Complex Messages in Long-Term Monitoring of Regal Fritillary (Speyeria idalia) (Lepidoptera: Nymphalidae) in the State of Wisconsin, USA, 1988–2015

The regal fritillary (“regal”) (Speyeria idalia) is endangered in Wisconsin, USA, and declining and at risk range-wide. During 1988–2015, we surveyed 24 known regal sites and >100 areas of potential habitat in Wisconsin. We recorded 9037 individuals in 742.7 km on the peak survey per year at occupied sites. At six sites surveyed over 5–25 years, we found regal fritillaries in only one year, mostly in the latter half of the study. The three populations in the state with more favorable trends than the median had a never-burned refugium and/or infrequent fire management. They also all had substantial amounts of grazing, haying, and/or mowing managements. Sites with trends below the regional median trend had frequent or moderate fire management, and either a diminishing never-burned refugium or none at all. Regal populations at sites with ≤15 ha of grassland have become undetectable. Nonetheless, Hogback, a slightly larger than 15 ha site, had the most favorable trend, a significant increase. Nearly all Wisconsin Regal populations known before 1990 declined to consistent non-findability, even though these were conserved sites. More favorable trends at more recently discovered populations may be attributable to species-specific habitat management protocols implemented in the 1990s. Two sites with better than median long-term trends represent the longest consistent land ownership of known Regal populations in the state. This wide range of population outcomes illustrates both the need for long-term monitoring and the challenges of explaining the outcomes. Despite evidence of increasing Regal dispersal, this species remains very localized, indicating the unsuitability of the wider landscape as regal habitat. The number of significantly declining or no longer detectable populations in Wisconsin indicates an ever more adverse landscape for this species. Sites will need to have habitat characteristics that are ever more optimal in a wide range of climatic conditions for Regal populations to persist.


Introduction
Surveying and monitoring are necessary components of conservation programs for rare or declining butterfly species [1,2]. However, determining a butterfly's status and trend is greatly complicated by large variation in abundance among generations attributable to climatic fluctuations, land use, and management activities, and sometimes other factors, such as disease, parasitoids, and density-dependent population regulation [3][4][5][6][7]. Our long-term study here exemplifies both the value of long-term monitoring data and the challenges of interpreting them. In Southern Wisconsin, we surveyed 28 sites because they either had Regal records (historically or currently) or appeared plausible as potential habitat for this and/or other prairie-specialized species. We also added one site complex in central Wisconsin (Buena Vista Grassland, Portage County) in 1997 specifically to search for regal fritillaries, as well as study other grassland butterflies and birds. At many other survey sites in Central and Northern Wisconsin, we had other primary survey goals but our visits there during summer also afforded valid assays for the regal fritillary: (1) six state-owned wildlife areas containing complexes of old fields (degraded grasslands) to study grassland species, especially birds; most were in counties within the regal's historic range [42][43][44]; (2) >125 pine-oak barrens (prairie-like herbaceous flora with trees and shrubs intermixed) in Central and Northwestern Wisconsin to study the Karner blue (Lycaeides melissa samuelis), federally listed as endangered [45], and other barrens butterflies [40,46,47]; and (3) >10 barrens and heaths across Northern Wisconsin (north of Karner blue range); and (4) 75 bogs and 20 bog roadsides in Northern Wisconsin to study their butterfly community, which includes other butterfly species that feed on violets as larvae [48][49][50]. Most sites were conserved lands in private or government ownership (e.g., In Southern Wisconsin, we surveyed 28 sites because they either had Regal records (historically or currently) or appeared plausible as potential habitat for this and/or other prairie-specialized species. We also added one site complex in central Wisconsin (Buena Vista Grassland, Portage County) in 1997 specifically to search for regal fritillaries, as well as study other grassland butterflies and birds. At many other survey sites in Central and Northern Wisconsin, we had other primary survey goals but our visits there during summer also afforded valid assays for the regal fritillary: (1) six state-owned wildlife areas containing complexes of old fields (degraded grasslands) to study grassland species, especially birds; most were in counties within the regal's historic range [42][43][44]; (2) >125 pine-oak barrens (prairie-like herbaceous flora with trees