An Ecological Comparison of Floristic Composition in Seasonal Semideciduous Forest in Southeast Brazil : Implications for Conservation

We examined floristic patterns of ten seasonal semideciduous forest sites in southeastern Brazil and conducted a central sampling of one hectare for each site, where we took samples and identified all individual living trees with DBH (diameter at breast height, 1.30 m) ≥ 4.8 cm. Arboreal flora totaled 242 species, 163 genera, and 58 families. Fabaceae (38 species) and Myrtaceae (20 species) were families with the largest number of species. Only Copaifera langsdorffii and Hymenaea courbaril occurred at all sites. Multivariate analysis (detrended correspondence analysis and cluster analysis) using two-way indicator species analysis (TWINSPAN) indicated the formation of a group containing seven fragments in which Siparuna guianensis was the indicator species. This analysis revealed that similarities between studied fragments were due mainly to the successional stage of the community.


Introduction
The extent of seasonal semideciduous forests (SSFs) in Brazil is underestimated because of its naturally fragmented distribution [1,2].Semideciduous seasonal forests occur along the contact zone between Atlantic forest and the diagonal of opened formations [3][4][5], comprising three different scenarios: (1) in northeastern Brazil, the SSFs form a marked belt (<50 km) in transition between the coastal rainforest and semiarid formations (Caatinga), but also occur in enclaves of montane forests, the altitude marsh [6,7]; (2) the transition between the Cerrado and the coastal Atlantic forest in southeastern Brazil involves an extensive occurrence of SSFs to the south, up to the east of Paraguay and northeast of Argentina, forming a complex mosaic with the Cerrado vegetation in the west; (3) in southeastern Brazil, a large Araucaria forest confronts subtropical coastal Atlantic forest, and the SSF appear to the west and south as a transition to the Chaco forests, and to the southeast with fields or the southern pampas [8,9], and beyond disjunct areas located in Mato Grosso and Tocantins states [10].
In southeast Brazil, SSFs are distributed widely in sites with a seasonal rainfall regime, characteristic of the Atlantic forest and Cerrado domains.In the Atlantic forest domain, SSF is the predominant typology, and in the Cerrado domain, SSF occurs in enclaves, associated with permanent or intermittent watercourses [12] and should be regarded as lato sensu Atlantic forest, since it presents a floristicstructural identity similar to forests of the Atlantic forest domain [12].
Semideciduous seasonal forests suffered the same degradation process as other Brazilian ecosystems.Since the 1970s, there was an accelerated replacement of natural vegetal formations to pasture and use for agriculture, transforming extensive areas into an important agriculture area for grain and fruit production and livestock [13].This rural reorganization was determined by the II National Development Plan (NDP), which collaborated for a modern agricultural deployment.With increasingly intense adoption of mechanization and land use, SSFs were drastically reduced, with only a few remaining in Southeast Brazil.Habitat fragmentation transforms the original landscape into different dynamic units that continually modify its structure [14].Furthermore, the occurrence of disturbance histories on different scales and heterogeneity environmental landscapes can influence the species composition of forest fragments [15] and form species richness patterns according to the fragment successional stage [16,17].
Therefore, owing to the similar process of disturbance in southeastern Brazilian forest formations and being located in the same catchment area, it is expected that the remaining SSF will have the same floristic pattern.
This work aims to analyze the floristic composition and regional richness in the arboreal species of ten fragments of SSF in southeastern Brazil, in order to answer the following questions: (1) what is the alpha diversity in sampled fragments?(2) is there a floristic pattern that represents the SSF in the studied region?(3) is the beta diversity in SSF similar to patterns found in other tropical forests?

Material and Methods
2.1.Floristic, Geographic, and Climate Data.We elaborate a floristic list from the tree species compilation in ten semideciduous seasonal forest fragments (according to Veloso classification [18]), distributed in five counties of southeastern Brazil (Table 1 1).
The studied region is part of a relief global set, named the "Domain of Central Brazil Tropical Plateaus" by Ab'Saber [19], and the "Plateaus and elevated plains of the Paraná sedimentary basin" by the RADAM project [20].The soils in the studied areas are predominantly red nonferric latossols (LV) [21].The predominant climate in the studied region is tropical savanna (Aw Megathermic), characterized, according to Köppen classification [22], by rainy summers and dry winters.The clearly seasonal climate has two well-defined seasons, where the winter season (April to September) has approximately six months of drought, and the summer season (October to March) is warm and rainy.The average annual temperature lies between 23 • C and 25 • C, with July the month with the lowest average temperature (16 • C).The annual rainfall ranges from 1160 to 1460 mm [23].

