Landscape context and selection for forest edge by breeding Brown-headed Cowbirds

Landscape Ecol (2007) 22:273 284 DOI 10.1007/s10980-006-9022-1 RESEARCH ARTICLE Landscape context and selection for forest edge by breeding Brown-headed Cowbirds Christine A. Howell Æ William D. Dijak Æ Frank R. Thompson III Received: 29 August 2005 / Accepted: 2 June 2006 / Published online: 19 October 2006 Ó Springer Science + Business Media B.V. 2006 C. A. Howell Department of Biology, University of Missouri-Saint Louis, 8001 Natural Bridge Rd., Saint Louis, Missouri 63121, USA Present address C. A. Howell (&) PRBO Conservation Science, 3820 Cypress Dr., Petaluma, CA 94954, USA e-mail: chowell@prbo.org W. D. Dijak Æ F. R. Thompson III USDA Forest Service North Central Forest Experiment Station, University of Missouri, 202 Natural Resources Bldg., Columbia, Missouri 65211, USA Abstract We evaluated support for four alternate hypotheses explaining the distribution of breeding Brown-headed Cowbirds (Molothrus ater) in forests at varying distances from the forest edge in three Midwestern USA landscapes with varying amounts of forest fragmentation (core forest area ranged from 5 to 70%). We focused on breeding cowbirds use of forest because of the risk of nest parasitism to forest-dwelling hosts and to identify factors affecting breeding cowbird habitat selection. We compared distances of cowbird locations in the forest from the forest edge ( edge distances ) to distances of random forest locations in the entire landscape or within individual cowbird home ranges. We analyzed 1322 locations of 84 cowbirds across three landscapes. We found support for the landscape context hypothesis that breeding cowbird preference for forest edge varied with landscape context. Ninety percent of cowbird locations were within 150 350 m of forest edge, despite the overall availability of forest at greater distances from edge (as far as 500 1450 m) both within cowbird home ranges and the entire forested landscape. Cowbird preference for edge varied by landscape context largely due to differences in the availability of forest edge. In a highly fragmented forest cowbirds utilized the entire forest and likely viewed it as all edge. In less fragmented forests, cowbirds preferred edge. We consider how variation in cowbird edge preference might relate to patterns in host abundance, host diversity, and host quality because cowbird movements indicate they are capable of using forest farther from edges. Keywords Brown-headed Cowbird Æ Edge effects Æ Forest fragmentation Æ Habitat selection Æ Home range Æ Illinois Æ Landscape context Æ Missouri Æ Molothrus ater Æ Neotropical migrant bird Introduction Fragmented forests have greater habitat discontinuity, greater edge density, reduced forest

274 Landscape Ecol (2007) 22:273 284 cover, reduced core area, and increased isolation of patches relative to more contiguous forests (Forman and Godron 1986; Howell et al. 2000). Fragmentation may have a negative effect on the fitness of organisms that are area sensitive or require forested habitats to breed (Robinson et al. 1995). Forests fragmented by agriculture in the Midwestern and Eastern United States tend to have more predators and Brown-headed Cowbirds (Molothrus ater) and hence, greater levels of songbird nest predation and brood parasitism than non-fragmented forests (Robinson et al. 1995; Hochachka et al. 1999; Thompson et al. 2000). Forests also provide important breeding habitat for Brown-headed Cowbirds (hereinafter cowbirds). Cowbirds are obligate brood parasites and thus able to spatially and temporally separate morning breeding activities (nest searching and laying) from afternoon feeding activities (Rothstein et al. 1984; Thompson 1994). Cowbirds prefer to forage in grasslands or other agricultural habitats (Thompson 1994; Morris and Thompson 1998; Gates and Evans 1998). In much of North America, however, female cowbirds prefer to parasitize hosts in forest or shrubland habitats (Thompson 1994; Hahn and Hatfield 1995; Gates and Evans 1998). Segregation of cowbird activities between forest and grassland habitats may be a relatively new behavior resulting from their range expansion east and west from the Great Plains, where there are few forests and there does not appear to be any segregation of breeding and feeding activities (Elliott 1980). In these new landscapes cowbirds likely penetrate the forest to take advantage of abundant and naïve hosts (Hahn and Hatfield 1995; Hosoi and Rothstein 2000). Searching for potential host nests, feeding, and avoiding predation are presumably essential fitness trade-offs for cowbirds. Their ability to locate multiple host nests, at the appropriate stages, has a direct effect on the cowbird s lifetime reproductive success (Ortega 1998; Hauber 2001). Longer commute distances may also diminish cowbird laying rates and reproductive success (Curson and Mathews 2003). Cowbirds should therefore select landscapes in which they minimize commuting distance and costs between breeding and feeding areas. Cowbirds commute, on average, between 1 and 4 km and a maximum of about 7 km between breeding and feeding areas (Rothstein et al. 1984; Thompson 1994; Gates and Evans 1998; but see Curson et al. 2000 for an exception). In forested habitats, cowbird abundance (Coker and Capen 1995; Donovan et al. 1997; Evans and Gates 1997) and host parasitism rates (Gates and Gysel 1978; Chasko and Gates 1982; Brittingham and Temple 1983; Morse and Robinson 1999) are often greater near forest edges that adjoin non-forest habitats. These edge effects could be the result of cowbirds limiting the distance they move away from their agricultural feeding habitats, or cowbirds responding to patterns in host density or habitat structure. Two of us (Thompson 1994; Thompson and Dijak 2000) conducted a radio-telemetry study of breeding female cowbirds to quantify spatial and temporal habitat use by cowbirds engaged in feeding, roosting, and breeding behaviors in three Midwestern landscapes with varying levels of forest cover (50 93%) and fragmentation. Cowbird movements between morning breeding and afternoon feeding locations were similar for all three landscapes, and there was a disproportionate use of forests by breeding cowbirds in all three landscapes (Thompson 1994; Thompson and Dijak 2000). Cowbird abundance, percent of nests parasitized, and cowbird eggs/nest were much greater, however, in the two fragmented sites (Jonesboro and Ashland) than the relatively nonfragmented Carr Creek site (Fig. 1; Thompson and Dijak 2000). In this paper we determined breeding female cowbirds selection of forest relative to its distance from edge using cowbird locations from Thompson s study (1994) and newly developed landscape data. We focus on breeding cowbirds use of forest to address how two relatively consistent cowbird behaviors (commuting distance and breeding preference for forest) were manifested in terms of cowbird locations within the forest relative to the forest edge. This new analysis will provide insight into the risk of parasitism for forest-dwelling hosts at a range of fragmentation levels, as well as the factors limiting cowbird numbers and affecting breeding habitat selection.

