Functional responses in habitat selection by tropical birds moving through fragmented forest

Summary

1. The ability of animals to move through a landscape is a fundamental determinant of population persistence in fragmented habitats. This movement can be affected by both the composition and configuration of the remaining habitat. To date, few studies have examined the habitat selection of animals moving in novel landscapes or addressed whether animals exhibit a functional response in selection as available habitat changes.

2. To assess habitat selection during movement, we translocated 60 individuals of two species of birds with differing forest dependency in three configuration treatments in a highly fragmented, tropical dry forest landscape: along a forested riparian corridor, along a fencerow (row of live trees) or across pasture. We closely followed the return routes of translocated birds to determine their choice of habitat and proximity to the forest edge. We then tested whether habitat composition or configuration (treatment) best explained individual variation in habitat selection.

3. Both species preferred habitat closer to the forest edge, but this preference was weaker in the forest specialist, the barred antshrike Thamnophilus doliatus. This species selected routes in forest habitat, which included riparian corridors, over fencerow and stepping stone habitat, which were all preferred over pasture habitat.

4. By contrast, the forest generalist, the rufous-naped wren Campylorhynchus rufinucha preferred forest equally to fencerow and stepping stone habitat over pasture habitat. For it, fencerows were selected more strongly than stepping stones.

5. Analysis of the individual variation in selection for forest habitat revealed that both species exhibited a functional response to habitat configuration, selecting forest more strongly in riparian corridor treatments where it was also more abundant. The forest specialist also reduced its preference for edge habitat in riparian corridor treatments.

6.Synthesis and applications. The unprecedented precision of our route information demonstrates the extent to which our forest specialist preferred to travel in forest relative to fencerow and stepping stone habitat. Functional responses to habitat configuration indicated that both species make more use of other habitats when forested corridors are not present. Stepping stones in particular may be important features to the conservation of forest birds in highly fragmented habitats.

Introduction

Habitat loss and fragmentation are widely acknowledged as major contributors to biodiversity loss worldwide. Many of the species affected by habitat loss dwell in forest, which is permanently cleared for urbanization, agriculture and industry. The long-term persistence of forest-dependent species in landscapes that have been anthropogenically fragmented is expected to depend partly on the success with which individuals move within and among subpopulations (Hanski 1998). The mechanistic basis of this movement is habitat selection and yet there have been very few detailed assessments of the way moving animals select habitat at the landscape scales at which habitat fragmentation occurs (Chetkiewicz, St. Clair & Boyce 2006).

Habitat selection by animals in fragmented habitats can be influenced by both the composition and configuration of the preferred habitat (e.g. Stubblefield, Vierling & Rumble 2006; Radford & Bennett 2007). In general, habitat configuration is expected to be more important to population persistence in landscapes in which small amounts of the original habitat remain (Fahrig 1998), such as those dominated by agriculture. The fundamental components of habitat configuration in these landscapes are those elements facilitating animal movement among isolated populations. Conduits for animal movement may be provided by riparian corridors (Beier & Noss 1998; Hawes et al. 2008; Lees & Peres 2008), fencerows (living fences; Rosenberg, Noon & Meslow 1997; Şekercioğlu et al. 2002) and stepping stones of individual trees (Fischer & Lindemayer 2002a; Boscolo et al. 2008; Manning, Gibbons & Lindenmayer 2009). All of these have been advocated for conservation planning (Rosenberg et al. 1997). These habitat elements typically create an abrupt edge with the surrounding matrix, therefore edge response is a fundamental part of habitat selection in highly fragmented habitats (McCollin 1998). To date, these responses have been studied in the context of distance from edges within home ranges (e.g. Desrochers & Fortin 2000; Hansbauer et al. 2008), or as directional response to patch edges (e.g. Schultz & Crone 2001) or corridor edges (e.g. Haddad 1999; Levey et al. 2005). Almost no studies have provided a detailed assessment of habitat selection and the effect of edges on animals travelling outside a home range at large spatial scales (>200 m) in fragmented landscapes (see Bakker 2006 for an exception).

