Response of butterflies to structural and resource boundaries


Correspondence author. E-mail:


1. Two aspects of landscape composition shape the behavioural response of animals to habitat heterogeneity: physical habitat structure and abundance of key resources. In general, within-habitat movement behaviour has been investigated in relation to resources, and preference at boundaries has been quantified in response to physical structure.

2. Habitat preference studies suggest that responses to resources vs. structure should differ, e.g. between male and female animals, and effects of responses to structure and resources may also interact. However, most studies of animal movement combine various aspects of behavioural responses to ‘habitat’, implicitly assuming that resources and structure are broadly equivalent.

3. We conducted a large-scale experiment of the movement of Fender’s blue (Icaricia icarioides fenderi), an endangered butterfly, to investigate butterfly response to physical structure of the landscape (prairie, open woods and dense woods) and to resources [presence or absence of Kincaid’s lupine, Lupinus oreganus (larval hostplant patches)]. The experiment included 606 butterfly flight paths across four habitat types and nine ecotones.

4. Responses to physical structure and resource patches were not congruent. Butterflies were attracted to resource patches within both prairies and open woods and moved more slowly when in resource patches. Butterflies tended to prefer prairie at prairie-forest edges but tended to move faster in prairies than in open woods. Physical structure and resources also interacted; butterflies did not respond to physical habitat structure when resource patches spanned prairie – open woods ecotones.

5. Even dense woods were not perfect barriers, in contrast to a large body of literature that assumes insects from open habitats will not enter dense forests.

6. Movement of both males and females responded to resources and structure. However, female butterflies had stronger responses to both resources and structure in most cases. Females had strongest response to resource (hostplant) patches at patch edges, whereas the strongest preference of males was to return to prairie from open forest.

7. If other species behave like Fender’s blue, then combining different definitions of ‘habitat’ (physical structure vs. resources), different aspects of movement (edge preference vs. within-habitat movement) and/or males and females within species could all lead to misleading conclusions. Our results highlight the importance of investigating these responses, and our study provides a framework for separating them in other systems.


Understanding animal movement in heterogeneous environments is one of the most challenging problems in ecology. Elucidating general conclusions about animal responses to different habitat types is problematic because definitions of habitat are often vague, awkward and confusing (Dennis 2010). Typical definitions of animal habitat include at least one of two aspects of landscape composition: physical structure and resource distribution (e.g. Odum 1971; Dennis & Sparks 2006). In some dispersal studies, resources are mapped, and boundaries between resource areas and non-resources areas are the boundary between habitat and matrix (e.g. Schultz & Crone 2001; Schtickzelle & Baguette 2003; Haynes & Cronin 2006; Chapman, Dytham & Oxford 2007). In other studies, structural aspects are used to define habitat, such as woods and meadows (e.g. Ries & Debinski 2001; Vos et al. 2007). In many of these cases, resources are found within one habitat type, but structural features, rather than resource distributions per se, are used to determine habitat boundaries (e.g. Roland, Keyghobadi & Fownes 2000; Goheen et al. 2003; Chin & Taylor 2009). Finally, resources may span multiple structural habitats types (Haddad 1999). Responses to resource vs. structural boundaries are rarely clearly delineated, and few studies investigate these attributes separately in the literature.

In relation to both physical structure and resources, animal responses are characterized by two aspects of movement: behaviour within patches and behaviour at boundaries (Schultz 1998; Haddad 1999; Chapman, Dytham & Oxford 2007; Ovaskainen et al. 2008b). The distinction between habitat-specific responses and responses to boundaries is important because these responses are not necessarily correlated and they have different implications for dispersal (Kuefler et al. 2010). Habitat-specific movement behaviour has been extensively studied with respect to resource aspects of habitat (e.g. Morris & Kareiva 1991; Haynes & Cronin 2006; Schtickzelle et al. 2007). In general, individuals move more slowly through high-quality habitat in response to dense resources (Dover 1997; Schultz & Crone 2001; Klaassen, Nolet & Bankert 2006), which increases the time they spend in that habitat type (Kareiva & Odell 1987; Morris & Kareiva 1991; Kuefler et al. 2010). In contrast, investigations into boundary response often focus on structural rather than resource aspects of habitat (Strayer et al. 2003; Ries et al. 2004; Jackson et al. 2009). The fact that habitat-specific movement has been most often studied in relation to resources, and edge behaviour in relation to structure, could reflect general patterns in nature, but it could also reflect researcher biases. To our knowledge, no previous studies have tested the relative importance of resources vs. physical structure in determining habitat-specific movement and edge behaviour in natural landscapes.

