Infl uence of environmental variation on Atlantic Forest tree-shrub-layer phytogeography in southeast Brazil

(Infl uence of environmental variation on Atlantic Forest tree-shrub-layer phytogeography in southeast Brazil). This work assessed data from 32 forest sites in Rio de Janeiro and São Paulo, Brazil, using multivariate analysis to answer the question: Are there fl oristic patterns of the Atlantic Forest tree-shrub layer related to the Serra do Mar and the width of coastal plains in the states of Rio de Janeiro and São Paulo? Three multivariate analyses were performed to investigate the relationship between the tree-shrub fl ora and environmental variation in these 32 study areas. Our analyses demonstrated the infl uence of geo-climatic variation on fl oristic differentiation of tree and shrub species in Atlantic Forest regions generating groups of areas based on similar biotic and abiotic characteristics. These groups support the existence of fl oristic patterns within the states of São Paulo and Rio de Janeiro and refl ect tree-shrub species substitution between the study areas as a consequence of annual rainfall, altitude, and mean annual temperature variation linked to a change in the position of the Serra do Mar and an increase in coastal plain width. Preferential species were cited for each group and should be considered in restoration and conservation programs for the phytoecological regions represented by the groups.


Introduction
Ecosystem fragmentation is one of the major causes of the loss of biodiversity on our planet.The Atlantic forest is the most threatened biome in Brazil, its degradation having been initiated in the 16 th century and accelerated with expansion of agriculture and fi xation of Europeans and Africans along the Brazilian coast (SOS Atlantic Forest & INPE 2002).Today this forest is one of the planet's 35 biodiversity hotspots, with high species richness and high levels of endemism, although it suffers from rapid reduction of intact habitats (Myers et al. 2000).There has been signifi cant recent interest in studies that assess the consequences of forest fragmentation on biodiversity conservation, as many forests have been reduced to disconnected areas (Schellas & Greenberg 1997, Viana et al. 1997).These studies have stimulated initiatives for the conservation of remnant areas, as in the case of the Atlantic forest that now occupies only eight percent of its original range (Galindo-Leal & Câmara 2003).
A number of floristic studies in remnant forests on the southeastern Brazilian coast have demonstrated the heterogeneity of the biome as a whole, with greatest similarity between regions of equal forest formation (Peixoto et al. 1995, Sanchez et al. 1999, Moreno et al. 2003, Peixoto et al. 2004, Gomes et al. 2005, Guedes-Bruni et al. 2006b).Studies employing multivariate analyses to facilitate identifi cation of ecological patterns in fl oristic data (Kent & Coker 1992) have concentrated on phytogeographic relationships and geo-climatic influences in the southeastern part of the country.These studies have revealed important information about the Atlantic forest in that region, such as the fl oristic differences between ombrophilous and seasonal forests in relation to rainfall; greater similarities between ombrophilous and seasonal forests than with other forest formations; differentiation between ombrophilous and seasonal forests that can be attributed to altitude and temperature; a "gap" in the ombrophilous forest in northern Rio de Janeiro state due to this region's seasonal climate, which cannot be defi ned as a proper disjunction in distribution of tree species; and the existence of three fl oristic blocks in Rio de Janeiro state based on tree species of the Fabaceae family (Oliveira-Filho & Fontes 2000, Oliveira-Filho et al. 2005, H.C.Lima, unpublished data).Given the value of this kind of information for decision makers and agents involved in planning conservation programs in southeastern Brazil, there is a real need for a better understanding of how the environment affects phytogeographic relationships between Atlantic forests remnants in Rio de Janeiro and São Paulo states.
Many of the remnant areas of Atlantic forest in Rio de Janeiro and São Paulo are located in the Serra do Mar and Mantiqueira mountain ranges, which are aligned subparallel to each other (Almeida & Carneiro 1998).The Serra do Mar mountain range crosses these two states in a northeasterly direction, stretching from the southern coast of São Paulo state to northern coastal Rio de Janeiro.In São Paulo the mountain range generally remains close to the coastline and forms a natural barrier capturing orographic rains that fall on the narrow coastal plain and mountain slopes facing the ocean (Barbosa 2007).At Itacuruçá (Rio de Janeiro), the Serra do Mar mountain range advances into the interior of the state, receiving different names as it stretches to northern Rio de Janeiro.The inland orientation of Serra do Mar allows occurrence of a wider coastal plain in Rio de Janeiro than in São Paulo, which is then gradually replaced by tablelands in northern Rio de Janeiro.Changes in the orientation of Serra do Mar mountain range and widening of the coastal plain are attributed to the Guanabara Graben, which was formed by the erosion and retreat of the Atlantic Plateau scarps during the Paleocene era (Almeida & Carneiro 1998).
