DOI: 10.1111/ddi.

12581

BIODIVERSITY RESEARCH

Dissecting a biodiversity hotspot: The importance of

environmentally marginal habitats in the Atlantic Forest
Domain of South America

Danilo M. Neves1 | Kyle G. Dexter2,3 | R. Toby Pennington3 | Arthur S. M. Valente4 |

Marcelo L. Bueno5 | Pedro V. Eisenlohr6 | Marco A.L. Fontes7 | Pedro L. S. Miranda2 |
Suzana N. Moreira8 | Vanessa L. Rezende8 | Felipe Z. Saiter9 | Ary T. Oliveira-Filho8

1
Department of Ecology and Evolutionary
Biology, University of Arizona, Tucson, AZ, Abstract
USA Aim: We aimed to assess the contribution of marginal habitats to the tree species rich-
2
School of GeoSciences, The University of
ness of the Mata Atlântica (Atlantic Forest) biodiversity hotspot. In addition, we aimed
Edinburgh, Edinburgh, UK
3 to determine which environmental factors drive the occurrence and distribution of
Royal Botanic Garden Edinburgh, Edinburgh,
UK these marginal habitats.
4
Instituto Estadual de Florestas, Minas Gerais, Location: The whole extension of the South American Atlantic Forest Domain plus
Ubá, Brazil
5
forest intrusions into the neighbouring Cerrado and Pampa Domains, which comprises
Laboratório de Ecologia e Evolução
de Plantas, Departamento de Biologia rain forests (“core” habitat) and five marginal habitats, namely high elevation forests,
Vegetal, Universidade Federal de Viçosa, rock outcrop dwarf-­
forests, riverine forests, semideciduous forests and restinga
Viçosa, Minas Gerais, Brazil
6 (coastal white-­sand woodlands).
Laboratório de Ecologia, Universidade do
Estado de Mato Grosso, Alta Floresta, Brazil Methods: We compiled a dataset containing 366,875 occurrence records of 4,431
7
Departamento de Ciências tree species from 1,753 site-­checklists, which were a priori classified into 10 main
Florestais, Universidade Federal de Lavras,
Lavras, Brazil
vegetation types. We then performed ordination analyses of the species-­by-­site ma-
8
Programa de Pós-Graduação em Biologia trix to assess the floristic consistency of this classification. In order to assess the rela-
Vegetal, Universidade Federal de Minas tive contribution of environmental predictors to the community turnover, we
Gerais, Belo Horizonte, Brazil
9 produced models using 26 climate and substrate-­related variables as environmental
Instituto Federal do Espírito Santo, Santa
Teresa, Espírito Santo, Brazil predictors.
Results: Ordination diagrams supported the floristic segregation of vegetation types,
Correspondence
Danilo M. Neves, Department of Ecology and with those considered as marginal habitats placed at the extremes of ordination axes.
Evolutionary Biology, University of Arizona,
These marginal habitats are associated with the harshest extremes of five limiting
Tucson, AZ, USA.
Emails: [email protected] and factors: temperature seasonality (high elevation and subtropical riverine forests),
[email protected]
flammability (rock outcrop dwarf-­forests), high salinity (restinga), water deficit sever-
Funding information ity (semideciduous forests) and waterlogged soils (tropical riverine forests).
Coordenação de Aperfeiçoamento de Pessoal
Importantly, 45% of all species endemic to the Atlantic Domain only occur in marginal
de Nível Superior, Grant/Award Number:
BEX 13197-13-4; Fundação de Amparo à habitats.
Pesquisa do Estado de Minas Gerais; Natural
Main conclusions: Our results showed the key role of the poorly protected marginal
Environment Research Council, Grant/Award
Number: NE/I028122/1; Conselho Nacional habitats in contributing to the high species richness of the Atlantic Domain. Various
de Desenvolvimento Científico e Tecnológico,
types of environmental harshness operate as environmental filters determining the
Grant/Award Number: 151002/2014-2 and
301644/88-8 distribution of the Atlantic Domain habitats. Our findings also stressed the importance
of fire, a previously neglected environmental factor.
Editor: Kenneth Feeley

898 | © 2017 John Wiley & Sons Ltd wileyonlinelibrary.com/journal/ddi Diversity and Distributions. 2017;23:898–909.
 14724642, 2017, 8, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ddi.12581 by CAPES, Wiley Online Library on [08/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
NEVES et al. | 899

KEYWORDS
campo rupestre, climate, conservation assessment, flammability, rain forests, restinga, stress
gradients, variation partitioning

1 | INTRODUCTION

The Atlantic Forest of South America, or the Mata Atlântica as it is

known in Brazil where it largely occurs, stretches for over 3,500 km
across equatorial, tropical and subtropical latitudes, and is renowned
world-­wide for being one of the 35 biodiversity hotspots for conserva-
tion prioritization (Myers, Mittermeier, Mittermeier, Fonseca, & Kent,
2000). Its importance is also demonstrated by its designation as one of
the five primary vegetation “Domains” of Brazil (Ab’Sáber, 2003; IBGE,
1993), the others being the Caatinga, Cerrado, Pampa and Amazon
Domains. The Atlantic Forest Domain (hereafter Atlantic Domain)
borders all the other Domains except for the Amazon. The prevailing
land cover of these bordering Domains are semi-­arid thorn woodlands
in the Caatinga, woody savannas in the Cerrado and prairies in the
Pampa. Species from rain forests, the habitat that originally prevailed
F I G U R E 1 Environmental variables (arrows) hypothesized in
in the Atlantic Domain, become a minor component of the landscape
Scarano (2009) as key factors limiting plant species distribution
in these neighbouring Domains, and they are only found in riverine or across the Atlantic Domain of South America. The harshest extremes
high elevation forest enclaves. give rise to distinct vegetation types, referred to as marginal habitats.
Environmental restriction to the establishment of the rain forest Coastal white-­sand woodlands are called restinga in Brazil
habitat is certainly operating at the boundaries of the Atlantic Domain.
In a seminal paper, Scarano (2009) proposed a list of five key fac- sectors of the Atlantic Domain, such as the south-­east (Oliveira-­Filho
tors limiting the occurrence and distribution of rain forest species in & Fontes, 2000; Eisenlohr & Oliveira-­Filho, 2015), the subtropical
the Atlantic Domain, which at its harshest extremes give rise to dis- South (Oliveira-­
Filho, Budke, Jarenkow, Eisenlohr, & Neves, 2015)
tinct habitats (one for each factor), referred to as marginal habitats. and the highly biodiverse central region in eastern Bahia state, north-­
Therefore, the rain forest is placed by Scarano (2009) as the “core” ex- eastern Brazil (Saiter, Eisenlohr, Barbosa, Thomas, & Oliveira-­Filho,
pression of the Atlantic Domain, where deep shade plays the chief role 2016). However, the whole of the Atlantic Domain has only been in-
as a limiting factor for competing plants. The five marginal habitats are vestigated for epiphytic angiosperms (Menini-­Neto, Furtado, Zappi,
high elevation forests, rock outcrop dwarf-­forests, riverine forests, sea- Oliveira-­Filho, & Forzza, 2016). Also, the Atlantic Domain is affected
sonally dry forests and restinga (coastal white-­sand woodlands). Most by fire in much of its distribution (Archibald, Lehmann, Goméz-­Dans,
of these marginal habitats have a relatively high density of trees and & Bradstock, 2013), though to a lesser extent than in surrounding
can be considered forests, albeit not as well developed structurally as Domains, such as in central (Cerrado woody savannas) and southern
rain forests. High elevation forests are primarily associated with frost, Brazil (Pampa prairies). Nevertheless, the potential effect of fire in
with secondary limitation imposed by drought (leeward rain-­shadow) limiting plant species distribution across the Atlantic Domain is yet to
and high-­light intensity. Cloud forests and Araucaria-­dominated for- be investigated. Here we bring together a novel and comprehensive
ests are the main vegetation types of highlands in the Atlantic Domain. dataset assembled on the composition of tree communities across the
Rock outcrop dwarf-­forests, found at lower elevations (and even at the whole Domain (c. 2,000 community surveys across core and marginal
seashore), are primarily limited by the paucity, or even lack, of soil and habitats, with >1,000 sites representing surveys not used in the afore-
related poor water retention. Meanwhile, riverine forests are associ- mentioned studies), combined with environmental data, focusing on
ated with waterlogging on lowland plains and riverbeds. Seasonally testing Scarano’s proposed limiting variables as well as factors that
dry forests (either deciduous or semideciduous) replace rain forests were neglected in previous studies (e.g., fire).
where seasonal rainfall regimes bring regular periods of drought. Besides the importance for community ecology, understanding
Finally, environmental harshness for restinga is primarily associated the degree to which limiting factors drive community differentiation
with salinity, with secondary limitations imposed by drought and low is inherently relevant for conservation. The Atlantic Domain houses
fertility in mineral nutrients (Scarano, 2009) (Figure 1). c.18,000 plant species (REFLORA, 2017), but the current high levels of
Within limited areas, some studies have confirmed the lead- fragmentation and the continuous habitat loss throughout the Domain
ing role of Scarano’s limiting factors as distribution filters for plants. have raised several concerns in the scientific community (Galindo-­Leal,
These studies addressed tree species composition for particular Jacobsen, Langhammer, & Olivieri, 2003; Joly, Metzger, & Tabarelli,
 14724642, 2017, 8, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ddi.12581 by CAPES, Wiley Online Library on [08/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
900 | NEVES et al.