For many species, assessing animal responses to habitat heterogeneity is also confounded by differential responses by males and females (Greenwood 1980; Ruckstuhl & Neuhaus 2005). Sex-specific differences in habitat preference have been observed in a wide range of species including ungulates, primates, bats, birds, reptiles, fishes, seals and insects (Ruckstuhl & Neuhaus 2005; Breed et al. 2006; Hahn & Reinhardt 2006). During the breeding season, females may be more driven by offspring needs (obtaining food to feed offspring and finding suitable sites for oviposition, nesting or rearing offspring), whereas males explore habitat more broadly to attract or find mates. For example, sex-specific differences in micro-/macro-habitat selection are common in birds and attributable to size-mediated dominance, reproductive role specialization and/or trophic niche divergence (Phillips et al. 2004). Similarly, sex differences in habitat preference in midges result open wind-sheltered rocks being dominated by males, while females congregate under rocks where oviposition occurs (Hahn & Reinhardt 2006). Conde et al. (2010) noted that differences between sexes are often ignored in habitat models; however, when behavioural differences do exist, the lack of discrimination could result in incorrect or misleading conclusions. Unfortunately, numerous dispersal-related studies combine sexes (e.g. Baguette, Petit & Queva 2000; Kuefler et al. 2010). If sex-specific habitat preferences translate to differences in movement behaviour, then combining sexes could lead to misleading conclusions about landscape scale processes.

In this study, we investigate movement behaviour of Fender’s blue butterfly (Icaricia icarioides fenderi = Plebejus icarioides fenderi) in relation to structural and resource boundaries. Previous research in this system (Schultz 1998; Schultz & Crone 2001) indicates strong responses to resource patches, defined by the butterfly’s larval hostplant, Kincaid’s lupine (Lupinus sulphureus spp. kincaidii = Lupinus oreganus). We focus on lupine hostplant as a fundamental resource with a known and quantifiable response by butterflies. Here, we evaluate the relative responses of butterflies to structural boundaries (forest edges) in relation to resource patches. Open woodlands support prairie vegetation between the trees, including resource patches. Dense woodlands have a shady understorey and do not support hostplant resources. We also test whether females are more responsive than males to resource and/or structural habitat features. Although females respond more strongly to resources, i.e. oviposition sites (Schultz & Crone 2001), there is less reason to anticipate that males and females would respond differently to forest edges, which are associated more with overall ease of movement and thermal environment. Together, these comparisons provide a much-needed evaluation of the extent to which animal movement reflects simple binary responses to habitat presence vs. absence, the typical assumption in dispersal ecology, as opposed to subtler behavioural responses to resources or physical habitat structure.

Materials and methods

Study species and system

Fender’s blue is a federally endangered butterfly, which persists in upland prairie remnants in Willamette Valley, OR, USA that maintain its larval hostplant, Kincaid’s lupine (federally threatened). The flight season for the butterfly is May and early June. This study took place in the Cardwell Hills region (44°35′N/123°25′W), c. 18 km west of Corvallis (Benton County, Oregon, Fig. 1). Remnant prairies in Cardwell Hills support c. 500–1500 Fender’s blue butterflies (C. B. Schultz, pers. obs), c. 15–20% of all known individuals of this endangered species (Fitzpatrick 2009; Hammond 2009). Rigorous surveys of population size have not been conducted in this area. Butterfly populations at this site are found in or near a range of structural habitat types, which can be broadly categorized as dense conifer forests, open woodlands and prairies. Previous anecdotal observation of Fender’s blue in this landscape suggested that the dense woods were a barrier to butterfly movement and raised concern that forest encroachment may further isolate remaining patches of butterfly habitat.

Figure 1.

 Map of study area in Cardwell Hills. Black triangles are butterfly release locations for behaviour experiments.