Considering that the fl oristic composition of ombrophilous and seasonal forests within the Atlantic forest biome are related to climate variation (Oliveira-Filho & Fontes 2000, Oliveira-Filho et al. 2005) and that environmental variation can infl uence climate (Leuschner 2000), the present study aims to answer the question: Are there fl oristic patterns of the Atlantic Forest tree-shrub layer related to Serra do Mar and widening of coastal plains in the states of Rio de Janeiro and São Paulo?In order to address to this question we tested the prediction that such fl oristic patterns do not exist, thus allowing us to accept or reject our null hypothesis.

Material and methods
Atlantic forest data -32 fl oristic studies from the states of Rio de Janeiro and São Paulo were selected (Fig. 1) to form two databases.The fi rst consists of a binary matrix (presence/absence) of the tree and shrub species in the 32 areas.The second consists of environmental and geographic information on altitude (a), average annual rainfall (b), average annual temperature (c), average annual temperature variation (d), distance from the ocean (e), latitude -UTM (f), and longitude -UTM (g) of these sites.When these geo-environmental variables were not specifi ed in the original publication, we consulted the authors of the papers, the National Meteorological Institute (2006) or DNMet (1992).The fi rst database was fully revised to eliminate synonyms and errors in the presence/absence of species that could infl uence the results.In its fi nal form, the matrix contained 1590 species.Multivariate analysis -canonical correspondence analysis (CCA) was performed (ter Braak 1995) using Canoco for Windows Version 4.5 software program (ter Braak & Smilauer 2002) to investigate the relationship between the tree fl ora (at the taxonomic level of species) and geo-environmental variables.After preliminary analysis of the data, variation in average annual temperatures (d) and distance from the ocean (e) were discarded from further analysis due to a weak correlation with the four axes ("intra-set" correlation < 0.4) since they poorly explain variation within the data.We opted to maintain species that occurred in at least one of the 32 areas in the database, as their removal greatly reduced the eigenvalues of the axes.However, species considered discrepant with the main cloud of points in the CCA were eliminated during the analysis (Kent & Coker 1992) so that the fi nal principal matrix was composed of 1412 species.A Monte Carlo permutation test was performed to assess the signifi cance of the correlations found by CCA.
We also performed a cluster analysis, using the Jaccard similarity coeffi cient, through unweighted pair-group method (UPGMA) or group average (McCune & Grace 2002), in order to verify the existence of groups Figure 1.Map indicating the location of the 32 sites analyzed in this study.
based on fl oristic similarities among the areas analyzed, using MVSP 3.1 -MultiVariate Statistical Package software program (Kovach 2004).TWINSPAN analysis with PC-ORD 4.10 program (McCune & Mefford 1999) allowed us to identify key species among the groups formed by the cluster analysis (Kent & Coker 1992).

Results and discussion
The Atlantic forest sites examined by multivariate analysis are listed in Table 1, and the results of the CCA analyses are shown in Tables 2 and 3.The eigenvalues encountered for the fi rst two axes of the CCA were relatively high (> 0.4), indicating changes in species composition along the gradients synthesized in these ordination axes (ter Braak 1995).The cumulative percentage of variance explained by the axes was low (17.6%),indicating a considerable amount of unexplained "noise".This type of situation is quite common for fl oristic data and does not diminish the signifi cance of the species-environment relationship (Oliveira-Filho & Fontes 2000).In fact, Pearson's correlations for taxa-environment were consistently high (Table 2) (r > 0.9).The Monte-Carlo permutation test indicated high signifi cance (<0.01) of the data expressed in the fi rst canonical axis, as well as for all of the ordination axes considered together (Table 2).