Information about habitat selection for animals moving at landscape scales is particularly important in the tropical dry forests of Mesoamerica where only c. 2% of the original forest remains (Janzen 1986). Pressure on the remaining forest is likely to continue as the agricultural land area expands to meet growing food demand (Tilman et al. 2001). Tropical birds, and insectivores in particular, are considered more sensitive to forest destruction than their temperate counterparts (Harris & Reed 2002; Stratford & Robinson 2005). One contributor to this difference may be greater sensitivity to forest edges (Lindell et al. 2007), which often present greater predation risk for forest birds (McCollin 1998).

The precise routes of birds moving through highly fragmented tropical landscapes is likely to depend on two main attributes: the type of habitat elements used and the selection or avoidance of habitat edges these elements create. Identifying the responses of birds to both attributes can be achieved using resource selection functions (hereafter RSFs; Manly et al. 2002), which quantify habitat preferences based on the habitats used relative to their availability. RSFs typically examine the mean habitat selection for a sample of individuals. However, this approach can obscure individual differences in selection stemming from variation in their habitat context. It is more likely that their habitat preferences vary with habitat abundance (Mysterud & Ims 1998). Mysterud & Ims (1998) termed this difference a functional response to habitat abundance and acknowledged that selection may also vary with the spatial arrangement, or configuration, of the habitat. Recent advances in the methodology used to analyse resource selection (Gillies et al. 2006) make it possible to explore individual-specific habitat selection and examine functional responses for multiple habitat types and continuous variables.

To better understand habitat and edge selection and to test whether moving birds exhibit functional responses to habitat configuration or composition, we translocated and then followed the returns of 60 individuals of two species of insectivorous forest birds as they travelled through highly fragmented tropical dry forest in Costa Rica. Translocations were aligned in three treatments of habitat configuration corresponding to common habitat elements in this and other highly fragmented, agriculturally dominated forests: riparian corridors, fencerows and open pasture. This experimental approach enabled us to collect information about habitat and edge selection by adult birds moving in novel habitat at a landscape scale. We assumed these movements would partially reflect the behaviour of the dispersing individuals that are so critical to the persistence of subpopulations (Levey et al. 2005). Using translocations allowed us to standardize the bird’s motivation for moving, anticipate the direction it would predominantly travel, and choose the configuration of the intervening habitat (Bélisle 2005).

Materials and methods

Study area

We followed the movement of two species in an agricultural landscape of northwestern Costa Rica near the town of Liberia. This landscape was once contiguous tropical dry forest, but is now dominated by cattle pasture. The remaining forest is often confined to riparian areas. The landscape also contains planted rows of individual trees (fencerows) at the edges of fields and individual trees or small patches of trees in the pasture that we term stepping stones. These are typically remnants of the original forest cover that have been retained as shade for the cattle. Fencerows and stepping stones are rare relative to forest habitat (Table 1). Forest habitat represents ∼38% of the study area of ∼125 km².

Table 1.   Habitat amounts in each of the three treatments, including means and standard deviations, measured in the 20 ellipses for each treatment

Treatment	Forest	Fencerows	Stepping stones	Total
All values are given as percentages.
Riparian corridor	35·3 ± 13·5	2·9 ± 2·1	1·9 ± 1·0	40·2 ± 13·7
Fencerow	21·8 ± 11·6	6·3 ± 2·9	2·6 ± 1·2	30·6 ± 10·9
Pasture	16·8 ± 7·9	3·1 ± 2·8	4·0 ± 2·1	23·8 ± 7·9

