Structure , diversity , and spatial patterns in a permanent plot of a high Restinga forest in Southeastern Brazil

(Structure, composition, and diversity of a high Restinga forest in the Southern Atlantic Forest) We assessed the structure, diversity and distribution of tree species in a 10.24-ha permanent plot of high Restinga forest (HRF) in southeastern Brazil. We sampled 15,040 individuals belonging to 45 families, 87 genera and 116 species (density= 1,468 trees ha-1, and basal area= 28.0 m2 ha-1). Mean richness was lower than other types of tropical forests and other HRFs that have been less intensively sampled, which is probably due to the younger age and hydromorphy of the soils in the study site. Tree density, basal area, species composition, and diversity all varied considerably depending on the size class considered. The studied plot had a density similar to other tropical forests but with a smaller basal area (fewer individuals > 60 cm in diameter), and higher density of smaller trees. Multi-stemmed individuals were not frequent (11% of the trees), though they were widespread among species (74% of the species). Spatial distribution was predominantly clumped, a pattern common to other types of tropical forests. Besides growing over white-sand soils, the PEIC plot and other white-sand forests of the world also had very similar patterns of community structure, richness and diversity. Therefore, we consider the HRF as a recognizable subtype of white-sand forest.


Introduction
One useful way to study tropical forests is via large permanent inventory plots (Condit 1998), specially designed for the study large-sized, long-lived organisms such as trees.Recently, numerous plots have been established in a variety of types of forests around the world (Losos et al. 2004).However, these permanent plots are restricted to particular types of soils, primarily oxisols and ultisols (Ashton 2004).Still, surprisingly diverse forests grow on a larger number of soil types, including relatively nutrient-poor, white-sand soils (Newbery 1991;Phillips et al. 1994).In Brazil, besides the widespread Amazon campinaranas (Anderson 1981), some coastal forests also grow on sandy soils (mainly spodosols), the Restinga forests.Th ese forests are known to be lower and less species rich than surrounding rain forests, although sp ecies rich Restinga forests may also be found (César & Monteiro 1995;Guedes et al. 2006;Negrelle 2006).
But what is the typical structure and composition of Restinga forests?How is it diff erent from forests that grow on non-sandy soils?How clumped are trees of diff erent species on such nutrient poor-soils?Is there any evidence of species being more associated to habitat characteristics than in forests on richer soils?Th e answer to these and many other questions are still unavailable because detailed accounts of Restinga forests are largely missing.Plots over 1-ha are rare (Assis et al. 2004;Negrelle 2006), and therefore there is not enough data to properly assess general community-level patterns, Restinga population distributions, habitat associations or other features related to the ecology of these forests.For instance, the occurrence of multi-stemmed trees, cited as being typical of Restinga species (Sugiyama 1998;Guedes et al. 2006), has never been properly assessed.In addition, this lack of knowledge also prevents a proper comparison of Restinga forests with other forest types.
Restinga forests are part of the Restinga vegetation complex (or simply Restinga), a typical feature of Brazilian coastal plains.By defi nition, it is the vegetation that grows over tertiary or quaternary marine sand deposits.So, although the Restinga forest may be seen as a climax formation in the Restinga complex, it is a very young type of forest, geologically speaking.Besides its young age, the limiting soil conditions makes it a very promising setting in which the mechanisms that control the fl oristic composition and structure of tropical forests can be better understand (Scarano 2009).For instance, a low degree of endemism is expected because soils are much younger than other forests on soils developed in situ (Anderson 1981).In this scenario, one may ask if the Restinga composition is the result of simple historical events or if there is local and specifi c ecological processes (e.g.competition, facilitation or herbivore pressure -Scarano 2002;Fine et al. 2009) that control species establishment and growth.Th erefore, the study of tropical forests under such extreme conditions can give us a broader perspective on their structure, composition, and diversity.
In addition, Restingas are very important systems to physical and biological conservation of Atlantic Forest areas.Besides the more obvious important linked to the prevention of sea erosion, Restingas can shelter considerable populations of species that are rare in other Atlantic forest types (Scarano 2009).Th is type of vegetation is restricted to narrow spaces along the Atlantic Ocean coastline, places where human settlements are major and constant threats to biodiversity (Rocha et al. 2007).More specifi cally, high Restinga forests (HRF) represent one of the most rare and threatened vegetation types of the already deforested Brazilian Atlantic Forest (Ribeiro et al. 2009).
Aiming to describe and understand more precisely the composition, structure, diversity, and spatial patterning of Restinga forests at diff erent scales, we present a study of a 10.24-ha plot established in the high Restinga forest of the Parque Estadual da Ilha do Cardoso (PEIC), at 25 o S latitude.Th e main goal of this study was to detect general community-level patterns of the plot tree community.We evaluated several diff erent characteristics of the community such as tree density, basal area, and species richness and how these characteristics vary among diff erent tree size classes.We also assess the frequency and specifi city of multi-stemmed individuals in the community.Spatial analyses were carried out for the most abundant species, aiming to describe general patterns of species aggregation.

