Disturbance-Mediated Drift in Tree Functional Groups in Amazonian Forest Fragments

BIOTROPICA *(*): ***–*** **** 10.1111/j.1744-7429.2007.00318.x

Disturbance-Mediated Drift in Tree Functional Groups in Amazonian Forest Fragments

Fernanda Michalski1,2,4, Ivone Nishi3, and Carlos A. Peres1

1Centre for Ecology, Evolution and Conservation, School of Environmental Sciences, University of East Anglia, Norwich NR4 7TJ, UK

2Instituto Pro-Carnıvoros, C.P. 10, Atibaia, SP-12940–970, Brazil

3Rua Palmeira, 318, Setor H, Cx.P. 91, Alta Floresta, MT-78580–000, Brazil

ABSTRACT

The interactive effects of forest disturbance and fragmentation on tropical tree assemblages remain poorly understood. We examined the effects of forest patch andlandscape metrics, and levels of forest disturbance on the patterns of floristic composition and abundance of tree functional groups within 21 forest fragments and twocontinuous forest sites in southern Brazilian Amazonia. Trees were sampled within 60 (10 × 250 m) plots placed in the core areas of the fragments. Tree assemblagecomposition and abundance were summarized using nonmetric multidimensional scaling (NMDS). Forest patch size explained 36.2 percent and 30 percent of thevariance in the proportion of small-seeded softwood and hardwood stems in the 21 forest patches, respectively. Large fragments retained a higher abundance ofhardwood tree species whereas small-seeded softwood trees appear to proliferate rapidly in small disturbed fragments. Generalized linear mixed models showed thattime since fragmentation had both positive and negative effects on the density of different functional groups of trees and on the ordination axes describing treeabundance. The composition and abundance of different tree genera were also related to time since fragmentation, distance to the nearest edge, and fire severity,despite the recent post-isolation history of the forest patches surveyed. Both the proliferation of fast-growing pioneer trees and the decline of hardwood trees found inour forest plots have profound consequences on the floristic composition, forest dynamics, carbon storage, and nutrient cycling in Amazonian forest fragments.

Abstract in Portuguese is available at http://www.blackwell-synergy.com/loi/btp.

Key words: floristic composition; fragmentation; pioneers; tree communities.

THE AMAZON BASIN CONTAINS OVER HALF OF THE WORLD’S RE-MAINING TROPICAL RAIN FORESTS, including some of the richestplant communities anywhere. The arborescent flora (≥ 10 cm instem diameter) of a single hectare often includes 200–300 speciesor more (Gentry 1982, ter Steege et al. 2000, Valencia et al. 2004).Lowland Amazonia is also experiencing the world’s fastest absoluterate of deforestation (Whitmore 1997, Laurance et al. 2001); 2.6million ha/yr in Brazilian Amazonia in 2003–2004 (INPE 2004).Since the 1970s, large-scale deforestation has been concentratedin the eastern and southern parts of Amazonia (Skole & Tucker1993, INPE 2004) often generating a highly fragmented forestlandscape containing thousands of forest remnants experiencing avariable history of internal disturbance (Peres & Michalski 2006).The expansion of the cattle and soybean industries has increaseddeforestation rates, and agricultural expansion will likely eliminatesome 40 percent of Amazonian forests by 2050 (Soares-Filho et al.2006).

Fragmentation alters forest dynamics, reducing live treebiomass and increasing tree mortality and turnover and atmo-spheric carbon emissions (Laurance et al. 1997, 2000; Nascimento& Laurance 2004). Because most tropical forest tree species have lowpopulation densities throughout their geographic ranges (Gentry1990, Hubbell & Foster 1996, Pitman et al. 1999), high tree mor-tality rates in forest fragments could result in the loss of many rare

Received 3 July 2006; revision accepted 28 December 2006.4 Corresponding author; Current address: Av. Mariland 1367/1001, Porto Ale-gre, RS-90440–191, Brazil. e-mail: [email protected]

species. Seedling abundance and survival in forest fragments may bereduced in several ways because of detrimental microclimatic con-ditions (Bruna 1999, Benıtez-Malvido & Martinez-Ramos 2003),changes in plant–animal interactions such as reduced animal dis-persal services (da Silva & Tabarelli 2000, Cordeiro & Howe 2003),and elevated seed predation (Terborgh et al. 2001).

Most forest fragmentation studies examining tree communitieshave focused on the impact of edge creation and vegetation changesalong the edges of forest fragments (Malcolm 1994, Fox et al. 1997,Laurance et al. 1998, Sizer & Tanner 1999, Hooper et al. 2004).However, fragmentation effects may be strong enough to promotelocal-to-regional scale extinctions events, and there is sufficient ev-idence to indicate that fragmentation does not occur alone, but isoften associated with other patterns of natural and anthropogenicforest disturbance (Tabarelli et al. 2004, Peres & Michalski 2006),which can aggravate threats to tree populations.

The southern Brazilian Amazon was subjected to massive de-forestation rates during the 1970s and 1980s following several large,central-government sponsored agricultural resettlement programssubsidized by generous fiscal incentives (Hecht 1993). Forest lossalong this section of the Amazonian “Arc of Deforestation” createdseveral types of landscape structure, ranging from a typical small-holder fish-bone pattern to those dominated by sizeable remnantswithin large cattle ranches (Oliveira-Filho & Metzger 2006). Theseforest remnants are widely variable in size, degree of connectivityand level of human disturbance, and have since been regularly sub-jected to timber extraction, fires, and hunting (Peres & Michalski2006), all of which can greatly compound the effects of habitat

C© 2007 The Author(s)Journal compilation C© 2007 by The Association for Tropical Biology and Conservation

1

2 Michalski, Nishi, and Peres

fragmentation (Nepstad et al. 1999, Peres 2001, Cochrane et al.2002).

Mechanized logging operations often exacerbate the effects offorest fragmentation by desiccating forests near roads, tracks, andlogging gaps (Verıssimo et al. 1995, 1998; Uhl et al. 1997). More-over, logging reduces canopy cover and produces large amountsof litterfall and other dead phytomass. This fuel material ren-ders fragments more susceptible to human-induced forest burning(Cochrane et al. 1999, Nepstad et al. 1999), which is often triggeredby ignition sources in adjacent cattle pastures, slash-and-burn plots,and commercial crops (Holdsworth & Uhl 1997, Gascon et al.2000, Barlow & Peres 2004). Forest burning depletes the soil seed-bank (Cochrane & Schulze 1999) and increases rates of seedling,sapling, and adult mortality due to lethal thermal stress and com-petition with lianas, vines, and early successional species (Kauffman1991, Holdsworth & Uhl 1997, Peres 1999, Perez-Salicrup 2001).Hunting can drive key populations of vertebrate seed dispersers tolocal extinction (Peres 2000) and consequently exacerbate dispersallimitation (Asquith et al. 1999, da Silva & Tabarelli 2000, Peres &Roosmalen 2002, Wright 2003).

The interactions among forest fragmentation, logging, andwildfires can have profound consequences on the taxonomic andfunctional composition of tree assemblages in forest fragments, yettheir underlying causes remain poorly understood. Here we examinethe combined effects of forest fragmentation and forest degradationon the functional composition of woody plants in a highly frag-mented forest landscape of southern Brazilian Amazonia. First, wedescribe the variation in the number of genera and density usingstandard species-area and density-area relationships. We then con-sider fragmentation effects on the structure of tree assemblages byexamining how these may interact with levels of human perturba-tion including logging and wildfires. We also examine how densitiesof small-seeded softwood and hardwood stems have responded toforest patch size and anthropogenic disturbances. Finally, we makesome general points about how patch and landscape scale habitatmetrics can explain patterns of floristic composition and regenera-tion in a heavily fragmented tropical forest landscape.

METHODS

STUDY AREA.—This study was conducted in the region of Alta Flo-resta, located in the northern Mato Grosso, southern Brazilian Ama-zonia (09◦53′ S, 56◦28′ W). The study area spans 4100 km2 (Fig. 1)at 200–300 m elevation (RADAMBRASIL 1983). Soils in the studyarea are predominantly Ultisols (Brazilian nomenclature: Podzolicovermelho-amarelo distrofico) followed by oxisols (Brazilian nomen-clature: Latossolo vermelho alico) (RADAMBRASIL 1983), whichare the predominant soil types in Brazilian Amazonia (Moraes et al.1995).