We analysed butterfly movement in relation to two components of landscape composition: hostplant presence and woodland structure. We mapped the distribution of Kincaid’s lupine (Magellan ProMark III GPS, Santa Clara, CA, USA) with 1 m accuracy and used these data to create a lupine distribution map in GIS. We quantified woodland structure in three categories: dense woods, open woods and prairies. Dense woods were dominated by mature Douglas Fir (Pseudotsuga menziesii) with little sun in the understorey. Open woods included conifer-encroached prairies in which openings were common and many of the conifers were small (<10 years old). Other areas of open woods were dominated by low-density deciduous riparian vegetation. Open woods maintained adequate sun for butterfly movement during the day. Prairies were entirely open, with occasional isolated trees. Maps of forest cover were created using aerial photographs (3Di West Inc, Eugene, OR, USA, July 2005) over which polygons of dense woods, open woods and prairie could be drawn with c. 2 m accuracy. Dense woods had canopy cover >90%; open woods had canopy cover 20–70%; and prairies had canopy cover of <10%. Habitat with canopy cover of 10–20% and 70–90% were considered open woods and were only a small percent of the landscape. Lupine patches occurred in both prairies and open woods (Fig. 1). Therefore, our study area consisted of five habitat types (PL, prairie with lupine; PN, prairie without lupine; WL, open woods with lupine; WN, open woods without lupine; and D, dense woods), and we analysed both responses to the presence of woodlands and how woodlands modified behavioural responses to lupine patches.

Movement behaviour

To assess the response of Fender’s blue butterflies to landscape composition, we conducted a behavioural experiment in which we released butterflies at nine boundary types, tracked their behaviour and mapped flight paths. Our behavioural approach was based on methods used and tested in our earlier studies of habitat-specific movement and boundary behaviour (Schultz 1998; Schultz & Crone 2001). In brief, to map flight paths, butterflies were captured, cooled and released at focal locations for observation, a technique which has little impact on subsequent movement (Schultz 1998). Butterflies were marked with a small dot of permanent powder to ensure that each observation included a unique individual. Butterflies were followed, and a flag was dropped every time the butterfly landed or every 20 s during flight for a maximum of 15 flags. When each flag was dropped, the time of the event and the butterfly’s behaviour were recorded on a hand-held computer. At the end of each observation day, locations of flags were recorded with 1 cm accuracy using a differential GPS (Magellan ProMark III GPS). To achieve this accuracy, we established an on-site base station, which recorded satellite data throughout the season. This procedure allows post-processing of data to 0·1 m accuracy. From these data, we estimated move length, time in flight and turning angle for each move (Turchin 1998, p. 99). In addition to mapping flag location, we noted presence of lupine and physical structures; field notes were systematically cross-checked with landscape maps to ensure map accuracy.

Experiments were designed to test behavioural response to the two primary habitat features: lupine patches and physical structure (Prairie/Dense woods and Prairie/Open Woods). Therefore, release sites differed by release habitat (PL, PN, WL and WN) and proximity to resource (lupine/non-lupine) or structural (dense woods, open woods and prairie) boundaries (Table 1). Release sites were spread across the landscape (Fig. 1). All releases occurred within 15 m of the boundary of interest based on past estimates of perceptual range (Schultz & Crone 2001). At each boundary type, we aimed to release 30 butterflies (15 males and 15 females) in each of two flight seasons, 2007 and 2008; actual sample sizes depended on variation in weather and butterfly population sizes throughout each flight season (see Results below). Using these flight paths, we estimated response to habitat boundaries, as well as habitat-specific movement parameters, once butterflies had left the edge. It was not physically possible to track butterflies in dense woods because they entered them at c. 5–10 m above the ground (C. B. Schultz, A.M.A. Franco and D. L. Roberts, pers. obs).

Table 1.   Boundary experiments and analyses. Experimental and simulated data are per-move probability of crossing boundary. Shaded rows are butterfly response to resource boundaries; non-shaded rows are butterfly responses to structural boundaries, as modified by the presence or absence of resources Thumbnail image of

Data analysis

Overall, our analysis was based on a correlated, random walk model of animal movement, with bias at patch edges. Our previous research with this species has shown that this framework adequately predicts large-scale movement of Fender’s blue butterflies, at least for female butterflies (cf. Schultz 1998; Schultz & Crone 2001; McIntire, Schultz & Crone 2007; Crone & Schultz 2008).

