wileyonlinelibrary.com/journal/ddi Diversity and Distributions. 2017;23:898–909.
© 2017 John Wiley & Sons Ltd
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
1Department of Ecology and Evolutionary
Biology, University of Arizona, Tucson, AZ,
2School of GeoSciences, The University of
Edinburgh, Edinburgh, UK
3Royal Botanic Garden Edinburgh, Edinburgh,
4Instituto Estadual de Florestas, Minas Gerais,
5Laboratório de Ecologia e Evolução
de Plantas, Departamento de Biologia
Vegetal, Universidade Federal de Viçosa,
Viçosa, Minas Gerais, Brazil
6Laboratório de Ecologia, Universidade do
Estado de Mato Grosso, Alta Floresta, Brazil
7Departamento de Ciências
Florestais, Universidade Federal de Lavras,
8Programa de Pós-Graduação em Biologia
Vegetal, Universidade Federal de Minas
Gerais, Belo Horizonte, Brazil
9Instituto Federal do Espírito Santo, Santa
Teresa, Espírito Santo, Brazil
Danilo M. Neves, Department of Ecology and
Evolutionary Biology, University of Arizona,
Tucson, AZ, USA.
Emails: firstname.lastname@example.org and
Coordenação de Aperfeiçoamento de Pessoal
de Nível Superior, Grant/Award Number:
BEX 13197-13-4; Fundação de Amparo à
Pesquisa do Estado de Minas Gerais; Natural
Environment Research Council, Grant/Award
Number: NE/I028122/1; Conselho Nacional
de Desenvolvimento Científico e Tecnológico,
Grant/Award Number: 151002/2014-2 and
Editor: Kenneth Feeley
Aim: We aimed to assess the contribution of marginal habitats to the tree species rich-
ness of the Mata Atlântica (Atlantic Forest) biodiversity hotspot. In addition, we aimed
to determine which environmental factors drive the occurrence and distribution of
these marginal habitats.
Location: The whole extension of the South American Atlantic Forest Domain plus
forest intrusions into the neighbouring Cerrado and Pampa Domains, which comprises
rain forests (“core” habitat) and five marginal habitats, namely high elevation forests,
rock outcrop dwarf- forests, riverine forests, semideciduous forests and restinga
(coastal white- sand woodlands).
Methods: We compiled a dataset containing 366,875 occurrence records of 4,431
tree species from 1,753 site- checklists, which were a priori classified into 10 main
vegetation types. We then performed ordination analyses of the species- by- site ma-
trix to assess the floristic consistency of this classification. In order to assess the rela-
tive contribution of environmental predictors to the community turnover, we
produced models using 26 climate and substrate- related variables as environmental
Results: Ordination diagrams supported the floristic segregation of vegetation types,
with those considered as marginal habitats placed at the extremes of ordination axes.
These marginal habitats are associated with the harshest extremes of five limiting
factors: temperature seasonality (high elevation and subtropical riverine forests),
flammability (rock outcrop dwarf- forests), high salinity (restinga), water deficit sever-
ity (semideciduous forests) and waterlogged soils (tropical riverine forests).
Importantly, 45% of all species endemic to the Atlantic Domain only occur in marginal
Main conclusions: Our results showed the key role of the poorly protected marginal
habitats in contributing to the high species richness of the Atlantic Domain. Various
types of environmental harshness operate as environmental filters determining the
distribution of the Atlantic Domain habitats. Our findings also stressed the importance
of fire, a previously neglected environmental factor.
NEVES Et al.
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
in the Atlantic Domain, become a minor component of the landscape
in these neighbouring Domains, and they are only found in riverine or
high elevation forest enclaves.
Environmental restriction to the establishment of the rain forest
habitat is certainly operating at the boundaries of the Atlantic Domain.
In a seminal paper, Scarano (2009) proposed a list of five key fac-
tors limiting the occurrence and distribution of rain forest species in
the Atlantic Domain, which at its harshest extremes give rise to dis-
tinct habitats (one for each factor), referred to as marginal habitats.
Therefore, the rain forest is placed by Scarano (2009) as the “core” ex-
pression of the Atlantic Domain, where deep shade plays the chief role
as a limiting factor for competing plants. The five marginal habitats are
high elevation forests, rock outcrop dwarf- forests, riverine forests, sea-
sonally dry forests and restinga (coastal white- sand woodlands). Most
of these marginal habitats have a relatively high density of trees and
can be considered forests, albeit not as well developed structurally as
rain forests. High elevation forests are primarily associated with frost,
with secondary limitation imposed by drought (leeward rain- shadow)
and high- light intensity. Cloud forests and Araucaria- dominated for-
ests are the main vegetation types of highlands in the Atlantic Domain.
Rock outcrop dwarf- forests, found at lower elevations (and even at the
seashore), are primarily limited by the paucity, or even lack, of soil and
related poor water retention. Meanwhile, riverine forests are associ-
ated with waterlogging on lowland plains and riverbeds. Seasonally
dry forests (either deciduous or semideciduous) replace rain forests
where seasonal rainfall regimes bring regular periods of drought.
Finally, environmental harshness for restinga is primarily associated
with salinity, with secondary limitations imposed by drought and low
fertility in mineral nutrients (Scarano, 2009) (Figure 1).
Within limited areas, some studies have confirmed the lead-
ing role of Scarano’s limiting factors as distribution filters for plants.
These studies addressed tree species composition for particular
sectors of the Atlantic Domain, such as the south- east (Oliveira- Filho
& Fontes, 2000; Eisenlohr & Oliveira- Filho, 2015), the subtropical
South (Oliveira- Filho, Budke, Jarenkow, Eisenlohr, & Neves, 2015)
and the highly biodiverse central region in eastern Bahia state, north-
eastern Brazil (Saiter, Eisenlohr, Barbosa, Thomas, & Oliveira- Filho,
2016). However, the whole of the Atlantic Domain has only been in-
vestigated for epiphytic angiosperms (Menini- Neto, Furtado, Zappi,
Oliveira- Filho, & Forzza, 2016). Also, the Atlantic Domain is affected
by fire in much of its distribution (Archibald, Lehmann, Goméz- Dans,
& Bradstock, 2013), though to a lesser extent than in surrounding
Domains, such as in central (Cerrado woody savannas) and southern
Brazil (Pampa prairies). Nevertheless, the potential effect of fire in
limiting plant species distribution across the Atlantic Domain is yet to
be investigated. Here we bring together a novel and comprehensive
dataset assembled on the composition of tree communities across the
whole Domain (c. 2,000 community surveys across core and marginal
habitats, with >1,000 sites representing surveys not used in the afore-
mentioned studies), combined with environmental data, focusing on
testing Scarano’s proposed limiting variables as well as factors that
were neglected in previous studies (e.g., fire).