The mean annual rainfall is 2350 mm, and the evapotranspi-ration is ∼1000 mm/yr, providing a 1350–1400 mm/yr surplus,except for the dry season (May–August), which typically results ina hydrological deficit of 250–300 mm (RADAMBRASIL 1983).

Forests in the study area are classified as open tropical submon-tane rain forest, with lianas (RADAMBRASIL 1983), and the for-est canopy often includes emergents such as Apuleia moralis, Cae-salpiniaceae (Amarelao), Bagassa guianensis, Moraceae (Tatajuba),Astronium gracilis, Anacardiaceae (Aroeira), and Bertholletia excelsa,Lecythidaceae (Brazil nut tree). A Landsat time series shows that thisonce entirely forested region has been subjected to high deforesta-tion rates since the early 1980s. As of 2004, only 32 percent of theoriginal forest cover remained in the Alta Floresta region south ofthe Teles Pires river. The resulting fragmented landscape containsforest patches of varying size, degree of connectivity, and distur-bance regimes surrounded by a matrix of managed cattle pastures.All 21 forest patches sampled (mean ± SD patch size = 1416 ±3834 ha, range = 2.0–14,480 ha) were entirely isolated from, orthinly connected to, neighboring forest areas. These were comparedwith two continuous undisturbed primary forest “control” sites lo-cated on either bank of the Teles Pires river (Fig. 1). All sites werelocated within 50 km of Alta Floresta and were accessible by riverand roads.

FLORISTIC INVENTORIES.—Forest patches were initially selected us-ing a 2001 Landsat image (scene 227/67) on the basis of their sizeand degree of isolation. To ensure homogeneity in any possible ef-fect of the surrounding matrix on tree species composition (e.g., treerecruitment: Nascimento et al. 2006), all selected fragments weresurrounded by actively managed pasture grazed by cattle. Duringa pilot study in 2002, we inspected all target fragments and inter-viewed their landowners, to ensure that they had not experiencedany habitat loss, logging, and wildfires for at least 1 yr prior tofloristic inventories. A total of 60 quarter-hectare (10 × 250 m)forest plots were then inventoried in June–December 2003 andJune–December 2004. Fifty-two of these plots were located withinvariable-sized fragments as follows: six plots in six fragments < 10 ha(mean ± SD forest patch size = 5.0 ± 1.8 ha, range = 2.4–7.3 ha),ten plots in five fragments of 14–26 ha (20.5 ± 4.6 ha, range =14.9–25.7 ha), 12 plots in four fragments of 86–142 ha (108.1 ±21.3 ha, range = 86.9–141.3 ha), 16 plots in four fragmentsof 915–1800 ha (915.5 ± 572.9 ha, range = 211.7–1763 ha),and eight plots in two fragments of 12,800–14,500 ha (12,758 ±1842 ha, range = 11,035–14,481 ha). The remaining eight plotswere sampled in the two continuous forest sites. To minimize edgeeffects, plots were placed at least 200 m from the nearest for-est border, but this was only possible in medium and large frag-ments and continuous forest sites. We identified to at least genuslevel all trees ≥ 10 cm dbh (diameter at breast height or abovethe tallest buttress) that lay within each 0.25-ha plot, excludingthose with more than half of their basal trunk falling outside theplot.

Trees were defined as dead if they were both leafless and ex-hibited a complete ring of dead cambium at breast height. The dbhof a total of 1358 dead and 6890 live trees contained within the60 quarter-hectare plots was measured. A total of 6807 (98.8%)live trees were identified to at least genus level. Trees were identi-fied using either sterile or fertile material collected from each treeand by double-checking their characters and local names (provided

Disturbance-Mediated Drift in Amazonian Trees 3

FIGURE 1. Location of the study region in Alta Floresta, northern Mato Grosso, Brazil, showing the 21 surveyed forest patches (solid areas) and two continuous,

undisturbed forest sites used as “controls” (solid circles). Forest patches < 30 ha are circled. Gray and white areas on either bank of the Teles Pires river represent forest

and nonforest cover, respectively.

by one of us [IN], a local botanist with > 15 yr of experience infloristic inventories in the region, and a local herbarium technicianfrom the local university [UNEMAT] with > 25 yr of experiencewith the local flora), and compared with those in tree field guides(Parrotta et al. 1995, Lorenzi 1998, Ribeiro et al. 1999, Lorenzi2000). We tested our technician’s tree identification skills using arandom set of 120 trees ≥ 10 cm dbh contained in a 1-ha tree plot,which had been independently number-tagged and identified onthe basis of complete vouchers deposited at the Federal Universityof Acre herbarium (Santos 2005). This showed that he correctlyidentified 100 percent of these trees to genus level, which givesus confidence that unvouchered identifications are reliable. Taxo-nomic resolution at the genus level was considered to be sufficientlyrobust for the purposes of this study because Amazonian tree con-geners are similar in many life history traits (Casper et al. 1992, terSteege & Hammond 2001, Chave et al., in press) and genus-levelidentification captures 80 percent of the taxonomic differentiationof Amazonian trees at the species level (Higgins & Ruokolainen2004). Furthermore, most tree genera in our study area are repre-sented by a single species; for example, only 12 of 57 tree generaoccurring in a completely herbarium-vouchered 1-ha floristic plotin Alta Floresta (containing 433 trees ≥ 10 cm dbh belonging to 74species and 31 families: Santos 2005) were represented by more thanone species. Voucher specimens for each genus were deposited in

our Fragmentation Dynamics Project plant collection, Alta Floresta,Brazil.

Because our focus here is on floristic composition, only datafrom live trees are analyzed. We also excluded from our analysis 1.2percent (N = 83) of the trees that we were unable to reliably identifyto genus. We assigned three functional types to 99.4 percent (N =6767) of all identified trees based on multiple forest inventoriesconducted throughout Amazonia (Henderson et al. 1995, FAO1997, Fearnside 1997, ter Steege & Hammond 2001, Baker et al.2004, Block 2004, FAO 2004). This included life history traitssuch as wood density (oven-dried weight divided by green volume)and seed mass that are important determinants of tree successionalstatus according to the following classification: (1) small-seededsoftwoods (SSWs): wood density < 0.7 g/cm3, seed mass < 0.1g; (2) large-seeded softwoods (LSWs): wood density (< 0.7 g/cm3,seed mass > 0.1 g; and (3) hardwoods (HWDs): wood density > 0.7g/cm3 (ter Steege & Hammond 2001, ter Steege et al. 2002). Thisfunctional classification, which is less subjective than the traditionalsubdivision into short-lived and long-lived pioneers, is based on thefacts that (1) all pioneers have an air-dry wood density below 700kg/m3 and (2) among the pioneers, the short-lived pioneers have thesmallest seeds (ter Steege 2000, ter Steege et al. 2002). Moreover,there is sufficient evidence indicating that seed mass variation withinNeotropical forest tree genera is lower than between genera (van

4 Michalski, Nishi, and Peres

Roosmalen 1985; Lorenzi 1998, 2000) and the use of only twoclasses minimizes any possibility of misclassification. For example,seed mass and wood density assigned to the level of genus has beenapplied to basin-wide studies across Amazonia (ter Steege et al.2006).

Evidence of selective logging activity (i.e., stumps) were foundwithin only 21 of the 60 forest plots surveyed, and the number ofstumps per plot ranged from 0 (0.61%) to 8 (6.56%) (Mean ±SD = 0.68 ± 1.44%, N = 60) of the total number of live anddead trees occurring within each plot. Continuous forest controlplots (N = 8) showed virtually no selective logging, with one stumpaccounting for a mean of only 0.081 percent (SD = 0.23) of thetotal number of live and dead trees sampled. Our compositionaldata were therefore not meaningfully affected by the within-plothistory of logging activity.

