Gap-crossing movements predict species occupancy in Amazonian forest fragments

Gap-crossing movements predict species occupancy in Amazonianforest fragments

Alexander C. Lees and Carlos A. Peres

A. C. Lees ([email protected]) and C. A. Peres, Centre for Ecology Evolution and Conservation, School of Environmental Sciences, Univ. of EastAnglia, Norwich, NR4 7TJ, UK.

In fragmented landscapes, species persistence within isolated habitat patches is governed by a myriad of species life-history, habitat patch and landscape characteristics. We investigated the inter-specific variation in non-forest gap-crossingabilities of an entire tropical forest-dependent avifauna. We then related this measure of dispersal ability to species life-history characteristics and occupancy data from 31 variable-sized forest patches sampled within the same fragmentedforest landscape. A total of 5436 gap-crossing movements of 231 forest-dependent bird species were observed across tenlinear forest gaps of varying widths, adjacent to large areas of undisturbed forest. Species persistence in isolated fragmentswas strongly linked to gap-crossing ability. The most capable gap-crossers were medium to large-bodied species in thelarge insectivore, frugivore and granivore guilds, matching the most prevalent subset of species in small forest patches.However, some competent gap-crossing species failed to occur in small patches, and minimum forest-patch arearequirements were more important in determining patch occupancy for these species. Narrow forest gaps (4�70 m)created by roads and power-lines may become territory boundaries, thereby eliminating home-range gap-crossingmovements for many forest species, but permit rarer dispersal events. Wider gaps (�70 m) may inhibit gap-crossingbehaviour for all but the most vagile species. Although patch size and quality may be the most important factors instructuring species assemblages in forest fragments, our results show that the degree of patch isolation and permeability ofthe surrounding matrix also explain which species can persist in forest isolates. Reducing the number and width of forest-dividing gaps; maintaining and/or creating forest corridors and increasing matrix permeability through the creation andmaintenance of ‘stepping-stone’ structures will maximise the species retention in fragmented tropical forest landscapes.

The extent to which a forest species can occupy forestfragments depends on its physiological and locomotorability to overcome the intervening open-habitat matrix;the size, state of degradation and connectivity of thefragment; and the nature of the surrounding non-forestlandscape (Taylor et al. 1993, Fahrig 2007). The point atwhich patch isolation effects become more important thanpatch size effects in a landscape depends on the dispersalcapabilities of any given species (Andren 1994, With et al.1997). A small forest fragment can be used by forest-dependent species with large area requirements if resourceswithin the patch can be supplemented by those ofneighbouring, but accessible forest patches (Dunninget al. 1992).

It has long been suggested that birds may be compara-tively less sensitive to habitat fragmentation, owing to theirhigh vagility (Ambuel and Temple 1983). Some species canfly across gaps of many tens of kilometres in width, andovercoming gaps narrower than one km may involvenegligible energetic costs. However, many forest speciesare closely associated with closed-canopy habitat and evadecanopy discontinuities, thereby potentially restricting thegap sizes they are willing to traverse (Grubb and Doherty

1999). This apparent reluctance to move into open habitatcould be explained by higher perceived predation risk (Limaand Dill 1990), a limited perceptual range within whichbirds can detect and identify other forest patches (Lima andZollner 1996), or naturally low rates of dispersal (Green-wood and Harvey 1982).

The frequency with which birds traverse linear gap stripsis influenced by both ‘gap avoidance’, whereby species avoidforest clearings (Greenberg 1989), and ‘edge avoidance’.For example, many species decline in abundance near forestedges (Laurance et al. 2004) and may only rarely cross gapsbecause they are rare outside core-forest areas. Forestfragment edges are exposed to various edge effects thatmay alter forest structure, light levels, thermal regimes,floristic composition, and invertebrate abundance (Kapos1989, Didham 1997, Laurance et al. 2002). Reproductivesuccess may be reduced in small forest patches for multiplereasons, such as smaller and less diverse food supplies(Burke and Nol 1998), elevated predation levels (Andren1992) and increased exposure to extreme environmentalconditions along forest edges (Murcia 1995). Forest birdresponses to fragmentation may be highly species/guildspecific. Large-bodied species (Grubb and Doherty 1999),

Oikos 118: 280�290, 2009

doi: 10.1111/j.1600-0706.2008.16842.x,

# 2009 The Authors. Journal compilation # 2009 Oikos

Subject Editor: Tim Benton. Accepted 29 July 2008

280

species with large spatial requirements (Dale et al. 1994)and habitat generalists (Sieving et al. 1996) are more likelyto cross gaps between forest patches. In contrast, smallunderstorey birds (Karr 1982), members of mixed-speciesflocks, army-ant followers, terrestrial species, solitary in-sectivores, and some forest specialists are usually reluctant tocross open areas (Sieving et al. 1996, Develey and Stouffer2001, Laurance et al. 2004). For example, local movementsof many understorey bird species in central Amazonia weresignificantly inhibited by narrow unpaved road clearings(Develey and Stouffer 2001, Laurance et al. 2004).

Two qualitatively distinct forms of gap-crossing arerecognized (Grubb and Doherty 1999) � ‘dispersal gap-crossing’ and ‘home-range gap-crossing’. The former occursduring movement between natal and breeding sites when anindividual must acquire its own territory. In many species,these movements may be very brief over a lifetime, involveovercoming significant swathes of hostile habitat, and maybe important for patch recolonization (Opdam 1991) andrescue effects (Brown and Kodric-Brown 1977) in ametapopulation. The second mechanism refers to morehabitual gap-crossing events between habitat patches in-cluded within an animal’s home range to meet basicmetabolic requirements.

