Direct versus indirect effects of habitat reduction on the loss of avian species from tropical forest fragments


Kenneth J. Feeley, Biology Department, Wake Forest University, PO Box 7325 Reynolda Station, Winston-Salem, NC 27109 USA. Tel: +617 777 4817


Tropical forest fragments typically decrease in avian diversity at rates inversely related to area. However, the mechanisms by which area reduction drives avian species loss remain poorly understood. Changes in habitat area may directly lead to species loss through stochastic fluctuations of reduced populations. Alternatively, area-dependent changes in top-down and bottom-up processes associated with fragmentation may indirectly lead to accelerated avian extinctions. For example, on land-bridge islands in Lago Guri, Venezuela, fragmentation has resulted in complex changes in the biotic environments through altered abundances of nest predators and generalist herbivores. Using path analysis, we quantified the relative importance of these indirect versus direct effects of area reduction on the rates of avian species loss from 11 fragments during the period 1993–2003. Area reduction had a direct effect on species loss but this was relatively minor compared with indirect effects, especially those mediated through changes in herbivore abundances: species loss was slowed on islands occupied by hyperdense howler monkeys and accelerated on islands with leaf-cutter ants but lacking howlers. The effects of herbivores on bird species loss are likely indirect and resulting from bottom-up processes. The primacy of indirect effects at Lago Guri suggests that the loss of species from forest fragments may be driven by active biotic processes (i.e. changes in trophic structure) and is not only a passive response to reduction in habitat area per se. These findings have important implications for the design and management of reserves aimed at protecting birds and other threatened species.


Previous studies of avian communities in remnant forest fragments have found that isolation and reduction of habitat area lead predictably to the loss of diversity (Diamond, 1972; Terborgh, 1974; Leck, 1979; Pimm, Moulton & Justice, 1994; Pimm & Askin, 1995; Meiri, Dayan & Simberloff, 2005; Sodhi et al., 2006; Stouffer et al., 2006). Furthermore, rates of species loss are typically inversely related to fragment area (Diamond, 1972; Brooks, Pimm & Oyugi, 1999; Ferraz et al., 2003, 2007). Yet despite the considerable number of reports documenting such patterns, the mechanisms behind species loss, and how they relate to area, remain poorly resolved.

Habitat fragmentation may directly drive the loss of avian species through reductions in population size and an associated increase in the risk of local extinction through stochastic failure. Simultaneously, habitat loss may also have important indirect effects on avian populations through changes in the abundance and/or composition of other species groups which in turn drive changes in the biotic environment. In temperate forests, for example, it is well established that fragmentation increases relative edge length resulting in increased exposure of forest-interior birds to edge-frequenting nest parasites and mesopredators (Brittingham & Temple, 1983; Wilcove, 1985; Andren & Angelstam, 1988; Heske, Robinson & Brawn, 1999). Exotic predators introduced by man have had devastating impacts on the birds of oceanic islands (Blackburn et al., 2004). Likewise, the role of increased nest predation in fragments has been surmised (Terborgh & Winter, 1980) or experimentally demonstrated (Loiselle & Hoppes, 1983; Sieving, 1992; Estrada, Rivera & Coates-Estrada, 2002b; Githiru, Lens & Cresswell, 2005) in the tropics. Habitat fragmentation may also lead to increased herbivory and seed predation (McEuen & Curran, 2006; Lopez & Terborgh, 2007) resulting in altered vegetation structure and productivity (Feeley & Terborgh, 2005) with subsequent effects on bird communities (McShea & Rappole, 2000; Feeley & Terborgh, 2006).

While area reduction clearly has both direct and indirect effects on species loss, quantifying the relative magnitude of these effects will have important implications for guiding conservation strategies. For example, if species loss is primarily indirect and mediated through changes in other faunal groups, then even large fragments may be insufficient to prevent the loss of avian diversity if poaching, cattle grazing, species introductions, etc. are not curtailed. Conversely, smaller fragments may potentially sustain viable bird populations if actions are taken to help minimize changes in other faunal groups (e.g., by reducing hunting and/or increasing connectivity and facilitating the movement of key species between fragments).