FOREST PATCH AND LANDSCAPE DATA.—Landscape variables were ex-tracted from a Landsat image (12/06/2004) using Fragstats(c) v. 3.3(McGarigal & Marks 1995) and ArcView 3.2. Following an unsu-pervised classification, eight land-cover classes were resolved (closed-canopy forest, open-canopy forest, disturbed forest, highly disturbedforest, managed pasture, unmanaged pasture, bare ground, and openwater). The image was registered and geo-referenced with a posi-tional error of < 10 m.

Five of the 21 forest patches surveyed were not completelyisolated, so to calculate their metrics, we artificially eroded thenarrowest, most disturbed forest cover connections (mean width± SE = 55.6 ± 8.2 m, N = 12), following Michalski and Peres(2005). For each forest patch, we measured size (based on all foresttypes) and for control sites, embedded within continuous forests,we assigned an arbitrary forest area of 144,805 ha, equivalent to oneorder magnitude larger than our largest fragment.

Time since fragmentation was obtained from 11 biannualLandsat images (1984–2004). This is defined as the time lag (yr)since the surrounding open-habitat matrix was initially formed byextensive clear-cutting of adjacent forest areas and then stabilizedover time so that further retractions in fragment size had been dis-continued. This variable essentially describes the postisolation ageof forest patches since the last major reduction in patch size, rangingfrom 1 to 20 yr (mean ± SD = 7.85 ± 6.02 yr).

Distance to nearest edge was defined as the mean distance (m)between the nearest forest edge and five points located every 50 malong the 10 × 250 m floristic plots. The intensity and extent ofselective logging and fire disturbance were quantified within eachforest site. As a part of an independent line-transect census samplingprogram (F. Michalski & C. Peres, pers. obs.), 3–6 transects of upto 5 km in length were cut across each of the forest fragments andcontinuous forest sites. Logging intensity was measured accordingto a rank variable (1–5) obtained during interviews based on thespecies profile, scale, and method of timber extraction; and threevariables obtained in situ: (1) number of logging and skid trails(created to extract timber) per kilometer of transect; (2) number ofstumps detected per kilometer of transect; and (3) total stump basalarea of logged trees for each plot. Severity of fire disturbance wasmeasured according to: (1) a rank variable (0–6) obtained during

interviews based on the history and extent of the burned area; (2)a rank variable (0–3) obtained during interviews of recurrent fires(number of ignition events); (3) fire penetration distance, or themaximum linear distance obtained in situ between forest edges andany evidence of fire (e.g., charred tree trunks); and (4) total areaburned, which was calculated in Fragstats(c) (McGarigal & Marks1995) on the basis of the fire penetration distance.

DATA ANALYSIS.—All 130 tree genera identified in our forest plotswere included in the analyses. Considering all possible pairwisecomparisons between any two 0.25-ha forest plots (N = 1770), wecalculated the Chao–Sørensen abundance-based estimator basedon the abundance of tree genera in each plot using ESTIMATES

(Colwell 2004). This estimator is considerably less biased than otherindices when a substantial proportion of species are missing frommost samples (Chao et al. 2005). This matrix was then contrastedwith the between-plot straight-line distance matrix, considering thegeometric centre of each plot, to test for spatial autocorrelation inthe tree community data using a Mantel test statistic (Mantel 1967)based on 5000 randomizations.

We performed a nonmetric multidimensional scaling (NMDS)ordination based on: (1) the floristic composition (Chao–Jaccardabundance-based estimator); and (2) the abundance (Chao–Sørensen abundance-based estimator) of all 60 forest plots. TheNMDS was restarted 50 times and undertaken using Primer v. 5(Clarke 1993).

To assess relationships between log10-transformed forest patcharea (ha) and mean (± SE) of density of trees, number of tree genera,and percentage of small-seeded softwoods, large-seeded softwoods,and hardwood stems, we performed linear regression models con-sidering only 52 plots in 21 forest fragments and excluding the“control” sites. The R2 value that we report is always the adjustedR2.

For each intercorrelated set of disturbance variables, we con-ducted a Principal Component Analysis (PCA) to derive a fewcomposite variables that retained most of the information in theoriginal data set. For the logging disturbance, the PCA axes summa-rized information on logging intensity (ranked based on interviews),number of logging roads and number of cut stumps per kilometerof transects sampled, and the total stump basal area. These vari-ables were moderately to highly correlated (mean ± SD Pearsoncorrelations = 0.694 ± 0.109, N = 15). Variables describing thehistory of fire disturbance, which were obtained from interviews orin situ, included the extent of fires, number of recurrent fires, firepenetration distance from forest edges, and total area burned. Thesevariables were moderately to highly correlated with a mean Pearsoncorrelation of 0.732 (SD = 0.137, N = 15).

We controlled for high levels of interdependence between patchand landscape variables by performing a Pearson correlation matrix,and excluding those variables intercorrelated by r > 0.70. Distanceto the nearest edge was highly correlated with forest patch area(log10 ha) so we used the residuals from the linear regression fromthese two variables in all further analyses of distance to the nearestedge.

Disturbance-Mediated Drift in Amazonian Trees 5

Using generalized linear mixed models (hereafter, GLMMs),we examined the effects of (1) forest patch and landscape variables(e.g., forest size, time since fragmentation, and distance to the nearestedge), and (2) disturbance variables (i.e., PCA axis 1 of logging andfire severity) on the density of small-seeded softwoods and hardwoodstems, and two ordination axes describing tree abundance in eachplot. GLMMs introduce higher-level random effects in addition tothe residual error term, and allow estimation of fixed effect andinteraction parameters while accounting for random effects at dif-ferent levels of the data. It was necessary to use mixed-effects modelsto account for the spatial correlation within the data. GLMMs werefitted to model the density of small-seeded softwood and hardwoodstems, and the two axes of the NMDS of floristic composition asfunction of landscape and disturbance variables using “site” as a ran-dom effect. Models were fitted within the R platform and selectedbased on an information-theoretic approach (Akaike’s Informa-tion Criterion—AIC) to evaluate our a priori models (Burnham &Anderson 2002).

When using parametric statistical tests, appropriate transfor-mations were used to stabilize variances and improve normality ofthe data. Data on time since fragmentation and forest patch areawere square-rooted and log-transformed, respectively.

We examined which subset of individual explanatory variablescould maximize the rank correlation between biotic and environ-mental similarity matrices and provide the best model to explain dif-ferences in floristic composition among the 60 forest plots using theBIOENV procedure (Clarke & Ainsworth 1993). BIOENV useda Spearman’s rank correlation coefficient (S) between the resultingranked similarity matrices of tree assemblage and correlation basedon a PCA of the normalized environmental variables. The proce-dure then calculates S between the tree abundance similarity matrixand all possible sets of variables. The combination of variables thathas the highest S-value is considered to provide the best support forthe abundance data.

FIGURE 2. Relationships between forest patch area (log10) and (A) stem density per hectare, and (B) number of tree genera in 21 forest fragments (gray circles) and

two continuous forest sites (solid circles). Multiple plots sampled in all but the smallest forest patches are represented by means ± SE.

RESULTS

FOREST PATH AREA AND FLORISTIC COMPOSITION.—A total of 8248trees were recorded in all 60 quarter-hectare forest plots, including6890 live trees, 998 fallen dead trees, and 360 standing dead trees.From this total, we retained for further analysis 6807 trees (or 98.8%of all live stems) belonging to 130 genera. Occurrence rates rangedwidely across different tree genera, from those occurring in all 60plots (e.g., Tetragastris, Burseraceae) to those occurring in a singleplot (see Appendix S1).

The 21 forest patches surveyed ranged in size from 2.0 to14,480 ha (mean ± SD = 1416 ± 3834 ha), and contained between21 and 36 of all 130 tree genera (28.5 ± 4.2 genera, N = 52). Incomparison, the eight plots inventoried in the two continuous forestsites contained between 23 and 41 genera (29.5 ± 5.4 genera), butwere not significantly richer than plots in forest fragments (t =0.694, df = 7, P = 0.510).

