INTRODUCTION
Tropical bird distribution and abundance patterns are often correlated to anthropogenic variation in forest cover (Gillespie Reference GILLESPIE2000, Restrepo & Gomez Reference RESTREPO and GOMEZ1998, Shahabuddin & Kumar Reference SHAHABUDDIN and KUMAR2006, Stouffer & Borges Reference STOUFFER, BORGES, Bierregaard, Gascon, Lovejoy and Mesquita2001). Researchers, however, are only beginning to describe such abundance patterns for tropical dry-forest birds (Gillespie Reference GILLESPIE2000, Shahabuddin & Kumar Reference SHAHABUDDIN and KUMAR2006). Few data have been published that characterize the structure and fauna of Panamanian dry forests (but see Griscom et al. Reference GRISCOM, KALKO and ASHTON2007). Additionally, the spatial scales at which abundance patterns correlate to tropical forest cover are poorly known (but see Pearman Reference PEARMAN2002 for scale of diversity patterns). Two goals of this study are to discover (1) species’ traits that mediate the relationship between forest cover and abundance, and (2) the spatial scale of such relationships. This information may lead to insights into the mechanisms that underlie abundance patterns (Levin Reference LEVIN1992, McGill et al. Reference MCGILL, ENQUIST, WEIHER and WESTOBY2006).
Spatial variation in Panamanian dry-forest bird communities might differ from that of tropical rain-forest bird communities for several reasons. Many Central American dry forests have been extremely fragmented and reduced in area for decades (Gillespie et al. Reference GILLESPIE, GRIJALVA and FARRIS2000). Extant fauna may be less sensitive to habitat fragmentation and edge effects than fauna found in more extensive and recently fragmented forests (Restrepo & Gomez Reference RESTREPO and GOMEZ1998, Stouffer & Borges Reference STOUFFER, BORGES, Bierregaard, Gascon, Lovejoy and Mesquita2001). Additionally, dry-forest avifauna might be less sensitive to variation in land cover and forest edges (Laurance et al. Reference LAURANCE, LOVEJOY, VASCONCELOS, BRUNA, DIDHAM, STOUFFER, GASCON, BIERREGAARD, LAURANCE and SAMPAIO2002) due to the large seasonal fluctuations in environmental conditions within dry forests, which have long (4–6 mo) and intense dry seasons and deciduous vegetation of reduced stature relative to vegetation of wetter forests (Holdridge et al. Reference HOLDRIDGE, GRENKE, HATHEWAY, LIANG and TOSI1971).
We employ a statistical model of local abundance along environmental gradients that accounts for detection error at greater distances from the observer and random Poisson error in counts (Royle et al. Reference ROYLE, DAWSON and BATES2004). This model diverges from traditional distance-sampling methods (Buckland et al. Reference BUCKLAND, ANDERSON, BURNHAM, LAAKE, BORCHERS and THOMAS2001) in several key ways. Traditional distance-sampling models often sacrifice fine-scale spatial information. Many sample points are lumped together to estimate abundance across sample sites where birds were detected (i.e. abundance conditioned on detection, Buckland et al. Reference BUCKLAND, ANDERSON, BURNHAM, LAAKE, BORCHERS and THOMAS2001). In the method of Royle et al. (Reference ROYLE, DAWSON and BATES2004), local abundance is estimated using all sample points, instead of only those where birds were detected. Thus the unconditional consideration of abundance in the model of Royle et al. (Reference ROYLE, DAWSON and BATES2004) is more biologically realistic, especially within a single habitat type, where absence is but the case when local abundance approaches zero.
