradual or abrupt edges) influences the phylogenetic relatedness between tree species in forest interiors and at forest edges within the Chamela-Cuixmala Biosphere Reserve, Mexico. In the same study region, tree communities in secondary forests of distinct ages are structured by a few phylogenetically closely related families, whereas old-growth forest tree communities are dominated by late-successional species from many distant families (Kembel and Hubbell, 2006; Kelly et al., 2008; Álvarez-Añorve et al., 2012). On the basis of this information and the fact that early successional species tend to dominate forest edges, we hypothesized that edge creation has an impact on the pathways of tree community assemblages, and that the magnitude of such changes differs according to matrix type.
MATERIALS AND METHODS
Study area
The study took place in the Chamela Biological Station (EBC) located in the central western coast of Jalisco, Mexico. The EBC is part of the Chamela-Cuixmala Biosphere Reserve covering a total area of 13145 ha of TDF associated with arroyo vegetation (Sánchez-Azofeifa et al., 2009). The landscape is characterized by low hills (120−250 m elevation) with little flat terrain. The average annual rainfall is 718 mm (350−1200 mm) with 80% of the rainfall occurring between July and November.
Tropical dry forest is the most extensive vegetation type in the region and is considered one of the most diverse in the Neotropics (Gentry, 1995). The flora of Chamela has at least 1,149 vascular plant species (10% endemic) in 572 genera and 125 families (Lott et al., 1987). The plant families with the greatest number of species are Bignoniaceae, Euphorbiaceae, Fabaceae and Rubiaceae (Lott, 2002). Average tree height is 10 m with a high stem density (trees 5−10 cm diameter at breast height (dbh)), but there are a low number of trees ≥25 cm dbh (Segura et al., 2003). The conversion of old-growth forest to cattle pastures is the main cause of TDF transformation (Maass, 1995). Anthropogenic activities within and around the Chamela-Cuixmala Biosphere Reserve have resulted in different vegetation matrices abutting the protected area (e.g., Sánchez-Azofeifa et al., 2009). These edges range from gradual (e.g., old fields and plantations) to abrupt (e.g., open pastures) and are sometimes exposed to additional perturbations, such as fires, herbicides, pesticides and cattle grazing and trampling.
Edge type and tree community structure
In the perimeter of the EBC, all trees ≥5 cm dbh within twelve 0.1 ha (50 × 20 m) blocks parallel to the forest edge were sampled and identified to species (following Lott, 2002). The blocks were located in four different habitats: (1) three blocks in forest edges abutting disturbed TDF vegetation with intruding cattle, (2) three blocks in edges abutting pastures with intruding cattle, (3) three blocks in forest edges abutting pastures without cattle and (4) three blocks in old-growth interior forest (control). Blocks in forest interiors were more than 500 m away from and parallel to the nearest forest edge and at least 100 m from each other, while edge plots were at least 100 m apart. Forest interior blocks were situated within thousands of hectares of continuous forest (Fig. 1). All forest clearings that originated the different edge-matrix types were at least 30 y old at the beginning of the study and maintained as clearings thereafter. All study sites were in an area of ca. 620 ha (Fig. 1). Old-growth forest interior blocks and different edge types presented lower similarity in tree species composition (Jaccardśs Index; Magurran, 2004) than did different edge types. Differences in species similarity were not correlated with distances between habitat types (Fig. 1; Gallardo-Vásquez, 2012).
Map of the Chamela-Cuixmala (Mexico) region showing the location of the 12 study blocks (0.1 ha each) along tropical dry forest edges abutting different vegetation matrices. All trees within each block were identified to species. For tree species, Jaccard indexes of similarity between habitat pairs are indicated beside the lines. Habitat-pair comparisons of tree species similarity between preserved forest and edge types are indicated by continuous lines; those between edge types are indicated by dotted lines. Abbreviations: PC, forest edges abutting pastures with cattle; PA, forest edges abutting pastures without cattle; DV, forest edges abutting disturbed dry forest vegetation with cattle; PF, old-growth forest interiors and/or preserved forest.
