Swainson's Warbler Nest-Site Selection in Eastern Arkansas

Abstract.

By affecting the probability of nest predation and brood parasitism, avian nest-site selection has significant implications for reproduction and fitness. Therefore, understanding factors associated with habitat use at the nest-site scale is imperative, especially for species of conservation concern. One such species, the Swainson's Warbler (Limnothlypis swainsonii), is a rare neotropical migrant that breeds mainly in bottomland hardwood forests throughout the southeastern United States, but relatively little is known about its nesting habitat. From 2004 through 2007, we studied nest-site selection of Swainson's Warblers at two study areas in eastern Arkansas. We found that, relative to paired random plots, nest sites were characterized by dense understory vegetation, high total canopy cover, abundant leaf litter, and high density of woody stems, especially of giant cane (Arundinaria gigantea), a once-abundant bamboo native to the southeastern U.S. Indeed, most nests (90%) were placed in giant cane either exclusively or in combination with other nest substrates. However, understory vegetation density and total canopy cover were the best predictors of nest sites. We suggest that management for Swainson's Warbler nesting habitat should focus on providing forests with uniformly dense understory vegetation and well-developed structure of the canopy and subcanopy. When possible, conservation efforts should focus on maintaining, enhancing, or restoring dense cane thickets.

Resumen.

Debido a que afecta la probabilidad de depredatión de los nidos y de parasitismo de cría, la selection de sitios de anidación en las aves tiene implicaciones importantes para la reproductión y la adecuación. Por lo tanto, es imperativo entender los factores asociados con el uso del habitat a la escala de los sitios de anidación, especialmente para las especies de interés en conservation. Una de dichas especies, Limnothlypis swainsonii, es un migrante neotropical raro que cría principalmente en bosques de bajura de maderas duras a través del sureste de los Estados Unidos, pero se conoce poco acerca de su hábitat de anidación. Entre 2004 y 2007, estudiamos la selection de sitios de anidación por parte de L. swainsonii en dos areas de estudio en el este de Arkansas. Encontramos que, en relatión con parcelas aleatorias pareadas, los sitios de anidación se caracterizaron por presentar vegetación densa en el sotobosque, una cobertura total del dosel alta, hojarasca abundante y una alta densidad de tallos leñosos, especialmente de una especie de bambú (Arundinaria gigantea) nativa del sureste de los Estados Unidos que anteriormente era abundante. De hecho, la mayoría de los nidos (90%) se ubicaron exclusivamente sobre plantas de A. gigantea o en combinatión con otros sustratos. Sin embargo, la densidad de la vegetation del sotobosque y la cobertura total del dosel fueron las variables que mejor predijeron la ubicación de los sitios de anidación. Sugerimos que el manejo del hábitat de anidación de L. swainsonii debería enfocarse en proveer bosques con una cobertura de vegetatión uniformemente densa en el sotobosque y una estructura bien desarrollada del dosel y el subdosel. Cuando sea posible, los esfuerzos de conservación deberían enfocarse en mantener, mejorar o restaurar los parches densos de A. gigantea.

Introduction

Habitat selection is important for migratory birds in that it may significantly affect their reproduction and survival (Cody 1985). Particularly, nest-site selection may affect individual fitness and population dynamics because predation is the primary cause of reproductive failure for most birds (Ricklefs 1969, Martin 1992). Indeed, availability of high-quality nest sites may be a key aspect of territory selection by passerines, and habitat features at the nest-site scale may have direct consequences for reproductive success by affecting the probability of Brown-headed Cowbird (Molothrus ater) parasitism and nest predation (e.g., Martin and Roper 1988, Steele 1993, Burhans 1997, Budnik et al. 2002). Birds may attempt to minimize risk of predation or parasitism by selecting sites that maximally conceal nests or sites surrounded by many substrates, which may decrease predators' searching efficiency (Martin and Roper 1988, Martin 1992, 1993). Because of the consequences for fitness, examination of nest-site selection is vital to understanding the process of habitat selection in birds and is critical for species of conservation concern for which managers seek to provide high-quality habitat (Martin 1992, Jones and Robertson 2001).