Here we investigate avian gap-crossing behaviour inBrazilian Amazonia, where 20 000 km of new forest edgesare created every year (M. Cochrane and D. Skole pers.comm.). A growing network of forest roads (which areusually narrow, unpaved, lightly travelled, and remote) arethe most important potential subdividing features alongnew deforestation frontiers. Although such small roads mayhave little ecological impact, over time, roads may bewidened and paved and other linear gaps opened upfollowing infrastructure ‘improvements’ such as the con-struction of gas and power-lines (Laurance et al. 2004).Road construction in remote areas catalyses human migra-tion, leading to accelerated land-use change and forestfragmentation. Large-scale deforestation in private land-holdings produces much larger gaps, which may be highlyirregular in size and shape, presenting considerable barriersto dispersal. A proliferation of planned developmentactivities and their associated infrastructure investments isexpected to impact Amazonia in the near-future. Mostnotable among these is the Initiative for the Integration ofthe Regional Infrastructure of South America (IIRSA)project, a blueprint to improve physical links betweenAmazonian countries via highways, waterways and energyprojects (Reed and de Souza 2005, Killeen 2007). This willresult in further massive subdivision and isolation of forestpatches (Laurance and Williamson 2001).

Our current knowledge of the gap-crossing behaviour ofNeotropical forest birds remains poor (Ferraz et al. 2007),and this study is the first to examine the relative gap-crossing capabilities of an entire avian assemblage in ahighly-fragmented tropical landscape using passive gap-crossing observations. Specifically, we examine (1) thevariation in gap-crossing behaviour exhibited by differentspecies and families, (2) the life-history correlates of gap-crossing behaviour, and (3) the relationship betweenobserved occupancy of different-sized forest patches andgap-crossing rate of species persisting in large patches.

Methods

Study landscape

The countryside around the town of Alta Floresta, State ofMato Grosso, Brazil (09853?S, 56828?W) lies along thecentre of the Amazonian ‘Arc of Deforestation’ andprovides an ideal landscape in which to study the effectsof primary forest fragmentation and perturbation (Fig. 1).For a fuller description of the study area, see Lees and Peres(2006).

From June to September 2004 and May to June 2006,930 unlimited-radius point-counts were completed at 31forest sites, including 30 fragments and one continuous,undisturbed forest site. This continuous forest and two largeforest fragments (�10 000 ha) were treated as pseudo-controls. The size of the 31 forest sites surveyed rangedfrom 1.2 to over 1 000 000 ha (largest true isolate�14 476ha; mean9SD�5647926 014 ha). Sites were initiallyselected using a georeferenced 2001 Landsat ETM image(scene 227/67), on the basis of their size, degree of isolation,degree of perturbation (well-preserved patches were invari-ably chosen) and nature of the surrounding habitat matrix(all patches were embedded within a matrix of managedcattle pasture). All sites were located within a 50-km radiusof Alta Floresta and were accessible by river, and paved orunpaved roads. GPS coordinates were obtained in situ tolocate all forest patches inventoried. Thirty point-counts of15 min were completed within each site over threeconsecutive mornings at points along transects spaced200 m apart wherever possible, but this distance wasinevitably reduced to 50 m in the smallest fragments dueto severe area constraints. An equal amount of samplingtime (10 point-counts per day) was allocated to fragmentsof all size classes and the continuous forest site, thus smallpatches were more exhaustively sampled than large ones andthe results should be regarded as conservative with respectto area effects. During point-counts, all bird species seen orheard within the boundaries of the fragment were system-atically recorded, which may include typical ‘matrix taxa’ ifthey occurred within the edge boundaries of the patch. Fora detailed description of the sampling techniques andlimitations, see Lees and Peres (2006, 2008a) and Peresand Michalski (2006).

Gap-crossing movements

From July to September 2006, 140 h of observations werecarried out at 10 observation stations overlooking gaps of4�425 m in width (mean9SD�959169 m) from theedge of undisturbed primary forest patches �1000 ha,which were likely to retain a near complete avifaunalassemblage, i.e. �90% of the expected forest speciesrichness (Lees and Peres 2006). All observations werecarried out by a single observer (AL) with over 22 yearsof field-ornithology experience, including �420 field-daysin the Alta Floresta region. All gap sites were located alongquiet unpaved forest roads, with maximum motor-vehicletraffic of 20 passes per day (mean�5.36, SD�5.06). Afteran initial trial period when observations were spread outacross all times of the day, sampling effort was restricted to

281

05:40�10:00 h (five visits per site) and 14:30�17:30 h (onevisit per site). The observer remained in a semi-concealedposition with a good field of visibility (1808) and recordedthe number, time, and height (understorey cross-overs,midstorey cross-overs, canopy cross-overs, above canopyflyovers) at which all forest birds crossed the gap. Observa-tions were carried out over a 200-m stretch of road, dividedup into four marked 50-m sections, and the distanceinterval (e.g. 100�200 m) at which each individual crossedwas recorded. There is an inherent size-bias in this method,however, as small birds crossing in the understorey are lesslikely to be detected and more difficult to identify thanlarge species crossing at canopy height. Although werecorded all gap-crossing events, we therefore excludedrecords of small-bodied understory species from distancesbeyond 50 m and all small-bodied canopy species from the100�150 m and 150�200 m distance classes. Similarly,because observations of species crossing very narrow gaps ator above canopy level were partly obstructed by the almostcontiguous canopy, only above-canopy passes within thefirst 100 m were retained in the analysis. In order tofacilitate comparison between all species and all sites wecalculated a rate per gap-perimeter distance per unitobservation time (passes m�1 h�1) during the peak activityperiod for all species.

Here we consider all forest-dependent taxa, excludingall non-forest taxa (Stotz et al. 1996), all obligate water-birds, raptors, nocturnal species and aerial insectivores, allof which were incompletely sampled. We assessed in-dividual species’ potential predisposition for gap-crossingusing seven independently derived morpho-ecologicaltraits: body mass (after Hilty 2002, Terborgh et al.1990); geographic range size (Birdlife International

2006); degree of primary forest dependence (Lees andPeres 2006); flocking tendency (after Hilty 2002, pers.obs.); number of habitats used, patchiness across a speciesrange and relative abundance (Stotz et al. 1996). Inaddition, we also used the number of forest patches inwhich a species was detected to test the hypothesis thatpatch occupancy may be inextricably linked to gap-crossing ability. In order to correct for any potentialeffect of abundance on gap-crossing rates, we derived anabundance index based on the number of detections ofeach species in nine forest patches and control sites largerthan 1000 ha, the minimum patch size of the source gap-crossing sites used in this analysis. However most of thedetections recorded in these surveys were recordedacoustically so that there may be a bias against certainnon/rarely-vocal canopy species such as cotingas. Wetherefore assume that species that were detected less oftenwere rarer than those recorded more regularly, althoughthis may not always be the case (MacKenzie et al. 2003).To prevent this possible bias, we present multi-modelinference analyses (MMI) both including and excludingour measure of abundance.