The gradient expressed in the fi rst canonical ordination axis is consistently correlated with longitude (r = 0.65), annual temperature (r = 0.62), and rainfall (r = -0.63).The second axis showed consistent correlations with latitude (r = 0.83), altitude (r = 0.68), and longitude (r = 0.64).Consistent correlations were also observed between altitude and annual temperature (r = -0.7);longitude and rainfall (r = -0.73);and latitude and longitude (r = 0.82) (Table 3).As such, the fi rst ordination axis mainly refl ects the change in fl oristic composition between the states of São Paulo and Rio de Janeiro, indicating a geographic gradient with higher average temperatures and lower altitude and rainfall with increasing longitude in a northeasterly direction.The second axis expresses primarily the gradual alteration of species related to changes in latitude and altitude (latitude diminishing towards the south, and altitude diminishing both towards the coastline and to the northeast).These results are in agreement with patterns observed in southeastern Brazil that indicated the infl uence of geo-environmental factors on the differentiation of the tree fl ora in that region (Oliveira-Filho et al. 2005).
The relationship between environmental variables and species composition of the 32 sites of Atlantic forest is illustrated by the CCA diagram (Fig. 2).Three distinct groups can be identifi ed: one group of areas (black squares) in the lower left quadrant are all located along (or very near to) the Serra do Mar range near the São Paulo coast, and they are united by a fl oristic pattern determined by high levels of rainfall, predominantly high temperatures, and low-to-medium altitudes; a group in the upper left quadrant (black triangles) formed by sites in both São Paulo and Rio de Janeiro is associated with high altitudes, low annual temperatures, and intermediate rainfall levels; a large group in the upper and lower right-hand quadrant (white circles) is formed by sites in the Serra do Mar (RJ), Guanabara Graben, coastal mountains, and northern region, all in Rio de Janeiro state, and are associated with low-to-average altitudes, lowto-intermediate rainfall, and high annual temperatures (Fig. 2).These results demonstrate that the fl oristic differentiation between areas of Atlantic forest in Rio de Janeiro and those in São Paulo are a result of differences in rainfall, altitude, and average annual temperature, and reinforces the role of geo-environmental variables in speciation (Gentry 1982, Huggett 1995, Oliveira-Filho et al. 2005).
Among the various environmental factors examined, rainfall plays a signifi cant role in the fl oristic differentiation between São Paulo and Rio de Janeiro.The importance of rainfall in determining fl oristic composition has already been confi rmed in extensive studies in the Neotropics and in the differentiation of ombrophilous and seasonal forests in the Atlantic forest biome of southeastern Brazil (Gentry 1982, Oliveira-Filho & Fontes 2000).Rainfall decreases from São Paulo to Rio de Janeiro following the Serra do Mar range (which crosses the coastal region of São Paulo in a northeasterly direction) as the slopes become more distant from the coastline, thus causing decreasing infl uence on orographic rainfall, as the range approaches southern Rio de Janeiro (and northeastern São Paulo state), and allowing widening of the coastal plains known as Guanabara Graben (Almeida & Carneiro 1998).This geographic situation translates into heavier rainfall on the coastal plains (and the slopes of the Serra do Mar range in São Paulo) and lighter rainfall in Rio de Janeiro state (Barbosa 2007).This variation in rainfall imposed by landform changes helps explain the differences observed between the Atlantic forest of São Paulo and Rio de Janeiro (Huggett 1995).
The heterogeneity of the Atlantic forest has been examined in various comparative studies employing cluster analysis (Peixoto et al. 2004, Carvalho et al. 2006b, Rolim et al. 2006).The use of this technique together with TWINSPAN analysis has confi rmed and refi ned information generated by CCA.The interpretation of the dendrogram obtained in the similarity analysis is a subjective task that depends greatly on the researcher who must subdivide groups if there is compelling ecological evidence (Kent & Coker 1992).As such, three consecutive cuts of the dendrogram defi ned the groups "A", "B", "C", "D", "E", "F", and "G".In the fi rst cut, two principal groups ("A" and "B") were identifi ed; in the second cut, group "B" was subdivided into "C", "D", and "E"; and in the third cut, "E" was subdivided into "F" and "G" (Fig. 3).
Group "A" was separated from the group of dense ombrophilous forests ("B") and is composed mainly of seasonal forests (Fig. 3).According to the CCA, the four sites in this group are related by high temperatures, low rainfall, and low altitude.Although the CGR site is geographically distant from the other sites in this group  4).