One such species of concern, the Swainson's Warbler (Limnothlypis swainsonii), is a medium-sized wood warbler that breeds in the southeastern United States and winters in the Caribbean basin and on the Yucatán peninsula (Brown and Dickson 1994). Because of multiple factors on the breeding and wintering grounds, Swainson's Warbler is among the species of greatest conservation concern both regionally and nationally (Hunter et al. 1993, Twedt et al. 1999, Rich et al. 2004). Rare, yet locally abundant in suitable habitat, Swainson's Warblers breed in bottomland hardwood forests with a dense understory (Meanley 1971). They often select higher ridges within bottomland forests that are flooded infrequently, less often alternative habitats including mid-seral-stage pine plantations and montane rhododendron and hemlock thickets (Meanley 1971, Henry 2004, Bassett-Touchell and Stouffer 2006, Lanham and Miller 2006). A secretive and poorly understood species, Swainson's Warbler remains little known with respect to nesting ecology, largely because of the difficulty of locating nests (Brown and Dickson 1994). In many years of studying Swainson's Warbler, Meanley (1945, 1966, 1969, 1971) found only about 30 nests (Graves 1992), and most information on Swainson's Warbler nest sites is largely descriptive and based on small sample sizes (Brown and Dickson 1994). Only two published studies have directly compared habitat at nest sites to surrounding areas. One was based on a small sample and focused exclusively on canebrakes (Thomas et al. 1996), the other was conducted in early-successional habitat in an intensively managed forest (Peters et al. 2005).

Many historical accounts of Swainson's Warbler mentioned giant cane (Arundinaria gigantea), a bamboo native to the southeastern United States, as a requirement or dominant component of occupied habitat (Brewster 1885, Meanley 1971). Rangewide studies, however, have confirmed that Swainson's Warblers occur routinely in habitat lacking cane (e.g., Meanley 1971, Graves 2002, Bednarz et al. 2005). Yet individuals may prefer cane (Wright 2002, Anich 2008, Brown et al. 2009), and little is known about the relative effects on fitness of habitats dominated by cane or by other species. Given the drastic reductions in the extent of bottomland hardwood forests, especially on the higher elevations within these forests on which cane was formerly abundant, understanding these preferences and relative quality of habitat is important for Swainson's Warbler conservation and management (Noss et al. 1995, Platt and Brantley 1997, Twedt and Loesch 1999).

To better understand factors influencing habitat use and reproduction in Swainson's Warbler as well as to provide information relevant for management, we studied the species' nest-site selection from 2004 to 2007. Given previous research on habitat use in this species and presumed adaptations for maximizing reproductive success, we expected nest sites to be characterized by a dense understory and canopy, which may conceal nests, a high density of woody stems, especially cane, which may provide suitable nest substrates, and abundant leaf litter, which may support abundant food resources (Brown 2008). In this paper, we focus exclusively on nest-site selection and consider the reproductive consequences of this selection elsewhere (Benson et al. 2010a).

Study Areas

We studied nest-site selection of Swainson's Warbler at two locations in eastern Arkansas, St. Francis National Forest (SFNF) and White River National Wildlife Refuge (WRNWR). Although similar in some respects, these locations differ in vegetation cover and structure. SFNF is located on the edge of Crowley's Ridge, an upland “island” in the floodplain, and supports a mix of upland and bottomland forest; at this location Swainson's Warblers occurred in forests transitional between these two types at the base of the ridge and in ravines within the upland forest. WRNWR is exclusively bottomland.

Located in Lee and Phillips counties of east-central Arkansas, SFNF includes over 8500 ha of upland and bottomland forest and is bordered by the Mississippi and St. Francis rivers on the east. This location is managed for multiple uses, including wildlife habitat, and management includes burning and timber harvest, although areas occupied by Swainson's Warblers are generally mature forest and too wet to be extensively burned. Areas used by Swainson's Warblers at SFNF are dominated by elm (Ulmus spp.), box elder (Acer negundo), sweetgum (Liquidambar styraciflua), maple (Acer spp.), oak (Quereus spp.), hickory (Carya spp.), and tulip tree (Liriodendron tulipifera). Greenbrier (Smilax spp.), poison ivy (Toxicodendron radicans), Virginia creeper (Parthenocissus quinquefolia), spicebush (Lindera benzoin), and pawpaw (Asimina triloba) dominate the understory; cane is relatively uncommon at this location.