Forest patch metrics

Following an unsupervised classification of a 2004 Landsatimage (ETM 227/067), we unambiguously resolved eightmutually exclusive land-cover classes (ranging from closed-canopy forest to bare-ground). The image was georefer-enced based on a 1996 satellite image with a 5-m accuracyand a spatial resolution of 15 m. Patch size and landscapemetrics were extracted from the image using Fragstats ver.

Figure 1. Location of the study region around Alta Floresta, northern Mato Grosso, Brazil showing the location of the 31 surveyed forestpatches (filled black or circled), the 10 gap-crossing sites (narrow white rectangles) and surrounding forest (gray) and non-forest matrix(white).

282

3.3 (McGarigal et al. 2002) and ArcView ver. 3.2. Forincompletely isolated forest patches surveyed (3 of 30patches), we artificially eroded the narrowest connections �usually consisting of riparian corridors to other forestpatches � in order to calculate the total patch area. Erosionof connections was carried out across the narrowest groupsof pixels. Patch metrics data for our largest site (locatedwithin vast tracts of continuous forest north of the TelesPires River) were based on an arbitrary area equivalent toone order of magnitude larger than our largest fragmentand linear regression models based on all other patches(Lees and Peres 2006). Gap widths were measured in situand also independently using our classified images withina GIS.

Data analysis

To assess the relationships between gap-width (log10 m)and number of bird species traversing gaps (all birds,passerines, non-passerines and at different forest stratalevels), we performed linear regression models consideringall 10 gap-crossing sites. We used Pearson correlations toinvestigate the relationship between patch occupancy andgap-crossing rate for all species. Interspecific variation inboth gap-crossing rates and distances crossed during eachgap-crossing event were compared by summing and thenlog-transforming (log10 x�1) all distances crossed by eachspecies. We log-transformed species body mass but useduntransformed values for all other species traits. Catego-rical variables (flock and primary-forest dependence) wereconverted into dummy variables for inclusion in theanalysis. We investigated the variation in the structure ofthe bird community crossing different-sized gaps usingnonmetric multidimensional scaling (NMDS) ordinations,with the Bray�Curtis dissimilarity measure (sqrt-trans-formed and unstandardised) using the PRIMER software(Carr 1996).

We used model selection methods based on Akaike’sinformation criterion (AIC: Burnham and Anderson 2002).To identify which life-history variables were the bestpredictors of gap-crossing rates, we used an all-subsetsapproach with general linear models. This MMI approachusing model-averaged parameter estimates typically pro-vides better precision and reduced bias relative to theestimator of a given parameter from only the AIC-selectedbest model. Models were first ranked by second-order AIC(AICc � corrected for small sample size) differences (Di;Burnham and Anderson, 2002); the lower the AICc valuethe more likely the model approximates the data. Modelswere ranked by rescaling the AICc values such that themodel with the minimum AICc had a value of 0. Relativelikelihood of each model in a candidate set was thenestimated with Akaike weights (wi; Burnham and Anderson2002). The wi values for all models in a candidate set sumto one. Akaike weights were used to generate model-averaged parameter estimates and confidence sets. A 95%confidence set of models was constructed beginning withthe model with the highest wi and then continuouslyadding the model with the next highest weight, until thecumulative sum of weights exceeded 0.95 (Burnham andAnderson 2002). We used the unconditional variance

estimator to calculate confidence intervals. Model averagingis based on wi values for each model, therefore includingmodel selection uncertainty in the estimate of eachparameter and its associated variance. We calculated therelative variable importance (swi) by summing the Akaikeweights for all models containing the variable (Burnhamand Anderson 2002). Data analyses were undertaken withthe STATA 8 �2 statistical package (STATA Corp., CollegeStation, TX).

We analyzed models incorporating four different mea-sures of gap-crossing: 1) gap-crossing rate (passes m�1

h�1); 2) gap-crossing rate, but controlling for speciesabundance within source patches; 3) maximum gap-widthcrossed; and 4) number of gap sites crossed. Finally, for31 species observed crossing gaps at least 20 times we used amore robust measure of gap-crossing ability that takes intoaccount the entire probability distribution of gap-crossingevents. We therefore modelled the effective gap-width(EGW) each species was able to traverse, which was derivedon the basis of the best model-fit derived within Distancever. 5.0 (Thomas et al. 2005). The variation in these EGWestimates was highly correlated with those from a lognormalfit and the Mean�1 SD of the distances observed in eachindependent gap-crossing event.

Results

Forest bird responses to gaps

We recorded 5436 gap-crossing movements, only 4479 ofwhich could be identified at the species-level (231 species;mean9SD�16.1944.3; range 1�484 individuals). Of all269 bird species recorded in the forest patch surveys, 74species were never recorded crossing gaps, whereas threespecies recorded during gap-crossing surveys were neverrecorded in the patches. Considering all 5436 observations,gap-width was a significant determinant of the rate (passesm�1 h�1) at which forest birds crossed forest gaps (R2�0.809, pB0.001, n�10). If these observations are splitinto either passerines or non-passerines then the relation-ships are highly significant in both cases but weaker for thelatter (passerines: R2�0.918, pB0.001; non-passerines:R2�0.554, p�0.008, Fig. 2). Gap-width was also asignificant determinant of the rate of canopy cross-overs(R2�0.740, pB0.001, f�0.055�0.013x, SE�0.0066),midstorey cross-overs (R2�0.861, pB0.001, f�0.033�0.0012x, SE�0.0038) and understorey cross-overs (R2�0.705, pB0.001, f�0.061�0.0232x, SE�0.0115), butnot of the rate of ‘canopy flyovers’ (R2��0.003, p�0.353; f�0.017�0.002x, SE�0.0048, Fig. 3). Abundancein control forest sites was not a significant predictor for thegap-crossing rate of individual species (R2�0.005, n�189, p�0.163).