Landscape Ecol (2007) 22:273 284 275 Commuting distance (m) Female cowbirds/ point count Cowbird eggs/ parasitized nest % parasitism 1500 1000 500 0 0.6 0.4 0.2 0 2.5 2 1.5 1 0.5 0 80 60 40 20 0 Carr Creek Jonesboro Ashland Fig. 1 Female cowbird mean commuting distances between breeding and feeding areas, detections per point count, number of cowbird eggs per parasitized nest, and host nest parasitism rates in three Midwestern landscapes. Sites are listed in order of increasing fragmentation from left to right. Data from Thompson and Dijak (2000) and Thompson et al. (2000) We used an information-theoretic approach to evaluate support for four alternate hypotheses. (1) The edge preference hypothesis cowbirds prefer forest edge in all landscapes, possibly because of habitat structure, host communities, or limited commuting distances (as suggested by Brittingham and Temple 1983). (2) The forest interior hypothesis cowbirds prefer forest interior (and avoid edge) because of the availability of naïve hosts in forest interiors (sensu Gustafson et al. 2002). (3) The landscape context hypothesis cowbird preferences for forest edges vary among different landscapes (Donovan et al. 1997). In fragmented landscapes where cowbirds are abundant and potentially limited by hosts, and forest distant from edges is rare, cowbirds will saturate the forest and there will be no indication of edge preference. However, in predominately forested landscapes where cowbird numbers are low and limited by feeding habitat, cowbirds will prefer forest edges. (4) The Null hypothesis cowbirds show no preference for forest in relation to its distance from edge in any landscape. Methods Study sites and landscape classification This study was conducted in three landscapes in Illinois and Missouri, USA. The Jonesboro landscape was in Union County, Illinois; the Carr Creek landscape was in Shannon, Reynolds, and Carter Counties, Missouri; and the Ashland landscape was in southern Boone County, Missouri. The landscapes varied in composition and structure but we ranked them in order of decreasing forest fragmentation (based on edge density, percent forest cover, and forest-core area) as (1) Ashland, (2) Jonesboro, and (3) Carr Creek (Table 1). We used the program FRAG- STATS (McGarigal et al. 2002) to calculate the following metrics for each landscape: total area, percent forest cover, number of forest patches within the study area, meters of forest edge per hectare of forest (edge density), number of core areas, and percent core area (where core area includes forested areas that were > 100 m from forest edge). We include core area metrics because of their importance to breeding forest songbird habitat selection, especially for areasensitive birds. Carr Creek had the lowest level of forest fragmentation as indicated by greater forest cover and core area and lesser edge density. Jonesboro had an intermediate level of forest fragmentation

276 Landscape Ecol (2007) 22:273 284 Table 1 Patterns of forest cover and landscape composition in three Midwestern landscapes in which the distribution of cowbird locations in the forest was related to distance to forest edge, 1991 1992 Variable Ashland, MO Jonesboro, IL Carr Creek, MO Forest cover patterns Total forested area (ha) 2808 12067 5247 Number of forest patches 23 54 55 Edge density (m/ha) 58 39 32 Core % of landscape 5 38 71 Number of core areas 31 55 27 Total core area index (%) 42 65 75 Landscape composition (%) Forest 50.0 55.0 93.1 Shrubland 2.1 2.0 0.5 Grassland 31.6 3.0 3.5 Rowcrop 13.3 35.6 0.2 Feedlot 0 0.01 0.01 Developed 1.8 3.0 1.1 relative to Ashland and Carr Creek; the percent forest cover at Jonesboro was equal to Ashland, but Ashland had greater edge density and much less core area indicating greater forest fragmentation. The non-forest portion of the landscape was dominated by grassland at Ashland and Carr Creek and rowcrop at Jonesboro. In all landscapes, however, forest was the most used habitat by cowbirds in the morning when engaged in breeding activities. Grassland, rowcrop, and feedlots were the most used habitat by feeding cowbirds in all landscapes; cowbirds commuted daily between the forested and non-forested portions of the landscape (Thompson 1994; Thompson and Dijak 2000). Additional information on cowbird movements, abundance, and parasitism levels in these landscapes is available from earlier studies (Fig. 1; Thompson 1994; Thompson and Dijak 2000; Thompson et al. 2000). We classified habitat types using aerial photographs (1:12,000) and our classifications were ground-truthed during data collection. Forested habitat included mature and young forest. Nonforest habitats included rowcrop, feedlot (animal pens in which livestock fed), grassland (primarily cool season pasture), shrubland, developed habitats, and roads within forests; thus edges were primarily anthropogenic. The boundaries of the three landscapes were defined by a minimum convex polygon that encompassed all locations of feeding, roosting, and breeding cowbirds at a site (Thompson and Dijak 2000). The size of the landscapes varied with Jonesboro having the largest overall area. This is due in part to Jonesboro birds roosting at sites in the Mississippi River floodplain more distant from daytime areas relative to other study sites (see Thompson 1994). The primary effect of having the distant roost sites at Jonesboro included is that the landcover composition of Jonesboro has more cropland because the roosts were in the Mississippi River floodplain; the forests were located in the adjacent uplands. We considered