WRNWR, located in Arkansas, Desha, Phillips, and Monroe counties of east-central Arkansas, is, at over 62 000 ha, among the largest remaining continuous tracts of bottomland hardwood forest in the Mississippi Alluvial Valley (Twedt and Loesch 1999). This area is managed primarily for wildlife habitat, and management includes hydrological manipulations and infrequent timber harvest. At WRNWR Swainson's Warbler habitat is infrequently flooded mature forest dominated by sugarberry (Celtis laevigata), sweetgum, box elder, elm, oak, American sycamore (Platanus occidentalis), and hickory. Dominant understory vegetation includes greenbrier, Virginia creeper, peppervine (Ampelopsis arborea), grape (Vitis spp.), spicebush, box elder, and often dense thickets of cane.

Methods

Nest Searching

After locating territorial males from previous research (Brown et al. 2009) and passive and song-playback surveys, we searched for Swainson's Warbler nests from late April to early August from 2004 through 2007. We located nests systematically, by first estimating the extent of each uniquely marked male's territory and then searching these territories with one to six observers, and opportunistically while engaging in other research. We searched for nests on the basis of the appearance and placement of nest structures (Brown and Dickson 1994) and by observing the behavior of individual birds. We returned to nests at 1- to 4-day intervals to determine their status (i.e., active or inactive) and fate. We defined active nests as those containing apparently viable eggs or young (i.e., not abandoned, failed, or successful prior to discovery).

Vegetation Sampling

After nesting was terminated, we returned to nests and recorded data on vegetation cover and structure by a modified BBIRD protocol (Martin et al. 1997). Likewise, we sampled a paired random site for each documented nest; these random sites were located within forest a random distance and direction between 25 and 75 m from each nest and were searched to confirm the absence of a Swainson's Warbler nest. We chose this distance to eliminate overlap between nest and random samples and to maximize the probability that random locations were within the home range of the nesting pair and so available to the birds for nest-site selection. The mean home range at these locations was 9.4 ha (range 1.6–30.8 ha; Anich et al. 2009); if home ranges are circular, the 75-m distance (1.8 ha) was reasonable for most if not all birds. Prior to sampling, we trained all observers to ensure consistency, and data for each combination of nest and paired random site were recorded by the same observers. Similarly, observers sampled approximately equal proportions of nests and random sites at both SFNF and WRNWR to avoid systematic bias in the results.

At each point, we recorded data at the nest-site scale (i.e., within a 5-m radius) and at the nest-patch scale (i.e., within an 11.3-m radius). At the nest-site scale, we estimated percent ground cover of grasses and sedges, forbs, shrubs, cane, total green vegetation, brush (downed woody vegetation <8 cm in diameter), leaf litter, logs (downed woody vegetation >8 cm in diameter), and bare ground in four quadrants of the circle. At a distance of 5 m in the four cardinal directions, we counted the number of cane, vine, and shrub stems within a 1-m2 quadrat at a height of 0.3 m, measured the leaf-litter depth to the nearest millimeter with a ruler, and estimated soil moisture on an 11-point scale (0 = completely dry, 10 = completely saturated) with a soil-moisture meter (Lincoln Irrigation, Inc., Lincoln, NE). For analyses, we used the average of the four measurements of cover, stem density, litter depth, and soil moisture. We estimated litter volume as plot area × percent leaf-litter cover × leaf-litter depth in meters. From the center of each plot, we estimated total canopy cover in the four cardinal directions by using a spherical densiometer at a height of about 1.3 m and recorded an ocular estimate of subcanopy cover. We estimated average overstory and subcanopy height within the nest patch with a clinometer. Also at the nest-patch scale, we counted saplings (>30 cm tall, <2.5 cm diameter; including shrubs), poles (2.5–8 cm diameter at breast height [dbh]), small trees (8–23 cm dbh), medium trees (23–38 cm dbh), large trees (>38 cm dbh), small snags (>1.4 m tall, <12 cm dbh), large snags (>1.4 m tall, >12 cm dbh), and vine tents, defined as conspicuous accumulations of terrestrial or hanging vines that form a dense and largely impenetrable canopy.