2014; Tabarelli, Pinto, Silva, Hirota, & Bedê, 2005; Tabarelli, Silva, increasingly larger areas towards the south and become wide enough
& Gascon, 2004). Therefore, we believe the time is ripe for studies to reach eastern Paraguay and north-­eastern Argentina. Araucaria-­
aiming to test the overall importance of environmental conditions in dominated forests are also a very important component of the sensu
controlling the occurrence and distribution of plant species across the lato concept, followed by coastal woodlands on white-­sand substrates
whole extent of the Atlantic Domain and, more importantly, across (termed restingas) and three highland dwarf-­
forests: rocky cloud
both its core and marginal habitats. dwarf-­forests, rocky semideciduous dwarf-­forests and rocky highland
We addressed the following questions: (1) Are the patterns of savannas (termed campos rupestres).
tree species distribution across the Atlantic Domain, and its intrusions The sensu latissimo concept of Atlantic Domain habitats proposed
into neighbouring Domains, limited by factors associated with water by Oliveira-­Filho et al. (2006) surpasses the geographical limits of the
deficit (via both soil depth and dry season), water excess (via water- Atlantic Domain to include riverine and deciduous forest tracts oc-
logging), frosts (via low temperature) and soil salinity? If previously curring in the neighbouring Domains as a secondary component of
unrecognized environmental conditions are the main factors explain- the landscape, though with a typically Atlantic Domain flora. In the
ing the patterns of tree species distribution, Scarano’s (2009) limiting present contribution, we adopt this concept because it allows a more
factors should account for a small proportion of the variation in com- complete inclusion of marginal habitats. However, deciduous forests
munity composition explained by environmental factors; (2) are these found in the Cerrado and Pampa Domains, one of the forest types in
limiting factors leading to floristically distinct marginal habitats? If the the sensu lato concept (IBGE, 1993), were not included in this contri-
community composition of the marginal habitats is simply a nested bution because previous studies (e.g., Eisenlohr & Oliveira-­Filho, 2015;
subset of the more diverse Atlantic Domain rain forest, species turn- Oliveira-­Filho et al., 2006) have demonstrated that their flora is dis-
over should account for a small fraction of the dissimilarity between tinct and more closely related to that of semi-­arid woodlands (e.g., in
rain forest and marginal habitats; and (3) what is the contribution of the Caatinga Domain).
these marginal habitats to the overall high species richness of the
Atlantic Domain?
2.2 | Dataset
We extracted the dataset from the NeoTropTree (NTT) database
2 | METHODS
(http://prof.icb.ufmg.br/treeatlan), which consists of tree species
checklists (trees defined here as freely standing woody plants >3 m
2.1 | Study area
in height) compiled for geo-­referenced sites, extending from southern
The Atlantic Forest, designated as one of the five phytogeographi- Florida (U.S.A.) and Mexico to Patagonia. NTT currently holds 5,126
cal “Domains” of Brazil (Ab’Sáber, 2003; IBGE, 1993), occurs primar- sites/checklists, 14,878 woody plant species and 920,129 occurrence
ily along the Atlantic coast and is bordered by the Pampa Domain records. A site/checklist in NTT is defined by a single habitat, following
(woody prairies) of southern Brazil and by the “dry diagonal,” a cor- the classification system proposed by Oliveira-­Filho (2015), contained
ridor that includes three other phytogeographical Domains: Caatinga in a circular area with a 10-­km diameter. Therefore, where two or more
(largely semi-­arid thorn woodlands) of north-­eastern Brazil, Cerrado habitats co-­occur in one 10-­km area, there may be two geographically
(largely woody savannas) of central Brazil, and Chaco (largely semi-­ overlapping sites in the NTT database, each for a distinct habitat.
arid thorn woodlands) of Paraguay–Argentina–Bolivia (IBGE, 1993, The data were originally compiled from an extensive survey of
Prado & Gibbs 1993, Neves, Dexter, Pennington, Bueno, & Oliveira-­ published and unpublished (e.g., PhD theses) literature, particularly
Filho, 2015). The South American Atlantic Forest Domain (hereafter those on woody plant community surveys and floristic inventories.
Atlantic Domain) has a history of controversies over its geographical Moreover, new species occurrence records obtained from both major
circumscription and associated terminology. The controversy may be herbaria and taxonomic monographs have been added to the check-
summarized by three main concepts of Atlantic Domain habitats: the lists when they were collected within the 10-­km diameter of the
sensu stricto, sensu lato and sensu latissimo concepts (Oliveira-­Filho, original NTT site and within the same habitat. All species and their
Jarenkow, & Rodal, 2006). The first, and most restrictive concept, in- occurrence records were checked regarding current taxonomic and
cludes only the tracts of rain forests that occur as a narrow band along geographical circumscriptions, as defined by the team of specialists
the coast (<100 km wide and up to 2500 m elevation) and stretches responsible for the online projects Flora do Brasil and Flora del Conosur
all through the Domain, though with two main interruptions, the São (available at http://floradobrasil.jbrj.gov.br/ and http://www.darwin.
Francisco Gap and Campos dos Goytacazes Gap. The former is a semi-­ edu.ar/, respectively). NTT does not, therefore, include occurrence
arid nucleus at the mouth of the São Francisco River (~10°30′S), and records with doubtful identification, location or habitat, nor sites with
the latter is a seasonally dry region extending from southern Espírito an indication of high anthropogenic disturbance. The latter is assessed
Santo to northern Rio de Janeiro (RJ) States, with its driest extreme at by taking into account the information available in the studies that
Cabo Frio/RJ (~22°50′S). comprise the checklists, and by direct observation of site surface on
The sensu lato concept of Atlantic Domain habitats, which is Google Earth©. It also excludes checklists with low species richness
currently prevalent, includes other habitats adjacent to rain forests, (<20 species), because this is often due to low sampling/collecting ef-
such as the much more extensive semideciduous forests that cover forts, which results in poor descriptive power.
 14724642, 2017, 8, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ddi.12581 by CAPES, Wiley Online Library on [08/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
NEVES et al. | 901