excluding roosting areas when defining study site boundaries, but we felt it was necessary to include all areas that cowbirds used or traveled across during the breeding season in our definition of available habitat (Jones 2001). While commuting from roost sites, cowbirds had opportunities to assess the habitat they traveled across for breeding and feeding opportunities. Cowbird movements between morning breeding and afternoon feeding locations were similar for all three landscapes (Thompson 1994). Cowbird and random locations Cowbird telemetry data were collected from 1991 to 1992 at the Carr Creek and Ashland sites, and in 1992 at the Jonesboro site. Individual adult female cowbirds were located one to three times over a 24 h period with relocation times stratified to equally sample cowbirds during morning, afternoon, and evening activities from 15 May to

Landscape Ecol (2007) 22:273 284 277 30 June each year. Cowbirds were visually observed at each location and the time, habitat, and behavior were recorded. Further methodological details are provided in Thompson (1994). We selected a subset of cowbird locations from individuals for which there were at least 8 locations collected in the forest between 500 and 1200 because our hypotheses addressed cowbird use of forest and this time period is when cowbirds were most likely laying or nest searching (Thompson 1994). At Carr Creek we had 424 observations of 33 birds. At Jonesboro we had 401 observations of 25 birds. At Ashland we had 497 observations of 26 birds. For each observation we calculated the distance to the closest forest edge in a GIS. We used random data points to evaluate habitat availability at the scale of the landscape and individual home range. At the landscape scale we generated random locations in the forest across the entire landscape and calculated distance to the closest forest edge. For Ashland, Jonesboro, and Carr Creek there were 645, 727, and 1256 random points respectively. At the home range scale we generated random locations in forest within the home range of each cowbird, determined by the minimum convex polygon method (Thompson and Dijak 2000), and calculated distance to the closest forest edge. For Ashland, Jonesboro, and Carr Creek there were 690, 874, and 986 random points respectively at the home range scale. Analyses We report cumulative frequency distributions of the distance to forest edge in 50 m intervals for cowbird locations, random locations within cowbird home ranges, and random locations across the entire landscape. We used an information-theoretic approach to determine support for our hypotheses. We developed five a priori general linear models, evaluated support for each model, and interpreted parameter estimates from the best supported model to determine support for our four hypotheses. The models were: 1. DIST = INTERCEPT, 2. DIST = INTERCEPT + TYPE, 3. DIST = INTERCEPT + LANDSCAPE, 4. DIST = INTERCEPT + TYPE + LANDSCAPE, 5. DIST = INTERCEPT + TYPE + LANDSCAPE + (TYPE LANDSCAPE), where DIST is distance to the nearest forest edge. TYPE was a categorical variable with three levels indicating if a location was a random forest location within a cowbird s home range, a random forest location selected from the entire landscape, or a cowbird location. LANDSCAPE was a categorical variable that indicated if a location was from Ashland, Carr Creek, or Jonesboro. Model coefficients are estimated for each level of a categorical variable minus one level that is treated as the reference category. We treated cowbird location as the reference category for TYPE and Carr Creek as the reference category for LANDSCAPE. Reference categories for models with interaction terms consisting of categorical variables are more complex and indicated in Table 2 for model 5. All models included a covariance parameter to acknowledge potential covariance among locations from the same cowbird; we assumed a compound symmetry covariance structure (repeated statement, SAS Institute 2000). Model 1 is a null model and we interpreted support for it as evidence for a lack of landscape and edge effects. We interpreted support for model 2 and a positive coefficient for TYPE = random within home range or TYPE = random within landscape (indicating random locations were, on average, further from the edge than cowbird locations) as support for the edge preference hypothesis. Correspondingly we interpreted support for model 2 and a negative coefficient for these two values of TYPE as support for the forest interior (edge avoidance) hypothesis. We interpreted support for model 5 as evidence of the landscape context hypothesis, that is edge preference or avoidance varied with landscape, because of the inclusion of the interaction term. Support for model 4 would be indicative of landscape and edge effects, but that edge effects were consistent across landscapes. Model 3 is of limited ecological interest because it