We estimated understory density by placing a 2.5-m-tall cover board at the plot's center and recording the percent of the board obscured by vegetation in five different height classes, 0–0.5 m, 0.5–1.0 m, 1.0–1.5 m, 1.5–2.0 m, and 2.0–2.5 m (Nudds 1977). We repeated this from both 5 and 11.3 m in the four cardinal directions. Our estimates of understory density for each height interval are based on the mean of the four horizontal measurements, and our overall density estimates are based on the mean of all 20 readings for each point. Because heterogeneity in vegetation structure may also be important to Swainson's Warbler, we calculated coefficients of variation (CV) for cane and total stem density, for understory-density readings at each point for the four cardinal directions averaged over all height intervals (horizontal CV), and the five vertical height intervals averaged over all cardinal directions (vertical CV).

Data Analyses

We investigated differences between nests and paired random sites with mixed-model ANOVA and Poisson regression (SAS PROC MIXED and GLIMMIX; Littell et al. 2006). Because habitats occupied by Swainson's Warblers at SFNF and WRNWR differed in structure and composition (Anich 2008), in addition to differences between point types (i.e., nest or random), we also examined the effect of location (i.e., SFNF or WRNWR) and the point type × location interaction. Our sample of points included nests that were known to be active and empty nests of unknown fate. To examine whether these empty nests were an unbiased sample with which to investigate nestsite selection (i.e., point type, location, or point type × location effects at active and inactive nests were consistent), we examined effects of the interactions active × point type, active × location, and active × point type × location on habitat variables with MANOVA (SAS PROC GLM; Littell et al. 2002:312).

To account for potential non-independence of data points, we considered models with nest and year as random effects and fit these models with variances homogeneous and heterogeneous by point type, location, or both point type and location. We selected the best model for variance structure with Akaike's information criterion adjusted for small sample sizes (AICc; Littell et al. 2006:343–411). We also fit geostatistical models that allowed independence to vary as a function of distance between data points and compared these models to the above random-effects models by using AIC_c (Littell et al. 2006:437–478, Christman 2008; for details see Benson 2008). Using the Kenward—Roger approximation of denominator degrees of freedom (Littell et al. 2006:188), we based inferences on the best-fit random-effects or spatial model (i.e., on the basis of AIC_c). We examined plots of residuals from our best-fit models to confirm our data approximated a normal distribution; for variables that did not conform to a normal distribution, particularly those with many zero values, we used generalized linear mixed models with the Poisson distribution (SAS PROC GLIMMIX; Littell et al. 2006:557–560). We set the significance level for all analyses at α = 0.05 and when the point type × location interaction was significant, we evaluated differences among point-type-by-location combinations with contrasts within SAS PROC MIXED or GLIMMIX (Quinn and Keough 2002:196199; Littell et al. 2006).

To ascertain which variables were most important for differentiating between nest and random sites, we generated 21 a priori models for nest-site selection based on factors we believed to be important for Swainson's Warbler habitat use or avoidance of nest predation on the basis of previous research (e.g., Meanley 1971, Martin 1992, 1993, Graves 2002, Bednarz et al. 2005) and personal observations (Table 1). Prior to testing models, we evaluated correlations among variables and did not include highly correlated variables (|r|>0.70) in the same model. We examined these a priori models by using conditional logistic regression, which accounts for the non-independence between nests and paired random sites (SAS PROC PHREG; Allison 1999, Bailey and Thompson 2007). We ranked candidate models by AIC_c and computed model weights and importance values from the sum of weights over all models in which a variable occurred (Burnham and Anderson 2002:168). Likewise, to account for model-selection uncertainty, we model-averaged the parameter estimates in the models accounting for 90% of the Akaike weight (Burnham and Anderson 2002:169) and present odds ratios and unconditional 95% confidence intervals based on these modelaveraged estimates (Burnham and Anderson 2002:149–167, Keating and Cherry 2004).

To further explore the relationships among variables and to visualize the extent of overlap among nest and random sites at our two locations, we used principal components analysis (SAS PROC FACTOR; McGarigal et al. 2000) on the variables included in our 21 a priori models (Table 1). We performed principal components analysis on the correlation matrix of variables and present loadings for all components with eigenvalues ≥1. We interpreted variables with loadings ≥ |0.5| as descriptors of each principal component (McGarigal et al. 2000).