Gap crossing and patch occupancy

We recorded a total of 269 forest bird species in the 31forest patches, species richness per patch varied between 16and 213 species (mean9SD�87.2953.6). Occurrence ofindividual species across the 31 forest patches was positivelycorrelated with the number of linear gaps across which the

283

same set of species were recorded (r�0.493, pB0.001,n�214). However, if this subset is split into non-passerines(r�0.626, pB0.001, n�75) and passerines (r�0.288,p�0.001, n�139) the relationship becomes muchstronger for non-passerines (Fig. 4, Table 1). Despiteconsiderable interspecific variation, minimum patch-arearequirements decreased with gap-crossing ability (Fig. 5).NMDS scores indicate that species assemblages crossing

gaps at different sites converged in similarity according togap width; 90.4% of the variation in the first NMDS axiscan be explained by gap-width (r�0.904, pB0.001, n�10). This suggests that there was a predictable subset ofspecies crossing gaps of different widths, and assemblagesconverged in similarity with increasing gap-widths as nestedsubsets of species capable of crossing increasingly wider gapsdecreased.

Gap crossing and species life-history traits

Bird species in different families exhibited differing gap-crossing abilities (Fig. 6). Highly mobile taxa such asparrots and other families were relatively insensitive to largegaps, whereas antbirds and other families with low flightcapacity were rarely observed crossing wide clearings. Themodel combinations revealed a high level of modeluncertainty, with 162 models in the 95% confidence setfor gap-crossing rate (without controlling for abundance)(swi�0.95); 280 models in the 95% confidence set forgap-crossing rate (controlling for abundance); 116 modelsin the 95% confidence set for number of gaps crossed; and280 models in the 95% confidence set for maximum gapcrossed. The best approximating single model for gap-crossing rate (without abundance control) contained totalnumber of fragments, body mass, flocking tendency, andnumber of habitats used but only provided an Akaikeweight of 0.0416 as strength of evidence, suggesting that

Figure 2. Effect of gap width on the species richness of forestbirds: passerines (solid squares) f�73.877�25.744x, SE�3.12and non-passerines (solid circles) f�51.663�13.163x, SE�6.12crossing variable-sized forest gaps.

Figure 3. Relationships between gap-crossing rates and gap-width for forest birds (A) flying over the forest canopy, (B) crossing gaps atcanopy height, (C) crossing gaps in the midstorey, and (D) crossing gaps in the understorey or at ground-level.

284

this model was not statistically distinguishable from lowerranking models. The second model, for instance, did notinclude flocking tendency. The top 11 models had DiB2and were therefore considered the most parsimonious.

Akaike weights provide an overall measure of importanceof the variables in explaining the data (Table 2, Fig. 7).Body mass was consistently ranked as the most importantpredictor variable across all four analyses, with larger-bodied

birds crossing gaps more frequently. Forest patch occu-pancy, the most important predictor of gap-crossing rate,was not important once abundance was controlled for.However, this was the most important predictor for boththe maximum width and number of gaps crossed, which areless susceptible to an abundance bias. Species occupying alarge number of patches also exhibited the highest gap-crossing rates. Primary forest dependence and flocking

Figure 4. Relationship between the number of gaps crossed and forest patch occupancy for (a) 75 non-passerine and (b) 139 passerinespecies.

Table 1. The 15 most and 15 least ubiquitous forest bird species from a sample of 31 forest patches surveyed, and their respective gap-crossing abilities, expressed as four different metrics.

Species Overall patchoccupancy

(%)

Number ofgap-crossing events(and total number

of individuals)

Maximumgap widthcrossed

(m)

Sum of allgap-crossing

distances(m)1

Mean�1 SD ofgap-crossingdistance (m)

Highest patch occupancy (15 species)Chestnut-fronted macaw Ara severa 100 94 (280) 425 12230 275.3Blue-headed parrot Pionus menstruus 100 163 (450) 425 16801 265.4Squirrel cuckoo Piaya cayana 100 15 (44) 75 222 �Yellow-crowned parrot Amazona ochrocephala 96.8 15 (40) 425 1895 �Yellow-tufted woodpecker Melanerpes cruentatus 96.8 15 (34) 425 727 �Chestnut-eared aracari Pteroglossus castanotis 93.5 22 (80) 425 1233 173.0Reddish hermit Phaethornis ruber 90.3 69 (146) 425 1472 98.6Lineated woodpecker Dryocopus lineatus 90.3 7 (13) 425 1092 �Buff-throated woodcreeper Xiphorhynchus guttatus 90.3 4 (12) 12 39 �Thrush-like wren Campylorhynchus turdinus 90.3 3 (3) 75 225 �Blue and yellow macaw Ara ararauna 87.0 25 (118) 425 3725 308.0White-eyed parakeet Aratinga leucophtalmus 87.0 84 (432) 425 8615 236.1Red-billed toucan Ramphastos tucanus 87.0 27 (76) 400 296 �White-shouldered antshrike Thamnophilus aethiops 87.0 7 (19) 7 46 �White-tailed trogon Trogon viridis 86.6 20 (2) 12 147 10.5

Lowest patch occupancy (15 species)Bare-faced currasow Crax fasciolata 3.2 0 (0) 0 0 �Scarlet-shouldered parrotlet Touit huettii 3.2 2 (5) 9 16 �Pavonine cuckoo Dromococcyx pavoninus 3.2 0 (0) 0 0 �Black-bellied thorntail Popelairia langsdorffi 3.2 5 (5) 7 27 �Brown-banded puffbird Notharchus ordii 3.2 0 (0) 0 0 �Rufous motmot Bariphthengus martii 6.5 0 (0) 0 0 �Peruvian recurvebill Simoxenops ucayalae 3.2 0 (0) 0 0 �Para foliage-gleaner Automolus paraensis 6.5 0 (0) 0 0 �Slender-billed xenops Xenops tenuirostris 3.2 0 (0) 0 0 �Grey-throated leaftosser Sclerurus albigularis 3.2 1 (1) 8 8 �Black-banded woodcreeper Dendrocolaptes picumnus 3.2 3 (3) 12 30 �Striated anthrush Chamaeza nobilis 3.2 0 (0) 0 0 �Variegated antpitta Grallaria varia 3.2 0 (0) 0 0 �Chestnut-belted gnateater Conopophaga aurita 3.2 0 (0) 0 0 �Yellow-shouldered grosbeak Parkerthraustes humeralis 3.2 0 (0) 0 0 �

1 These values are expressed as zero for species that were never observed crossing any gap sites.

285

tendency were also consistently among the most highly-ranked predictors, with forest-dependent and obligate flock-following birds being less likely to cross gaps.

