Local and landscape effects on butterfly density in northern Idaho grasslands and forests
- First Online:
- Cite this article as:
- Pocewicz, A., Morgan, P. & Eigenbrode, S.D. J Insect Conserv (2009) 13: 593. doi:10.1007/s10841-008-9209-7
- 164 Views
Understanding butterfly response to landscape context can inform conservation management and planning. We tested whether local-scale resources (host and nectar plants, canopy cover) or landscape context, measured at two scales, better explained the densities of four butterfly species. The density of Coenonympha tullia, which has host plants strongly associated with grassland habitats, was positively correlated with the amount of grassland in 0.5- and 1-km radius landscapes and only occurred in forests when they bordered grasslands. For the other species, Celastrina ladon, Cupido amyntula, and Vanessa cardui, local-scale resources better explained butterfly densities, emphasizing the importance of local habitat quality for butterflies. These three species also used host plants that were distributed more heterogeneously within and among habitat types. Our findings demonstrate the importance of host plant spatial distributions when determining the scale at which butterfly density relates to resources, and we recommend that both these distributions and landscape context be evaluated when developing butterfly monitoring programs, managing for species of concern, or modeling potential habitat.
KeywordsConservation planningDistance samplingEdge effectsHost plantLandscape context
Understanding how species use resources across landscapes can improve models of predicted habitat, which are often used for conservation planning (e.g., Scott et al. 1993; Groves et al. 2002), or to project effects of anticipated land use changes on wildlife (e.g., Schumaker et al. 2004; Gude et al. 2007). The inclusion of butterflies in such analyses and plans is valuable because of their importance as pollinators (Proctor et al. 1996; Ehrlich 2003) and prey for vertebrate species (Guppy and Shepard 2001), as well as for the attention they bring to invertebrate conservation (New et al. 1995). While habitat loss or degradation is known to directly affect occurrences of butterfly larval host plants and adult nectar sources (e.g., New et al. 1995; Schultz and Dlugosch 1999; Weiss 1999), the indirect effects of landscape context on butterflies remain less clear. One recent study found that butterfly abundance was influenced by landscape context, with the relative importance of local resources or landscape context differing among grassland butterfly species (Davis et al. 2007). In other cases landscape diversity or composition has been positively correlated with butterfly species richness but has not been related to butterfly abundance (Krauss et al. 2003; Öckinger and Smith 2006). Boundaries between land cover types have also affected how some butterfly species use the landscape (Roland et al. 2000; Ries and Debinski 2001; Dover and Settele 2008). Generally, local habitat quality has had greater effects on butterfly species composition and abundance than various measures of landscape context, including the proportion of urbanization (Collinge et al. 2003), isolation of habitat patches (Thomas et al. 2001), the amount of habitat fragmentation (Stoner and Joern 2004), the diversity of habitats (Weibull and Ostman 2003), and the amount of habitat (Bergman et al. 2008).
It can be challenging to identify the scale at which butterflies use resources through pooled analyses of species richness or overall abundance, as has been the focus of some studies (Collinge et al. 2003; Weibull and Ostman 2003; Bergman et al. 2008). Although many individual butterfly species have been studied extensively, they have typically been host plant specialists having a limited distribution of host plants (Dennis et al. 2003) and which are already threatened (e.g., Grundel et al. 1998; Schultz and Dlugosch 1999). In the absence of this amount of data for all species, linking the importance of local versus landscape resources to butterfly life history characteristics and the abundance and distribution of important resources may reveal general patterns that could be applied to the conservation of species with similar characteristics.
We evaluated the amount of variation in the densities of four butterfly species that was explained by local habitat resources versus landscape context within a landscape dominated by grassland and forest in northern Idaho. Local habitat resources measured were larval host plants, adult nectar plants, and tree canopy cover (microclimate). Landscape context was determined based on the amount of grassland or forest within nested circular landscapes of 0.5- and 1.0-km radii surrounding each local habitat sampled. Additionally, we evaluated how butterfly densities varied across cover type boundaries, where one cover type was the species’ preferred habitat. While the species studied here are not currently of conservation concern, they differed in grassland or forest habitat type preference, numbers of host plant species used and other characteristics that may inform the conservation of similar but declining species. This study addressed two questions, relating the findings to butterfly and habitat resource characteristics and broader conservation implications. First, do local habitat resources or landscape context explain more variation in butterfly density? Second, how do butterfly populations respond to adjacent land cover types that differ from their preferred habitat?