Besides the importance for community ecology, understanding
the degree to which limiting factors drive community differentiation
is inherently relevant for conservation. The Atlantic Domain houses
c.18,000 plant species (REFLORA, 2017), but the current high levels of
fragmentation and the continuous habitat loss throughout the Domain
have raised several concerns in the scientific community (Galindo- Leal,
Jacobsen, Langhammer, & Olivieri, 2003; Joly, Metzger, & Tabarelli,
FIGURE1 Environmental variables (arrows) hypothesized in
Scarano (2009) as key factors limiting plant species distribution
across the Atlantic Domain of South America. The harshest extremes
give rise to distinct vegetation types, referred to as marginal habitats.
Coastal white- sand woodlands are called restinga in Brazil
campo rupestre, climate, conservation assessment, flammability, rain forests, restinga, stress
gradients, variation partitioning
NEVES Et al.
2014; Tabarelli, Pinto, Silva, Hirota, & Bedê, 2005; Tabarelli, Silva,
& Gascon, 2004). Therefore, we believe the time is ripe for studies
aiming to test the overall importance of environmental conditions in
controlling the occurrence and distribution of plant species across the
whole extent of the Atlantic Domain and, more importantly, across
both its core and marginal habitats.
We addressed the following questions: (1) Are the patterns of
tree species distribution across the Atlantic Domain, and its intrusions
into neighbouring Domains, limited by factors associated with water
deficit (via both soil depth and dry season), water excess (via water-
logging), frosts (via low temperature) and soil salinity? If previously
unrecognized environmental conditions are the main factors explain-
ing the patterns of tree species distribution, Scarano’s (2009) limiting
factors should account for a small proportion of the variation in com-
munity composition explained by environmental factors; (2) are these
limiting factors leading to floristically distinct marginal habitats? If the
community composition of the marginal habitats is simply a nested
subset of the more diverse Atlantic Domain rain forest, species turn-
over should account for a small fraction of the dissimilarity between
rain forest and marginal habitats; and (3) what is the contribution of
these marginal habitats to the overall high species richness of the
2 | METHODS
2.1 | Study area
The Atlantic Forest, designated as one of the five phytogeographi-
cal “Domains” of Brazil (Ab’Sáber, 2003; IBGE, 1993), occurs primar-
ily along the Atlantic coast and is bordered by the Pampa Domain
(woody prairies) of southern Brazil and by the “dry diagonal,” a cor-
ridor that includes three other phytogeographical Domains: Caatinga
(largely semi- arid thorn woodlands) of north- eastern Brazil, Cerrado
(largely woody savannas) of central Brazil, and Chaco (largely semi-
arid thorn woodlands) of Paraguay–Argentina–Bolivia (IBGE, 1993,
Prado & Gibbs 1993, Neves, Dexter, Pennington, Bueno, & Oliveira-
Filho, 2015). The South American Atlantic Forest Domain (hereafter
Atlantic Domain) has a history of controversies over its geographical
circumscription and associated terminology. The controversy may be
summarized by three main concepts of Atlantic Domain habitats: the
sensu stricto, sensu lato and sensu latissimo concepts (Oliveira- Filho,
Jarenkow, & Rodal, 2006). The first, and most restrictive concept, in-
cludes only the tracts of rain forests that occur as a narrow band along
the coast (<100 km wide and up to 2500 m elevation) and stretches
all through the Domain, though with two main interruptions, the São
Francisco Gap and Campos dos Goytacazes Gap. The former is a semi-
arid nucleus at the mouth of the São Francisco River (~10°30′S), and
the latter is a seasonally dry region extending from southern Espírito
Santo to northern Rio de Janeiro (RJ) States, with its driest extreme at
Cabo Frio/RJ (~22°50′S).
The sensu lato concept of Atlantic Domain habitats, which is
currently prevalent, includes other habitats adjacent to rain forests,
such as the much more extensive semideciduous forests that cover
increasingly larger areas towards the south and become wide enough
to reach eastern Paraguay and north- eastern Argentina. Araucaria-
dominated forests are also a very important component of the sensu
lato concept, followed by coastal woodlands on white- sand substrates
(termed restingas) and three highland dwarf- forests: rocky cloud
dwarf- forests, rocky semideciduous dwarf- forests and rocky highland
savannas (termed campos rupestres).
The sensu latissimo concept of Atlantic Domain habitats proposed
by Oliveira- Filho et al. (2006) surpasses the geographical limits of the
Atlantic Domain to include riverine and deciduous forest tracts oc-
curring in the neighbouring Domains as a secondary component of
the landscape, though with a typically Atlantic Domain flora. In the
present contribution, we adopt this concept because it allows a more
complete inclusion of marginal habitats. However, deciduous forests
found in the Cerrado and Pampa Domains, one of the forest types in
the sensu lato concept (IBGE, 1993), were not included in this contri-
bution because previous studies (e.g., Eisenlohr & Oliveira- Filho, 2015;
Oliveira- Filho et al., 2006) have demonstrated that their flora is dis-
tinct and more closely related to that of semi- arid woodlands (e.g., in
the Caatinga Domain).
2.2 | Dataset
We extracted the dataset from the NeoTropTree (NTT) database
(http://prof.icb.ufmg.br/treeatlan), which consists of tree species
checklists (trees defined here as freely standing woody plants >3 m
in height) compiled for geo- referenced sites, extending from southern
Florida (U.S.A.) and Mexico to Patagonia. NTT currently holds 5,126
sites/checklists, 14,878 woody plant species and 920,129 occurrence
records. A site/checklist in NTT is defined by a single habitat, following
the classification system proposed by Oliveira- Filho (2015), contained
in a circular area with a 10- km diameter. Therefore, where two or more
habitats co- occur in one 10- km area, there may be two geographically
overlapping sites in the NTT database, each for a distinct habitat.