The disturbed TDF vegetation abutting the forest edge is similar in height to the forest interior blocks. However, the understory is highly disturbed due to the extraction of nontimber forest products, and cattle grazing and trampling. In the study region cattle damage vegetation and soil properties, impacting forest regeneration (García-Oliva et al., 1999). Cattle can also act as surrogate seed dispersers for native, exotic and invasive plant species that may change successional pathways in the dry forests (Janzen and Martin, 1982). Apart from cattle intrusion, no extra pressures on the vegetation such as fires and selective logging were observed at the forest edge. In forest edges bordering pastures without cattle, no cattle have been allowed for at least 15 y prior to the current study. These pastures are used to grow fodder to feed cattle in nearby pastures; once grasses are harvested, the pastures are occasionally burnt (every 2 to 3 y) to promote the growth of new grasses and prevent forest succession, as is the case for the pastures with grazing cattle (Gallardo-Vásquez, personal observations).
In the pasture with cattle matrix, cattle are present for most part of the year (ca. 8 mo). In addition, an exotic tree species, Mimosa arenosa (Willd.) Poir (Fabaceae), was found in forest edges abutting this matrix type, though at small diameter classes (1–4.9 cm dbh) and therefore was not considered in the analysis. This legume is an invasive species of old fields and is known to arrest forest succession in the study region (Romero-Duque et al., 2007). Another type of disturbance is hurricanes. Although infrequent in the study region, their occurrence damages the vegetation within the EBC reserve and along forest edges (Benítez-Malvido, personal observations). Around 4% of adult trees in continuous forest were damaged by hurricane Jova in October 2011 (R. Ahedo, UNAM, unpublished data).
Edge effects in the dry forest of Chamela
The empirical evidence so far indicates that for different parameters edge effects at Chamela reach up to 20 m toward interior forest. Temperature and photosynthetically active radiation increase toward the edge, whereas soil infiltration decreases (Nava-Cruz et al., 2006). Furthermore, edge creation disrupts P dynamics within the first 10 m toward forest interior, which could have implications for nutrient intake and productivity (Toledo-Aceves and García-Oliva, 2008). In the study habitats, no significant edge–interior differences in juvenile and adult tree density (1–4.9 cm and ≥5 cm dbh, respectively) and in the basal area have been detected. Species richness and diversity of adult trees were greater in forest interior blocks only compared to the disturbed vegetation-cattle edges (Table 1 and Gallardo-Vásquez, 2012). For many individuals with a small dbh, taxonomic identification was uncertain (with many identified only as morphospecies) and therefore only large trees ≥5 cm dbh) were considered in this study.
| Habitat type | Number of observed species | Shannon index of diversity | Number of individuals/ha | Mean (± SD) basal area (m2/ha) |
|---|---|---|---|---|
| Forest interior | 72b | 3.6b | 1270 | 16.6 (1.8) |
| Disturbed vegetation with cattle | 48a | 3.1a | 1187 | 16.7 (3.4) |
| Pasture without cattle | 57ab | 3.4ab | 1133 | 16.4 (2.0) |
| Pasture with cattle | 52ab | 3.5ab | 1220 | 14.7 (4.2) |
- Note: Each habitat type was replicated three times (n =12). Significant differences are indicated with different letters (GLIM, t ≥2, P < 0.05).
Data analyses
We characterized tree communities along a disturbance gradient in terms of the following: (1) species composition, (2) phylogenetic composition, and (3) phylogenetic structure. Previous data analysis showed that generally species density of adults did not differ significantly among habitat types, and therefore species density did not affect the results of our analysis (Table 1). In addition, we compared the proportion of early- and late-successional tree species among habitats with GLIM for binomial errors following Crawley (1993).
Tree species composition
We evaluated among-habitat differences in terms of species composition by mapping dissimilarities in plant assemblages with relation to species identity and abundance through a nonmetric multidimensional scaling ordination (NMDS) (Kruskal, 1964). The NMDS is an iterative method of ordination considered to be one of the most effective for analyzing community data (McCune and Grace, 2002). For this analysis, we employed a matrix containing the number of sampled individuals for each species in each sampling plot (n = 12). Based on this matrix, we then built a matrix of distances among plant assemblages using the Bray-Curtis coefficient (Magurran, 2004). Prior to the conformation of the distance matrix, for all sites we log-transformed each value of species abundance to reduce the bias toward the species with the largest differences in abundance (Kindt and Coe, 2005).
We employed the stress value, expressed on a scale from 0–100, to evaluate how successfully the distances between sites in the ordination space reflect the between-sites distances in the original space (distance matrix). Lower values of stress indicate a more reliable ordination. Finally, we used the “envfit” function from the R package “vegan” to test for among-habitats differences (Oksanen, 2011; Oksanen et al., 2012). This function calculates the centroids for each habitat type in the ordination space, and evaluates whether the observed difference among centroids is greater than expected by chance. The significance (alpha = 0.05) of the differences was evaluated through a randomization test (10000 permutations).