Here we quantify the relative importance of the direct versus indirect effects of area contraction on rates of avian species loss and local extinction using long-term data from land-bridge islands in Lago Guri, Venezuela, where fragmentation has caused dramatic changes in the abundance of many important faunal groups (Terborgh et al., 2006; Terborgh & Feeley, 2008). We conclude that the direct link between habitat area and the rate at which avian species are being lost is largely overshadowed by the indirect effects of area reduction as mediated through changes in the abundance of nest predators and especially herbivores.


Lago Guri is a hydroelectric reservoir that was created in 1986 following construction of the Raul Leoni Dam on the lower Rio Caroni in east-central Venezuela (Bolivar State, 7°20′N, 62°48′W). The inundation of over 4300 km2 of hilly terrain resulted in the fragmentation of once continuous forest into hundreds of isolated land-bridge islands (Morales & Gorzula, 1986). These islands serve as model forest fragments and offer several distinct advantages over mainland fragmentation systems in that there are a large number of replicate patches that were all isolated from a single continuous forest at the same time by a single disturbance event (Diamond, 2001). In addition, because the matrix is water, the biotic and abiotic influences of the matrix should not vary greatly between fragments (a factor which has potentially confounded many previous studies investigating the responses of faunal communities to fragmentation; Antongiovanni & Metzger, 2005; Stouffer et al., 2006). Finally, movement of non-flying organisms between patches is greatly reduced compared with mainland systems thus increasing the confidence in vertebrate and invertebrate censuses (Terborgh et al., 1997a).

We conducted our research on a set of 11 islands. Study islands ranged in area from <1 ha to over 200 ha and were isolated from the mainland by between 0.1 and 6.2 km (Table 1). Areas and distances differ from those reported in earlier publications due to the time of measurement and differences in measurement techniques (the area estimates used here are based on measurements of Landsat images from 1990; Feeley, Gillespie & Terborgh, 2005). We did not include the mainland and one small island, Perimetro, included in the original 1993 censuses (Terborgh, Lopez & Tello, 1997b) because they suffered ground fires during the research period (Terborgh et al., 2006).

Table 1.   Characteristics of Lago Guri islands including rates of species change in bird species richness (kSR) and local extinction (kEXT)
Island nameArea
Distance to
Density of
(ind ha−1)b
Density of
Atta sp.
(colonies ha−1)b
  • a

    Area and distance estimates differ from previous publications due to refined techniques.

  • b

    b Density estimates based on previous publications have been adjusted to account for area changes.

Danto Machado211.41.6290.20.01−0.09−0.13

The habitat of the study islands is semi-deciduous tropical dry forest (Huber, 1986) with a canopy height ranging from 15 to 20 m and occasional emergents reaching over 25 m. The islands generally have low topographic profiles and clay-rich oxysol soils, with the exception of two islands, Miedo and Lomo, which were steeper and rocky. The mean annual temperature recorded at the dam is 27.6 °C and mean annual precipitation ranges from 800 to 1200 mm with the majority of rain falling during a 6 month summer wet season (May–October). During the dry season, the water table at Lago Guri can drop by >10 m, exposing large expanses of muddy lake bottom. All islands are surrounded by belts of dead trees killed by inundation.

The selection of study islands at Lago Guri took advantage of a preliminary understanding of area-dependent changes in the composition and/or abundance of different species groups (Terborgh et al., 2001, 2006; Terborgh & Feeley, in press). The largest island, Danto Machado possessed a mostly intact faunal community including resident populations of all but the topmost predators (jaguar Panthera onca; puma Felis concolor and harpy eagle Harpia harpyja), although all three of these species were recorded as transient visitors. The three medium islands (9.8–11.5 ha) supported several mid-sized mammals [howler monkey Alouatta seniculus (one island); olive capuchin Cebus olivaceus (two islands); agouti: Dasyprocta spp. and armadillo Dasypus novemcinctus] but lacked all top predators (Terborgh et al., 2001). The faunal communities of small islands (0.6–3.2 ha) were even further reduced, lacking all predators of large vertebrates and supporting greatly elevated densities of generalist herbivores: howler monkeys; common iguanas Iguana iguana and leaf-cutter ants Acromyrmex spp. and Atta spp. The latter at up to 100 times the density observed on the nearby mainland (Rao, 2000).