278 Landscape Ecol (2007) 22:273 284 Table 2 Support for models predicting distance to forest edge of cowbird locations versus random points in three Midwestern landscapes Model 2Log e likelihood # Model parameters AIC DAIC w i LANDSCAPE a TYPE b 87537.5 11 87559.5 0.0 1.000 LANDSCAPE + TYPE 87685.7 7 87699.7 140.2 0.000 TYPE 87763.3 5 87773.3 213.8 0.000 LANDSCAPE 88073.6 5 88083.6 524.1 0.000 NULL c 88171.9 3 88177.9 618.4 0.000 Models are ranked in ascending order by Akaike s Information Criterion (AIC); models with lower AIC and DAIC, and a greater Akaike weight (w i ) have more substantial support a Category levels were Ashland, MO; Jonesboro, IL; Carr Creek, MO b Category levels were random from within home range, random from landscape, and cowbird locations; TYPE = cowbird locations was used as the reference category in the model c Null model only included an intercept pools all types of locations to examine the effect of landscape, but we included it so all possible combinations of the factors were represented by models. We fit models using distance to edge for random points generated within each cowbird home range as well as with random points generated throughout the landscape to see if this difference in scale affected our conclusions. We used PROC MIXED (SAS Institute 2000) to estimate each model. In all models we defined individual cowbirds as the subjects nested within landscapes, each cowbird location or random location as repeated measures on the subject, and assumed a constant covariance structure among repeated measures on a subject (Litell et al. 1996). We report parameter estimates and least squares means for the factor levels in the bestsupported model. We calculated Akaike s information criteria (AIC), DAIC, and Akaike weights (w i ) to identify the best models (Burnham and Anderson 2002). The model with the smallest AIC is the best approximating model for the data; Akaike weights represent the likelihood of a given model and evidence ratios can be constructed as the ratio of weights for the two models being compared (Burnham and Anderson 2002). Burnham and Anderson (2002) recommend assessing the goodness-of-fit of the global model in the set of candidate models; we performed a likelihood ratio test comparing the fullest model (LANDSCAPE TYPE) to the null model. We report parameter estimates and their SE for the best-supported model; these estimates were conditional on that model being the best model (Burnham and Anderson 2002). We determined the number of individual cowbirds in each landscape that preferred forest edge, forest interior, or had no preference. We estimated least-square means and confidence intervals with the above model for distance to forest edge of cowbird locations and random locations within the home range of each cowbird. If the mean distance to edge of a cowbird s location was less than the mean distance to edge of random points and 95% CI did not overlap we concluded that individual preferred forest edge, if a cowbird s locations were farther from edge than random points we concluded they preferred interior, if confidence intervals overlapped we concluded no preference. Results Cowbird locations in forest The cumulative distribution of distance to forest edge varied among cowbird locations, random locations in the landscape, and random locations in cowbird home ranges (Fig. 2). At Carr Creek (the site with the least fragmentation), 90% of the cowbird observations were within 350 m of the forest edge, with some cowbirds ranging 800 m into the forest (where the maximum distance to forest edge in the landscape was 1450 m

Landscape Ecol (2007) 22:273 284 279 Cowbird edge preference Fig. 2 Cumulative frequency distribution of distances from forest edge (in 50 m categories) for actual cowbird locations, random points within cowbird home ranges, and random points in the forest for Carr Creek (part A), Jonesboro (part B), and Ashland (part C) [Fig. 2a]). At Jonesboro (the site with intermediate fragmentation), 90% of the cowbird observations were within 200 m of the forest edge, with some cowbirds ranging up to 600 m into the forest (where the forest maximum distance to forest edge in the landscape was 1250 m [Fig. 2b]). At Ashland (the site with greatest fragmentation), all cowbird observations were within 350 m of the forest edge (Fig. 2c). The frequency distributions for distance to edge of random points within home ranges and within the overall forest at Ashland were similar and ranged up to a maximum of 500 m from forest edge (reflecting the high degree of fragmentation at this site), although there were very few forested areas at distances exceeding 300 m from the forest edge. The LANDSCAPE TYPE model (model 5) had overwhelming support (w i = 1.0, and DAIC for next closest model = 140.2; Table 2). The likelihood ratio test indicated the LAND- SCAPE TYPE model was much better than the Null (intercept only) model (v 2 8 = 634.4, P < 0.0001). Model coefficients indicate that both random locations within home ranges and from the landscape were a greater distance from edge than cowbird locations and that all locations pooled were closer to edges in the Ashland landscape than other landscapes (Table 3). However, because the interaction effects were strongly supported (Tables 2, 3) interpretation should focus on these rather than the main effects. Coefficient estimates indicate that random locations from within the home range and from the landscape were closest to the edge in the Ashland landscape (Table 3). Patterns in the least-square means estimated from the model also demonstrated that the magnitude of the difference in distance to edge between cowbird and random points varied by landscape and that mean cowbird distance to edge increased from Ashland to Jonesboro to Carr Creek (Fig. 3). Preference of individual cowbirds for forest edge varied greatly among the landscapes. Nearly all individuals at Ashland exhibited no preference for forest close to edges while at Jonesboro and Carr Creek most individuals preferred forest closer to edge (Fig. 4). Only one bird (from Carr Creek) exhibited a preference for forest interior. Discussion We found strong support for the landscape context hypothesis; cowbird preference for forest close to edge is dependent on landscape context. Specifically, in the most fragmented landscape cowbirds did not prefer edge while in the two landscapes with lesser levels of fragmentation cowbirds preferred forest closer to edge. However in an observational study such as ours, it is impossible to isolate the mechanism for this result because fragmented forests occur in landscapes