Translocations and tracking

We followed moving forest birds in real time at a landscape scale, collecting information about their route and habitat use. We translocated 30 barred antshrikes (Thamnophilus doliatus, hereafter antshrikes) and 30 rufous-naped wrens (Campylorhynchus rufinucha, hereafter wrens) between 0·7 and 1·9 km. Both are common insectivores that hold territories year-round. Antshrikes are forest specialists in the tropical dry forest of Costa Rica, typically being found only in the understorey of the most intact forest in this region, whereas rufous-naped wrens are forest generalists in the dry forests of this region, being found in a range of forest types from degraded to intact (Stiles & Skutch 1989). Individuals were moved from unique forested territories to unique release locations (>100 m from the nearest release for the same species) in one of three treatments: along a riparian corridor, along a fencerow or across pasture (10 individuals per species per treatment; Fig. 1). These treatments represent the type of habitat between the capture and release locations. Translocation treatments for each distance were chosen a priori from aerial photos. We conducted translocations from June to August 2000 and January to June 2002. All individuals were caught by 09:40 local time (mean capture time = 06:59 h ± 65 min) by attracting them into a mistnet with a playback of a conspecific song. We moved male antshrikes and both male and female adults of the monomorphic wrens. We marked individuals with unique coloured leg bands and attached a radio transmitter using eyelash adhesive to trimmed feathers on the backs of translocated individuals. Birds were captured in forest and released in fencerow or forest habitat. Birds that were moved across pasture were released in patches >3 ha in area. Due to the rarity of fencerows in the study area, the same fencerow was used for two treatments (one of each species) on three occasions. Thus, 17 fencerows were used for 20 translocations. In these cases, we moved an individual of each species differing distances which resulted in 10 unique fencerows for each species. Most wrens (23 of 30) were sexed by extracting DNA from a whole tail feather (Griffiths et al. 1998). The remaining individuals were sexed by comparing their weight, tarsus length and exposed culmen length to measurements of individuals of known sex using a discriminant function analysis (sensuDesrochers 1990). We translocated 14 female and 16 male wrens.

Figure 1.  Example translocations for each of the three treatments overlaid on land cover imagery. Darker areas are trees and lighter areas are pasture. Capture locations are white circles and the release locations are white squares. The return paths of the birds are depicted with a black line. Available habitat and edge distances were sampled within the ellipses for each bird. All 60 paths are available in Gillies & St. Clair (2008).

Following release, we recorded with radio-telemetry and hand-held GPS units (Trimble GeoExplorer 3 Sunnyvale, California, USA) the location of each translocated bird approximately every 15 min (mean = 14·8 min ± 8·2 min standard deviation) for up to 4 days and daily thereafter for 10 days or until they returned, whichever was earlier. We used transmitters that were 0·6 g or 1·3 g (Holohil Systems Ltd, Carp, Ontorio, Canada; models LB-2 and BD-2G) that represented <5% of an individual’s mass (mean mass for antshrikes = 26·2 g and wrens = 34·8 g). Two observers closely followed individuals by simultaneously triangulating their location from a mean distance of 27 ± 13 m using a Telonics RA-14K antenna and Communications Specialist Inc. R-1000 receiver. These positions provided trajectories of moving birds, which we used to assess the habitat used during movement. Because we only had two observers, we were not able to estimate our telemetry error, but we expect it to be small because we were relatively close to the bird and we do not expect any bias in the error. One bird from each species was translocated in each treatment at each of 10 distances (0·7–1·3 km in 0·1 km intervals, then 1·45, 1·6 and 1·9 km). Even the shortest translocations were well outside the home range of these birds. Home range size for two antshrikes that returned were 0·58 and 0·30 ha (mean = 0·44, minimum convex polygons, n = 26 and 20, radius if circular = 37 m) and for four wrens at 0·33, 0·31, 0·36 and 1·20 ha (mean = 0·55, n = 28, 18, 16, 17, radius = 42 m). These are similar to published results for cogener of both species, where home range radius was ∼60 m for a cogener to the antshrike in Brazil (T. caerulescens; Duca, Guerra & Marini 2006) and ∼75 m for a cogener to the wren in Venezuela (C. nuchalis; Yaber & Rabenold 2002). Mean return times were 33·7 daylight hours for antshrikes and 26·2 daylight hours for the wrens (Gillies & St. Clair 2008). Riparian corridors were typically between 50 and 150 m wide. Fencerows were typically 15–30 m wide with little understorey.

