Spatial patterns of tree community dynamics are detectable in a small ( 4 ha ) and disturbed fragment of the Brazilian Atlantic forest

(Spatial patterns of tree community dynamics are detectable in a small (4 ha) and disturbed fragment of the Brazilian Atlantic forest) The dynamics of the tree community of a fragment of tropical semideciduous forest was investigated in south-eastern Brazil. Surveys were carried out in 2000 and 2005 in 29 20 × 20 m plots. The main purpose was to assess the possibility of detecting spatial patterns of dynamics that could be related to the heterogeneity of both the environment and disturbance history in a small forest area (4.0 ha). Rates of mortality and recruitment of trees and gain and loss of basal area were obtained for the whole sample, four pre-defi ned soil habitats, diameter classes and twelve tree populations. The tree community was rather unstable in the period, once mortality rates surpassed recruitment rates and loss rates of basal area surpassed gain rates all-over the area denoting the prevalence of a degradation process, possibly triggered by a clear felling in 1985 that reduced the area of the fragment in 26%. The tree community dynamics showed no spatial autocorrelation but was not evenly distributed throughout the fragment. This spatial heterogeneity was chiefl y determined by the disturbance history of each site while environmental heterogeneity played a secondary but signifi cant role. The main causes of disturbance heterogeneity were the extension of the adjacent felled area, cattle trampling and selective logging. The environmental variables that most strongly correlated with the variations of the dynamics rates were those related to the availability of light, water and mineral nutrients. In contrast with the overall trends, three understory species expanded in the period, possibly at the expense of the steeply declining density of mid-sized trees.


Introduction
Processes of tree community dynamics in tropical forests frequently show a striking variation over small distances that produces spatial heterogeneity of both physiognomic structures and floristic composition (Whitmore 1989;Lieberman & Lieberman 1989;1994;Felfi li 1995;Guilherme et al. 2004;Appolinário et al. 2005).This heterogeneity results from a combination or interaction of several factors, among which variations of the physical environment and disturbance regimes are prominent (Martinez-Ramos 1985;Denslow 1987).A background source of variation is the substrate, which produces spatial and temporal heterogeneity of ground water and mineral nutrients availabilities (Pinto & Hay 2005).Disturbances cause an additional heterogeneity, modifying various aspects of the physical environment, notably the intensity and quality of light, temperature and humidity (Caswell & Cohen 1991;Burslem & Whitmore 1999).In fact, local disturbance history has been reported as one of the main forces driving the dynamics of tropical tree communities, and disturbances caused by tree falls, in particular, were postulated by some authors as a leading factor in maintaining the high species diversity of tropical forests (Hubbell 2001;Bell et al. 2006).Regeneration cycles triggered by tree-falls would result in a picture of tropical forests as ever-changing mosaics composed of patches at different successional stages and variable in both size and species composition (Richards 1979;Oldeman 1983;1989).
The approach to tree species distribution and diversity in tropical forests in articulation with environmental heterogeneity, in both space and time, is ultimately based on the classic view of plant community organization mediated by ecological niches.This perception has been seriously challenged since Hubbell (2001) proposed his Unified Neutral Theory of Biodiversity and Biogeography (UNTB) and the chain of stochastic events has gained increasing relevance in 'explaining' species distribution, dynamics and diversity in tropical forests leaving a minor role for species differentiation with respect to the physical environment.
At this critical historical age when human kind is seeking ways to conserve the vanishing biological diversity of the remaining tropical forests, it is of utmost importance to understand the mechanisms driving dynamics processes of tree communities and populations, particularly those confi ned in forest fragments.Large scale and long-term monitoring of permanent plots has been shown to yield promising results in the study of dynamic processes and related factors (Whitmore 1988;Condit 1995;Condit et al. 1996;Tomás 1996).The ongoing theoretical discussion, in particular, gives relevance to the assessment of the relative role played by both stochastic events and environmental variables as factors determining tree species distribution in space and time.In the present paper we analyze the shortterm (5 yr) dynamics of the tree community of a small (4 ha) and disturbed forest fragment in south-eastern Brazil with the purpose of seeking for spatial patterns related to environmental variables.The hypothesis was that, despite its small area and severe disturbance history, the spatial variation of tree community dynamics is affected by both substrate heterogeneity and disturbance distribution.

