Introduction

Anthropogenic disturbances are causing severe impacts on biodiversity, especially in the tropics (Malhi et al. 2014). Of particular concern are the changes in species composition (β-diversity) between sites (Socolar et al. 2016). For example, deforestation (Arroyo-Rodríguez et al. 2013; Morante-Filho et al. 2016) and climate change (Keil et al. 2012; Dornelas et al. 2014) are known to affect β-diversity of different organisms across different spatial scales. However, our knowledge on this topic is far from complete, as contrasting responses of β-diversity to disturbance have been documented (Socolar et al. 2016).

Both positive and negative responses of taxonomic β-diversity to disturbance have been documented depending on processes acting at different scales (Arroyo-Rodríguez et al. 2013; Filgueiras et al. 2016; Socolar et al. 2016). For example, human disturbances can benefit some disturbance-adapted species (winners), which may proliferate and replace many disturbance-sensitive species (losers), thereby decreasing β-diversity (biotic homogenization) at landscape and regional scales (Lôbo et al. 2011). Yet, some landscape-scale disturbances, such as forest loss and isolation, can limit dispersal movements, leading to compositional differentiations (higher β-diversity) between localities (Liu et al. 2015; Filgueiras et al. 2016). Such differentiation can also be caused by differences between sites in disturbance regime and/or biotic and abiotic conditions (Tscharntke et al. 2012; Arroyo-Rodríguez et al. 2013).

Changes in taxonomic diversity (i.e., diversity metrics that consider species and their abundances, but not their phylogenetic relatedness) can also alter the patterns of phylogenetic diversity and structure (Webb 2000; Cisneros et al. 2015). As many phenotypic, behavioral, and phenological traits can co-occur in close relatives (i.e., high phylogenetic conservatism), preserving phylogenetic diversity is needed to decrease the chance of losing unique traits, thus increasing ecological resilience and functioning (Cadotte et al. 2012; Wang et al. 2015). As argued by Cisneros et al. (2015, p. 2) the phylogenetic diversity “may represent the long-term evolutionary potential of a biota to respond or adapt to current and future environments”. Yet, the relationship between taxonomic and phylogenetic diversity is highly variable, and depends on the phylogenetic topology and geographical distribution of species (Rodrigues et al. 2005), and on the extent to which traits that influence species distribution along environmental gradients are conserved along the phylogeny (Webb et al. 2002). This is why some recent studies have explored and compared the taxonomic, functional and phylogenetic responses of assemblages (Devictor et al. 2010; Guilhaumon et al. 2015; Chai et al. 2016). Although some studies show congruence between taxonomic and phylogenetic responses (Rodrigues and Gaston 2002), these two dimensions of diversity do not always align (Devictor et al. 2010) given that communities interact in complex ways across environmental gradients and geographical distances (Graham and Fine 2008).

Deforestation is a major threat to global biodiversity (Arroyo-Rodríguez et al. 2020). However, tropical forests are also increasingly threatened by post-deforestation disturbances (e.g., hunting, overgrazing, selective logging and extraction of firewood and non-timber forest products), generally known as chronic disturbances (Martorel and Peters 2005; Barlow et al. 2016; Arroyo-Rodríguez et al. 2017). Chronic disturbances are particularly important in seasonally dry tropical forests (SDTFs), as nearly a third of the global human population lives in or near these forests, using natural resources for their livelihood (Miles et al. 2006). SDTFs are also expected to experience major climate changes, including rising temperature and decreasing annual precipitation in the coming years (Magrin et al. 2014), thus imposing novel and strong environmental pressures to the remaining biota (Choat et al. 2012; Rugemalila et al. 2016; Rito et al. 2017a). In fact, both chronic disturbances and climate change have been reported as important drivers of community assembly in SDTFs (Ribeiro et al. 2015, 2016; Rito et al. 2017a). Yet, to our knowledge, no study to date has assessed the response of taxonomic and phylogenetic β-diversity to chronic disturbances and water scarcity in this species-rich but vanishing ecosystem.

