Does climate influence spatial variation of tree species alpha diversity in the Brazilian Atlantic rainforest ?

We tested the hypothesis that the variation in tree species alpha diversity is driven by climate in the Brazilian Atlantic Rainforest (ARF). Considering 139 samples of trees with DBH ≥ 4.8 cm, we correlated alpha diversity measures (Shannon heterogeneity index H', Chao I richness estimator, and Simpson concentration index C) with climate variables (perhumidity index, mean annual rainfall, and mean annual temperature) and spatial variables (latitude, longitude, and altitude). Using CCA, multiple regression analysis and RDA procedures, we found a positive relationship between latitude, longitude, and altitude with Shannon’s diversity index and Chao I richness estimator, and a negative relationship with Simpson concentration index. Over 75% of the variation remained unexplained and were attributed to stochastic processes. These results indicate that climate has a very weak influence on tree species alpha diversity, which is more influenced by spatial variation in the ARF. We propose that the current tree species alpha diversity could be a result of the history of the ARF during the Cenozoic, when geological events and climate oscillations could have triggered biogeographic processes, such as alternating episodes of vicariance and dispersal, which would have lead to the great diversity of species and heterogeneity across the geographic space observed today.

Among them, climatic variables have been identified among the major abiotic factors related to variation in species richness and alpha diversity in tropical forests (GENTRY, 1988).Besides climate, latitudinal variations also play a fundamental role in species richness variation and alpha diversity (OLIVEIRA-FILHO et al., 2005).One of the most species-rich forests in the world is the Brazilian Atlantic Rainforest (ARF), but information on how climate influences alpha diversity in this forest is still lacking.
The Atlantic Forest sensu stricto (JOLY et al., 1992), or Atlantic Rain Forest (ARF) (IBGE, 2012), extends from 3° to 30°S, along almost the entire Brazilian coastline.It is considered one of the 35 global biodiversity hotspots, with high species richness and endemism (WERNECK et al., 2011), and is one of the most endangered areas in the world (MYERS et al., 2000;ORME et al., 2005).Although the influence of climate and soil is undeniably important for species diversification, there is no evaluation on the effects on values of alpha diversity across the entire ARF range.Indeed, there are only few studies on the role of climate on the ARF diversity, but most have a local scope, and some controversy has arisen about which factors are the most important to explain the remaining ARF diversity.For instance, Gentry (1988) compared species richness and Shannon diversity index with climate and other environmental variables for several neotropical sites and found a direct relationship with annual rainfall for most areas, but in Linhares, Espírito Santo State, Brazil, an area with average annual rainfall of 1,400 mm, it was higher than expected.Oliveira-Filho and Fontes (2000) and Oliveira-Filho et al. (2005) found a latitudinal correlation between tree species richness and variations in temperature and rainfall.
Although some effort has been made to describe patterns of species distribution in the ARF, many gaps still remain, especially those dealing with species richness distribution along the total area of ARF.Except for Oliveira-Filho and Fontes (2000), there is a general lack of investigations using quantitative analytical tools to associate diversity patterns with climatic variations in the ARF.Considering local tree communities across the entire range of the ARF, this study tested the following predictions: 1) The spatial variation in tree species alpha diversity is associated with one or more abiotic environmental variables; 2) The variation in different diversity indices is related to the same environmental variables across the entire ARF range; 3) The diversity variation, expressed by different indices, has a gradient across the entire ARF range.

Study Area
This study considered the Phytoecological Region of the Atlantic Rainforest (ARF) the area mapped by the Brazilian Institute of Geography and Statistics (IBGE,2012), extending from 3° to 30°S (CÂMARA, 2005) along the coastline.Its elevations range from the sea level to higher than 2,700 m.This area is characterized by a high degree of endemism in plants and animals (MYERS et al., 2000;WERNECK et al., 2011).Its tree formations are classified according to altitude ranges, whose lower and upper limits vary according to latitude (IBGE, 2012): Alluvial (Riparian) Forest, Lowland Forest (5 to 30 or 100 m altitude, depending on latitude), Submontane Forest (30 to 400 or 600 m), Montane Forest (400 to 1000 of 2000 m), and Altimontane Forest (above Montane Forest).
Two predominant climate regimes are found in the ARF: the humid tropical coastal and the subtropical (NETO; NERY, 2005).The humid tropical coastal regime is prevalent in the states of Ceará to São Paulo (approximately from 3° to 22°S), while the subtropical regime, with a subtropical thermal seasonality, prevails from the state of São Paulo to Rio Grande do Sul (approximately from 22° to 30°S).Major landforms found in the ARF are mountain ranges (serras), coastal plains (tabuleiros), and plateaus (chapadas) (BACKES; IRGANG, 2004).