Data Collection and Analysis
. We conducted a central sampling of one hectare from each fragment of SSF, in an attempt to exclude edge effect and to obtain uniform samples, avoiding the ecotones of the adjacent formations.Central sampling gave priority to sites without the influence of watercourses (gallery forest), savanna formations (cerrado sensu stricto), and others forestry formations (dry seasonal forest and cerradão), since these vegetation contact areas vary considerably between fragments and would lead to changes in the species listing of each area.In each central sampling, all individual living trees with a DBH (diameter at breast height, 1.30 m) of ≥ 4.8 cm were recorded and identified.
We identified the species using literature, queries in herbarium, and specialists.For specific binomial formation, we employed the w3 Tropicos database [24].We deposited fertile material to the Uberlandia Federal University herbaria (Herbarium Uberlandensis: HUFU).Vegetative samples of all species were deposited at the Laboratory of Plant Ecology at the same university.Species were classified into families, according to the Angiosperm Phylogeny Group III system [25].
We compiled all species present in the inventories to undertake the floristic composition analysis.We carried out the analysis by preparing a binary database (presence/absence), containing the tree species listed in each fragment.We only considered individual trees identified at the species level in this database composition.For the Alpha diversity evaluation, we used the Shannon diversity index (H ) and the Pielou evenness index (J ).
We used the binary database for the measurement of richness analysis, and also for the total number of species projection, we used first-and second-order, nonparametric jackknife estimators, in addition to Chao and Chao 2 estimators, which were capable of projecting total species richness from the samples of species richness [26], with 5,000 randomizations.We conducted the analysis using the EstimateS 8.0 program [27].