Here, we assessed the patterns of taxonomic and phylogenetic β-diversity across 19 woody plant communities in the Caatinga—the largest SDTF in South America (Silva et al. 2017). This forest has a high level of endemism, and the two most representative families are Fabaceae and Euphorbiaceae (Queiroz et al. 2017), which are also very representative in Neotropical Dry Forests, especially Fabaceae (Gentry 1995). Some species from these families are known to be more adapted (and resistant) to disturbance and are frequently associated with degraded and/or early successional forest stages (e.g., Rito et al. 2017b; Martínez-Ramos et al. 2021). As the Caatinga forest has experienced a long history of chronic anthropogenic disturbance and encompasses high heterogeneity in water availability (Silva et al. 2017), we evaluated how variations in water deficit and disturbance variables (including human access and grazing pressure) affect the taxonomic and phylogenetic β-diversity of plant communities.

We hypothesized that, at the landscape scale, chronic disturbance and water deficit are critical environmental stress factors that can prevent the presence of some species, and promote the proliferation of others (e.g., stress-tolerant species adapted to harsh conditions). A previous study in the region shows that sites with lower environmental harshness (i.e., with higher precipitation and lower disturbance) have higher numbers of rare plant species, whereas more abundant stress-tolerant species are usually distributed along a wider range of the environmental gradient (Rito et al. 2017a). Thus, we also assess how β-diversity patterns are influenced by species rarity and dominance, predicting that taxonomic β-diversity is higher when considering rare than dominant species.

Our expectations for phylogenetic β-diversity depend on species divergence times. If rare species have older divergence times, we predict higher phylogenetic β-diversity values, and thus lower differences between taxonomic and phylogenetic β-diversity, than if they are more recent (i.e., Paleoendemics vs Neoendemics, sensu Graham and Fine 2008). Regarding the effect of environmental variables on taxonomic and phylogenetic β-diversity, we predicted increasing β-diversity between sites with increasing differences in chronic disturbance and water deficit. Yet, as chronic disturbance can be measured with different metrics (e.g. Martorell and Peters 2005; Leal et al. 2014; Ribeiro et al. 2015; Rito et al. 2017a; Câmara et al. 2018; Arnan et al. 2018), and we do not know a priori which metrics can be more important, we did an exploratory analysis with six metrics of chronic disturbance to identify the strongest predictors of taxonomic and phylogenetic β-diversity. In any case, we further posit that the magnitude of these relationships depends on the relative abundance of species, such that stronger relationships are expected when all species are considered (including rare species), than when only dominant species are considered. We further predicted higher taxonomic and phylogenetic β-diversity values between sites more isolated from each other, given the importance of dispersal limitation in promoting the compositional differentiation of plant communities (Arroyo-Rodríguez et al. 2013). All expectations are under the assumption that plant traits associated with vulnerability to extirpation are phylogenetically conserved (i.e., restricted to some clades), such that those clades would respond consistently to changes in environmental gradients.

Methods

Study area

The Caatinga vegetation is a mosaic of seasonally dry tropical forests and shrub stands that covers 912,529 km2 and correspond to 10.7% of the Brazilian territory (Silva et al. 2017). It has been used by human populations since the mid-sixteenth century, and nowadays > 50% of all Caatinga region is degraded by human activities (Antongiovanni et al. 2018). Remaining vegetation offers livelihood support for a dense and low-income population, resulting in chronic disturbances such as selective logging, extraction of firewood and non-timber forest products, grazing by livestock and hunting (Ribeiro et al. 2015). The Caatinga shows a very high annual and interannual seasonality in precipitation, with the majority of rainfall concentrated in three months per year, and with severe droughts that may extend for years (Silva et al. 2017). Mean annual precipitation varies between 450 and 1100 mm, and mean annual temperature between 23 and 27 °C. The Intergovernmental Panel on Climatic Change forecasted an increase of 1.8 °C to 4 °C of temperature and a reduction of 22% in rainfall by 2100 in this region (Magrin et al. 2014), electing Caatinga as one of the top six ecosystems with the largest intrinsic vulnerability to climate variability (Seddon et al. 2016). In this ecosystem there is no natural fire regime.