Only 31 species crossed gaps on �20 occasions, and thebest measure, explaining most of the variance in theireffective gap-width was the mean�1SD of gap-widthscrossed. Subsequent analyses considering only this subset ofspecies produced weaker models fits: the best model forgap-crossing rate had an Akaike weight of only 0.0267 asstrength of evidence estimator. However, the best predictorswere entirely consistent with the subset identified inprevious models considering a much larger set of species.

Discussion

Our results are consistent with most tropical forest studieselsewhere which suggest that gap-crossing behaviour is bothhighly species and guild specific (Dale et al. 1994, Develeyand Stouffer 2001, Laurance et al. 2004). Habitat general-ists and edge-species were most willing to cross-gaps,whereas forest specialists were inhibited from crossingeven narrow gaps. The most adept subset of gap-crossingspecies largely matched that of highly ubiquitous speciesexhibiting high occupancy rates even in very isolated smallfragments. The exceptions to this rule are a few large-bodiedarea-sensitive species such as canopy cotingas and cracids.Bird species exhibiting the double jeopardy traits of lowgap-crossing ability but requiring large areas are likely to bemost susceptible to local extinction in a highly fragmentedforest landscape. These species would also violate a moregeneral positive correlation between gap-crossing ability andarea requirements (Dale et al. 1994, Pereira et al. 2004).

Species persisting in forest fragments north of Manaus,Brazil, are generally less mobile, rarely cross gaps, yet candisperse farther from fragments than in continuous forest(Van Houtan et al. 2007). It could be counter-intuitivelyargued therefore that species that fail to disperse too far

within the continuous forest respond better to landscapefragmentation (Fahrig 2007). Conversely, extinction-pronegregarious species, such as ant-followers and mixed-speciesflock-followers, were found to undertake more long-rangemovements. However, the narrow subset of species sampledin the mark-recapture Manaus study was predominantlycomprised of small understorey passerines and near-passer-ines that make up the bulk of mist-net captures. Yet birdcommunities in small forest fragments are typically domi-nated by ‘transient’ medium-sized frugivores/omnivores andedge-specialists endowed with high flight-capacity that canpersist therein by using multiple patches in order to meettheir daily metabolic requirements (Lees and Peres 2008b).Dispersal is likely to represent a more critical element of thelife-history of species tracking spatially patchy and ephem-eral resources (e.g. fruits) and many of these species maytherefore be evolutionary pre-adapted to fragmented land-scapes. However, even for highly vagile species, theenergetic costs of trap-lining multiple patches may representa mechanism by which fragmentation reduces the repro-ductive potential of species occupying small and isolated(but potentially high-habitat quality) patches (Hinsley2000). Understorey dwelling species were most affectedby increasingly wider gaps, whereas medium-large bodiedcanopy species that make daily long-distance foragingforays, flying high above the canopy (e.g. psittacids),showed little or no evidence of reduced movement ratesacross wider gaps.

Whilst avian assemblage similarity increased with in-creasing gap-width, assemblage similarity decreased withdecreasing forest fragment size (Lees and Peres 2006).Smaller-bodied species persisting in small forest fragmentstypically comprise a subset of ruderal and/or edge specieswhose niche requirements are either met by a small patch orsubsidized by the non-forest matrix (Lees and Peres 2006,2008b). Gap-crossing may therefore be less important forthis subset of ‘resident’ species (Van Houtan et al. 2007).Some small-bodied species such as disturbance-toleranttanagers were among the most frequently recorded gap-crossers, and these species were also over-represented inother studies (Silva et al. 1996). Fahrig (2007) suggestedthat species that evolved in patchy habitats surrounded byan inhospitable matrix should have stronger boundaryresponses than those that evolved in areas of continuoushigh-quality habitat, within which long-range movementsare less risky. Movement probabilities for these speciesshould be just low enough to permit recolonization oftemporarily vacant patches and displacement distancesshould be relatively short due to the high risk of movement.However, species that evolved using patchy resources mayhave a tendency to vacate patches in a newly fragmentedhabitat, which may result in many individuals neverreaching the next patch (Fahrig 2007).

In the Alta Floresta landscape, the degree of primaryforest dependence (and number of habitats used) bydifferent forest species are extremely important in structur-ing avian communities within forest patches (Lees and Peres2006). Primary forest dependence was a powerful predictorof gap-crossing ability (and patch occupancy) but therewere exceptions to this rule. For example, some uncommonprimary forest species, such as great jacamar Jacameropsaurea and black-girdled barbet Capito dayi, were recorded

Figure 5. Relationship between gap-crossing rates (log10-trans-formed sum of the distance traversed in all gap-crossing events)and minimum size of forest patches occupied for 269 forest birdspecies, 74 of which were never observed crossing gaps (solidtriangles). Gap-crossing observations were obtained for theremaining 194 species (solid circles).

286

crossing gaps of up to 75 m on multiple occasions, butneither species was recorded in forest patches B20 ha.Therefore, although these species are physically capable ofreaching weakly-isolated small patches, the patches them-selves may not meet their minimum ecological require-ments.