We quantified habitat-specific movement in the four habitat types (PL, PN, WL and WN) by assessing individual moves along the flight path in terms of length (hereafter, ‘move length’), flight time (flag to flag time in flight in seconds) and direction (converted to turning angle, i.e. the difference between the direction of each step and the previous step in a flight path). Because of directional bias at lupine patch edges (Schultz & Crone 2001), all moves in PN and WN within 15 m of a lupine boundary were removed from analyses of habitat-specific parameters. We analysed move lengths and turning angles as a function of physical structure, lupine presence, sex (male vs. female), and year. Year effects were included because 2007 was a relatively warm, sunny flight season and 2008 was a relatively cold, cloudy flight season [May 2007 average high temperature = 20·8 °C with 17 clear, 10 partly cloudy and four cloudy days; May 2008 average high = 19·4 °C with eight clear, 11 partly cloudy and 12 cloudy days, National Climatic Data Center (NCDC), available from: accessed May 21, 2010]. One key constraint of this type of data is that individual moves are not independent. Instead, each path is independent and a sample of that individual’s behaviour (Turchin 1998). One solution to this is using individual as a random effect (Gillies et al. 2006). Therefore, to account for possible differences among individual butterflies, analyses were implemented as mixed models lmer function and lme4 package in R (Bates, Maechler & Dai 2008; R Development Core Team 2008) with individual as a random factor. Move length and flight time data were log-transformed prior to analysis. Turning angles were analysed as cos (θ) (following Turchin 1998), then rescaled from 0 to 1 and arcsin-squareroot transformed to be approximately normal. In addition to analysing individual move parameters, we calculated habitat-specific diffusion coefficients (following methods in Turchin 1998) with confidence limits by bootstrapping flight paths with replacement. Individual butterflies were randomly selected with replacement, and then habitat- and sex-specific diffusion coefficients were calculated. Diffusion rate is an integrative measure that combines observed differences in move lengths, turning angles and time in flight. Slower diffusion suggests that butterflies spend more time in a particular habitat type, and therefore move through that habitat type less quickly (Turchin 1991). In making inferences from these analyses, we concluded groups differed significantly when the mean for one group does not overlap the 95% confidence interval for the other group and vice versa.

To investigate behaviour at boundaries, we used a failure time analysis assuming a constant fail rate (exponential distribution of failure times) to assess the per-move probability that a butterfly will cross a boundary. For each butterfly released within 15 m of a boundary, we recorded the first step at which it (i) crossed the boundary (paths were scored as failures at this point); or (ii) left the ecotone (moved to >15 m from the boundary) or were lost by observers (paths were scored as ‘censored’ after this point). Failure time analysis was implemented as a mixed-effects binomial model (glmer function in lme4), with individual as a random effect. This random effect accounts for differences among flight paths, including autocorrelation in move direction and other differences among individual butterflies. The expected failure times from random movement depend on landscape configuration, e.g. relative to a straight-line boundary, we expect higher edge-crossing rates at concave edges and lower crossing rates at convex edges. Expected failure times also cannot be calculated analytically because, although diffusion approximates correlated random walk at long time-scales, it does not over the course of short flight paths such as these (see, e.g. Ovaskainen & Crone 2010). Therefore, we tested edge preference at habitat boundaries by comparing observed failure times to simulated failure times, based on expectations from correlated random walk behaviour. Specifically, simulated flight paths were started from the release point of each butterfly (see Fig. 1), projected forward using movement parameters for the appropriate habitat type × sex × year combination for that butterfly, scored as failed or censored and analysed as described previously. We compared the actual failure rates to the distribution of failure rates from 500 simulated data sets, for each habitat type × sex combination. Sample sizes did not permit separating data by years. As above, we concluded that groups differed significantly when the mean for one group did not overlap the 95% confidence interval of the other group and vice versa.