Multivariate Analyses.
We did similarity and sorting analysis utilizing the FITOPAC SHELL 1.6.4program [28], using an absolute density matrix for the species of ten sites, considering only those species with two or more occurrences, since species that had just one occurrence did not contribute to the floristic site ordering for floristic similarity.For categorical floristic data (presence and absence), we analyzed the similarity between sites, using the same absolute density matrix, transformed into presence/absence.We used the Sørensen similarity coefficient [29] and the unweighted pair group method with arithmetic mean (UPGMA) for a dendrogram graphical representation.
We applied multivariate analysis to quantitative tree species floristic data (abundance) present in all the compared fragments.For this, we realized a data ordination by using the detrended correspondence analysis (DCA, [30]).In addition, to define predominantly the indicator and preferred species of the floristic groups, based on frequency and density (species with n ≥ 5 individuals), we used dichotomous hierarchical division using TWINSPAN (two-way indicator species analysis, [31]), from an absolute data matrix and its frequency at the 10 sites, with a cut level of 0, 2, 5, and 10.These analyses were made by PC-ord for Windows program version 4.0 [32].
Most well-represented genera are in Fabaceae Faboideae (13), Rubiaceae (13), Myrtaceae (10), Annonaceae (8), and Euphorbiaceae, Fabaceae Mimosoideae, and Malvaceae, with seven genera each.Approximately 30% of families were International Journal of Forestry Research represented by only one genus.The best represented genera in terms of species number were Aspidosperma, Machaerium, and Ficus, with seven species each, followed by Ocotea, Cordia, Inga, Trichilia, and Casearia with four species each.The total species richness, designed by jackknife estimators 1 and 2 and Chao 1 and 2 estimators, showed a similar pattern (273, 284, 285, and 268 species, resp.).Results suggest high regional tree species richness.The 242 species found approximates of richness designed for estimators, demonstrating sampling sufficiency among the ten hectares studied (Figure 2).The Shannon's diversity index (H ) for each site showed diversity values between 2.92 and 3.97, with the Pielou's evenness index (J ) between 0.73 and 0.87 (Table 2).However, when we considered the ten sites as a single sample, Shannon's diversity index (H ) was 4.62, and Pielou's evenness index was 0.84.This value can be attributed to the sampling covering quite heterogeneous sites, allowing differences in remnant structure and floristic composition and, consequently, increasing beta diversity.
The floristic similarity analysis between sites showed a similar pattern in seven of the ten sites examined (Figure 3).Accordingly, there was the formation of a floristically cohesive group (Group 1), where sites 3, 6, 7, 8, 9, and 10 were presented with values exceeding 0.5, indicating sites that were floristically similar.Also, a second group was formed with sites 1 and 4, whereas site 2 was floristically different from all the other sites (Figure 3).
Detrended correspondence analysis, DCA, presented eigenvalues of 0.61 and 0.25 in the two first ordination axes,  explaining about 85% of the total data variation (Figure 4).The diagram demonstrated a large group formed from seven sites.Structural ordination, on the left, contemplates the same sites forming Group 1 by grouping analysis.These seven sites indicate the floristic pattern representative of the studied SSF.The second DCA axis was associated strongly with the sampled fragments' conservation degree.The group formed on the center left is characterized by intermediate conservation stages, while other sites on the top right are characterized by more advanced conservation stages (Group 2).On the lower right is the initial stage (site 2).Therefore, the three fragments on the right do not form a cohesive group.In addition, these three sites are those that are more distant from each other.Thus, the structural floristic similarity of the seven areas above may have been established on a geographical position basis.The division made by TWINSPAN corroborated the ordination results by DCA and the Sørensen coefficient (Figure 3).The first division by TWINSPAN separated the seven SSF representative sites in this region from the three other fragments, owing to the absence of Siparuna guianensis, a species regarded as an indicator, which only occurs in the group formed on the other side of the dichotomy (Figure 5).In the second division, sites 3 and 8 were separated from the other fragments (Figure 5).The lower levels of division by TWINSPAN do not reveal new groups that make sense.