We test four major predictions about how the abundance of Panamanian dry-forest birds changes along two types of spatial gradient of forest cover: per cent forest cover at three scales and distance to forest edge. First, we hypothesize that the abundance of forest-preferring species increases in dry-forest interiors and where forest coverage is high. We hypothesize secondly, however, that many species of dry-forest birds do not show abundance variation along gradients of forest cover and distance to forest edge, in contrast to avifauna of wet forests where nearly all common species may exhibit such patterns (Stouffer & Borges Reference STOUFFER, BORGES, Bierregaard, Gascon, Lovejoy and Mesquita2001). Third, we predict that in dry forests the abundance of habitat generalist species commonly found in open habitats declines along gradients of increasing distance to forest edge and forest coverage. As insectivores have been reported to be the feeding guild most sensitive to anthropogenic change in dry forests (Shahabuddin & Kumar Reference SHAHABUDDIN and KUMAR2006), we predict that insectivore abundance is positively correlated to forest cover and distance to forest edge.
METHODS
Study sites
The Panamanian tropical dry forest region covers ~5100 km2 (Olson et al. Reference OLSON, DINERSTEIN, WIKRAMANAYAKE, BURGESS, POWELL, UNDERWOOD, D'AMICO, ITOUA, STRAND, MORRISON, LOUCKS, ALLNUTT, RICKETTS, KURA, LAMOREUX, WETTENGEL, HEDAO and KASSEM2001). This region is isolated from more extensive Latin American dry forests to the north (Costa Rica) and south (Ecuador) along the coast of the Pacific Ocean. As of 1992, the dry-forest municipalities of Los Santos, Herrera and Coclé provinces (where study sites were located) had 0.8% forest coverage (Autoridad Nacional del Ambiente, http://www.anam.gob.pa/Sif%202002/index.htm).
In November and December 2004 we conducted a search for fragments of tropical dry forest using ground surveys, aerial photographs (Instituto Geográfico Nacional of Panama), satellite imagery, and scientist and resident contacts. Forest fragments were selected for study based on their relatively large size, level of protection, and/or accessibility. The five fragments studied, ranging in size from 80 to 1115 ha, represent nearly all of the largest and best-protected remaining fragments of tropical dry forest in Panama (Figure 1, Appendix 1). For reference, vegetation structure was measured at a randomly located 0.1-ha plot within each fragment following the method of Gentry (Reference GENTRY1982, Appendix 1). Mean annual precipitation at sites ranged from 1134–1702 mm y−1 (Panama: Dirección de Estadística y Censo 1970–2002), placing sites within the tropical dry-forest life zone of Holdridge et al. (Reference HOLDRIDGE, GRENKE, HATHEWAY, LIANG and TOSI1971). The matrix surrounding forest fragments was composed principally of cattle pasture, with smaller areas of crops, mangroves and secondary scrub growth.
Map of the study region showing forest coverage, with an inset map of Panama with the study region indicated. Study fragments are indicated. The thick black line represents the approximate limit of Panamanian dry forests. The more extensive forests near river mouths are mostly mangroves. Data are taken from Landsat ETM+ scenes from 2001–2002.
Forest cover
Forest fragment edges were identified using aerial photography and ground surveys with a GPS unit (Brunton Inc.). Maps of forest coverage of the region were created using Landsat ETM+ images from 2001–2002 with pixels of 30 × 30 m (U.S. Geological Survey Global Landcover Facility). We created ground-truthed training sites. Using GIS software (Idrisi Andes), we then classified pixels using standardized minimum distance to class mean and also based on ground observations and aerial photography. The map of forest cover correctly classified 92% of random forest sample sites. We used GIS software (Arc GIS, ESRI) to calculate the percentage of land covered by forest within 500 m, 1000 m and 2000 m of each bird sampling location. We also calculated the distance from each forest point count (the location of the observer) to the nearest forest edge. In calculating these metrics we did not distinguish between dry forest and mangrove forests that adjoined two of the fragments.