We observed six gap-crossing events by cinereousThamnomanes caesius and saturnine T. saturninus an-tshrikes. These species are flock-leaders of understoreymixed-species flocks, a guild of species considered to beparticularly fragmentation-sensitive (Stratford and Stouffer1999, Laurance et al. 2004, Laurance and Gomez 2005).Mixed-species understorey flocks were regularly recorded

approaching forest edges, but never recorded crossing gaps,suggesting that most flock territories were bounded by roads(Develey and Stouffer 2001). Gap-crossing events wererarely recorded for the obligate ant-following antbirds (e.g.bare-eyed antbird Rhegmatorhina gymnops: five events;black-spotted bare-eye Phlegopsis nigromaculata: fourevents). However, members of this guild are naturallyuncommon and, as their gap-crossing rates were greaterthan that of the much more abundant Thamnomanesantshrikes, they may be less gap-shy than understoreyflock-followers. Even the 100�400 m wide Teles PiresRiver in Alta Floresta does not represent an effective fluvialbarrier to gene flow in black-spotted bare-eyes, whilst many

Figure 6. Relationships between species richness and gap-crossing rate for different forest bird groups: (a)�pigeons (Columbiformes),(b)�parrots (Psittaciformes), (c)�hummingbirds (Trochilidae), (d)�trogons (Trogonidae), e�toucans and aracaris (Ramphastidae),(f)�antbirds (Thamnophilidae), g�tanagers (Thraupidae), (h)�New World blackbirds (Icteridae).

287

other small understorey passerines show considerablegenetic and morphological differentiation on either side ofthis river (Bates et al. 2004). This observation is all themore important because rivers are thought to be lesspermeable than other barriers to birds (St. Clair 2003).

Reluctance to cross gaps may also stem from differentmechanisms of edge avoidance. For example, diurnalraptors and other predators regularly hunt along forestedges (Lees unpubl.). This edge-related risk is likely to bemost pronounced in small passerines as bird vulnerability topredation by raptors is size-dependent (Gotmark and Post1996). It is also conceivable that aggressive territorial

behaviour by invasive ruderal species, such as house wrenTroglodytes aedon and silver-beaked tanager Ramphoceluscarbo, could potentially strengthen edge-avoidance in someinterior species before they have the opportunity to attempta crossing. Critically, it is difficult to address the question ofwhether many birds are actually unable to cross smallclearings, or whether reduced movements across roads (Railet al. 1997, Develey and Stouffer 2001, Laurance et al.2004) simply reflect an alignment of individual homeranges along road edges. Normal territorial defence byresident species would therefore tend to reduce road-crossing movements (Bakker and van Vuren 2004). How-ever, for some guilds such as mixed-species flock-followers,regeneration of tall regrowth along road verges ‘softens’ theforest edge, such that the entire road can be incorporatedinto the flock’s home range (Develey and Stouffer 2001).

Despite our exceptionally robust survey effort, we likelywitnessed only a small number of rare dispersal events. Suchlong-range dispersal events in rare species may regularlyoccur but at a relatively low frequency (Murray 1967).Most our observations may therefore relate to routinehome-range commuting across gaps, rather than dispersalgap-crossing (Grubb and Doherty 1999). However, con-sidering the congruence between our results and those ofprevious studies of gap-crossing behaviour and sensitivity tofragmentation, they likely reflect the general trends in forestbird responses to non-forest gaps. Our findings reinforceLaurance and Gomez’s (2005) conclusions based ontranslocation experiments that narrow forest roads mayinhibit regular movement patterns in some understoreyforest birds, but do not present a major barrier to rarerdispersal movements and ultimately meta-population dy-namics. Wider gaps, however, may represent a near-complete barrier to gene-flow for all but the mostcompetent gap-crossers. Moreover, species that are typicallypoor gap-crossers are also most range-restricted and there-fore of most conservation concern. Ultimately, speciespersisting in heavily fragmented landscapes must eithertolerate small forest patches or be adept at moving betweenthem, whereas the most extinction-prone species are thosethat are not evolutionarily predisposed to satisfy eitherof these requirements.

Tropical deforestation frontiers such as southern Ama-zonia have already become heavily fragmented landscapes,where movements of forest wildlife between small popula-tions can be severely restricted. For example, the mean gap-width between neighbouring forest fragments �1 ha (n�17 211) distributed throughout a 437 000-km2 region ofAmazonian Mato Grosso is 3.91 km (Irgang et al. 2007).Such prohibitively wide gaps will virtually preclude patch-matrix movements of all but the most mobile species. At apolicy level, the impacts of roads on sensitive bird speciescan be mitigated against by permitting forest regenerationalong road verges and establishing continuous canopy coverover narrow forest roads. Across wider gaps filled by aninhospitable matrix, the maintenance of ‘keystone struc-tures’ (Tews et al. 2004), such as isolated clusters of trees orthe creation of live fences (Leon and Harvey 2006) mayfacilitate successful matrix negotiation for commuters andrare dispersal events. These ‘stepping-stones’ are, however,no substitute for either maintaining or recreating appro-priate structural and functional connectivity within the

Table 2. Model weights from information-theoretic analysis of patchand life-history characteristics of birds crossing variably sized forestgaps, from a total of 1023 possible models.

Variable aswi b SE

Basic modelNumber of habitats 0.466 0.000484 0.000708Geographic range size 0.292 0.0000000000509 0.000000000144bPrimary forest

dependence1,20.388 0.000875 0.00165

Primary forestdependence2,3

0.374 0.000885 0.00153

Flocking1 0.810 �0.000979 0.00178Flocking2 0.390 0.00375 0.00246Patchiness 0.332 0.000658 0.00133Relative abundance 0.287 �0.000197 0.000570Body mass 0.995 �0.00560 0.00159Total fragments 1.00 0.000718 0.000112

Abundance controlNumber of habitats 0.340 0.000333 0.00503Geographic range size 0.637 0.000000000564 0.00000000844Primary forest

dependence1,20.490 0.00543 0.0812

Primary forestdependence2,3

0.744 0.00224 0.0335

cFlocking1 0.960 0.000412 0.00650Flocking2 0.250 0.00780 0.1166Patchiness 0.271 �0.000339 0.00538Relative abundance 0.826 0.00360 0.0538Body mass 0.952 �0.00606 0.0906Total fragments 0.327 �0.0000425 0.000643