Study area
Th is study was conducted in the Parque Estadual da Ilha do Cardoso (PEIC), a 13,500-ha continental island located in Cananéia County, in the extreme southern tip of São Paulo State, southeastern Brazil (Fig. 1).Although at 25 o S latitude, the climate is tropical and wet with no dry season (mean annual water surplus= 1,330 mm -Af sensu Köp pen (1948) climate classifi cation).Mean annual temperature is 22.4 o C (average daily maximum and minimum are 26.1 and 18.0 o C), mean minimum temperature of the coldest month is 12.6 o C, and average annual precipitation is 2,261 mm.Only one month (August) has an average precipitation below 100 mm and the average rainfall for the three driest months is 314 mm (DAEE/SP, 2010).Data collection was carried out in a 10.24-ha (320×320 m) permanent plot established in 2002 (25 o 04'41"S and 47 o 55'53"W), subdivided into 256 subplots of 20×20 m (400 m 2 ).Th e terrain is fl at (altitude between 3 and 8 m above sea level) and developed over the sedimentary rocks of the Cananéia Formation (Upper Pleistocene).Th e soils are young (around 5,000 years), sandy (90-100%, mainly fi ne-grained sand), strongly acidic (pH in water between 3.9-5.0),hydromorphic, and with variable content of organic matter (Gomes et al. 2007).Th e plot soil survey classifi ed them as aquodic quartzipsamments (lower parts, periodic fl ooded), wet spodosols (arenic/oxyaquic alorthod and histic alaquod) and thionic histosols (terric sulfi saprists -soil maps, profi les and details on physical and chemical properties can be found in Gomes et al. 2007).Th e plot vegetation represents one type of the Atlantic Forest, the high Restinga forest (HRF), a type of forest which grows over marine sand deposits and is equivalent to Lowland tropical rain forest (Holdridge et al. 1971).Vegetation over marine sand deposits covers approximately 18% of the island.Th e PEIC has no main geographical barriers except rivers separating it from the continent, which is one of the largest and better conserved landscapes of the Atlantic Forest (Ribeiro et al. 2009).Th erefore, there is no evidence of dispersal limitation from the continent and other forest types of the island.In addition, there is no evidence of any recent human disturbance inside the PEIC plot, although it is possible that low intensity selective timber extraction occurred before the establishment of the park in 1962.

Tree census
Tree census followed the Center for Tropical Forest Science protocol (Condit 1998) except that it only included individuals with stem girth at breast height ≥ 15 cm, equivalent to diameter at breast height (dbh) ≥ 4.8 cm.All individuals were tagged, mapped and identifi ed to the species level.Th e diameter of each stem of each individual was measured to the nearest 1 mm and the number of stems of multistemmed individuals (i.e. with ramifi cations of the trunk below 1.3 m) was counted.Tree height was measured to the nearest 0.5 m for most individuals and was used to estimate mean and maximum canopy heights.Species identifi cations were based on several reference collections and the relevant literature.Vouchers were deposited in the Universidade de São Paulo (Campus Piracicaba) herbarium (ESA) under the collection numbers of D. Sampaio.We followed the APG II ( 2003) system.Th e plot was fi rst censused in 2002 and then again in 2004/2005; because of improved species-level identifi cations in the second census, data presented here refer to the individuals found alive during the second census.