Study sites

The study sites are described elsewhere (Rito et al. 2017a), but a brief overview is given here. We assessed the Caatinga vegetation of the Catimbau National Park, Pernambuco State, Brazil (8°24′00″ and 8°36′35″ S; 37°0′30″ and 37°1′40″ W). The Park covers ca. 60,000 ha, most of them (70% of the area) dominated by sand quartzolic soils (Rito et al. 2017a). Precipitation is highly variable within the Park (510 to 1100 mm). Local human populations are very poor (IBGE 2019), and usually extract forest products for animal and human food and fuelwood (Rito et al. 2017a). The extensive grazing by stock (goats and cattle) is also common, and is considered among the most important human-caused disturbances to vegetation in the region, creating a mosaic of patches exposed to different levels of chronic disturbance. We used RapidEye satellite imagery and maps of soil to select forest sites that covered a wide range of disturbance and precipitation levels in the study area (Rito et al. 2017a). Then, we established 19 20 × 50-m plots within a 21,430-ha landscape (minimum inter-plot Euclidian distance = 2 km) dominated by old-growth vegetation and with the same relief and soil type (sand soil) (Rito et al. 2017a). As plant species composition can depend on soil types (Moro et al 2015), and the Park is dominated by sand soils, we controled for the potential confounding effect of soil type by locating all sample plots in sand soils.

Chronic disturbance measures

We assessed different indicators of chronic disturbance that have been described as important drivers of human disturbance in tropical forests (e.g., Martorell and Peters 2005; Leal et al. 2014; Ribeiro et al. 2015; Rito et al. 2017a), including indirect and direct (field) measures. The indirect measures of disturbance included: (1) proximity of each plot to the nearest village weighted by the number of houses (Sagar et al. 2003; Martorell and Peters 2005; Leal et al. 2014); (2) proximity of each plot to the nearest road (Sagar et al. 2003; Leal et al. 2014); (3) number of people living in the nearest ranch; (4) number of goats; and (5) number of cattle in the nearest ranch that have access to the plot area. We also estimated directly the wood extraction in the field by using the live stem extraction for fuelwood in the plot (i.e., total basal area of stems cut in the plot) (additional details on the methods used to measure chronic disturbances are included in Supplementary Material Appendix 1).

Annual climatic water deficit data

Annual water availability is an important driver of plant composition and structure in Caatinga (Rito et al. 2017a; Silva et al. 2017) and tropical dry forests in general (Choat et al. 2012). To determine water availability within the plots, we considered the average annual climatic water deficit (D). D represents the potential additional evaporative demand not met by available water based on energy input and precipitation (Stephenson 1998; Lutz et al. 2010). It was calculated based on 30-arc-seconds (1 km) resolution maps of long-term average annual potential evapotranspiration and actual evapotranspiration (CGIAR-CSI’s Global Aridity and PET Database and Global High-Resolution Soil–Water Balance database; Trabucco and Zomer 2009, 2010). These maps are generated using temperature and precipitation data from WorldClim global climate data repository (www.worldclim.org). For each plot, we calculated the difference between annual potential evapotranspiration and actual evapotranspiration to obtain D values. All measures were performed with ArcGIS 10.0 (ESRI 2011). We choose D to represent the water stress sufered by Caatinga among other strongly correlated climatic variables, including: (1) mean annual precipitation; (2) mean annual climatic water deficit; (3) precipitation of driest quarter; and (4) precipitation of driest month (r > 0.85, p < 0. 001, in all cases).

Plant surveys and measures of taxonomic and phylogeny diversity

As described elsewhere (Rito et al. 2017a), we sampled all shrubs and trees with diameter at soil height ≥ 3 cm and total height ≥ 1 m in the 19 0.1-ha plots (see the complete database in Rito et al. 2016). For multi-stem individuals, we summed the diameter at soil height of all stems and considered the individual in the database if total diameter summed ≥ 3 cm. We recorded a total of 5,118 individuals belonging to 104 species (23.3 ± 8.5 species per plot; mean ± SD) and 31 families.