For our analyses, densities of leaf-cutter ant colonies were taken from Rao (2000) or Terborgh et al. (2001) and modified according to the revised area estimates. For the remaining islands, we conducted counts in 2000. Howler monkey densities were determined through visual censuses conducted annually throughout the late 1990s and early 2000s (Terborgh et al., 1997a, 2001). These estimates were reinforced by counts of howler monkeys encountered during avian censuses. While common iguanas were present on all islands, their cryptic habits precluded accurate density estimates and thus they were omitted from the analyses. Likewise, we were unable to determine the precise numbers of capuchins inhabiting the islands so we instead used artificial nest assays to estimate relative rates of nest predation. In June of 2000, we set out 12 grass-wicker nests on each of four small and three medium islands and 24 nests on Danto Machado (logistical constraints prevented running assays on the remaining three islands). Each nest was tied to a low shrub or sapling away from any trails and stocked with two fresh quail eggs (standard precautions were taken to minimize the risk of predators following human cues to nests). We checked the status of all nests after 9 days and if any egg was found damaged or missing, the nest was considered to have been predated. We then calculated relative nest predation as the ratio of predated to total nests (Hensler & Nichols, 1981).

The avian communities of all sites were monitored over a period of 10 years (1993, 2000, 2001, 2002 and 2003) using standard spot-mapping techniques. We have presented the details of our census methods elsewhere (Terborgh et al., 1997b; Feeley, 2003) and hence we provide only a brief description here.

Each year, we censused the small study islands between three and five times each, and the medium and large islands between five and nine times each. Small and medium islands were censused in their entireties while Danto Machado was sub-sampled using a 26 ha plot. Following the protocol of Terborgh et al. (1997b), we identified and mapped all vocalizations and sightings of birds. In the analyses presented here, we only included species considered as forest-interior residents. To be considered a resident, a pair/male had to be observed during at least two-third of the census visits in a given year (Supplementary Material Appendix S1). Pairs observed during fewer visits were classified as visitors and excluded from the analyses (Terborgh et al., 1997b). The stands of snags that surround the islands form a distinct matrix from the island interiors and are used by a number of bird species for nesting, foraging and roosting (e.g. some falcons, woodpeckers, flycatchers, martins, swallows). Because these ‘edge’ species are not dependent on the forest patches, they were excluded from the analyses. Likewise, we excluded all aquatic species as well as species observed to utilize large territories incorporating multiple islands (e.g. raptors, macaws, parrots, large pigeons, large woodpeckers, etc.).

In the initial study of the bird communities at Lago Guri, Terborgh et al. (1997b) classified large pigeons (Columba spp.) as forest-interior residents and the house wren Trogolydes aedon as an edge species. Here we excluded large pigeons from our analyses (because they were frequently observed flying between islands) but included the house wren after observing that they forage primarily in the forest interior (K. J Feeley, D. Lebbin & J. Hardesty, unpubl. data). We have revised the diversity values for 1993 accordingly.

Species loss from recently isolated fragments is often modeled using the exponential decay function


where S0 is the number of species inhabiting an area before fragmentation, St is the number of species present after t years and k is a decay constant describing the rate of species change (Diamond, 1972; Brooks et al., 1999; Ferraz et al., 2003). We determined the k values for the Lago Guri islands by fitting exponential curves to the number of species present (n+1) in the community at each census and calculating the slope (k) of the curve for each island from equation (1) (Supplementary Material Appendix S2).

In calculating k, we used two different methods to tabulate St. First, we used the total number of species present in year t, a measure that incorporates the effect of colonizations in sustaining diversity. Calculated as such, k represents the rate of change in species richness (kSR) and local extinctions are subsumed in the turnover of species between censuses.

The second tabulation of St counted only those species present in 1993 that remained on an island in subsequent censuses (kEXT). Therefore kEXT provides an estimate of local extinction in the absence of colonization. kEXT is a conservative estimate of local extinction since species may have gone extinct from and subsequently recolonized islands during the 7-year interval between censuses in 1993 and 2000. This ‘in-and-out effect’ (Diamond & May, 1977) masks colonization, local extinction and turnover events and therefore can result in underestimations of these rates.