Materials and methods
Study area and habitat classifi cation -The study was carried out in a forest fragment known as Mata da Lagoa situated in Lavras, Minas Gerais state, south-eastern Brazil (21°13'11"S, 44°58'15"W) at 884 m of altitude.The fragment area was 4.0 ha during the study period (2000)(2001)(2002)(2003)(2004)(2005), but was approximately 5.4 ha in 1986 when a clear felling took place at its edges causing a 26% reduction of the area (Fig. 1).
Mata da Lagoa is a fragment of montane semi-deciduous seasonal forest according to the IBGE classifi cation of Brazilian vegetation (Veloso et al. 1991) and is also included in the wide-sense defi nition of Atlantic Forest (Oliveira-Filho & Fontes 2000).The fragment is surrounded by pastures and isolated from its nearest neighbor fragment by ca. 1 km.The drainage duct of a nearby reservoir fl ows close to southern edge of the fragment.Evident current impacts to the forest were those caused by selective logging and cattle trampling in the search for shade and shelter.
The climate of the region is classified as Köppen's Cwb type (mesothermic with wet summers and dry winters).The mean annual temperature and precipitation are 19.3 °C and 1493.2 mm respectively, with 66% of the annual rainfall concentrated between November and February (Vilela & Ramalho 1979).Following the Brazilian system of soil classifi cation (Embrapa 2000) the two occurring soil types are Red-Yellow Dystrophic Argisols (PVAd) and Red Eutroferric Nitosols (NVef).
As described in Machado et al. (2004), we classifi ed the 29 sample plots into four soil habitats based on the two soil types, Argisols and Nitosols, and three drainage classes: 'moderately drained', 'well drained' and 'highly drained', which were assigned to plots following the criteria proposed by Embrapa (2000).The resulting soil habitats (Fig. 1) were: Upper Argisols (well drained), n = 5; Lower Argisols (moderately drained), n = 9; Upper Nitosols (highly drained), n = 5; e Lower Nitosols (well drained), n = 10 plots.Because Machado et al. (2008) demonstrated that several tree community features differed signifi cantly among these soil habitats without spatial dependence, we used them to investigate whether dynamics rates differed accordingly.Tree community surveys -In the years 2000 and 2005 we carried out two tree surveys in 29 permanent plots with 20 × 20 m of dimensions totaling an area of 1.16 ha (29.1% of the fragment area).We distributed the plots systematically in eight parallel transects crossing the fragment from border to border up the main slope (Fig. 1).Transects were placed 30 m apart and plots along transects were spaced 10 m apart.As described in Machado et al. (2004), between April and May 2000 we carried out the fi rst survey of all live individual trees with a diameter at breast height (dbh) ≥ 5 cm, labeled them with numbered aluminum tags, identifi ed them to species, measured their circumference at breast height (cbh) and estimated their total height.Five years later, between April and May 2005, we accomplished the second tree survey during which we registered the dead, re-measured the survivors, and labeled, measured and identifi ed the recruits (trees that surpassed dbh ≥ 5 cm in the period).The following death categories were registered for dead trees: fallen, standing, felled and broken.Dynamics rates -In order to detect changes in species richness, we contrasted species gains and losses in the total sample and its four soil habitats by means of comparisons of Poisson counts (Zar 1996).We also compared the species richness of the two surveys in each soil habitat by means of species rarefaction curves with increasing sample size produced by the means and 95% confi dence limits of the number of species of all possible combinations of sample plots necessary to produce each sample size (Arrhenius 1921;McCune & Mefford 1999).Differences among species richness curves were assessed through their confi dence limits.