Results

From 2004 to 2007, we found 269 nests, of which 135 were active (i.e., had apparently viable eggs or young). Fifty-two nests successfully fledged young, and analyses of factors influencing nest survival and fledgling production are presented elsewhere (Benson et al. 2010a). Empty nests likely represented early failed breeding attempts, and in some cases possibly early successful attempts or nests built and abandoned prior to laying. Nonetheless, all Swainson's Warbler nests represented breeding attempts from the year in which they were found, as nests rarely persist between breeding seasons and are clearly recognizable as >1 year old when they do persist (Benson, pers. obs.). We recorded habitat data at 224 nests (including all active nests), 13 in 2004, 64 in 2005, 92 in 2006, and 55 in 2007. Of 269 nests, 241 (90%) were placed in cane exclusively (222; 83%) or in combination with other vegetation (e.g., shrubs or vines; 19; 7%). There was no significant effect of active × point type (Wilks' λ = 0.93, F_43,322 = 0.5, P = 0.99), active × location (Wilks' λ = 0.87, F_43,322 = 1.1, P = 0.31), or active × point type × location (Wilks' λ = 0.91, F_43,322 = 0.7, P = 0.91), suggesting that any habitat differences between active and inactive nests are consistent for point type, location, and the point type × location interaction.

Nests and randomly selected sites differed conspicuously (Tables 2 and 3). At both SFNF and WRNWR litter depth and volume were greater at nest sites than at paired random sites, although litter cover values were similar (Table 2). Likewise, nest sites had greater density and cover of cane, greater density of vine and total stems, and greater average understory height but lower cover of green vegetation, grasses, forbs, and vines than the random sites. At the nest-patch (11.3 m) scale, nests had greater total canopy and subcanopy cover but lower canopy and subcanopy height than non-nest patches (Table 3). For all height intervals understory density was greater at both the nest-site and nest-patch scales, and these areas were uniformly dense relative to random areas (i.e., lower CV values; Tables 2 and 3, Fig. 1).

In addition to differences between nest and random points, there were several habitat differences between SFNF and WRNWR. SFNF had lower depth, cover, and volume of litter, greater shrub-stem density and patchiness in total stem density, and greater cover of green vegetation, grasses, and forbs than did WRNWR (Table 2). SFNF was also characterized by greater subcanopy cover, lower subcanopy height, and greater abundance of saplings and poles but lower abundance of medium-sized trees and large snags (Table 3).

Some habitat differences between nests and random sites at the two locations were not consistent. Although at both locations nest sites had greater cane-stem density and cane cover than random sites, the density and cover at nests were greater at WRNWR than at SFNF. At SFNF patchiness of cane stems (CV) was greater at nests than at random sites, likely because cane-stem density was uniformly low in randomly selected areas away from nests (Table 2). Because of overall low shrub-stem density at WRNWR, there was no difference in shrub-stem density between nests and random sites at that location, but at SFNF density was lower at nests than at random sites. Additionally, at SFNF the abundance of small trees was lower at nest than at random patches; at WRNWR the difference was less extreme (Table 3). At SFNF, small-snag abundance was greater in random patches than in nest patches, and vine tents were more common in nest patches than random patches, but at WRNWR there was no difference in either of these variables (Table 3). Understory density at nests was similar at both locations, but random locations were denser at WRNWR than at SFNF (Tables 2 and 3, Fig. 1).

We had complete data for all variables of interest and performed conditional logistic-regression analyses for 208 nests and their associated random sites. All of our a priori models were better predictors of nest sites than the null model and, despite some model-selection uncertainty, several variables emerged as consistent predictors of nest sites (Table 4). Mean understory density, total canopy cover, cane-stem density, and litter volume all appeared in the models accounting for the top 90% of the Akaike weight, with importance values of 1.00, 0.99, 0.28, and 0.32, respectively. All other variables had importance values of <0.10. On the basis of model-averaged parameter estimates, only the odds ratios for average understory density and total canopy cover had unconditional 95%) confidence intervals not including 1.0 (Fig. 2A), with greater values of understory density and canopy cover being associated with an increased relative probability of selecting habitat at nest sites over random sites (Fig. 2). Cases where understory density and total canopy cover were greater at random than at nest locations were rare, and, correspondingly, the relative probability of these situations occurring at nest sites was close to zero (Fig. 2B). However, larger values for both variables at nests than at random locations were associated with a near-certain relative probability of nest-site occurrence. Although not the best model, the stem-type model fit better than a total-stems model with odds-ratio estimates from the stem-type model of 1.88 (1.55–2.27 [95% CI]), 1.35 (1.04–1.75), and 1.19 (1.06–1.33) for cane, shrub, and vine stems, respectively, and an odds ratio from the total-stems model of 1.40 (1.26–1.55).