Forest structure and diversity
In addition to the total plot density and basal area, the relative values of density and basal area were calculated for each species following Mueller-Dombois & Ellenberg (1974).Stems were assumed to have a cylindrical shape to calculate tree basal area.In order to assess forest stratifi cation we divided the sample according to diff erent diameter size classes: ≥ 4.8 cm; ≥ 10 cm; ≥ 20 cm; and ≥ 30 cm dbh.For multi-stemmed individuals, the diameter of the largest stem was used to determine the size class.We estimated the maximum richness using diff erent non-parametric estimators (e.g.Michaelis-Menten, fi rst and second order Jackknife), performed using 5,000 randomizations by the EstimateS soft ware (Colwell 2006).Th e choice of using the fi rst order Jackknife to discuss the results was made empiri- cally, based on the number of samples required to estimate the total plot richness and on the tendency of stabilization of the curve, following Colwell and Coddington´s (1994) recommendations.Average values of density, basal area, richness, singletons, and Fisher´s alpha and Berger-parker index (Magurran 2004) per hectare were estimated from all possible 1-ha square plots that could be fi t in the 10.24-ha plot (n= 144 plots).Th e number of singletons per hectare is used here as a practical defi nition of rare species and the Berger-parker index refers to the relative abundance of the most common species.All the analyses were completed for the four size classes in order to assess possible diff erences in the vertical structure of the forest.

Multi-stemmed individuals
Th e numbers of stems per individual were evaluated for each species, to detect if the distribution of the number of individuals in diff erent stem categories (1 to 10 stems per individual) was signifi cantly diff erent than the distribution observed for the community as a whole.Th e distribution observed for each focal species was compared to 1000 random distributions, obtained by randomly selecting from the community a number of individuals equal to the abundance of the focal species in the plot.For each category of number of stem per individual, we observed the proportion of random distributions (simulations) smaller than the number observed for the species in the category.When this proportion was equal to or greater than 95%, the observed number was considered to be diff erent than the expected by chance for the whole community (randomization test sensu Manly 2006).To test if the species with the highest proportions of multi-stemmed individuals were also the most abundant species in the plot, we compared the density of each species that had more multi-stemmed individuals than expected by chance with the density of all other species in the plot.Simulation procedures were also used to run this test.In both analyses, we excluded some palms (Euterpe edulis, Geonoma schottiana and Syagrus romanzoffi ana) that have strictly monopodial growth habits and one gymnosperm species (Podocarpus sellowii).

Spatial distribution
To describe the spatial patterns of species, the Morisita Index of Dispersion (ID, Krebs 1999) was calculated based on the 20×20 m subplots for each species with more than 70 individuals (n = 40 species).Th e observed value of the index for each focal species was compared with a null model that assumed complete spatial randomness of the individuals (Poisson distribution).Th is model was generated from 1000 simulations for the same number of individuals of the focal species.For each simulation, the null ID value was calculated and the distribution of values obtained was used to calculate the probability of the type I error on the statistical inference (Manly 2006).Th is approach was only used in the ≥ 4.8 and ≥ 10 cm dbh size classes because of the small number of individuals per species in the bigger size classes.Th e routines used to generate the null models of multi-stemmed individuals (described above) and of complete spatial randomness were developed in R (R Development Core Team 2008) and are available upon request from the authors.
To explore spatial patterns at scales larger than 20 m, we carried out spatial autocorrelation analyses.For the same 40 species with n > 70, the Moran´s index (I) was calculated for diff erent distance classes up to 200 m (correlograms, Legendre & Legendre 1998) based on the distance between the 20×20 m subplots.Th e global signifi cance of the correlogram was tested using the Bonferroni method (Legendre & Fortin 1989) and in cases of signifi cant correlograms, the values of the coeffi cients were tested for each scale individually.Th is test was performed using a Monte Carlo permutation routine that assigned randomly the observed densities for each subplot 1000 times to construct the confi dence envelopes around the null model at each given scale.Positive and signifi cant values indicate positive autocorrelation at a given distance class, whereas negative and signifi cant values indicate the opposite.Th e absence of signifi cant values indicates a random or regular distribution at the given distance class.Finally, to illustrate the existing types of spatial distribution, species were separated into general patterns of spatial distribution (see results for patterns description).Th e assignment of species to each pattern was based on the visual inspection of the correlogram and of the distribution map of the species in the plot.Th e defi nition of patterns was subjective but we do believe that they are representative examples and that they may be useful to assess what mechanisms are infl uencing species distributions.