The data were originally compiled from an extensive survey of
published and unpublished (e.g., PhD theses) literature, particularly
those on woody plant community surveys and floristic inventories.
Moreover, new species occurrence records obtained from both major
herbaria and taxonomic monographs have been added to the check-
lists when they were collected within the 10- km diameter of the
original NTT site and within the same habitat. All species and their
occurrence records were checked regarding current taxonomic and
geographical circumscriptions, as defined by the team of specialists
responsible for the online projects Flora do Brasil and Flora del Conosur
(available at http://floradobrasil.jbrj.gov.br/ and http://www.darwin.
edu.ar/, respectively). NTT does not, therefore, include occurrence
records with doubtful identification, location or habitat, nor sites with
an indication of high anthropogenic disturbance. The latter is assessed
by taking into account the information available in the studies that
comprise the checklists, and by direct observation of site surface on
Google Earth©. It also excludes checklists with low species richness
(<20 species), because this is often due to low sampling/collecting ef-
forts, which results in poor descriptive power.
NEVES Et al.
This study used a subset of tree inventories from the NTT data-
base, consisting of 328 rain forest sites and 1,425 sites representing
the limiting environmental factors and marginal habitats proposed by
Scarano (2009), namely seasonally dry (663 semideciduous forests),
high elevation (193 Araucaria- dominated forests and 61 cloud forests),
rock outcrops (49 rocky cloud dwarf- forests, 31 rocky semideciduous
dwarf- forests and 41 campos rupestres), high salinity (181 restingas—
with only forests and dwarf- forests of the mosaic included) and water-
logged soils (133 tropical riverine forests and 73 subtropical riverine
forests). Note that marginal habitats associated with seasonal drought
and high salinity are represented by one vegetation type, whereas
high elevation, rock outcrops and waterlogged soils are represented
by more than one vegetation type. The final species matrix contained
presence/absence data for 4,431 tree species across 1,753 sites, with
a total of 366,875 presences (see Figure 2a and b).
The NTT database also included 26 environmental variables for all
its sites, derived from multiple sources (at a 30 arc- second resolution;
detailed below). The resolution used in this study was particularly ap-
propriate (1 km2) because all sites are more than 1 km distant from each
other (only 124 of 1,753 sites are less than 5 km distant from another
site, and the mean distance between all sites is >1,000 km). Elevation at
the NTT site centre was used as an integrative environmental variable.
Mean annual temperature, mean daily temperature range, isothermality,
temperature seasonality, maximum temperature of the warmest month,
minimum temperature of the coldest month, temperature annual range,
mean annual precipitation, precipitation of the wettest month, precipi-
tation of the driest month and precipitation seasonality were obtained
from WorldClim 1.4 data layers (Hijmans, Cameron, Parra, Jones, &
Jarvis, 2005). WorldClim monthly temperatures and precipitation were
also interpolated to obtain values for 5- day intervals by applying sinusoi-
dal functions centred at day 15 of each month. These functions yielded
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
variables: duration (days) and severity (days) of both the water deficit
and water excess periods. Frost frequency (days) and cloud interception
(mm) were obtained from interpolating known values as response vari-
ables (data obtained from 135 and 57 Brazilian Meteorological Stations
measuring frost frequency and cloud interception, respectively) with
elevation, latitude and the WorldClim layers as predicting variables.
Potential evapotranspiration (mm) and the aridity index (annual precip-
itation/potential evapotranspiration) were obtained from Zomer et al.
(2007), Zomer, Trabucco, Bossio, van Straaten, and Verchot (2008).
Surface rockiness (% exposed rock), soil coarseness (% sand), soil
fertility (% base saturation) and soil salinity (ds/m) were obtained from
the Harmonized World Soil Database v 1.2 (available at http://www.
nized-world-soil-database-v12/en/) and ranked afterwards by mid-
class percentage. The use of classes was adopted to add robustness
to the data because of the high local soil heterogeneity that makes
raw figures unrealistic. Soil drainage classes were obtained following
EMBRAPA’s protocol (Santos et al., 2013), which combines soil type,
texture and depth with landforms. Soil drainage classes, mean annual
precipitation (Hijmans et al., 2005) and the aforementioned indices of
water deficit and excess were also combined to produce a hyperseason-
ality index. Grass coverage (%) was used as a proxy of fire return interval
(i.e., frequency). Previous studies give support to grass coverage as a
good proxy of fire frequency (Archibald et al., 2013; Hoffmann et al.,
2012; Lehmann et al., 2014), although further quantification of fire re-
gime is clearly needed (c.f. Archibald et al., 2013). Grass coverage was
obtained by direct observation of site surface on Google Earth© images
in five 100 × 100 m areas, one at the central coordinates of the NTT site
and four at 2.5 km away from it and towards the NE, SW, NW and SE.
Further details of NTT history, products and protocols can be
found at http://prof.icb.ufmg.br/treeatlan.
2.3 | Analyses of community turnover
We first explored the patterns of floristic differentiation between rain
forest and marginal habitats by performing non- metric multidimen-
sional scaling (NMDS) (McCune & Grace, 2002). We then assessed
FIGURE2 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
NEVES Et al.
the relative importance of turnover and nestedness to floristic differ-
entiation between rain forest and each of the marginal habitats. This
analysis was performed by first calculating Jaccard pairwise distances,
which range from 0 (identical in community composition) to 1 (com-
pletely different in community composition). These pairwise distances
are then decomposed into dissimilarity due to species turnover (i.e.,
only compositional changes) and dissimilarity due to differences in
species richness. The latter is the difference between Jaccard dis-
tance and the dissimilarity due to species turnover (Baselga, 2010).
The ordination and the dissimilarity partitioning analyses were con-
ducted in the statistical packages vegan (Oksanen et al., 2016) and
betapart (Baselga & Orme, 2012), respectively, both in the R Statistical
Environment (R Development Core Team, 2015).