Sampling sites were chosen to include both mature forest and one of four open-canopy land cover types (four replicates each): recently harvested young forest, agricultural fields of wheat or Kentucky bluegrass, previously cultivated grasslands, and unplowed grasslands. The mature forest stands were dominated by ponderosa pine (Pinus ponderosa Dougl. ex P. & C. Laws) or Douglas-fir (Pseudotsuga menziesii var. glauca Beissn.). The young forests had been clear-cut or seed–tree harvested within the previous 2–10 years and had very low tree canopy cover. The previously cultivated grasslands were enrolled in the conservation reserve program and dominated by non-native pasture grasses. Native grasses and forbs dominated the unplowed grasslands, which included Palouse prairie and forest meadows.
There were only two locations in the study area where Palouse prairie adjoined forest and only two forest meadows, so all were included. All four sites with recently harvested forest had to be located on the University of Idaho Experimental Forest, because other forest landowners could not assure that sites would not be disturbed by timber harvesting during this study. The remaining eight sites were chosen by identifying boundaries without roads or streams between forest and agricultural fields or previously cultivated grasslands on recent digital imagery. Sites were located at least 1 km apart and ranged in elevation from 820 to 1,050 m.
Within each sampling site, the mature forest and open-canopy cover types shared a boundary. Line transects were established to cross this boundary, with 50 m of transect located in the open-canopy cover type and 100 m located in the mature forest. Transects were twice as long in forests, because it was expected that lower butterfly abundance would require more survey area to obtain a sufficient number of observations for density estimation. Three transect lines were randomly placed at each site, at least 40 m apart along the boundary and at least 50 m from any other edge of the forest, such as a road or another land cover type.
Butterfly surveys and density estimation
Each transect line was surveyed for butterflies five times per year from the middle of April to late July in 2004 and 2005, approximately every 3 weeks. Butterfly species numbers and abundance peak in late May to early June in these habitats. In 2004, sites were surveyed in a random order during the first visit and then this order was maintained as closely as possible during the next four visits. In 2005, sites were surveyed in the opposite order so that all sites were surveyed during both the earliest and latest parts of the sampling season. Surveys were completed between 09:30 and 17:30, under sunny conditions with air temperatures greater than 15.5°C, and when wind speeds were below 8 km per hour.
Characteristics of the four butterfly species studied
Host plant distributionb (%)
Number of host plant species
Larval host plant species occurring in sampling sitesc
Celastrina ladon Cramer (Lepidoptera: Lycaenidae)
Ceanothus sanguineus, Holodiscus discolor, Physocarpus malvaceus, Spiraea betufolia
Cupido amyntula Boisduval (Lepidoptera: Lycaenidae)
Western tailed blue
Astragalus canadensis, Lathyrus sp., Vicia americana
Coenonympha tullia Müller (Lepidoptera: Nymphalidae)
Agrostis alba, A. exarata, A. microphylla, A. stolonifera, Festuca idahoensis, F. occidentalis, F. scabrella, Melica subulata, Poa bulbosa, P. pratensis, P. sandbergii
Vanessa cardui L. (Lepidoptera: Nymphalidae)
Achillea millefolium, Anaphalis margaritacea, Cirsium arvense, C. vulgare, Helianthella uniflora, Trifolium douglasii, T. pretense, T. repens, Plantago lanceolata, Rumex acetosa, Fragaria vesca, F. virginiana, Prunus emarginata, P. virginiana
Butterfly densities were estimated by mature forest or open-canopy cover types for each sampling transect using the program Distance 5.0 (Buckland et al. 2004). Mature forest and open-canopy habitats were separated because of large differences in our ability to see butterflies in these two environments. Statistical models were used to estimate the probability of detecting butterflies based on their perpendicular distances from the sampling transects, which produced detection functions and density estimates. We were unable to obtain adequate model fits at the transect level for each of the individual species or these four species pooled. Therefore, we used all of the butterfly observations from our sampling sites to fit the models, pooled across the 10 sampling dates. The models were based on a total of 1,098 observations and 37 species.