We used path analysis to determine the relative strength of direct versus indirect effects of habitat loss in determining the rates of change in avian species richness (kSR) and rates of local avian extinctions (kEXT). According to the model, area (log10 transformed) has a direct effect on avian species loss (i.e. increased probability of stochastic extinction). Area also affects the abundance of generalist herbivores (howler monkeys and leaf-cutter ants) and rates of nest predation (the effects of area on the abundance of nest predators and herbivores are potentially due to the altered abundance of higher predators; Rao, 2000). The resultant changes in nest predation and herbivory may in turn affect the rates of species loss. We do not contend that the effects of herbivores on bird communities are direct (especially given that most birds inhabiting Lago Guri islands are insectivorous). Rather, as discussed below, these effects are themselves hypothesized to be indirect and driven through unquantified bottom-up effects. In path analysis the relative magnitude of all effects is calculated as partial regression coefficients (β) accounting for the contribution of all other pathways. Accordingly, the strength of any indirect path is then the product of its components and the total effect is the sum of all involved paths (Sokal & Rohlf, 1995).


The rates of change in species richness (kSR) ranged from 0.05 on Lomo to −0.21 on Colon Island (a positive kSR indicates an increase in species richness over the 10 years study period) and extinction rates (kEXT) ranged from −0.11 on Panorama and Iguana Islands to −0.24 on Colon Island (Table 1).

There was an insignificant positive direct effect of log area on kEXT (β=0.05, P=0.434) and in the case of kSR the direct effect of area remained insignificant but was negative (β=−0.37, P=0.889; Fig. 1).

Figure 1.

 Path diagrams showing the direct and indirect effects of island area on kSR (top) and kEXT (bottom) at Lago Guri. Curved lines depict paths which themselves are hypothesized to be indirect. The effect of area on the density of mesopredators (nest predation), howler monkeys (Alouatta seniculus) and leaf-cutter ants (Atta sp.) is possibly mediated through changes in abundance of higher predators, and the effects of howler monkeys and leaf-cutter ants on k are possibly mediated through changes in plant productivity and resource availability. Numbers are path coefficients and the significance of each path is indicated by the line type.

Area reduction had positive effects on the densities of both howler monkeys and leaf-cutter ants (i.e. negative regression coefficients indicating increased densities on smaller islands; Fig. 1) The density of leaf-cutter ant colonies on the smallest island was approximately two orders of magnitude greater than on Danto Machado (β=−0.68, P=0.016). Likewise the densities of howler monkeys was up to 43 times greater on small islands than on Danto Machado (β=−0.56, P=0.059). The possible causes of these hyperabundances have been explored elsewhere (Terborgh et al., 1997a, 2001; Rao, 2000; Terborgh & Feeley, in press). Leaf-cutter ants and howler monkeys had opposing effects on the bird communities. Rates of species loss and extinction tended to decrease with howler monkey density (βSR=0.92, P=0.133; and βEXT=1.28, P=0.023) but increased with leaf-cutter densities (βSR=−1.61, P=0.047; and βEXT=−1.43, P=0.028). Consequently, the indirect effects of area on kSR and kEXT as mediated through howler monkeys was positive (βSR=0.52 and βEXT=0.72) but the indirect effect of area as mediated through leaf cutters was negative (βSR=−1.09 and βEXT=−0.97). Nest predation tended to increase with log island area (β=0.53, P=0.107) and had an insignificant negative effect on kSR (β=−0.12, P=0.835), but an insignificant positive effect on kEXT (β=0.07, P=0.852).

Combined, the indirect effects of area mediated through changes in nest predation pressure and the densities of leaf-cutter ants and howler monkeys were 1.48 times stronger than the direct effects of area on kSR (β indirect=0.52) and 7.25 times stronger than the direct effects on kEXT (β indirect=0.29; Fig. 1). Accounting for both direct and indirect effects, the total effect of area was positive for kSR and kEXT such that overall, larger islands tended to have slower rates of species loss (βSR=0.173 and βEXT=0.163).


