Metacommunity structure, additive partitioning and environmental drivers of woody plants diversity in the Brazilian Atlantic Forest

1 INTRODUCTION

For over a century, ecologists have sought to answer questions regarding the processes that sort species to local assemblages and how they vary across space (Cavender-Bares, Kozak, Fine, & Kembel, 2009; Clements, 1916; Diamond, 1975; Gaston, 2000; Gleason, 1926; Holyoak, Leibold, & Holt, 2005; McGill, 2014; Ricklefs, 1987; Tilman, 1982). Studies have shown that processes that promote diversity may change across spatial scales (e.g., Cavender-Bares et al., 2009; Davies et al., 2005; Diez, Sullivan, Hulme, Edwards, & Duncan, 2008; McGill, 2014; Silvertown, Dodd, Gowing, Lawson, & McConway, 2006). Thus, to advance ecology, rather than ask which drivers are most important for biodiversity distribution, we need to understand which drivers are most important at a given spatial scale (Gotelli, Dorazio, Ellison, & Grossman, 2010; McGill, 2014).

The metacommunity concept is an important framework to understand community assembly because it explicitly recognizes that patterns of species distribution and diversity among communities are influenced by both environmental and dispersal processes (e.g., Fernandes, Henriques-Silva, Penha, Zuanon, & Peres-Neto, 2014), and can be applied across different spatial scales (Leibold et al., 2004; Meynard et al., 2013; Presley & Willig, 2010). Assessing community structure and species distribution patterns using the metacommunity framework (Leibold & Mikkelson, 2002; Presley, Higgins, & Willig, 2010) has recently helped ecologists to understand patterns and processes structuring diversity (Heino et al., 2015; Henriques-Silva, Lindo, & Peres-Neto, 2013; Keith, Newton, Morecroft, Golicher, & Bullock, 2011; Meynard et al., 2013; de la Sancha, Higgins, Presley, & Strauss, 2014). Recent examples include assessments of the relative importance of environmental and spatial properties in species turnover (Henriques-Silva et al., 2013; Meynard et al., 2013), differences of community structure among clades (Heino et al., 2015; de la Sancha et al., 2014) and along ecological succession (Fernandes et al., 2014; Keith et al., 2011). In all those studies, the key to understand processes that drive metacommunities is their emergent spatial structure (Presley et al., 2010).

As proposed by Leibold & Mikkelson, 2002 and later expanded by Presley et al. (2010), the elements of metacommunity structure (EMS) is a useful framework to assess the underlying processes of metacommunity assembly at different spatial scales (Henriques-Silva et al., 2013; de la Sancha et al., 2014). The EMS allows identifying which idealized structure (i.e., checkerboard, nestedness, Clementsian, Gleasonian, evenly spaced or random distributions) best characterizes an empirical metacommunity based in three elements: coherence, turnover and boundary clumping (Leibold & Mikkelson, 2002; Figure 1). The coherence expresses the level in which species are affected by the same environmental gradient; turnover represents the way species composition changes along the gradient; and boundary clumping measures the degree of concordance among the boundaries of the species range along the gradient.

Idealized metacommunities emerging from the combination of these elements present a unique theoretical underpinning (Clements, 1916; Dallas, 2014; Diamond, 1975; Gleason, 1926; Patterson & Atmar, 1986; Presley et al., 2010; Tilman, 1982) that facilitates the interpretation of processes regulating the metacommunity and affecting its collective properties (e.g., species richness) (Figure 1). A checkerboard pattern indicates that multiple pairs of species co-occur less than expected by chance (Diamond, 1975). This pattern can arise from various mechanisms that often act in concert, such as interspecific competition, different habitat preferences and dispersal limitation associated to ecological and historical factors (Gotelli & McCabe, 2002; Heino et al., 2015). Nested metacommunities occur when species in the poorest sites are comprised of subsets of the species found in richer sites and when widely distributed species present ranges encompassing those of more narrowly distributed species (Patterson & Atmar, 1986). In general, habitat area, isolation and environmental suitability are typical correlates with nestedness (Wright, Patterson, Mikkelson, Cutler, & Atmar, 1998). Clementsian gradients consist of communities with tight association among species, behaving like “super organisms” and presenting discrete community boundaries (Clements, 1916). In contrast, Gleasonian gradients describe communities as mere collections of species, with individualistic responses to underlying environmental gradients, and whose ranges happen to overlap (Gleason, 1926). In evenly spaced gradients, the idealized model of the species range boundaries is hyperdispersed (i.e., the distribution of species ranges is more dispersed than expected by chance) along the underlying environmental gradient (Tilman, 1982), indicating maximal differences among species in environmental tolerances (Presley & Willig, 2010). A random structure characterizes metacommunities in which species respond to different environmental gradients (Presley et al., 2010), revealing no coherence among species distributions within the metacommunity. Contrasting these different idealized models simultaneously may reveal more about potential mechanisms that structure metacommunities than studying each model in isolation (Dallas & Presley, 2014; Heino et al., 2015; Leibold & Mikkelson, 2002; Meynard et al., 2013;Presley et al., 2011). The metacommunity emergent structure can improve the efficiency of conservation efforts by helping select target areas for biodiversity conservation. For example, maximizing the conservation in a metacommunity with the Clementsian structure (Clements, 1916) requires the preservation of communities that represent all the “super organisms” (i.e., all the groups of species with discrete distribution boundaries) compounding this metacommunity. On the other hand, in a metacommunity with a nested structure, conservation efforts may focus on richer sites as they tend to contain most species of the metacommunity (Patterson & Atmar, 1986).