Land cover information

Land cover information for the study area was developed from a series of high-resolution (∼1 m pixel size) infrared images taken by the Airborne Sensor Facility at the National Aeronautics and Space Administration as part of the CARTA program during March 2003 (NASA 2003). Images were orthorectified using a digital elevation model and the coordinates of known locations in the field with the OrthoBASE package in ERDAS IMAGINE 8 (ERDAS Inc 2002). Land cover was delineated on these images using arcgis 9 (ESRI 2005) as pasture, forest, fencerow or stepping stone habitat. Riparian corridors were classified as a component of forest habitat. The available habitat for each individual was measured inside an ellipse with foci on the release and capture points and an eccentricity of 1·4. This ellipse shape was chosen by rescaling all movement paths to a common capture and release point to approximate the region in which these birds typically moved after release. For the habitat selection analysis, used locations were intersected with the land cover and we sampled the available habitat at 1000 random locations within each individual’s ellipse using ‘Hawth’s Tools’ (Beyer 2007). Similarly, we measured the distance to the forest edge for only those observed locations that occurred in forest habitat and we sampled the available distance from the forest edge at 200 random points in forest within each individual’s ellipse.

Data analysis

Habitat selection and edge selection were both analysed using RSF (Manly et al. 2002) using mixed-effects logistic regression. The RSF for habitat selection compared the used habitat types to their availability. The RSF for forest edge selection examined whether the birds selected or avoided areas closer to the forest edge. We included random effects in all models to account for the correlation inherent in taking multiple samples from the same individual and to correct for differing samples sizes among individuals (Gillies et al. 2006). A random intercept for each individual helps to account for differences among individuals in the number of used and available points. A random coefficient allows individuals to vary from the population coefficient estimate in the strength of selection for a covariate. This approach also made it possible to assess individual-specific responses to habitat variables in addition to the estimate of population-level responses (Gillies et al. 2006).

Statistical models for the individual-specific (random) coefficients used linear regression and were built using forward step-wise entry of variables (P < 0·1 for addition). Because n = 30 for these two analyses, we felt it was inappropriate to include all the candidate covariates in a single model (Table 2). We considered variables to be statistically significant at P < 0·05, but used P < 0·1 for addition to include variables that may indicate a trend that would be significant with a larger sample size. No variables in the final model were correlated at >0·5. Variables describing the amount of forest cover and the mean distance from the edge for the available points were included to investigate how birds responded to habitat availability in the habitat and edge selection, respectively. We tested for the addition of interaction terms once there were no more univariate variables to add. For the analysis of edge selection, we included only those individuals with 10 or more locations in forest (antshrikes n = 29; wrens n = 26). We removed one outlier from the analysis of the antshrike edge selection coefficients because its value for edge selection was greater than five standard deviations from the mean for the other individuals in that treatment. Mean available distances to the forest edge ranged from 16 to 141 m (mean = 38 m) for the antshrikes and 8–76 m (mean = 33 m) for the wrens. All analyses were performed using stata 10.0 (Statacorp 2007). The mixed-effects logistic regressions used the GLLAMM package in Stata and an exchangeable within-group correlation structure (Rabe-Hesketh, Skrondal & Pickles 2004) and the analyses of the individual-specific coefficients used linear regression. Post hoc tests for group membership used the test procedure in Stata (Statacorp 2007).

Table 2.   Candidate variables for inclusion in the forward step-wise addition of variables to the models explaining individual-specific (random) coefficients for habitat and edge selection. Measures of cover were the proportion of the ellipse around the capture and release points that contained that habitat (forest, fencerow or stepping stone). Pasture treatments were the reference category to which the other two treatment variables were compared. We considered treatment to be a measure of habitat configuration and land cover as measures of habitat composition

Analysis	Candidate variable
Habitat selection	Treatment (2 levels; riparian corridor and fencerow)
	Forest cover
	Treatment × forest cover
	Fencerow cover
	Stepping stone cover
	Whether returned
	Translocation distance
	Sex (wrens only)
Edge selection	Treatment (2 levels; riparian corridor and fencerow)
	Mean available distance to edge
	Treatment × mean available distance
	Forest cover
	Fencerow cover
	Stepping stone cover
	Whether returned
	Translocation distance
	Sex (wrens only)

Results

Habitat selection

The habitat selection analyses included a single categorical variable describing the habitat elements birds could select in their return journeys: forest, fencerow, stepping stones, with pasture as the reference category. There were 2441 used locations for the antshrikes and 2295 for the wrens, with a range of 12–175 per individual for the antshrikes and 11–248 for the wrens.