Database
From the literature, we selected published phytosociological tables (quantitative surveys) that sampled trees with DBH (diameter at breast height) ≥ 4.8 cm in the entire ARF range.The phytosociological table represents species lists with the respective species phytosociological descriptors, such as density per area, frequency per sampling unit, and basal area per terrain surface area (CAIAFA;MARTINS, 2007).Since the same study may present results for more than one site, that is, may present more than one phytosociological table, the final total number of phytosociological tables in our dataset was greater than the number of papers we consulted.The tables were then classified and coded according to the IBGE (2012) geographical region and the state in which the survey was conducted.
We took as a starpoint the database FITOGEO (SCUDELLER; MARTINS, 2003), which includes quantitative samplings performed in the ARF from 1946 to early 2005.Then, we surveyed the literature for papers concerning quantitative sampling of trees with DBH ≥ 4.8 cm published from January 2005 to January 2007 in peer-reviewed journals.We added these data to the FITOGEO database to build our own dataset.Afterwards, our final dataset had a total of 139 phytosociological tables from 79 surveys (Figure 1), 2,168 species (six exotic species were removed from the database), 449 genera, 100 families, and 126,238 individuals.We arranged the species in families according to APG III (2009) and checked for updated binomials at http://mobot.mobot.org/W3T/Search/vast.html and http: //www.ipni.org/ipini/ipni.html.

Alpha diversity and climate variables
We used the following three indices to measure alpha diversity of the ARF (Table 1): Shannon's diversity index (H'), Chao I richness estimator (Chao I), and Simpson's concentration index (C).All indices were calculated using the Species Diversity and Richness 4 software (SEABY; HENDERSON, 2006).
Shannon diversity index (H') is a measure of heterogeneity that considers both the relative abundance of individuals and the number of species in a sample.It calculates the best relationship between species richness and evenness (STOCKER et al., 1985).Simpson concentration index (C) is a measure of the concentration of relative abundance on the species (SIMPSON, 1949).It is based on evenness and calculates the probability that two individuals randomly and independently taken from one community belong to the same species.The Chao I richness estimator (CHAO, 1984) uses information on the distribution of rare species in the sample, i.e., those represented by only one or two individuals.The greater the number of rare species in the sample, the greater is the probability that species other than those represented in the sample occur in the area (GOTELLI;COLWELL, 2001).
We obtained climate variables from DIVA-GIS 5.2 (HIJMANS et al., 2005) and used rainfall data to calculate the Perhumidity Index according to WALSH (1996).Monthly Perhumidity Indices may vary from -24 to +24, depending on the average rainfall.A very wet month (p > 200 mm) receives a score +2, a humid month (100 ≤ p ≤ 200 mm) receives a score of +1, a dry month (50 ≤ p ≤ 99 mm) has a score of -1, and a very dry month (p <50 mm) a score of -2.However, when a dry or very dry month follows a humid or very wet month, its score is increased by -0.5 and -1.5, respectively, since the availability of water in the soil may still be high.Therefore, the smaller the Perhumidity Index, the drier and more seasonal the climate is.Then, we checked for multicollinearity among all variables available in DIVA-GIS plus the Perhumidity Index by means of a Principal Component Analysis.Among the 19 climate variables analysed, only four (Altitude, Perhumidity, Average Annual Rainfall, and Mean Annual Temperature) showed no collinearity and were thus selected for our analyses.In addition to those four variables, we included Latitude (LAT), Longitude (LONG), and the Perhumidity Index (PER).