We observed 259 butterflies in 2007 and 347 in 2008 including 5708 flight moves that were used to characterize boundary responses and relationships between behaviour and habitat structure, lupine presence and sex. We observed butterfly flight from 27 April 2007–2 June 2007 and 14 May 2008–20 June 2008. Males and females were observed throughout the season with median observation dates of 17 May 2007 and 30 May 2008 for males and 19 May 2007 and 6 June 2008 for females. All three movement variables (move lengths, flight times and turning angles) depended on physical structure, hostplant resources or their interaction (Table 2, Fig. 2). In addition, movement parameters and landscape effects also differed between males and females, among individuals within sexes and between years (significant main effects and interactions of sex and year, and large random effects of individual, Table 2). In addition, butterflies had distinct preferences at patch edges that often differed between males and females (Table 1).

Table 2.   Analysis of habitat-specific movement parameters
FactorsLengthFlight timeTurning angle
  1. Non-significant interactions were removed from the final model (‘–’; note that 2-way interactions had to be retained if they were included in significant 3-way interactions).

  2. σID = standard deviation among flight paths; σresid = residual standard deviation.

  3. aCalculated from F-tests of the full model to a reduced model with that factor removed, analogous to type II tests in anova. Models were fitted using maximum likelihood, and F-statistics and P-values for these tests were obtained using anova() function, with each effect (in turn) set as the last factor in the model, relative to the appropriate reduced model.

Physical structure32253·20·072532250·00·991227546·70·0099
Resource × Structure322527542·60·1095
Resource × Sex322515·10·0001322533·3<0·000127545·10·0234
Resource × Year32255·20·022932255·20·023127540·10·8016
Structure × Sex32255·60·0177
Structure × Year27541·10·2846
Sex × Year5960·50·4982
Resource × Sex × Year32256·90·0085
Resource × Structure × Year27547·20·0073
 σIDσresid σIDσresid σIDσresid 
Random effects0·4850·914 0·2480·682 0·0900·399 
Figure 2.

 Examples of butterfly flight paths in response to resource and structural boundaries. As in Fig. 1, stippled area is lupine; light shaded area is open woods; white area is prairie. Female 8212 (solid line) and male 8231 (dotted line) were released inside lupine in the open woods (a). Female 8329 and male 8328 were released in prairie near the boundary with open woods (b).

Within-habitat movement

Butterflies strongly modified their movement in response to resources (Table 2, Fig. 3a). Move lengths inside lupine were consistently shorter than moves outside lupine, regardless of presence of trees. Differences in move lengths between lupine and non-lupine tended to be smaller in 2008 then 2007. Marginally, significant main effects of physical structure on move lengths were dominated by longer moves by males in prairie than open woods, and negligible responses by females to physical structure. In addition, males had longer move lengths than females inside lupine patches in prairies, outside lupine patches in prairies and inside lupine patches in open woods.

Figure 3.

 Within-habitat movement behaviour for 2007 and 2008. Move length (a), time in flight (b), turning angle (c) and diffusion rate (d). Error bars are 95% confidence intervals.

Butterfly flight times differed in response to resource patches but not physical structure (Table 2, Fig. 3b). Butterflies tended to land more frequently inside lupine than outside lupine patches, leading to shorter flight times. In prairies, butterflies tended to land more frequently when they were inside than outside lupine patches (though this trend was non-significant for males in 2008). In open woods, butterflies tended to land less frequently outside lupine, though males in 2008 showed the opposite pattern (resource × sex × year interactions, Table 2). In addition, male butterflies spent more time in flight than females in all habitat types except open woods outside lupine in 2008, and flight times were longer in 2007 than 2008.

Turning angles were larger inside than outside lupine patches, and larger in open woods than prairies (significant main effects of resource and physical structure, Table 2; though, this difference was not significant for all pairwise comparisons, Fig. 3c). Across all habitat types, female butterflies tended to have straighter flight paths (lower turning angles) than males (main effect of Sex, Table 2), and their turning angles were less responsive to lupine patches (Resource × Sex interaction, Table 2). In 2007, both males and females had high turning angles in open woods (lupine × structure × year interaction, Table 2).

Diffusion rates were higher outside than inside lupine patches (non-overlapping 95% bootstrapped confidence limits for most pairwise contrasts, Fig. 3d). Diffusion within lupine patches did not depend on forest structure (PL vs. WL, Fig. 3d). Outside lupine patches, diffusion rates in prairies were higher than in the open woods in 2007 but not in 2008.