Considering all possible pairwise comparisons between forestplots, the Chao–Sørensen abundance-based similarity at the level oftree genera was on average 0.809 (SD = 0.120, range = 0.281–1.00,N = 1770). Mantel tests indicated that the tree assemblage structureacross the 60 forest plots was influenced by large-scale spatial effects.The straight-line distance between any two floristic plots variedfrom 0.07 to 73.5 km (26.3 ± 14.6 km, N = 1770), resulting insignificant spatial autocorrelation (r = 0.688, P < 0.001). Plotscloser to one another, which were usually placed within the sameforest patch, were significantly more similar than more distant plots.We are therefore unable to consider plots sampling the same patchas spatially independent in any further analyses.

EFFECTS OF FOREST PATCH SIZE.—There was no effect of forest patcharea on the mean number of live stems per hectare (R2 = 0.0, F1,19 =0.84, P = 0.371; Fig. 2A). However, contrary to most species-areacurves, there was a marginally negative relationship between forest

6 Michalski, Nishi, and Peres

patch area and the mean number of tree genera contained within 21forest patches (R2 = 0.177, F1,19 = 5.31, P = 0.033), with smallfragments containing a slightly higher number of tree genera thanmedium and large fragments (Fig. 2B).

The percentages of small-seeded softwood and hardwood stemsin each plot were clearly affected in opposite directions by forestpatch area. Forest plots in small forest patches contained a greaterabundance of mean small-seeded softwood trees compared to largepatches (R2 = 0.362, F1,19 = 12.37, P = 0.002; Fig. 3A). Con-versely, hardwood trees were significantly more abundant in largeforest patches (R2 = 0.300, F1,19 = 9.56, P = 0.006; Fig. 3C),whereas the abundance of large-seeded softwoods did not show anysignificant relationship with forest patch area (R2 = 0.021, F1,19 =1.42, P = 0.247; Fig. 3B).

The first and second axes of the NMDS based on the Chao–Sørensen abundance-based similarity matrix of 130 tree genera inall 60 forest plots (stress value = 0.29) suggested that the data setis highly heterogeneous. Indeed, the wide positional scatter of ourforest sites in the NMDS plots indicates poor congruence in thepatterns of floristic composition and abundance.

PREDICTORS OF FUNCTIONAL GROUPS.—The first axes derived fromthe PCAs of logging intensity and burn severity explained 77.2percent and 80.1 percent of the maximum amount of variation ineach data set, respectively. The first PCA axis of logging intensitywas positively related to the most logging-disturbed sites, whereasthe first PCA axis of fire severity was related to the most fire-disturbed sites, with greater positive values associated with a largernumber of recurrent fire events and proportionally larger areas ofburnt forest. Time since fragmentation and forest size were the mostcommon predictors retained in the GLMMs having opposite effectson the density of small-seeded softwood and hardwood stems, andon explaining the second axis of the ordination describing treeabundance. Fire severity had a marginal negative effect on the firstaxis of the ordination. Distance to the nearest edge had a negativeeffect on the density of small-seeded softwoods, but had a positiveeffect on the density of hardwood stems (Table 1).

The BIOENV procedure based on the Chao–Sørensenabundance-based similarity matrix resulted in a maximum corre-lation of 0.177 with the patch and disturbance variables with fourvariables selected (patch size, time since fragmentation, distance tothe nearest edge, and fire severity). Overall, these four variables andlogging intensity best explained patterns of tree abundance acrossall 60 forest plots (Table 2).

DISCUSSION

As far as we are aware, this study is based on the largest numberof tropical forest fragments of widely variable sizes and disturbancehistories, where tree assemblage composition has been quantified.Although we do not have soil samples from the location of ourforest plots, there is no evidence to suggest that major differencesin floristic composition across sites were due to pre-existing dif-ferences in soil types or other baseline environmental gradients inthe study region. Moreover, although soil types can affect floristic

FIGURE 3. Relationships between forest patch area (log10) and the abundance

of (A) small-seeded softwood stems (SSW), (B) large-seeded softwood stems

(LSW), and (C) hardwood stems (HWD) in 21 forest fragments (gray circles)

and two continuous forest sites (solid circles). Multiple plots sampled in all but

the smallest forest patches are represented by means ± SE.

composition (Poulsen et al. 2006), there is no evidence of edaphiceffects on the relative abundance of major tree functional groups(Laurance et al. 2006). As recently as 1976, the Alta Floresta regionwas entirely covered by undisturbed Amazonian terra firme forest ofrather similar physiognomy (RADAMBRASIL 1983, Soares-Filho1998, Oliveira-Filho & Metzger 2006). This apparent landscape-scale uniformity in the distribution of tree genera is at odds with

Disturbance-Mediated Drift in Amazonian Trees 7

TABLE 1. Minimum generalized linear mixed models showing significant land-

scape and disturbance predictors of the density of small-seeded softwood

(SSW) and hardwood (HWD) trees, and two ordination axes describ-

ing tree assemblage structure within 60 0.25-ha forest plots sampled in

21 forest fragments and two continuous sites.

Effect a Estimate SE z-value P

Density of SSW

Forest area −0.105 0.032 −3.266 0.001

Time since fragmentation 0.167 0.041 4.017 0.000

Distance to the nearest edge −0.526 0.126 −4.175 0.000

Logging intensity 0.130 0.044 2.998 0.003

Akaike’s information criterion

(AIC) = 183.4

Density of HWD

Forest area 0.046 0.024 1.940 0.053

Time since fragmentation −0.090 0.032 −2.840 0.005

Distance to the nearest edge 0.211 0.109 1.940 0.052

Akaike’s information criterion

(AIC) = 150.2

NMDS Axis 1

Fire severity −0.211 0.187 −1.129 0.259

Akaike’s information criterion

(AIC) = 35.9

NMDS Axis 2

Time since fragmentation −0.079 0.115 −0.690 0.490

Akaike’s information criterion

(AIC) = 28.1

aForest area: size of forest fragment or assigned value for continuous forests

(log10); Time since fragmentation (square-root years); Distance to the

nearest edge: residuals of mean distance to the nearest forest edge (see text);

Logging intensity: PCA 1 of logging intensity; and Fire severity: PCA 1 of

fire severity.

the significant spatial structure uncovered in the data set. However,this is probably best explained by our sampling design, which in-cluded some level of within-plot replication including 2–4 plots inall continuous forest sites and fragments that were sufficiently largeto accommodate multiple plots. Nearest-neighboring plots weretherefore more likely to be located in the same fragment (61.7%of all 60 quarter-hectare plots), and consequently shared a simi-lar postisolation disturbance history. Differences in local floristiccomposition are most likely explained by an interaction betweenpatch- and landscape-scale environmental variables and the historyof anthropogenic disturbance within each patch. We first turn toconventional patch area effects, and then explore how other vari-ables may interact with patch area to determine the composition oftrees and regeneration guilds.

EFFECTS OF PATCH SIZE.—Forest patch area alone is one of a num-ber of key patch metrics that could influence the shape and slopeof species-area relationships and reflects primarily the effects ofhabitat loss, rather than increased exposure to habitat degradation,

TABLE 2. Relationships between environmental variables and abundance of tree

genera occurring in 60 quarter-hectare forest plots using the BIOENV

procedure (see text). Analysis include up to five environmental variables

that best explain patterns of abundance. Resulting values of the 10 best

results are listed as Spearman rank correlation coefficients (S).

Number of Weighted Variables

variables Spearman (S) chosen1

4 0.177 Patcha, TFb, DEc, FSe

5 0.175 Patcha, TFb, DEc, LId, FSe FSc

3 0.174 Patcha, DEc, FSe

4 0.170 Patcha, DEc, LId, FSe

4 0.147 Patcha, TFb, LId, FSe

3 0.146 Patcha, TFb

, DEc

3 0.144 Patcha, TFb, FSe

3 0.144 Patcha, LId, FSe

4 0.144 Patcha, TFb, DEc, LId

2 0.138 Patcha, DEc

1 Environmental variables selected by BIOENV – aPatch: Forest patch area;bTF: Time since fragmentation (square-root years); cDE: Residuals of mean

distance to the nearest forest edge (see text); dLI: PCA 1 of logging intensity;eFS: PCA 1 of fire severity.

and human disturbances. Contrary to the vast majority of species-area relationships, our results indicate that patch area was not thestrongest predictor of floristic richness, explaining only 17.7 per-cent of the overall variation in number of tree genera across allforest plots. This could be explained by the fact that species-area“relaxation” in long-lived trees would require a long time to becomediscernible (Chambers et al. 1998, Laurance et al. 2004), and manyold-growth trees were still retained in small fragments. Our sys-tematic placement of forest plots in core areas of sufficiently largefragments could also contribute to this pattern. Forest patch areawas a negative predictor of the density of small-seeded softwoodstems, but a positive predictor of the density of hardwood stems.The negative responses of small-seeded pioneer trees to forest patcharea were largely associated with the high abundance of this guild insmall forest fragments. For example, this was the case of Cecropia,a widely known fast-colonizing tree genus in Neotropical forestcanopy gaps and second-growth habitats, which was particularlycommon in heavily disturbed, small forest patches.