This study used a subset of tree inventories from the NTT data- (mm) were obtained from interpolating known values as response vari-
base, consisting of 328 rain forest sites and 1,425 sites representing ables (data obtained from 135 and 57 Brazilian Meteorological Stations
the limiting environmental factors and marginal habitats proposed by measuring frost frequency and cloud interception, respectively) with
Scarano (2009), namely seasonally dry (663 semideciduous forests), elevation, latitude and the WorldClim layers as predicting variables.
high elevation (193 Araucaria-­dominated forests and 61 cloud forests), Potential evapotranspiration (mm) and the aridity index (annual precip-
rock outcrops (49 rocky cloud dwarf-­forests, 31 rocky semideciduous itation/potential evapotranspiration) were obtained from Zomer et al.
dwarf-­forests and 41 campos rupestres), high salinity (181 restingas— (2007), Zomer, Trabucco, Bossio, van Straaten, and Verchot (2008).
with only forests and dwarf-­forests of the mosaic included) and water- Surface rockiness (% exposed rock), soil coarseness (% sand), soil
logged soils (133 tropical riverine forests and 73 subtropical riverine fertility (% base saturation) and soil salinity (ds/m) were obtained from
forests). Note that marginal habitats associated with seasonal drought the Harmonized World Soil Database v 1.2 (available at http://www.
and high salinity are represented by one vegetation type, whereas fao.org/soils-portal/soil-survey/soil-maps-and-databases/harmo-
high elevation, rock outcrops and waterlogged soils are represented nized-world-soil-database-v12/en/) and ranked afterwards by mid-­
by more than one vegetation type. The final species matrix contained class percentage. The use of classes was adopted to add robustness
presence/absence data for 4,431 tree species across 1,753 sites, with to the data because of the high local soil heterogeneity that makes
a total of 366,875 presences (see Figure 2a and b). raw figures unrealistic. Soil drainage classes were obtained following
The NTT database also included 26 environmental variables for all EMBRAPA’s protocol (Santos et al., 2013), which combines soil type,
its sites, derived from multiple sources (at a 30 arc-­second resolution; texture and depth with landforms. Soil drainage classes, mean annual
detailed below). The resolution used in this study was particularly ap- precipitation (Hijmans et al., 2005) and the aforementioned indices of
propriate (1 km2) because all sites are more than 1 km distant from each water deficit and excess were also combined to produce a hyperseason-
other (only 124 of 1,753 sites are less than 5 km distant from another ality index. Grass coverage (%) was used as a proxy of fire return interval
site, and the mean distance between all sites is >1,000 km). Elevation at (i.e., frequency). Previous studies give support to grass coverage as a
the NTT site centre was used as an integrative environmental variable. good proxy of fire frequency (Archibald et al., 2013; Hoffmann et al.,
Mean annual temperature, mean daily temperature range, isothermality, 2012; Lehmann et al., 2014), although further quantification of fire re-
temperature seasonality, maximum temperature of the warmest month, gime is clearly needed (c.f. Archibald et al., 2013). Grass coverage was
minimum temperature of the coldest month, temperature annual range, obtained by direct observation of site surface on Google Earth© images
mean annual precipitation, precipitation of the wettest month, precipi- in five 100 × 100 m areas, one at the central coordinates of the NTT site
tation of the driest month and precipitation seasonality were obtained and four at 2.5 km away from it and towards the NE, SW, NW and SE.
from WorldClim 1.4 data layers (Hijmans, Cameron, Parra, Jones, & Further details of NTT history, products and protocols can be
Jarvis, 2005). WorldClim monthly temperatures and precipitation were found at http://prof.icb.ufmg.br/treeatlan.
also interpolated to obtain values for 5-­day intervals by applying sinusoi-
dal functions centred at day 15 of each month. These functions yielded
2.3 | Analyses of community turnover
values for days 1, 5, 10, 20, 25 and 30, which were used to generate
Walter’s Climate Diagrams (Walter, 1985) and, thus, four additional We first explored the patterns of floristic differentiation between rain
variables: duration (days) and severity (days) of both the water deficit forest and marginal habitats by performing non-­metric multidimen-
and water excess periods. Frost frequency (days) and cloud interception sional scaling (NMDS) (McCune & Grace, 2002). We then assessed

F I G U R E 2 Distribution of 1,753
Atlantic Domain sites with their a priori
classification into vegetation types
(symbols). Variations in (a) temperature
seasonality (standard deviation × 100) and
(b) water deficit severity (mm) were fitted
across geographical space by generalized
additive models. Dashed lines represent
Brazilian state borders
 14724642, 2017, 8, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ddi.12581 by CAPES, Wiley Online Library on [08/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
902 | NEVES et al.