Phylogenetic structure analysis
For the ecophylogenetic analysis of plant assemblages we built a phylogenetic tree including all surveyed plant species (Fig. 2). The phylogenetic tree was built by using the desktop Phylomatic (Webb et al., 2008; available at website http://svn.phylodiversity.net/tot/megatrees/ [accessed 15 January 2012]), utilizing as bone tree the R20100701 maximally resolved super tree of angiosperms. We assigned the branch length of the resultant phylogenetic tree with the BLADJ function implemented in the Phylomatic, using as a reference the node age database of Wikström et al. (2001). The BLADJ function sets the unknown branch lengths by evenly placing the undated node between dated nodes or between dated nodes and the terminals. This algorithm minimizes the variance in branch length within the constraints of dated nodes. The final tree is ultrametric with branch lengths in units of time (millions of years). In subsequent analyses we did not consider the species Gyrocarpus jatrophifolius Domin (Hernandiaceae) which was an extreme outlier (poorly phylogenetically related to the rest of the species, with a very long branch length, and very rare, represented by one individual; see Dinnage, 2009).
Chronogram containing all plant species sampled along tropical dry forest edges and interior forest. Branch lengths represent millions of years. The genera and species names are separated by an underscore. The nodes representing the most specious plant families and genera (represented by at least two species) are marked with a number as follows: Anacardiaceae (1), Apocynaceae (2), Capparaceae (3), Cinchonoideae (4), Euphorbiaceae (5), Fabaceae (6), Malvaceae (7), Moraceae (8), Polygonaceae (9), Rubiaceae (10), and Salicaceae (11). In addition, plant genera represented by at least two species include the following: Bursera (12), Caesalpinia (13), Capparis (14), Ceiba (15), Cordia (16), Croton (17), Jatropha (18), Lonchocarpus (19), Randia (20), Tabebuia (21) and Trichilia (22). Scale bar = 20 myr.
Phylogenetic composition analysis
We distinguished between phylogenetic composition and phylogenetic structure analysis following Dinnage (2009). Phylogenetic composition analysis incorporates the phylogenetic relatedness information into traditional methods of community analyses (i.e., ordinations), while phylogenetic structure analysis summarizes the phylogenetic information contained in the communities and is analogous to the traditional measures of species diversity. Basing ordination and classification methods on intersample distances that reflect net phylogenetic dissimilarity, rather than on Euclidean distance in N-dimensional species space, offers a means to display the phylogenetic relations among sample blocks. Such methods can reveal meaningful ecological relationships hidden by standard, nonphylogenetic methods; blocks sharing many genera should still cluster even if they share none of the same species (Webb et al., 2008).
We compared tree assemblages associated with the different habitat types in terms of their phylogenetic composition by mapping their dissimilarities (phylogenetic distances) in relation to their phylogenetic nodal structure in an ordination space (NMDS, explained under Tree species composition above). In this analysis the species are not considered as independent entities. What is considered instead is the species degree of phylogenetic relatedness (Dinnage, 2009). For this analysis we generated a matrix where each column represents a node of the phylogenetic tree, and each row represents a study site. We then, filled the matrix, considering how many species representing each node where found on each site. This information was obtained with the R package “caper” (Orme et al., 2012).
Based on the generated matrix, we built a distance matrix, used as an input for the NMDS. For the construction of the distance matrix we employed the Bray-Curtis coefficient, which considers in its calculation not only which node is represented in each sampling site, but also how many species are representing the node. Before the conformation of the distance matrix, we standardized the original matrix by log-transforming each value. This transformation reduces the bias of the Bray-Curtis coefficient produced by the nodes with the largest among-sites differences in terms of the number of species representing them (Kindt and Coe, 2005). To test for differences among plant assemblages associated with different habitat types, we employed the “envfit” function from the R package “vegan” (Oksanen et al., 2012). The significance (alpha = 0.05) of the differences among the centroids for each type of habitat (explained under Edge type and tree community structure above) was evaluated with 10000 permutations.