Bird sampling
During May and June 2005, the peak breeding season for birds in lowland Central America (Stiles & Skutch Reference STILES and SKUTCH1989), an experienced observer (J. R. Lasky) distance-sampled birds from point counts (Buckland et al. Reference BUCKLAND, ANDERSON, BURNHAM, LAAKE, BORCHERS and THOMAS2001). Counts lasted 10 min (Hutto et al. Reference HUTTO, PLETSCHET and HENDRICKS1986), during which time all individual birds identified by sight or sound within a radius of 50 m and their distance from the counting point estimated to the nearest 5 m were recorded. After counts were completed, distances were placed into four bins for statistical analyses: 0–12.5, 12.5–25, 25–37.5 and 37.5–50 m. Distance estimation by the observer was verified before counts began by comparing distance estimates to actual measurements of distance to stationary singing birds. To avoid counting the same individual twice, multiple individuals of the same species were recorded only if all were observed simultaneously. Point counts were located at least 100 m apart and 100 m from the forest edge. Thirty point counts were conducted in each fragment. Only non-aquatic and diurnally active species were counted. We completed counts within 2 h of sunrise and conducted counts over two periods of three consecutive days at each of the five fragments, alternating between fragments every 3 d.
Data analyses
Species habitat affinities are often reported qualitatively and represent expert opinion of general habitat-abundance relationships across entire countries (Ridgely & Gwynne Reference RIDGELY and GWYNNE1989, Skutch Reference SKUTCH1954, Stiles & Skutch Reference STILES and SKUTCH1989, Wetmore Reference WETMORE1965). These affinities have successfully predicted patterns of forest bird diversity relative to forest cover in Costa Rican and Nicaraguan dry-forest reserves (Gillespie & Walter Reference GILLESPIE and WALTER2001), although related patterns of abundance and their scale are unknown. Based on habitat associations reported in the literature (Ridgely & Gwynne Reference RIDGELY and GWYNNE1989, Skutch Reference SKUTCH1954, Stiles & Skutch Reference STILES and SKUTCH1989, Wetmore Reference WETMORE1965) and the expertise of an ornithologist with over 15-y experience throughout Panama (G. Angehr, pers. comm.), we divided species habitat associations into three main categories: species primarily found in forest and woodland habitats, species primarily found in open habitats (e.g. grasslands, savanna), and intermediate species falling into neither preceding category. In order to determine if the abundance of uncommon forest birds varied with landscape structure, we lumped the abundances of those forest species detected fewer than 15 times (i.e. too few to model species-specific abundance). We also classified species into dietary guilds (insectivore, insectivore–frugivore, frugivore, granivore, omnivore and raptor), based on feeding habits reported in the literature (Poulin et al. Reference POULIN, LEFEBVRE and MCNEIL1994, Skutch Reference SKUTCH1954, Stiles & Skutch Reference STILES and SKUTCH1989, Trainer & Will Reference TRAINER and WILL1984, Wetmore Reference WETMORE1965). The abundances of insectivore–frugivores, which mostly consume insects, were combined with the abundances of insectivores for analysis. Additionally, the abundances of nectarivore–frugivores were included in nectarivore abundance.
In order to account for sampling error that results from imperfect detection and random error in a Poisson process, we employed the Poisson and negative binomial regression models of Royle et al. (Reference ROYLE, DAWSON and BATES2004) that model distance decay in detection probability and the effects of local environmental covariates on local abundance. The distance sampling method assumes that detection probability decays over distance from the observer (Buckland et al. Reference BUCKLAND, ANDERSON, BURNHAM, LAAKE, BORCHERS and THOMAS2001). The model of Royle et al. (Reference ROYLE, DAWSON and BATES2004) estimates biologically interpretable, species-specific parameters for detection probability, mean local abundance, the effect of covariates on abundance, and in the case of the negative binomial, an over-dispersion parameter.
All models and statistical tests were executed with the open-source software R (R Development Core Team, www.r-project.org). We logarithmically transformed distances from point count to forest edge data and logit-transformed per cent forest cover in order to make covariate distributions more normal. All covariates were scaled to have mean equal to 0 and standard deviation equal to 1.