the relative importance of turnover and nestedness to floristic differ- SDMTools package (VanDerWal, Falconi, Januchowski, Shoo, & Storlie,
entiation between rain forest and each of the marginal habitats. This 2014) in the R Statistical Environment—and identified configuration
analysis was performed by first calculating Jaccard pairwise distances, metrics of landscapes (e.g., patch area, edge perimeter) for 95% of our
which range from 0 (identical in community composition) to 1 (com- sites using the vegetation map of the Brazilian Atlantic Domain (http://
pletely different in community composition). These pairwise distances mapas.sosma.org.br/). We found that the effect of habitat quality was
are then decomposed into dissimilarity due to species turnover (i.e., negligible in explaining variation in tree community composition across
only compositional changes) and dissimilarity due to differences in rain forests and marginal habitats (see Table S1 for further details).
species richness. The latter is the difference between Jaccard dis-
tance and the dissimilarity due to species turnover (Baselga, 2010).
2.4 | Conservation assessment
The ordination and the dissimilarity partitioning analyses were con-
ducted in the statistical packages vegan (Oksanen et al., 2016) and We assessed how well the floristic diversity is captured in our data-
betapart (Baselga & Orme, 2012), respectively, both in the R Statistical set by calculating the expected species accumulation curves for rain
Environment (R Development Core Team, 2015). forest and marginal habitats, using sample-­based rarefaction (Colwell
We assessed whether Scarano’s (2009) limiting factors are the key et al., 2012) with the “specaccum” function in the statistical package
environmental factors driving variation in community composition, vegan (Oksanen et al., 2016). We also explored levels of endemism for
and then explored the results visually by plotting the habitats in geo- Atlantic Domain habitats. We obtained the lists of endemic species
graphical or ordination (NMDS) space and then fitting the values of the (woody + non-woody) from Reflora (http://floradobrasil.jbrj.gov.br),
most important environmental variables via generalized additive mod- which is the most comprehensive study of the patterns of plant spe-
els (GAM) and generalized linear models (GLM), respectively. This rou- cies richness and endemism for phytogeographical Domains in east-
tine follows methods similar to those proposed by Blanchet, Legendre, ern South America. Afterwards, we conducted an assessment of the
and Borcard (2008) and Legendre, Borcard, and Roberts (2012), which conservation status of the Atlantic Domain habitats by overlaying the
comprise (1) the exclusion of 300 singletons (species found at a single distribution of our 1,753 sites on to the coverage of protected areas
site), as they commonly increase the noise in most analyses without across South America. We used conservation units from the World
contributing information (Lepš & Šmilauer, 2003); (2) the Hellinger Database on Protected Areas (IUCN & UNEP—WCMC, www.pro-
transformation of the binary presence/absence data (Legendre & tectedplanet.net) and Cadastro Nacional de Unidades de Conservação
Gallagher, 2001), which reduces the effect of widespread species; (3) (Ministério do Meio Ambiente—Brazil, www.mapas.mma.gov.br).
the independent compilation of significant spatial and environmental Species accumulation curves are provided for rain forest and marginal
variables through a forward selection method for redundancy analy- habitats as SI (Fig. S1).
sis (RDA), after first checking that the respective global models were Lastly, we used the main environmental variables emerging from
significant (Blanchet et al., 2008); (4) an additional and progressive the community turnover models to create site groups discriminating
elimination of collinear variables based on their variance inflation fac- the marginal habitats and then processed the species matrix follow-
tor (VIF) and ecological relevance, until maintaining only those with ing the procedure proposed by Tichý and Chytrý (2006) to produce
VIF <4 (Quinn & Keough, 2002); and (5) an RDA-­based partitioning of sets of diagnostic species, which are provided as supporting informa-
variation in the community composition matrix due to environmental tion (Table S2). This procedure is particularly suitable to quantify the
variables, spatial autocorrelation and their combined, statistically indis- ­fidelity of species to groups that have unequal sizes, that is, ­different
tinguishable effects. As spatial variables, we used principal coordinates numbers of sampling units, as is the case with our study. After the
of neighbour matrices (PCNMs; Borcard, Legendre, Avois-­Jacquet, & groups are equalized, a coefficient of fidelity is calculated and the
Tuomisto, 2004), which represent the spatial structure of the sampling ­significance of each diagnostic species is obtained with 999 Monte
units at multiple spatial scales without considering any environmental Carlo permutations.
variation (Borcard, Legendre, & Drapeau, 1992; Borcard et al., 2004;
Legendre et al., 2002). We tested the overall significance of the envi-
3 | RESULTS
ronmental fraction (controlled for spatial autocorrelation) by applying
ANOVA permutation tests (999 permutations) for RDA (Peres-­Neto,
3.1 | Floristic patterns
Legendre, Dray, & Borcard, 2006). The variable selection, variation
partitioning, NMDS, GLM and GAM analyses were conducted using The distribution of the sites in the ordination space yielded by NMDS
the fields (Nychka, Furrer, Paige, & Sain, 2015), spacemakeR (Dray, (Figure 3a and b) largely segregated rain forests and marginal habitats.
2010) and vegan (Oksanen et al., 2016) packages in the R Statistical The ordination placed “marginal” vegetation types at the extremes of
Environment (the variation partitioning script is available as support- the first three ordination axes. Axis 1 segregated, at negative scores,
ing information). The maps were designed using the package maptools the shoreline-­associated restinga and, at positive scores, the vegetation
(Lewin-­Koh & Bivand, 2012) in the R Statistical Environment. types associated with low-­temperature extremes of higher elevations
We also calculated patch statistics to test whether floristic dif- and latitudes further from the equator (Araucaria-­dominated forests and
ferentiation can be modulated by habitat quality (a proxy for anthro- subtropical riverine forests). Axis 2 segregated, at positive scores, veg-
pogenic effect). We used the PatchStat function—available in the etation types associated with rock outcrops (rocky cloud dwarf-­forests,
 14724642, 2017, 8, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ddi.12581 by CAPES, Wiley Online Library on [08/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
NEVES et al. | 903

F I G U R E 3 Ordination of 1,753 Atlantic Domain sites yielded by non-­metric multidimensional scaling (NMDS) of their tree species
composition with their a priori classification into vegetation types (symbols). Diagrams are provided for axes 1 × 2 (a) and 1 × 3 (b). Arrows in
each diagram represent the correlations between the most explanatory environmental variables and ordination scores. TempSeas, temperature
seasonality; DaysFrost, days of frost; Salinity, soil salinity; GrassCover, grass coverage; HyperSeas, water hyperseasonality; PrecAnn, mean
annual precipitation

F I G U R E 4 Decomposition of the pairwise floristic dissimilarity of rain forest and marginal habitat sites of the South American Atlantic
Domain (e.g., bullets in the Araucaria-­dominated triangle represent pairwise dissimilarities between each of the 193 Araucaria-­dominated
sites and all the 328 rain forest sites, i.e., 63,304 pairwise dissimilarity values). Numbers represent the mean turnover (%) and nestedness (%)
components of the Jaccard dissimilarity for each marginal habitat

rocky semideciduous dwarf-­forests and campos rupestres). Axis 1 fur- The floristic composition of marginal habitats is not simply a nested
ther segregated rock outcrop vegetation types into warmer sites (rocky subset of the more species rich rain forest. The turnover component
semideciduous dwarf-­forests and campos rupestres), at positive scores, accounts for most of the floristic dissimilarity of each marginal habi-
and colder sites (rocky cloud dwarf-­forests), at negative scores. Axis tat in relation to rain forests (Figure 4). Nestedness is higher than the
3 placed the habitat associated with seasonal drought (semideciduous turnover component in very few cases (i.e., few marginal habitat sites
forests) at intermediate scores and the habitat associated with water- are simply a subset of another rain forest site; see semideciduous for-
logged soils at positive scores (tropical riverine forests). est triangle in Figure 4). More specifically, vegetation types associated
 14724642, 2017, 8, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ddi.12581 by CAPES, Wiley Online Library on [08/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
904 | NEVES et al.

T A B L E 1 Variables selected for the

adj. R2 cum. ∆AIC F VIF
analysis of environmental controls of tree
Temperature seasonality 0.068 −508.02 128.96 3.51 community composition in the Atlantic
Grass coverage 0.174 −716.16 34.28 1.28 Domain of South America.