280 Landscape Ecol (2007) 22:273 284 Table 3 Parameter estimates for the best supported general linear model predicting distance to forest edge of cowbird locations, random points located within home ranges, and random points within the forested landscape for three Midwestern landscapes Effect Category level Estimate SE P-value a Intercept 117.2 9.90 < 0.001 Type Random from home range 118.3 10.08 < 0.001 Type Random from landscape 186.2 9.76 < 0.001 Type Real (cowbird) b Landscape Ashland 51.6 14.04 < 0.001 Landscape Jonesboro 24.4 14.71 0.101 Landscape Carr Creek Landscape Type Random from home range Ashland 93.1 14.73 < 0.001 Landscape Type Random from home range Jonesboro 16.0 14.98 0.222 Landscape Type Random from home range Carr Creek Landscape Type Random from landscape Ashland 155.9 14.68 < 0.001 Landscape Type Random from landscape Jonesboro 48.0 15.02 0.002 Landscape Type Random from landscape Carr Creek Landscape Type Real Ashland Landscape Type Real Jonesboro Landscape Type Real Carr Creek Estimates represent the difference in distance from the reference level for each effect. Sites are listed in order of decreasing fragmentation for each level of TYPE a P-value for null hypothesis test that the parameter estimate = 0 b Parameter not estimated because category level was the reference level and linearly dependent on other parameter estimates with abundant cowbirds and have limited availability of forests distant from edges. Hypothetically, an increase in cowbird abundance in the landscape might force cowbirds to use more of the forest to find hosts and this could result in an Distance from forest edge (m) 350 300 250 200 150 100 50 0 Cowbirds Home Range Random Forest Random Ashland Jonesboro Carr Creek Landscape Fig. 3 Mean distance from forest edge for cowbirds, random points within home ranges, and random points within the landscape in three Midwestern landscapes. Sites are listed in order of decreasing fragmentation. Values for distance from edge are least squares means after adjusting for the effects of landscape and location type. Vertical bars indicate 95% confidence intervals observed lack of preference for forest close to edge (e.g. Jensen and Cully 2005). A lack of preference for forest close to edge at our most fragmented site, however, likely resulted from no change in behavior by cowbirds but a change in availability there was very little forest distant Number of Individuals 30 20 10 0 Edge Preference No Preference Interior Preference Ashland Jonesboro Carr Creek Landscape Fig. 4 Number of individual birds exhibiting preferences for forest edge, forest interior, or no preference in three Midwestern landscapes. Sites are listed in order of decreasing fragmentation

Landscape Ecol (2007) 22:273 284 281 from edge (interior) in the landscape. Thus in this context cowbirds perceived the entire forest at Ashland as all edge. Therefore, although we found strong support for the landscape context hypothesis, this result was primarily due to differences in the availability of forest distant from edge (i.e. core forest) in the three landscapes, and not a change in cowbird behavior. The absence of forest interior at Ashland is a design constraint in our study, but also reflects the true nature of highly fragmented forests which, by definition, lack forest interior. These are often the areas of greatest conservation concern. Studies measuring parasitism rates have similarly observed lack of edge effects in highly fragmented landscapes. For example, Friesen et al. (1999) did not detect significant differences in parasitism rates between small and large fragments in a highly fragmented landscape where parasitism rates were high in both types of fragments. In another study of a highly fragmented study site where the maximum distance to forest edge was 250 m, parasitism rates were also high across sites (Robinson 1992). Nest parasitism data from Ashland indicate high parasitism rates and high numbers of cowbird eggs per parasitized nest (Fig. 1). Collectively these results suggest that cowbirds view highly fragmented forests as all edge. In moderately to greatly fragmented landscapes with abundant cowbirds, competition for hosts may be high (Donovan et al. 2000) so that cowbirds saturate the available forest in search of hosts, or perhaps even prefer interior because of host quality or lower nest predation there (Donovan et al. 2000; Gustafson et al. 2002). Nest parasitism data and numbers of cowbird eggs per parasitized nest at Ashland (Fig. 1) may indicate increased competition for hosts at Ashland. We cannot differentiate whether the patterns observed at Ashland occurred because none of the forest was distant from the edge (a difference in availability) or because cowbirds were abundant, competition for hosts was high, and cowbirds saturated the forest. However, we believe both issues are interrelated and both contributed to the observed pattern at Ashland. Our results underscore the importance of adopting a landscape perspective when considering animal behavior and the effects of edge and forest fragmentation (Wiens et al. 1993; Donovan et al. 1997; Howell et al. 2000; Stephens et al. 2003). If we had