Behaviour at boundaries

Consistent with our earlier work, butterflies biased movement towards lupine at patch edges (Table 1, Fig. 4). From outside lupine patches, female butterflies were strongly biased to return to lupine (PN→PL and WN→WL responses). Males showed a similar tendency; though, their bias was not statistically significant for either release habitat, and significantly lower than female bias when butterflies were released in the woods (WN→WL response, Fig. 4). Female bias was more than twice as strong when butterflies were released in open woods than in prairies (PN→PL vs. WN→WL response), but this distinction was not statistically significant. When inside lupine patches, females were more likely to stay in lupine than males (PL→PN and WL→WN responses).

Figure 4.

 Butterfly behaviour preference: Observed preference minus simulated preference from a habitat-specific correlated random walk. Females are open circles; males are stars. Mean and 95% confidence intervals from observed data relative to 500 simulated paths at each release point. See text for details.

Both dense and open woods repelled butterflies relative to expected movement from a correlated random walk (PL→D, PN→D, PL→WL and PN→WN responses). This difference was not statistically significant for all barrier types, and woods were never absolute barriers; even when they were statistically significant barriers, a small proportion of butterflies entered both dense and open woods. Females crossed all woodland boundaries significantly less than would be expected based from random movement, except for crossing from prairie to open woods when released in lupine (PL→WL). For males, the probability of entering the dense woods differed from correlated random walk only in non-lupine areas. From inside open woods, butterflies were attracted to the open prairies (WN→PN response). Females were less likely than males to cross into dense woods when inside lupine in the prairie (PL→D response) or to cross into open woods when outside lupine (PN→WN response).

The presence of resources affected responses to structural boundaries. The responses of male and female butterflies to woodland boundaries were more pronounced outside lupine patches than within lupine patches (significant differences for all paired contrasts except PL→WL vs. PN→WN for males, Fig. 4). At open woods boundaries that intersected lupine patches (PL→WL), observed behaviour did not differ from random. However, the probability of leaving lupine patches was similar in the prairies and inside the open woods (PL→PN and WL→WN). Together, these results suggest that butterflies were more focused on the resource patches and less influenced by woodland structure.


Both physical structure and resource patches influence behaviour of Fender’s blue butterflies. Response to both structural and resource boundaries is not surprising given the substantial literature on the influence of physical structure on boundary responses (e.g. Ries & Debinski 2001) and prior research demonstrating the attraction to food resources in this and other species (Schultz & Crone 2001; Schtickzelle & Baguette 2003; Haynes & Cronin 2006; Chapman, Dytham & Oxford 2007). However, no prior studies have investigated the combined influence of these factors in natural landscapes. Our results show that butterflies respond differently to structural vs. resource boundaries. For example, both open and dense woods are partial boundaries to movement by female butterflies, but the boundary response disappears when resources are found near the boundary with open woods. Thus, the combined influence of these factors cannot be inferred from responses to either factor in isolation. The implicit assumption in most other studies is that a priori designation of habitat based on physical structure or resources alone will adequately characterize movement (but see discussion of Turlure et al. 2011, below).

Although resources and structure both influence movement in this species, resources appear to affect movement more than structure (Table 2 higher F values, Fig. 3 more noticeable effect of lupine and Fig. 4 larger response to lupine boundaries). Prior studies across a wide range of taxa indicate that animals have reduced diffusion rates, slower movements and more sinuous paths in favourable habitats (e.g. Lima 1983; Senft et al. 1987; Odendaal, Turchin & Stermitz 1989; Fortin, Morales & Boyce 2005; Frair et al. 2005; Klaassen, Nolet & Bankert 2006; Schtickzelle et al. 2007; Phillips et al. 2009), consistent with optimal foraging theory (Zollner & Lima 1999) A smaller but growing number of studies have demonstrated biased movement towards favourable habitat after crossing a habitat boundary (Ries, Debinski & Wieland 2001; Schtickzelle & Baguette 2003, Crone & Schultz 2008; Delattre et al. 2010). Still, stronger responses to resource patch edges than structural barriers are surprising, because most studies have focused on movement in response to structural barriers (e.g. studies such as Ries & Debinski 2001 and Haddad 1999 vs. studies such as Schtickzelle et al. 2007). Haddad (1999) concluded that structural boundaries limited movement of some butterfly species but not others. He observed that sleepy orange (Eurema nicippe) and cloudless sulphur (Phoebis sennae), both of which are open habitat species, flew away from a woodland border, while spicebush swallowtail (Papilio troilus), a generalist, did not display a boundary response at the woodland edge. In Haddad’s study, butterfly movement parallels presence of resources. Sleepy orange and cloudless sulphur avoid woods, but there are no hostplants in the woods; spicebush swallowtail enters woods, and its resources are spread across both fields and woods. In our case, we observe Fender’s blue crossing structural boundaries both in the presence and absence of resources (prairie–open woods boundaries and prairie–dense woods boundaries), suggesting that crossing boundaries do not always parallel resource presence.