Taxonomic and phylogenetic β-diversity was measured with Hill numbers (Jost 2007; Chao et al. 2010; Chiu et al. 2014). Diversity measures based on Hill numbers obey the doubling rule, i.e., “if we have N equally large, equally diverse groups with no species in common, the diversity of the pooled groups must be N times the diversity of a single group” (sensu Chao et al. 2010). Other diversity metrics, such as those based on entropy measures, do not obey this principle, potentially leading to serious errors in interpretations (Jost 2009). Also, both taxonomic and phylogenetic diversity based on Hill numbers are expressed in effective number of species, which allows direct comparisons of different diversity metrics in the same graph (i.e., so-called ‘diversity profiles’). We computed taxonomic β-diversity (q\(D\)β) using a multiplicative diversity partitioning: q\(D\)β = q\(D\)γ/q\(D\)α; i.e., β-diversity is simply the gamma to mean alpha ratio. The details of the calculation of mean alpha (q\(D\)α) and gamma (q\(D\)γ) diversity components can be found elsewhere (Jost 2007). q\(D\)β gives the effective number of distinct assemblages, and ranges from 1, when all assemblages are compositionally identical, to N assemblages, where all assemblages are compositionally different from each other. We computed q\(D\)β for two orders of q (0 and 2), which are related to well-known diversity metrics, namely: species richness (q = 0) and inverse Simpson concentration (q = 2) (Jost 2007). When q = 0 (0\(D\)β), the calculation of all components of diversity (α, γ, and β) does not consider the abundances of species, and consequently results in a diversity metric strongly sensitive to the number of rare species. When q = 2 (2\(D\)β), very abundant species are weighted more strongly, and the metric can be interpreted as the diversity of dominant species in the community.

Phylogenetic β-diversity was also computed using a multiplicative diversity partition of Hill numbers: q\(\overline{D}\)β(T) = q\(\overline{D}\)γ(T)/ q\(\overline{D}\)α(T). Alpha (q\(\overline{D}\)α(T)) and gamma (q\(\overline{D}\)γ(T)) components give the mean phylogenetic diversity for the interval [− T,0] or mean effective number of species over T years. These phylogenetic β-diversity metrics incorporate information about the pattern of phylogenetic tree branches, relative branch lengths, and the relative abundance of each branch segment represented in the present day in the sample plots (see Chiu et al. 2014). Therefore, q\(\overline{D}\)β(T) gives the effective number of assemblages that are phylogenetically distinct. If all assemblages are similar in species composition and abundance, q\(\overline{D}\)β(T) will be equal to 1, and q\(\overline{D}\)β(T) = q\(D\)β. In contrast, if they are totally distinct in terms of phylogeny, q\(\overline{D}\)β(T) = q\(D\)β = N. q\(\overline{D}\)β(T) was also computed for q = 0 and q = 2, which are more sensitive to rare and dominant species, respectively (Chao et al. 2010; Chiu et al. 2014). To compute q\(D\)β and q\(\overline{D}\)β(T) we used the package entropart for R (Marcon and Herault 2015).

To calculate the phylogenetic metrics, a time-calibrated phylogeny was adopted (Fig. 1; Supplementary Material Table S1). For this, we identified all individuals and produced a species list based on the APG IV (Byng et al. 2016) classification. We searched GenBank (http://www.ncbi.nlm.nih.gov/genbank/) for three genes sequence used in published angiosperm phylogenies: ribulose-bisphosphate carboxylase gene (rbcL), maturase K (matK) and 5.8 S ribosomal RNA gene and performed a divergence-time and phylogenetic analyses using Markov chain Monte Carlo (MCMC) methods (additional details on the methods used to construct the time-calibrated phylogeny are included in Supplementary Material Appendix 1).

Fig. 1
figure 1

Bayesian tree for plant species recorded in 19 plots located within the Catimbau National Park, Brazil. The Bayesian tree was estimated based on three DNA regions: maturase K (matK), 5.8S ribosomal RNA gene (5.8S), and ribulose-1,5-carboxylase/bisphosphate gene (rbcL). Amborella trichocarpa was used as outgroup species to root the tree. The geologic time scale is presented in millions of years ago. Circles indicate species occurrence in plots: red circles = one plot, blue circles = two plots and green circles = three plots. Asterisks indicate species with ≤ 5 individuals

Full size image

Data analyses

We first evaluated the accuracy of plant inventories with the coverage estimator (\(\hat{C}\)n) recommended by Chao and Jost (2012) using the entropart package in R (Marcon and Herault 2015). This estimator calculates the proportion of the total number of individuals in an assemblage that belong to the species represented in the sample. Sample coverage was very high in all sites (> 0.93), indicating that our sampling effort was adequate, and that our diversity estimates are not biased by differences in sample coverage among sites (Chao and Jost 2012; Chao et al. 2014). The following analyses were therefore done with observed values, and not with rarefied or extrapolated values, as recommended by Chao and Jost (2012) for assemblages with similarly high sample coverage.