Patch area effect presented here could be disaggregated intoseveral other effects since tree plots in small and medium forestfragments were often sampled in areas near forest edges. In fact,small forest fragments, particularly those < 10 ha, can effectivelyconsist entirely of edge habitat, and given time may not be struc-turally or floristically distinguishable from forest edges (Zuidemaet al. 1996, Viana et al. 1997). Edge effects have a number ofconsequences to forest functioning and composition including mi-croclimatic changes, increase in wind shear and turbulence, elevatedtree mortality, increase in liana abundances, and changes in litterfalland nutrient cycling (Laurance et al. 2002, 2006). The germination

8 Michalski, Nishi, and Peres

of many pioneer species can be facilitated by the lateral penetra-tion of light and wind along forest edges (Pearson et al. 2002) andincreased soil irradiance and temperature fluctuations (Didham &Lawton 1999, Laurance 2004). Moreover, plots near edges prob-ably intercepted a greater seed rain from early successional plantsthriving in peripheral treefall gaps and along forest margins (Janzen1983, Grau 2004, Nascimento et al. 2006). Small-seeded softwoodtrees thus rapidly increased in abundance in small- to medium-sizedfragments in response to these changes, despite the relatively recenthistory of forest loss and fragmentation in Alta Floresta. Similar re-sults were found in central Brazilian Amazonia, where fast-growingsuccessional tree species proliferated rapidly, tracking a decline ofold-growth trees (Laurance et al. 2006).

Curiously, our continuous forest sites did not follow the trendof large forest patches with regard to stem density, number of genera,and percentage of small-seeded softwood and hardwood stems (Figs.2 and 3). This cannot be attributed to logging disturbance in anyof our large fragments and control sites, as there was virtually noevidence of selective logging in our forest control plots, with a totalof one stump accounting for a mean of only 0.081 percent (SD =0.23) of the total number of live and dead trees sampled. Therefore,we attribute this variation to the natural heterogeneity of the areassampled, and plots in continuous forest sites were located fartherapart than those in forest fragments.

PATCH-LEVEL ANTHROPOGENIC DISTURBANCE.—Our results indi-cate that the recent history of human perturbation within forestpatches was also important determinants of the patterns of abun-dance for several genera and regeneration guilds. Time since frag-mentation was a negative predictor of the density of hardwoodstems, but a positive predictor of the density of small-seeded soft-woods. In the Alta Floresta region, recently isolated fragments innew deforestation frontiers often remained unlogged and relativelyundisturbed at the time of sampling because timber extraction usu-ally lagged behind new pulses of clear-cutting for cattle pasture ex-pansion. The positive relationship between small-seeded softwoodsand fragment age is entirely consistent with the gradual prolif-eration over time of this functional group in increasingly olderfragments.

Logging intensity was only a positive predictor of the abun-dance of small-seeded softwoods. Although our forest plots exhib-ited signs of virtually no logging activity, within-patch logging in-tensity was also retained in the BIOENV analysis as one of the bestvariables explaining patterns of tree abundance. The appearance inneighboring logging gaps of fast-growing pioneer trees, which wererarely logged in Alta Floresta, could greatly affect their abundancein logged sites (Pearson et al. 2002).

Nevertheless, logging practices in Alta Floresta have alreadydepleted stocks of the most commercially valuable timber speciesand are currently relatively unselective. In fact, most recurrent tim-ber offtakes over the last 25 yr have undergone a succession of lessselective logging cycles that target an increasingly larger numberof tree species of lower market value. However, undisturbed forestareas containing commercially valuable species farther from the mu-

nicipal center of our study region are still protected due to relativelydifficult access and high extraction costs.

In any case, logging often predisposes seasonally dry Amazo-nian forests to more severe wildfires (Cochrane et al. 1999) thatmay eventually affect even the core of remaining fragments, therebyrapidly degrading the last remaining areas of forest interior charac-terized by high tree basal area and a shaded undergrowth underneathclosed-canopy (Barlow & Peres 2004). This is consistent with thefact that fire severity was retained in the model of the first ordinationaxes of tree abundance. Tree species containing latex and resins canbe more common in heavily disturbed forests (Hedge et al. 1998),as confirmed in our study, because these traits may protect treesfrom heat stress or postburn attacks by pathogens. Conversely, latexand resins may exacerbate tree mortality by increasing bark flamma-bility as many resins produced by Protium and Hymenae trees areflammable (Barlow et al. 2003). Previous conjectures that buttressesmay act to protect vital cambium tissues from fire damage, aidingthe probability of tree survival (Uhl & Kauffman 1990, Pinard &Huffman 1997), could not be confirmed in our study. Althoughsome morphological traits could influence fire tolerance, local fireseverity was detrimental to some taxa and had a negative effect onthe first axis of the ordination. Although changes in forest structureare beyond the scope of this study, the total basal area of live treesin our forest plots was negatively associated with the first PCA axisof fire severity (R2 = 0.054, F1,58 = 4.35, P = 0.041). This ishighlighted in our BIOENV analysis and is consistent with markedpulses of tree mortality induced by surface fires elsewhere in theAmazon (Barlow et al. 2003).

CONSERVATION IMPLICATIONS.—We have shown that the relativeabundance of tree functional groups in forest fragments of southernAmazonia is a function of both habitat loss and the subsequent pro-cess of fragmentation, and that nonrandom drift in tree assemblagecomposition in remaining forest patches can take place within afew years or decades of postisolation patch exploitation and dis-turbance. However, in working tropical forest landscapes such asAlta Floresta, area and edge effects operate concurrently with otherhuman-induced disturbances (Peres 2001, Michalski & Peres 2005,Peres & Michalski 2006), which when acting in concert with habitatfragmentation can amplify the threats to tree species typical of ma-ture Neotropical forests (e.g., Tabarelli et al. 2004). Deterministicdrifts in plant community composition is likely to be even greaterfor other woody and nonwoody plant taxa beyond the scope ofthis study, including coarse herbs, shrubs, epiphytes, and small treesthat reach reproductive maturity in the forest understory, as manyof these taxa have faster generation times and are more sensitive tomicroclimatic conditions.

Logging activity combined with the dead phytomass fuel loadresulting from elevated rates of tree mortality render forest frag-ments more susceptible to surface fires (Holdsworth & Uhl 1997,Cochrane et al. 1999). This is especially the case of regions like AltaFloresta where forest remnants usually interface a matrix dominatedby cattle pastures in which weed control is deliberately managed byseasonal fires. In fact, time since fragmentation and fire severity

Disturbance-Mediated Drift in Amazonian Trees 9

appeared to be the most important determinants of the tree com-position and regeneration guild structure in Alta Floresta, and thefuture floristic composition of remaining fragments will also dependon the tolerance of individual saplings and mature trees to fire con-tact (Barlow & Peres 2004).

The proliferation of pioneer trees in forest fragments, and thecorrelated decline of many hardwood trees have important implica-tions for forest ecosystems, not least because fast-growing pioneershave a substantially lower wood density (Chambers et al. 2000) andthus store less carbon per wood volume than slow-growing matureforest trees. Softwood trees also tend to be smaller and retain alower biomass than hardwood trees. In addition, the smaller carbonstocks in stands dominated by early successional trees have a fasterturnover because they are short-lived compared to old-growth trees(Laurance et al. 2004). Finally, a proliferation of early successionaltree taxa near fragment edges may increase forest vulnerability tosurface fires by increasing litterfall accumulation (Vasconcelos &Luizao 2004), which can propagate through dense litter fuel on theforest floor (Cochrane et al. 1999).