The forest specialist antshrikes preferred all three habitats relative to pasture (Fig. 2; Table 3). Predictably, they exhibited significantly stronger selection for forest than for either fencerow or stepping stone habitat. The forest generalist wrens also preferred all three habitats relative to pasture (Fig. 2; Table 3), but this generalist species selected fencerow habitat more strongly than stepping stone habitat and showed an intermediate preference for forest habitat, which did not differ significantly from either fencerow or stepping stone habitat (Table 3). Based on the selection coefficients in Table 3, if all four habitats were equally available, the distribution of a hypothetical individual’s points in pasture, forest, fencerow and stepping stone habitat would be 0·2%, 53%, 25% and 22% for antshrikes and 0·2%, 31%, 44% and 26% for wrens. For both species, we included in these initial models a random intercept and a random coefficient for forest habitat which revealed substantial individual variation in the strength of selection (Fig. 2).

Figure 2.  Coefficient values for habitat selection for both species of the three habitat types. Pasture was the reference category. Error bars are standard errors on the coefficient estimates (Table 3). Open circles are the individual-specific values of selection for forest used in the examination of functional responses.

Table 3.   Habitat and edge selection logistic regression model results, including coefficients, standard errors, significance, and random effect variance. Pasture habitat provides the reference category to which the other habitat types were compared. Superscript letters denote group membership for post hoc comparisons (P ≥ 0·05). Coefficient values indicate the direction and magnitude of selection for habitat type and distance from the forest edge

Analysis	Species	Variable	Coefficient	SE	P	Random effect variance
Habitat selection	Antshrikes	Forest	6·48^A	0·52	<0·001	5·42
		Fencerow	5·23^B	0·22	<0·001
		Stepping stone	5·08^B	0·22	<0·001
		Constant	−7·99	0·47	<0·001	4·03
	Wrens	Forest	5·48^AB	0·48	<0·001	5·11
		Fencerow	6·04^A	0·21	<0·001
		Stepping stone	5·24^B	0·22	<0·001
		Constant	−7·64	0·37	<0·001	2·44
Edge selection	Antshrikes	Distance to edge	–0·0185	0·0040	<0·001	0·00825
		Constant	−0·748	0·165	<0·001	0·942
	Wrens	Distance to edge	−0·0285	0·0088	0·001	0·00154
		Constant	−1·231	0·243	<0·001	1·284

To better understand whether habitat composition (forest cover) or landscape configuration (treatment) was affecting the strength of an individual’s selection for forest, we next regressed their individual-specific coefficients against our candidate predictor variables (Table 2). For both species, treatment and whether the bird returned were significant predictors of the variation in selection for forest (Table 4; overall model for antshrikes F_3,26 = 11·55, P = 0·0001, r² = 0·57; wrens F_4,25 = 9·41, P = 0·0001, r² = 0·60). This effect of forest configuration indicates that selection for forest habitat increased in riparian corridor treatments and decreased in fencerow treatments relative to pasture treatments. Put another way, these birds exhibited greater preference for forest habitat when forested routes were available, the requirement of a functional response. When they were in a pasture or, especially, a fencerow, treatment, they made less use of forest. Non-returning birds selected forest more strongly than returning birds of both species. Finally, female wrens selected forest more strongly than males.