Salinity 0.199 −767.24 27 2.04

Water deficit severity 0.209 −787.86 22.65 3.13
Hyperseasonality 0.222 −816.58 15.42 3.82
Mean annual precipitation 0.234 −840.26 13.41 2.57
Days of frost 0.242 −856.91 8.87 1.76
Elevation 0.251 −863.48 8.52 3.83
Temperature daily range 0.251 −875.73 7.8 2.64
Cloud interception 0.257 −887 4.89 3.27
Soil fertility 0.26 −892.36 4.6 1.46
Water excess duration 0.263 −896.43 3.73 3.11
Sandiness 0.264 −897.48 3 1.74

The variables shown were selected through a forward selection method for redundancy analysis and
are ordered by the amount of explained variation in species composition across rain forest and marginal
habitats. Goodness-­of-­fit of the predictor variables was assessed through adjusted coefficients of de-
termination, Akaike information criterion (AIC), F-­values and significance tests (p < .01 in all cases). VIF,
variance inflation factor, obtained using the r-­squared value of the regression of one variable against all
other explanatory variables. adj. R2 cum. = cumulative adjusted coefficient of correlation.

with rock outcrops (including campos rupestre) have the highest frac- the Atlantic Domain, where water deficit severity and mean annual
tion of dissimilarity attributed to turnover while restinga and subtropi- precipitation, proxies for drought-­stress, explained the floristic differ-
cal riverine forest have the lowest fraction attributed to turnover. entiation of everwet vegetation types, namely rain forest, cloud forests
and rocky cloud dwarf-­forests, from campos rupestres, semideciduous
forests, rocky semideciduous dwarf-­forests and tropical riverine for-
3.2 | Variation partitioning analyses
ests (Figures 2b and 3b). At the harshest extreme of the drought-­stress
The forward selection procedure retained 13 environmental vari- gradient (Figure 3b), water-­related hyperseasonality (i.e., ranging from
ables in the model to explain the variation in tree species composition water shortage to soil waterlogging) segregates campo rupestres
(Table 1). In partitioning the variation explained by the retained en- and tropical riverine forests from semideciduous forests. These fac-
vironmental and spatial predictors, we found that the environmental tors represent the seven most explanatory environmental variables
fraction explained 27% of the variation, 5% of which was independ- (Table 1) and they accounted for a large fraction of the variation in
ent of spatial autocorrelation (p < .01). The environmental predictors community composition attributed to environmental predictors (ad-
could not account for a spatially structured variation of 12% (p < .01), justed R2 = .242; Table 1), which is nearly the same as the value for
and 61% of the variation remained unexplained (see discussion for all 13 variables retained in the variation partitioning model (adjusted
more details). R2 = .264; Table 1).
The harshest extremes of the retained environmental variables
(Table 1) do lead to distinct habitats, treated here in the context of
3.3 | Conservation assessment
“marginal” vegetation types. A north to south increase in temperature
seasonality was congruent with a latitudinal gradient in community The species accumulation curves showed a levelling off at larger
turnover, which represents the floristic differentiation of Araucaria-­ sample sizes for all vegetation types, although no curve actually
dominated forests and subtropical riverine forests (Figures 2a and reached an asymptote. Species accumulation curves levelled off less
3a) from all other vegetation types. Grass coverage, a proxy for fire in vegetation types associated with rock outcrops (including campos
frequency (see Methods), was congruent with the floristic differentia- rupestres) and in Araucaria-­dominated forest (see Fig. S1). Because
tion of the vegetation types associated with rock outcrops (including the overall floristic dissimilarity between cloud forests and rain for-
campos rupestres) from all other vegetation types (Figure 3a). Within ests was relatively low (Figure 3), we assessed the rates of endemism
the rock outcrop habitat, the frequency of frost was associated with considering these two vegetation types as “core” habitats (wet for-
the floristic differentiation of rocky cloud dwarf-­forests from the other ests in Table 2 and Figure 5). Despite the fact that wet forests have
rocky vegetation types. Soil salinity was congruent with a coast to in- twice as much protection as marginal habitats (45% and 26%, re-
land gradient in community turnover, which represents the floristic spectively; Table 2 and Figures 5, 6 and 7), almost half of all species
differentiation of restinga from all other vegetation types (Figure 3a). endemic to the Atlantic Domain are only found in marginal habitats
Another coast to inland gradient is evident in the tropical section of (Table 2).
 14724642, 2017, 8, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ddi.12581 by CAPES, Wiley Online Library on [08/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
NEVES et al. | 905

4 | DISCUSSION
Wet forests (rain forest + cloud forest) and marginal habitats of the South American Atlantic Domain ranked by their level of endemism in plant species (total endemics/total species

PA (%)

45
54
73
51
19
52
17
21
1
Both the variation partitioning and the ordination support the impor-

PA = percentage of NeoTropTree sites in protected areas (see Figures 5, 6 and 7). Lists of plant species (woody + non-woody) were obtained from the Reflora project (http://floradobrasil.jbrj.gov.br).
tance of the set of limiting conditions proposed by Scarano (2009) as
the factors controlling tree community composition of rain forests and

41
39
21
12
7
1
6
4
–
%
marginal habitats, which are treated here in the context of “marginal”
vegetation types (question 1). We also showed that these limiting fac-
endemics

3,939 tors lead to floristically distinct tree communities, thus indicating that
1,968
448
298
247
8
87
104
Total

–
the marginal habitats are not simply a nested subset of the more diverse
Atlantic Domain rain forest (question 2). In fact, marginal habitats shelter
nearly half the endemic plant species in the Atlantic Domain (question 3).
Gymnosperms

–
–
–
–
–
–
–
1
–
4.1 | Limiting factors
A north to south increase in temperature seasonality is the major factor
Pteridophyta

associated with a wide-­scale floristic differentiation between tropical

habitats and those that are mainly comprised of cold-­tolerant species
(see Figure 2a and Table 1). Interestingly, this is consistent even within
199
15
19
1
4

6
2
–

the subtropical section of the Atlantic Domain (Oliveira-­Filho et al.,

2015), where variation in community composition along the tempera-
Angiosperms

ture seasonality gradient is congruent with increasing foliage decidu-

Endemics

ousness, a trait associated with frost-­tolerance (Oliveira-­Filho et al.,

3,740
1,953
429
297
243
8
81
101

2015). A similar trend in species turnover and foliage deciduousness is

–

also found in the tropical and equatorial sections of the Atlantic Domain,
but the main driving force there is rainfall seasonality and the associ-
ated dry season (Eisenlohr & Oliveira-­Filho, 2015; Saiter et al., 2016).
9,695
4,993
2,136
2,530
3,528

1,507
2,561
900

234
Total

Contrary to our expectations, temperature seasonality showed stronger

explanatory power than the frequency of frosts, believed to be a chief
factor limiting species distribution across temperature gradients (see
Gymnosperms

Oliveira-­Filho et al., 2015; Rundel, Smith, & Meinzer, 1994; Scarano,

2009; Zanne et al., 2014). Nevertheless, within rock outcrop habitats
(Figure 3b), the occurrence of frost in rocky cloud dwarf-­forests seems
to be limiting the establishment of species from campos rupestres and
2
–
2
2
1
1
4
5
1

rocky semideciduous dwarf-­forests, suggesting that the frequency of

frosts is an important factor underpinning the distribution of marginal
Pteridophyta

habitats in the Atlantic Domain, though at smaller spatial scales.

Periods of water shortage represented by seasonal droughts are
755

165

155
57
97
38

21

61
2

indeed the chief factor driving species turnover in the tropical and
equatorial sections of the Atlantic Domain (see Figure 2b), while other
local factors may also affect water availability to plants (Pontara et al.,
Angiosperms

2016). The substrate often either favours or restricts water drainage

8,938
4,936
2,037
2,490
3,362

1,348
2,495
878

231

via landforms and soil depth and texture, while strong winds may add
All

to the water deficit stress, particular nearer to the coast, where restin-
gas occur. In this coastal marginal habitat, which was identified as one
rocky semideciduous dwarf-­forest

of the most floristically differentiated (see Figure 3a), the stress due
to water deficit is increased by a sandy substrate with high salinity,
Araucaria-­dominated forest

subtropical riverine forest

and by salt spray coming directly from the ocean (Cerqueira, 2000).
rocky cloud dwarf-­forest

tropical riverine forest

semideciduous forest

In addition, although nutrient-­poor soils prevail all over the Domain,

the edaphic conditions in restingas represent an extreme of particu-
campos rupestres

larly low soil fertility (most NTT sites of the dataset were classified as
wet forests
TABLE 2

“dystrophic” while most restingas were “hyperdystrophic”).

richness).

restinga

When assessing whether soil waterlogging leads to a floristically

distinct marginal habitat, we found that the intrusions of riverine
 14724642, 2017, 8, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ddi.12581 by CAPES, Wiley Online Library on [08/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
906 | NEVES et al.