Another ecological approach that can improve conservation efforts in a multiscale scenario is diversity partitioning. Biodiversity defined in ecological terms refers to the variety and distribution of species or vegetation types (Noss, 1990) and can be decomposed in three different components: diversity representing the total species pool in a region (γ), within-communities (α) and between-communities (β). This diversity partitioning was originally coined by Whittaker (1960, 1972) and has since been widely used in applied and theoretical ecology, in which several metrics have been developed based on this hierarchical concept (Anderson et al., 2011; Tuomisto, 2010a,b). The hierarchical partitioning of diversity is a promising approach to understand patterns of species distribution in compartmentalized systems across spatial scales (Gering, Crist, & Veech, 2003). It allows analysing the contribution of each diversity component (α, β) for the total diversity in hierarchical studies with multiscale nested sampling (Lande, 1996) because it identifies the scales in which the diversity is higher or lesser than expected by chance. The hierarchical partitioning of diversity enables evidencing sources of heterogeneity within a megadiverse metacommunity and can provide insights into how conservation efforts could be most cost effective (Gering et al., 2003; Ribeiro, Prado, Brown, & Freitas, 2008; Summerville, Boulware, Veech, & Crist, 2003). Therefore, it is possible to reveal the most important scales contributing for the total diversity in a given metacommunity. Thus, additive partitioning of diversity coupled with description of EMS has the potential to improve our understanding of mechanisms that sort species from the regional pool to local assemblages (de la Sancha et al., 2014). In short, EMS allows the characterization of how communities are structured and the diversity partitioning reveals the scales promoting the biodiversity.

The environmental heterogeneity within a metacommunity directly affects its total biodiversity. As the community composition is driven by, among others, the environmental tolerance of each organism (e.g., Weiher & Keddy, 1995), species turnover (β-diversity) is expected to be higher between areas with different environmental conditions than between areas that have similar environments (Tuomisto, Ruokolainen, & Yli-Halla, 2003). Thus, a metacommunity encompassing high environmental heterogeneity also tends to have a higher diversity. Despite this known relationship between the environment and β-diversity, the way in which species composition changes across communities is not necessarily predicted by their differences in environmental conditions. In other words, geographically close metacommunities under different environmental conditions can exhibit either the same or different pattern of species turnover through its main gradient (Heino et al., 2015; Henriques-Silva et al., 2013). In this way, the knowledge of the metacommunity structure patterns of a heterogeneous and biodiverse system can shed light on processes underlying the relationship between species richness and environmental heterogeneity. Furthermore, it can help conservation efforts as it characterizes how the diversity is distributed and how the environment influences it. This can provide basis for, as example, the choice on how conservation areas should be distributed to maximize diversity.