Values of q\(D\)β and q\(\overline{D}\)β(T) were calculated via two ways: for the whole landscape (i.e., considering 19 plots) and for pairs of plots. When considering pairs of plots (i.e., matrix of dissimilarity), we calculated 95% confidence intervals to test for differences between taxonomic and phylogenetic diversity and between diversity values in different orders of q.

To evaluate the effect of chronic disturbance, water deficit and site location on taxonomic and phylogenetic β-diversity, generalized dissimilarity models were adopted (GDM; Ferrier et al. 2007). GDM is a nonlinear matrix regression technique that analyze patterns of compositional dissimilarity between pairs of locations in relation to environmental dissimilarity and geographical distance (Ferrier et al. 2007). The nonlinearities in compositional dissimilarity occur because it can increase non-monotonically with increase of environmental distance and because the compositional turnover rates may vary across environmental gradients (Ferrier et al. 2007). The GDM is based on a curvilinear relationship, where the predicted ecological distance, called the linear predictor, includes the adjustment of the ecological distance functions of the predictor variables by maximum likelihood. Each predictor variable has a function between partial ecological distance (dissimilarity) and gradient. When the coefficient is zero, it means that the predictor variable is not able to explain the biological dissimilarity (Ferrier et al. 2007).

To reduce the number of predictor matrices in the complete model, we first used the gdm function from the gdm package in R (Manion et al. 2018) to select those chronic disturbance predictors with higher explanatory power. We used univariate GDMs to relate each response matrix to each predictor matrix (i.e., 4 β-diversity metrics × 6 predictors = 24 models). We then selected those chronic disturbance predictors for which the sum of function coefficients (i.e., I-splines) was higher than zero (Supplementary Material Table S2). In particular, we excluded proximity of each plot to the nearest village, number of people living in the nearest ranch, and number of goats. Then, we used the gdm function from the gdm package to assess the relative effect of water deficit, three metrics of chronic disturbance (proximity to nearest road, wood extraction in the plot, and number of cattle in the nearest ranch) on each β-diversity metric with multivariate GDMs, accounted spatial autocorrelation by including the geographical distance as a predictor variable. We used the gdm.varImp function of the gdm package to test the complete model significance, to quantify variable importance (explained deviance) and test for statistical significance for each predictor separately using matrix permutation with 999 randomizations. The significance of the complete model is tested by comparing the deviance explained of a full model using un-permuted environmental data to the distribution of deviance explained values from GDM fit to “n” permuted tables. The variable’s significance is assessed by the same process repeated for each predictor individually (i.e., only the data for the tested variable is permuted) and its importance is quantified as the percent change in deviance explained between a model fit with and without that variable (Ferrier et al. 2007).

Results

A total of 5,118 individuals from 104 species were recorded across 19 0.1-ha plots. A high percentage of these species can be considered rare, as 43% were restricted to 1–2 plots, and 28% showed < 6 individuals. Rarity was present in all major clades (Fig. 1), but the richest families (Fabaceae, Euphorbiaceae and Myrtaceae) contributed less to this group (Fig. 1; Supplementary Material Fig. S1). In fact, rarity was concentrated in species-poor families such as Boraginaceae, Celastraceae, Sapotaceae and Salicaceae.

At the landscape scale, taxonomic and phylogenetic β-diversity were notably higher when considering all species (including rare species) than when considering only the dominant species, although this pattern was stronger for the phylogenetic dimension of β-diversity (Fig. 2a). The mean taxonomic β-diversity between pairs was similarly high when considering all species (0Dβ-plot = 1.7 ± 0.1; mean ± DE) and dominant species (2Dβ-plot = 1.7 ± 0.2), but mean phylogenetic β-diversity was notably higher when considering all species [0\(\overline{D}\)β(T)plot = 1.4 ± 0.1] than when only considering the dominant species [2\(\overline{D}\) β(T)plot = 1.1 ± 0.0], thus indicating that rare species increase total phylogenetic β-diversity (Fig. 2b).