F I G U R E 5 Conservation assessment of wet forests (rain + cloud), rocky cloud dwarf-­forest and Araucaria-­dominated forests of the South
American Atlantic Domain. Black bullets represent woody plant communities occurring within protected areas. Grey areas represent the current
network of protected areas across South America. Dashed lines represent Brazilian state borders

F I G U R E 6 Conservation assessment of campo rupestre, semideciduous forests and rocky semideciduous dwarf-­forests of the South
American Atlantic Domain. Black bullets represent woody plant communities occurring within protected areas. Grey areas represent the current
network of protected areas across South America. Dashed lines represent Brazilian state borders

forests into poorly drained soils of the Cerrado Domain showed only beds is known as one of the main sources of floristic differentiation
a weak differentiation from their neighbouring semideciduous forests among habitat types and an important driver of tree species distribu-
(see Figure 3). Kurtz, Valentin, and Scarano (2015) also found that riv- tion patterns (Wittmann et al., 2013), whereas in the Atlantic Domain,
erine habitats of the Atlantic Domain are indistinguishable as a floristic the tiny areas of riverine forest are swamped with immigration from
unit from non-­flooded habitats, and that their flora is essentially an the non-­flooded habitats. On the other hand, the intrusions of sub-
extract of the regional species pool. These trends may result from a tropical riverine forests into poorly drained soils of the Pampa Domain
particular feature of the Atlantic Domain. Unlike the Amazon Domain, seems to have a comparatively stronger floristic differentiation (see
where a wide net of rivers lead to large areas of seasonally flooded Figure 3a), but primarily associated with high temperature seasonality.
habitats, rivers in the Atlantic Domain represent a minor component For campos rupestres we were able to document fire as an im-
of the landscape. In the Amazon, seasonal flooding over wide alluvial portant factor limiting tree species distribution across the Atlantic
 14724642, 2017, 8, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ddi.12581 by CAPES, Wiley Online Library on [08/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
NEVES et al. | 907

F I G U R E 7 Conservation assessment of restinga, subtropical riverine forests and tropical riverine forests of the South American Atlantic
Domain. Black bullets represent woody plant communities occurring within protected areas. Grey areas represent the current network of
protected areas across South America. Dashed lines represent Brazilian state borders. Coastal white-­sand woodlands are called restinga in Brazil

Domain (see Figure 3a). This is consistent with previous studies show- their spatial interdigitation (e.g., in south-­eastern Brazil; Figure 2), we
ing that forest-­savanna boundaries in tropical savannas are driven believe it is more parsimonious to attribute the positive spatial auto-
by fire, though generally in interaction with other factors (Archibald correlation, a proxy of distance decay in community similarity (Nekola
et al., 2013; Dantas, Batalha, & Pausas, 2013; Hoffmann et al., 2012). & White, 1999), to niche-­based controls (e.g., unmeasured spatially
Within the Atlantic Domain, however, fire frequency is low relative to structured variables describing environmental conditions, natural en-
the surrounding savanna formations (see detailed maps in Archibald emies and competition). Regarding the large fraction of unexplained
et al., 2013) and has therefore been neglected in previous studies. variation, it may suggest that ecological drift (cf. Hubbell, 2001) is driv-
Nevertheless, here we show that fire is actually an important com- ing stochastic rearrangements of species distribution ranges through
ponent shaping macroscale patterns of floristic variation across the time. However, a high proportion of unexplained variation, ranging
Atlantic Domain and thus deserves further attention. The congruence from 40% to 80% (e.g., Legendre et al., 2009; Neves et al., 2015; re-
between floristic turnover and grass coverage, a proxy for fire fre- viewed by Soininen, 2014), is a common outcome in studies of floristic
quency, across rocky semideciduous dwarf-­forests and campos rupes- composition over similar spatial scales, and could also be attributed
tres (Figure 3a) indicates that fire plays a key role in determining the to statistical noise (ter Braak, 1986; Guisan, Weiss, & Weiss, 1999) or
mosaic of rock outcrop habitats in the Atlantic Domain. Rocky semide- unmeasured non-­spatially structured environmental conditions.
ciduous dwarf-­forests seem to represent a transition between rain
forests and campos rupestres (see Figure 3a), which is likely to be medi-
4.3 | Conservation implications
ated by fire history and local factors contributing to either increase or
decrease flammability, particularly topography and soil depth. Here we showed the uneven distribution of protected areas across
the Atlantic Domain with wet forests having twice as much protec-
tion. Marginal habitats receive considerably lower protection, despite
4.2 | Spatial structure and unexplained variation
harbouring almost half of the 7,099 species endemic to the Atlantic
While the relevance of the environmental fraction in controlling com- Domain. These 3,160 endemic species are not found anywhere else
munity turnover was straightforward to interpret, the variation that in the world, including in the rain forests of the Atlantic Domain. This
either remained unexplained or was attributed to spatial structure demonstrates that different marginal habitats, characterized by envi-
independent of the measured environmental factors (61% and 12%, ronmental harshness, underpin the patterns of high species richness
respectively) deserves further attention. Rain forests and marginal across the Atlantic Domain as a whole. Therefore, we emphasize that
habitats are often geographically segregated (Figure 2), suggesting these marginal habitats need better consideration by conservation-
that there may be a role for spatially structured dispersal limitation ists and biodiversity scientists, based on their (1) high level of en-
and historical biogeography in driving some of the observed floristic demism; (2) lower level of protection; and (3) less data (see species
differentiation. However, given the clear floristic segregation of rock accumulation curves of vegetation types associated with rock out-
outcrop dwarf-­forests from semideciduous and rain forests, despite crops in Fig. S1).
 14724642, 2017, 8, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ddi.12581 by CAPES, Wiley Online Library on [08/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
908 | NEVES et al.