Given the levels of biodiversity and environmental heterogeneity, the Atlantic Forest (hereafter AF) biome can be a good case study for characterizing biodiversity using the metacommunity approach. The AF is a tropical forest marked by high endemism and intense anthropogenic disturbance, which distinguishes it as a world biodiversity hotspot (Mittermeier et al., 2004; Myers, Mittermeier, Mittermeier, da Fonseca, & Kent, 2000). It has been defined as a complex biome mainly comprised of three forest types (Brazilian Federal Law 11.428/2006; Oliveira-Filho & Fontes, 2000): the Dense Rain forest, the Mixed Rain forest and the Seasonal forest. It has been suggested that historical processes related to dispersal limitation and speciation possibly resulted in different phylogenetic structure of communities in each forest type (Duarte, Bergamin, Marcilio-Silva, Seger, & Marques, 2014). Also, previous studies have evidenced rainfall gradients as the major drivers of plant species turnover in the AF (Kamino, Oliveira-filho, & Stehmann, 2008; Marques, Swaine, & Liebsch, 2011; Oliveira-Filho & Fontes, 2000; Oliveira-Filho et al., 2005; Santos, Serafim, & Sano, 2011). However, it has been shown that different environmental factors drive the communities’ structures across different AF forest types (Oliveira-Filho, Budke, Jarenkow, Eisenlohr, & Neves, 2013), such as annual temperature (in south-eastern Dense forest; Oliveira-Filho & Fontes, 2000), duration of the dry season (in south-eastern Seasonal forest; Oliveira-Filho & Fontes, 2000), rainfall seasonality (in south-eastern Seasonal forest; Kamino et al., 2008; Santos et al., 2011) and soil sandiness (lowland forests within the south-eastern and south Dense Rain forest; Marques et al., 2011). However, a link between environmental factors, their imprint on plant metacommunity patterns and relative importance in different AF types is still needed to better understand the relationship between environmental heterogeneity and the high diversity found in this biome.

To this aim, we used a comprehensive database of tropical trees and shrubs, including the presence/absence of 2,071 species across 177 sites from the Brazilian Atlantic Forest. We applied the EMS framework and the diversity partition approach to determine the community structure and the contribution of each scale (sites, among sites and among forest types) for the high levels of diversity found in the AF. We also assessed the role of climatic variables as drivers of community structure across the entire AF biome and in each of its forest types.

2 METHODS

2.1 The Atlantic Forest biome

The AF extends along the Brazilian coast and a few inland portions of Paraguay and Argentina, originally covering more than 1 million km². Latitudes range from 5°N to 33°S, longitudes from 35°W to 52°W and altitudes from 0 to 2,200 m (IBGE, 1992). This biome encompasses three different main forest types (Figure 2). The Dense Rain forests (hereafter Dense Forests) are associated with the Atlantic Ocean and are distributed in lowlands (until 50 m a.s.l.) and slopes (50–2,200 m a.s.l.) from north-eastern to southern Brazil. Average temperature ranges from 22 to 25°C, and the climate is generally hot and wet in lowlands and colder and wetter in the slopes (IBGE, 1992; Oliveira-Filho & Fontes, 2000). Mixed Rain forests (hereafter Mixed forests), also known as Araucaria forests because of the remarkable presence of the conifer Araucaria angustifolia, constitute the main forest type on the highland plateau in southern Brazil at elevations above 500 m a.s.l. (Hueck, 1972). These forests are in regions of tropical and subtropical humid climates without pronounced dry periods, where the annual mean temperature ranges from 12 to 18°C (Behling, 2002). Finally, Seasonal forests are related to the hinterland of the Parana River basin across south and south-east Brazil. These forests are characterized by two distinct seasons with marked alternation from tropical climate with intense rainfalls and temperatures around 22°C in the summer to subtropical climate with lower temperatures around 15°C and scarce precipitation in the winter. Many of its canopy trees are deciduous (20%–50%), with leaves falling during the colder and drier period (IBGE, 1992), thus characterizing this forest type as semi-deciduous.

2.2 Woody species occurrence in the Atlantic Forest

To compile the checklist of the AF species, we used a comprehensive dataset provided by Bergamin et al. (2015) of all published forest surveys from 1994 to 2014 led in the south and south-east of Brazil. We complemented this dataset searching for studies published until March 2015 in all regions encompassing the AF. We considered only studies that used the cut-off diameter at the breast height (DBH) ≥4.8 cm, to standardize the sample criteria. Our dataset includes a list of 2,071 tree and shrub species from 177 localities (Figure 2) organized in the presence/absence matrix. The species list was checked for recent synonyms in the Missouri Botanical Garden (http://www.tropicos.org), The Plant List (http://www.theplantlist.org/), and Flora do Brasil databases (http://floradobrasil.jbrj.gov.br), all accessed in February 2016 (Details about the dataset are provided in Appendix S1).