and shrubs intermixed) in Central and Northwestern Wisconsin to study the Karner blue (Lycaeides melissa samuelis), federally listed as endangered [45], and other barrens butterflies [40,46,47]; and (3) >10 barrens and heaths across Northern Wisconsin (north of Karner blue range); and (4) 75 bogs and 20 bog roadsides in Northern Wisconsin to study their butterfly community, which includes other butterfly species that feed on violets as larvae [48][49][50]. Most sites were conserved lands in private or government ownership (e.g., preserves, federal refuges, state parks, state wildlife areas, state, and county forest) or public rights-of-way. Table 1. Descriptive statistics for each site (county in parentheses) on the number of years during 1988-2015 peak counts were obtained by one surveyor or two surveyors. Before 1998, all peak counts by a single surveyor occurred in years with two surveyor visits during the flight period. A single site of untilled grassland (in terms of contiguous tracts with the same ownership) was subdivided into multiple study sites if ≥200 m of wetland, woodland, intensive continuous farm grazing, and/or tilled land intervened among our sampling areas. However, we maintained separate time series for Thousand's II and Thomson (subsequent acquisition) Prairies in Iowa County, even though they are about 100 m apart, because of their different survey histories (1990 on at the former, 1992 on at the latter). Buena Vista Grassland Wildlife Area in Portage County is a complex of eight sites (24-1350 ha) aggregated into one of the largest grassland complexes east of the Mississippi River, with about 5000 ha of public land and a large amount of surrounding private grassland [51].

Site (County) N Surveyors Years of Peak Counts by N Surveyors
Our long-term monitoring sites include all sites known in 1990-1992 to have Regal records during that period, as well as many of the Regal sites discovered subsequently. We discontinued consecutive-year surveying at Spring Green in 1999 because we had never found any regal fritillaries there; the last report was in 1990 [52,53]. However, we resumed surveying there in some years from 2005 to 2015, primarily for other target species. One population (in St. Croix County) was beyond the scope of our scheduling constraints to continue visiting after 2000. We surveyed here during 1998-2000 (three regal individuals found in 1998, one in 1999, and zero in 2000).

Butterfly Surveys
We conducted butterfly transect surveys along similar routes within each site each visit, similar to Pollard [54], as described in Swengel [39,40] and Swengel and Swengel [13,15,32,33]. We counted all adult butterflies observed ahead and to the sides, to the limit of species identification (possibly with binoculars after detection) and our ability to track individuals. We walked at a slow pace (1.5-2 km/h) on parallel routes 5-10 m apart, or down the middle of this route corridor if surveying alone. We tried to have both of us surveying together as much as possible for methodological consistency. Surveys by one person were conducted, however, to fill data gaps caused by scheduling constraints and poor weather on days when both surveyors could be fielded (Table 1). All surveys in this study were conducted by one or both of the authors, which may afford more methodological consistency than if all surveys were conducted by two people, but with variation in personnel.
We designated a new sampling unit whenever the vegetation along the route varied by management, type (wetland, wet, wet-mesic, mesic, dry-mesic, dry, "extra" dry sand), vegetative quality (amount of brush and diversity and abundance of native and exotic flora), and/or canopy. Starting in 2000, we also directly estimated percent shrub cover per unit. Routes crossed rather than followed ecotones and management boundaries to reduce edge effects, and were designed to minimize number of unit changes during the survey while covering representative areas of the site. Survey distance was approximately similar within unit among survey dates, and was measured based on topographic maps, landmarks, and property markers aligned to section (square mile) lines. At Buena Vista it was feasible to standardize the survey route to an 800 m length, typically a square (200 m per side) set in the center of a 40 acre (16.2 ha) square block of land. Some units had to be divided in half (two 400 m transects) when a management treatment occurred in only half the unit. At all other sites analyzed in this study, unit size and route length varied due to vegetative characteristics and size of land management treatments. For each unit survey, we recorded temperature and time spent surveying, and we estimated wind speed, percent time the sun was shining, and percent cloud cover. We surveyed in a wide range of times of day and weather, occasionally in intermittent light drizzle, if butterfly activity was apparent, but not in continuous rain. We kept data separate by unit survey.
