Estimating the annual distribution of monarch butterflies in Canada over 16 years using citizen science data

Monarch butterflies (Danaus plexippus, Linnaeus, 1758) in North America are known for their long-distance movements among the United States, Canada, and Mexico. Monarchs are comprised of one genetic population with two migratory flyways separated by the Rocky Mountains (Brower 1995; Lyons et al. 2012). The eastern North American population overwinters in the highlands of central Mexico and migrates to breeding areas in the midwestern and eastern United States and Canada over successive breeding generations before a final generation migrates back to Mexico (Brower 1995; Flockhart et al. 2013). The western population overwinters in coastal areas of California and some individuals migrate north over multiple breeding generations to all western states and occasionally reaching very southern portions of Canada in British Columbia before a final generation migrates back to coastal California (Brower 1995; Dingle et al. 2005; James et al. 2018). Both migratory populations have declined over the last several decades because of multiple threats (Brower et al. 2012; Schultz et al. 2017), and monitoring their distribution and abundance is a necessary aspect of their ongoing management (Commission for Environmental Cooperation 2008).

Their small population size and large fluctuations in population growth rate (Brower et al. 2012; Flockhart et al. 2015; Semmens et al. 2016; Schultz et al. 2017) were primary reasons why monarchs were designated a species of ‘special concern’ in Canada in 1997, a status that was reconfirmed in 2010 (COSEWIC 2010). Monarch population trajectories currently meet the threshold to be considered endangered in Canada and the Committee on the Status of Endangered Wildlife in Canada has recommended this listing to the federal government for a classification decision due in 2019 (COSEWIC 2016). Both North American populations are at risk from multiple factors such as disease, extreme weather, and overwinter habitat loss (Brower et al. 2012). There is increasing evidence to support that population viability is three times more sensitive to breeding habitat loss compared with habitat loss on the wintering grounds (Flockhart et al. 2015) and, therefore, the monarch population decline is driven, at least in part, by decreasing abundance of their obligate host plants, milkweeds (Asclepias spp.) across the United States and Canada (Pleasants and Oberhauser 2013; Flockhart et al. 2015; Stenoien et al. 2016; Oberhauser et al. 2017; Thogmartin et al. 2017a). Prioritizing locations based on the greatest probability of occurrence for breeding habitat restoration in Canada, therefore, relies on spatially explicit mapping the distribution of monarchs across space and time that extends beyond simple generalized maps of distribution.

A variety of long-term citizen science monitoring programs across North America provides data to map the distribution and phenology of many organisms. Monarchs are the focus of multiple citizen science programs that document locations, migration patterns, reproduction, and disease rates (e.g., Journey North, Monarch Alert, Monarch Watch; for a more complete list see, Ries and Oberhauser 2015). Citizen science programs that document monarch occurrence data may record first monarch observations of the season (e.g., Journey North; Davis and Howard 2005) or a compilation of observations throughout the spring and summer for monarchs as well as other North American butterflies (e.g., eButterfly; Prudic et al. 2017). All programs record the date and location where monarchs are observed, giving rise to so-called presence-only data, and have been used to document migration routes (Howard and Davis 2009), migration speed (Davis and Howard 2005), and breeding distributions (Batalden et al. 2007; Flockhart et al. 2013). Presence-only citizen science data from different programs, if not biased to only include specific locations or portions of the annual cycle, can be combined to increase sample sizes to document annual variation in distribution patterns inherent in insects. Understanding annual variation in distribution would provide valuable information to aid in management and conservation planning (Warren et al. 2001).

The annual breeding distribution of monarchs in Canada depends on the colonization of individuals that migrate from the United States (Brower 1995) and the conditions (habitat, physiological, and geographic) under which such migrations occur (Brown et al. 1996). The eastern North American monarch butterfly population often reaches into southern parts of Canada as far west as Alberta (Brower 1995; Flockhart et al. 2013; Flockhart et al. 2019). In some extreme cases, individuals observed early in the breeding season have likely migrated directly from their overwintering grounds in Mexico (Miller et al. 2012). In contrast, monarchs that reach the extreme southern portions of British Columbia are likely from the western North American population, which overwinters in California (Environment and Climate Change Canada 2016; Yang et al. 2016; James et al. 2018). Introgression among the eastern and western populations (Lyons et al. 2012) implies that climatic, physiologic, and geographic factors may consistently influence migration and hence colonization of the breeding distribution in Canada. Monarch breeding distribution may correlate with open habitats typical of grassland ecoregions (Oberhauser et al. 2001), geographic limits of migration and recruitment (McKinnon et al. 2010), or climatic thresholds for flight in insects. Additionally, these factors may not be mutually exclusive if, for instance, host plant availability may be limited by climate (Lemoine 2015). Because Canada legislates species at risk, including monarchs, independently of the US or Mexico, effective conservation planning requires understanding the extent of the breeding range in Canada, the natural fluctuation from year to year (Brown et al. 1996; Prysby and Oberhauser 2004), and what might be the primary drivers of annual variability.