We calculated dynamics rates for tree assemblages (total sample, soil habitats, individual plots, diameter classes and tree populations) in the fi veyear interval (2000)(2001)(2002)(2003)(2004)(2005) for two aspects of tree dynamics: demography (tree counts) and basal area represented by the sum of sectional areas of individual trees at breast height.Based on Sheil et al. (1995Sheil et al. ( , 2000)), we assumed that changes in population size per unit time were a constant proportion to the initial population size and calculated the mean annual rates of tree mortality (M) and recruitment (R), and loss (L) and gain (G) of tree basal area as follows: where t is the time interval between surveys (5 yr), N 0 and N t are the initial and fi nal counts of individual trees, N m and N r are the number of dead and recruited trees, AB 0 and AB t are the initial and fi nal basal areas, AB m and AB d are the basal areas lost by tree death and decrement (tree decay and partial stem loss), and AB r and AB g are the basal areas gained by recruited trees and increment (growth) of surviving trees.
Additionally, in order to represent global dynamics, we obtained turnover rates from averaging mortality and recruitment rates as well as gain and loss rates (Oliveira-Filho et al. 1997;Werneck & Franceschinelli 2004).Following Korning and Balslev (1994), we also obtained net rates of change for both tree counts (Ch N ) and basal area (Ch AB ) using the expressions: x 100, and Ch AB = [(AB t /AB 0 )1/t -1] x 100.
In the analysis of tree dynamics per size class, we adopted the diameter classes of increasing amplitude proposed by Appolinário et al. (2005) to make up for the strong decrease in density in the larger size classes typical of the negative exponential distribution known as J-reverse.We then counted up the number of trees that experienced the following events between the two surveys: (a) remained in the class, (b) death, (c) recruitment, (d) ingrowth and (e) outgrowth.The latter two are inter-class migration events and can be either progressive or regressive (Lieberman et al. 1985).Based on the distribution of diameters in 2000, we used chisquared tests to assess whether the frequencies of both live trees in 2005 and dead trees were independent of the diameter classes.We used comparison of Poisson counts (Zar 1996) to assess the balance between the number of tree admissions (recruits + ingrowth) and exclusions (dead + emigrants) in each diameter class.
In order to seek for changing trends in the species abundance distribution we also obtained dynamics rates for the 12 species with 20 or more individuals in the two surveys, including the dead and recruited ones.We assessed changing trends in population demography through comparison of Poisson counts (Zar 1996) of recruited and dead trees.We classifi ed the species into the fi ve regeneration guilds proposed by Lieberman and Lieberman (1987) to help interpretation of community changes through the trends observed for particular populations (Gentry & Terborgh 1990).Environmental variables -Five environmental variables were chosen to investigate their correlation to dynamics rates: (a) drainage classes converted into an ordinal scale to express water availability; (b) soil pH, to express mineral nutrients availability; (c) leaf area index (LAI), to express light availability in the understory; (d) total tree basal area in 2000, to express saturation of standing biomass; and (e) edge factor, to express edge effects.
We calculated the 'edge factor' for each plot following Oliveira-Filho et al. (1997) to evaluate the infl uence of the proximity of the forest edge.The edge factor of a plot is the total length of one or more segments of the fragment perimeter enclosed by a circle of radius R whose origin is at the plot center.R should represent the maximum reach of edge effects on the trees of a forest fragment.We chose R = 100 m because it has proven effective in detecting edge effects in the tree community structure of six other fragments in the region (see Carvalho et al. 2007).
Following the procedures described by Pezzopane et al. (2002), we estimated the canopy cover of each plot through measures of LAI (leaf area index) obtained with the help of two LI-2050 sensors connected to LI-2000 data-loggers from LI-COR.We installed one sensor in a nearby open area and the other within the forest.We made all readings at 1.30 m from the ground and in overcast days, preferentially early in the morning or late in the afternoon, due to the diffuse light condition.We obtained a mean LAI value for each plot from readings made at four different positions.All readings took place in September 2004 coinciding with the peak of the leaf-fall season, i.e. maximum canopy openness.