In the first year or two of surveying at a site, the route within a unit or number of units surveyed might have been notably smaller than once we standardized the route for all subsequent long-term monitoring surveys. This initial pilot period occurred at Muralt Bluff (1988)(1989), Spring Green (1989), and Pine Island site 1 (dog training area) in 1993-1994, as described previously [33]. After this initial pilot period, when we added units, we maintained them as separate time series. For example, we maintained Pine Island site 4 as a separate time series even though this area is adjoining to the east of site 1, the dog training area.
A unit's management was coded based on treatments observed or evident during the study, including combinations (e.g., burning + mowing) as appropriate and following information available from the agencies that owned and/or managed the sites [32,33]. Fires typically occurred in a rotation of units burned in different years, possibly with some mowing or brush-cutting by hand or machine too. Non-broadcast managements (e.g., hand-cutting of brush) were counted as a treatment only in years when substantial alteration of vegetation occurred. Sites with no active broadcast management conducted or otherwise evident were categorized as "idle" (long-term unmanaged). In the early years of this study, Hogback was a non-conserved site with continuous moderate dairy grazing but cattle grazing did not occur after 1997, following conservation acquisition [33,55]. Cattle grazing (sometimes with horses, also) at Buena Vista was growing-season long, usually for one or two years at a time [15]. We observed cattle grazing at one other regal site (Thomson subsequent, 2012-2014).
Throughout the study, we conducted surveys on multiple dates each year at sites with Regal populations. We started no later than late June (except in the very cool growing season of 2004, when we began on 2 July) and continuing through July, to ensure we did not miss the onset of the regal's main flight period. We were also attentive each year in the regal's historic Wisconsin range to adults of the Aphrodite fritillary (S. aphrodite), a more widespread congenor similar but slightly earlier in flight period timing [33,56]. We also noted other butterfly species that emerge shortly before the regal fritillary as cues to check regal sites.
We required no ethical approval or research permits for this study because we did not handle or experiment on any animals and our survey sites were either open to public visitation or we obtained landowner permission to visit them.

Data Analysis
All statistics were calculated using ABstat 7.20 [57], with statistical significance set at two-tailed p < 0.05. Since we obtained significant results at a frequency well above expected from spurious Type I statistical error, we did not lower the critical p value further, as more Type II errors (biologically meaningful, but not statistically significant, patterns) would be generated than Type I errors eliminated. All statistical tests in this study are non-parametric, which do not require any assumptions about how the data are distributed (e.g., normality). All correlations were done with the Spearman rank correlation. To test for significant differences between categories, we used the Mann-Whitney U test.
To compare relative abundance, we calculated observation rates as total individuals per total survey distance per site (sum of units), using the peak survey at the site per year. Using the one peak survey during the main flight period avoids pseudoreplication (counting the same individual in more than one value in the dependent variable) and has been adequate for producing representative indices for comparisons of relative abundance within and among sites [35,47,58]. We calculated the median observation rate per year for pools of sites surveyed each year during 1990-2015 (N = 4 sites), 1992-2015 (N = 5 sites), and 1997-2015 (N = 7 sites) to serve as regional population abundance indices. At the scales of the site, site complex, and pools of sites, we correlated relative abundance with year to calculate trend. We calculated the median observation rate per year for pools of sites surveyed each year.
At Buena Vista Grassland, adequate samples were available to compare different methods of calculating abundance. On the scale of this entire site complex, we calculated relative abundance indices for mutually exclusive sets of long-term monitoring units (N = 2 units), and Southwest (N = 2 units). We correlated these four geographically segregated time series with year (trend) and in pair-wise correlations with each other. The spatial range of units surveyed in 1997 was not sufficient to include that year in this analysis by geography.