Blanchet, F. G., Legendre, P., & Borcard, D. (2008). Forward selection of

4.4 | Concluding remarks explanatory variables. Ecology, 89, 2623–2632.
Borcard, D., Legendre, P., Avois-Jacquet, C., & Tuomisto, H. (2004).
The distribution of the Atlantic Forest marginal habitats is asso-
Dissecting the spatial structure of ecological data at multiple scales.
ciated with low-­temperature extremes (i.e., ranging from winter Ecology, 85, 1826–1832.
frosts to summer maxima higher than 40°C), soil salinity, drought-­ Borcard, D., Legendre, P., & Drapeau, P. (1992). Partialling out the spatial
stress and soil waterlogging. Additionally, grass coverage, a proxy component of ecological variation. Ecology, 73, 1045–1055.
ter Braak, C. J. F. (1986). Canonical correspondence analysis: A new eigen-
for flammability and a previously unappreciated environmental
vector technique for multivariate direct gradient analysis. Ecology, 67,
factor in the Atlantic Domain, is amongst the principal factors ex- 1167–1179.
plaining the patterns of tree species distribution. For conservation Cerqueira, R. (2000). Biogeografia das restingas. In F. A. Esteves & L. D.
purposes, the restinga is strikingly distinct both floristically and Lacerda (Eds.), Ecologia de restingas e lagoas costeiras (pp. 65–75).
Macaé, Brasil: NUPEM/UFRJ.
environmentally (see Figures 3a and b), suggesting the need for
Colwell, R. K., Chao, A., Gotelli, N. J., Lin, S.-Y., Mao, C. X., Chazdon, R. L.,
further investigation. If restingas are indeed a distinct phytogeo- & Longino, J. T. (2012). Models and estimators linking individual-­based
graphical region, instead of an extension of rain forests into saline and sample-­based rarefaction, extrapolation and comparison of assem-
white-­sand environments, they may be much more threatened than blages. Journal of Plant Ecology, 5, 3–21.
Dantas, V. L., Batalha, M. A., & Pausas, J. G. (2013). Fire drives functional
assumed based upon classifications that places these two habitats
thresholds on the savanna–forest transition. Ecology, 94, 2454–2463.
together. Restinga has suffered massive fragmentation due to high
Dray, S. (2010). SpacemakeR: Spatial modelling. R package version 473
human occupation in coastal areas and a rapidly developing tour- 0.0–5/r101. Retrieved from: http://R-Forge.R-project.org/projects/
ism industry. sedar/
Eisenlohr, P. V., & Oliveira-Filho, A. T. (2015). Revisiting patterns of tree
species composition and their driving forces in the Atlantic forests of
ACKNOWLE DG E MEN TS southeastern Brazil. Biotropica, 47, 689–701.
Galindo-Leal, C., Jacobsen, T. R., Langhammer, P. F., & Olivieri, S. (2003).
D.M.N., K.G.D. and R.T.P. were supported by the National State of the hotspots: The dynamics of biodiversity loss. In C. Galindo-
Environmental Research Council (grant NE/I028122/1). AOF and Leal C & I. G. de Câmara (Eds.), The Atlantic Forest of South America:
Biodiversity status, threats, and outlook (pp. 12–23). Washington: Center
MLB were supported by the Conselho Nacional de Desenvolvimento
for Applied Biodiversity Science and Island Press.
Científico e Tecnológico—Brazil (CNPq) (grants 301644/88-­
8 and Guisan, A., Weiss, S. B., & Weiss, A. D. (1999). GLM versus CCA spatial
151002/2014-­
2, respectively). ASMV and SNM were supported modeling of plant species distribution. Plant Ecology, 143, 107–122.
by the Fundação de Amparo à Pesquisa de Minas Gerais – Brazil Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G., & Jarvis, A. (2005).
Very high resolution interpolated climate surfaces for global land areas.
(FAPEMIG). PLSM thanks the Coordenação de Aperfeiçoamento de
International Journal of Climatology, 25, 1965–1978.
Pessoal de Nível Superior—Brazil (CAPES) for supporting a full PhD Hoffmann, W. A., Geiger, E. L., Gotsch, S. G., Rossatto, D. R., Silva, L. C. R.,
at the University of Edinburgh under the Science Without Borders Lau, O. L., … Franco, A. C. (2012). Ecological thresholds at the savanna-­
Programme (grant BEX 13197-­13-­4). forest boundary: How plant traits, resources and fire govern the distri-
bution of tropical biomes. Ecology Letters, 15, 759–768.
Hubbell, S. P. (2001). The unified neutral theory of biodiversity and biogeogra-
AUTHOR CONTRI B UTI O N S phy. Princeton, NJ: Princeton University Press.
IBGE (1993). Mapa de Vegetação do Brasil. Rio de Janeiro: Fundação
A.O.F compiled the database and conceived the idea; D.M.N. and Instituto Brasileiro de Geografia e Estatística, Ministério da Agricultura.
Joly, C. A., Metzger, J. P., & Tabarelli, M. (2014). Experiences from the
K.G.D. designed the manuscript; D.M.N. analysed the data; D.M.N. and
Brazilian Atlantic forest: Ecological findings and conservation initia-
A.O.F. led the writing with substantial input from K.G.D. and R.T.P. All
tives. New Phytologist, 204, 459–473.
authors commented on the manuscript and approved the final version. Kurtz, B. C., Valentin, J. L., & Scarano, F. R. (2015). Are the Neotropical
swamp forests a distinguishable forest type? Patterns from Southeast
and Southern Brazil. Edinburgh Journal of Botany, 72, 191–208.
DATA ACCE SSI B I LI TY Legendre, P., Borcard, D., & Roberts, D. W. (2012). Variation partitioning
involving orthogonal spatial eigenfunction submodels. Ecology, 93,
Additional accessibility data are provided as supporting information.
1234–1240.
Legendre, P., Dale, M. R. T., Fortin, M. J., Gurevitch, J., Hohn, M., & Myers,
D. (2002). The consequences of spatial structure for the design and
REFERENCES
analysis of ecological field surveys. Ecography, 25, 601–615.
Ab’Sáber, A. N. (2003). Os domínios de natureza no Brasil: Potencialidades Legendre, P., & Gallagher, E. D. (2001). Ecological meaningful transforma-
paisagísticas. São Paulo: Ateliê Editorial. tions for ordination of species data. Oecologia, 129, 271–280.
Archibald, S., Lehmann, C. E. R., Goméz-Dans, J. L., & Bradstock, R. A. Legendre, P., Mi, X., Ren, H., Ma, K., Yu, M., Sun, I., & He, F. (2009).
(2013). Defining pyromes and global syndromes of fire regimes. Partitioning beta diversity in a subtropical broad-­leaved forest of China.
Proceedings of the National Academy of Sciences of the United States of Ecology, 90, 663–674.
America, 110, 6442–6447. Lehmann, C. E. R., Anderson, T. M., Sankaran, M., Higgins, S. I., Archibald,
Baselga, A. (2010). Partitioning the turnover and nestedness components S., Hoffmann, W. A., … Bond, W. J. (2014). Savanna vegetation-­fire-­
of beta diversity. Global Ecology and Biogeography, 19, 134–143. climate relationships differ among continents. Science, 343, 548–552.
Baselga, A., & Orme, D.l. (2012). Betapart: An R package for the study of Lepš, J., & Šmilauer, J. P. (2003). Multivariate analysis of ecological data using
beta diversity. Methods in Ecology and Evolution, 3, 808–812. CANOCO. Cambridge, UK: Cambridge University Press.
 14724642, 2017, 8, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/ddi.12581 by CAPES, Wiley Online Library on [08/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
NEVES et al. | 909