Forest diversity and composition
We sampled 15,040 individuals that represented a total ba sal area of 290 m 2 (Tab.1).Th ese individuals belonged to 45 families, 87 genera and 116 species.Based on this sample, a total of 126 ± 3.1 (Jackknife 1) tree species was estimated for the HRF of the PEIC.Fisher´s alpha was 17.1 ± 0.6 (Shannon= 3.58) for the entire plot, but the value was slightly greater when calculated for samples of 1-ha (17.9).In the 1-ha subsamples, the number of singletons (13.7 singletons ha -1 ) was greater than that obtained for the 10.24-ha plot (9 singletons).In contrast, the Berger-Parker index of dominance was more or less the same independent of the sample size.Th e rarefaction curve indicated 5.5 ha as the area at which 95% of the total richness was found.In addition, the number of singletons leveled-off around nine aft er about 6.5 ha.
Arecaceae was the most abundant family, followed by Myrtaceae, Lauraceae, Annonaceae and Rubiaceae.Together they accounted for 58.8% of the sampled individuals.Th e most abundant species in the plot was Euterpe edulis with 293.8 ± 154.9 trees ha -1 , and a frequency of 99.6%.Th e 40 most abundant species in the plot are listed in Tab. 2. Calophyllum brasiliense, Ocotea pulchella, Albizia pedicellaris, Tapirira guianensis, Euterpe edulis and Manilkara subsericea were the species with the largest contribution to basal area (50.2%).Myrtaceae was the richest family (21% of species), followed by Lauraceae (10%), and Fabaceae (5%).Twenty fi ve families had just one species in the plot.Th e full list of species names and abundances from the 2004/2005 census can be found at http://ecologia.ib.usp.br/labtrop/doku.php?id=labtrop:01_projetos:02_pp:pp_peic:pp_peic#lista_ de_espécies.

Forest structure
Th e increase in tree diameter from 4.8 to 10 cm resulted in the exclusion of 8,811 individuals, that is, almost three fi ft hs of all individuals in the sample (58.6%).In turn, the total number of singletons increased with the size class considered.Th ere was a tendency for higher dive rsity and smaller dominance in the ≥ 10 cm size class (Tab.1).With respect to dominance, there was a shift in the dominant (i.e.more-abundant species) species among the size classes.Euterpe edulis was dominant in the size class ≥ 4.8 cm, Ocotea pulchella was dominant in the mid-size classes (≥ 10 and ≥ 20 cm) and Calophyllum brasiliense was dominant above 30 cm dbh.Th e six species with the greatest contribution to basal area (cited above) were also the most common species in the plot canopy (78.3% of trees ≥ 30 cm dbh).Albizia pedicellaris attained the greater average and overall dbh in the plot (34.4 and 85.4 cm, respectively), followed by Calophyllum brasiliense (mean dbh = 33.1 cm).Together, these two species accounted for 82% of trees ≥ 50 cm dbh in the plot.Besides these two species, only fi ve others exceeded 50 cm dbh (i.e.Eugenia sulcata, Myrcia glabra, Manilkara subsericea, Ocotea pulchella, and Tapirira guianensis).All species with individuals present in the plot canopy (dbh ≥ 30 cm) had individuals < 10 cm dbh.In contrast, 26 species did not exceed 15 cTh ere was considerable variance in abundance and, consequently, in species rank order among the diff erent size classes (Tab.2).Many species´ (45%) abundances gradually declined with increasing size class (e.g.Euterpe edulis, Xylopia langsdorffi ana, Amaioua intermedia, Andira anthelmia and Scheffl era angustissima).Small-sized species, as expected, declined drastically with an increase in size class (25% of species), with no or extremely low contributions to the > 20 cm dbh size class (e.g.Miconia cubatanensis, Marlierea racemosa, Guatteria australis, Garcinia gardneriana, Alibertia myrciifolia, and Ilex amara).Other species, however, presented a smoother decline in abundance with size classes, refl ecting peaks of abundance at larger size classes (17.5% of species) or abundance values more similar among classes (12.5%).Ocotea pulchella, for instance, had more individuals > 40 cm than < 10 cm dbh, showing a peak of abundance between 20 and 25 cm dbh.Examples of these general patterns of species diameter distribution are illustrated in Fig. 2.