We assessed whether Scarano’s (2009) limiting factors are the key
environmental factors driving variation in community composition,
and then explored the results visually by plotting the habitats in geo-
graphical or ordination (NMDS) space and then fitting the values of the
most important environmental variables via generalized additive mod-
els (GAM) and generalized linear models (GLM), respectively. This rou-
tine follows methods similar to those proposed by Blanchet, Legendre,
and Borcard (2008) and Legendre, Borcard, and Roberts (2012), which
comprise (1) the exclusion of 300 singletons (species found at a single
site), as they commonly increase the noise in most analyses without
contributing information (Lepš & Šmilauer, 2003); (2) the Hellinger
transformation of the binary presence/absence data (Legendre &
Gallagher, 2001), which reduces the effect of widespread species; (3)
the independent compilation of significant spatial and environmental
variables through a forward selection method for redundancy analy-
sis (RDA), after first checking that the respective global models were
significant (Blanchet et al., 2008); (4) an additional and progressive
elimination of collinear variables based on their variance inflation fac-
tor (VIF) and ecological relevance, until maintaining only those with
VIF <4 (Quinn & Keough, 2002); and (5) an RDA- based partitioning of
variation in the community composition matrix due to environmental
variables, spatial autocorrelation and their combined, statistically indis-
tinguishable effects. As spatial variables, we used principal coordinates
of neighbour matrices (PCNMs; Borcard, Legendre, Avois- Jacquet, &
Tuomisto, 2004), which represent the spatial structure of the sampling
units at multiple spatial scales without considering any environmental
variation (Borcard, Legendre, & Drapeau, 1992; Borcard et al., 2004;
Legendre et al., 2002). We tested the overall significance of the envi-
ronmental fraction (controlled for spatial autocorrelation) by applying
ANOVA permutation tests (999 permutations) for RDA (Peres- Neto,
Legendre, Dray, & Borcard, 2006). The variable selection, variation
partitioning, NMDS, GLM and GAM analyses were conducted using
the fields (Nychka, Furrer, Paige, & Sain, 2015), spacemakeR (Dray,
2010) and vegan (Oksanen et al., 2016) packages in the R Statistical
Environment (the variation partitioning script is available as support-
ing information). The maps were designed using the package maptools
(Lewin- Koh & Bivand, 2012) in the R Statistical Environment.
We also calculated patch statistics to test whether floristic dif-
ferentiation can be modulated by habitat quality (a proxy for anthro-
pogenic effect). We used the PatchStat function—available in the
SDMTools package (VanDerWal, Falconi, Januchowski, Shoo, & Storlie,
2014) in the R Statistical Environment—and identified configuration
metrics of landscapes (e.g., patch area, edge perimeter) for 95% of our
sites using the vegetation map of the Brazilian Atlantic Domain (http://
mapas.sosma.org.br/). We found that the effect of habitat quality was
negligible in explaining variation in tree community composition across
rain forests and marginal habitats (see Table S1 for further details).
2.4 | Conservation assessment
We assessed how well the floristic diversity is captured in our data-
set by calculating the expected species accumulation curves for rain
forest and marginal habitats, using sample- based rarefaction (Colwell
et al., 2012) with the “specaccum” function in the statistical package
vegan (Oksanen et al., 2016). We also explored levels of endemism for
Atlantic Domain habitats. We obtained the lists of endemic species
(woody + non-woody) from Reflora (http://floradobrasil.jbrj.gov.br),
which is the most comprehensive study of the patterns of plant spe-
cies richness and endemism for phytogeographical Domains in east-
ern South America. Afterwards, we conducted an assessment of the
conservation status of the Atlantic Domain habitats by overlaying the
distribution of our 1,753 sites on to the coverage of protected areas
across South America. We used conservation units from the World
Database on Protected Areas (IUCN & UNEP—WCMC, www.pro-
tectedplanet.net) and Cadastro Nacional de Unidades de Conservação
(Ministério do Meio Ambiente—Brazil, www.mapas.mma.gov.br).
Species accumulation curves are provided for rain forest and marginal
habitats as SI (Fig. S1).
Lastly, we used the main environmental variables emerging from
the community turnover models to create site groups discriminating
the marginal habitats and then processed the species matrix follow-
ing the procedure proposed by Tichý and Chytrý (2006) to produce
sets of diagnostic species, which are provided as supporting informa-
tion (Table S2). This procedure is particularly suitable to quantify the
fidelity of species to groups that have unequal sizes, that is, different
numbers of sampling units, as is the case with our study. After the
groups are equalized, a coefficient of fidelity is calculated and the
significance of each diagnostic species is obtained with 999 Monte
3 | RESULTS
3.1 | Floristic patterns
The distribution of the sites in the ordination space yielded by NMDS
(Figure 3a and b) largely segregated rain forests and marginal habitats.
The ordination placed “marginal” vegetation types at the extremes of
the first three ordination axes. Axis 1 segregated, at negative scores,
the shoreline- associated restinga and, at positive scores, the vegetation
types associated with low- temperature extremes of higher elevations
and latitudes further from the equator (Araucaria- dominated forests and
subtropical riverine forests). Axis 2 segregated, at positive scores, veg-
etation types associated with rock outcrops (rocky cloud dwarf- forests,
NEVES Et al.
rocky semideciduous dwarf- forests and campos rupestres). Axis 1 fur-
ther segregated rock outcrop vegetation types into warmer sites (rocky
semideciduous dwarf- forests and campos rupestres), at positive scores,
and colder sites (rocky cloud dwarf- forests), at negative scores. Axis
3 placed the habitat associated with seasonal drought (semideciduous
forests) at intermediate scores and the habitat associated with water-
logged soils at positive scores (tropical riverine forests).
The floristic composition of marginal habitats is not simply a nested
subset of the more species rich rain forest. The turnover component
accounts for most of the floristic dissimilarity of each marginal habi-
tat in relation to rain forests (Figure 4). Nestedness is higher than the
turnover component in very few cases (i.e., few marginal habitat sites
are simply a subset of another rain forest site; see semideciduous for-
est triangle in Figure 4). More specifically, vegetation types associated
FIGURE4 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
FIGURE3 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
NEVES Et al.
with rock outcrops (including campos rupestre) have the highest frac-
tion of dissimilarity attributed to turnover while restinga and subtropi-
cal riverine forest have the lowest fraction attributed to turnover.