For the comparison of plant assemblages in terms of their phylogenetic structure, we employed the two metrics proposed by Helmus et al. (2007a) which incorporate phylogenies into typical measures of community structure: species variability and species evenness. For the conformation of these metrics, Helmus et al. considered the values of some hypothetical unselected trait (neutral) that evolves randomly and independently among separate phylogenetic lineages (Brownian motion model; sensu Felsenstein, 1985). They then used the branch lengths of the community phylogenies as an indicator of the expected variance in the trait value for each species, determining in this way the degree of covariance among the species trait values, and the degree of relatedness among species.
The phylogenetic species variability metric (PSV) measures the degree of phylogenetic relatedness among the species in the community. The values of this metric are limited to between 0 and 1, and tend to approach 0 when the variability in the hypothesized trait decreases, indicating an increase in the degree of relatedness among species. On the other hand, the phylogenetic species evenness metric (PSE) is a transformation of PSV to incorporate species abundances. This metric is considered both a measure of phylogenetic and species evenness. The values are between 0 and 1, reaching the maximum when all species are phylogenetically even, and they also have even abundances. Both metrics (PSV and PSE) are useful for comparisons among different habitat types because their values are independent of species richness, which is likely to differ among habitats (Helmus et al., 2007a, b).
We analyzed the phylogenetic structure of tree communities along edges and interior forest in two ways. Firstly, we compared PSV and PSE raw values among habitats through an analysis of variance (ANOVA). This analysis allowed us to evaluate the overall pattern of phylogenetic structure among habitat types (Dinnage, 2009). Secondly, we evaluated how PSV and PSE means (considering all sampling sites) deviate from their expected values based in null model distributions that consider the same species pool for each habitat. This analysis allowed us to make inferences about the forces determining the structure of plant communities (Helmus et al., 2007a). We generated the PSV and PSE distribution of means by using two null models (following Helmus et al., 2007a): (1) a model maintaining the species frequency of occurrence, but making each community have the same expected species richness; and (2) a model maintaining the sampled species richness, but making each species have the same expected frequency of occurrence. Metrics calculation and statistical analysis were performed with R (R Development Core Team, 2011) with the help of the package “picante” (Kembel et al., 2010).
RESULTS
Overall, compared to forest interiors, edge exposure changes the phylogenetic composition of tree communities in the tropical dry forest of Chamela. Furthermore, the magnitude of such divergences among edge types and forest interiors was associated with the presence of cattle.
Tree species composition
We considered only three axes of the plant assemblage ordination in terms of their species composition (Fig. 3) because additional dimensions did not substantially diminish the stress value (6.402). According to Kruskal and Clarke's rule of thumb for interpreting the final stress (McCune and Grace, 2002), the stress value corresponding to the final ordination (6.402) supports a good ordination with no real risk of drawing false inferences (for stress values ranging from 5 to 10).
Ordination of tree species assemblages along tropical dry forest edges abutting different vegetation matrices compared to forest interiors in terms of their tree species composition. NMDS1–NMDS3 are axes resulting from the nonmetric multidimensional ordination. Abbreviations: PC, the centroid in the ordination of forest edges abutting pastures with cattle (n = 3); PA, the centroid in the ordination of forest edges abutting pastures without cattle (n = 3); DV, the centroid in the ordination of forest edges abutting disturbed dry forest vegetation with cattle (n = 3); PF, the centroid in the ordination of old-growth forest interiors and/or preserved forest (n = 3).
The axis showing the higher variation among sites (axis 1in Fig. 3) evidences a gradient of cattle influence on the species composition of tree assemblages: Assemblages with higher influence of cattle (PC, pasture and cattle raising) tended to present lower scores while those not affected by cattle (PF, preserved forest) tended to present higher scores The differences in the centroids among habitats were statistically significant (r2 = 0.581, P = 0.004).
Phylogenetic composition
We also considered three axes for the ordination of plant assemblages according to their phylogenetic composition (Fig. 4). Additional dimensions did not significantly reduce the stress value (5.261). This stress value, according to Kruskal and Clarke's rules of thumb, supports an ordination satisfactorily representing the degree of relatedness between assemblages in terms of their phylogenetic composition (McCune and Grace, 2002).
Ordination of tree species assemblages along edges abutting different vegetation matrices compared to forest interiors in terms of their phylogenetic composition. NMDS1–NMDS3 are axes resulting from the nonmetric multidimensional ordination. Abbreviations: PC, the centroid in the ordination of forest edges abutting pastures with cattle (n = 3); PA, the centroid in the ordination of forest edges abutting pastures without cattle (n = 3); DV, the centroid in the ordination of forest edges abutting disturbed dry forest vegetation with cattle (n = 3); PF, the centroid in the ordination of old-growth forest interiors (n = 3).