Maximum gap crossedNumber of habitats 0.677 0.0467 0.0470Range size 0.274 �0.00000000139 0.00000000541Primary forest

dependence1,20.863 0.460 0.145

Primary forestdependence2,3

0.991 0.207 0.119

Flocking1 0.533 0.0112 0.0429Flocking2 0.303 0.0690 0.0877Patchiness 0.255 �0.00175 0.0273Relative abundance 0.263 0.00246 0.0183Body mass 1.000 0.310 0.0681Total fragments 1.000 0.0211 0.00478

Total number of gaps crossedNumber of habitats 0.501 �0.110 0.0470Geographic range size 0.376 0.0000000230 0.00000000541Primary forest

dependence1,20.966 1.589 0.145

Primary forestdependence2,3

0.988 1.140 0.119

Flocking1 0.440 �0.126 0.0429Flocking2 0.379 0.173 0.0877Patchiness 0.379 0.0606 0.0273Relative abundance 0.263 0.000948 0.0183Mass 0.684 0.338 0.0681Total fragments 1.000 0.134 0.00478

aswi (Akaike weights) for all models with a given variablebsuperscripts denote categorical binary ‘dummy’ variables; primaryforest dependence 1,2, for species most dependent on primary forestand primary forest dependence2,3 for less primary forest dependentspeciescFlocking1 including all obligate flock-following species and Flock-ing2 all facultative flock-following species

288

landscape (Lees and Peres 2008a), which will be critical forthe persistence of the most gap-shy and matrix-intolerantspecies in fragmented forest landscapes. Birds are generallyconsidered to be an extremely vagile group; the effects offragmentation will be undoubtedly more severe in mostforest organisms such as large-seeded plants, flightlessinvertebrates, and non-volant arboreal vertebrates.

Acknowledgements � This study was funded by the NaturalEnvironment Research Council (UK) and a small grant from theCenter for Applied Biodiversity Sciences, Conservation Interna-tional (USA). We thank Vitoria da Riva Carvalho and theFundacao Ecologica Cristalino for critical support during thestudy; Geraldo Araujo, Simon Mahood and Fernanda Michalskifor logistical help, Katherine Boughey, Toby Gardner and JosBarlow for comments on the manuscript and all the landownersand people of Alta Floresta for their unreserved cooperation.

References

Ambuel, B. and Temple, S. A. 1983. Area-dependent changes inthe bird communities and vegetation of southern Wisconsinforests. � Ecology 64: 1057�1068.

Andren, H. 1992 Corvid density and nest predation in relation toforest fragmentation: a landscape perspective. � Ecology 73:794�804.

Andren, H. 1994. Effects of habitat fragmentation on birds andmammals in landscapes with different proportions of suitablehabitat: a review. � Oikos 71: 355�366.

Bakker, V. J. and van Vuren, D. H. 2004. Gap-crossing decisionsby the red squirrel, a forest-dependent small mammal.� Conserv. Biol. 18: 689�697.

Bates, J. M. et al. 2004. Avian mitochondrial DNA sequencedivergence across a headwater stream of the Rio Tapajos, amajor Amazonian river. �J. Ornithol. 145: 199�205.

BirdLife International. 2006. Data Zone species accountsBhttp://www.birdlife.org/datazone/index.html�. AccessedSeptember 2006.

Brown, J. H. and Kodric-Brown, A. 1977. Turnover rates ininsular biogeography: effect of immigration on extinction.� Ecology 58: 445�449.

Burke, D. M. and Nol, E. 1998. Influence of food abundance,nest-site habitat, and forest fragmentation on breeding oven-birds. � Auk 115: 96�104.

Burnham, K. P. and Anderson, D. R. 2002. Model selection andmultimodel inferences. A practical information-theoretic ap-proach (2nd ed.). � Springer.

Carr, M. R. 1996. PRIMER: Plymouth routines in multivariateecological research. � Plymouth Mar. Lab., Plymouth, UK.

Dale, V. H. et al. 1994. Relating patterns of land-use change tofaunal biodiversity in the central Amazon. � Conserv. Biol. 8:1027�1036.

Develey, P. F. and Stouffer, P. C. 2001. Effects of roads onmovements by understory birds in mixed-species flocks incentral Amazonian Brazil. � Conserv. Biol. 15: 1416�1422.

Didham, R. K. 1997. The influence of edge effects and forestfragmentation on leaf-litter invertebrates in central Amazonia.� In: Laurance, W. F. and Bierregaard Jr., R. O. (eds),Tropical forest remnants: ecology, management, andconservation of fragmented communities. Univ. of ChicagoPress, pp. 55�70.

Dunning, J. B. et al. 1992. Ecological processes that affectpopulations in complex landscapes. � Oikos 65: 169�175.

Fahrig, L. 2007. Non-optimal animal movement in human-alteredlandscapes. � Funct. Ecol. 21: 1003�1015.

Ferraz, G. et al. 2007. A large-scale deforestation experiment:effects of patch area and isolation on Amazon birds. � Science315: 238�241.

Figure 7. Ranking of explanatory variables according to the sum of the Akaike weights (swi) for each variable, (a) gap-crossing rateregardless of abundance, (b) gap-crossing rate accounting for species abundance, (c) maximum gap-width crossed, and (d) number of gapscrossed.

289

Gotmark, F. and Post, P. 1996. Prey selection by sparrowhawks,Accipiter nisus: relative predation risk for breeding passerinebirds in relation to their size, ecology and behaviour. � Philos.Trans. R. Soc. Lond. B 351: 1559�1577.

Greenberg, R. 1989. Neophobia, aversion to open space, andecological plasticity in song and swamp sparrows. � Can. J.Zool. 67: 1194�1199.

Greenwood, P. J. and Harvey, P. H. 1982. The natal and breedingdispersal of birds. � Annu. Rev. Ecol. Syst. 13: 1�21.

Grubb, T. J. and Doherty, P. F. 1999. On home-range gapcrossing. � Auk 116: 618�628.