The forest and open-canopy detection function models were stratified by butterfly size, because size, more than color or behavior, affected our ability to see butterflies in the field. Butterfly sizes were determined based on the midpoint of the wingspans reported in Opler et al. (2007). The size classes were small (1.7–3.3 cm wingspan midpoint), medium (3.4–5.0 cm), and large (5.1–8.5 cm). Celastrina ladon and Cupido amyntula were categorized as small (2.9 and 2.6 cm midpoints, respectively), Coenonympha tullia was categorized as medium (3.6 cm midpoint), and Vanessa cardui was categorized as large (6.2 cm midpoint).
We also considered two covariates that potentially affect the rate of detection: observer and average shrub height within a 1-m radius of each butterfly observation location. We evaluated a series of detection function models with varying truncation distances (furthest perpendicular distance of a butterfly from the sampling transect), covariates, key functions (half-normal, hazard rate), and series expansion terms (cosine, polynomial). We selected the models having the lowest Akaike’s information criterion (AIC) (Burham and Anderson 1998), quantile–quantile plots with good fits, and χ2 goodness-of-fit P-values > 0.05.
Measuring vegetation and landscape composition
We measured grass, forb, and shrub cover by species in four plot locations along each sampling transect. We used these data to calculate the average host and nectar plant percent cover for the mature forest and open-canopy portions of each transect. Grasses in agricultural fields were not counted as host plants for C. tullia. Sampling plots were centered on the butterfly transect line and extended 10 m in either direction, parallel with the forest/open-canopy boundary. Most butterfly observations also occurred within 10 m of this transect line (Fig. 2). Vegetation sampling occurred during a 6-week period in May and June 2004. Shrub percent cover was measured along the 20-m plot based on the length of foliage intercepting a measuring tape (Elzinga et al. 2001). Percent cover of forbs and grasses was measured within two quadrats (2 m × 2 m) within each 20 m × 2 m plot. Percent cover was measured as the vertical projection of the foliage from the ground, as viewed from above (Elzinga et al. 2001), using 13 standard cover classes (Caratti 2004) and was estimated by the same individual at all plots. Larval host plant species present in the study sites for each butterfly species are summarized in Table 1, along with summaries of their distributions.
We determined which plant species were potential nectar sources for butterflies based on field observations (Pocewicz 2006) and from a literature review focused on butterfly and plant genera occurring in our sampling sites (Opler and Krizek 1984; Schultz and Dlugosch 1999; Pyle 2002; Tooker et al. 2002; Tudor et al. 2004). We designated 136 of the 175 species of recorded flowering forbs and shrubs as potential butterfly nectar sources. Although butterfly size and proboscis length affect nectar plant choices (Opler and Krizek 1984; Corbet 2000), we did not have sufficient data to confidently assign different nectar plants to each butterfly species. Although percent vegetative cover is only coarsely related to availability of floral nectar resources, for our purposes we deemed this measure adequate because of the stark differences among our sites. Some of our sites were dominated by grasses providing no nectar while others had a diverse community of flowering forbs.
Tree canopy percent cover was recorded along the transect lines using a spherical densiometer at two open-canopy locations and five closed-canopy forest locations. Average percent cover was calculated for the mature forest and open-canopy portions of each transect. We also measured canopy cover at each location where a butterfly was recorded.