Multi-stemmed individuals
We found 84 species (72%) and 1,623 individuals (10.8%) with two or more stems (mean ± sd: 1.17 ± 0.60 stems individual -1 ).Only 233 trees (1.5%) of 36 species (31%) had more than three stems.Eighteen species had signifi cantly more stem ramifi cations than expected for the community as a whole, whereas another 26 species had fewer multi-stemmed individuals than expected.Marlierea racemosa had the highest number of stems (10) and also the highest proportion of multi-stemmed individuals (50.5%).Th e mean abundance of species with signifi cantly more stems per individual (185.1 ± 256.7 trees) was marginally greater (p = 0.041) than the abundance of the other species in the plot (95.8 ± 154.3 trees).Among the species with signifi cantly fewer multi-stemmed individuals, only Blepharocalyx salicifolius did not have any multi-stemmed individuals in the plot.Some of the significantly less-ramifi ed species had high densities in the plot (n < 500), such as Amaioua Scheffl era angustissima, Andira anthelmia, Ocotea pulchella, and Tapirira guianensis (Tab.3), besides the monopodial palm Euterpe edulis.

Spatial distribution
Among the 40 most abundant species, dispersion patterns of three (Albizia pedicellaris, Myrcia brasiliensis and Ormosia arborea, pattern A in Tab. 2) were not signifi cantly diff erent from random at the 20 m scale.At larger scales, three other species (Scheffl era angustissima, Myrcia sp. and Ilex amara) were also not clumped.All other species had clumped pat-terns at one or more distance classes, with Myrcia bicarinata, Myrcia ilheosensis, Myrcia multifl ora, Marlierea racemosa and Ilex theezans having the highest aggregation indexes.In the size class ≥ 10 cm dbh, the random distribution was maintained for Albizia pedicellaris and Myrcia brasiliensis.Th e species Pera glabrata, clumped in the ≥ 4.8 cm size class, was not diff erent from random in the ≥ 10 cm class.Other species were still clumped but had considerable increases (e.g.Myrcia racemosa and Myrcia multifl ora) or decreases (e.g.Byrsonima ligustrifolia and Ilex theezans) in the Morisita index of dispersion in the size class ≥ 10 cm dbh.Visual inspection of the auto-correlograms and the distribution maps of the most abundant species in the plot (n > 70 individuals) suggested the existence of fi ve general distribution patterns (Fig. 3 and Tab. 2).For certain species the individuals were clumped in small patches (e.g.Ilex theezans, pattern B), whereas others were concentrated in larger and specifi c areas of the plot (e.g.Ocotea aciphylla and Ternstroemia brasiliensis, patterns C and D).Besides a signifi cant auto-correlation at some distances, eight species did not display an evident pattern of distribution (e.g.Calophyllum brasiliensis, pattern E).It is noteworthy that several Myrtaceae species (e.g.Myrcia ilheosensis, Myrcia multifl ora and Myrcia bicarinata) were the most strongly clumped species on the plot.