Geostatistical treatment and correlation analyses -Because sampled events were not spatially randomized in the plots and, perhaps, not independently distributed, we applied a geostatistical analysis to incorporate an eventual dependence residual between plots.Spatial variables corresponding to coordinates x and y of the plot centers were used to defi ne a network of points.We then performed a preliminary exploratory analysis to observe the general spatial pattern of environmental variables and dynamics rates.The spatial dependency was analyzed by means of semivariogram adjustment using isotropic models to test the functions (Burrough and McKonnell, 1998).We found no structural function with a semivariance capable of being modeled in all variographic analyses indicating that there was no spatial dependency for all variables of both sets, dynamics and environmental.
Given the results of geostatistical treatments, we then analyzed all variables through classical methods of least squares parametric statistics, but only after performing Bartlett normality tests and applying the arc-sin √ x transformation to variables expressed by proportions (Zar 1996).We performed comparisons among the four soil habitats using analyses of variance coupled with Tukey-Kramer tests and assessed the relationship between dynamics rates and the five environmental variables using Pearson's correlation for soil pH, leaf area index, tree basal area and edge factor, and Spearman's (non-parametric) correlation for drainage classes because of its ordinal nature (Zar 1996).We adopted the signifi cance threshold of α < 0.05/5 (i.e.α < 0.001) to account for the Bonferroni correction (Miller 1991) for fi ve independent variables.Due to evident contrasting patterns related to size classes we split dynamics variables of tree demography into two pre-defi ned diameter classes: 5 ≤ dbh < 10 cm and dbh ≥ 10 cm.

Results
Changes in species richness were generally small in the whole sample as were in its four soil habitats (Tab.1).The balance between the numbers of lost and gained species was not signifi cant in all cases (total sample: Z = 1.41,Upper Argisols: Z = 0.0, Upper Nitosols: Z = 0.33, Lower Nitosols: Z = 1.13) but the Lower Argisols, where the number of species lost was signifi cantly greater than the number gained (13 : 1; Z = 3.21, p = 0.002).The same trends were confi rmed by the species accumulation curves (Fig. 2), which indicate a signifi cant species loss in the Lower Argisols and no signifi cant change in the other habitats.
There was a net decline of both tree density and biomass (expressed as basal area) in all soil habitats (Tab.1).Despite the overall decline in the number of trees due to a higher number of dead over recruited trees, this was signifi cantly unbalanced only for the whole sample (Z = 3.14; p < 0.002) and Lower Argisols (Z = 2.98; p < 0.005).The number of dead and recruited trees did not differ signifi cantly in the Upper Argisols (Z = 1.54; p > 0.1), Upper Nitosols (Z = 1.18; p > 0.1) and Lower Nitosols (Z = 0.76; p > 0.1).Thus the Lower Argisols habitat was probably the main responsible for the predominance of dead over recruited trees.In addition, the frequencies of surviving, dead and recruited trees were not independent of the four soils habitats (3 × 4 contingency * p <0,01; ns = non signifi cant table: χ 2 = 26.81,p < 0.001).Nevertheless, the frequencies of surviving trees in all soil habitats and of surviving, dead and recruited trees in the two Nitosols habitats did not differ signifi cantly from the total sample.Contrasts were therefore restricted to the frequencies of recruited and dead trees in the two Argisols habitats.The frequencies of recruited and dead trees were both higher than expected in the Upper Argisols (χ 2 = 8.52, p < 0.01) and lower than expected in the Lower Argisols (χ 2 = 12.89, p < 0.001).This suggests that tree community dynamics was more accelerated in the Upper Argisols and less in the Lower Argisols, with respect to the other habitats, as confi rmed by the comparison of dynamics rates.No signifi cant difference was found among soil habitats in their rates of tree mortality (F = 2.79, p > 0.06) and net decline in tree density (F = 0.60, p > 0.6), but signifi cant differences were found for recruitment (F = 4.64, p < 0.01) and turnover rates (F = 5.95, p <0.003), with higher values in the Upper than in the Lower Argisols, but no contrast with respect to the Nitosols habitats.Different patterns emerged in tree community dynamics expressed as tree basal area (Tab.1).No significant difference was found among soil habitats in their rates of net decline (F = 0.70, p > 0.5) and loss of basal area (F = 0.55, p > 0.6).Nevertheless, the habitats differed in their rates of gain in tree basal area (F = 0.70, p > 0.5), which were higher in the Upper Nitosols than in the Lower Nitosols and Lower Argisols, with Upper Argisols in an intermediate position and no signifi cant differences with the other three habitats.