Group "B" is composed of dense ombrophilous forests in the states of São Paulo and Rio de Janeiro (Fig. 3) with typical species such as Alchornea triplinervia (Spreng.)Müll.Arg., Cabralea cangerana Saldanha, Cariniana estrellensis (Raddi) Kuntze, Chrysophyllum flexuosum Mart., Euterpe edulis Mart., and Licania kunthiana Hook.f. (Table 4).The inclusion of the MPS site (classifi ed as a semideciduous seasonal forest by Veloso et al. 1991) in group "B", as well as its proximity to the dense ombrophilous forests in the CCA diagram (Fig. 2), suggests that the fl ora is strongly infl uenced by the surrounding forest formations (Serra do Mar and Serra da Mantiqueira) and helps explain its inclusion in this group.The CCA diagram attributes greater rainfall, intermediate-to-high altitudes, and lower annual temperatures to group "B".It appears that these factors constitute the principal forces of fl oristic differentiation between the sites in groups "A" and "B".
Group "C" comprises areas of dense montane ombrophilous forest -DMOF (Fig. 3) and demonstrated the same tendency in CCA (Fig. 2).The similarities between the high montane forests of São Paulo and Rio de Janeiro draw attention to the relationship between the forests in these two mountain chains (Serra do Mar and Mantiqueira) that are separated by the Paraíba do Sul River valley.This observation was also supported by investigations undertaken   in southeastern Brazil (Oliveira-Filho & Fontes 2000), in the region of Bacias do Leste (Oliveira-Filho et al. 2005), and in Rio de Janeiro state (H.C.Lima, unpublished data).
Although the Tinguá site is classifi ed as a dense montane ombrophilous forest (Veloso et al. 1991) it was grouped with forests found at medium-to-low altitudes.This may be the result of sampling in Tinguá, which was concentrated at lower elevations (Rodrigues, unpublished data).The areas included in group "D" are located along the coast of São Paulo in the Serra do Mar range (with the exception of BSD in the Serra de Paranapiacaba range).These same localities were also grouped by CCA, mainly because of high rainfall, high temperatures, and low-tomedium altitudes (Fig. 2).The formation of group "D", and its distinction from group "E" in the cluster analysis, confi rms a fl oristic differentiation between dense ombrophilous forests in São Paulo and Rio de Janeiro (Fig. 3).This differentiation may be related principally to rainfall, as suggested by the CCA diagram.The separation between the ombrophilous forests in the states of São Paulo and Rio de Janeiro had been detected in earlier studies (Oliveira-Filho & Fontes 2000, Oliveira-Filho et al. 2005, Rolim et al. 2006).The present study emphasizes the determinant role of landscape in the fl oristic composition of the Atlantic forest in these states through its infl uence on regional rainfall patterns (Almeida & Carneiro 1998, Barbosa 2007) that are in turn responsible for the observed difference in species composition.Some of the species indicated as typical for group "D" through TWINSPAN analysis include Ilex dumosa Reissek, Ixora burchelliana Müll.Arg., Pseudopiptadenia warmingii (Benth.)G.P. Lewis & M.P. Lima, and Roupala paulensis Sleumer.(Table 4).
Group "E" unites dense ombrophilous forests at lowto-medium altitudes in Rio de Janeiro state having typical species such as Cordia trichotoma (Vell.)Arráb.ex Steud., Guarea kunthiana A. Juss., Miconia prasina (Sw.)DC., Ocotea schottii (Meisn.)Mez, Pradosia kuhlmannii Toledo, Sorocea hilarii Gaudich., and Xylopia sericea A. St.-Hil.(Table 4).As was suggested earlier, the fl oristic differences between groups "E" and "D" appear to result mainly from differences in rainfall between São Paulo and Rio de Janeiro.However, according to the CCA diagram (Fig. 2), it is also possible to detect the infl uence of higher annual temperatures on the sites in Rio de Janeiro included in group "E", as they are often located on the coastal plain region in the Guanabara Graben.As such, the fl oristic differences between groups "E" and "D" can be attributed to variation in rainfall and annual temperatures -reinforcing the infl uence of landscape features on the differentiation of the tree-shrub composition of forests in Rio de Janeiro and São Paulo.Group "E" can be further subdivided into "F" and "G".This subdivision refl ects a tendency towards a fl oristic differentiation between areas in the Guanabara Graben and those on the slopes of the   4).However, a clear distinction between these two groups, as detected by similarity analysis (Fig. 3), cannot be found when observing the CCA diagram (Fig. 2).This may be the result of differences between the types of analyses utilized.The CCA considers species as well as environmental variability, while similarity analysis considers only species.As such, the sites clustered by the dendrogram into groups "F" and "G" may not necessarily be closely aligned in the CCA diagram.Nonetheless, it is possible to perceive a general tendency for the areas composing group "F" to have greater altitudes but colder annual temperatures, while the members of group "G" have a greater tendency to occur at lower altitudes in areas that have higher annual temperatures (Fig. 2).
