Abstract
Monarch butterflies (Danaus plexippus) in eastern North America must make frequent stops to rest and refuel during their annual migration. During these stopovers, monarchs form communal roosts, which are often observed by laypersons. Journey North is a citizen science program that compiles roost observations, and we examined these data in an attempt to identify habitat characteristics of roosts. From each observation we extracted information on the type of vegetation used, and we used GIS and a national landcover data set to determine land cover characteristics within a 10 km radius of the roost. Ninety-seven percent of roosts were reported on trees; most were in pines and conifers, maples, oaks, pecans and willows. Conifers and maples were used most often in northern flyway regions, while pecans and oaks were more-frequently used in southern regions. No one landcover type was directly associated with roost sites, although there was more open water near roost sites than around random sites. Roosts in southern Texas were associated primarily with grasslands, but this was not the case elsewhere. Considering the large variety of tree types used and the diversity of landcover types around roost sites, monarchs appear highly-adaptable in terms of roost site selection.
1. Introduction
Research on one of the world’s most famous insects, the monarch butterfly (Danaus plexippus, Figure 1), has benefitted greatly from numerous citizen science programs in North America devoted to tracking this species at various life stages. The attention given to this insect no doubt stems from its large size, easily identifiable orange and black colors (Figure 1), and its well-known and spectacular migrations, which are unique among butterflies. All of these factors make this butterfly extremely charismatic, and this helps to promote public participation in various citizen science programs. For example, the larval stages of this insect are monitored each summer by volunteers of the Monarch Larval Monitoring Project (http://www.mlmp.org/), and these data have been used to document geographic and temporal variation in population recruitment [1, 2]. In the western North American population, volunteers count numbers of adult monarchs that overwinter in clusters along the California coast (Western Monarch Thanksgiving Count), and a recent analysis of these data showed the importance of climatic conditions at the natal sites for predicting overwintering numbers [3]. There is another citizen science program whereby volunteers submit samples of a monarch-specific protozoan parasite (MonarchHealth; http://www.monarchparasites.org/), which has led to the identification of trends in disease prevalence during the summer and fall [4]. Finally, numerous scientific investigations have made use of data from a citizen science program called Journey North (http://www.learner.org/jnorth/), which asks volunteers in North America to report sightings of adult monarchs during the winter [5], during the spring migration [6–8], and during the fall migration when monarchs from the eastern population are travelling to their Mexican overwintering site [9]. The primary fall sightings are of nocturnal roosts, which monarchs form during their southward migration [10], and that are easily recognized by laypersons, since they often consist of hundreds or thousands of monarchs (Figure 2).
(a)
(b)
(c)
(d)
Monarch roosts can be considered stopover sites, which are essentially places where migratory animals pause during their journey to rest and/or refuel. Like most migratory organisms, monarchs utilize stopover sites to feed and deposit fat reserves [11] and to rest at night. Moreover for monarchs, depositing fat reserves during the migration not only provides fuel for the flight, but is essential to their overwintering survival [12]. As such, determining where stopover sites are for monarchs is an important issue in conserving their migration [9]. Further, while there is a wealth of research into stopover ecology of migrating birds (e.g., [13–18]), there are comparatively few studies examining the nature of stopover behavior in monarchs [19–21]. Moreover, there are no published studies where monarch stopover habitat is documented, other than anecdotal observations of roost trees [10]. In fact, it is not known even if monarchs select specific large- or small-scale habitat features at all when they stop or if roost site selection is completely random. Prior examination of roost observations indicated that few locations are utilized by monarchs for roosting year after year [9], which argues for the latter scenario, although more thorough investigation on this idea is warranted. Furthermore, like most migratory animals, monarchs must face continually changing landscapes throughout the entire flyway, including prairies and farmland in the American Midwest, deciduous forests in the eastern seaboard, and dry scrublands in Texas and northern Mexico. Given these changing landscapes they encounter, how then would their stopover habitat preferences (if there are any) change as they progress southward?
The Journey North roost observation database is uniquely positioned to offer insights into this question. When volunteers observe a migratory roost, they not only report the location and date, but are also encouraged to submit general observations on the roost, such as the type of tree or vegetation in which the roost was observed. In this study, we screened four years of Journey North’s migratory roost sightings (from eastern North America only) and, from these records, we recorded the type of tree (or other vegetation) in which the roost was observed. We also examined the landscape-level features of the roost site using a GIS approach; here we compared the land use surrounding each roost location to those of randomly selected locations at similar stages of the migration, which we arbitrarily divided into five flyway regions. Our goals for this study were to (1) document the large- and small-scale habitat preferences of monarchs at roosting sites and (2) determine if monarchs display a uniform preference for specific stopover habitats throughout the migration flyway or does their preference change as the migration progresses. Results from this study will not only further scientific understanding of monarch butterfly migration, but should also be relevant to the science of animal migration in general. In fact, to our knowledge this study is the first to examine how stopover habitat preferences of a single migratory animal vary throughout an entire migration flyway.