Forest diversity and composition
Th e mean richness obtained (79 species ha -1 ) was lower than other Brazilian sites of HRF (92 and 128 species ha -1 - Assis et al. 2004 andNegrelle 2006, respectively; note that both studies used non-co ntiguous plots).Even the studies with much smaller sampling eff orts attained similar or higher species richness (César & Monteiro 1995;Guedes et al. 2006).One possible explanation for this result may lie in the particularities of the PEIC plot substratum, which is young (< 5000 years), very acid, hydromorphic and periodically fl ooded (Gomes et al. 2007).Besides the little time that species had for colonization, such features impose important constraints on the establishment of some species (Dumont et al. 1990;Marques et al. 2009).Lower richness was also found when comparing trees ≥ 10 cm dbh of two nearby non-HRF sites (109 -Melo 2000; and ca.108 species ha -1 -R.A.F.Lima, unpublished data).
Th e number of rare species (i.e.singletons) per hectare represented 17% of the plot richness, a value quite similar to other HRFs (14 to 17%, Assis et al. 2004;Negrelle 2006) and lower than that observed values for forests on non-sandy soils (> 23% - Melo 2000, Melo et al. 2000;Guilherme et al. 2004;Losos 2004).In addition, Fisher´s alpha were much smaller than in 10 other tropical forest plots (values from 23 to 181) evaluated by Condit et al. (2004), except by the Mudumalai dry deciduous forest plot (5.9).Th erefore, lower richness, rarity and diversity may be characteristic to HRFs relative to forests on richer soils.However, studies with standardized sampling methods and in HRF on soils with diff erent features (age, acidity and/or fl ooding regime) are still needed to test this suggestion.
Arecaceae was the most abundant family, a common fi nding in many other Atlantic Forests.Exceptions are found where there is illegal palm-heart harvesting (César & Monteiro 1995;Guilherme et al. 2004;Guedes et al. 2006), an activity not seen inside the plot.Myrtaceae also deserves special mention because it harbored twice as many species as the second richest family of the plot.Th is pattern was not observed in other CTFS plots (Ashton et al. 2004), or inventories of forests on white-sand soils in South America (Phillips & Miller 2002) and Asia (Proctor et al. 1983;Newbery 1991;Davies & Becker 1996), where Myrtaceae does not exceed 7% of the total tree richness.However, nearby Atlantic Forest types have similar richness per family, with similar or higher Myrtaceae richness (Melo et al. 2000;Guilherme et al. 2004), suggesting an important eff ect of the geographic region on the general pattern of HRF fl oristic composition.
To the best of our knowledge, this is the largest survey ever carried out in a high restinga forest or in any other forest on white-sand soils.Magurran (2004) has suggested that sample suffi ciency is attained when the species ac-cumulation curve crosses the Michaelis-Menten model curve, which was observed around 4.4 ha.Th is also can be interpreted as the sampling area needed to confi dently estimate the richness observed in the plot.According to the fi rst and second order Jackknife estimators, these areas would be 3.4 and 2 ha, respectively.Th e suggestion of Colwell & Coddington (1994) that sample suffi ciency is attained when all species have at least two individuals proved to be impracticable because the number of singletons virtually leveled-off aft er about 6.5 ha.Th is indicates that the presence of naturally rare species (≤1 trees ha -1 ) in the PEIC plot is not merely a sampling artifact.Th erefore, the 10.24-ha plot was appropriate to describe the composition of the studied HRF, but it was insuffi cient to characterize the population structure of rare species.Anyhow, these results indicate the use of contiguous plots of at least 6 ha to properly assess species richness and composition of HRF on similar soils.
Forest structure -Th e PEIC plot had a density similar to other non-HRF Atlantic Forest sites (Melo 2000, R.A.F. Lima, unpublished data) but with less basal area.In addition, among the other 42 sites consulted (e.g.Philips et al. 1994;Losos et al. 2004), only four had basal areas smaller than the one found here, three of them also being found on whitesand soils.Th is result corroborates the connection between forest structure and soil properties suggested by Phillips et al. (1994) and Uhl & Murphy (1981).Th e pattern found here was also found by other authors when comparing whitesand forests to other types of tropical forests (Proctor et al. 1983;Davies & Becker 1996;Nebel et al. 2001;Miyamoto et al. 2003) and it is primarily due to the higher proportion of trees < 10 cm dbh and to the smaller proportion of large trees.In the PEIC plot the mean number of trees ≥ 60 cm dbh per hectare was six, whereas in other forest types it varied between eight and 33 trees (mean ± sd: 19 ± 7, Losos et al. 2004).Th ree factors are probably impeding species to attain large sizes: low soil fertility, higher light availability at ground level, which may induce individuals to cease vertical growth earlier, and lower potential size of Restinga species (Luizão et al. 2007;Scarano 2009).Future studies monitoring Restinga species growth controlling these three factors would be particularly useful to better understand the mechanisms giving rise to structure in such forests.
Th e considerable change in species rank order among size classes, including changes in the most abundant species in the sample, confi rms that the contribution of a given species also depends on the tree-diameter cutoff (Valencia et al. 2004;Kenfack et al. 2007).And these changes are not related exclusively to the presence of small-sized species in the understory.Diff erent from expected, almost one third of the species did not have a constant decrease in the number of individuals with an increase in diameter.Th is result may be a remnant of recent sucessional changes in the community.For instance, species that had peaks of density in larger size classes (e.g.Ternstroemia brasiliensis, Ocotea pulchella, Myrcia ilheosensis, Ilex theezans) are also abun-dant in forest borders or in younger successional stages of Restinga vegetation of PEIC (Sugiyama 1998).On the other hand, species that had densities more constant among size classes were oft en emergent tree species (e.g.Calophyllum brasilense, Albizia pedicellaris, Manilkara subsericea), which may have regeneration strategies diff erent from other species, including diff erent mortality and growth rates and/ or seed production strategies (Clark & Clark 1987;Swaine & Hall 1988).So, the successional stage of the community (i.e.age of the substratum) and the regeneration strategies of its species are perhaps the two main determinants of the vertical composition and structure of HRFs.