Tree density distributions per diameter (dbh) classes did not change markedly between the two surveys in the four soil habitats except for the declining density of trees of the 10 ≤ dbh < 20 cm class in all habitats but the Upper Nitosols (Fig. 3).Diameter distributions in 2005 did not differ signifi cantly from expected frequencies based on 2000 (Tab.2) for the whole sample (χ 2 = 7.19, p > 0.06) as well as for its soil habitats (Upper Argisols: χ 2 = 4.73, p > 0.10; Lower Argisols: χ 2 = 2.56, p > 0.10; Upper Nitosols: χ 2 = 0.30, p > 0.10; Lower Nitosols: χ 2 = 2.18, p > 0.10).Despite this, the concentration of tree density reduction in the 10 ≤ dbh < 20 cm class was confi rmed by the imbalance between the number of inter-class immigrants (ingrowth and recruits) and emigrants (dead and outgrowth), which was signifi cant only for this dbh class in all soil habitats but the Upper Nitosols.
The proportion of dead trees differed signifi cantly among dbh classes in the Lower Argisols (χ 2 = 6.65, p < 0.05) and Lower Nitosols (χ 2 = 5.74, p < 0.10) but not in the Upper Argisols (χ 2 = 2.81, p > 0.10), Upper Nitosols (χ 2 = 0.28, p > 0.10) and the whole sample (χ 2 = 5.18, p > 0.10).Dead trees in the Lower Argisols were in smaller number than expected in the 5 ≤ dbh < 10 cm class and larger than expected in the 10 ≤ dbh < 20 cm class.In the Lower Nitosols dead trees were in larger number than expected in the 5 ≤ dbh < 10 cm class.Differences were not signifi cant for the other dbh classes of either habitat.Death mortality categories were not independent of soil habitats (Tab.3).Signifi cant differences were found for felled and broken trees, which occurred at higher numbers than expected in the Lower Nitosols.Four out of fi ve environmental variables were signifi cantly correlated with at least one of the six dynamics rates (Tab.4); the edge factor yielded no signifi cant correlation.Tree mortality rates in the 5 ≤ dbh < 10 cm class were positively correlated with soil drainage classes and negatively with leaf area index values.Mortality rates in the dbh ≥ 10 cm class were negatively correlated with initial tree basal areas.Tree recruitment rates were positively correlated with drainage classes and soil pH for trees of the 5 ≤ dbh < 10 cm and dbh ≥ 10 cm classes, respectively.Gain rates of tree basal area were negatively correlated with initial tree basal areas while loss rates of tree basal area were positively correlated with drainage classes.
Significant trends emerged from the analysis of population dynamics of the 12 most abundant tree species (Tab.5).Allophylus edulis, Mollinedia widgrenii and Galipea jasminiflora, all sub-canopy species were the only to show a significantly higher number of recruited over dead trees and are probably expanding in tree density.The former two also showed a net increase in basal area in the period.The opposite trend, i.e. significantly higher number of dead over recruited trees, was registered for five other species: Piptadenia gonoacantha, Sebastiania commersoniana, Machaerium stipitatum, Dendropanax cuneatum and Platycyamus regnellii.The four remaining species, Acacia polyphylla, Cupania vernalis, Eugenia acutata and Myrciaria floribunda showed no significant pattern.