Fig. 2
figure 2

Plant β-diversity in 19 0.1-ha plots with different water deficit levels and exposed to chronic anthropogenic disturbance in the Catimbau National Park, Brazil. Taxonomic and phylogenetic β-diversity among all plots in the landscape (a), and between pairs of plots (b) are shown for two orders of q (0 and 2), which determines the sensitivity of the diversity measure to relative abundances. Boxplots show median (bold horizontal line), first and third quartiles (lower and upper box limits, respectively). The lower and upper whiskers represent minimum and maximum values and open circles represent outliers

Full size image

The full GDM models were significant (p < 0.05) for taxonomic β-diversity (0Dβ-plot = 41.7%; 2Dβ-plot = 21.4%) and phylogenetic β-diversity of q = 0 (0\(\overline{D}\)β(T)plot = 16.8%), but not for the phylogenetic β-diversity of q = 2 (2\(\overline{D}\)β(T)plot = 17.9%; p = 0.05). Water deficit and number of cattle were the strongest predictors of taxonomic and phylogenetic β-diversity between plots for all species (q = 0), but not for dominant (q = 2) species (Table 1). These results were not biased by differences in variation among predictor variables (Supplementary Material Appendix 2; Table S3). In particular, water deficit was the best predictor of compositional dissimilarity for all species (q = 0), as it was that most contributes to the explanation of the predicted ecological distance for both, taxonomic and phylogenetic β-diversity (Function coefficient value: 0Dβ-plot = 0.57 and 0\(\overline{D}\)β(T) plot = 0.25 from total; Fig. 3a, n). Number of cattle was the second-best predictor, with 0.36 and 0.16 function coefficient value for the same variables, respectively. Finally, geographical distance was positively related only to taxonomic β-diversity, independently of the q order, but the deviance explained by this predictor was relatively low (Table 1; Function coefficient value: 0Dβ-plot = 0.18 and 2Dβ-plot = 0.28 from total; Fig. 3f, l).

Table 1 Generalized dissimilarity models using matrix permutation showing the effect of water deficit, three indicators of chronic anthropogenic disturbance and inter-plot isolation distance on the taxonomic and phylogenetic β-diversity of plant communities in the Brazilian Caatinga
Full size table
Fig. 3
figure 3

Results for taxonomic and phylogenetic β-diversity from generalized dissimilarity modelling (GDM). We modeled the effect of water deficit, three indicators of chronic anthropogenic disturbance and inter-plot isolation distance on plant β-diversity in the Catimbau National Park, Brazil. Partial ecological distance was determined by summing the coefficients of the I-splines from multivariate generalized dissimilarity models. The I-splines are partial regression fits and represents the importance of each predictor in determining patterns of β-diversity and total amount of compositional dissimilarity associated with each variable (i.e. the greater the ecological distance, the greater the explanatory power of the predictor). The shape of each function indicates how the rate of compositional dissimilarity varies along gradients and the maximum height reached by each curve indicates the total amount of dissimilarity associated with each variable, holding all other variables constant. The graphs a, g, m and s represent the relationship between Observed compositional dissimilarity (Obs. Comp. dissimilarity; i.e. each point represents the dissimilarity between pairs of localities and can vary from 0 to 1) and the predicted ecological distance of complete model (i.e. including water deficit, three indicators of chronic disturbance and geographical distance). If the curve reaches an asymptote, it means saturation of dissimilarity, that is, the variation of predicted ecological distance beyond it will not cause an increase in dissimilarity. Taxonomic β-diversity (qDβ) and phylogenetic β-diversity [q\(\overline{D}\)β (T)] were calculated considering two orders q (0 and 2), which determine the sensitivity of each β-diversity component to the relative abundances (see “Methods” section). Dark blue = 0Dβ (a to f), light blue = 2Dβ (g to l), dark red = [0\(\overline{D}\)β (T)] (m to r) and light red = [2\(\overline{D}\)β (T)] (s to x)