only conducted our study at Ashland, as opposed to replicating it across landscapes representing a gradient of fragmentation, we might have incorrectly concluded that cowbirds do not prefer edge. Landscape effects may be especially important considerations for cowbirds because they commute daily over several kilometers (Thompson 1994; Curson et al. 2000). Cowbirds with long distances commutes (approximately 12 km) had reduced numbers of post-ovulatory follicles and lower egg laying rates relative to cowbirds with short distance commutes (2 km) in a New Mexico study (Curson and Mathews 2003). The observed differences among landscapes could have also arisen from differences in landscape composition, not just levels of forest fragmentation. We think this is unlikely because even though landscapes differed in the amount of nonforest cover that was grassland versus rowcrop, these habitat types were used similarly by cowbirds. Cowbirds prefer to forage in the afternoon in feedlot, grassland, or rowcrop habitats (Thompson and Dijak 2000). All landscapes had low proportions of feedlot habitat (0 0.01%). For the two landscapes with greater amounts of forest fragmentation, the combined percent of rowcrop and grassland habitats (which were utilized as cowbird feeding habitat) was 38.6% at Jonesboro and 43.9% at Ashland (Table 1). These two habitat types combined accounted for 50.6 51.3% of the afternoon feeding observations (Thompson and Dijak 2000). Our results indicate that cowbird-mediated edge effects in forests may extend up to and beyond 350 m of the forest edge. Ninety percent of cowbird locations in forest were within 150 350 m of forest edge, despite the availability of forest habitat at greater distances from edge (as far as 500 1450 m) both within cowbird home ranges and the landscape. Where more forest was available (Jonesboro), or where the forest was less fragmented (Carr Creek), cowbirds used forest more distant from the edge than at Ashland but they still exhibited a preference for forest closer to edges. Cowbirds were found up to 800 m

282 Landscape Ecol (2007) 22:273 284 into the forest depending on the landscape, but generally were found within 350 m of the forest edge (Fig. 2). While cowbirds were never found > 800 m from the forest edge, < 5% of the forested landscape existed at that distance. In these three landscapes previously documented mean movements of cowbirds from morning breeding to afternoon feeding sites are large (1020 1262 m, Fig. 1) relative to distances to forest edge. Based on the magnitude of these movements we do not believe cowbirds were limited in their ability to penetrate the forest nor were they forced to breed in forest immediately adjacent to feeding habitat in non-forest habitat. Moreover, the commute distances in our landscapes (< 2 km) were not large enough to lead to the reproductive costs found in other studies (12 km; Curson and Mathews 2003). It is reasonable to ask why cowbirds preferred forest edge, especially since they do not appear limited by commuting distance. Breeding cowbirds may select edges because of a greater density of hosts, because of greater species richness of hosts, because of habitat structure at edges, because of resources in non-forested habitats adjacent to forest edges, or because of a combination of these factors. The habitat structure hypothesis suggests that breeding cowbirds key in on the physical and structural attributes of forest edges. Forest edges, including even relatively minor breaks in the canopy, may be more efficient for host nest searching (Brittingham and Temple 1983; Robinson and Wilcove 1994). This mechanism seemed unlikely in our study because while cowbirds preferred forest closer to edges, much of their use of forest was at distances we believe too distant for vegetation structure to be affected by the edge. Alternately, cowbirds may prefer forest edges merely because they are actually interested in the habitat on the other side of the forest ecotone. This spillover hypothesis (see Manolis et al. 2002) seems an unlikely mechanism in our study because cowbirds spent relatively little time in the habitats adjacent to forest during the morning breeding period and cowbird commuting distances are greater than typical cowbird distances to edge (Thompson 1994). Under the host-density and host-richness hypotheses cowbirds prefer habitats with high host nest densities or host species richness, and edges provide these conditions. We did not survey host abundance or richness as part of this study so we could not directly address this hypothesis. Companion studies in these same landscapes found that host species composition was similar with Wood Thrush (Hylocichla mustelina), Ovenbird (Seiurus aurocapillus), Acadian Flycathcer (Empidonax virescens), and Red-eyed Vireo (Vireo olivaceus) as frequent parasite hosts in all three landscapes (Thompson et al. 2000). However, we cannot posteriorly evaluate how host density varied in relationship to distance to edge. Other studies, however, have related host abundance to edge distance. Several studies have found greater densities and richness of hosts near edges as well as greater parasitism levels (Gates and Gysel 1978; Chasko and Gates 1982; Brittingham and Temple 1983; Flaspohler et al. 2001), where other studies have found no differences in host density and edge distance in fragmented (Hanowski et al. 1995; Hahn and Hatfield 1995) and non-fragmented forests (King et al. 1996). Robinson and Wilcove (1994) reported greater densities of Acadian Flycatcher and Wood Thrush near edges at