Our analysis in a heavily fragmented region of the “defor-estation arc” of Brazilian Amazonia provides rare evidence of thesynergistic effects between forest fragmentation and forest distur-bance on the floristic composition and regeneration guild structurein a large number of forest fragments. It also indicates relativelyrapid community-level floristic responses in increasingly disturbedforest fragments to wholesale changes in the surrounding landscapestructure. This highlights the importance of understanding howhuman-induced disturbances can interact with habitat fragmenta-tion per se.

ACKNOWLEDGMENTS

This study was funded by NERC (UK) through a grant (2001/834)to CAP and by WWF Brazil and USAID (US) through a grant(NT 746/2003). We thank Conservation International for addi-tional support. F. Michalski is funded by a PhD studentship fromthe Brazilian Ministry of Education (CAPES). Fieldwork in AltaFloresta would not have been possible without the generous coop-eration of numerous landowners. We are deeply indebted to all ofthem, but especially to Romildo Candido da Silva and Vitoria daRiva. Idelfonso Dourado contributed tirelessly with the vernacu-lar identification of several tree species. Geraldo Araujo and AlexGrandini Araujo provided invaluable assistance during the field-work. We thank Robin Chazdon, Bill Laurance, Hans ter Steege,and two anonymous referees for constructive comments on themanuscript.

SUPPLEMENTARY MATERIAL

The following supplementary material is available for this articleonline at: www.blackwell-synergy. com/toc/btp

Appendix S1. Tree morphospecies and genera inventoried inthe 60 0.25-ha forest plots in Alta Floresta, Mato Grosso, Brazil.

LITERATURE CITED

ASQUITH, N. M., J. TERBORGH, E. ARNOLD, AND M. RIVEROS. 1999. The fruitsthe agouti ate: Hymenae courbaril seed fate when its disperser is absent.J. Trop. Ecol. 15: 229–235.

BAKER, T. R., O. L. PHILLIPS, Y. MALHI, S. ALMEIDA, L. ARROYO, A. DI FIORE,T. ERWIN, T. J. KILLEEN, S. G. LAURANCE, W. F. LAURANCE, S. L. LEWIS,J. LLOYD, A. MONTEAGUDO, D. A. NEILL, S. PATINO, N. C. A. PITMAN,J. N. M. SILVA, AND R. V. MARTINEZ. 2004. Variation in wood densitydetermines spatial patterns in Amazonian forest biomass. Global ChangeBiol. 10: 545–562.

BARLOW, J., AND C. A. PERES. 2004. Ecological responses to El Nino-induced sur-face fires in central Amazonia: Management implications for flammabletropical forests. Philosophic. Trans. R. Soc. Lond. B. 359: 367–380.

BARLOW, J., B. O. LAGAN, AND C. A. PERES. 2003. Morphological correlates offire-induced tree mortality in a central Amazonian forest. J. Trop. Ecol.19: 291–299.

BENITEZ-MALVIDO, J., AND M. MARTINEZ-RAMOS. 2003. Influence of edge expo-sure on tree seedling species recruitment in tropical rain forest fragments.Biotropica 35: 530–541.

BLOCK, A. 2004. Gottinger Mahhacksler Tritucap und Forstmulcher—NichtBrennende Flachenvorbereitung am Beispiel der Zona Bragantina, Nord-Ost-Amazonien, Brasilien. PhD Dissertation, University of Gottingen,Germany.

BRUNA, E. M. 1999. Seed germination in rainforest fragments. Nature 402: 139.BURNHAM, K. P., AND D. R. ANDERSON. 2002. Model selection and multimodel

inference: A practical-theoretic approach, 2nd edition. Springer-Verlag,New York, New York.

CASPER, B. B., S. B. HEARD, AND V. APANIUS. 1992. Ecological correlates ofsingle-seededness in a woody tropical flora. Oecologia 90: 212–217.

CHAMBERS, J. Q., N. HIGUCHI, AND J. P. SCHIMEL. 1998. Ancient trees inAmazonia. Nature 391: 135–136.

CHAMBERS, J. Q., N. HIGUCHI, J. P. SCHIMEL, L. V. FERREIRA, AND J. M.MELACK. 2000. Decomposition and carbon cycling of dead trees intropical forests of the central Amazon. Oecologia 122: 380–388.

CHAO, A., R. L. CHAZDON, R. K. COLWELL, AND T.–J. SHEN. 2005. A newstatistical approach for assessing similarity of species composition withincidence and abundance data. Ecol. Lett. 8: 148–159.

CHAVE, J., H. C. MULLER-LANDAU, T. R. BAKER, T. A. EASDALE, H. TER STEEGE,AND C. O. WEBB. In press. Regional and phylogenetic variation in wooddensity among 2456 Neotropical tree species. Ecol. Appl.

CLARKE, K. R. 1993. Nonparametric multivariate analyses of changes in com-munity structure. Aust. J. Ecol. 18: 117–143.

CLARKE, R. R., AND M. AINSWORTH. 1993. A method of linking multivariatecommunity structure to environmental variables. Mar. Ecol. Prog. Ser.92: 205–219.

COCHRANE, M. A., AND M. D. SCHULZE. 1999. Fire as recurrent event in atropical forest of the eastern Amazon: Effects on forest structure, biomass,and species composition. Biotropica 31: 2–16.

COCHRANE, M. A., A. ALENCAR, M. D. SCHULZE, C. M. SOUZA, D. C. NEPSTAD,P. LEFEBVRE, AND E. A. DAVIDSON, E. A.. 1999. Positive feedbacks inthe fire dynamic of closed canopy tropical forests. Science 284: 1832–1835.

COCHRANE, M. A., D. L. SKOLE, E. A. T. MATRICARDI, C. BARBER, AND W.CHOMENTOWSKI. 2002. Interaction and synergy between selective log-ging, forest fragmentation and fire disturbance in Tropical forests: Casestudy Mato Grosso, Brazil. CGCEO Research Advances No. RA03–02 \ w. East Lansing, Michigan.

COLWELL, R. K. 2004. Estimates: Statistical estimation of species rich-ness and shared species from samples, Version 7.5. Available athttp://viceroy.eeb.uconn.edu/estimates. Persistent URL http://purl. ocl.org/estimates.

CORDEIRO, N. J., AND H. F. HOWE. 2003. Forest fragmentation severs mutualismbetween seed dispersers and an endemic African tree. Proc. Natl. Acad.Sci. USA 100: 14052–14056.

10 Michalski, Nishi, and Peres

DA SILVA, J. M. C., AND M. TABARELLI. 2000. Tree species impoverishment andthe future flora of the Atlantic forest of northeastern Brazil. Nature 404:72–74.

DIDHAM, R. K., AND J. H. LAWTON. 1999. Edge structure determines the mag-nitude of changes in microclimate and vegetation structure in tropicalforest fragments. Biotropica 31: 17–30.

FEARNSIDE, P. M. 1997. Wood density for estimating forest biomass in BrazilianAmazonia. For. Ecol. Manage. 90: 59–87.

FOOD AND AGRICULTURE ORGANIZATION OF THE UNITED NATIONS (FAO).1997. Estimating biomass and biomass change of tropical forests:A primer. Available at http://www.fao.org/docrep/W4095E/w4095e00.HTM.

FOOD AND AGRICULTURE ORGANIZATION OF THE UNITED NATIONS

(FAO). 2004. Global forest resources assessment update 2005—Guidelines for country reporting to FRA 2005. Available athttp://www.fao.org/docrep/007/j2132e/J2132e00.HTM.

FOX, B. J., J. E. TAYLOR, M. D. FOX, AND C. WILLIAMS. 1997. Vegetationchanges across edges of rainforest remnants. Biol. Conserv. 82: 1–13.

GASCON, C., B. WILLIAMSON, AND G. A. B. DA FONSECA. 2000. Receding forestedges and vanishing reserves. Science 288: 1356–1358.