Data analysis
To test our hypotheses, Canonical Correspondence Analysis (CCA), (TER BRAAK, 1995), multiple regressions and partial redundancy analysis (BORCARD et al., 1992) were run.A matrix of the localities per diversity values (columns) and another with the localities per environmental variables were used for the CCA.The CCA was used to identify environmental variables associated with diversity indices.Because of large differences in scale among the variables, the data were normalized using a ranging transformation.To test the significance of CCA correlations, a Monte Carlo permutation test (999 permutations) was performed.All the analyses and transformations were performed in Fitopac Shell 2 (SHEPHERD; URBANETZ, 2010).
Using multiple regression analysis, minimal models were obtained by the stepwise method in Systat 10, permitting the removal of the non significant variables (p > 0.05).Samples generating high levels of residuals and representing outliers (Stantard Deviation > 2.0) were removed from the analysis.Analysis of values for T g 1 and T g 2 showed the need to transform H', C, and Chao I variables to base 10 logarithm.With the transformation of the dependent variables, all explanatory variables were also transformed.Since some samples showed negative values, we added 19 to all results, thus obtaining positive values for all samples.
We conducted a partial redundancy analysis (RDA) (LEGENDRE; LEGENDRE, 1998) to investigate which factors (environmental, spatial or stochastic) most determine the variation of alpha diversity values across the ARF.As factor-variables, we used a matrix of localities per environmental data (Altitude, Perhumidity, mean annual rainfall and mean annual temperature) and a matrix of localities per spatial data (latitude and longitude).As response-variable, we used a matrix with localities per alpha diversity indices (H ', Chao and C).Analyses were performed using the Varpart function of the Vegan package in R (R DEVELOPMENT CORE TEAM, 2010).

Results
The highest values of heterogeneity (Shannon H'), richness estimate (Chao I) and the lowest values of abundance concentration (Simpson C) occur unevenly at latitudes corresponding to the middle range of the ARF (Table 1).Since the ARF ranges from 3 o to 30 o south latitude, the greatest diversity values are found around 19 o S.These latitudes correspond to the Doce River valley.
The canonical eigenvalues for all axes of the CCA were low (less than 18%), indicating a weak relationship between diversity indices and environmental variables (Figure 2).The Monte Carlo permutation test showed a correlation between diversity indices and environmental variables for the first two canonical axes (Axis 1. eigenvalue = 0.0259, p = 0.0010; Axis 2. eigenvalue = 0.0025, p = 0.0350).Percentage of the explained variance for the first axis was 16.17% and for the second axis, 1.59%.The sum of the eigenvalues for axes 1 and 2 explained only 17.76% of the correlation between environmental variables and diversity indices.The sum of the non-canonical eigenvalues accounted for 83% of the explanatory power of the results, i.e., 83% of the results could not be explained by the correlation model of species diversity and climatic variables.
We performed the CCA in two modes, one considering the diversity indices, and the other considering the environmental variables.Both types of ordination provided similar results (Figure 2).Spatial variables (latitude, longitude, altitude) were opposed to climatic variables (mean annual temperature, mean total annual rainfall and perhumidity index).This means that as latitude, longitude and altitude increase, the mean annual temperature, the mean annual total rainfall and the perhumidity index decrease, indicating that drier, colder and more seasonal climates prevail at greater latitudes, longitudes and altitudes.In both ordinations, Shannon H' and Chao I were positively correlated with the spatial variables, whereas Simpson C was positively correlated with the climatic variables.This means that richness (Chao I) and heterogeneity (Shannon H') vary with latitude, longitude and altitude, whereas the abundance concentration in few species (Simpson C) varies as climate varies.
For the Shannon's diversity index (H'), the model obtained by multiple regression retained only variables of altitude, latitude, and longitude (R = 0.462; R 2 = 0.214; SD = 0.088, ŷ = 2.335 + 0.049 ALT + 0.236 LAT -1.341 LONG). .By applying the partial redundancy analysis (RDA), we found that the variation of alpha diversity in the ARF was not significantly explained by pure environmental variables and spatial components, or by the spatially structured environmental variation (Figure 3).The greatest explanatory power came from the stochastic processes, presented as residual in our results.