Using Akaike's Information Criterion (AIC), we compared null models of abundance with models containing each covariate of forest cover or the distance from the observer's location in the point count to nearest forest edge. Null abundance models estimated parameters for detection and fragment mean abundance, but did not include parameters for forest cover. Abundance was modelled for species and functional groups found in at least four of the five forest fragments and detected at least 15 times. Because there were 30 point counts within each of five dry-forest fragments, we estimated a mean abundance for each fragment in all models to account for block effects. For species that were found in only four of the five forest fragments, we excluded sample points from the fragment where it was absent because we cannot stably estimate a parameter that results in a mean log abundance equal to zero. Non-linear minimization of deviance was used to estimate parameter values to maximize joint likelihood of the data across all sites (Royle et al. Reference ROYLE, DAWSON and BATES2004).
Models selected based on AIC that included a parameter for forest cover or distance from point count to forest edge were tested for spatial autocorrelation in the deviates (−2 × log(likelihood)) at each sample site, corresponding to autocorrelation in the data not predicted by the model. Using the Mantel test, a Euclidean distance matrix of the site-specific deviates was compared with a Euclidean distance matrix of the geographical distance between sites. The null hypothesis that geographical distance is not correlated to deviate distance was tested among the sample sites of each fragment (α = 0.05). A goodness-of-fit test using parametric bootstrapping was used to verify that the model's likelihood did not deviate significantly from its respective Poisson or negative binomial distribution (α = 0.05).
RESULTS
We observed a total of 2117 individual birds and 71 species at 150 point counts (Appendix 2). Models of local abundance were compared for each of the 21 species (totalling 1429 individuals) that occurred in at least four of the five forest fragments and were detected at least 15 times (Table 1).
Selected models of log abundance and model parameters for dry-forest bird species and group abundances, with the habitat, diet guild, and number of detections (N) of each species or group. The model column gives the type of forest cover parameter in the abundance model selected by AIC, or ‘null’ if a model without a forest cover parameter was selected, and the error distribution of the selected model in parentheses. Parameters are α 1, the slope of the transformed and scaled covariate effect, and k, the negative binomial over-dispersion parameter. Null abundance models (without a parameter for forest cover, α 1) were selected for the species and groups below. The other α parameters are fragment mean abundance (individuals ha−1; A = Achotines, B = Bosque Común, C = Cerro Cercao, M = Cenegón de Mangle and Z = Cerro Zuela). If species did not occur at a site, N/A is written. Habitat preferences are F = forest/woodland, I = intermediate and O = open habitats. Guild abbreviations are ins = insectivore, fru = frugivore, nec = nectarivore, gra = granivore, omn = omnivore and rap = raptor.
Nectarivore abundance decreased as forest coverage within 1000 m increased (Table 1, Figure 2). For open-habitat species, abundance slightly increased as forest cover within 500 m increased (Figure 3), while a null model of abundance was favoured for total forest-preferring birds and total forest-preferring birds of species detected fewer than 15 times (Table 1). Null models were also selected for frugivore and insectivore abundances. Granivores were comprised almost wholly of Leptotila verreauxi, so we did not model the abundance of this group. We detected very few raptors and did not model the abundance of this group.
Estimated abundance relationships with forest coverage within a 1000-m radius of the sample points for total nectarivores (a), Amazilia edward (b), and Polioptila plumbea (c). Count values for each patch are shown, converted from raw individuals per count to individuals ha−1 (circles = Achotines, triangles pointing up = Cenegón de Mangle, diamonds = Bosque Común, triangles pointing down = Cerro Cercao, squares = Cerro Zuela). The estimated abundance curves are shown for Achotines and the range of variation in forest coverage for sample points there. Note that mean abundance parameters are estimated for each patch, the curves only show estimates for Achotines.