It was not valid to make one-and two-surveyor counts comparable by multiplying one-surveyor counts by two. On two-surveyor counts, we surveyed together in one party as done in the Christmas Bird Count and 4th of July Butterfly Count [59], and had one recorder to eliminate double-counting when both surveyors observed the same butterfly. Two surveyors would not necessarily record twice as many butterflies as one surveyor. It was also undesirable to limit analyses only to two-surveyor peak counts, as some of these were lower than one-surveyor counts at some sites in some years. It is not possible to know how many butterflies would have been recorded if both of us had been present for those one-surveyor counts. In prior analyses, the number of surveyors did not appear to have significant impacts on the statistical outcomes [15,33]. As a result, we included one-surveyor surveys in analysis without making any adjustments for comparability to two-surveyor surveys.
Instead, we disclose here which sites had how many one-surveyor surveys ( Table 1). All long-term sites, except Pine Island, were primarily surveyed with two surveyors. They also had a two-surveyor count in years when the peak came from a count by one surveyor, except for most sites with one surveyor in 2012 and all sites with one surveyor 2013-2015. Muralt Bluff, Oliver, and the Thomson complex were primarily surveyed by two surveyors until 2012-2015. However, regal fritillaries at Muralt Bluff and Oliver had appeared to crash prior to 2012. Thus, comparing abundances from more recent one-surveyor results to prior two-surveyor results would likely have the most potential effect on trend analyses at the Thomson complex and Barneveld. Most surveys at Pine Island were done by one surveyor. The other sites primarily covered by one surveyor were either sites more recently added to the study or marginal for the regal fritillary with few records since 1990. Since comparability among sites in abundance would be also be affected by number of surveyors, we did no tests of differences in abundance among site complexes.
Starting in 2012, we participated in a single survey day per year at Fort McCoy, a large military reservation with >24,000 ha in Monroe County [60]. The target species was another prairie specialist butterfly, but the timing occurred during the main regal flight period. Due to safety and access issues, at least one staff biologist accompanied us, as well as other agency staff and/or volunteers. We recorded all butterfly observations by all observers in a unit as a single total of observations and survey effort per unit. Due to variation in units and routes within units surveyed per year, we calculated a single population index per survey day as total regal individuals per estimated total distance surveyed by the group (not distance walked by each individual surveyor). This method of calculating survey effort (by group or "party" surveying together, rather than by individual surveyor) is comparable to methods of the 4th of July Butterfly Count Program [59]. Due to the roughness of the data and few years from Fort McCoy, we provide summary statistics but no statistical analysis of this dataset. We also surveyed other prairie and grassland sites both in the known current range (in, and south of, counties with populations as mapped in Figure 1) and north of that range in Wisconsin. At a number of sites with plausible habitat, we did not find any regal fritillaries and know of no such records by anyone else (Appendix A Table A1).                     At six sites surveyed for at least five years (14-25 years at five sites), we found any regal fritillaries in only one year ( Figure 9). Although one of these records occurred in 1989, they are skewed to the 21st century (five of six sites) and later in the time series (the last 20% of the time series at 4/6 sites). Half of these records occurred in 2012, the year with the earliest (warmest) seasonal development in this study as evidenced by flight period data ( Table 2). A few other sporadic records have also been reported by volunteers in the last decade ( Table 3).

Patterns of Occurrence
The only records for the regal fritillary in Wisconsin in the 4th of July Butterfly Count Program are from five count circles, and only in one year per circle (Table 4). These records are similarly biased later in the time series: 4/5 are in the latter 50% of the published time series (2/3 when excluding the Baraboo and Wazee Counts, which are also represented in Figure 9).  (Table 4). At six sites surveyed for at least five years (14-25 years at five sites), we found any regal fritillaries in only one year ( Figure 9). Although one of these records occurred in 1989, they are skewed to the 21st century (five of six sites) and later in the time series (the last 20% of the time series at 4/6 sites). Half of these records occurred in 2012, the year with the earliest (warmest) seasonal development in this study as evidenced by flight period data ( Table 2). A few other sporadic records have also been reported by volunteers in the last decade ( Table 3).
The only records for the regal fritillary in Wisconsin in the 4th of July Butterfly Count Program are from five count circles, and only in one year per circle (Table 4). These records are similarly biased later in the time series: 4/5 are in the latter 50% of the published time series (2/3 when excluding the Baraboo and Wazee Counts, which are also represented in Figure 9). At six sites surveyed for at least five years (14-25 years at five sites), we found any regal fritillaries in only one year ( Figure 9). Although one of these records occurred in 1989, they are skewed to the 21st century (five of six sites) and later in the time series (the last 20% of the time series at 4/6 sites). Half of these records occurred in 2012, the year with the earliest (warmest) seasonal development in this study as evidenced by flight period data ( Table 2). A few other sporadic records have also been reported by volunteers in the last decade ( Table 3).