Full size image

The highest taxonomic and phylogenetic β-diversity values are represented by the highest function slope coefficients (Fig. 3). Regarding this, the highest differences in species composition along the water deficit gradient were found above ~ 600 mm for taxonomic and up to ~ 900 mm for phylogenetic β-diversity (Fig. 3b, n). The highest differences in species composition along the number of cattle gradient were found above ~ 70 for both, taxonomic and phylogenetic β-diversity (Fig. 3c, o). Finally, the highest differences in species composition along the geographical distance gradient was found up to ~ 20 km when considering taxonomic β-diversity (both 0Dβ-plot and 2Dβ-plot; Fig. 3f, l).

Discussion

Our results indicate that, as predicted, plant taxonomic and phylogenetic β-diversity at the landscape scale is higher when considering rare species than when considering only the dominant species. This can be related to the distribution and abundance of a few taxa from the most species-rich families (e.g., Fabaceae and Euphorbiaceae), which appear to do relatively well in the landscape, decreasing β-diversity when considering dominant species. Yet, as most species are rare, and rarity is distributed across most taxa, phylogenentic β-diversity was relatively higher when considering all species, including rare ones. Our findings also support the hypothesis that chronic disturbance and water deficit are critical environmental stress factors that shape plant β-diversity in the Caatinga dry forest of the Catimbau National Park, Brazil. In particular, both taxonomic and phylogenetic β-diversity between sites were significantly related to inter-site differences in water deficit and number of cattle, and inter-site isolation was another complementary driver structuring plant communities.

As expected, the magnitude of β-diversity responses depends on relative species abundance. Specifically, the impact of water availability and cattle grazing on taxonomic and phylogenetic β-diversity is only evident when assessing all species (q = 0), but not when considering dominant species (q = 2). This not only suggests that rare species are more susceptible to these environmental drivers than dominant species (Jost 2007; Chao et al. 2014; Chase et al. 2018), but that the responses can be mainly detected when considering larger sampling efforts: i.e., there is a scale-dependence in plant β-diversity responses (Chase et al. 2018). In other words, it seems that water availability and cattle grazing do not cause shifts in dominance and interspecific aggregations from site to site (see below), but by moving from small to large scale there is a relatively higher compositional difference in rare species associated with these environmental pressures. This is consistent with the high number of rare species we detected, including large trees associated with more humid stands in the Caatinga dry forest, such as Guapira graciliflora, Handroanthus impetiginosus and Libidibia ferrea (Rito et al. 2017a) and those sensitive to chronic disturbances (Ribeiro et al. 2016).

In general, we found consistent taxonomic and phylogenetic β-diversity responses to chronic disturbances and water availability, thus suggesting that there is a relatively high phylogenetic conservatism in the traits that confer vulnerability to these environmental forces. This finding can be related to the fact that those species that have more restrict distribution along the environmental gradients share ecological traits (e.g., low abundance, low tolerance to disturbance and/or drought) can co-occur in close relatives (e.g., Myrtaceae, Boraginaceae, Salicaceae, Sapotaceae, Apocynaceae, Annonaceae, Senna species; Rito et al. 2016; Fig. 1) and are more prone to extinction. As consequence, chronic disturbances and water availability show similar impact on both taxonomic and phylogenetic β-diversity. Other studies have also found similar responses to chronic disturbances and water deficit in phylogenetically related species (Martínez-Blancas et al. 2018; Aguirre-Gutiérrez et al. 2020).

The importance of water availability for plant assemblages is known to be positively related to plant species richness (Francis and Currie 2003; Hawkins et al. 2003). We showed in a previous study that precipitation is a major driver of local (α) diversity patterns in Caatinga, and that plant responses to disturbances can actually be mediated by precipitation (Rito et al. 2017a). Thus, differences in water availability may accentuate the species turnover among sites, especially for those species that are more physiologically restricted (Esquivel-Muelbert et al. 2017). Consistent with this hypothesis, taxonomic β-diversity of dominant species was not related to water deficit. We expected stronger effects on rare species than on dominant species, because rare species seem to be more dependent on water availability (Rito et al. 2017a). Also, dominant species are expected to be widely distributed, thus reducing compositional differences along the water deficit gradient. Consistent with this prediction, we found that the group of dominant species (n = 14 species) is not limited to a certain range of water deficit, but most species occupy several plots along this gradient (Supplementary Material Fig. S2).