Discussion
Sylvigenesis is the chain of structural changes that a certain forest area undergoes after a disturbance event disrupts tree community biomass and organization (Oldeman 1983;1989;Nascimento & Viana 1999).This regeneration process may be classifi ed into phases that are recognized by some tree community features, particularly size distribution, species composition and dynamic trends.Changes in tree demography and biomass, traditionally expressed by density and basal area, may stand for key diagnostic features for recognizing regeneration phases.We here propose a hypothetical model of sylvigenesis based on both aspects of tree community dynamics and on the manifold possible paths connecting the regeneration phases (Fig. 4).The Stability phase corresponds to the 'climax' or 'mature' stage of a forest area or eco-unit, where both density and basal area fl uctuate moderately under a low impact disturbance regime, i.e. without the formation of canopy gaps (e.g., branch-falls and death of sub-canopy trees).This phase may be disrupted by a moderate to severe disturbance event that opens the forest canopy and causes a cutback in both tree density and basal area, characterizing the Degradation phase.This phase can be either very brief, as happens after sudden multiple treefalls caused by wind storms, or last for longer time spans, as when canopy gaps are formed by the gradual crumbling of standing trees and when tree death escalates, because the area is reached by the permeating effects of a nearby disturbance (e.g., edge-effects).Once disturbance effects are ceased, forest building takes place through either of two paths.In the fi rst, an Early Building phase, with increasing tree density and basal area, ensues immediately.In the second path, this phase is preceded by an Early Building with Degradation phase during which the loss of biomass persists, caused by the death of larger trees, while a new generation of smaller trees is recruited (Oldeman 1983).The declining basal area is therefore accompanied by increasing density.The ensuing Late Building phase is characterized by the opposite trend, i.e. increasing basal area and decreasing density, the so-called 'self-thinning' process (Oliveira-Filho et al. 1997), when a few trees that survive the escalating competition accumulate biomass and grow bigger while many others die, most of which are small.As a result, the forest structure changes from a high density of small trees to a lower tree density with increasing concentration of biomass towards larger trees (the J-reverse distribution).Additional disturbance may intervene during any of the building phases and cause setbacks in the regeneration process and, in strict terms, tree community structure hardly ever attains a long-lasting stability phase (Swaine & Hall 1988).
From the perspective of the above model one can easily infer that small and disturbed man-made forest fragments such as Mata da Lagoa are quite unstable and degradation and building phases probably predominate all-over the area for a period of unforeseen duration.Man-made fragments of tropical forests undergo far-reaching and long-lasting changes in both species composition and physiognomic structure caused by both edge-effects and eco-geographical isolation (Noss & Csuti 1997;Pulliam & Dunning 1997;Laurance 2000;Laurance et al. 2002).As usual in other small forest fragments lodged within intensively farmed areas, Mata da Lagoa has been subjected    interval, is the strongest candidate among the disturbance forces causing generalized degradation.The impact of this felling probably accentuated edge-effects over the whole fragment, as the area is small enough to be totally made up of forest edges if we take into account that strong edgeeffects in tropical forests are detectable at distances of up to 200 m from the edges and two or more decades after edge creation (Laurance & Yensen 1991;Laurance 2000).Another evidence of the greater impact of the fragment reduction was the generalized concentration of decreasing tree density in the second diameter class (10 ≤ dbh < 20 cm).This probably refl ects a past disturbance event that affected mostly the seedlings and saplings in the understory and created a recruiting gap that smoothes down along time but also appears at increasingly larger size classes.Seedlings, saplings and smaller trees are actually the most affected immediately after edge creation (Viana et al. 1997;Laurance et al. 1998;2002).Other relevant and additional sources of disturbance were cattle trampling and selective logging but their effects were not evenly distributed, as evidenced by the higher density of cattle tracks in the two Upper soil habitats and higher proportion of felled trees in the Lower Nitosols.Nearness was probably the key-reason in both cases, because the Upper habitats are adjacent to the cattle-grazing pastures while the Lower Nitosols are more accessible for collectors of both wood and palm-heart of Syagrus romanzoffi ana.The higher frequency of broken trees in this site is probably due to the 'domino effect' (Crawford & Yaoung 1998) triggered by fall of large trees; half of tree deaths in the larger diameter class took place in the Lower Nitosols.Additional spatial heterogeneity was registered for other dynamics trends despite the generalized drop of both tree density and basal area in the Mata da Lagoa.Generally speaking, tree community dynamics was more rapid in the Upper habitats than in the Lower ones, though this was clearer for the Argisols.Spatial heterogeneity of disturbance may be once more evoked here because forest edge areas felled in 1985 were much larger near the Upper habitats.In addition, these habitats are more trampled by cattle.Disturbance impacts are therefore more severe in the Upper habitats and this probably gives rise to more unstable and quickly changing community than in the other habitats.
