Correlates of plant β-diversity in Atlantic Forest patches in the Pernambuco Endemism Centre, Northeastern Brazil

Abstract

Understanding how vegetation structure and floristic composition vary across landscapes is fundamental to understand ecological patterns and for designing conservation actions. In a patch-landscape approach, we assessed the β-diversity (q0 order – rare species, q1 order – common species, and q2 order – dominant species) of plants between forest patches and surveyed plots in Atlantic Forest patches located in the Pernambuco Endemism Centre, northeastern Brazil. Furthermore, we tested the influence of predictor variables linked to landscape (forest cover and edge density) and habitat (basal area), as well as the geographical distance between forest patches and plots on the β-diversity in each forest patch and plot. We measured and identified a total of 1,682 individuals (trees and lianas), corresponding to 248 species, 116 genera, and 56 families in 10 plots (50 × 2 m) from each forest patch. The β-diversity presented lower values for the Mata de Água Azul patch at a landscape scale (i.e., between forest patches) and Mata dos Macacos patch at a site scale (i.e., between plots) for all orders. Geographical distance positively influenced the β-diversity at the landscape scale, and higher turnover between plots (e.g., within forest patches) was positively associated with differences in geographical distance, edge density, forest cover, and basal area. Our results indicate the need to conserve forest patches distributed across a wide area (distant sites) that encompass different landscape contexts with different vegetation structures, in order to conserve greater floristic diversity.

Introduction

The world is experiencing increasing habitat loss and fragmentation (Segan et al. 2016), which are mainly caused by anthropogenic actions (Haddad et al. 2015, Kleinschroth & Healey 2017). These actions have contributed to emptying, simplifying, and reducing the area of the tropical forests (Edwards et al. 2019), as well as modifying the composition and configuration of landscapes (Theobald et al. 2020). This widespread habitat-altering process can be observed throughout the entire Brazilian Atlantic Forest (Lira et al. 2021), resulting in a large number of small and isolated forest patches, varying in shape, inserted into different types of matrices (Melo et al. 2013). The effects of these processes in the forest patches found in these landscapes – mainly a reduction of overall forest cover and high levels of fragmentation – have increased the similarity of vegetation structure and floristic composition within and between forest patches (Thier & Wesenberg 2016, Salomão et al. 2019).

Pronounced edge effects, high levels of fragmentation, and human-modified matrices are highly related to the extirpation of plants from a local viewpoint, limiting essential ecological processes such as seed dispersal, seedling establishment, and plant survival in small forest patches (Santos et al. 2008). Additionally, in small and isolated forest patches, plant species with particular traits, such as large seeds, shade tolerance, and emergent trees, have suffered from the deleterious effects of habitat loss and fragmentation (Tabarelli et al. 2012). Moreover, anthropogenic disturbances directly associated with plant communities (e.g., selective and indiscriminate logging, fire) potentiate the extirpation of some species. These impacts can lead to the homogenization of diverse elements of the community (Lôbo et al. 2011), such as the floristic composition and forest structure within and among forest patches at different scales (Arroyo-Rodríguez et al. 2013). Nevertheless, some species are adapted to survive in small forest patches in fragmented landscapes, despite a pronounced edge effect (Santos et al. 2012).

Among the parameters of a determined community, β-diversity is primordial for understanding the organization of species (Flohre et al. 2011). β-diversity patterns are intrinsically associated with the processes managed at local (niche structure, biological interactions) and regional (dispersal, extinction, colonization) scales (Lawton 1999). In this scenario, β-diversity can increase (floristic differentiation) or decrease (floristic homogenization), and a multiscale approach is, perhaps, more suitable for visualizing such shifts (Harrison & Cornell 2008, Karp et al. 2012). Besides possessing a high species richness at a local scale, tropical forests also have high rates of species turnover (Macía et al. 2007). This high turnover is linked to environmental variability within and between sites, as well as geographical distance, which tends to limit dispersal (Myers et al. 2013). Thus, the increase of species turnover between forest patches (i.e., floristic differentiation) is often positively associated with the geographic distance between forest patches (Arroyo-Rodríguez et al. 2013). Similarly, landscapes with lower forest cover are associated, in some contexts, with more isolated forest patches (Lira et al. 2012), which affects species occurrence (Boscolo & Metzger 2011; Damschen et al. 2006), species turnover (Morante-Filho et al. 2016), and overall species diversity in the landscape (Martensen et al. 2008). Additionally, reduced forest cover results in increased forest edges that demonstrate different abiotic characteristics, resulting in changes in floristic composition compared to the interior of forest patches (Oliveira et al. 2004). Thus, edge density should influence the turnover between landscapes. Finally, changes in local conditions, especially regarding anthropogenic impacts, should affect both forest structure, such as plant basal area, and the composition of species which thrive in such environmental conditions, thus affecting the β-diversity between forest patches that experience different environmental conditions and anthropogenic pressure (Burrascano et al. 2008).