Similarly to what we observed when analyzing species composition, along axis 1 of the ordination (Fig. 4) we found a gradient of cattle influence on phylogenetic composition. The assemblages with higher influence of cattle (PC, pasture with cattle) presented the lowest scores, whereas the assemblages with no influence of cattle (PF, preserved forest) presented higher scores. Differences in the centroids corresponding to each habitat type were also statistically significant (r2 = 0.520, P = 0.009).
On the other hand, we did not find significant differences in the phylogenetic structure among the plant assemblages associated with different types of edges. The PSV and PSE values were statistically indistinguishable among the four habitats (Table 2). In addition, the results of null model tests showed a random phylogenetic structure of the tree assemblages. The overall mean values for the metrics PSV (0.794) and PSE (0.759) did not differ significantly from the values expected by chance. These results were obtained under the two null models used.
| Phylogenetic Structure Metrics | ||
|---|---|---|
| Habitat types | PSV | PSE |
| Pasture with cattle | ||
| PC1 | 0.740 | 0.722 |
| PC2 | 0.810 | 0.798 |
| PC3 | 0.737 | 0.563 |
| Mean ± SE | 0.762 (± 0.024) | 0.694 (± 0.069) |
| Pasture without cattle | ||
| PA1 | 0.829 | 0.786 |
| PA2 | 0.796 | 0.753 |
| PA3 | 0.789 | 0.778 |
| Mean ± SE | 0.805 (± 0.012) | 0.772 (± 0.010) |
| Disturbed vegetation with cattle | ||
| DV1 | 0.809 | 0.783 |
| DV2 | 0.802 | 0.810 |
| DV3 | 0.797 | 0.728 |
| Mean ± SE | 0.803 (± 0.003) | 0.774 (± 0.024) |
| Preserved forest | ||
| PF1 | 0.797 | 0.779 |
| PF2 | 0.822 | 0.806 |
| PF3 | 0.799 | 0.796 |
| Mean ± SE | 0.806 (± 0.008) | 0.794 (± 0.008) |
| F; P-value | 1.729; 0.238 | 1.179; 0.377 |
- Note: The phylogenetic species variability metric (PSV) and the phylogenetic species evenness metric (PSE) values did not differ significantly among habitat types. The null model tests showed a random phylogenetic structure of the plant assemblages. Forest edges abutting different matrices are indicated as follows: PC = pasture and cattle (n = 3); PA = pasture (n = 3); DV = disturbed vegetation with cattle (n = 3); PF = preserved or interior forest (n = 3) used as experimental control.
Edge type and tree community associations
In edge blocks, the dominant species were generally early successional trees whereas the dominant species in forest interior blocks were late-successional species (i.e., old-growth forest species). The proportion of early successional species was significantly greater (χ2 = 53.3, P < 0.001) in pastures and disturbed vegetation with cattle (ca. 30%) than in pastures without cattle and in interior forests (ca. 15%). The most common species in each habitat type were the following: Caesalpinia platyloba S. Watson (Fabaceae), exclusively found in interior forest; Cordia alliodora (Ruiz & Pav.) Oken an early succesional species, present in all habitats but among the most abundant species in all edge types. Lonchocarpus constrictus Pittier (Fabaceae), an early successional species and C. alliodora were the most abundant tree species in edges abutting pastures with cattle (see Álvarez-Añorve et al., 2012).
We found that tree species phylogenetic composition depended on edge type and therefore on the presence of cattle (Fig. 5). The phylogenetic composition of the most abundant plant families and associated genera largely associated were the following: (1) in edges abutting pastures and disturbed vegetation with cattle the families Anacardiaceae, Malvaceae and Moraceae, and the genera Bursera, Ceiba and Jatropha, (2) in edges abutting disturbed vegetation with cattle, and pastures without cattle the families Apocynaceae, Euphorbiaceae, Fabaceae, Polygonaceae, Salicaceae, and the genera Caesalpinia, Lonchocarpus and Randia, and (3) in forest interiors and pasture edges without cattle the families Capparaceae and Cinchonoideae, and the genera Capparis, Cordia, Jatropha, Tabebuia and Trichilia. These associations do not mean that plant families and genera were exclusively found in a particular habitat but that they were more abundant therein.