The only records for the regal fritillary in Wisconsin in the 4th of July Butterfly Count Program are from five count circles, and only in one year per circle (Table 4). These records are similarly biased later in the time series: 4/5 are in the latter 50% of the published time series (2/3 when excluding the Baraboo and Wazee Counts, which are also represented in Figure 9).  (Table 4).  (Table 4).

Population Trends
The populations in our long-term monitoring sites varied greatly in trend over time (Table 5). Hogback had the most favorable trend (significant increase), both in comparison of trends calculated with all years of data available and when limited to 1997-2015 to control for time period among sites.
The poorest outcome occurred at formerly reliable populations that declined to consistent non-detectability (Figure 2). At Muralt Bluff, we found none in the last five years, and only a single individual once during 2009-2015 (in 2010). Oliver was reported to have reliable regal detection prior to this study [53] but we only found inconsistent regal occurrence, and none during 2005-2015. Muralt Bluff had a strongly significant negative trend but Oliver had only a non-significant decline because we detected so few individuals and so sporadically, although these were entirely in the first 3/5 of the study (Table 5). Spring Green formerly had a reliable population prior to this study [53]. However, with only one year of regal observation in our surveys (Figure 9), we did not calculate a trend for this site. We do not know whether Pine Island 2 (west) ever had a reliable population. We found the Regal fritillary there in only two years (1995-1996) during 1993-2015 ( Figure 5). Pine Island 1 (dog training area) declined significantly to inconsistent detection recently. However, an area adjacent to this but not surveyed as many years (site 4, east) has continued to have reliable findability in recent years, but in low numbers ( Figure 5).
Since regal fritillaries occurred throughout the very large area of the Buena Vista Grassland site complex, we used several strategies for assessing occurrence and abundance. On the scale of this entire site complex (Figure 7), a variety of independent measures of regal presence and abundance covaried very strongly (Table 6). However, when the long-term monitoring units at Buena Vista were grouped geographically into four areas, only two of six pair-wise correlations of abundance were significant, although all were positive (Table 7). These different areas of Buena Vista had different long-term trends (Table 8), from no trend (Southeast) to mild non-significant negative trends (West Central, Southwest), to significant decrease (North). Regal abundance was significantly highest in the West Central area, while the other areas did not differ significantly from each other in abundance (Table 9, Figure 8).
This variation in trend among sites in a complex is also apparent at the Thomson complex. Thomson (original) has increased (significantly so over the entire study period), while Thousand's II and Thomson (subsequent) had less favorable trends (Table 5): a non-significant positive trend at the former and a negative trend (significantly so during 1997-2015) at the latter. It remains possible that the four recent years of one-surveyor surveys here made the calculation of long-term trend less favorable than it would have been with two-surveyor surveys. Thus, these results should be viewed as rougher than at the other sites. But our highest regal count ever in the 26-year time series at Thomson (original) was by one surveyor in 2015 (Figure 3). Table 5. Spearman rank correlations of trend (abundance vs. year). These were calculated for populations (sites where regal fritillaries were recorded in more than one year in this study) surveyed >12 years, with county in parentheses. NS = not significant (p > 0.10). p < 0.10 is considered near significant.   Fort McCoy data are rougher due to variation among years in number and identity of surveyors and in locations surveyed at this large site complex. The relative abundance and direction of change in abundance among years were generally consistent with results at Buena Vista, also in Central Wisconsin (Figure 8). The abundance index in 2016 was 16.2 (the highest index in this time series), when one staff biologist and no one else surveyed with us. The high 2016 abundance indicates there was not a consistent decline at this site in the last five years.