Multi-stemmed individuals
Th e occurrence of multi-stemmed individuals has been noted by diff erent authors as a common feature of Restinga species (Sugiyama 1998;Guedes et al. 2006) and as an important regeneration mechanism in Restinga forests (Assumpção & Nascimento 2000).Indeed, 74% of the species showed at least one multi-stemmed individual in the PEIC plot, a proportion that was higher than the one found for three other types of tropical forests of southeastern Brazil (ca.44 -68% -Lima et al., unpublished data).Nevertheless, the total number of multi-stemmed individuals was quite close to the numbers obtained for these forests, showing no evidence that this phenomenon is more common in the HRF tree individuals.Why there would be more species with multi-stemmed individuals in the HRF than in other comparable types of forest remains unclear.
Th ere was evidence that the average abundance of species with more multi-stemmed individuals was larger than other species, suggesting that this trait may infl uence species survival.For instance, a larger number of stems per individual may be benefi cial in terms of light interception effi ciency and survival aft er physical damage, an advantage which is well documented for species able to survive and resprout aft er being damaged (Paciorek et al. 2000;Martini et al. 2008).However, among the ten most abundant species in the plot (n > 400) only three had a greater number of stems than expected.Hence, the ability to produce multi-stemmed trees is not mandatory for successful establishment, at least in terms of density.In addition, resprouting aft er damage is not the only reason why individuals have more than one stem.Multi-stemmed individuals, for instance, may be linked to a higher availability of light that would stimulate a premature ramifi cation.In any case, our prediction is that species with a high proportion of multi-stemmed individuals would show a smaller number of stems per individual in shaded environments.And if the species also have a lower density of individuals, then we could start to infer about the competitive benefi ts of multi-stemmed individuals.

Spatial distribution
Although some studies indicate that species´ dispersion patterns are predominantly aggregated (e.g.Armesto et al. 1986), there is little information on multi-species patterns of dispersion at diff erent scales.Li et al. (2009) found that 92% of species were aggregated at scales <50 m, and Condit et al. (2000) found similar results for scales <30 m.Th ese scales correspond to our patterns B and C, but in the PEIC plot 25% of the most abundant species were aggregated at scales >70 m.Clumped distributions are generally the result of association to discrete habitats, patterns of seed dispersal, interspecifi c interactions (e.g.competition, facilitation), and/or expanding or contracting distributions (Hutchings 1997;Valencia et al. 2004;Law et al. 2009;Li et al. 2009), whereas large clumps can be related to environmental heterogeneity.In the PEIC plot, species distributed over large patches (patterns C and D) suggested species associated with habitat, probably soil.Species with distributions practically opposite in respect to particular places were relatively common and intriguing (e.g.Ocotea pulchella and Ternstroemia brasiliensis).Such places may indicate (un)favorable abiotic conditions where more stress-tolerants do well, or where more competitive species (e.g.ground bromeliads) may exclude others.Or it may indicate an association with resource availability (nutrients or light).It is noteworthy that similar distribution patterns can be generated from diff erent mechanisms (Harms et al. 2001).
In forests with restrictive soil conditions, such as the HRF, patterns of aggregation are particularly interesting.Higher aggregation in the larger size classes, for instance, would indicate a greater degree of habitat association (Condit et al. 2000).Examples of this result in the PEIC plot are Myrcia racemosa and Myrcia multifl ora.Th e opposite could indicate dispersal-limited species, such as Byrsonima ligustrifolia and Ilex theezans.Although the number of individuals in the larger size classes was limited to properly assess such processes, this approach should be extended to a larger number of species, perhaps including individuals < 4.8 cm dbh.Condit et al. (2000) suggested that changes in the patterns of aggregation among size classes are small (although the proportion of species with signifi cant values of aggregation presented by these authors was considerably lower for dbh ≥ 10 cm).Examples from other types of forests also suggest that the abiotic conditions are not able to explain the distribution of the majority of forest species (Harms et al. 2001;Valencia et al. 2004).Nevertheless, in the PEIC plot we believe that there is a higher degree of habitat association, particularly regarding the type of soil, fl ooding regime, and canopy gap distribution that combined generate very distinct habitats for tree establishment.