3.2 | Variation partitioning analyses
The forward selection procedure retained 13 environmental vari-
ables in the model to explain the variation in tree species composition
(Table 1). In partitioning the variation explained by the retained en-
vironmental and spatial predictors, we found that the environmental
fraction explained 27% of the variation, 5% of which was independ-
ent of spatial autocorrelation (p < .01). The environmental predictors
could not account for a spatially structured variation of 12% (p < .01),
and 61% of the variation remained unexplained (see discussion for
The harshest extremes of the retained environmental variables
(Table 1) do lead to distinct habitats, treated here in the context of
“marginal” vegetation types. A north to south increase in temperature
seasonality was congruent with a latitudinal gradient in community
turnover, which represents the floristic differentiation of Araucaria-
dominated forests and subtropical riverine forests (Figures 2a and
3a) from all other vegetation types. Grass coverage, a proxy for fire
frequency (see Methods), was congruent with the floristic differentia-
tion of the vegetation types associated with rock outcrops (including
campos rupestres) from all other vegetation types (Figure 3a). Within
the rock outcrop habitat, the frequency of frost was associated with
the floristic differentiation of rocky cloud dwarf- forests from the other
rocky vegetation types. Soil salinity was congruent with a coast to in-
land gradient in community turnover, which represents the floristic
differentiation of restinga from all other vegetation types (Figure 3a).
Another coast to inland gradient is evident in the tropical section of
the Atlantic Domain, where water deficit severity and mean annual
precipitation, proxies for drought- stress, explained the floristic differ-
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-
ests (Figures 2b and 3b). At the harshest extreme of the drought- stress
gradient (Figure 3b), water- related hyperseasonality (i.e., ranging from
water shortage to soil waterlogging) segregates campo rupestres
and tropical riverine forests from semideciduous forests. These fac-
tors represent the seven most explanatory environmental variables
(Table 1) and they accounted for a large fraction of the variation in
community composition attributed to environmental predictors (ad-
justed R2 = .242; Table 1), which is nearly the same as the value for
all 13 variables retained in the variation partitioning model (adjusted
R2 = .264; Table 1).
3.3 | Conservation assessment
The species accumulation curves showed a levelling off at larger
sample sizes for all vegetation types, although no curve actually
reached an asymptote. Species accumulation curves levelled off less
in vegetation types associated with rock outcrops (including campos
rupestres) and in Araucaria- dominated forest (see Fig. S1). Because
the overall floristic dissimilarity between cloud forests and rain for-
ests was relatively low (Figure 3), we assessed the rates of endemism
considering these two vegetation types as “core” habitats (wet for-
ests in Table 2 and Figure 5). Despite the fact that wet forests have
twice as much protection as marginal habitats (45% and 26%, re-
spectively; Table 2 and Figures 5, 6 and 7), almost half of all species
endemic to the Atlantic Domain are only found in marginal habitats
adj. R2 cum. ∆AIC FVIF
Temperature seasonality 0.068 −508.02 128.96 3.51
Grass coverage 0.174 −716.16 34.28 1.28
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.
TABLE1 Variables selected for the
analysis of environmental controls of tree
community composition in the Atlantic
Domain of South America.
NEVES Et al.
4 | DISCUSSION
Both the variation partitioning and the ordination support the impor-
tance of the set of limiting conditions proposed by Scarano (2009) as
the factors controlling tree community composition of rain forests and
marginal habitats, which are treated here in the context of “marginal”
vegetation types (question 1). We also showed that these limiting fac-
tors lead to floristically distinct tree communities, thus indicating that
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).
4.1 | Limiting factors
A north to south increase in temperature seasonality is the major factor
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
the subtropical section of the Atlantic Domain (Oliveira- Filho et al.,
2015), where variation in community composition along the tempera-
ture seasonality gradient is congruent with increasing foliage decidu-
ousness, a trait associated with frost- tolerance (Oliveira- Filho et al.,
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).
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
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
rocky semideciduous dwarf- forests, suggesting that the frequency of
frosts is an important factor underpinning the distribution of marginal
habitats in the Atlantic Domain, though at smaller spatial scales.
Periods of water shortage represented by seasonal droughts are
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.,
2016). The substrate often either favours or restricts water drainage
via landforms and soil depth and texture, while strong winds may add
to the water deficit stress, particular nearer to the coast, where restin-
gas occur. In this coastal marginal habitat, which was identified as one
of the most floristically differentiated (see Figure 3a), the stress due
to water deficit is increased by a sandy substrate with high salinity,
and by salt spray coming directly from the ocean (Cerqueira, 2000).
In addition, although nutrient- poor soils prevail all over the Domain,
the edaphic conditions in restingas represent an extreme of particu-
larly low soil fertility (most NTT sites of the dataset were classified as
“dystrophic” while most restingas were “hyperdystrophic”).
When assessing whether soil waterlogging leads to a floristically
distinct marginal habitat, we found that the intrusions of riverine
TABLE2 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
Angiosperms Pteridophyta Gymnosperms Total Angiosperms Pteridophyta Gymnosperms
endemics % PA (%)
wet forests 8,938 755 2 9,695 3,740 199 – 3,939 41 45
campos rupestres 4,936 57 – 4,993 1,953 15 – 1,968 39 54
rocky cloud dwarf- forest 2,037 97 2 2,136 429 19 – 448 21 73
restinga 2,490 38 2 2,530 297 1 – 298 12 51
semideciduous forest 3,362 165 1 3,528 243 4 – 247 7 19
rocky semideciduous dwarf- forest 878 21 1 900 8 – – 8 1 52
Araucaria- dominated forest 1,348 155 4 1,507 81 6 – 87 6 17
tropical riverine forest 2,495 61 5 2,561 101 2 1 104 4 21
subtropical riverine forest 231 2 1 234 – – – – – 1
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).