Lewin-Koh, N. J., & Bivand, R. (2012). Maptools: Tools for reading and han- Santos, H. G., Jacomine, P. K. T., Anjos, L. H. C., Oliveira, V. A., Lumbreras,
dling spatial objects. R package version 0.8–17/r238. Retrieved from: J. F., Coelho, M. R., … Oliveira, J. B. (2013). Sistema brasileiro de classifi-
http://R-Forge.R project.org/projects/maptools/ cação de solos, 3rd ed. Brasília: Embrapa.
McCune, B., & Grace, J. B. (2002). Analysis of ecological communities. Scarano, F. R. (2009). Plant communities at the periphery of the Atlantic
Gleneden Beach, Oregon: MjM Software Design. rain forest: Rare-­species bias and its risks for conservation. Biological
Menini-Neto, L., Furtado, S. G., Zappi, D., Oliveira-Filho, A. T., & Forzza, Conservation, 142, 1201–1208.
R. C. (2016). Biogeography of epiphytic Angiosperms in the Brazilian Soininen, J. (2014). A quantitative analysis of species sorting across organ-
Atlantic forest, a world biodiversity hotspot. Brazilian Journal of Botany, isms and ecosystems. Ecology, 95, 3284–3292.
39, 261–273. Tabarelli, M., Pinto, L. P., Silva, J. M. C., Hirota, M., & Bedê, L. (2005).
Myers, N., Mittermeier, R. A., Mittermeier, C. G., Fonseca, G. A. B., & Kent, Challenges and opportunities for biodiversity conservation in the
J. (2000). Biodiversity hotspots for conservation priorities. Nature, 403, Brazilian Atlantic Forest. Conservation Biology, 19, 695–700.
853–858. Tabarelli, M., Silva, J. M. C., & Gascon, C. (2004). Forest fragmentation, syn-
Nekola, J. C., & White, P. S. (1999). The distance decay of similarity in bioge- ergisms and the impoverishment of neotropical forests. Biodiversity and
ography and ecology. Journal of Biogeography, 26, 867–878. Conservation, 13, 1419–1425.
Neves, D. M., Dexter, K. G., Pennington, R. T., Bueno, M. L., & Oliveira-Filho, Tichý, L., & Chytrý, M. (2006). Statistical determination of diagnostic spe-
A. T. (2015). Environmental and historical controls of floristic composi- cies for site groups of unequal size. Journal of Vegetation Science, 17,
tion across the South American Dry Diagonal. Journal of Biogeography, 809–818.
42, 1566–1576. VanDerWal, J., Falconi, L., Januchowski, S., Shoo, L., & Storlie, C. (2014).
Nychka, D., Furrer, R., Paige, J., & Sain, S. (2015). Fields: Tools for spatial SDMTools: Species distribution modelling tools: Tools for processing
data. R package version 8.3-5. Retrieved from: http://CRAN.R-project. data associated with species distribution modelling exercises. R pack-
org/package=fields age version 1.1-221. Retrieved from: https://CRAN.R-project.org/
Oksanen, J., Blanchet, F. G., Kindt, R., Legendre, P., Minchin, P. R., O’Hara, R. B., package=SDMTools
… Wagner, H. (2016). Vegan: Community ecology package. R package ver- Walter, H. (1985). Vegetation of the earth and ecological systems of the
sion 2.0–3. Retrieved from: http://CRAN.R-project.org/package=vegan geo-biosphere, 3rd ed. Berlin: Springer-Verlag.
Oliveira-Filho, A. T. (2015). Um Sistema de classificação fisionômico-ecológica Wittmann, F., Householder, E., Piedade, M. T. F., Assis, R. L., Schöngart, J.,
da vegetação Neotropical. In P. V. Eisenlohr, J. M. Felfili, M. M. R. F. Melo, Parolin, P., & Junk, W. J. (2013). Habitat specificity, endemism and the
L. A. Andrade & J. A. A. Meira-Neto (Eds.), Fitossociologia no Brasil: Métodos neotropical distribution of Amazonian white-­water floodplain trees.
e estudos de casos, Vol. 2 (pp. 452–473). Viçosa, Brazil: Editora UFV. Ecography, 36, 690–707.
Oliveira-Filho, A. T., Budke, J. C., Jarenkow, J. A., Eisenlohr, P. V., & Neves, Zanne, A. E., Tank, D. C., Cornwell, W. K., Eastman, J. M., Smith, S. A.,
D. R. M. (2015). Delving into the variations in tree species composition FitzJohn, R. G., … Beaulieu, J. M. (2014). Three keys to the radiation of
and richness across South American subtropical Atlantic and Pampean angiosperms into freezing environments. Nature, 506, 89–92.
forests. Journal of Plant Ecology, 8, 242–260. Zomer, R. J., Bossio, D. A., Trabucco, A., Yuanjie, L., Gupta, D. C., & Singh,
Oliveira-Filho, A. T., Jarenkow, J. A., & Rodal, M. J. N. (2006). Floristic rela- V. P. (2007). Trees and water: Smallholder agroforestry on irrigated lands
tionships of seasonally dry forests of eastern South America based on in northern India. IWMI Research Report 122. Colombo, Sri Lanka:
tree species distribution patterns. In R. T. Pennington, J. A. Ratter, & G. International Water Management Institute.
P. Lewis (Eds.), Neotropical savannas and dry forests: Plant diversity, bioge- Zomer, R. J., Trabucco, A., Bossio, D. A., van Straaten, O., & Verchot, L. V.
ography and conservation (pp. 151–184). Boca Raton, USA: CRC Press. (2008). Climate change mitigation: A spatial analysis of global land suit-
Oliveira-Filho, A. T., & Fontes, M. A. L. (2000). Patterns of floristic differen- ability for clean development mechanism afforestation and reforesta-
tiation among Atlantic Forests in Southeastern Brazil and the influence tion. Agriculture Ecosystems and Environment, 126, 67–80.
of climate. Biotropica, 32, 793–810.
Peres-Neto, P. R., Legendre, P., Dray, S., & Borcard, D. (2006). Variation
partitioning of species data matrices: Estimation and comparisons of B I O S KETC H
fractions. Ecology, 87, 2614–2625.
Pontara, V., Bueno, M. L., Garcia, L. E., Oliveira-Filho, A. T., Pennington, R. Danilo M. Neves is a postdoctoral research fellow at the University
T., Burslem, D. F. R. P., & Lemos-Filho, J. P. (2016). Fine-­scale variation of Arizona. He is interested in the evolutionary dimension of
in topography and seasonality determine radial growth of an endan-
community ecology, with an emphasis on historical biogeography of
gered tree in Brazilian Atlantic forest. Plant and Soil, 403, 115–128.
Prado, D. E., & Gibbs, P. E. (1993). Patterns of species distribution in the terrestrial biomes.
dry seasonal forests of South America. Annals of the Missouri Botanical
Garden, 80, 902–927.
Quinn, G. P., & Keough, M. J. (2002). Experimental design and data analysis
for biologists. Cambridge, UK: Cambridge University Press. S U P P O RT I NG I NFO R M AT I O N
R Core Team (2015). R: A language and environment for statistical comput-
Additional Supporting Information may be found online in the
ing. Version 3.1.0. Vienna: R Foundation for Statistical Computing.
Retrieved from: http://www.Rproject.org/ supporting information tab for this article.
REFLORA (2017). Lista de espécies da flora do Brasil. Rio de Janeiro, Brazil:
Jardim Botânico do Rio de Janeiro. Retrieved from: http://floradobrasil.
jbrj.gov.br/ (accessed 19.01.17). How to cite this article: Neves DM, Dexter KG, Pennington RT,
Rundel, P. W., Smith, A. P., & Meinzer, F. C. (1994). Tropical alpine environments: et al. Dissecting a biodiversity hotspot: The importance of
Plant form and function. Cambridge, UK: Cambridge University Press. environmentally marginal habitats in the Atlantic Forest
Saiter, F. Z., Eisenlohr, P. V., Barbosa, M. R., Thomas, W. W., & Oliveira-Filho,
Domain of South America. Diversity Distrib. 2017;23:898–909.
A. T. (2016). From evergreen to deciduous tropical forests: How energy-­
water balance, temperature, and space influence the tree species com- https://doi.org/10.1111/ddi.12581
position in a high diversity region. Plant Ecology & Diversity, 9, 45–54.