Estimated abundance relationships with forest coverage within a 500-m radius of the sample points for total open-habitat species (a) and Elaenia flavogaster (b). Count values for each patch are shown, converted from raw individuals per count to individuals ha−1 (circles = Achotines, triangles pointing up = Cenegón de Mangle, diamonds = Bosque Común, triangles pointing down = Cerro Cercao, squares = Cerro Zuela). The estimated abundance curves are shown for Cenegón de Mangle and the range of variation in forest coverage for sample points there. Note that mean abundance parameters are estimated for each patch, the curves only show estimates for Cenegón de Mangle.
Of the 21 species for which we modelled abundance, models incorporating a habitat spatial structure covariate were selected for 12 species, five of which had habitat effects of the opposite sign as expected based on habitat association groups (Table 1). The distance to forest edge and three forest cover covariates were all selected in at least one species abundance model (Figures 2–5). Three of the five species of forest-habitat birds, four of seven species of open habitat, and five of nine intermediate habitat species with sufficient detections to model species abundance had non-null abundance models selected.
Estimated abundance relationships with forest coverage within a 2000-m radius of the sample points for Leptotila verreauxi (a), Thamnophilus doliatus (b), Myiodynastes maculatus (c), and Thryothorus rufalbus (d). Count values for each patch are shown, converted from raw individuals per count to individuals ha−1 (circles = Achotines, triangles pointing up = Cenegón de Mangle, diamonds = Bosque Común, triangles pointing down = Cerro Cercao, squares = Cerro Zuela). The estimated abundance curves are shown for Cerro Zuela and the range of variation in forest coverage for sample points there. Note that mean abundance parameters are estimated for each patch, the curves only show estimates for Cerro Zuela.
Estimated abundance relationships with distance to forest edge from the sample points for Brotogeris jugularis (a), Lepidopyga coeruleogularis (b), Vireo flavoviridis (c), and Euphonia leuteicapilla (d). Count values for each patch are shown, converted from raw individuals per count to individuals ha−1 (circles = Achotines, triangles pointing up = Cenegón de Mangle, diamonds = Bosque Común, triangles pointing down = Cerro Cercao, squares = Cerro Zuela). The estimated abundance curves are shown for Bosque Común and the range of variation in distance to forest edge for sample points there. Note that mean abundance parameters are estimated for each patch, the curves only show estimates for Bosque Común.
Mantel tests for spatial autocorrelation among deviates from the likelihood models showed no significant spatial autocorrelation (α = 0.05) in all but two instances: at the Achotines fragment for Leptotila verreauxi, and at the Cerro Zuela fragment for Tolmomyias sulphurescens. Because of the rarity of these instances and the fact that these fragments comprise a small proportion of the total observations for these species (2% of L. verreauxi and 17% of T. sulphurescens observations), we give no further consideration to spatial autocorrelation.