Most specious plant families and genera associated with different habitat types. The centroids for each edge type and preserved forest (in bold), resulting from the ordination of plant assemblages according to their phylogenetic composition. The plant family and genus scores were calculated by computing the weighted average scores with the help of the function “wascores” from the R package “vegan” (Oksanen 2011, Oksanen et al., 2012). Plant families represented by at least two species are the following: Anacardiaceae (1), Apocynaceae (2), Capparaceae (3), Cinchonoideae (4), Euphorbiaceae (5), Fabaceae (6), Malvaceae (7), Moraceae (8), Polygonaceae (9), Rubiaceae (10), and Salicaceae (11). Plant genera represented by at least two species include the following: Bursera (12), Caesalpinia (13), Capparis (14), Ceiba (15), Cordia (16), Croton (17), Jatropha (18), Lonchocarpus (19), Randia (20), Tabebuia (21) and Trichilia (22). Abbreviations: PC, forest edges abutting pastures with cattle (n = 3); PA, forest edges abutting pastures without cattle (n = 3); DV, forest edges abutting disturbed dry forest vegetation with cattle (n = 3); PF, old-growth forest interiors and/or preserved forest (n = 3).
DISCUSSION
As expected, the structure and composition of the vegetation matrix, but also management practices affect the dynamics of plant colonization and extinction along edges. In tropical wet forests, fragments that have experienced a similar management of the matrix tend to converge in tree species composition (Laurance et al., 2007). In contrast, our data showed that the presence of cattle apparently exerts a stronger effect on tree species composition assemblages than does vegetation structure of the matrix. The NMDS ordination clearly segregated floras at edges abutting a matrix with cattle from interior forest, and from the pasture matrix where cattle were absent. Disturbed dry forest vegetation provides shade and shelter and likely more food resources to cattle than do open pastures, and therefore might be used more intensively.
As in TDF, in several vegetation types of the dry tropics (e.g., Caatinga, Cerrado and Cerradão in Brazil, and Chacośs semiarid forest in Argentina) no edge effects or slight edge-related changes have been detected in microclimate and/or in the structure, diversity and composition of the vegetation (López de Casenave et al., 1995; Lima-Ribeiro, 2008; Santos and Santos, 2008; Sampaio and Scariot, 2011). The observed changes have been attributed to management of the matrix and forest fragments (e.g., cattle grazing, fires, selective logging and wood collection, pesticides, etc.), edge age, and changes in seed dispersal by vertebrates. In this study, generally, tree species composition but not diversity was impacted by edge effects as has been observed in a tropical dry forest of central Brazil (Sampaio and Scariot, 2011).
Phylogenetic composition along tropical dry forest edges
As compared to old-growth forest interiors, forest edges have different trajectories of change not only in species composition but also in phylogenetic species relatedness (phylogenetic composition). Edges abutting different matrix types were dominated by early successional tree species (e.g., C. alliodora and L. constrictus). These results are in accord with previous studies in TDF at different successional stages (in Mexico, and in dry forests of Brazil and Costa Rica) where species and phylogenetic composition differed between early and intermediate-successional stages and preserved old-growth forest sites (Álvarez-Añorve, 2012). In our study, these changes were magnified by the presence of cattle. Although cattle have been considered as an effective seed dispersal vector for many plant species in defaunated TDF (Janzen and Martin, 1982; Guimarães et al., 2008), our results suggest that the presence of cattle intensifies the negative effects of edge creation on tree communities.
Several tree species of TDF produce anemochorous seeds that can travel large distances, becoming the initial colonizers in disturbed areas such as forest edges (Janzen, 1988). Wind-dispersed species such as C. alliodora and L. constrictus dominated the tree community at forest edges. Forests dominated by wind-dispersed trees are relatively hostile to animals and are likely to persist for years (Janzen, 1988). Richness and abundance of large-seeded zoochorous tree species may be lower along forest edges (Melo et al., 2006). In fact, Spondias purpurea L. (Anacardiaceae) a large-seeded animal dispersed species and key source of food and water for a myriad of animals in the study region, was not recorded in any of the forest edges. Change in the composition of functional guilds together with species invasions (e.g., M. arenosa) could accumulate gradually, leading to persistent divergences in tree community composition and functioning.