We digitized land use types from 1-m pixel resolution 2004 National Agriculture Imagery Program (NAIP) digital orthophotos using ArcGIS (Environmental Systems Research Institute, Redlands, CA, USA). Classes included young forest, mature forest, agriculture, previously cultivated grasslands, and unplowed grasslands. From this map we extracted circular landscapes of 0.5-km radius (78 ha area) and 1-km radius (314 ha area), centered on each sampling site (Fig. 1). These landscapes were spatially independent, with exception of two pairs of landscapes that overlapped slightly at the 1-km scale (Fig. 1).
To guide how we measured butterfly response to landscape context, we determined the primary habitat preference for each species and considered the area of that habitat within each landscape. First, we tested for differences in butterfly density between the two major habitat types in the study area, grassland and forest. Second, we tested whether butterfly species having higher densities in grasslands differed between previously plowed and unplowed grasslands and whether those having higher densities in forests differed between young and mature forests. We used Welch’s t-tests for samples with unequal variances in the open-source statistical language R (R Development Core Team 2007).
Fitting statistical models
To determine the relative importance of local habitat and landscape composition, we fit statistical models of density for each butterfly species that included the following local-scale variables: host plant percent cover, nectar plant percent cover, and tree canopy percent cover, as well as landscape composition, at either a 0.5- or 1-km radius. For those butterfly species for which both local and landscape-scale variables were statistically significant, we also fit separate models with local or landscape variables.
We used linear mixed-effects models, using the lme function of R (Pinheiro and Bates 2000). This approach allowed us to take advantage of data variation at the transect level, while a typical linear model would require averaging to the site level. The mixed-effects models adjusted for potential correlation among transects within each sampling site by including sampling site as a random effect. We used ANOVA with marginal or partial sums of squares to test the statistical significance of each predictor variable (α = 0.05). Square root transformations were applied to the butterfly density response variables to meet model assumptions. We used an R2 analog for mixed-effects models as a measure of model fit. This R2 value is equal to the variance of the response variable minus the variance of the residuals, divided by variance of the response variable.
We fit additional models to test for differences in butterfly density in mature forests adjacent to young forests, grasslands, or agricultural fields. For forest species, this tested whether adjacent open-canopy cover types influenced density within mature forests. For grassland species, this tested whether density in mature forests was influenced by adjacency to grasslands. The same model structure described above was used. The response variable was the square root of butterfly density in mature forest, and the predictor variables were host plant percent cover, nectar plant percent cover, and tree canopy percent cover, followed by adjacent cover type. We used sequential ANOVA to test whether there was any variation in butterfly density explained by adjacent cover type that was not explained by local habitat variables.
Butterfly habitat preferences
Local habitat versus landscape composition
P-values from marginal ANOVA tests based on models of butterfly density including local habitat and/or landscape composition predictor variables
Host plant % cover
Nectar plant % cover
Tree canopy % cover
Grassland (0.5 km)
Grassland (1 km)
Forest (0.5 km)
Forest (1 km)
Adjacent cover type effects
The adjacent open-canopy cover type did not explain significant variation in density of the other three species within mature forest. The statistical significance of local habitat variables was the same within forests as overall for C. ladon (Table 2). For C. amyntula there was a negative correlation with tree canopy cover (P = <0.05) and no significant correlation with nectar plant cover, as in the analysis including all sampled habitats. There was also no significant correlation between V. cardui density and host plant percent cover within mature forests.
Landscape composition and adjacent cover type effects
The spatial scale at which butterfly density related to habitat resources differed by species. Local habitat variables explained more variation in butterfly density than did landscape composition for all but the grassland species C. tullia, whether landscape effects were assessed based on the cover types bordering the sampled preferred habitat or the proportion of preferred habitat within a larger surrounding area surrounding the sampled habitat. C. tullia may have been sensitive to landscape context because of the predominance of its host plants (all within Poaceae) in the grassland cover type. The response of C. tullia to landscape context, whether measured as immediately adjacent cover type or broader landscape composition, may reflect responses to its host plants or to characteristics of habitats or habitat edges associated with these host plants (Ries and Sisk 2008), or both.