There are still knowledge shortfalls regarding the quality of the forest patches in the Pernambuco Endemism Centre, located in the Atlantic Forest north of the São Francisco River, with the occurrence of an Endangered primate, the blonde capuchin monkey (Sapajus flavius) (Souza-Alves et al. 2018). The natural habitat in this region is severely reduced, fragmented, and homogenized (Silva & Tabarelli 2000; Lôbo et al. 2011; Mendes Pontes et al. 2016), and is home to several threatened plant and animal species, such as red-handed howler monkeys (Alouatta belzebul) and Brazilwood (Paubrasilia echinata) (Fialho et al. 2014, Gagnon et al. 2020). Additionally, assessing the floristic composition and vegetation structure of the forest patches where blonde capuchin monkeys occur is one of the goals of the National Action Plan for the conservation of primates in northeastern Brazil (Brasil 2011). Therefore, understanding the vegetation structure and composition of forest patches is necessary for understanding ecological patterns and outlining conservation strategies (Oliveira & Amaral 2004).

Considering that species turnover may be related to landscape variables (Arroyo-Rodríguez et al. 2013), in this study we tested the influence of metrics related to landscape composition (forest cover) and configuration (edge density), and geographical distance between Atlantic Forest patches in the state of Pernambuco, northeast Brazil, on β-diversity in order to understand the potential variation between forest patches. We predict that greater differences in forest cover, edge density, and larger geographical distance between landscapes will result in greater floristic differentiation between patches in the landscape (Oliveira et al. 2004; Arroyo-Rodríguez et al. 2013). We also tested whether the β-diversity between forest patches (and sampling plots) is related to differences in vegetation structure (assessed through plant basal area), predicting that higher β-diversity will be observed between patches/plots when vegetation structures demonstrate a greater level of differentiation.

Methods

Study area

This study was conducted in five Atlantic Forest patches in the state of Pernambuco, northeastern Brazil. The sampled forest patches present different characteristics in terms of size (215 ha – 3,000 ha), shape, elevation, topography, matrix type, ownership (public/private), and level of legal protection (Figure 1; Supplementary Information I for more details). All the studied areas are present in the Atlantic Forest domain north of the São Francisco River, more precisely, within the limits of the Pernambuco Endemism Centre (Fialho et al. 2014).

Fig. 1 Location of the sampled Atlantic Forest patches and landscape characterization for each forest patch. The black circle represents the 2-km radius buffer created for extracting the landscape metrics.

Data collection

The floristic composition and vegetation structure data were surveyed from May to November 2018. We surveyed ten 50 m × 2 m (0.1 ha) plots in each forest patch (Gentry 1982). The location of the first surveyed plot in each forest patch was determined using satellite images from Google Earth and was established at least 100 m from the forest edge. After this, we established each new plot at least 500 metres from the previous plot. This procedure was used due to the ease of implementation and low cost. We measured the diameter at breast height (DBH) and quantified all trees and lianas rooted within the plot limits. For this, we established a DBH ≥ 2.5 cm threshold for trees/shrubs and DBH ≥ 1.0 cm threshold for lianas (Gerwing et al. 2006, Gentry 1982). We collected fertile branches from the individuals that could not be identified during the fieldwork for later identification, which was carried out in the Geraldo Mariz herbarium of the Federal University of Pernambuco and the Lauro Pires Xavier herbarium of the Federal University of Paraíba. Both herbaria possess large databases derived from this region. The botanical classification system used in this study followed The World Flora Online (www.worldfloraonline.org).