Jonesboro but parasitism levels were uniformly high across the study area. A study of nesting Kentucky Warbler s (Oporornis formosus) at Jonesboro, however, found that nest parasitism level decreased over a 2 km distance from a known cowbird feeding site but did not differ relative to other edges in the forest (Morse and Robinson 1999). Similarly Hahn and Hatfield (1995) found more forest host nests within 50 m of the forest edge than at further distances but parasitism rates were not greater near the edge. In addition to debate concerning the occurrence of deleterious edge effects and distribution of hosts, there is also disagreement about the relevant scale for edge effects (Brittingham and Temple 1983; Paton 1994; Morse and Robinson 1999; Gustafson et al. 2002). The appropriate scale to examine edge effects depends on the organism and its dispersal abilities (Murcia 1995). For example, nesting songbirds have much smaller home ranges than highly mobile cowbirds (Donovan et al. 1997; Stephens et al. 2003). Landscape context may affect our ability to detect cowbird-mediated edge

Landscape Ecol (2007) 22:273 284 283 effects and the ecologically relevant distance to edge for cowbirds and their hosts may vary depending on the landscape, habitat type, or region (Hahn and Hatfield 1995; Friesen et al. 1999; Thompson et al. 2000; Stephens et al. 2003). Similarly Stephens et al. (2003) found that the effects of habitat fragmentation (including edge effects) on avian nest success were more likely to be detected at landscape scales. This study provides important information about the forest locations preferred by breeding cowbirds and how use and availability of forest edge varies with the landscape. Our data also provide important information for managers contemplating or modeling the potential impacts of forest fragmentation alternatives (sensu Temple and Cary 1988; Gustafson et al. 2002). Our results, however, are based on our assumption that cowbird breeding habitat preference, relative to distance to edge in the forest, is relevant to host parasitism risk. As our purpose was to examine cowbird use of forest, we did not include shrub or grassland (non-agricultural) habitats or hosts, both of which were limited in the landscapes studied. Conclusions Breeding cowbird preference for forest edge varied with the landscape context largely due to differences in the availability of forest edge. In a highly fragmented forest, such as Ashland, all forest is near edge and used by cowbirds. In less fragmented forests, cowbirds preferred forest closer to edge. While our study did not directly address the mechanisms for cowbird edge preferences we believe they relate to patterns in host abundance and diversity because cowbird movements clearly indicate they are capable of using forest farther from edges if they choose to. The variation in edge preference and availability, cowbird numbers, and parasitism levels in these landscapes demonstrates how a focus on local patch or edge effects could be ineffective for conservation. The most effective means to mitigate cowbird parasitism is to take a landscape approach and manage landscape composition to maintain low cowbird numbers. We strongly recommend that future studies investigating cowbird-mediated edge effects report the maximum distance to edge in order to provide landscape context for their study. Acknowledgments We thank G. Krause, S. Romano, S. Latta, E. Kinman, and A. Scheurlein for advice on our analyses. E. Gustafson, F. Huettmann, S. Latta, and D. Burhans commented on earlier drafts of the manuscript. We thank B. Edmond, L. Fray, T. Fredrickson, J. Gardner, B. Hartsell, C. Newbold, A. Taylor, and R. Weidel for assistance on the original radio-tracking study. This research was funded by the USDA Forest Service North Central Research Station, the Mark Twain National Forest, and the Shawnee National Forest. CAH was supported by a NSF BioInformatics Post-Doctoral Fellowship. This is PRBO contribution #1506. References Brittingham MC, Temple SA (1983) Have cowbirds caused forest songbirds to decline? BioScience 33:31 35 Burnham KP, Anderson DR (2002) Model selection and inference: a practical information-theoretic approach, 2nd edn. Springer-Verlag, New York, NY, USA Chasko GG, Gates JE (1982) Habitat availability along a transmission-line corridor in an oak-hickory forest region. Wildl Monogr 82 Coker DR, Capen DE (1995) Landscape-level habitat use by Brown-headed Cowbirds in Vermont. J Wildl Manage 59:631 637 Curson DR, Goguen CB, Mathews NE (2000) Long-distance commuting by Brown-headed Cowbirds in New Mexico. Auk 117:795 799 Curson DR, Mathews NE (2003) Reproductive costs of commuting flights in Brown-headed Cowbirds. J Wildl Manage 67:520 529 Donovan TM, Thompson FR III, Faaborg J (2000) Cowbird distribution at different scales of fragmentation: trade-offs between breeding and feeding opportunities. In: Cooke T, Robinson SK, Rothstein SI, Sealy SG, Smith JNM (eds) Ecology and management of cowbirds. University of Texas Press, Austin, Texas, USA, pp 255 264 Donovan TM, Jones PW, Annand EM, Thompson FR III (1997) Variation in local scale edge effects: mechanisms and landscape context. Ecology 78:2064 2075 Elliott PF (1980) Evolution of promiscuity in the Brownheaded cowbird. Condor 82:138 141 Evans DR, Gates JE (1997) Cowbird selection of breeding areas: the role of habitat and bird species abundance. Wilson Bull 109:470 480 Flaspohler DJ, Temple SA, Rosenfield RN (2001) Speciesspecific edge effects on nest success and breeding bird density in a forested landscape. Ecol Appl 11:32 46 Forman RTT, Godron M (1986) Landscape ecology. J. Wiley & Sons, New York, NY, USA Friesen L, Cadman MD, Mackay RJ (1999) Nesting success of neotropical migrant songbirds in a highly fragmented landscape. Conserv Biol 13:338 346