Six principal components, accounting for 71% of the variation, had eigenvalues ≥1 (Table 5). The first component (PC1) explained primarily variation in cane-stem density, understory density, and understory uniformity; the second component (PC2) described variation in subcanopy cover and forest height; the third component (PC3) was primarily related to density of stems other than cane (Table 5). PC1 separated nests and random sites with relatively little overlap, whereas nests and random sites overlapped extensively for PC2 and PC3 (Fig. 3). Similarly, there was extensive overlap between SFNF and WRNWR for the top three components.

Discussion

At our two mature-forest sites, Swainson's Warbler nest sites were characterized by a dense understory with abundant shade and leaf litter, sparse cover of herbaceous vegetation, and high stem density (especially of cane stems) relative to randomly selected sites. This pattern is largely consistent with what is known about Swainson's Warbler habitat use in general (Meanley 1971, Eddleman et al. 1980, Graves 2002) and the few previous studies of habitat at this species' nests (Thomas et al. 1996, Henry 2004, Peters et al. 2005). In early-successional bottomland forests in South Carolina, nest sites had greater cane density and cover, litter cover, and understory density than did other points in the territory or outside it; midstory and overstory height were lower at nests than at these other points (Peters et al. 2005). Similarly, in Missouri canebrakes nest sites had greater litter and less grass cover than random plots (Thomas et al. 1996), and in Louisiana bottomland forests nest sites had greater stem and understory density than did points elsewhere in the territory (Henry 2004). Contrasting with our results, Henry (2004) found that nest sites and points in the territory away from the nest had similar leaf-litter depth and cover of herbaceous vegetation and canopy. Similarly, Peters et al. (2005) observed greater vine cover at nests than at other sites, whereas we observed the opposite pattern. This difference was likely related to differences between our study areas in management history and consequently the age and floristic composition of forest stands. Indeed, in a more intensively managed forest in eastern Arkansas, we have observed that vines are a more important component of Swainson's Warbler habitat than at our mature-forest sites (Benson and Bednarz, unpubl. data).

Our results suggest that when cane is present in a home range, Swainson's Warblers often use it for nest sites. Indeed, most nests were partially or completely placed in cane (90%)), and cane has been recognized as a Swainson's Warbler nest substrate even in areas where its density is relatively low (Peters 1999, Henry 2004, Thompson 2005). This is consistent with previous findings that, on average, areas used by Swainson's Warblers have more cane than unused areas and cane is often among the best predictors of Swainson's Warbler habitat use (Wright 2002, Somershoe et al. 2003, Peters et al. 2005, Brown et al. 2009).

Although not the best model, density of cane stems was a better predictor of nest sites than stem type, which includes shrub and vine stems, or all woody stems combined. Cane-stem density likely was not a better predictor of nest sites than understory density because our sampling did not incorporate cane height. As observed in Missouri (Thomas et al. 1996), our nest sites were associated not only with cane but, more specifically, with tall cane. Therefore, understory density, which incorporates aspects of both stem density and height, was a better predictor of nest sites even though this dense vegetation structure was most often created by cane. In fact, cane density at nest sites was a better predictor of understory density than the density of any other stem type (cane stems: r = 0.41, P < 0.001; shrub stems: r = 0.00, P = 0.98; vine stems: r = -0.07, P = 0.31), and both cane-stem density and understory density were combined into a single principal component which effectively differentiated nest sites from random sites.