Concluding remarks
Th e structural features observed in the Atlantic HRF were very close to those found for other white-sand forests around the world.Common in South America and Asia, and known by several local names (e.g.heath forests, campinaranas, kerangas, varillal alto -Anderson 1981), white-sand forests develop on acid, sandy soils with low availability of nutrients (Proctor et al. 1983;Ashton 2004;Luizão et al. 2007).Th ese features act simultaneously to generate forests whose characteristics are shorter stature, lower species diversity, and roots more concentrated in the superfi cial soil horizons (Davies & Becker 1996;Miyamoto et al. 2003;2007).For these reasons, we recognize the Atlantic HRF as a type of white-sand forest.Although forest structure and probably functioning are similar among white-sand forests, species composition seems to be linked to local biogeographical issues.Th is is in accordance with the Walter´s biome classifi cation (Walter 1986), in which the HRF should be classifi ed as a Psammobiome I, i.e. a pedobiome within the Tropical Rain Forest Zonobiome, which is functionally and structurally determined by the sandy soil condition.Finally, we do believe that this similarity may be extended to other Neotropical forests over white-sand soils, such as the Brazilian muçunungas (Simonelli et al. 2008) or the 'river' HRF of Peru (Nebel et al. 2001).

Figure 2 .
Figure 2. Overall size distribution and examples of species with diff erent patterns of distribution in the PEIC plot.Dbh classes divided in 5 cm intervals.

Figure 3 .
Figure 3. Illustrative maps of the general distribution patterns suggested for the most abundant tree species in the 256 subplots (20×20 m) surveyed.Circles vary in size according to tree size.More dark colors indicate subplots with higher values of abundance, while the white color indicate subplots with zero individuals of a given tree species.Legend to patterns: A -no spatial autocorrelation (A); B -small isolated patches (B); C -more than one clump forming large stripes (C1) or larger patches more or less isolated (C2); D -big patches covering large areas of the plot (D); E -and no evident pattern (E).

Table 1 .
Total and per hectare values of the main structure and diversity parameters in diff erent size classes in a 10.24-ha plot of high restinga forest.n = total number of individuals; BA= Basal area; Singletons= species with only one individual in the sample.Brackets refer to the 2.5 and 97.5 percentiles of the distribution obtained from 144 one-hectare plots inside the 10.24-ha plot.

Table 2 .
Th e 40 most abundant species found in a 10.24-ha plot of high restinga forest and their correspondent abundance, aggregation (Morisita´s index of dispersion) and distribution pattern (DP).In parentheses, the species rank order in the size classes considered (species with the same abundance received the same rank to avoid overestimation of the real number of ranks per size class).See text for more detail on the distribution patterns of species.Legend to distribution patterns: A -no spatial autocorrelation; B -small isolated patches; C -more than one clump forming large stripes or larger patches more or less isolated; D -big patches covering large areas of the plot; E -no evident pattern.

Table 3 .
Number of individuals per category of number of stems for species with more or less multi-stemmed individuals than expected by chance.