NEVES Et al.
forests into poorly drained soils of the Cerrado Domain showed only
a weak differentiation from their neighbouring semideciduous forests
(see Figure 3). Kurtz, Valentin, and Scarano (2015) also found that riv-
erine habitats of the Atlantic Domain are indistinguishable as a floristic
unit from non- flooded habitats, and that their flora is essentially an
extract of the regional species pool. These trends may result from a
particular feature of the Atlantic Domain. Unlike the Amazon Domain,
where a wide net of rivers lead to large areas of seasonally flooded
habitats, rivers in the Atlantic Domain represent a minor component
of the landscape. In the Amazon, seasonal flooding over wide alluvial
beds is known as one of the main sources of floristic differentiation
among habitat types and an important driver of tree species distribu-
tion patterns (Wittmann et al., 2013), whereas in the Atlantic Domain,
the tiny areas of riverine forest are swamped with immigration from
the non- flooded habitats. On the other hand, the intrusions of sub-
tropical riverine forests into poorly drained soils of the Pampa Domain
seems to have a comparatively stronger floristic differentiation (see
Figure 3a), but primarily associated with high temperature seasonality.
For campos rupestres we were able to document fire as an im-
portant factor limiting tree species distribution across the Atlantic
FIGURE5 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
FIGURE6 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
NEVES Et al.
Domain (see Figure 3a). This is consistent with previous studies show-
ing that forest- savanna boundaries in tropical savannas are driven
by fire, though generally in interaction with other factors (Archibald
et al., 2013; Dantas, Batalha, & Pausas, 2013; Hoffmann et al., 2012).
Within the Atlantic Domain, however, fire frequency is low relative to
the surrounding savanna formations (see detailed maps in Archibald
et al., 2013) and has therefore been neglected in previous studies.
Nevertheless, here we show that fire is actually an important com-
ponent shaping macroscale patterns of floristic variation across the
Atlantic Domain and thus deserves further attention. The congruence
between floristic turnover and grass coverage, a proxy for fire fre-
quency, across rocky semideciduous dwarf- forests and campos rupes-
tres (Figure 3a) indicates that fire plays a key role in determining the
mosaic of rock outcrop habitats in the Atlantic Domain. Rocky semide-
ciduous dwarf- forests seem to represent a transition between rain
forests and campos rupestres (see Figure 3a), which is likely to be medi-
ated by fire history and local factors contributing to either increase or
decrease flammability, particularly topography and soil depth.
4.2 | Spatial structure and unexplained variation
While the relevance of the environmental fraction in controlling com-
munity turnover was straightforward to interpret, the variation that
either remained unexplained or was attributed to spatial structure
independent of the measured environmental factors (61% and 12%,
respectively) deserves further attention. Rain forests and marginal
habitats are often geographically segregated (Figure 2), suggesting
that there may be a role for spatially structured dispersal limitation
and historical biogeography in driving some of the observed floristic
differentiation. However, given the clear floristic segregation of rock
outcrop dwarf- forests from semideciduous and rain forests, despite
their spatial interdigitation (e.g., in south- eastern Brazil; Figure 2), we
believe it is more parsimonious to attribute the positive spatial auto-
correlation, a proxy of distance decay in community similarity (Nekola
& White, 1999), to niche- based controls (e.g., unmeasured spatially
structured variables describing environmental conditions, natural en-
emies and competition). Regarding the large fraction of unexplained
variation, it may suggest that ecological drift (cf. Hubbell, 2001) is driv-
ing stochastic rearrangements of species distribution ranges through
time. However, a high proportion of unexplained variation, ranging
from 40% to 80% (e.g., Legendre et al., 2009; Neves et al., 2015; re-
viewed by Soininen, 2014), is a common outcome in studies of floristic
composition over similar spatial scales, and could also be attributed
to statistical noise (ter Braak, 1986; Guisan, Weiss, & Weiss, 1999) or
unmeasured non- spatially structured environmental conditions.
4.3 | Conservation implications
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
harbouring almost half of the 7,099 species endemic to the Atlantic
Domain. These 3,160 endemic species are not found anywhere else
in the world, including in the rain forests of the Atlantic Domain. This
demonstrates that different marginal habitats, characterized by envi-
ronmental harshness, underpin the patterns of high species richness
across the Atlantic Domain as a whole. Therefore, we emphasize that
these marginal habitats need better consideration by conservation-
ists and biodiversity scientists, based on their (1) high level of en-
demism; (2) lower level of protection; and (3) less data (see species
accumulation curves of vegetation types associated with rock out-
crops in Fig. S1).
FIGURE7 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
NEVES Et al.
4.4 | Concluding remarks
The distribution of the Atlantic Forest marginal habitats is asso-
ciated with low- temperature extremes (i.e., ranging from winter
frosts to summer maxima higher than 40°C), soil salinity, drought-
stress and soil waterlogging. Additionally, grass coverage, a proxy
for flammability and a previously unappreciated environmental
factor in the Atlantic Domain, is amongst the principal factors ex-
plaining the patterns of tree species distribution. For conservation
purposes, the restinga is strikingly distinct both floristically and
environmentally (see Figures 3a and b), suggesting the need for
further investigation. If restingas are indeed a distinct phytogeo-
graphical region, instead of an extension of rain forests into saline
white- sand environments, they may be much more threatened than
assumed based upon classifications that places these two habitats
together. Restinga has suffered massive fragmentation due to high
human occupation in coastal areas and a rapidly developing tour-
D.M.N., K.G.D. and R.T.P. were supported by the National
Environmental Research Council (grant NE/I028122/1). AOF and
MLB were supported by the Conselho Nacional de Desenvolvimento
Científico e Tecnológico—Brazil (CNPq) (grants 301644/88- 8 and
151002/2014- 2, respectively). ASMV and SNM were supported
by the Fundação de Amparo à Pesquisa de Minas Gerais – Brazil
(FAPEMIG). PLSM thanks the Coordenação de Aperfeiçoamento de
Pessoal de Nível Superior—Brazil (CAPES) for supporting a full PhD
at the University of Edinburgh under the Science Without Borders
Programme (grant BEX 13197- 13- 4).
A.O.F compiled the database and conceived the idea; D.M.N. and
K.G.D. designed the manuscript; D.M.N. analysed the data; D.M.N. and
A.O.F. led the writing with substantial input from K.G.D. and R.T.P. All
authors commented on the manuscript and approved the final version.