It is important to note that only qualitative data was used in the present study, and that quantitative data is generally more reliable as it takes both the occurrence and the abundance of species into consideration (Kent & Coker 1992, McCune & Grace 2002).Despite that our results demonstrate the existence of important fl oristic patterns in the Atlantic Forest tree-shrub layer in Rio de Janeiro and São Paulo states, indicating the substitution of species across these two states, which corroborates many aspects previously published in the scientifi c literature (Oliveira-Filho & Fontes 2000, Moreno et al. 2003, Peixoto et al. 2004, Carvalho et al. 2006b, Rolim et al. 2006).These changes in fl oristic composition are related to transition of the Serra do Mar Range landscape as it moves further from the coastline (in a northeasterly direction), with consequent widening of the coastal plain (Guanabara Graben).The observed patterns of fl oristic differentiation were mainly attributed to high rainfall along the coast of São Paulo state, to high and medium average annual temperatures on coastal plain and mountain slopes in Rio de Janeiro, and to infl uence of altitudinal variation (Gentry 1982, Webster 1995, Oliveira-Filho et al. 2005).Because of these patterns we were able to subdivide the 32 areas analyzed into seven different phytoecological regions of Atlantic Forest through the generation of groups, each of which had several preferential species selected on the basis of TWINSPAN results.Therefore, the information provided by this analysis (Table 4) is a valuable guideline for elaborating a list of species to be used in environmental restoration programs in the seven phytoecological regions here applied.This type of data can be very important for orienting conservation efforts in the Atlantic forest, although efforts should be made to address phytogeographical questions using quantitative data, since this information is more accurate and allows the confi rmation or rejection of some of the patterns detected in current studies.
Table 1.Code designation, bibliography, and general data concerning 32 areas analyzed in the present study located in the states of Rio de Janeiro and São Paulo, Brazil.Temp.= average annual temperature, Rain = average annual rain (mm), Method = methodology, Total sample = total area (m2) or numbers of points, S = number of species.* Forest formations: FOD = dense ombrophilous forest, FES = semi-deciduous seasonal forest, TB = lowland forests (Terras Baixas), SM = submontane, M = Montane.Methodology applied: P = plots, T = transects, PQ = point quadrant, CL = free walks.** unpublished data.central Rio de Janeiro), and is classifi ed as a dense submontane ombrophilous forest (Veloso et al. 1991), it has environmental traits that lend a seasonal character to the site and contribute to differentiating it from other areas in São Paulo and Rio de Janeiro (as has been demonstrated in studies by Oliveira-Filho & Fontes 2000; Moreno et al. 2003) -but makes it very similar to the rest of group "A" (Fig. 2).Group "A" has an associated fl ora composed of species such as Alseis pickelii Pilger & Schmale, Acosmium lentiscifolium Vogel, Metrodorea brevifolia Engl., Pereskia grandifolia Haw., and Trichilia pseudostipularis (A.Juss.) C. DC., which were identified based on the results of TWINSPAN (Table

Figure 2 .
Figure 2. Diagram of canonical correspondence analysis (CCA).The grouping tendencies, demonstrating fl oristic patterns generated by geoenvironmental variation, are denoted by different symbols.

Figure 3 .
Figure 3. Cluster dendrogram of 32 sites of Atlantic forest in Rio de Janeiro and São Paulo.Associations are generated by group average after calculation of sites Jaccard coeffi cient similarity matrix.Groups are indicated by letters A,B, C, D, E, F and G. Since there was no dendrogram similarity value smaller than 4% or higher than 42%, axis range is shown from 2,5% to 42,5% to facilitate visualization.

Table 2 .
Summary of the results of the canonical correspondence analysis (CCA) and Monte Carlo test with the arboreal fl ora and environmental variable in 32 areas of Atlantic Forest in the states of Rio de Janeiro and São Paulo.

Table 3 .
Results of the canonical correspondence analysis (CCA), providing intra-set and inter-set correlations for the fi rst two axes, and the correlations between the environmental variables.Alt.= altitude, Prec.= average annual precipitation, Temp.= average annual temperature, Lat.= latitude; Long.= longitude.