Increasing structural complexity reduces movement for both walking (e.g. Stevens et al. 2004; Schooley & Wiens 2005; Prevedello, Forero-Medina & Vieira 2010) and flying (e.g. Ricketts 2001; Ross, Matter & Roland 2005; Kuefler et al. 2010) organisms. Therefore, we were surprised that diffusion rates in lupine patches did not differ between prairies and open woods. Outside lupine patches, diffusion rates in open woods were slower than in open prairie in 2007, consistent with other studies, but quite similar in 2008. Because 2008 was an unusually cloudy spring (see Materials and Methods) and weather conditions in 2007 were more typical, it may be that Fender’s blue movement generally has movement slower in woods than open prairie, consistent with other species. One explanation for the differences between these years is that butterflies change their behaviour in response to inter-annual differences in thermal environments. In either case, our work demonstrates that structural complexity does not always reduce movement of open habitat species. Turlure et al. (2011) reached a similar conclusion, using experimental manipulations of structure at the resource patch boundary. They observed that the addition of structure had little effect on butterfly movement behaviour. In our study, physical structure is provided by tall and sometimes dense woods, which are challenging to experimentally manipulate; whereas Turlure et al. manipulate a structurally simpler system but are able to assess the response of multiple species while controlling for structural changes.

Males and females responded differently to resource and structural boundaries. Overall, females tended to be more responsive to resources (e.g. bias to return to lupine and tendency to stay in lupine, Fig. 4), and males tended to explore habitat more broadly with more limited response to both resources and structure in most cases (e.g. movement across open woods boundaries when outside lupine and movement into dense woods when in lupine; Fig. 4). This result is consistent with a large body of habitat selection research in behavioural ecology, in which males and females have different habitat preferences, often in relation to selecting sites for offspring (females) vs. finding mates (males; e.g., Mysterud & Ims 1998; Dover & Rowlingson 2005; Lawson Handley & Perrin 2007; Gillies & St Clair 2010; Nesti, Posillico & Lovari 2010; Tirpak et al. 2010). For example, Gillies & St Clair (2010) observe that forest inhabiting female rufous-naped wrens (Campylorhynchus rufinucha) had higher selection for forest habitat than the males. They suggest this difference reflects territorial behaviour of males and fitness consequences for males with respect to finding a mate. Nonetheless, numerous dispersal-related studies combine sexes (e.g. Baguette, Petit & Queva 2000; Ries & Debinski 2001; Nowicki et al. 2005; Stasek, Bean & Crist 2008; Kuefler et al. 2010) In other cases, studies are limited to either males (e.g. barred antshrikes Thamnophilus doliatus in Gillies & St Clair 2010 or females (e.g. female Glanville fritillary Melitaea cinxia) Ovaskainen et al. 2008a) Other studies in which landscape connectivity treats sexes separately have similarly concluded that differences in dispersal behaviour between males and females have important consequences for predicting levels of functional connectivity across the landscape (Lawson Handley & Perrin 2007; Ovaskainen et al. 2008a). In their review of sex-biased dispersal in mammals, Lawson Handley & Perrin (2007) noted that sex-biased dispersal is a pervasive feature of mammalian life history with greater dispersal by males than females. If males generally tend to broadly explore habitat, and females are responsive to resource patches, then pooling males and females, or only studying one sex, could lead to misleading conclusions about the factors that determine movement response to habitat.