Additional accessibility data are provided as supporting information.
Ab’Sáber, A. N. (2003). Os domínios de natureza no Brasil: Potencialidades
paisagísticas. São Paulo: Ateliê Editorial.
Archibald, S., Lehmann, C. E. R., Goméz-Dans, J. L., & Bradstock, R. A.
(2013). Defining pyromes and global syndromes of fire regimes.
Proceedings of the National Academy of Sciences of the United States of
America, 110, 6442–6447.
Baselga, A. (2010). Partitioning the turnover and nestedness components
of beta diversity. Global Ecology and Biogeography, 19, 134–143.
Baselga, A., & Orme, D.l. (2012). Betapart: An R package for the study of
beta diversity. Methods in Ecology and Evolution, 3, 808–812.
Blanchet, F. G., Legendre, P., & Borcard, D. (2008). Forward selection of
explanatory variables. Ecology, 89, 2623–2632.
Borcard, D., Legendre, P., Avois-Jacquet, C., & Tuomisto, H. (2004).
Dissecting the spatial structure of ecological data at multiple scales.
Ecology, 85, 1826–1832.
Borcard, D., Legendre, P., & Drapeau, P. (1992). Partialling out the spatial
component of ecological variation. Ecology, 73, 1045–1055.
ter Braak, C. J. F. (1986). Canonical correspondence analysis: A new eigen-
vector technique for multivariate direct gradient analysis. Ecology, 67,
Cerqueira, R. (2000). Biogeografia das restingas. In F. A. Esteves & L. D.
Lacerda (Eds.), Ecologia de restingas e lagoas costeiras (pp. 65–75).
Macaé, Brasil: NUPEM/UFRJ.
Colwell, R. K., Chao, A., Gotelli, N. J., Lin, S.-Y., Mao, C. X., Chazdon, R. L.,
& Longino, J. T. (2012). Models and estimators linking individual- based
and sample- based rarefaction, extrapolation and comparison of assem-
blages. Journal of Plant Ecology, 5, 3–21.
Dantas, V. L., Batalha, M. A., & Pausas, J. G. (2013). Fire drives functional
thresholds on the savanna–forest transition. Ecology, 94, 2454–2463.
Dray, S. (2010). SpacemakeR: Spatial modelling. R package version 473
0.0–5/r101. Retrieved from: http://R-Forge.R-project.org/projects/
Eisenlohr, P. V., & Oliveira-Filho, A. T. (2015). Revisiting patterns of tree
species composition and their driving forces in the Atlantic forests of
southeastern Brazil. Biotropica, 47, 689–701.
Galindo-Leal, C., Jacobsen, T. R., Langhammer, P. F., & Olivieri, S. (2003).
State of the hotspots: The dynamics of biodiversity loss. In C. Galindo-
Leal C & I. G. de Câmara (Eds.), The Atlantic Forest of South America:
Biodiversity status, threats, and outlook (pp. 12–23). Washington: Center
for Applied Biodiversity Science and Island Press.
Guisan, A., Weiss, S. B., & Weiss, A. D. (1999). GLM versus CCA spatial
modeling of plant species distribution. Plant Ecology, 143, 107–122.
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.
International Journal of Climatology, 25, 1965–1978.
Hoffmann, W. A., Geiger, E. L., Gotsch, S. G., Rossatto, D. R., Silva, L. C. R.,
Lau, O. L., … Franco, A. C. (2012). Ecological thresholds at the savanna-
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-
phy. Princeton, NJ: Princeton University Press.
IBGE (1993). Mapa de Vegetação do Brasil. Rio de Janeiro: Fundação
Instituto Brasileiro de Geografia e Estatística, Ministério da Agricultura.
Joly, C. A., Metzger, J. P., & Tabarelli, M. (2014). Experiences from the
Brazilian Atlantic forest: Ecological findings and conservation initia-
tives. New Phytologist, 204, 459–473.
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.
Legendre, P., Borcard, D., & Roberts, D. W. (2012). Variation partitioning
involving orthogonal spatial eigenfunction submodels. Ecology, 93,
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
analysis of ecological field surveys. Ecography, 25, 601–615.
Legendre, P., & Gallagher, E. D. (2001). Ecological meaningful transforma-
tions for ordination of species data. Oecologia, 129, 271–280.
Legendre, P., Mi, X., Ren, H., Ma, K., Yu, M., Sun, I., & He, F. (2009).
Partitioning beta diversity in a subtropical broad- leaved forest of China.
Ecology, 90, 663–674.
Lehmann, C. E. R., Anderson, T. M., Sankaran, M., Higgins, S. I., Archibald,
S., Hoffmann, W. A., … Bond, W. J. (2014). Savanna vegetation- fire-
climate relationships differ among continents. Science, 343, 548–552.
Lepš, J., & Šmilauer, J. P. (2003). Multivariate analysis of ecological data using
CANOCO. Cambridge, UK: Cambridge University Press.
NEVES Et al.
Lewin-Koh, N. J., & Bivand, R. (2012). Maptools: Tools for reading and han-
dling spatial objects. R package version 0.8–17/r238. Retrieved from:
McCune, B., & Grace, J. B. (2002). Analysis of ecological communities.
Gleneden Beach, Oregon: MjM Software Design.
Menini-Neto, L., Furtado, S. G., Zappi, D., Oliveira-Filho, A. T., & Forzza,
R. C. (2016). Biogeography of epiphytic Angiosperms in the Brazilian
Atlantic forest, a world biodiversity hotspot. Brazilian Journal of Botany,
Myers, N., Mittermeier, R. A., Mittermeier, C. G., Fonseca, G. A. B., & Kent,
J. (2000). Biodiversity hotspots for conservation priorities. Nature, 403,
Nekola, J. C., & White, P. S. (1999). The distance decay of similarity in bioge-
ography and ecology. Journal of Biogeography, 26, 867–878.
Neves, D. M., Dexter, K. G., Pennington, R. T., Bueno, M. L., & Oliveira-Filho,
A. T. (2015). Environmental and historical controls of floristic composi-
tion across the South American Dry Diagonal. Journal of Biogeography,
Nychka, D., Furrer, R., Paige, J., & Sain, S. (2015). Fields: Tools for spatial
data. R package version 8.3-5. Retrieved from: http://CRAN.R-project.