GENTRY, A. 1982. Patterns of Neotropical plant species diversity. Evol. Biol. 15:1–84.

GENTRY, A. H. 1990. Floristic similarities and differences between southernCentral America and upper and central Amazonia. In A. H. Gentry(Ed.). Four Neotropical forests, pp. 141–157. Yale University Press,New Haven, Connecticut.

GRAU, H. R. 2004. Landscape context and local-scale environment influenceregeneration of pioneer tree species in treefall gaps. Biotropica 36: 52–59.

HECHT, S. B.. 1993. The logic of livestock and deforestation in Amazonia.Bioscience 43: 687–695.

HEDGE, V., M. D. S. CHANDRAN, AND M. GADGIL. 1998. Variation in thicknessin a tropical forest community of Western Ghats in India. Funct. Ecol.12: 313–318.

HENDERSON, A., G. GALEANO, AND R. BERNAL. 1995. Field guide to the palmsof the Americas. Princeton University Press, Princeton, New Jersey.

HIGGINS, M. A., AND K. RUOKOLAINEN. 2004. Rapid tropical forest inven-tory: A comparison of techniques based on inventory data from westernAmazonia. Conserv. Biol. 18: 799–811.

HOLDSWORTH, A. R., AND C. UHL. 1997. Fire in Amazonian selectively loggedrain forest and the potential for fire reduction. Ecol. Appl. 7: 713–725.

HOOPER, E. R, P. LEGENDRE, AND R. CONDIT. 2004. Factors affecting commu-nity composition of forest regeneration in deforested, abandoned landin Panama. Ecology 85: 3313–3326.

HUBBELL, S. P., AND R. B. FOSTER. 1996. Commonness and rarity in a Neotrop-ical forest: Implications for tropical tree conservation. In M. E. Soule(Ed.). Conservation biology: the science of scarcity and diversity, pp.205–223. Sinauer Associates, Sunderland, Massachusetts.

INPE. 2004. Projeto PRODES. Available from http://www.obt.inpe.br/prodes/.JANZEN, D. H. 1983. No park is an island: Increase in interference from outside

as park size decreases. Oikos 41: 402–410.KAUFFMAN, J. B. 1991. Survival by sprouting following fire in tropical forests of

the eastern Amazon. Biotropica 23: 219–224.LAURANCE, W. F. 2004. Forest-climate interactions in fragmented tropical land-

scapes. Philosophic. Trans. R. Soc. Lond. B 359: 345–352.LAURANCE, W. F., M. A. COCHRANE, S. BERGEN, P. M. FEARNSIDE, P. DE-

LAMONICA, C. BARBER, S. D’ANGELO, AND T. FERNANDES. 2001. Thefuture of the Brazilian Amazon. Science 291: 438–439.

LAURANCE, W. F., P. DELAMONICA, S. G. LAURANCE, H. L. VASCONCELOS, AND

T. E. LOVEJOY. 2000. Rainforest fragmentation kills big trees. Nature404: 836.

LAURANCE, W. F., L. V. FERREIRA, J. M. RANKIN-DE-MERONA, AND S. G.LAURANCE. 1998. Rain forest fragmentation and the dynamics of Ama-zonian tree communities. Ecology 79: 2032–2040.

LAURANCE, W. F., S. G. LAURANCE, L. V. FERREIRA, J. M. RANKIN-DE-MERONA,C. GASCON, AND T. E. LOVEJOY. 1997. Biomass collapse in Amazonianforest fragments. Science 278: 1117–1118.

LAURANCE, W. F., T. E. LOVEJOY, H. L. VASCONCELOS, E. M. BRUNA, R. K.DIDHAM, P. C. STOUFFER, C. GASCON, R. O. BIERREGAARD, S. G.LAURANCE, AND E. SAMPAIO. 2002. Ecosystem decay of Amazonianforest fragments: A 22-year investigation. Conserv. Biol. 16: 605–618.

LAURANCE, W. F., H. E. M. NASCIMENTO, S. G. LAURANCE, A. C. ANDRADE, P.M. FEARNSIDE, J. E. L. RIBEIRO, AND R. L. CAPRETZ. 2006. Rain forestfragmentation and the proliferation of successional trees. Ecology 87:469–482.

LAURANCE, W. F., H. E. M. NASCIMENTO, S. G. LAURANCE, R. CONDIT, S.D’ANGELO, AND A. ANDRADE. 2004. Inferred longevity of Amazonianrainforest trees based on a long-term demographic study. For. Ecol.Manage. 190: 131–143.

LORENZI, H.. 1998. Arvores Brasileiras—Manual de identificacao e cultivo deplantas arboreas nativas do Brasil. Vol. 2. Instituto Plantarum, Sao Paulo,Brazil.

LORENZI, H. 2000. Arvores Brasileiras—Manual de identificacao e cultivo deplantas arboreas nativas do Brasil. Vol. 1. Instituto Plantarum, Sao Paulo,Brazil.

MACHADO, R. B. 2004. Principais especies florestais de uso madeireiro exploradasno municıpio de Alta Floresta. Graduate dissertation, UNEMAT, AltaFloresta, MT, Brazil.

MALCOLM, J. R. 1994. Edge effects in central Amazonian forest fragments.Ecology 75: 2438–2445.

MANTEL, N. 1967. The detection of disease clustering and a generalized regres-sion approach. Cancer Res. 27: 209–220.

MCGARIGAL, K., AND B. J. MARKS. 1995. FRAGSTATS: Spatial pat-tern analysis program for quantifying landscape structure. USDAFor.Serv.Gen.Tech.Rep. PNW-351.

MICHALSKI, F., AND C. A. PERES. 2005. Anthropogenic determinants of pri-mate and carnivore local extinctions in a fragmented forest landscape ofsouthern Amazonia. Biol. Conserv. 124: 383–396.

MORAES, J. L., C. C. CERRI, J. M. MELILLO, D. KICKLIGHTER, C. NEILL, D. L.SKOLE, AND P. A. STEUDLER. 1995. Soil carbon stocks of the BrazilianAmazon basin. Soil Sci. Soc. Am. J. 59: 244–247.

NASCIMENTO, H. E. M., AND W. F. LAURANCE. 2004. Biomass dynamics inAmazonian forest fragments. Ecol. Appl. 14: S127–S138.

NASCIMENTO, H. E. M., A. C. S. ANDRADE, J. L. C. CAMARGO, W. F. LAURANCE,S. G. LAURANCE, AND J. E. L. RIBEIRO. 2006. Effects of the surroundingmatrix on tree recruitment in Amazonian forest fragments. Conserv.Biol. 20: 853–860.

NEPSTAD, D. C., A. VERISSIMO, A. ALENCAR, C. NOBRE, E. LIMA, P. LEFEBRE, P.SCHLESINGER, C. POTTER, P. MOUTINHO, E. MENDOZA, M. COCHRANE,AND V. BROOKS. 1999. Large-scale impoverishment of Amazonian forestsby logging and fire. Nature 398: 505–508.

OLIVEIRA-FILHO, F. J. B., AND J. P. METZGER. In press. Thresholds in land-scape structure for three common deforestation patterns in the BrazilianAmazon. Landscape Ecol.

PARROTTA, J. A., J. K. FRANCIS, AND R. R. DE ALMEIDA. 1995. Trees of theTapajos: A photographic field guide. USDA Forest Service, Washington,DC. General technical Report II TF-1.

PEARSON, T. R. H., D. BURSLEM, C. MULLINS, AND J. W. DALLING. 2002.Germination ecology of Neotropical pioneers: Interacting effects of en-vironmental conditions and seed size. Ecology 83: 2798–2807.

PERES, C. A. 1999. Ground fires as agents of mortality in a Central Amazonianforest. J. Trop. Ecol. 15: 535–541.

PERES, C. A. 2000. Effects of subsistence hunting on vertebrate communitystructure in Amazonian forests. Conserv. Biol. 14: 240–253.

PERES, C. A. 2001. Synergistic effects of subsistence hunting and habitat frag-mentation on Amazonian forest vertebrates. Conserv. Biol. 15: 1490–1505.