C. tullia only occurred in mature forests when those forests bordered its preferred grassland habitats. It appears that the resources in the forests were at best supplementary to those available in the preferred grassland habitat for this species. In such situations Ries and Sisk (2008) predict a transitional response to the habitat edge, which would cause C. tullia densities to decrease with distance into the forest from the forest-grassland edge. We did not find this pattern within the 100-m distance from the forest edge that we sampled. Within these forests, C. tullia density was positively correlated with host plant cover and negatively correlated with tree canopy cover. We had expected that abundance of species preferring forest habitats would be lower in sites adjacent to non-preferred habitat, particularly agricultural fields lacking host and nectar plant resources. However, both C. ladon and C. amyntula were equally likely to occur in mature forests adjacent to agricultural fields, young forests, and grasslands.
Although C. ladon and C. amyntula both showed a significant preference for forest habitat, they used few host plant species, and these plant species were not consistently associated with the forest habitat type class. Rather, their densities were best explained by finer-scale variation in abundance of host and nectar plants among sites. In contrast to the other three species, V. cardui, although a host plant generalist, is exceptionally mobile, migrating from the southern US or northern Mexico to this area in northern Idaho (Guppy and Shepard 2001; Opler et al. 2007). Its extreme vagility may contribute to its lack of detectable habitat preference or response to landscape composition in our study.
The relative abundance of forest or grassland habitats within the landscapes studied may have also contributed to species-specific sensitivities to landscape composition. The generally lower availability of grassland in the study area as a whole may account for the sensitivity of C. tullia to these resources. In contrast, the generally greater availability of forest habitat within our studied landscapes may account for the similar abundance of the forest species among these landscapes. The relatively small size of our circular landscapes may also have influenced our results, especially for the more vagile V. cardui. We were unable to consider landscapes larger than 1 km in diameter because some our sampling sites were close to one another.
While we focused on common drivers of butterfly occurrence to describe local habitat, including host and nectar plants and tree canopy cover, other factors may also contribute to variation in butterfly density. These include the availability of litter, water, sources of salts and minerals, and roosting, mate-location, and basking sites (Guppy and Shepard 2001; Dennis et al. 2006; Davis et al. 2007). Topography and proximity to streams also influence butterfly occurrence (Kuefler and Haddad 2006). In addition, our estimates of local availability of host plants and nectar sources may have lacked precision because they were based on sampling rather than a census. As mentioned previously, we estimated nectar sources based on percent cover of nectar plants but did not directly measure flower abundance or nectar quality (Holl 1995; Schultz and Dlugosch 1999).
Our findings suggest that host plant availability strongly influences butterfly abundance in complex landscapes, resulting in species-specific patterns of response to local and landscape-scale variation in land cover. The relatively greater importance of local habitat variables as compared with landscape composition for three of four species studied reemphasizes the importance of local habitat quality for butterflies (e.g., Schultz and Dlugosch 1999; Thomas et al. 2001). Yet our findings also showed that landscape context does influence the abundance of some butterfly species, consistent with previous research (Davis et al. 2007). The densities of species such as C. tullia, with host plants that are strongly associated with particular cover types, can respond to landscape context at a larger spatial scale than species such as C. amyntula and C. ladon, with host plants that are distributed heterogeneously within and among cover types. We conclude that the scale of host plant spatial distributions should be considered when assessing potential effects of landscape context on butterflies. Based on our findings, we also recommend that the influence of landscape context be evaluated when developing butterfly monitoring programs, managing for species of concern, or modeling potential habitat.
We thank Kelsey Sherich, Jennifer Szarkowski, and Sean Irby for their help with data collection, Jon Shepard for assistance with butterfly identification, Pam Brunsfeld, Matt Parks, and Jim Kingery for plant identification assistance, and Andrew Robinson and Kirk Steinhorst for statistical advice. This research would not have been possible without permission to access to lands owned by ten private landowners and the University of Idaho Experimental Forest. Financial support was provided by the USDA McIntire-Stennis Program, the National Science Foundation IGERT grant 0014304, and the University of Idaho.