2. Methods
2.1. Roost Observations
We examined roost observations from the Journey North program between 2005 and 2008 (Figure 3), which are accessible online in the archives section of the program (http://www.learner.org/jnorth/maps/archives.html). For the purposes of this study, we focused on the primary flyway only (the central flyway) and did not consider observations from the Atlantic flyway [9], since very few tagged monarchs from that region are ever recovered in Mexico [20, 22, 23]. Each roost observation in the database is associated with a date (of the first night of observation), and latitude and longitude (of the zip code of the observer’s mailing address, see below). While all roost observations have at least these components, observers are also encouraged to record notes about the roost and even take pictures, which are also archived with the sightings. For this study we screened these written notes and recorded what the monarchs were reported roosting on (i.e., tree, shrub, etc.). Moreover, since the aim of this study was to compare roost characteristics along the migratory flyway, we arbitrarily created 5 “flyway regions” of 4° latitude blocks that encompassed the majority of the flyway and roost observations in Canada and the United States (Figure 3). We then categorized the roost observation data (type of vegetation, etc.) into these regions based on the latitude of the observation.
2.2. Landscape Features of Roost Sites
The latitude and longitude associated with roost observations were imported into ArcGIS for analyses of land use surrounding roost sites. We point out that the coordinates of roosts in the Journey North database are not necessarily for the roost tree itself; when new participants sign up, they are asked to report their home address, and from this information, coordinates are generated by Journey North personnel using a database of coordinates for North American zip (postal) codes. This practice was started for ease of overlaying points on an online, continent-scale map and since most observers do not know their latitude and longitude. For our purposes this means that the coordinates for any given observation could be centered on a point several kilometers distant from the roost (the center of the zip code). However, we attempted to minimize this problem by creating a buffer around each point with a 10 km radius (314 km2), and evaluating the land cover within this area, which should be large enough to encompass the roost itself. The average area of zip codes in the United States is 222.7 km2 [24], and for urban areas that have multiple zip codes in the same city this number is likely to be much smaller, which only improves the chance that the buffer encompasses the roost. To minimize spatial autocorrelation, we eliminated all duplicate coordinates of roosts that were reported in the same city (which happens when two separate observers reported the same roost, from a roost being spotted in multiple years, or from two roosts sighted near one another). This left 310 spatially independent roost observations for analysis. In addition, we randomly selected a series of points () throughout each flyway region for comparison to the monarch-selected locations. For this we generated a minimum convex polygon around the entire flyway and, within that area, randomly generated points within ArcGIS, preventing points from occurring within 20 km of one another.
To evaluate land cover characteristics of both the monarch-selected and random locations we overlaid a national land cover map [25] where land cover had been digitally classified into 19 categories (though for the purposes of this project we only considered 7 of the largest categories—deciduous forest, coniferous forest, cropland, grassland, urban, open water, and wetland). Then, we calculated the percent cover of each category within the buffered area surrounding each point.
2.3. Data Analyses
There were 217 observations where the type of vegetation was specified. Using these data we compared the frequency of the most commonly reported tree species across flyway regions using chi-square statistics. Using the large-scale land cover data set containing both monarch-selected sites () and randomly selected locations (), we used logistic regression to simultaneously examine the effects of each land use category and flyway region (predictor variables) on whether a location was monarch-selected or random (response variable). We also included two-way interaction terms between each land use category and flyway region to determine if habitat preferences vary throughout the flyway. The full model with all main and interaction effects was simplified using likelihood ratio tests (Δ deviance) to evaluate the importance of nonsignificant terms following Crawley [26]. Nested models without interaction terms were compared against the full model prior to the removal of any main effects. Significance of terms remaining in the final model are reported based on Wald . All analyses were conducted using Statistica 6.1 software [27].
3. Results
3.1. Small-Scale Roost Habitat Characteristics
Of all roost observations where the type of vegetation was specified (), 97.7% of the roosts were reported on trees, with the remainder being on herbaceous vegetation, including two observations of monarchs roosting on seaside goldenrod (Solidago sempervirens), and one each of common groundsel (Senecio vulgaris), beach grass (Ammophila sp.), and golden crownbeard (Verbesina encelioides). There was no clear preference for one tree type; there were a total of 38 tree species reported overall (as hosting roosts) in the four years examined. The 10 most common tree species reported are listed in Table 1, broken down by flyway region. The most frequently reported trees included pines or other conifers (21.8%), maple species (20.7%), followed by oaks (15.6%), pecans (14.5%), and willows (7.8%). Collectively, these made up 80.4% of the observations (where the tree type was specified). The frequency of these 5 tree types (i.e., their use as roost sites) appeared to vary across the flyway regions (Table 1); a 5 × 5 contingency table analysis based on these top five rows revealed that these frequencies differed significantly (df = 16, , ). In general, pines/conifers and maples were used most often in the northern areas of the flyway, while pecans and oaks were more frequently used in the southern regions.