Regarding grazing activities pressure, cattle seem to have a greater impact than goats on changes in Caatinga woody flora. Empirical evidence suggests that larger herbivores have stronger impacts on plant communities than smaller herbivores (Bakker et al. 2006). Therefore, although goats are more generalist herbivores than cattle (Coppock et al. 1986), cattle exert greater soil compaction pressure (77.5% more than goat; Silva et al. 2014). It is therefore likely that differences in cattle grazing activities (i.e., trampling, soil compaction and mineralization by deposition of urine and feces) among sites create heterogeneous environments that may affect seed germination and seedling recruitment, generating increases in β-diversity between sites (de Bello et al. 2007). The fact that this pattern was not found when assessing dominant species (Table 1) suggests that plant traits (e.g., germination ability, desiccation tolerance, sprout ability) associated with vulnerability to extirpation in sites exposed to grazing activities are probably convergent or have low phylogenetic signal to dominant species (Santo-Silva et al. 2018). However, this hypothesis needs to be tested in future studies. Yet, there are 15 exclusive rare species (≤ 5 individuals per plot) in plots associated with high cattle number (i.e., n > 70; Fig. 3c). All of them are large tree species (from Anacardiaceae, Combretaceae, Boraginaceae, Salicaceae, Sapotaceae families) that could be previous stablished to cattle introduction and/or have conspicuous protection mechanisms against herbivory. Indeed, adult individuals are more prone to survive and persist in a context of chronic disturbance than seedlings ones (see Rubio-Méndez et al. 2018). Additionally, within these rare species set (including trees and some few shrubs) the majority have chemical (e.g., Anadenathera columbrina, Mimosa tenuiflora, Thiloa glaucocarpa, Jatropha mollisima) or physical (e.g., Sideroxylon obtusifolium, Xylosma ciliatifolia, Pilosocereus gounellei, Cereus jamacaru, Mimosa ophtalmocentra) protection to cattle’s herbivory. Probably, these strategies have benefited the species and favored their establishment and/or persistence over shrub/dominant species that does not have protection mechanisms.

Plant assemblages are not only affected by environmental stress factors, but they can also be shaped by dispersal limitation. Many factors can limit the dispersal capacity of plant species, including lower levels of self-incompatibility, a tendency toward asexual reproductive pathways, a short flowering season, and limited seed dispersal capacity (Kunin and Gaston 1993; Murray et al. 2002). Our findings indicate that β-diversity increases sharply when comparing sites up to ~ 20 km apart, suggesting that this is the range of distances within which dispersal limitation has a larger influence in shaping our plant communities (Fig. 3f, l). Interestingly, the fact that isolation distance showed a weaker effect on phylogenetic than on taxonomic β-diversity suggests that there may be a low phylogenetic signal in traits related to dispersal (Arroyo-Rodríguez et al. 2012; Santo-Silva et al. 2018)—a very important avenue for future research.

Conservation implications

Our findings have important applied implications to the appropriate management and conservation of this (and potentially others) tropical dry forests. Differences among localities in these global change factors (i.e., human disturbances and climatic change) increase compositional differences of species and lineages, thus changing landscape-scale (γ) taxonomic and phylogenetic diversity. This suggests that conserving forest areas across the whole Caatinga region, even though exposed to different disturbance intensities and climate variables, and preventing the loss of rare species is critically needed to maintain the compositional differentiation of biotic assemblages, and thus the total (γ) species diversity in this rich but vanishing tropical ecosystem. This can be done by applying management strategies related to the so-called biodiversity-friendly (sensu Melo et al. 2013) or optimal landscapes (sensu Arroyo-Rodriguez et al. 2020). These strategies include the maintenance of a relatively high forest cover, most of it in a many relatively small forests patches scattered through the landscape to preserve different assemblages, and thus β-diversity (Arroyo-Rodriguez et al. 2020). This will favor the ecological resilience of this species-rich biome, as preserving biologically and functionally distinct components of this ecosystem also favoring the maintenance of ecosystem services.