Working in a natural landscape has limitations with respect to experimental design. Most noticeably, lupine patches crossed open woods boundaries. Therefore, although within-patch movement analyses (Table 2, Fig. 3) represent a full factorial design of structure × resource presence, boundary release analyses do not. For example, we could not examine the response of butterflies released in WN to PL edges, because such edges do not exist in the Cardwell Hills landscape. Thus, boundary responses should be interpreted qualitatively (e.g. WN–PN responses differ from WL–PL responses), rather than an ‘interaction’ in the statistical sense of additivity, which are difficult to interpret from unbalanced designs. One possible solution to the constraints of natural landscapes would be to create experimental landscapes. For example, Turlure et al. (2011) experimentally manipulated habitat within a flight arena, so they were able to construct habitat in which resource and structural boundaries co-occur. However, their design was limited to a much smaller range of habitat types, and such experiments would rarely be feasible at all for forested landscapes (with large, long-lived species) such as Cardwell Hills, as opposed to their scrub/bog interfaces. Still, both of our studies reached the same general conclusion that responses to hostplant densities seem to be stronger than effects of habitat structure.

Because we did not assess butterfly density, we cannot assess density-related influences on dispersal in this study. In prior studies of butterfly dispersal, both positive (Enfjall & Leimar 2005) and negative (Baguette, Clobert & Schtickzelle 2011) influences of density have been observed. Therefore, we do not have an a priori expectation about the potential influence of conspecific density on Fender’s blue dispersal behaviour. Broadly speaking, the Fender’s blue is largely considered a prairie species (USFWS 2010), which implies that densities are higher in open areas (PN and PL as opposed to WN and WL). In a very general sense, our results (specifically, lowest diffusion rates in PL patches) imply butterflies are not repelled by higher densities, at least not within the range of densities they experienced during our study period. Similarly, larger diffusion coefficients and lower resource affinities very generally suggest stronger dispersal in males than females; though, in this case, our past work suggests short-term movement may not scale as well to long-distance dispersal for males as for females (Schultz & Crone 2001, and T.L. Hicks, E.E. Crone and C.B. Schultz, unpubl. data).

Landscape composition influences animal dispersal (Bowler & Benton 2005; Fahrig 2007; Nathan 2008; Schick et al. 2008; Brouwers & Newton 2009; Dover & Settele 2009). Considerable attention over the past few decades has focused on movement through matrix and primary habitat and across habitat boundaries (Cadenasso et al. 2003; Ries et al. 2004; Prevedello & Vieira 2010). In a recent review, Prevedello & Vieira (2010) identified 104 studies that investigated the influence of multiple matrix types and primary habitat on animal movement. Although most studies reviewed detected some responses to matrix types, these responses were idiosyncratic, and patterns across species remain elusive. We conclude that this lack of generality stems at least in part from differences in definitions of habitat type, which are often poorly delineated and inconsistent between studies. Predicting animal movement through heterogeneous environments requires going beyond characterizing habitat types as simple binary variables such as matrix and habitat and instead requires delineating behaviourally important features of the landscape (Morales & Ellner 2002; Dennis 2010 and references therein).


First, we thank the private landowners in Cardwell Hills for permission to study Fender’s blue butterflies and instrumental support for the study, including Charlie and Rich Clark, PK and Dai Crisp, Karen Fleck-Harding, Lorin and Josh Lidell, William Pearcy and Amy Schoener. Without their help, cooperation and commitment to conservation, this research would not have been possible. In addition, numerous people helped in the field, with GIS mapping and with other aspects of this project, including: Dina Roberts, who ran the 2007 field season, Alexa Carleton, Michele Hansen, Alan Kirschbaum, Angela Little, Alex Martin, Alice Monteiro, Carolyn Menke, Price Sheppy and Kiya Wilson. This project was financially supported by US Fish and Wildlife Service and Washington State University Vancouver. We also thank members of the UKPopNet (NERC R8-H12-01 and English Nature) working group ‘Bayesian distribution models: dynamics, processes and projections’ for support and useful discussions. A special thanks to Barb Anderson for her help with Arcinfo AMLs. We also thank Nicolas Schtickzelle and anonymous reviewers for additional comments.