DISCUSSION
Habitat-affinity groups were very poor predictors of correlations between abundance and gradients of distance to forest edge and forest cover. The abundances of total forest birds, of total uncommon forest species, and of some forest species did not decline along such gradients, while the abundances of some open habitat species did decline along such gradients. Although we expected total forest-bird abundance to decline with decreasing forest cover and distance to forest edge, we expected many species of dry-forest bird would not respond to such gradients. Overall, the abundances of only 33% of bird species with sufficient detections to model decreased with decreased forest cover and distance to forest edge. In dry forests in Costa Rica and Nicaragua, 55% of forest-preferring bird species were restricted to the five forest reserves over 1000 ha out of eight in total, while only 8% were restricted to the three smaller reserves (Gillespie Reference GILLESPIE2000). The relative insensitivity of Panamanian dry-forest birds could be due to the spatiotemporal scale of deforestation and fragmentation in the heavily deforested Panamanian dry-forest region. The time lapse after major deforestation in the region, which occurred at least 20 y before our study (Autoridad Nacional del Ambiente, http://www.anam.gob.pa/Sif%202002/index.htm), may have weakened correlations between abundance and variation in forest cover. Restrepo & Gomez (Reference RESTREPO and GOMEZ1998) found that lower abundances of certain guilds at new forest edges (<12 y old) did not occur at old edges (>40 y old). The dry-forest region in Panama is heavily deforested, even relative to deforested dry regions elsewhere in Central America (Gillespie Reference GILLESPIE2000), and the only large tract of dry forest is located at Cerro Zuela (Figure 1). As a result, species sensitive to extensive habitat loss and fragmentation may have already gone extinct in the region, leaving behind more generalist species that tolerate a wide range of conditions (Owens & Bennett Reference OWENS and BENNETT2000). None of the species we recorded is restricted to undisturbed primary forest, which does not exist in the dry-forest region of Panama. Manacus vitellinus, reported as occurring in the study region and being sensitive to deforestation in 1989 (Ridgely & Gwynne Reference RIDGELY and GWYNNE1989), was not recorded in the 9 mo J. R. Lasky spent in the field in many dry-forest fragments, nor in riparian forests near Achotines (Griscom et al. Reference GRISCOM, KALKO and ASHTON2007), and may be extirpated from Panamanian dry forests. Continuous monitoring of the region's avifauna would determine if populations of other forest-preferring species are in decline, although total forest cover in the region appears to have stabilized (Autoridad Nacional del Ambiente, http://www.anam.gob.pa/Sif%202002/index.htm).
While it has been established that open-habitat species are also found within dry forests (Ridgely & Gwynne Reference RIDGELY and GWYNNE1989), an increase in their abundance within dry forests as forest coverage increases has not been reported. We classified species as having open-habitat affinities based on literature that described abundance patterns across all habitat types at national scales, while our data describe abundance patterns within dry forests. The unexpected patterns we observed may be due to the difference in scale and the range of habitat variation studied, as the environmental factors most closely related to abundance variation may change across spatial scales (Diez & Pulliam Reference DIEZ and PULLIAM2007). Composition of adjoining matrix habitats likely played little role in affecting the abundance patterns observed, as nearly all forest was bordered by active pasture with small vegetable plots, except for a rice paddy bordering ~500 m of the edge of Cerro Zuela. Future research should determine how the demographic processes of these open-habitat species vary from dry forests to adjacent scrub and agricultural fields.
Nectarivores were the only feeding guild for which the selected abundance model contained a parameter for variation in forest cover or distance to forest edge. As forest coverage within a 1000-m radius of sample points increased, nectarivore abundance decreased. Patterns of nectarivore abundance may be related to floral resource density (Cotton Reference COTTON2007). Floral resources may be more abundant in wet-forest edges than interiors (Linhart et al. Reference LINHART, FEINSINGER, BEACH, BUSBY, MURRAY and ZUCHOWSKI1987), which may explain the apparent lack of sensitivity to forest fragmentation in trochilid nectarivores (Gillespie Reference GILLESPIE2000, Pearman Reference PEARMAN2002, Petit et al. Reference PETIT, PETIT, CHRISTIAN and POWELL1999, Restrepo & Gomez Reference RESTREPO and GOMEZ1998, Stouffer & Bierregaard Reference STOUFFER and BIERREGAARD1995a). Increases in forest coverage could decrease forest edge habitat and floral resources and create the pattern we observed. However, the estimation of abundance of highly mobile trochilids is notoriously difficult. Past estimates of trochilid abundance using mist-netting in Panamanian wet forests (~2 individuals ha−1, Robinson et al. Reference ROBINSON, BRAWN and ROBINSON2000) are much lower than our estimates (Table 1), possibly reflecting counting individuals in multiple point counts. However, we have no reason to believe our estimates are biased relative to variation in forest cover at the scales studied. Additional research should determine which specific demographic processes vary spatially for the abundance-forest cover correlations we found.