Both land-bridge islands and mainland habitat fragments start out with faunal communities characteristic of the surrounding landscape but gradually lose species subsequent to isolation (Diamond, Bishop & Balen, 1987; Sodhi et al., 2006). This process of species loss, or ‘relaxation’, is commonly modeled as a direct consequence of reduced habitat area (Brown, 1971; Diamond, 1972; Terborgh, 1974; Ferraz et al., 2003). While it is possible that area per se is the principal causal factor and that species loss is truly a passive process [e.g., driven by stochastic population fluctuations, as postulated by Karr (1982) for extinctions of birds on Barro Colorado Island, Panama], it is also possible that fragmentation also has an important indirect effects on bird communities due to changes in the presence/absence or abundances of other faunal groups and/or changes in habitat structure. Our results from a decade of monitoring bird communities on land-bridge islands in Lago Guri strongly support this hypothesis and indicate that the influence of area on species loss is primarily indirect and mediated through altered trophic interactions.

At Lago Guri, the density of generalist herbivores is greatly elevated on small islands, presumably due to the absence of predators (Rao, 2000; Terborgh et al., 2001). The consequences of increased herbivore densities are idiosyncratic, depending strongly on the ecologies of the specific herbivores involved. Reduced rates of avian extinction/species loss were associated with hyperabundant howler monkeys. With a different subset of Lago Guri islands we demonstrated that tree growth rates were six times greater on small islands supporting hyperabundant howler monkeys than on similar-sized islands lacking howlers (Feeley & Terborgh, 2005). The positive influence of howler monkeys on the bird communities may be due to increased primary productivity stimulated by accelerated nutrient cycling through urine and feces (Feeley, 2005; Feeley & Terborgh, 2005). As plant productivity increases, the amount of resources available to birds will increase either via increased seed/fruit/nectar production or increased insect abundance leading to the greater persistence of bird species (Feeley & Terborgh, 2006). Alternatively, howler monkeys may somehow be altering the physical structure of the forests in such a way as to increase avian nesting/foraging success and in turn causing decreased extinction rates.

In contrast with the positive effect of howler monkeys, increased abundances of leaf-cutter ants is associated with accelerated bird species loss/local extinction. Leaf-cutter ants deposit leaves and flowers in underground fungal gardens and refuse chambers that can be up to 5 m below the surface, thereby sequestering nutrients below the root zone, especially in systems with fluctuating water tables (Haines, 1983) such as Lago Guri. In addition to reducing nutrient availability, leaf cutters feed heavily on seedlings and saplings and thereby directly depress tree recruitment rates (Rao, Terborgh & Nunez, 2001; Terborgh et al., 2006; Lopez & Terborgh, 2007;Fig. 2). Elevated densities of leaf-cutter ants will consequently result in diminished productivity and changes to habitat structure that act to accelerate bird extinctions.

Figure 2.

 The photograph on the left shows the dense forest habitat typical of the large Lago Guri islands supporting ‘normal’ densities of generalist herbivores. This is in sharp contrast to the right-hand photograph which shows the relatively barren habitat found on islands with elevated densities of leaf-cutter ants (Atta sp.). Changes in habitat structure/quality such these may be accelerating the loss of bird species from small islands.

Contrary to our original expectations, we found only a weak effect of nest predation on relaxation rates (and in the case of kSR, greater predation was actually associated with decreased extinction rates). One possible explanation for this result is that the rate of predation on artificial nests may not accurately reflect true rates of nest predation (Robinson, Styrsky & Brawn, 2005). Alternatively, the lack of a predation effect could reflect the early loss of the most predation-sensitive bird species. When our research began in 1993, medium-sized islands that supported high densities of nest predators (e.g. capuchin monkeys) were already strikingly deficient in birds (Terborgh et al., 1997b). At that time, Lomo and Coral (with capuchins) supported just 0.8 and 1.2 pairs of birds ha−1, respectively, whereas Panorama (without capuchins) supported 4.3 pairs ha−1; small islands supported 8.9 pairs ha−1 and Danto Machado supported 8.3 pairs ha−1 (Terborgh et al., 1997b). Many vulnerable bird species may thus have already been lost by our first round of censuses when the islands were already 7 years old and hence the absence of subsequent effects on these islands. Early loss of the most predation-sensitive species would help explain why species richness remained relatively stable on medium islands over the decade of observation. In sharp contrast to the medium islands, Danto Machado suffered very high rates of species decline, losing over 29 species (57% of the initial species complement) during the 10-year period. These losses coincided with an apparent increase in the density of capuchin monkeys inhabiting the island. Troops of capuchin monkeys were rarely encountered during the initial 1993 censuses of Danto Machado but encounter rates increased sharply over the study period (unpubl. data).