Table 4.   Linear regression models predicting the individual-specific coefficients for forest habitat from the habitat selection analysis. Treatment coefficient superscript letters denote group membership (P ≥ 0·05). For the categorical variable treatment, pastures provide the reference category. Coefficient values indicate the direction and magnitude of change in selection for forest habitat

Species	Measure	Variable	Coefficient	SE	P
Antshrikes	Forest	Treatment-riparian corridor	2·54^A	0·73	0·002
	Coefficients	Treatment-fencerow	−1·47^B	0·70	0·045
		Returned	−1·38	0·61	0·032
		Constant	0·47	0·54	0·393
Wrens	Forest	Treatment-riparian corridor	1·86^A	0·72	0·016
	Coefficients	Treatment−fencerow	−1·61^B	0·70	0·029
		Returned	−1·74	0·59	0·007
		Sex – males	−1·33	0·59	0·034
		Constant	1·78	0·63	0·009

Edge selection

There were a total of 2224 used points in forest for the antshrikes and 1669 for the wrens with which we examined selection for edge proximity. Sample sizes ranged from 11 to 171 per individual for the antshrikes and 11–202 for the wrens. Both species selected positions closer to the forest edge relative to the random locations, but this preference was stronger in the wrens than the antshrikes (Table 3).

For the antshrikes, the individual-specific coefficients for edge selection were related to treatment, which is our measure of habitat configuration (Table 5; overall model F_2,25 = 6·59, P = 0·005, r² = 0·35). The other candidate variables (Table 2) were not significant predictors and were not added to the model. Antshrikes selected edges more strongly in pasture and fencerow treatments than riparian corridor treatments. In riparian corridor treatments, their selection for edge was neutral, they neither selected nor avoided edge habitat when they had a forested route to their territory. For wrens, increases in forest cover were slightly associated with a lesser preference for edge habitat, but this effect was shy of conventional statistical significance (Table 5; overall model F_1,22 = 3·88, P = 0·0617, r² = 0·15). In landscapes with low forest cover, wrens selected edge habitat, but their response to edges became approximately neutral in landscapes with higher amounts of forest cover (max. = 58% forest cover). In sum, wrens selected edges in landscapes with less forest and the antshrikes were less selective of edge or neutral in the more forested corridor treatments.

Table 5.   Linear regression models predicting the individual-specific coefficients from the edge selection analysis. Treatment coefficient superscript letters denote group membership (P ≥ 0·05) and the pasture treatment is the reference category. Coefficient values indicate the direction and magnitude of change in selection for distance from the forest edge

Species	Variable	Coefficient	SE	P
Antshrikes	Treatment-riparian corridor	0·0495^A	0·0145	0·002
	Treatment-fencerow	0·0103^B	0·0149	0·494
	Constant	−0·0319	0·0105	0·006
Wrens	Forest cover	0·090	0·046	0·062
	Constant	−0·023	0·014	0·105

Discussion

Closely following the movement of translocated birds allowed us to examine the habitat preferences and edge selection of two species of tropical forest birds as they moved through novel, fragmented landscapes. Differences in their habitat selection were consistent with differences in what is known of their breeding habitat. The forest specialist selected more forested locations. The generalist selected the three habitats similarly although it preferred fencerow habitat to stepping stone habitat. While habitat preferences are typically related to food availability (Buler, Moore & Woltmann 2007), few studies have examined the habitat selection of moving birds. We assume that preference for forest by the forest specialist is related to both food availability and the need for security cover while moving. Indeed, animals may select habitat for movement in a similar way to selection for breeding or home range purposes (Chetkiewicz et al. 2006). This may explain why understorey and terrestrial insectivores are consistently some of the most sensitive species to fragmentation. Their unwillingness to enter or cross open habitat (Harris & Reed 2002; Castellón & Sieving 2006; Stouffer et al. 2006; Moore et al. 2008) or to travel in areas with little understorey (Sieving, Willson & de Santo 2000) may be due to the fact that these areas do not contain usable habitat for foraging or other activities. Preference for fencerow habitat by the generalist suggests that these birds selected these habitats to provide a direct route of mostly continuous trees through the open landscape matrix, but it may also be that their more generalist foraging needs make the habitat more suitable overall. Despite these average differences between the two species, there was large individual variation in the strength of selection for forest cover within species.