284 Landscape Ecol (2007) 22:273 284 Gates JE, Evans DR (1998) Cowbirds breeding in the Central Appalachians: spatial and temporal patterns and habitat selection. Ecol Appl 8:27 40 Gates JE, Gysel LW (1978) Avian nest dispersion and fledging success in field-forest ecotones. Ecology 59:871 883 Gustafson EJ, Knutson MG, Niemi GJ, Friberg M (2002) Evaluation of spatial models to predict vulnerability of forest birds to brood parasitism by cowbirds. Ecol Appl 12:412 426 Hahn DC, Hatfield JS (1995) Parasitism at the landscape scale: Cowbirds prefer forests. Conserv Biol 9:1415 1424 Hanowski JM, Niemi GJ, Blake JG (1995) Seasonal abundance and composition of forest bird communities adjacent to a right-of-way in northern forests USA. In: Doucet GJ, Seguin C, Giguere M (eds) Environmental concerns of rights-of-way management. Hydro-Quebec, Montreal, Canada, pp 276 283 Hauber M (2001) Site selection and repeatability in Brown-headed Cowbird (Molothrus ater) parasitism of Eastern Pheobe (Sayornis pheobe) nests. Can J Zool 79:1518 1523 Hochachka WM, Martin TE, Artman V, Smith CR, Hejl SJ, Andersen DE, Curson D, Petit L, Mathews N, Donovan T, Klaas EE, Wood PB, Manolis JC, McFarland KP, Nichols JV, Bednarz JC, Evans DM, Duguay JP, Garner S, Tewksbury J, Purcell KL, Faaborg J, Goguen CB, Rimmer C, Dettmers R, Knutson M, Collazo JA, Garner L, Whitehead D, Geupel G (1999) Scale dependence in the effects of forest coverage on parasitization by Brown-headed Cowbirds. Stud Avian Biol 18:80 88 Hosoi SA, Rothstein SI (2000) Nest desertion and cowbird parasitism: evidence for evolved responses and evolutionary lag. Anim Behav 59:823 840 Howell CA, Latta SC, Donovan TM, Porneluzi PA, Parks GR, Faaborg J (2000) Landscape effects mediate breeding bird abundance in Midwestern forests. Landsc Ecol 15:547 562 Jensen WE, Cully JF Jr (2005) Density-dependent habitat selection by Brown-headed Cowbirds (Molothrus ater) in tallgrass prairie. Oecologia 142:136 149 Jones J (2001) Habitat selection studies in avian ecology: a critical review. Auk 118:557 562 King DI, Griffin CR, Degraaf RM (1996) Effects of clearcutting on habitat use and reproductive success of the Ovenbird in forested landscapes. Conserv Biol 10:1380 1386 Litell RCC, Milliken GA, Stroup WW, Wolfinger RD (1996) SAS system for mixed models. SAS Institute, Cary, North Carolina, USA Manolis JC, Anderson DE, Cuthbert FJ (2002) Edge effect on nesting success of ground nesting birds near regenerating clearcuts in a forest-dominated landscape. Auk 119:955 970 McGarigal K, Cushman SA, Neel MC, Ene E (2002) FRAGSTATS: spatial pattern analysis program for categorical maps. Computer software program available at: http://www.umass.edu/landeco/research/fragstats/fragstats.html Morris DL, Thompson FR III (1998) Effects of habitat and invertebrate density on abundance and foraging behavior of Brown-headed Cowbirds. Auk 115:376 385 Morse SF, Robinson SK (1999) Nesting success of a neotropical migrant in a multiple-use, forested landscape. Conserv Biol 13:327 337 Murcia C (1995) Edge effects in fragmented forests: implications for conservation. Trends Ecol Evol Biol 10:58 62 Ortega CP (1998) Cowbirds and other brood parasites. The University of Arizona Press, Tucson, Arizona, USA Paton PWC (1994) The effect of edge on avian nest success: how strong is the evidence? Conserv Biol 8:17 26 Robinson SK (1992) Population dynamics of breeding Neotropical migrants in a fragmented Illinois landscape. In: Hagan JM, Johnston DW (eds) Ecology and conservation of neotropical migrant land birds. Smithsonian Institution Press, Washington, DC, USA, pp 408 418 Robinson SK, Thompson FR III, Donovan TM, Whitehead D, Faaborg J (1995) Regional forest fragmentation and the nesting success of migratory birds. Science 267:1987 1990 Robinson SK, Wilcove DS (1994) Forest fragmentation in the temperate zone and its effects on migratory songbirds. Bird Conserv Int 4:233 250 Rothstein SI, Verner J, Stevens E (1984) Radio-tracking confirms a unique diurnal pattern of spatial occurrence in the parasitic Brown-headed Cowbird. Ecology 65:77 88 SAS Institute (2000) Version 8.02. Cary, North Carolina, USA Stephens SE, Koons DN, Rotella JJ, Willey DW (2003) Effects of habitat fragmentation on avian nesting success: a review of the evidence at multiple spatial scales. Biol Conserv 111:101 110 Temple SA, Cary JR (1988) Modeling dynamics of habitatinterior bird populations in fragmented landscapes. Conserv Biol 2:340 347 Thompson FR III (1994) Temporal and spatial patterns of breeding Brown-headed Cowbirds in the Midwestern United States. Auk 111:979 990 Thompson FR III, Dijak WD (2000) Movements, home range, and habitat preferences of female Brownheaded Cowbirds in three midwestern landscapes. In: Cooke T, Robinson SK, Rothstein SI, Sealy SG, Smith JNM (eds) Ecology and management of cowbirds. University of Texas Press, Austin, Texas, USA, pp 100 109 Thompson FR III, Robinson SK, Donovan TM, Faaborg J, Whitehead DR (2000) Biogeographic, landscape, and local factors affecting cowbird abundance and host parasitism levels. In: Cooke T, Robinson SK, Rothstein SI, Sealy SG, Smith JNM (eds) Ecology and management of cowbirds. University of Texas Press, Austin, Texas, USA, pp 271 279 Wiens JA, Stenseth NC, Horne BV, Ims RA (1993) Ecological mechanisms and landscape ecology. Oikos 66:369 380