Oksanen, J., Blanchet, F. G., Kindt, R., Legendre, P., Minchin, P. R., O’Hara, R. B.,
… Wagner, H. (2016). Vegan: Community ecology package. R package ver-
sion 2.0–3. Retrieved from: http://CRAN.R-project.org/package=vegan
Oliveira-Filho, A. T. (2015). Um Sistema de classificação fisionômico-ecológica
da vegetação Neotropical. In P. V. Eisenlohr, J. M. Felfili, M. M. R. F. Melo,
L. A. Andrade & J. A. A. Meira-Neto (Eds.), Fitossociologia no Brasil: Métodos
e estudos de casos, Vol. 2 (pp. 452–473). Viçosa, Brazil: Editora UFV.
Oliveira-Filho, A. T., Budke, J. C., Jarenkow, J. A., Eisenlohr, P. V., & Neves,
D. R. M. (2015). Delving into the variations in tree species composition
and richness across South American subtropical Atlantic and Pampean
forests. Journal of Plant Ecology, 8, 242–260.
Oliveira-Filho, A. T., Jarenkow, J. A., & Rodal, M. J. N. (2006). Floristic rela-
tionships of seasonally dry forests of eastern South America based on
tree species distribution patterns. In R. T. Pennington, J. A. Ratter, & G.
P. Lewis (Eds.), Neotropical savannas and dry forests: Plant diversity, bioge-
ography and conservation (pp. 151–184). Boca Raton, USA: CRC Press.
Oliveira-Filho, A. T., & Fontes, M. A. L. (2000). Patterns of floristic differen-
tiation among Atlantic Forests in Southeastern Brazil and the influence
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
fractions. Ecology, 87, 2614–2625.
Pontara, V., Bueno, M. L., Garcia, L. E., Oliveira-Filho, A. T., Pennington, R.
T., Burslem, D. F. R. P., & Lemos-Filho, J. P. (2016). Fine- scale variation
in topography and seasonality determine radial growth of an endan-
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
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.
R Core Team (2015). R: A language and environment for statistical comput-
ing. Version 3.1.0. Vienna: R Foundation for Statistical Computing.
Retrieved from: http://www.Rproject.org/
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).
Rundel, P. W., Smith, A. P., & Meinzer, F. C. (1994). Tropical alpine environments:
Plant form and function. Cambridge, UK: Cambridge University Press.
Saiter, F. Z., Eisenlohr, P. V., Barbosa, M. R., Thomas, W. W., & Oliveira-Filho,
A. T. (2016). From evergreen to deciduous tropical forests: How energy-
water balance, temperature, and space influence the tree species com-
position in a high diversity region. Plant Ecology & Diversity, 9, 45–54.
Santos, H. G., Jacomine, P. K. T., Anjos, L. H. C., Oliveira, V. A., Lumbreras,
J. F., Coelho, M. R., … Oliveira, J. B. (2013). Sistema brasileiro de classifi-
cação de solos, 3rd ed. Brasília: Embrapa.
Scarano, F. R. (2009). Plant communities at the periphery of the Atlantic
rain forest: Rare- species bias and its risks for conservation. Biological
Conservation, 142, 1201–1208.
Soininen, J. (2014). A quantitative analysis of species sorting across organ-
isms and ecosystems. Ecology, 95, 3284–3292.
Tabarelli, M., Pinto, L. P., Silva, J. M. C., Hirota, M., & Bedê, L. (2005).
Challenges and opportunities for biodiversity conservation in the
Brazilian Atlantic Forest. Conservation Biology, 19, 695–700.
Tabarelli, M., Silva, J. M. C., & Gascon, C. (2004). Forest fragmentation, syn-
ergisms and the impoverishment of neotropical forests. Biodiversity and
Conservation, 13, 1419–1425.
Tichý, L., & Chytrý, M. (2006). Statistical determination of diagnostic spe-
cies for site groups of unequal size. Journal of Vegetation Science, 17,
VanDerWal, J., Falconi, L., Januchowski, S., Shoo, L., & Storlie, C. (2014).
SDMTools: Species distribution modelling tools: Tools for processing
data associated with species distribution modelling exercises. R pack-
age version 1.1-221. Retrieved from: https://CRAN.R-project.org/
Walter, H. (1985). Vegetation of the earth and ecological systems of the
geo-biosphere, 3rd ed. Berlin: Springer-Verlag.
Wittmann, F., Householder, E., Piedade, M. T. F., Assis, R. L., Schöngart, J.,
Parolin, P., & Junk, W. J. (2013). Habitat specificity, endemism and the
neotropical distribution of Amazonian white- water floodplain trees.
Ecography, 36, 690–707.
Zanne, A. E., Tank, D. C., Cornwell, W. K., Eastman, J. M., Smith, S. A.,
FitzJohn, R. G., … Beaulieu, J. M. (2014). Three keys to the radiation of
angiosperms into freezing environments. Nature, 506, 89–92.
Zomer, R. J., Bossio, D. A., Trabucco, A., Yuanjie, L., Gupta, D. C., & Singh,
V. P. (2007). Trees and water: Smallholder agroforestry on irrigated lands
in northern India. IWMI Research Report 122. Colombo, Sri Lanka:
International Water Management Institute.
Zomer, R. J., Trabucco, A., Bossio, D. A., van Straaten, O., & Verchot, L. V.
(2008). Climate change mitigation: A spatial analysis of global land suit-
ability for clean development mechanism afforestation and reforesta-
tion. Agriculture Ecosystems and Environment, 126, 67–80.
Danilo M. Neves is a postdoctoral research fellow at the University
of Arizona. He is interested in the evolutionary dimension of
community ecology, with an emphasis on historical biogeography of
Additional Supporting Information may be found online in the
supporting information tab for this article.
How to cite this article: Neves DM, Dexter KG, Pennington RT,
et al. Dissecting a biodiversity hotspot: The importance of
environmentally marginal habitats in the Atlantic Forest
Domain of South America. Diversity Distrib. 2017;23:898–909.