The application of random effects to our models of habitat selection made it possible to examine the sources of variation among individuals and determine whether selection changes with availability (Gillies et al. 2006), the essence of a functional response (Mysterud & Ims 1998). Although there has been much attention paid to the relative effects of habitat composition and configuration on abundance of forest birds on breeding territories (e.g. Betts et al. 2006; Radford & Bennett 2007), this is, to our knowledge, the first investigation of their relative importance to habitat selection by animals moving in a novel landscape. Our analyses of the individual-specific coefficients suggested that both species adjusted their selection of habitat elements in response to habitat configuration, but not habitat composition. By contrast, when predicting returns rather than habitat selection, the results from previous translocation studies suggested that composition was important in some cases (Bélisle, Desrochers & Fortin 2001; Gobeil & Villard 2002), but configuration affected the return of birds in other contexts (Bélisle & St. Clair 2001; Bélisle et al. 2001; Gillies & St. Clair 2008). In this study, both species used non-forest habitat (fencerows and stepping stones) more when they did not have a direct forested route home, possibly adjusting their habitat selection as a compromise to be able to return home. This trade-off is suggested by the observation that returning birds of both species had weaker selection for forest than did non-returning birds. Consistent variation among individuals in the nature of habitat selection could be indicative of temperament (sensuReale et al. 2007) and may be an important predictor of dispersal success in fragmented habitats. Male wrens also had weaker selection for forest habitat than did females. The sex difference may reflect greater fitness consequences of territory loss for males. In a cooperatively breeding congener, males had higher reproductive success if they stayed on territories to inherit breeding positions than if they dispersed from their natal territory (Yaber & Rabenold 2002). This effect was reversed in females. Thus males may have stronger motivation to return to their territories to retain their breeding position rather than move to a new territory.

There were differences between species in edge selection similar to the overall measures of habitat selection. Other studies have examined edge selection on breeding territories (e.g. Restrepo & Gómez 1998; Laurance 2004; Hansbauer et al. 2008) and winter home ranges (Desrochers & Fortin 2000), but it has been difficult to collect such information from forest birds as they move through novel landscapes, a context with much relevance to the value of corridors for conservation. In our study, both the specialist and generalist exhibited preference for areas closer to the edge, although this preference was weaker in the forest specialist. Other work in more humid areas of the Neotropics has found that insectivores avoid areas near the forest edge (Restrepo & Gómez 1998; Laurance 2004) or authors have reported responses that differ among species (Hansbauer et al. 2008). Nonetheless, there is some consistency in the extent of edge avoidance within guilds. Similar to our results, Laurance (2004) found that midstorey insectivores, which occupy a more generalized niche, were indifferent to edges, whereas solitary understorey species – specialists – actively avoided edges. However, much variation in edge preference in our study was attributable to individuals and this variation is typically obscured in assessments of the mean response of all individuals.

Examination of the correlates of individual-specific coefficients for edge selection in our study revealed how individuals adjust their edge selection to broad landscape differences and provides some insight into the way in which corridors may facilitate bird movement more generally. By moving closer to edges in landscapes without a direct forested corridor home (in pasture and fencerow treatments), the forest specialist exhibited a functional response to habitat configuration, but not to the availability of distances to the edge or measures of habitat composition. In riparian corridor treatments, their behaviour was neutral, neither selecting nor avoiding edge. By contrast, Levey et al. (2005) found that edges channelled birds along the edges of corridors, a result consistent with the findings of Desrochers & Fortin (2000) who suggested that observed behaviour at edges would channel bird movement in corridors. Similarly, Haddad (1999) found evidence for reflection off boundaries facilitating corridor function in butterflies.

In our study, it appears that specialist birds moved closer to the forest edge, and hence potentially were directed by it, only when they did not have the option of travelling directly home via riparian corridors. Within corridors, they travelled at a greater distance from the edge (mean = 34 m, range 17–59 m, n = 10), where it was less likely that they were channelled by edge habitat, but successfully used other mechanisms to home. For the generalist, the slight trend for edge selection to decline with increasing forest cover (the only measure of habitat composition that neared significance) suggests they avoided edges when they had the option of doing so. Simulations by Zollner & Lima (2005) suggest there is a trade-off between greater predation risk at edges and increased perceptual range at edges, which aids navigation. If predation risk is higher at forest edges (McCollin 1998), the presence of riparian corridors in this landscape may have allowed birds to travel with less risk relative to fencerow or pasture configurations.

