Landscape-level impact of tropical forest loss and fragmentation on bird occurrence in eastern Guatemala

Abstract

Tropical forest destruction and fragmentation of habitat patches may reduce population persistence at the landscape level. Given the complex nature of simultaneously evaluating the effects of these factors on biotic populations, statistical presence/absence modelling has become an important tool in conservation biology. This study uses logistic regression to evaluate the independent effects of tropical forest cover and fragmentation on bird occurrence in eastern Guatemala. Logistic regression models were constructed for 10 species with varying response to habitat alteration. Predictive variables quantified forest cover, fragmentation and their interaction at three different radii (200, 500 and 1000 m scales) of 112 points where presence of target species was determined. Most species elicited a response to the 1000 m scale, which was greater than most species’ reported territory size. Thus, their presence at the landscape scale is probably regulated by extra-territorial phenomena, such as dispersal. Although proportion of forest cover was the most important predictor of species’ presence, there was strong evidence of area-independent and -dependent fragmentation effects on species presence, results that contrast with other studies from northernmost latitudes. Species’ habitat breadth was positively correlated with AIC model values, indicating a better fit for species more restricted to tropical forest. Species with a narrower habitat breadth also elicited stronger negative responses to forest loss. Habitat breadth is thus a simple measure that can be directly related to species’ vulnerability to landscape modification. Model predictive accuracy was acceptable for 4 of 10 species, which were in turn those with narrower habitat breadths.

Introduction

Habitat destruction and fragmentation caused by wood extraction, forest plantations, crops and cattle ranching are the greatest threats to the maintenance of biological diversity in the Neotropics (Wilcove, 1985, Jullien and Thiollay, 1996, Turner, 1996), where the loss of forests is occurring at an accelerated rate (Bolin, 1977, Sader and Joyce, 1988, Kaimowitz, 1996, FAO, 2005). Habitat modification at the landscape scale mainly consists of two aspects: the loss of habitat per se and habitat fragmentation (Fahrig, 2003). The effects of habitat loss are obvious: when habitat is removed from the landscape, animals subsequently displaced can also be eliminated, producing a population decline (Bender et al., 1998). Fragmentation effects are less obvious, but could nonetheless be very important. Habitat fragmentation determines the subdivision and reduction in habitat area, a potential increase of edge effects and changes in the surrounding habitat matrix, effects by which landscape continuity can be reduced and thus threaten the survival of sensitive species (With, 1997).

At present, there is no general consensus concerning the relative importance of cover or configuration in determing species presence and survival, in particular regarding bird species persistence in modified landscapes (McGarigal and McComb, 1995, Villard et al., 1999, Fahrig, 2003, Westphal et al., 2003). It is important to distinguish between pure habitat loss and true fragmentation effects when assessing bird population responses, because resultant land management and conservation recommendations may differ (Schmiegelow and Mönkkönen, 2002).

Fragmentation effects on bird communities have been well documented in the boreal temperate zone (Robbins et al., 1989, Boulinier et al., 1998, Burke and Nol, 2000). For forest-interior species, fragmentation reduces their abundance and richness (Askins et al., 1990). In the Neotropics, studies of the effect of habitat fragmentation undertaken by the Biological Dynamics of Forest Fragments Project (BDFFP) have demonstrated that terrestrial insectivorous birds are particularly sensitive to disturbance, evidenced by the reduction in their abundance as patch size decreases and their local extinction in the smallest fragments (Stouffer and Bierregard, 1995, Gascon et al., 1999). Besides these few very important studies, other research on the effect of forest fragmentation on Neotropical birds, particularly at the landscape scale, is virtually non-existent, and the aforementioned studies are patch-oriented, as opposed to being landscape-oriented. More importantly, results from landscape-level effects of fragmentation demonstrated for temperate zone birds should not be extended directly to tropical birds, due to the fact that these possess different evolutionary and life histories, which make them vulnerable to a different suite of threats than those normally considered for birds from temperate regions (Stratford and Robinson, 2005). Species in temperate regions are under strong selection pressure from abiotic factors (e.g. climate) whereas in tropical regions biotic selection pressures are most important. Interactions with other species (plant and animal) have played a key role in shaping the behavioral adaptations of tropical birds (Stutchbury and Morton, 2001). Tropical birds have high nest predation, high adult survival and small clutch sizes (Skutch, 1954, Skutch, 1960, Skutch, 1969, Ricklefs, 1969, Stutchbury and Morton, 2001), as opposed to temperate birds. Food is abundant throughout the year in the tropics (Janzen, 1973, Janzen, 1975), and this year-round availability means that most species are residents (non-migratory) and many that eat insects defend territories all year (Stutchbury and Morton, 2001). Also, unlike birds of temperate zones, tropical birds breed at all times of the year, breeding seasons are more protracted (Ricklefs, 1969), and are timed to coincide with fruit or insect abundance or reduced predation pressure, not climate per se (Stutchbury and Morton, 2001).