PERES, C. A., AND F. MICHALSKI. 2006. Synergistic effects of habitat disturbanceand hunting in Amazonian forest fragments. In W. F. Laurance and

Disturbance-Mediated Drift in Amazonian Trees 11

C. A. Peres (Eds.). Emerging threats to tropical forests, pp: 105–127.University of Chicago Press, Chicago.

PERES, C. A., AND VAN M. ROOSMALEN. 2002. Patterns of primate frugivory inAmazonia and the Guianan shield: Implications to the demography oflarge-seeded plants in overhunted tropical forests. In D. Levey, W. Silvaand M. Galetti (Eds.). Seed dispersal and frugivory: Ecology, evolutionand conservation, pp. 407–423. CAB International, Wallingford, UK.

PEREZ-SALICRUP, D. R. 2001. Effect of liana cutting on tree regeneration in aliana forest in Amazonian Bolivia. Ecology 82: 389–396.

PINARD, M. A., AND J. HUFFMAN. 1997. Fire resistance and bark propertiesof trees in a seasonally dry forest in eastern Bolivia. J. Trop. Ecol. 13:727–740.

PITMAN, N. C. A., J. TERBORGH, M. R. SILMAN, AND P. NUEZ. 1999. Tree speciesdistributions in an upper Amazonian forest. Ecology 80: 2651–2661.

POULSEN, A. D., H. TUOMISTO, AND H. BALSLEV. 2006. Edaphic and floristicvariation within a 1-ha plot of lowland Amazonian rain forest. Biotropica38: 468–478.

PROJETO RADAMBRASIL. 1983. Folha SC.21 Juruena. Levantamento de re-cursos naturais, vol. 20, Ministerio das Minas e Energia, Rio de Janeiro,RJ, Brazil.

RIBEIRO, J. E. L. da S., M. J. G. HOPKINS, A. VICENTINI, C. A. SOTHERS, M. A.DA S. COSTA, J. M. DE BRITO, M. A. D. DE SOUZA, L. H. P. MARTINS,L. G. LOHMANN, P. A. C. L. ASSUNCAO, E. da C. PEREIRA, C. F. daSILVA, M. R. MESQUITA, AND L. C. PROCOPIO. 1999. Flora da ReservaDucke—Guia de identificacao das plantas vasculares de uma floresta deterra-firme na Amazonia Central. INPA-DFID, Manaus, Brazil.

SANTOS, V. A. 2005. Analise florıstica e estrutural de uma floresta ombrofilaaberta primaria do Parque Estadual Cristalino, Alta Floresta—MT,Brasil. MSc Dissertation, Universidade Estadual do Mato Grosso (UN-EMAT), Alta Floresta, Brazil.

SIZER, N., AND E. V. J. TANNER. 1999. Responses of woody plant seedlings toedge formation in a lowland tropical rainforest, Amazonia. Biol. Conserv.91: 135–142.

SKOLE, D., AND C. J. TUCKER. 1993. Tropical deforestation and habitat frag-mentation in the Amazon: Satellite data from 1978 to 1988. Science260: 1905–1910.

SOARES-FILHO, B. S. 1998. Modelagem da dinamica de paisagem de umaregiao de fronteira de colonizacao amazonica. PhD Dissertation, EscolaPolitecnica da Universidade de Sao Paulo, Sao Paulo, Brazil.

SOARES-FILHO, B. S., D. C. NEPSTAD, L. M. CURRAN, G. C. CERQUEIRA, R.A. GARCIA, C. A. RAMOS, E. VOLL, A. MCDONALD, P. LEFEBVRE, AND

P. SCHLESINGER. 2006. Modelling conservation in the Amazon basin.Nature 440: 520–523.

TABARELLI, M., J. M. C. DA SILVA, AND C. GASCON. 2004. Forest fragmentation,synergisms and the impoverishment of Neotropical forests. Biodivers.Conserv. 13: 1419–1425.

TER STEEGE, H. 2000. Plant diversity in Guyana. With recommendations fora national-protected area strategy. Tropenbos Series 18. The TropenbosFoundation, Wageningen, The Netherlands.

TER STEEGE, H., and D. S. HAMMOND. 2001. Character convergence, diversityand disturbance in tropical rain forest in Guyana. Ecology 82: 3197–3212.

TER STEEGE, H., N. C. A. PITMAN, O. L. PHILLIPS, J. CHAVE, D. SABATIER, A.DUQUE, J. F. MOLINO, M. F. PREVOST, R. SPICHIGER, H. CASTELLANOS,P. VON HILDEBRAND, AND R. VASQUEZ. 2006. Continental-scale patternsof canopy tree composition and function across Amazonia. Nature 443:444–447.

TER STEEGE, H., D. SABATIER, H. CASTELLANOS, T. VAN ANDEL, J. DUIVENVO-ORDEN, A. A. DE OLIVEIRA, R. EK, R. LILWAH, P. MAAS, AND S. MORI.2000. An analysis of the floristic composition and diversity of Amazo-nian forests including those of the Guiana Shield. J. Trop. Ecol. 16:801–828.

TER STEEGE, H., I. WELCH, AND R. ZAGT. 2002. Long-term effect of timberharvesting in the Bartica Triangle, Central Guyana. For. Ecol. Manage.170: 127–144.

TERBORGH, J., L. LOPEZ, P. NUNEZ, M. RAO, G. SHAHABUDDIN, G. ORIHUELA,M. RIVEROS, R. ASCANIO, G. ADLER, T. LAMBERT, AND L. BALBAS.2001. Ecological meltdown in predator-free forest fragments. Science294: 1923–1926.

UHL, C., AND J. B. KAUFFMAN. 1990. Deforestation, fire susceptibility, andpotential tree response to fire in the Eastern Amazon. Ecology 71: 437–449.

UHL, C., P. BARRETO, A. VERISSIMO, E. VIDAL, P. AMARAL, A. C. BARROS, C.SOUZA, J. JOHNS, AND J. GERWING. 1997. Natural resource managementin the Brazilian Amazon: An integrated research approach. Science 47:160–168.

VALENCIA, R., R. B. FOSTER, G. VILLA, R. CONDIT, J. C. SVENNING, C. HERNAN-DEZ, K. ROMOLEROUX, E. LOSOS, E. MAGARD, AND H. BALSLEV. 2004.Tree species distributions and local habitat variation in the Amazon:Large forest plot in eastern Ecuador. J. Ecol. 92: 214–229.

VAN ROOSMALEN, M. G. M.. 1985. Fruits of the Guianan flora. Institute ofSystematic Botany, Utrecht University, The Netherlands.

VASCONCELOS, H. L., AND F. J. LUIZAO. 2004. Litter production and litternutrient concentrations in a fragmented Amazonian landscape. Ecol.Appl. 14: 884–892.

VERISSIMO, A., P. Barreto, R. TARIFA, AND C. UHL. 1995. Extraction of a high-value natural resource in Amazonia: The case of mahogany. For. Ecol.Manage. 72: 39–60.

VERISSIMO, A., C. SOUZA, S. STONE, AND C. UHL. 1998. Zoning of timberextraction in the Brazilian Amazon. Conserv. Biol. 12: 128–136.

VIANA, V. M., A. J. TABANEZ, AND J. L. BATISTA. 1997. Dynamics and restorationof forest fragments in the Brazilian Atlantic forest. In W. F. Lauranceand R. O. Bierregaard Jr. (Eds.). Tropical forest remnants: Ecology,management, and conservation of fragmented communities, pp. 351–365. University of Chicago Press, Chicago, Illinois.

WHITMORE, T. C. 1997. Tropical forest disturbance, disappearance, and speciesloss. In (W. F. Laurance and R. O. Bierregaard Jr. (Eds.). Tropical For-est Remnants: Ecology, management, and conservation of fragmentedcommunities, pp.3–12. University of Chicago Press, Chicago, Illinois.

WRIGHT, S. J. 2003. The myriad consequences of hunting for vertebrates andplants in tropical forests. Perspect. Plant Ecol. 6: 73–86.

ZUIDEMA, P. A., J. A. SAYER, AND W. DIJKMAN. 1996. Forest fragmentation andbiodiversity: The case for intermediate-sized conservation areas. Environ.Conserv. 23: 290–297.