Predictor variables

We adopted a patch-landscape approach; that is, we collected response variables from each focal forest patch and landscape variables within a given radius from the geographic centre of each focal forest patch (Arroyo-Rodríguez & Fahrig 2014). We used the data collected from vegetational plots (see above) to characterize vegetation structure at the patch scale using basal area as a predictor variable. Thus, we used the software Fitopac version 2.1 (Shepherd 2010) to extract the basal area for each plot. The landscape metrics were measured considering a 2-km radius buffer (total area: 12.7 km-2 – Figure 1) created from the centroid of each surveyed forest patch. The buffer size was defined based on the mean daily path length of blonde capuchin monkeys (ca. 2-km: Rodrigues 2013). Furthermore, we used 2-km buffers to avoid overlapping between the landscapes.

We extracted landscape composition (forest cover in km²) and configuration (edge density) metrics from within these buffers. Edge density was assessed from a division of the total perimeter of forested areas (km) by the total buffer area (km²). We used high-resolution images (30 m × 30 m) obtained from the MapBiomas Project version 5.0 (Souza et al. 2020) to extract landscape metrics from four of the surveyed forest patches (Mata de Bujari [MDB], Mata da Divisa [MDD], Mata dos Macacos [MDM], and Mata de Água Azul [MAA]) using ArcGis 10.2 (ESRI, Redlands, USA). When checking the classified images from MapBiomas in Google Earth Pro, we noticed that one of the forest patches, Mata dos Oito Porcos (MOP), presented an inconsistency in the MapBiomas classification, where areas of banana monoculture were classified as forested areas. To avoid this inconsistency, we evaluated the landscape metrics for the MOP through images from Google Earth Pro for the years 2018–2019. We obtained the geographical distance between forest patches and plots from a distance matrix using the geographical data (lat, long) of the centroid of each forest patch/plot.

Statistical analysis

To verify the floristic dissimilarities between forest patches, we conducted an Analysis of Similarity (ANOSIM) with 9999 permutations using a matrix of dissimilarities based on the Bray-Curtis index and considering species abundance. The R values indicate the biological importance of the differences, ranging from −1 to 1. Values greater than 0 indicate differences between groups, while 0 represents the absence of differences between groups. A negative R value indicates that dissimilarities within groups are greater than the dissimilarities between them (Clarke & Warwick 2001). ANOSIM was performed using the software PAST 2.16 (Hammer et al. 2001). We also created rank-abundance graphs for each forest patch aiming to identify the rare and dominant species (Moreno et al. 2017). This ranking allows us to describe the structure of the community and infer about the potential effects of anthropogenic disturbance on species dominance (Arroyo-Rodríguez et al. 2013).

Before running the richness and diversity analyses, we searched for similarities in the sample coverage between forest patches. We assessed the sample coverage within each forest patch using the coverage estimator suggested by Chao & Jost (2012), which is a less biased estimator since it uses the total number of individuals within a community belonging to the species represented in the sample. We considered a minimum coverage value for all forest patches of 0.80 to obtain reliable and comparable estimates based on the equal coverage between them (Chao et al. 2014; Chao & Jost 2012). All forest patches presented values greater than the determined value, thereby indicating that the estimates were adequate for the desired analyses (MDM = 0.93; MAA = 0.84; MDD = 0.91; MDB = 0.95 and MOP = 0.86). We performed the analysis of the sample coverage using the iNEXT online tool (Chao et al. 2016).

We analysed and compared the β-diversity within forest patches (i.e., between plots) and between forest patches. Patterns of plant β-diversity were analysed with multiplicative diversity decompositions of Hill numbers: ^qD_β = ^qD_γ/^qD_α, where ^qD_γ refers to the observed total (gamma) diversity, and ^qD_α refers to the mean local (alpha) diversity within the study communities (plots or patches). ^qD_β is interpreted as the ‘effective number of completely distinct communities’, as it ranges between 1 (when all communities are identical) and N (i.e., the number of communities) when all communities are completely different from each other (Jost 2007; Tuomisto 2010). As described by Jost (2007, 2010), α-diversity is independent of β-diversity and sample size. Nevertheless, it depends on the parameter q, which determines the sensitivity of the measurement of species’ relative abundances (Jost 2007; Tuomisto 2010). We considered β-diversities of order 0 (⁰D_β), 1 (¹D_β) and 2 (²D_β). While ⁰D_β does not consider species’ abundances, thus giving disproportionate weight to rare species, ¹D_β weighs each species according to its abundance in the community, measuring the turnover of ‘typical’ species; finally, ²D_β favours very abundant species and can be interpreted as the turnover of ‘dominant’ species (Jost 2007; Tuomisto 2010). These three β-diversity measures were calculated using raw estimators with the entropart package (Marcon & Hérault 2015) for RStudio version 1.2.5 (R Core Team 2019). Thus, we calculated the β-diversity for each patch (^qβ_{patch =} ^qγ_land/^qα_patch) and each plot (^qβ_{plot =} ^qγ_patch/^qα_plot). In order to evaluate the importance of β-diversity differences between spatial scales, we compared the relative compositional dissimilarities between communities using the transformation of ^qD_β proposed by Jost (2006) for communities with a different number of samples (i.e., forest patch: N = 5, plot: N = 50): ^qDS = 1 - [(1/ ^qD_β – 1/N)/(1 – 1/N)]. ^qDS ranges between 0, when all samples are identical, and 1, when all samples are completely different.