In the compiled dataset, the number of sites and the sampled area in each forest type is proportional to its original cover (Appendix S1). The majority of the AF sites have between 30 to 89 woody species. The total number of species is higher in the Dense forest (S = 1,591) compared to the Seasonal (S = 1,174) and Mixed (S = 683) forests; however, differences among forest types are not statistically significant correcting for sampled area (ANOVA, p = .068; Figure 3a,b).

During data compilation we considered only peer-reviewed studies and dissertations that used standardized methods. Although we recognize that the dataset may have limitations due to the total number of sites in comparison to the large geographical extent of the AF, the dataset is relatively well-dispersed geographically and proportional to the original cover of forest types. To ensure that sample size was not an issue, we repetitively subsampled the data and compared results to the full dataset (see below). Therefore, we are confident that it constitutes a realistic sample of this highly fragmented tropical forest.

2.4 Additive partitioning of diversity

We tested the uniformity of species diversity (S) across spatial scales using the additive partitioning of total diversity (Lande, 1996), which considers that mean alpha (α) and beta (β) diversities add up to gamma (γ) diversity. Thus, β-diversity is measured in the same biodiversity dimension as α and γ, defined here as species richness (S) (Lande, 1996). This additive procedure is then extended across multiple scales in a hierarchical sampling design with i = 1, 2, 3, …, m levels of sampling (Crist et al., 2003), which in our case was: local (within sites, α-diversity), regional (among sites in each forest subtype, β1) and geographical (between subtypes, β2). Samples in lower hierarchical levels are nested within higher level units (e.g., sites are nested within forest types and forest types are nested within the AF). At each level i, αi denotes the average diversity found within samples. At the highest sampling level, the diversity components are calculated as:

For each lower sampling level as:

Then, the additive partition of diversity is:

Average alpha components (αm) are uniformly weighted to calculate it as simple average. The expected diversity components are calculated 999 times by individual based randomization of the community data matrix.

We computed the additive diversity partitioning in the R environment (R Core Team, 2014) using the function “adipart” (follows Crist et al., 2003) from the “vegan” (version 2.0-8) package.

3 RESULTS

3.2 Additive partitioning of diversity

For the total woody species’ diversity (γ = 2,071 species), the hierarchical partitioning showed that each site (α), on average, retains more diversity than expected by the null hypothesis—that assumes equity among scales (p < .001, Figure 4). The β-diversities among sites within forest types (β1) are lower than expected by chance (p < .001). On the other hand, the observed among-forest type β-diversities (β2) were higher than expected by the null hypothesis (p < .001).

4 DISCUSSION

The AF is a continental biome where tree species’ distributions and processes of community assembly are affected by different factors and mechanisms acting at different scales. At the biome scale, the emergent metacommunity structure comprised of nested subsets of species correlated mostly with temperature. We found this structure pervasive across forest types, but while species composition in Dense and Seasonal forests was determined mainly by temperature, precipitation variables were more important in the Mixed forest. This difference leads to high species diversity on a geographical scale (among forest types).

The AF presented the same idealized metacommunity structure at the biome and geographical (forest types) scales. This indicates that the species composition changes in a similar way along the main gradient at different spatial scales (Meynard et al., 2013). The nested structure presented by the AF and all forest types is a structure in which the species of species-poor sites represent subsets of species found in species-rich sites (Leibold & Mikkelson, 2002). In other words, this structure evidences that the ranges of species with restricted distributions along the environmental gradient are contained within the ranges of widely distributed species (Presley et al., 2010). The nested structure also indicates that the communities have some degree of similarity in species composition, which is corroborated by the lower than expected beta diversity within forest types (β1). Indeed, Eisenlohr and Oliveira-Filho (2015) showed that around 30% of the species occurring in a given AF forest type are common to the entire AF, and approximately 40%–55% of the species of a given AF forest type are not restricted to that type.