3.2. Landscape Characteristics of Roost Sites
Of the 7 land use categories evaluated in both monarch-selected locations and random ones, the majority (~50%) of the landscapes across most flyway regions were composed of crops (Figure 4), which makes sense given that much of the central flyway traverses the agricultural region of the American Midwest (Figure 3). Following that were the broadleaf (deciduous) forest and grasslands categories. Visual comparison of the breakdown of all land use categories at monarch-selected sites (Figure 4(a)) versus random sites in the same region (Figure 4(b)) gives the impression that land use at random sites is fairly uniform throughout the flyway while that of actual roost sites varies to some degree. In particular, there appeared to be a distinct shift in the relative proportions of land use in the two southernmost regions (northern and southern Texas). In region 4, most of the land around selected roost sites was cropland (67%), while in the last region, 61% of the land around roost sites was grassland, compared to 15% around random locations in that region.
(a)
(b)
In the logistic regression model examining large-scale land use at monarch-selected sites versus random ones the results were complex. The probability of a monarch roost appeared to depend on the amount of deciduous forest, urban area, open water area, and wetland area around the site (all significant main effects; Table 2). In direct comparison of land use between roost sites and random sites, it appears that monarch-selected sites had less overall deciduous forest cover than random sites, more urban area, a higher percentage of open water nearby, and less wetland cover than random locations (Figure 5). Further, there were significant interaction effects (i.e., meaning that the strength of the main effect depended on the flyway region) in the percent deciduous forest, grassland, and urban area (Table 2).
(a)
(b)
(c)
(d)
(e)
(f)
(g)
4. Discussion
Places where monarch butterflies stop during their migration represent important links between breeding and overwintering areas, and identifying habitat requirements of roosting monarchs is therefore a key component to our understanding of this phenomenon. The ephemeral nature of migratory roosts [9], plus their broad geographic scope, makes them difficult to study using conventional scientific methodology. However, by using observations made by this nationwide network of citizen scientists, we hope to have made the first steps in addressing this question. For example, while monarch roosts were nearly exclusively on trees, we found no overwhelming preference for a tree species or type (i.e., conifer versus deciduous), other than a general tendency for maples and conifers in the north and pecans and oaks in the south (Table 1). A tendency to use males was also casually noted by F. A. and N. R. Urquhart [32] who were located in the northernmost region. When one considers the entire flyway however, given the diverse branch and leaf morphology of the various trees reported as used, it appears that monarchs are highly adaptable in terms of their roost tree use. This is also evidenced by the pictures submitted by Journey North participants (Figure 2); one can see that monarchs are capable of settling on a wide variety of branch structures (from needles to small-leafed trees to large-leafed trees).
Similar to the small-scale patterns obtained, from a large-scale habitat perspective, there was no one land cover type that best predicted the location of monarch roosts throughout the flyway; there was statistically significant difference in the proportion of multiple land cover types between monarch-selected and random sites (Table 2), and there were multiple interactions with flyway region. These complex results make interpreting these data difficult. One clear pattern in the land cover analyses was that, in nearly all flyway regions, there was a greater proportion of open water in the (314 km2) roost area than around random locations (Figures 4 and 5). This land cover category would include large rivers, ponds, and lakes. It may be that monarchs use these land features as beacons while searching from the air for potential roost sites, as these areas would tend to be lush with vegetation and possibly support a variety of nectaring plants. Conversely, monarchs roosted in areas that had significantly less wetland land cover than random sites did (Figure 5). Here we can only speculate as to the reason for this dichotomy; it may be that such areas are not as visible from the air as are open water bodies.
There were certain land cover patterns uncovered that may have resulted from inherent biases in observer distribution; most Journey North participants tend to live in urban areas (E. Howard, unpublished data). For example, monarch roosts had a higher proportion of urban land use around them than did random sites (Figure 5). This was probably an artifact of the tendency for most observers to live in or near cities (i.e., fewer observers in rural areas means fewer roost sightings). Similarly, in nearly all flyway regions, roost sites tended to have less deciduous forest area around them than did random sites (Figure 5), which could indicate either a general avoidance of these land types for roosting purposes, or (more likely) that roosts in these habitats are not often spotted by laypeople. A roost of several hundred monarchs in a forest would not stand out as readily as one in a lone tree in the middle of a cornfield.