The abundances of most individual insectivore species and lumped insectivores were not correlated to forest cover gradients. In Indian dry forests, however, insectivores were the only guild that showed negative correlations to localized anthropogenic disturbance (Shahabuddin & Kumar Reference SHAHABUDDIN and KUMAR2006). This result, however, is not necessarily contradictory to ours, as we focused on large-scale gradients of forest cover. In wet forests insectivores are often the avian guild most sensitive to forest edges (Laurance Reference LAURANCE2004, Şekercioğlu et al. Reference ŞEKERCIOĞLU, EHRLICH, DAILY, AYGEN, GOEHRING and SAND2002, Stouffer & Bierregaard Reference STOUFFER and BIERREGAARD1995b), but the most sensitive group of insectivores, the terrestrial insectivores, is depauperate in Central American dry forests (Gillespie Reference GILLESPIE2000).
Species abundances and distributions may be affected by both habitat loss and habitat fragmentation, which often occur simultaneously (Fahrig Reference FAHRIG1997, Trzcinski et al. Reference TRZCINSKI, FAHRIG and MERRIAM1999). Of the five species abundance models with negative effects of deforestation or fragmentation, two had lower abundances closer to forest edges (i.e. a measure of fragmentation), while the others declined with decreasing forest coverage (i.e. a measure of habitat loss). Because the abundances of so few species declined along such gradients, we cannot evaluate whether habitat loss or forest fragmentation plays a larger role in altering bird communities in dry forests (Fahrig Reference FAHRIG1997).
This study is the first to model bird abundance along gradients of forest cover and distance to forest edge in Latin American dry forests. We have shown how the abundances of 12 bird species and two functional groups vary within Panamanian dry forest along gradients of forest cover, while the abundances of nine species and three functional groups did not significantly vary along the gradients studied. In general, species’ habitat preferences were poor predictors of spatial patterns of abundance, possibly because of the long history of deforestation in the region that may have already resulted in the local extinction of sensitive species. The use of distance sampling and Poisson and negative binomial regression to site-specific environmental covariates represents a technical advance in understanding spatial variation in abundance (Royle et al. Reference ROYLE, DAWSON and BATES2004). Studies of bird community abundance patterns (such as ours) provide a richer understanding of spatial ecology than do studies of simpler presence/absence patterns.
ACKNOWLEDGEMENTS
We thank Peter Collier whose fieldwork was essential. The comments of anonymous reviewers, M. Brown, and the Keitt lab were extremely helpful. The input of J. Andrew Royle was vital. We also thank George Angehr and the Smithsonian Tropical Research Institute, Vernon Scholey and the staff of the Achotines Laboratory, the Cultural Affairs Office of the U.S. consulate in Panama, the Instituto Geográfico Nacional of Panama, and the Autoridad Nacional del Ambiente of Panama for assistance and permission. Landowners gave permission and are greatly appreciated. The Department of Biology of Kenyon College and William Barney III provided equipment. Funding was provided by a Fulbright Research Fellowship and a University Fellowship from the University of Texas at Austin.
Study forest fragment characteristics. Vegetation data are from a series of transects covering 0.1 ha (Gentry Reference GENTRY1982).
Species (AOU 1998, Banks et al. Reference BANKS, CICERO, DUNN, KRATTER, RASMUSSEN, REMSEN, RISING and STOTZ2006) recorded during point counts in five fragments (A = Achotines, B = Bosque Común, C = Cerro Cercao, M = Cenegón de Mangle and Z = Cerro Zuela). Species are listed in the order of the AOU check-list. Total counts for each fragment where they were detected, habitat preference, and foraging guild are shown. Guild abbreviations are ins = insectivore, fru = frugivore, nec = nectarivore, gra = granivore, omn = omnivore, rap = raptor and sca = scavenger. Habitat abbreviations are F = forest preferring, I = intermediate habitat preferring and O = open habitat preferring.