Discussion
Studies conducted in tropical forests showed close relationships between climatic variables and variations in species richness and diversity (PAUSAS; AUSTIN, 2001).Although these relationships are well established in the literature and are important to predict species richness in a site, some issues have been raised on which environmental variables best explain species distribution.
Comparative studies have associated the increase in species richness with average annual total rainfall, number of dry days in the year, altitude, mean annual temperature, and latitudinal variation (SCUDELLER et al. 2001, OLIVEIRA-FILHO et al., 2005).Average annual rainfall is considered one of the most important environmental factors in determining species diversity in tropical forests (GENTRY, 1982;1988;CLINEBELL II et al., 1995, OLIVEIRA-FILHO;FONTES, 2000;LINDER, 2001SCUDELLER et al, 2001).Nevertheless, our results showed no direct relationship of this variable with the increase or reduction in alpha diversity in the ARF.Walsh (1996) considered annual rainfall values of at least 1700 mm, with a dry season either absent or short, as requirements for a tropical rain forest with its expected high species diversity.Nevertheless, our findings showed that the degree of variation in the mean annual rainfall may not be such a significant factor in determining species alpha diversity in the ARF.
Altitude had a positive relationship with the Shannon diversity index and the Chao I richness estimator.Influence of altitude on the distribution of tree species in the ARF is already well known (OLIVEIRA-FILHO et al., 2005).The results of Gentry (1988) diverge from ours, since this author registered that the site diversity in tropical regions tends to decrease with increasing altitude, though that pattern is not necessarily valid for all forests.Oliveira-Filho and Fontes (2000) reported a gradient of species, genera, and families in the ARF, regulated by variations in altitude, and distinguishing the ARF of the Northeast and Southeast Brazil (Espírito Santo and Bahia) from the ARF of the Southeastern and Southern Brazil (Rio de Janeiro, São Paulo, and Paraná).Scudeller et al. (2001) verified that altitude represents a strong environmental variable determining tree species distribution in the São Paulo State.Oliveira-Filho et al. (2005) registered that variations in altitude are strongly correlated with the internal differentiation of both the ARF and the Seasonal Semidecidious Forests.
According to these findings, physiognomic variations in the Phyto-ecological Region of the ARF would be a result of environmental conditions associated with altitudinal and latitudinal variations and that higher species diversity at high altitudes is due to the altitudinal gradient of diversity, with the montane and altimontane forests presenting greater species diversity than the lowland and lowermontane forests.Another important factor to consider in the distribution of the ARF is the type of relief.In Southeastern Brazil, the diversity center of the ARF (R.M.CERQUEIRA, unpublished data), the main form of relief is mountain ranges (serras), whereas in the Northeast prevail coastal plains (tabuleiros) and plateaus (chapadas).Areas with predominance of mountain ranges, usually located at higher altitudes, have sites with better preserved vegetation probably because of difficulties of access and mechanization for land use.Therefore, altitude, an important variable in determining species diversity in the ARF, should be considered within the context of variations in landforms.
Both Shannon diversity index (H') and Chao I richness estimator (Chao I) were positively correlated with increasing latitude.Our results contrast with the general trend of increasing biodiversity with reducing latitude.Fischer (1960), Pianka (1966), Gentry (1988), Rosenzweig (1995), among others, associated increasing biodiversity of the sites with decreasing latitude.Although we registered a positive correlation of diversity indices with latitude in the ARF, we believe that this correlation is because the richness center of the ARF is located in the Southeastern Region, as firstly indicated by Smith (1962) and confirmed by R.M. Cerqueira (unpublished data).According to R.M. Cerqueira (unpublished data), the higher alpha diversity in the central portion of ARF could be the outcome of historical factors (climatic fluctuations of the Quaternary) and associated biogeographical processes (alternating vicariance and dispersal) that occurred in this phytoecological region, which caused alternating retraction and expansion of forest areas in the ARF.The higher values of the Shannon's diversity index (H') in the ARF middle range could have biased the results of the canonical correlation and regression, without necessarily implying that areas of higher latitudes have greater biodiversity.
The negative correlation between longitude and species richness was analyzed in different countries and in different geographic regions of Brazil, but no investigation of this correlation had been carried out so far for the total extension of the ARF in Brazil.By analyzing samples from the ARF and from the Seasonal Semidecidious Forests in the São Paulo State, R.J. Oliveira (unpublished data) found a negative correlation of tree species richness with distance from the Atlantic.The author concluded that the direct correlation of species richness with latitude and annual thermal amplitude may be a consequence of the correlation of these variables with the distance from the Atlantic.Thus, the author found an indirect effect of the distance from the Atlantic and its associated humidity gradient.O' Brien (1993) described a pattern of longitudinal variation for the woody flora richness in South Africa, with climate accounting for the 78% of that variation.Although all these authors investigated directly the variation in the number of species, not the composite indices that we used, our findings of the longitudinal pattern of decreasing tree species diversity, with a gradual replacement of species along the coastal-inland gradient, corroborated other studies performed in tropical forests.
The Simpson concentration index (C) was negatively correlated with latitude and altitude, the opposite to that found for Shannon heterogeneity H' and the Chao I richness estimator.Communities with high diversity tend to show great evenness (MARTINS; SANTOS, 1999) and species richness.Our findings show that more diverse areas tend to have higher species evenness, with less concentration of abundance; whereas those with lower species diversity tend to have lower evenness and greater abundance concentration.Also, our results show that high H' and Chao I and small C vary in this way as latitude, longitude and altitude vary in the ARF.
Spatial variables alone (latitude, longitude, and altitude) were retained for our minimum models of multiple regression, indicating that the space has a greater explanatory power than environmental factors for species diversity variation in the ARF.However, in the partial redundancy analysis (RDA), the space had little power to explain the variation of alpha diversity values.Stochastic processes were determinants and explained more than 75% of the variation of the alpha diversity.It is more likely that the current distribution patterns of tree species in the ARF are a result of vicariance and dispersal events caused by climatic oscillations and geological events in the Cenozoic (ANDRADE-LIMA, 1982; BIGARELLA; ANDRADE-LIMA, 1982) than resulting from limitations imposed by variations in climatic factors.Scudeller et al. (2001) showed a significant negative relationship between floristic similarity and geographical distance in samples from the ARF: the farther two communities are, the less similar they are.This pattern may be indicative of dispersal limitation, a basic process in stochastic models for community assembly (HUBBELL, 2001).Species of wide distribution may have great ability for dispersal and tolerance to different types of environments, whereas species of restricted distribution tend to have limited dispersal ability (BOULANGEAT et al., 2012).The dispersal ability may influence species replacement rate and is directly associated with distance, being independent from local weather conditions (SVENNING; SKOV, 2004).Plants with limited dispersal range are called stenotopic, almost 54% of the tree species in the ARF are characterized as such (CAIAFA; MARTINS, 2010).R.M. Cerqueira (unpublished data) have shown that most of the analyzed tree taxa (families, genera, and species) in the ARF have low relative constancy (occurrence in less than 20% of the total number of samples).Altogether, these findings suggest that ARF tree species have reduced dispersal ability and tend to remain in the place of origin.