To identify the correlates of the turnover component of β-diversity between forest patches and sampling plots, we ran ordinary least squares models with all possible combinations of predictor variables and assessed model support through their AICc (Akaike Information Criterion corrected for small samples – Burnham & Anderson 2002), selecting the model with greatest support. Specifically for this analysis, we used the Jaccard index to represent the dissimilarity between forest patches and plots (Carvalho et al. 2012), which was calculated in the package BAT (Cardoso et al. 2015) and was decomposed into turnover and nestedness components (Carvalho et al. 2012). Given that most of the β-diversity was represented by the turnover component (see results), and that the nestedness component is also related to the α-diversity of the forest patches, we focused our analyses only on the correlates of the β-diversity turnover component. We included the difference in basal area (patch variable), edge density, forest cover (landscape variables), and geographic distance as possible correlates of plant turnover. We calculated the Euclidean distance between forest patches and plots for each of these variables in the package vegan (Oksanen et al. 2019). We used RStudio software to run these analyses. The level of statistical significance throughout was p < 0.05.

Results

Overview

We recorded 1,682 individuals (trees = 1,463 and lianas = 219), grouped in 56 families, 116 genera, and 248 species (including 95 morphospecies; Supplementary Information II), with a mean of 33 ± 8 families, 52 ± 16 genera, and 78 ± 31 species per forest patch (Supplementary Information II). A more detailed description of the plant assemblage of the forest patches and their characteristics is provided in Supplementary Information II. The global analysis using ANOSIM showed differences in floristic composition and abundance among the forest patches (R = 0.52, p < 0.0001). When the pair-to-pair comparison was carried out between the forest patches, a difference was found in all possible pairs of combinations between the patches (Table 1). Most of the difference between forest patches (90.4%) was due to species turnover. A similar pattern was found when considering the surveyed plots: turnover represented 80.3% of the β-diversity among plots.

Table 1. Results, R and (p-values), of the ANOSIM considering the pairwise dissimilarity in species richness between the forest patches with the occurrence of the blonde capuchin monkey (Sapajus flavius) in the state of Pernambuco, northeastern Brazil. MDM = Mata dos Macacos, MAA = Mata de Água Azul, MDD = Mata da Divisa, MDB = Mata de Bujari, MOP = Mata dos Oito Porcos

Correlates of plant β-diversity between forest patches and sampling plots

The β-diversity was lower for MAA at a landscape scale (i.e., between forest patches, ^qβ_patch) and MDM at a site scale (i.e., among plots, ^qβ_plot) for all q orders (Supplemental Information III). Geographical distance was the only variable that explained the turnover component of the β-diversity between forest patches (Table 2), although the null model also presented high support (ΔAICc = 0.87). All the other models had little support (i.e., ΔAICc ≥2.52). The model containing only geographical distance explained 42.8% of the variance in the turnover component of the β-diversity between forest patches (Table 2).

Table 2. Results of the best models that explained the turnover component of the β-diversity between the surveyed forest patches and between the surveyed plots in the Atlantic Forest of the Pernambuco Endemism Centre, northeastern Brazil

All the predictor variables (geographic distance, Δbasal area, Δedge density, and Δforest cover) positively affected the turnover component of the β-diversity between surveyed plots and were included in the model with highest support (Table 2). Thus, the turnover was greater between the plots that were more distant from each other and that had more contrasting vegetation structure, forest cover, and edge amounts in their landscape. Among these variables, geographical distance had the strongest effect on the turnover between plots (Table 2). The model containing all the variables explained 14.8% of the variance in the β-diversity turnover component between sampling plots (Table 2). All the other models had little support (i.e., ΔAICc ≥4.36).