As with predation, the absence of a strong direct effect of area on extinction rates may be due in part to the rapid loss of area-sensitive species before the start of censuses. Several of the islands included in this study are smaller than the typical territories of tropical forest-interior bird species (Terborgh, 1990) and thus it is probable that a high proportion of the most sensitive species were lost near-instantaneously from small islands following inundation. Consistent with this hypothesis, some of the species persisting on the islands are known to utilize secondary habitats throughout much of their ranges (e.g. T. aedon, Melanerpes rubricapillus; Meyer de Schauensee & Phelps, 1978; Feeley et al., 2007) and thus may be capable of persisting on small islands by an ability to exploit edges as well as interior habitat (Antongiovanni & Metzger, 2005; but see Harris & Pimm, 2004). Several other species persisting on the small islands are highly vagile and therefore populations on small islands may be sinks (Pulliam, 1988) maintained through high rates of colonization and the rescue effect (Brown & Kodric-Brown, 1977).

While the conditions at Lago Guri are perhaps extreme, we contend that the unleashing of scale-dependent forces that distort normal trophic interactions is a typical consequence of habitat fragmentation, even in mainland systems. For example, hyperdense populations of howler monkeys and other generalist herbivores is a common phenomenon in forest fragments throughout much of the tropics, having been reported from Panama, Brazil, French Guiana and Mexico (Milton, 1982; Lovejoy et al., 1984; Schwarzkopf & Rylands, 1989; Asquith, Wright & Clauss, 1997; Estrada et al., 1999, 2002a; Leigh et al., 2002; Pozo-Montuy & Serio-Silva, 2007). As we have shown here, distortions such as these can in turn have significant consequences for the persistence of birds and other important taxonomic groups. Altered abundances of herbivores (McShea & Rappole, 2000), nest predators and parasites (Wilcove, 1985; Robinson et al., 1995; Crooks & Soulè, 1999; Soule & Terborgh, 1999; Schmidt, 2003), as well as commensals (Bierregaard & Lovejoy, 1989; Stouffer & Bierregaard, 1995) can all have important impacts on bird communities through bottom-up and top-down effects. In this light, previous investigations of bird (and other) communities in fragmented systems that have used area as the sole or principle explanatory variable should perhaps be reevaluated. It is possible that the direct effects of area have been widely confounded with the more subtle indirect effects of distorted trophic interactions that inevitably accompany fragmentation.

Understanding the relative importance of direct and indirect effects of area contraction on bird communities will have important implications. Indirect effects of altered trophic interactions differ from direct effects of area per se in that the former are often non-continuous, depending on the presence/absence of key species such as predators and/or mesopredators and the release into hyperabundance of others, such as generalist herbivores. As such, if the impacts of habitat loss on bird persistence are primarily indirect, as our results suggest, conservation strategies will have to be modified accordingly. For example, in order to mediate the impacts of human activities on faunal communities it will be necessary not only to maximize the areas of preserved habitats, but also to minimize the associated distortions in trophic interactions. While this will pose a daunting challenge given the high sensitivity of many ecologically important species (such as large predators) and the synergy between fragmentation and other anthropogenic disturbances (Terborgh, 1974; Laurance, 2001; Peres, 2001; Wright & Duber, 2001) headway may be made through increased protection against poaching or by increasing connectivity between fragments (Dobson et al., 1999).


We thank EDELCA (Electrificación del Caroní), particularly L. Balbas, for longstanding support of the project and its many personnel. For assistance in the field, we are grateful to L. Lopez, J. Tello, G. Orihuela, D. Lebbin, J. Hardesty, D. Escalasans, J. Rowland and L. Davenport. The generous financial support of the Chapman Memorial Fund, Cooper Ornithological Society, Duke University, Georgia Audubon Society, MacArthur Foundation and National Science Foundation (DEB-9707281, DEB01-08107) is gratefully acknowledged.