Comparisons of Site Characteristics
We identified the median abundance index per year for sites surveyed each year in the period ( Figure 10). This median abundance was low early in the 1990s, peaked in the middle of the study, and was persistently low again in recent years ( Figure 10). The two longest running of these median time series had mild non-significant negative trends (Table 10). The shortest series began in the period of peak abundance and had a significant negative trend (Table 10). Table 10. Spearman rank correlations of trend (abundance vs. year) in median abundance of pools of sites monitored each year during the period (Figure 10). NS = not significant (p > 0.10). We ordinated sites by trend (increasing, relatively stable, decreasing) and classified them by site factors (patch size, fire and other management activities) for the seven long-term monitoring sites surveyed 1997-2015 (Table 11). Four sites had a more favorable trend than the regional median trend (Table 10). But one of these (Oliver) had 11 years of a zero abundance index at the end of its time series. As a result, Oliver had the negative outcome of appearing no longer to support a reliably detectable population. The three other populations had a never-burned refugium and/or infrequent fire either throughout or later in the study period. They also all had substantial amounts of alternative managements (e.g., grazing, haying, mowing). While Thomson (original) had frequent fire (with 100% fire) early in the study, this was mitigated by being embedded in a site complex that was not 100% burned at once. This site also had large amounts of mowing management throughout the study, and very little fire in the second half of the study. The sites with poorer trends than the median had either frequent or moderate fire. They also had either a diminishing never-burned refugium or none at all. All of the smallest sites (15 ha of grassland or less, with no plantings either) have become undetectable recently. Nonetheless, Hogback was just slightly larger than 15 ha and has had the most favorable positive trend. Table 11. Characteristics of sites analyzed for trend during 1997-2015 (Table 5). Sites are listed from most positive (+) to most negative (−) trend, with the trend of the median abundance for these sites (negative at p < 0.05, per Table 10) also indicated. The bottom site has no calculated trend because no regal fritillaries were recorded there during this period. was not a consistent decline at this site in the last five years.

Comparisons of Site Characteristics
We identified the median abundance index per year for sites surveyed each year in the period ( Figure 10). This median abundance was low early in the 1990s, peaked in the middle of the study, and was persistently low again in recent years ( Figure 10). The two longest running of these median time series had mild non-significant negative trends (Table 10). The shortest series began in the period of peak abundance and had a significant negative trend (Table 10).

Patterns of Occurrence
On the one hand, many sources of data (Tables 3 and 4, Appendix A Table A1) indicate the overall rarity and localization of the regal fritillary in Wisconsin ( Figure 1) and range-wide [37]. An observer must visit relatively few and very particular sites to find this butterfly reliably, which rarely turns up elsewhere in the landscape.
On the other hand, there is some evidence that regal fritillaries occasionally disperse out from a population. Regal occurrence in only one of many years of our surveys at a site ( Figure 9) is consistent with similar single-date or single-year records for an area reported by others in the 4th of July Butterfly Count Program (Table 4) and other volunteer data ( Table 3). The clustering of most of these sporadic records in and adjoining counties with regal populations (Figure 1) supports that these sporadic records may be symptomatic of dispersal from a relatively near population. In addition, proximity of some sporadic records to riverways (as evident by serpentine county and state borders in Figure 1) suggests that these topographic features may serve as dispersal corridors.
These sporadic records appear to have increased in Wisconsin in recent years, both in our surveys ( Figure 9) and in the 4th of July Count Program (Table 4). In both these data sources, effort is accounted for (i.e., years with zero observations are documented). Thus, this increase in sporadic records appears to be a true pattern and not just a consequence of increased survey effort (more observers and/or more field days) and/or increased interest in reporting the species. Since Wisconsin is at the northern edge of the species' range, this may be a possible consequence of climate change. Many butterfly species have been documented to increase in abundance in their northern range, and expand their range, in recent decades in association with warming climate [61][62][63][64][65][66].
While these sporadic records indicate the potential for regal fritillaries to disperse out of localized population areas, they also indicate very limited effectiveness of this dispersal to found reliably detectable populations afterwards. This assessment is based on well surveyed sites rarely having a consistently detectable regal population discovered in them many years after consistent surveying would have been adequate for finding regal fritillaries in them (Figure 9 and Appendix A Table A1).