A nested subset structure may arise due to different factors operating at local and regional scales such as metacommunity biogeographical factors (i.e., species probability of extinction, colonization and speciation) as well as some filters that species need to pass in order to establish in a site (i.e., sampling filter, distance filter, habitat filter and area filter; see Wright et al., 1998 for details). Given that we evaluated the AF metacommunity from the species range perspective (see Presley et al., 2010 for a discussion on this matter), we believe that the nested structure detected here may be due to the nested habitat hypothesis (Hylander, Nilsson, Gunnar Jonsson, & Göthner, 2005). In short, this hypothesis predicts that all species within a metacommunity have the optima of their realized niche on the same end of the main gradient. The nested structure arises due to differences in each species capability of establishing along non-optimal parts of this gradient (Hylander et al., 2005). Therefore, the nested structure of the AF's communities suggests that they are composed of species with different tolerances, from environmental generalist with wide distributions, to specialist with narrow environmental and geographical ranges.

It is worth noting that the AF and its forest types do not constitute perfect nested systems as some species are restricted to specific sites and the metacommunity structure presented a pattern of clumped species loss. These patterns are probably due to similarities of species’ environmental requirements and dispersal, which reflected in association among species’ geographical distributions (Presley et al., 2010). This can shede light as to why, even with lots of species with restricted distribution (~22% of each forest type composition), the best predicted metacommunity structure that fits the AF and its forest types is the nested subset structure.

Temperature-related factors were found to be the main drivers of the entire AF woody plant metacommunity. De la Sancha et al. (2014) also found temperature as the main driver of AF small mammals’ metacommunity structure, emphasizing the importance of temperature gradients as a driver of species composition in the AF. This contradicts, in part, results from previous studies (Oliveira-Filho & Fontes, 2000) that found precipitation as the major driver of south-eastern AF structure at the species level whereas temperature was important at the genus and family levels. However, disentangling this forest complex and analysing each forest type separately revealed that the Mixed forest distinguishes from the others by presenting an opposite relationship with temperature seasonality and being mainly affected by precipitation factors. Species composition of the Mixed forest is remarkably influenced by the presence of basal clades of woody plants (Duarte et al., 2014) that are characterized by limited water efficiency (Boyce et al., 2009; Brodribb & Field, 2010) and thus constrained by water availability (Duarte, 2011). Taken together, these facts may explain why precipitation was found to be a major driver of the Mixed forest metacommunity structure and stress the singularities of each forest type comprising the AF biome.

The relationship between the main structure of the AF metacommunity and its different forest types with geographical location (latitude and longitude) reveals the role of spatially structured factors on the species composition. Previous researches on the south-eastern AF (Eisenlohr & Oliveira-Filho, 2015; Oliveira-Filho & Fontes, 2000) have shown that the distance from the sea influences floristic composition. The Brazilian coastal mountain range Serra do Mar and the inland plateau create a relatively abrupt relief transition from coastland to hinterland. This altitudinal transition is related to changes in climatic factors, including temperature, frost events, atmospheric pressure, solar radiation, air masses speed and rainfall patterns (Jones, 1992). The moist air masses from the Atlantic Ocean face a barrier on the coastal mountain range and sudden changes in atmospheric pressure and air speed leads the greater part of these air masses to precipitate over the Serra do Mar seaward side. This coastland region has the highest mean annual rainfall of the entire AF range (up to 3,600 mm), while in the inland plateau, with typical seasonal climates, mean annual rainfall ranges from 1,300 to 1,600 mm (Oliveira-Filho & Fontes, 2000).

Our findings also indicate that the main processes promoting biodiversity act at local (site) and geographical (among forest types) scales. The number of species within sites (α) and the turnover of species between forest types (β2) maintain the high diversity in the AF. The high β2-diversity can be related to the environmental heterogeneity, as suggested by the variation found in the environmental drivers among forest types. Also, the high species turnover between the AF forest types (β2) can be due to their distinct biogeographical history as suggested by their contrasting phylogenetic patterns (Duarte et al., 2014). The high α-diversity indicates that the communities in the AF, in general, are composed of more species than expected by chance. At this small scale, microclimate variation and species interactions are hypothesized to play major roles in the process of community assembly (Cavender-Bares et al., 2009; McGill, 2014).

BIOSKETCHES

Vinicius Marcilio-Silva is a biologist with main interest in plant ecology and evolution. His work focuses on processes underlying the community assembly in tropical forests with implications for the theoretical ecology and conservation.

Márcia Cristina Mendes Marques and Victor P. Zwiener are biologists with focus in plant ecology. Their research interest covers various aspects of ecology ranging from how different processes affect tropical plant diversity across multiple scales to applied ecology, providing ecological grounds to management actions.

Author contributions: All authors collected the data, conceived the ideas and wrote the manuscript; VMS performed statistical analysis.