Despite all this, one can not disregard that substrate heterogeneity may also be involved as, for instance, water availability is lower in the Upper habitats and soil drainage classes were signifi cantly correlated with both mortality and recruitment rates of trees in the 5 ≤ dbh < 10 cm class as were gain rates of tree basal area.Thus, the lower was the water availability the higher were the rates of mortality and recruitment of smaller trees while larger ones (dbh ≥ 10 cm) were unaffected.Smaller trees probably have less access to deeper ground water stocks than larger ones, and this is particularly critical in Brazilian seasonal forests that experience a water shortage season (Furtini et al. 2003).As a consequence, the higher water shortage of the Upper habitats may have contributed to locally intensify the disturbance effects.In a similar way, the increasing mortality rates of smaller trees with increasing light availability probably refl ected the higher canopy openness of more disturbed sites, which may, in turn, have aggravated the water shortage during the dry seasons (Sterck & Bongers 2001).Water-and light-related factors would then be inextricably linked with disturbance at affecting the dynamics of smaller trees.If no correlation was found with the edge factor, this must have to do with the small size of the fragment that assigns the status of forest edge to the whole area.
The only substrate-related variable correlated with tree dynamics apparently with no connection with disturbance was the increasing recruitment rates of larger trees (dbh ≥ 10 cm) with increasing soil pH.The higher nutrient availability of Nitosols is probably contributing to the higher survival and growth rates of trees above a certain size.The dynamics of larger trees was also related to the initial basal area, which represents the standing tree biomass at the beginning of the study period and, putatively, the competition environment.Contrary to what was expected (Machado et al. 2008), the higher was the initial basal area of a site the lower were both the mortality rates of larger trees and the loss rates of tree basal area.In fact, as Degradation phases prevail all over the fragment, competition was certainly not the main force driving the dynamics of larger trees as expected in Building phases.The trend probably refl ects, once more, different levels of disturbance and sites with higher initial basal areas and lower loss and mortality rates simply represent the less advanced stages of forest degradation.
Tropical forests normally experience very low rates of change in species composition and fi ve years of observation is a very short time to reveal signifi cant species replacement for long-living organisms such as trees (Swaine & Lieberman 1987;Condit et al. 1992).As in the present case, studies of short-term dynamics of tropical forests generally record less than 10% of species disappearing and appearing between surveys and, obviously, among those represented by few individuals (Taylor et al. 1996;Kellman et al. 1998;Nascimento et al. 1999;Werneck et al. 2000;Pinto & Hay 2005).Despite all this, a signifi cant decrease in species richness was detected in the Lower Argisols and this was probably related to the lower recruitment rates at this habitat.The relevant fact here is that forest degradation processes caused by severe impacts may result in local species extinction in a very short time span.
Tree population changes of more abundant species are much easier to detect in short-term studies than changes in species richness.Five out of the 12 analyzed populations showed the same degradation trend of the whole community and four populations showed no signifi cant change.Both groups included species of all regeneration guilds.The only noteworthy trend was that the three species expanding in the period, contrasting with the overall degradation trend, were all sub-canopy dwellers.This trend is possibly related to the signifi cant drop in the density of trees of the 10 ≤ dbh < 20 cm class that could have opened spaces for small subcanopy trees.On the other hand, there is also the possibility that this trend is merely a result of selective logging of trees belonging to this size class because they are preferred for fi rewood, tool handles, props and poles.