Phylogenetic structure along tropical dry forest edges
Overall, we found that plant assemblages at Chamela have a random phylogenetic structure in old-growth forest and forest edges, regardless of matrix type. This suggests that tree communities are mainly assembled by stochastic processes, as assumed by the neutral theory (Hubbell, 2001). Phylogenetic randomness in the old-growth forest is expected because early successional species belonging to a few families are uncommon, while late-successional species from many distant families dominate the tree community (Álvarez-Añorve, 2012). In fact, in habitats showing strong physical differences between their canopies and open areas, facilitation mechanisms can increase the phylogenetic diversity of plant communities (Valiente-Banuet and Verdú, 2007). Long-lived late-successional species persisting after edge creation could have had important implications in the phylogenetic structure of the tree community assemblage in our study. In other words, although edges bordering different matrices may diverge in species composition over time, they held random subsets of clades characteristic of old-growth preserved forest.
The adult tree assemblage is less dynamic in ecological and phylogenetic terms than the advanced regeneration assemblages (i.e., seedlings and saplings in the understory) (Letcher, 2010). Because of recruitment limitations along forest edges, it is likely that the advanced regeneration community changes according to the type of abutting matrix. Further studies are needed to detect edge effects on phylogenetic assemblages at early life stages (Letcher, 2010; Arroyo-Rodríguez et al., 2012). For instance, diversity of seedling species decreased toward forest edge in a TDF in central Brazil (Sampaio and Scariot, 2011), whereas in the current study we recorded an invasive exotic (M. arenosa) along edges bordering pastures with cattle, though at smaller diameter classes (<5 cm dbh). The maintenance of phylogenetic diversity and structure of tree assemblages at different life stages along dry forests edges is critical for ecosystem functioning because it ensures the representation of sufficient ecological strategies in an ecosystem under constant external pressures (Cavender-Bares et al., 2009).
Implications for conservation
Tropical dry forest is one of the most threatened terrestrial ecosystems in the Neotropics (Miles et al., 2006). In Mexico, TDF is the most abundant tropical forest vegetation type, but it is undergoing rapid rates of deforestation and/or land cover change. Annually, 1.4–1.9% of all TDF is converted to cattle pastures and other land uses (Trejo and Dirzo, 2000). As TDF is increasingly deforested and fragmented, dry forest landscapes are dominated by edge-affected habitats. The remaining TDF tracts in Chamela are mainly surrounded by open pastures and old fields. Once planted, pastures are never fertilized, but are maintained by burning cycles carried out every 2 or 3 years. Although fires have been shown to favor germination in some TDF species (Otterstrom et al., 2006), their impact on the vegetation is largely negative because fires destroy secondary regrowth, and the seed and seedling banks along edges, and increase the intensity of edge-related microclimatic stress and forest degradation.
It is important to emphasize that we evaluated edge effects caused by relatively small perforations within the natural old-growth forest vegetation (20–100 ha forest clearings; Fig. 1). Edge effects on the tree community and on other components of TDF biota will certainly be stronger in larger forest clearings such as those caused by intensified agriculture, power lines and wide paved or unpaved roads. Furthermore, edge effects are expected to increase as time elapses and habitat fragmentation proceeds in the study region (Goosem, 2000; Fletcher, 2005). The preservation of the native tree cover on the sides of the roads and along the edges of other kinds of forest clearings will help to diminish edge effects by facilitating the maintenance of canopy cover, and by shortening the distance seed dispersers and pollinator vectors have to move when crossing forest openings (Forman and Alexander, 1998). Continuous monitoring of edge vegetation along the Chamela-Cuixmala Biosphere Reserve should be implemented to detect other edge-related changes such as species invasions and the introduction of exotic pests and diseases, or the synergisms between edge-creation and other types of disturbances such as hurricanes.
One of the major conclusions of this study is that the phylogenetic approach contrasts with the vegetation structural assessment of TDF edge effects (i.e., basal area and species richness), indicating that edge effects in the study area are more profound than previously documented. Compared to previous studies on phylogenetic tree assemblages in fragmented landscapes, our study has the strength of using large tracts of preserved forest with intact stand structure and faunal assemblages as experimental controls (see Santos et al., 2010; Arroyo-Rodríguez et al., 2012). Furthermore, all edge types in the study have been in a stable landscape for 30 years, which reinforces the certainty of our findings on tree community pathways. Due to TDF's less complex structure and height, edge effects in TDF are not as conspicuous at a glance as they are in the rain forests. Phylogenetic analysis appears to be an important tool to determine the magnitude and implications of edge creation in this endangered tropical ecosystem.