Hilty, S. L. 2002. Birds of Venezuela. � Princeton Univ. Press.Hinsley, S. A. 2000. The costs of multiple patch use by birds.

� Landscape Ecol. 15: 765�775.Irgang, G. L. et al. 2007. Analise da fragmentacao da paisagem na

Amazonia Mato-Grossense. � Unpubl. Rep. Inst. Centro deVida, Alta Floresta, Brazil.

Kapos, V. 1989. Effects of isolation on the water status of forestpatches in the Brazilian Amazon. � J. Trop. Ecol. 5: 173�185.

Karr, J. R. 1982. Population variability and extinction in theavifauna of a tropical land bridge island. � Ecology 63: 1975�1978.

Killeen, T. J. 2007. A perfect storm in the Amazon wilderness:development and conservation in the context of the Initiativefor the Integration of the Regional Infrastructure of SouthAmerica (IIRSA). Adv. Appl. Biodiv. Ser., No. 7. � Center forAppl. Biodiv. Sci. (CABS), Conserv. Int., USA.

Laurance, W. F. and Williamson, G. B. 2001. Positive feedbacksamong forest fragmentation, drought, and climate change inthe Amazon. � Conserv. Biol. 15: 1529�1535.

Laurance, W. F. et al. 2002. Ecosystem decay of Amazonian forestfragments: a 22-year investigation. � Conserv. Biol. 16: 605�618.

Laurance, S. G. W. and Gomez, M. S. 2005. Clearing width andmovements of understory rainforest birds. � Biotropica 37:149�152.

Laurance, S. G. W. et al. 2004. Effects of road clearings onmovement patterns of understory rainforest birds in centralAmazonia. � Conserv. Biol. 18: 1099�1109.

Leon, M. C. and Harvey, C. A. 2006. Live fences and landscapeconnectivity in a neotropical agricultural landscape. � Agrofor.Syst. 68: 15�26.

Lees, A. C. and Peres, C. A. 2006. Rapid avifaunal collapse alongthe Amazonian deforestation frontier. � Biol. Conserv. 133:198�211.

Lees, A. C. and Peres, C. A. 2008a. Conservation value of remnantriparian forest corridors of varying quality for Amazonian birdsand mammals. � Conserv. Biol. 22: 439�449.

Lees, A. C. and Peres, C. A. 2008b. Avian life history determinantsof local extinction risk in a fragmented neotropical forestlandscape. � Anim. Conserv. 11: 128�137.

Lima, S. L. and Dill, L. M. 1990. Behavioral decisions made underthe risk of predation: a review and prospectus. � Can. J. Zool.68: 619�640.

Lima, S. L. and Zollner, P. A. 1996. Towards a behavioral ecologyof ecological landscapes. � Trends Ecol. Evol. 11: 131�135.

MacKenzie, D. I. et al. 2003. Estimating site occupancy,colonization, and local extinction when a species is detectedimperfectly. � Ecology 84: 2200�2207.

McGarigal, K. et al. 2002. FRAGSTATS: spatial pattern analysisprogram for categorical maps. � Univ. of Massachusetts,Amherst. Bwww.umass.edu/landeco/research/fragstats/fragstats.html� accessed September 2006.

Murcia, C. 1995. Edge effects in fragmented forests: implicationsfor conservation. � Trends Ecol. Evol. 10: 58�62.

Murray, B. G. Jr. 1967. Dispersal in vertebrates. � Ecology 48:975�978.

Opdam, P. 1991. Metapopulation theory and habitat fragmenta-tion: a review of Holarctic breeding bird studies. � LandscapeEcol. 5: 93�106.

Pereira, H. M. et al. 2004. A framework for assessing thevulnerability of species to land-use change. � Ecol. Appl. 14:730�742.

Peres, C. A. and Michalski, F. 2006. Synergistic effects of habitatdisturbance and hunting in Amazonian forest fragments. � In:Laurance, W. F. and Peres, C. A. (eds), Emerging threats totropical forests. Univ. of Chicago Press, pp. 105�127.

Rail, J. F. et al. 1997. Territorial responses of boreal forest birds tohabitat gaps. � Condor 99: 976�980.

Reid, J. and De Souza, W. D. Jr 2005. Infrastructure andconservation policy in Brazil. � Conserv. Biol. 19: 740�746.

Sieving, K. E. et al. 1996. Habitat barriers to movement ofunderstory birds in fragmented south-temperate rainforest.� Auk 113: 944�949.

Silva, J. M. C. et al. 1996. Plant succession, landscape manage-ment, and the ecology of frugivorous birds in abandonedAmazonian pastures. � Conserv. Biol. 10: 491�503.

St. Clair, C. C. 2003. Comparative permeability of roads, rivers,and meadows to forest songbirds in Banff National Park.� Conserv. Biol. 17: 1�11.

Stotz, D. F. et al. 1996. Neotropical birds: ecology andconservation. � Univ. of Chicago Press.

Stratford, J. A. and Stouffer, P. C. 1999. Local extinctions ofterrestrial insectivorous birds in a fragmented landscape nearManaus Brazil. � Conserv. Biol. 13: 1416�1423.

Taylor, P. D. et al. 1993. Connectivity is a vital element oflandscape structure. � Oikos 68: 571�573.

Terborgh, J. et al. 1990. Structure and organization of anAmazonian forest bird community. � Ecol. Monogr. 60:213�238.

Tews, J. et al. 2004. Animal species diversity driven by habitatheterogeneity/diversity: the importance of keystone structures.� J. Biogeogr. 31: 79�92.

Thomas, L. et al. 2005. Distance 5.0. Rel. 4. � Res. Unit forWildlife Population Assessment, Univ. of St. Andrews, UK.Bhttp://www.ruwpa.st-and.ac.uk/distance/� accessed Sep-tember 2007.

Van Houtan, K. S. et al. 2007. Dispersal of Amazonian birds incontinuous and fragmented forest. � Ecol. Lett. 10: 219�229.

With, K. A. et al. 1997. Landscape connectivity and populationdistributions in heterogeneous environments. � Oikos 78:151�169.

290