Discussion

Our findings reveal that the studied forest patches in the state of Pernambuco differ in floristic composition and that geographical distance is a key correlater of plant species turnover between forest patches and plots. Additionally, the difference in forest structure (basal area) and in landscape context (edge density and forest cover) was also related to a greater turnover between the sampling plots.

The geographical distance had a strong and positive effect on plant species turnover between forest patches. This result may be related to the limited interchanging of seeds (and species) between forest patches, which is observed between more distant sites (Condit et al. 2002; Myers et al. 2013). Indeed, a lack of potential seed dispersers could enhance the turnover between more distant sites, although in our study site there is still a large number of birds and terrestrial mammals which may be able to carry seeds between the forest patches (Lobo-Araújo et al. 2013; Mendes Pontes et al. 2016; Pereira et al. 2016; Campos et al. 2018; Garbino et al. 2018). The effect of geographical distance on species turnover is also explained by the fact that distant areas will likely have more contrasting environmental (e.g., soil, climate, topography) and anthropogenic (e.g., different disturbance regimes) contexts which will affect floristic composition (Pyke et al. 2001; Arroyo-Rodríguez et al. 2013; Garibaldi et al. 2014).

In addition to geographic distance and landscape context, Δbasal area was related to the turnover between the surveyed plots. Different plant species have different allometry (Bohlman & O’Brien 2006) and respond differently to environmental conditions (John et al. 2007, Bohlman et al. 2008) and to disturbance (Sagar et al. 2003, Kumar & Ram 2005). Since there is some variation in environmental conditions and disturbance levels within the forest patches (i.e., between plots), this should affect both the vegetation basal area and floristic composition.

We also found that differences in edge density in the landscape were related to the turnover between forest plots. The edge effect influences many abiotic conditions of the forest patches, such as temperature, light, wind and humidity (Didham & Lawton 1999; Davies-Colley et al. 2000), which also affect floristic composition (Oliveira et al. 2004). These effects shape the turnover between plots. This variable was assessed with buffers around the centroid of the forest patches and thus may also influence the turnover between forest patches; however, we were not able to detect this effect due to our small sample size.

ΔForest cover also affected plant species turnover between sampled plots. This result may be explained by the fact that changes in forest cover affect plant functional groups, where small-seeded softwood pioneer species colonize small forest patches in low forest cover landscapes more efficiently, while hardwood climax trees are more common in landscapes with greater forest cover (Michalski et al. 2007). Thus, floristic composition varies not only in the edge-interior gradient within forest plots (Oliveira et al. 2004) but also over larger scales considering different levels of forest cover in the landscape.

The fact that some of the species present in our sample were not identified has a limited effect on our conclusions. Firstly, the unidentified species were invariably rare, accounting for only a small proportion of the sampled individuals (<5%). Additionally, we collected samples of these individuals, which allowed us to confirm that they correspond to different morphospecies. Thus, this limitation does not change the turnover patterns that we examined in our analysis.

Studies related to the community structure of plants and inventories are an important tool for assessing the succession processes and conservation statuses of forest patches (Turchetto et al. 2017). Overall, our results indicate the need to conserve forest patches spread across a wide area (distant sites), encompassing different landscape contexts with different vegetation structures, in order to conserve greater floristic diversity. This strategy is in accordance with the current Brazilian law (federal law no. 12,651/2012), which determines that 20% of native vegetation cover must be maintained in each rural property in the Atlantic Forest as a Legal Reserve. This legal requirement should result in the maintenance of forest patches in different landscape contexts and spread throughout the whole biome area. However, landowners frequently do not comply with this determination and Brazilian lawmakers often try to change this conservation requirement (Carvalho et al. 2019; Ferrante and Fearnside 2019). Here, we provide evidence to support the current law and the need to ensure law enforcement in order to conserve the great biodiversity of the Atlantic Forest (Tabarelli et al. 2005), especially that of the threatened Pernambuco Endemism Centre.