The fi ndings of the present study showed that processes of tree community dynamics can be spatially heterogeneous even in small forest fragments where the same dynamics trend is observed all over the area.This spatial heterogeneity was chiefl y determined by the particular disturbance history of each site while environmental heterogeneity may have played a secondary but signifi cant role.Apart from this, as shown in similar studies (Oliveira-Filho et al. 2004;Appolinario et al. 2005) it was also clear, that the causes of variations in community dynamics in both space and time are often obscure particularly when the interacting factors involve past disturbance events which are diffi cult to recover with satisfactory precision and detail.

Figure 1 .
Figure 1.(A) Surface diagram showing the topography of Mata da Lagoa and the distribution of the 29 sample plots (20 × 20 m) and two soil types.(B) Map showing the area of Mata da Lagoa in 2000-2005 (inner polygon), the area cleared in the 1980's (dashed) and the classifi cation of sample plots according to four soil habitats: UA = Upper Argisols; LA = Lower Argisols; UN = Upper Nitosols; LN = Lower Nitosols.

Figure 2 .
Figure 2. Species accumulation curves with increasing number of 20 × 20 m sample plots in the four soil habitats and two surveys (2000 e 2005) carried out in Mata da Lagoa.Curves representing 95% confi dence limits are given only for the Lower Argisols, for clarity.

Figure 3 .
Figure 3. Distribution of tree density per diameter class in the four soil habitats and two surveys (2000 e 2005) carried out in Mata da Lagoa.

Figure 4 .
Figure 4. Flow diagram representing the interactions among hypothetical forest regeneration phases and their respective tree community dynamics trends expressed as changing demography (density) and biomass (basal area).

Table 1 .
Tree community dynamics in the Mata da Lagoa, between 2000 e 2005, given for the Whole sample and its four soil habitats.Values within parentheses are expected frequencies and those within brackets are the means of the N sample plots (20 × 20 m) of each soil habitat.p is the signifi cance of statistical tests (see text).Means followed by the same lower-case letters do not differ signifi cantly in Tukey-Kramer tests.

Table 2 .
Tree community dynamics per diameter class (dbh) in the Mata da Lagoa, between 2000 e 2005, given for the whole sample and its four soil habitats.Values within parentheses are expected frequencies of trees in 2005 and dead trees based on the frequencies of trees in 2000.Mort.= mortality rates; inter-class migrants: outgr.= outgrowth, recr.= recruits, ingr.= ingrowth.
to a great number of additional impacts, particularly cattle trampling, selective logging, burning of adjacent pastures, and forest clearing at the edges.Therefore, it is no surprise that Degradation phases prevailed all over the area between 2000 and 2005 with a generalized cutback of both tree density and basal area.The prevalence of Degradation phases entails that drawbacks promoted by disturbance impacts prevailed over tree regeneration.Nevertheless it is almost impossible to tell apart the effects of either past or current disturbances, i.e. occurring prior to or between the two surveys.The 25% reduction of the fragment area through forest felling performed in 1985, i.e. fi fteen years before the studied

Table 3 .
Contingency table with the frequencies of dead trees in each death category and soil habitat of the Mata da Lagoa.Values within parentheses are expected frequencies.

Table 4 .
Coeffi cients of correlation and their signifi cance (p <0.01) for the relationships between dynamics rates and fi ve environmental variables in N = 29 sample plots (20 × 20 m) used to survey the Mata da Lagoa.Coeffi cients are non-parametric (Spearman's) for correlations with drainage classes and parametric (Pearson's) for the others.Tree mortality and recruitment rates are split into two diameter (dbh) classes.

Table 5 .
Population dynamics for 12 tree species surveyed in 2000 and 2005 in the Mata da Lagoa, expressed as number of trees and basal areas.Regeneration (Regen. ).