Conclusion
Regardless of which diversity index (Shannon H', Chao I, or Simpson C) or community attribute (species richness, or species richness and abundance) were used in our analyses, we found similar results, indicating that our findings are consistent.The Atlantic Rainforest is characterized by high diversity of tree species.A very small fraction of the tree species diversity in the ARF is conditioned by climate, a greater portion (but still small) is conditioned by the spatial variables of altitude, latitude and longitude, and most part (over 75%) is conditioned by stochastic factors.These results indicate that climate has a weak influence on the values of diversity indices, which vary with spatial variation, but are mostly explained by historical events and biogeographic processes that are manifestations of stochasticity.We found a positive correlation between alpha diversity and latitude, probably due to the location of a richness center in the middle range of the ARF.The current status of the ARF diversity could be a result of regional biogeographic processes triggered by stochastic geological events and climate oscillation occurring over time and resulting in alternating processes of vicariance and dispersal, which have influenced the present patterns in tree species diversity.Environmental variables, such as mean annual rainfall, perhumidity, and mean annual temperature, although considered important in studies performed so far in tropical forests, showed a very weak influence on the values of alpha diversity in the ARF.

Figure 2 .
Figure 2. CCA ordination diagram for axes 1 and 2, based on scores for the diversity indices (WA) and environmental variables (LC).H': Shannon diversity index; CHAO: Chao I richness estimator; C: Simpson concentration index.Environmental variables: ALT: altitude; LAT: latitude; LONG: longitude; TMAnual: mean annual temperature, Perhumidity index and Mean Annual rainfall.Longitude showed a negative relationship with increasing H' values.To obtain the minimal model for the Chao I richness estimator, four phytosociological tables [SE-SP(2), S-SC(41), NE-CE(72), and SE-SP(86)] were removed from the analysis due to high levels of residuals, and because they represented outliers.The minimal model retained the same variables, showing a

Figure 3 .
Figure 3. Variation partitioning by RDA for different indices of alpha diversity (Shannon H', Chao I, and Simpson C) in the Atlantic Rainforest.The numbers over the bars indicate the variation explained by pure environmental factors, by pure spatial variables, by spatially structured environmental variation, and the residue.Most variation (over 75%) remained unexplained.