Discussion
Studies conducted in areas of SSF in Brazil presented Fabaceae, Myrtaceae, Rubiaceae, Annonaceae, Lauraceae, and Meliaceae as the families with greater species richness [33][34][35][36].In São Paulo state SSF, as well as the greater richness of families found in this study, there is also an emphasis on Melastomataceae and Solanaceae [37,38].In northeastern Brazil, the most prominent families in terms of species number in SSF sites are Fabaceae Mimosoideae, Euphorbiaceae, Fabaceae Faboideae, Myrtaceae, and Rubiaceae [39].The pattern in families and genera richness found here corroborates that described by Oliveira Filho and Fontes [40] for SSF domains in southeastern Brazil, with significant tree species richness in Fabaceae (all subfamilies), Lauraceae, Myrtaceae, Moraceae, and Rubiaceae and genres Aspidosperma, Ficus, Machaerium, and Ocotea.
Copaifera langsdorffii and H. courbaril are wide-ranging adaptive species, occurring in various forest formations of the Cerrado region, as well as in other geographical provinces in Brazil and are, therefore, considered to be habitat generalists [41].Apuleia leiocarpa and Luehea grandiflora are also among the most frequent species of SSF, according to a study presented by Ferreira J únior et al. [42], which brought together 15 studies in southeastern Brazil.The tree species composition studied in SSF showed a relationship with lowland seasonal forests of Brazil's eastern Minas Gerais state, based on the indicator species analysis made by Oliveira-Filho and Fontes [40].Of the 57 species identified by these authors as eastern Minas Gerais region indicators, 36 (63.1%) were recorded in the sample sites presented here.This data corroborates the idea that interior SSF fragments may have a floristic connection with the Atlantic forest that is able to expand its distribution in sites with strong seasonality by means of gallery forests [43].
The presence of unique species to each location highlights the heterogeneity between samples (sites), reflected in increased beta diversity.We should point out that the limited number of individuals in some populations may impair the biological conservation of many species, causing serious difficulties for their preservation.However, some species, such as Galipea jasminiflora and Micrandra elata, occurred in only one fragment, but with a high density.This observation acquires relevance in terms of plant population conservation.The conservation of the remaining forest, with different structures and consequent higher beta diversity, can be a relevant parameter to the decision for choosing priority areas for conservation in SSF [44].
Diversity values are among those frequently found for SSF and Atlantic rainforest, which generally vary between 3.2 and 4.2 [45].However, they are lower than those found in other tropical forests [46].The lower values for evenness found in fragments indicate a relatively high concentration in density of a small number of species, which dominate the tree community.The predominance of a few species, in number or biomass in a community (also known as ecological dominance), is not uncommon in tropical forests [47].
A variation in similarity values occurs as a result of different conditions in the use and historic occupation of each fragment and acts as an important force capable of modifying plant communities through spatial and temporal heterogeneity, thus determining the community composition and structure [16,48].The separation of site 2 from other areas, along with Group 2, can be related to different stages of forest succession.Site 2 presents in an initial successional stage owing to an intense human processes.This fragment, although situated in the countryside, reveals recent historical disturbance as a result of selective logging and the presence of cattle, and the surrounding matrix is formed entirely by agricultural sites [49].Selective logging enables the creation of gaps within the fragment and the establishment of pioneering species.As Whitmore states [47,50], pioneer trees and shrubs need high light and temperature intensities for seed germination and seedling establishment and growth.Group 2 (sites 1 and 4) is formed by the more well-preserved fragments among those studied and is characterized by the presence and dominance of individuals with a large basal area.In fragment 4, for example, Micrandra elata was sampled with a density of 118 ind•ha −1 and a basal area of 24.51 m 2 •ha −1 [51], whereas in fragment 1, Psidium sartorianum was the species with the highest dominance [52].
Although the floristic similarity between the fragments can be related to the geographical proximity, as already reported by MaCdonald et al. [53], we verified that similarity relations between studied sites are established mainly as a result of the community successional stage.This same pattern was also found by Durigan et al. [17]; that is, within the same vegetal formation, communities at similar stages of the successional process tend to have similar flora.
This analysis showed high beta diversity of the studied SSF arboreal flora, mainly as a result of the coverage of heterogeneous areas in relation to fragments at a successional stage.The largest SSF sampling in the Cerrado domain led to the registration and distribution of several plant populations, which had previously only been described for the Atlantic forests domain, demonstrating not only a lack of studies  like this, but also corroborating the idea of strong floristic connections with rainforests in the east of the country [40].
The fragmentation process started in the 1970s in the studied region of Brazil, especially as a result of the incentive given to landowners by the public authority for the advancement of occupied areas for agricultural and livestock activities, which had a direct consequence for the formation of many smaller forest fragments.Fragmentation possibly created similar processes in forests' successional development, which has conditioned a floristic pattern as found for seven of the ten sites surveyed.

Figure 1 :
Figure 1: Location of the ten sites of seasonal semideciduous forest studied in the region of Triângulo Mineiro, Minas Gerais state, southeastern Brazil.

Figure 2 :
Figure 2: Sampling sufficiency designed by "jackknife" estimator for the ten sites of seasonal semideciduous forest in southeastern Brazil.Dashed line: standard deviation.

Figure 3 :
Figure 3: Dendrogram of similarity (Sørensen coefficient) produced by cluster analysis (UPGMA connection method) of tree species composition among the ten sites of seasonal semideciduous forest in southeastern Brazil.

Figure 5 :
Figure 5: Classification by TWINSPAN of ten sites of seasonal semideciduous forest in southeastern Brazil.

Table 1 :
Information of ten seasonal semideciduous forest fragments in southeastern Brazil analyzed regarding regional tree flora composition.

Table 2 :
Number of families and species, Shannon diversity index (H ), and Pielou evenness index (J ) from ten sites of seasonal semideciduous forest in southeastern Brazil.

Table 3 :
[11]ies list sampled in ten sites of seasonal semideciduous forest of southeastern Brazil.NI = number of individuals; RF = frequency in the ten studied fragments; * First record for the Triangulo Mineiro region, * * First record for the Minas Gerais state in seasonal semideciduous forests, according[11].