In addition to the significant main effects of land cover types, the logistic regression model revealed several significant interaction effects with flyway region and certain land cover types (Table 2, Figure 5), which indicates the effect of the land cover in question varied depending on the flyway region. In other words, there was some degree of change to the land cover preference (or avoidance) throughout the flyway. For example, the tendency for roosts to be associated with greater urban area was stronger in the northern regions than the southern (Figure 5), and the avoidance of deciduous forest was most pronounced in the southern regions. Further, there was a statistical effect of grassland, but it depended on the flyway region; in the southernmost region in particular, monarch roosts were in areas with 61% grassland, but this habitat was not widespread throughout that region (random sites had 15% grassland; Figure 4). In this region (southern Texas), monarchs appear to be drawn to this type of large-scale habitat.
The collective results from both the small- and large-scale analyses in this study should have conservation implications, but not in the manner we anticipated. Conserving migratory habitats is an important issue for all migrant species [28]. With this issue in mind, we had attempted to ascertain if there were certain types of small- or large-scale habitat features that could be identified as being important to monarchs for at least one part of the stopover, their overnight roosting. However, the data we gathered in this effort did not point to a select few landscape features or roost tree types, but instead indicate that monarchs are capable of using a wide variety of habitats for roosting. Knowing this, it then becomes challenging, from a management perspective, to pinpoint what could be done to conserve important stopover areas. It may be that conservation efforts need to be targeted at the areas where it is most clear what (primary) habitats monarchs are using, such as in southern Texas, and less so in areas where there are less “habitat” preferences that can be identified, such as in the northern flyway regions. Texas is also an area of high importance for monarch migration, since, here, monarchs deposit considerable amounts of fat [29], which they will use to sustain themselves over the winter [12].
Throughout this paper we have stressed that we would not have been able to study this phenomenon of migratory roosting by any other means than by using Journey North’s nationwide citizen science network. We recognize, however, that this data set was not ideal for answering our original questions regarding habitat characteristics of roosts and that there are areas where this program could be strengthened. Moreover, by highlighting these limitations (below), managers of other citizen science projects may be able to learn from these problematic issues, which may be common to many programs. Perhaps the largest drawback of the Journey North program is the fact that the coordinates of all sightings, including those of “roosts,” are of the geographic center of the zip code from the observer’s address, which could be several kilometers away from the actual roost. There is no remedy for this problem, unless observers take GPS readings of roost trees, which would certainly be difficult to implement into the Journey North protocol. It would also have been helpful if the protocol for reporting roosts included providing information about the surrounding habitat, such as how many trees are nearby (and not being used by monarchs) and what species they are. This would have allowed for more direct evaluation of roost tree “preference” at the sites where monarchs stop over (i.e., by comparing trees that were used to those that were not). Furthermore, in addition to habitat data, one area where Journey North could strengthen its protocol is in the reporting of the size of the roosts (i.e., number of monarchs). Many people state in their notes that they saw “hundreds” or (very often) “thousands” of monarchs in the roost observed. Estimating numbers of clustering monarchs is notoriously difficult, even for trained scientists (e.g., [30]), so this would also be difficult to implement in the current protocol. However, if actual numbers were associated with roost observations (and if we were confident in their accuracy), it would theoretically allow for annual estimates of the size of the entire migratory generation. Estimating long-term trends in abundance for this population is something that has been attempted with other data sets, but with inconsistent results [31, 32]. With the vast number of observers in the Journey North program, such data would undoubtedly be of value in this regard.
Finally, this study may well represent the first-ever examination of habitat requirements of a single migratory organism across an entire migration flyway. Such an approach allows us to identify any changes in habitat requirements at different stages of the migration. And indeed, although the monarchs’ “habitat preferences” during migration appear to be broad, we did see certain changes in large- and small-scale habitat preferences as the migration advanced southward. The next step may be to assess the availability of nectaring sources at all stages of the migration, especially in the latter stages of the migration where monarchs are accumulating the most fat [29]. Additional questions regarding roosting or stopover behavior could also be addressed in the future, using citizen science observational data or direct study at specific stopover sites [20, 21]. Thanks to the efforts of hundreds of dedicated and observant people in North America who participate in citizen science programs, the answers to these and other questions are now within reach.
Acknowledgments
This project could not have been completed without the contributions of the thousands of Journey North participants who watch the skies each fall and faithfully submit roost observations. Lincoln Brower has provided expert advice to the Journey North program over the years. Funding for Journey North was provided by the Annenberg Foundation. The authors thank the members of the MonarchNet working group (Karen Oberhauser, Sonia Altizer, Leslie Ries, Dennis Frey, Becky Bartel, Elise Zipkin, James Battin, and Rebecca Batalden) for helpful discussion about the project, as well as two anonymous reviewers for suggestions for improvement on the paper.