In some areas, stems other than cane provide appropriately dense structure, and Swainson's Warblers breed in areas where cane is rare or absent (Meanley 1971, Graves 2002). At our study areas, plants other than cane rarely provided the suitably dense understory that Swainson's Warblers prefer, and no other plant species was as consistently associated with nest sites. However, Swainson's Warblers use other nest substrates in the absence of cane or, even if cane is present, when other plant species provide suitable structure. Although cane was used in Louisiana and South Carolina, most nests in these locations were found on other substrates (Peters 1999, Henry 2004, Thompson 2005). In an Appalachian population, 81% of nests were located in small eastern hemlocks (Tsuga canadensis; Lanham and Miller 2006). Similarly, in another area of eastern Arkansas where we have studied Swainson's Warbler (Benson 2008), cane is relatively sparse yet Swainson's Warbler densities are greater than at either SFNF or WRNWR, likely because greenbrier provides a uniformly dense understory at this location.

Even though all of our models were good predictors of nest sites, the best variables were understory density and total canopy cover. The importance of these variables for Swainson's Warbler was established in previous studies (e.g., Meanley 1971, Thomas et al. 1996, Somershoe et al. 2003, Bednarz et al. 2005), with total canopy cover among the best predictors of presence (Mitchell et al. 2001, Brown et al. 2009). In addition to providing concealment, canopy cover may be associated with increased abundance or diversity of arthropods on which the warbler feeds (Brown 2008). Leaf-litter volume, although important for Swainson's Warbler habitat use in general (e.g., Graves 1998, Bednarz et al. 2005, Brown et al. 2009) seemed less important at nests than concealment and abundance of potential nesting substrates. However, cane-stem density and understory density are also good predictors of arthropod richness and abundance (Brown 2008).

Habitat features selected for nest sites are not necessarily adaptive, especially when human activities may decouple historically reliable cues related to reproductive success, resulting in possible ecological traps (e.g., Gates and Gysel 1978, Robertson and Hutto 2006, Arlt and Pärt 2007). Therefore, the consequences of nest-site selection for reproductive success are also important to understand. At these two locations, Swainson's Warbler nests in dense areas were less likely to receive multiple cowbird eggs, which cause complete reproductive failure (Benson 2008), and less likely to be depredated by cowbirds (Benson et al. 2010b). However, increased cane-stem density was weakly associated with lower nest survival, especially at WRNWR; nests in sites dominated by plants other than cane had somewhat greater survival at SFNF (Benson 2008, Benson et al. 2010a) and were less likely to be depredated by ratsnakes (Elaphe obsoleta) at WRNWR (Benson et al., 2010b). Although some patterns linking habitat features and reproductive success were discernible, none were especially strong predictors of productivity, likely because habitats are highly modified and contain complex predator communities (Benson et al. 2010a, b).

To promote Swainson's Warbler conservation, management should focus on relatively high-elevation bottomland hardwood forests free from the adverse effects of frequent flooding (Graves 2001, Twedt et al. 2006). Management of these areas should include creating and maintaining forests with dense understories and high total canopy cover. In many areas, this dense understory structure can be created silviculturally through canopy disturbance (e.g., Hetzel and Leberg 2006, Twedt and Wilson 2007). Although managers should focus efforts on maintaining or enhancing existing forested canebrakes where practical, canopy disturbances that promote development of dense shrub or vine growth should also benefit Swainson's Warbler and are likely more effective in many cases, given the difficulties associated with cane restoration. Given the reliance of Swainson's Warblers on dense understory and abundant canopy cover, managed forests should likely be maintained as a mosaic of stands of different ages to ensure landscape-level availability of suitable habitat. Managers should avoid extirpating existing isolated populations.

Acknowledgments

Funding for this research was provided by the Arkansas Game and Fish Commission (AGFC) and U.S. Fish and Wildlife Service (USFWS) through a State Wildlife Grant. We thank C. Rideout and L. Barnhill, formerly of AGFC, for continued support of this research and help securing funding. R. Hines, C. Hunter, and S. Reagan of USFWS provided invaluable assistance. P. Hamel, G. Graves, T. Risch, R. Grippo, M. Patten, and an anonymous reviewer provided valuable comments on an earlier version of the manuscript. We thank C. Roa, J. Sardell, M. Albrechtsen, K. Ballantyne, D. Baxter, E. Huskinson, J. Jackson, K. Jones, C. McCarroll, J. O'Connell, B. Paterson, A. St.-Pierre, D. Townsend, and W. Edwards for providing valuable field assistance.

Literature Cited