In this study, we use the citizen science programs eButterfly (Prudic et al. 2017) and Journey North (Davis and Howard 2005) to estimate the annual breeding distribution of monarch butterflies in Canada between 2000 and 2015. In doing so, we test predictions from several hypotheses to explain variation in monarch occurrence (Table 1). Our results provide the geographic context for decision-makers to identifying priority areas to engage in restoration and other conservation-related activities.

Across 16 years (2000–2015), the number of observations of monarchs in Canada per year ranged from 187 in 2000 to 2827 in 2012 (Table 2). Monarch breeding distribution in Canada varied widely among years (Fig. 1). The breeding distribution in western Canada (Fig. 1a) was an order of magnitude less than in eastern Canada (Fig. 1b). When considered at the 0.05 probability of occurrence level, the annual mean breeding distribution in Canada was 484 943 km² (standard deviation (SD) = 329 105 km²). After removing the two years with below-average distribution estimates (2002: 173 449 km²; 2004: 144 542 km²; Fig. 2a) and the two years with above-average estimates (2007: 1 082 494 km²; 2012: 1 425 835 km²; Fig. 2c), the mean breeding distribution in Canada was 411 064 km² (SD = 97 188 km²; e.g., 2010, Fig. 2b). Overall, the core areas of the breeding distribution in Canada occurred in southern Ontario, southern Quebec, and to a lesser extent southern Manitoba (Fig. S1). Across all years, monarchs were predicted to occur with a high probability of occurrence (>0.5) in Ontario and only occurred rarely at this probability in Quebec, New Brunswick, Nova Scotia, and Manitoba (Fig. 3a). When considering probability of occurrences >0.25 (Fig. 3b) the southern portions of all provinces were represented in the predicted breeding distribution in at least one year, whereas when considering probability of occurrences >0.05 the southern portions of all provinces were represented in the breeding distribution of multiple years (Fig. 3c).

Overall, all variables considered were supported in the top model in at least one year (Table 3) and model selection uncertainty was low in all years except in 2000, 2012, and 2015 (Table S1). Monarch distribution was best explained by geographic variables followed by habitat variables (Table 3). Latitude (16 of 16 years) and longitude (15 of 16 years) were the most common variables found in top models, whereas the least common variables supported included linear (3 of 16 years) and quadratic (1 of 16 years) effects of mean temperature and a quadratic effect of precipitation (2 of 16 years). Parameter estimates for human population density were positive in all models, which support the notion that human presence needs to be controlled for when using citizen science data to predict monarch distributions (Table 3; Fig. 3d).

There was no correspondence between the preceding or forthcoming overwintering population estimates in California and Mexico and the estimated breeding distribution in western (Fig. S2a) and eastern Canada (Fig. S2b). However, there was a relationship between a larger estimated breeding distribution and the number of observations of monarchs in Canada (Fig. S3).

The core breeding distribution of monarchs in Canada includes portions of southern Ontario (south of Sudbury and Ottawa to the USA border), Quebec (south of Québec City including Montreal and Sherbrooke), and Manitoba (south of Winnipeg). In some years, monarchs were predicted to range over more than a million square kilometers that included southern portions of all provinces. Monarchs in eastern Canada occupied an area that was an order of magnitude larger than the area occupied by monarchs in western Canada. Patterns of occurrence in Canada were best described by geographic limits that are presumably related to limits on northward migration in the late summer or limits in the availability of milkweeds (Asclepias spp.), the obligate host plants (Lemoine 2015). Given the inter-annual variation in breeding distribution, all portions of southern Canada should be considered when developing long-term management strategies for this species; however, the population in eastern Canada occurs each year over a much larger area.

The distribution of monarchs was best explained by geographic limits of migration described with relationships between geographic variables of latitude, longitude, elevation, and slope. These are fixed attributes compared with regional weather and vegetation patterns that vary annually or seasonally and are known predictors of monarch abundance and distribution (Zipkin et al. 2012; Flockhart et al. 2013; Flockhart et al. 2017; Saunders et al. 2018). Therefore, variation in migration timing and geographic distribution in Canada could be dependent on environmental conditions in the United States experienced by monarchs during earlier portions of the annual cycle (Zipkin et al. 2012; Saunders et al. 2018). Monarch butterfly development and survival is known to be impacted by local temperature and precipitation. Although we did not see their effect at the broad spatial scale of our modeling efforts, both temperature and precipitation may be important within the breeding population and warrant future research at a finer geographic scale than performed in this study.