Fencerow and stepping stones were not the preferred habitat for the forest specialist, but they were on par with forest habitat for the forest generalist and they were important for the specialist outside the riparian corridor treatments. These findings may have important implications for the retention of these features. Fencerows and stepping stones are relatively rare habitats compared to forest and are easily influenced by human activity. Fencerows are planted by farmers at the edges of their fields whereas stepping stones are created primarily when farmers leave large trees or patches of trees from the original forest as shade for cattle. Interviews with farmers in our study area indicated that some of them planned to intensify their operations, using methods that would require fields with fewer or no obstacles to accommodate the use of machinery (C. Gillies, unpublished data). This could result in the clearing of stepping stones especially, which has been observed in other areas following agricultural intensification for mechanized irrigation (Maron & Fitzsimmons 2007). Beyond their value to travelling birds (this study; Boscolo et al. 2008), these remnant trees can also be very important for birds living in the agricultural landscape (Fischer & Lindemayer 2002b; Şekercioğlu et al. 2007) and may assist species adapting to climate change (Manning et al. 2009). Fischer, Lindenmayer & Manning (2006) promoted the retention of stepping stones as a general principle of conservation in agricultural landscapes. Because many of the existing stepping stone are in cattle pasture where natural regeneration is unlikely (Manning, Fischer & Lindenmayer 2006), active planting of stepping stones is likely to be needed to retain them in the landscape. The loss of stepping stones could have substantial impacts on the permeability to forest birds of this landscape and they are likely to be similarly important in other landscapes, particularly for forest generalists.

Although our information was collected from birds moving under an artificial stimulus (translocation and homing), the differences we reported among and within species are probably suggestive of the ways habitat fragmentation influences dispersing tropical birds. However, our results are likely to be conservative. A dispersing bird would have less motivation to reach a final destination than would birds returning to a territory and mate. The forest specialist in particular is likely to take fewer risks while dispersing; selecting forest habitat more strongly and avoiding edge habitat. Forest specialists in more humid forests, which have denser canopy cover, may exhibit even stronger selection for forested habitat.

In conclusion, this study provides some of the first detailed information on habitat and edge selection and associated functional responses for animals moving at a landscape scale. We have demonstrated that forested habitat is likely to be critical to the movement of a forest specialist bird, with conservative potential to generalize to other forest-dependent species. The presence of forested corridors increased selection for forest habitat. The shifts in habitat and edge selection (i.e. the functional responses) suggest that habitat configuration is important to the movement of birds in highly fragmented habitats. For the forest generalist, both fencerows and stepping stones were used for movement. The specialist travelling in landscapes with little forest also used stepping stones, suggesting that their retention will be important to the conservation of forest-dependent species in highly fragmented landscapes (Fischer et al. 2006; Manning et al. 2006). Finally, our results suggest that conservation planning will benefit from exploring individual variation in behaviour rather than assuming all individuals behave similarly (sensuReale et al. 2007).

Acknowledgements

We wish to thank E. Carman, M. Gamboa-Poveda and S. Perez-Brenes for superb help in the field and J. Zook, R. Blanco, the Guanacaste Conservation Area, and G. A. Sanchez-Azofeifa for support throughout the project. Comments from E. Bayne, J. Barlow, E. Crone, J. Groom, S. Hannon, M. Lewis and three anonymous reviewers improved the manuscript. We also extend special thanks to the landowners who gave us permission to conduct this work on their land. This project was supported by grants from the Animal Behavior Society, American Ornithologists’ Union, American Wildlife Research Foundation, Association of Field Ornithologists, Canada Foundation for Innovation, Fund in Support of International Development Activities at the University of Alberta, the International Development Research Centre, the Natural Science and Engineering Research Council (NSERC) and National Geographic Society Committee for Research and Exploration. CSG was supported by scholarships from the NSERC and the Province of Alberta.