Typical tropical species also have a large number of complex and unique adaptations (Stutchbury and Morton, 2001). Many tropical birds tend to be sedentary or specialized in their habitat requirements, and such species are expected to have lower persistence in fragments, and are also the ones that are more extinction prone (Mac Nally et al., 2000, Sekercioglu, 2007, Sekercioglu and Sodhi, 2007). Area needs may be greater in the tropics, due to larger territory size (Terborgh et al., 1990, Robinson et al., 2000), and greater areas are often related to particular behavioral adaptations, like obligate army-ant following, and interspecific flocking (Terborgh et al., 1990). Species with these particular adaptations are often the first to disappear from fragments (Stouffer and Bierregard, 1995, Van Houtan et al., 2006, Ferraz et al., 2007). With respect to forest isolation through forest fragmentation, poor colonizers that rarely cross open areas, have relatively low patch occupancy in Amazonia, and thus are more vulnerable to fragmentation effects (Ferraz et al., 2007).

Given the complex nature of simultaneously evaluating the effects of landscape-level cover and configuration on population persistence, statistical presence/absence modelling (as a special case of generalized linear models, or GLM's) has become an important tool in conservation biology, improving the understanding of factors that determine the distribution of habitats, the species that occupy them and the human factors that lead to changes in the habitat (Guisan and Zimmermann, 2000, Guisan et al., 2002). Models represent ecological processes in simplified (Fig. 1, Fig. 2), general ways that provide information on those factors responsible for the observed patterns (Johnson and Omland, 2004, Austin, 2007).

Habitat fragmentation indices are usually strongly correlated with habitat amount, thus making it difficult to evaluate area-independent fragmentation effects on species presence (Fahrig, 2003). This means that in real-world situations, habitat fragmentation usually increases with decreasing habitat cover (see Figure 2.5, page 19, in Lindenmayer and Fischer, 2006). Thus, most empirical studies have attempted to distinguish between these effects using statistical regression models (McGarigal and McComb, 1995, Trzcinski et al., 1999, Villard et al., 1999, Westphal et al., 2003). Our main objective was to evaluate the relative and independent contributions of landscape-scale cover and fragmentation of tropical forest in determining the presence of forest-dependent birds in the Caribbean region of Guatemala. To examine the potential relation between species’ occurrence and landscape indices, we constructed occupancy-based habitat models using logistic regression, with forest area, fragmentation and their interaction as predictive variables. These relationships were examined at three different scales (200-, 500- and 1000-m scales).

Following Betts et al. (2006b), models with area and fragmentation variables at different scales were related to a series of hypotheses concerning population responses to landscape modification. At the smallest scale, we sought a response related to the “random sample hypothesis” (Haila, 1983), which states that only factors (specifically the amount of suitable habitat) at the scale of an individual territory are important in determining habitat quality. This hypothesis also implies that species have high dispersal capabilities, and are able to find patches with a size that corresponds to their territory size. Alternatively, a response related to scales greater than territory size would correspond to the so-called “landscape composition hypothesis”, which states that species respond to the amount of suitable habitat, but at larger spatial scales than the individual territory (Fahrig, 2003, Betts et al., 2006b). We also sought evidence for the “landscape fragmentation hypothesis”, which expects populations to decline with increasing fragmentation, independent of the effects of habitat loss (Villard et al., 1999, Betts et al., 2006b). Finally, an area-dependent effect of fragmentation is described by the “non-linear landscape fragmentation hypothesis”, which states that landscape fragmentation is important only below some critical amount of habitat (Andrén, 1994b, Fahrig, 1998, Betts et al., 2006b). This would result in multiplicative (non-linear) effects of fragmentation on habitat loss, i.e., a statistical interaction between landscape configuration and composition (Trzcinski et al., 1999, Betts et al., 2006b).

In theoretical studies, the relative importance of fragmentation appears to depend on the life-history traits of hypothetical species (Fahrig, 1998, With and King, 1999). Mixed results of fragmentation studies have been attributed to differences in such traits (Ewers and Didham, 2005). We investigated the relationship of species responses to fragmentation and one particular life-history trait, habitat breadth, because of its fundamental association to species’ dispersal ability. Assuming that dispersal is a key aspect of species’ persistence in fragmented landscapes, the important question is whether a species is able to disperse through (and survive in) the landscape matrix (Opdam, 1991, Fahrig, 2001, Ricketts, 2001). Terrestrial habitats are often surrounded by a complex mosaic of other cover types, which may differ in their resistance to the movement of individuals among habitat patches (Gustafson and Gardner, 1996, Ricketts, 2001, Laurance et al., 2002). A species’ ability to use matrix habitats may affect their vulnerability in fragmented landscapes (Gascon et al., 1999). A direct prediction of this hypothesis is that a species with a wider habitat breadth will find less resistance to move through the matrix and therefore will be less susceptible to fragmentation effects. In terms of resources, species that are able to exploit secondary habitats might have higher survival probability (at the individual and population level), supplementing their resource intake by using a substitutable resource in nearby patches of secondary habitats (landscape supplementation, Dunning et al., 1992).

In our modelling approach, we used a combination of traditional null hypothesis tests (NHT) and information-theoretic model-selection approaches, specifically the Akaike Information Criterion (AIC), for model-selection (Luckacs et al., 2007, Stephens et al., 2005). The Kullback–Leibler information-theoretic (KLIT) framework cannot assess model goodness-of-fit, which in contrast is provided for by the NHT framework. Alternatively, KLIT methods allowed us to compare models of different scales, comparisons not possible with inferential approaches: with NHT, only comparisons of reduced models which are nested within full models are possible (Agresti, 1996). From a practical conservation perspective, we evaluated the model's ability in predicting species presence and absence, and discuss their applicability to the study region.