For monarchs in eastern Canada that migrate north in the spring, the first observations that occur in southern Ontario are of individuals moving north through Michigan and Ohio that largely originate from the central and southern United States (Miller et al. 2012). In some years, first observations may occur in late April and increase over the next two months suggesting colonization from areas farther south until approximately July when monarchs have reached their breeding distribution limit (Flockhart et al. 2013). In June, the first observations also occur in southern Manitoba and these butterflies likely originate from the US Midwest via Minnesota and North Dakota. Therefore, while monarchs colonize Canada at two different locations the origins of these individuals also occur from the midwestern and southern regions of the United States where regional weather and habitat conditions will have influenced population processes. In western Canada, monarchs occur in the Fraser River Valley in the Lower mainland and in Douglas-fir/bunchgrass ecosystems in the Central Interior. Monarchs in the west, therefore, likely fly north up the Okanagan River valley in the summer.

Locations for habitat restoration should be prioritized based on where monarchs have a high probability of occurrence annually, occur regularly each year, and contribute to maximum monarch population growth. Despite interannual variation in the extent of monarch breeding distribution in Canada, the core of the breeding distribution is in eastern Canada, primarily southern Ontario, and to a less extent southern Quebec and Manitoba (Brower 1995; Wassenaar and Hobson 1998; Flockhart et al. 2013). Southern Ontario is densely populated and all areas in the core of the breeding range are found in land uses that are predominantly agricultural. Given the tendency for monarchs to lay more eggs per host plant in agricultural areas compared with other land uses in the US Midwest (Oberhauser et al. 2001; Pleasants and Oberhauser 2013) and in southern Ontario (Pitman et al. 2018), management strategies that engage with the agricultural sector will almost certainly be required (Thogmartin et al. 2017b; Pitman et al. 2018). For example, one strategy may include compensation for delayed harvest to ensure host plant establishment or monarch recruitment (Drechsler et al. 2007). Alternatively, another option in agricultural landscapes would be to promote habitat restoration in roadsides (Kasten et al. 2016; Pitman et al. 2018). Although the immediate focus of habitat restoration for monarch butterflies in Canada should prioritize productive habitats in southern Ontario, it must be recognized that those potential gains may only contribute marginally to the viability of monarchs in eastern North America (Flockhart et al. 2015). Given the large-scale restoration of host plants necessary to recover monarchs, a diverse portfolio of strategies must be considered across an appropriate geographic extent (Thogmartin et al. 2017b).

There are a few considerations to note when interpreting the results of our work. Our data come from two citizen science programs and their use assumes that sampling occurs randomly throughout the landscape and that detection probability is constant across sites. Citizen science observations are rarely collected from random locations or uniformly across a landscape and, therefore, should be controlled in some manner during analysis (Yackulic et al. 2013). Several approaches have been proposed to account for biased sampling (Phillips et al. 2009). For monarchs, variation in sampling intensity is correlated with the number of potential observers in a given area then controlling for human population density is one manner to account for this bias when mapping breeding distributions (Flockhart et al. 2013). Additionally, the distribution of citizen science observations some years invokes limits to the geographic extent or resolution of the environmental variables that could be considered to explain monarch distributions, especially when considering species distributions at large geographic scales. Detection probability is rarely perfect for any organism but as we are using presence-only observation, any detection probability below unity results in a reduced sample size of occurrence records. However, if detection probability is heterogeneous across the landscape, then it likely correlates with our explanatory variables, which would confound our interpretation of the probability of occurrence with detection probability (Yackulic et al. 2013). Systematic surveys for monarchs in other citizen science projects have focused on monarchs that allow for effort to be controlled, presence-absence data are recorded, and surveys that occur repeatedly over the breeding season (e.g., Monarch Larvae Monitoring Program, Mission Monarch; Ries and Oberhauser 2015) would allow the concurrent analysis of the probability of occurrence accounting for variation in sampling intensity and detection probability (Yackulic et al. 2013). The effects of these limitations in our study, if they exist, would be to lower the probabilities of occurrence for monarchs across Canada and thereby reduce the annual breeding distribution area, but these effects would presumably occur across all years and, therefore, would not have a large impact on variability of the breeding distribution area over time. Future analyses should consider a range of distribution models of higher resolution using measures of monarch density based on counts conducted at precisely georeferenced locations.

Our quantitative assessment of the probability of occurrence of monarchs in Canada provides important information to develop successful management strategies because it sets the geographic context for management decisions. The next step is to prioritize management actions while accounting for variation in use of habitat in different land use by monarchs (Pitman et al. 2018), differential recruitment of offspring in these habitats (Oberhauser et al. 2001), and the associated costs of restoring habitat that will maximize population growth of monarchs. Our study supports efforts to identify strategic locations for restoration that will be necessary in regional and national conservation plans for monarch butterflies in Canada.

We thank E. Howard and the large number of observers that have provided their monarch observations through citizen science programs such as Journey North and eButterfly focused on butterflies and monarchs, specifically. DTTF was supported by a Liber Ero Postdoctoral Fellowship and DRN was supported by a University Research Chair from the University of Guelph.