Fort McCoy is a probable exception. Either the regal fritillary colonized at this northern range edge recently, or a population occurred here all along but was undetected despite a number of knowledgeable surveyors on the site for many years. Some of those prior surveys included specific searches for this species [67], while others were targeting summer Karner blues [68][69][70] or grassland birds [60] in areas that now have regal records. Although the data in 2012-2015 suggest a decline at Fort McCoy (Figure 8), the population index in 2016 was just above the highest value in 2012-2015, suggesting overall stability in this rough dataset. The much more noteworthy result is that the regal fritillary has reliably occurred here each year since being discovered in July 2010 (Tim Wilder, pers. comm.; Figure 8). Not graphed are the years of non-detection in prior years. Thus, any positive population index now represents an increase over the prior situation of not being discovered.
This outcome at Fort McCoy fits an expectation of increasing abundance and/or range expansion at a species' northern range margin because of warming climate [61,62,64,66]. However, Fort McCoy also fits the definition of high-quality regal habitat as defined by midwestern regal populations: large untilled grassland with a mix of both uplands and swales and unintensive land uses compatible with maintaining both the native flora and the butterflies themselves in all their life stages [5,13,31,34,38]. While fire occurs at Fort McCoy, the fire return interval varies greatly, from annually in weapon firing zones, to frequent, moderate, infrequent, and never (as defined in Table 11). Thus, increased observation of the regal fritillary at Fort McCoy supports that suitable habitat conditions are essential for butterflies to persist successfully in the landscape [65,[71][72][73][74][75].
These complex messages of extreme rarity and localization contrasting with dispersal and colonization have also been documented elsewhere. In isolated Iowa prairies, regal fritillaries overwhelmingly turned back at habitat margins instead of dispersing out of the site into non-habitat [76]. The regal fritillary has also been documented as localized and declining in Illinois and Indiana [77]. However, this butterfly has recently and unexpectedly expanded in the highly-developed landscape of Northeastern Illinois and Northwestern Indiana recently [77]. In a large-scale conservation effort there, tilled fields around remnant prairies were acquired for re-vegetation to native prairie flora. The two violet species deliberately included in the diverse plantings established only sparsely. However, a ruderal annual violet unexpectedly established profusely in both the plantings and in unplanted old fields that developed once tilling ceased. Regal fritillaries have recently increased and expanded within the project area outside the prairie remnants, and achieved equal abundances in prairie plantings and old fields. This butterfly also dispersed and colonized on a landscape scale of 5-10 km and more to re-occupy historic sites as well as colonize other sites. On the one hand, these results support that the regal fritillary is usually localized because of the paucity of suitable vegetation in compatible land uses and managements, but when required floristic resources consistently abound on a landscape scale, the regal fritillary is able to respond and colonize suitable habitat patches.

Flight Period Characteristics
At Buena Vista, the flight span was longer when peak dates were earlier-i.e., in warmer years (Table 2). Thus, survey timing must not only be shifted earlier or later due to climatic variation, but the length of the survey period also varies from year to year. Independent of that, flight period span also correlated positively with abundance. Thus, phenology is not the only factor determining how long the flight period (and survey season) is in a year. More analysis is needed to develop predictions of regal abundance in the next flight period in relation to observed climate conditions during the immature life stages. Until then, it is easier to anticipate when the peak flight period will occur, based on seasonal development and first observed dates, than how long the flight will last in the summer based on anticipated regal abundance (e.g., for planning personnel schedules for presence/absence surveying).

Population Trends
Median abundance ( Figure 10) is useful for comparing which sites had more or less favorable trends than the median regional value (Tables 5 and 11). Collectively, these population values exhibited high interannual variability, with highest abundance in the middle of the study (early 2000s). Most sites individually also exhibited a very wide range of abundance as well (Figure 2: Muralt Bluff; Figure 3, Figure 4, and Figures 6-8). This suggests that regional climatic variation has been one important determinant of regal abundance in Wisconsin.
However, site-specific factors of habitat and land use are also strongly implicated in regal population outcomes. For example, a few populations had consistently lower population values throughout the study period, in contrast to relatively high median abundance values ( Figure 10). This includes Pine Island ( Figure 5), a large site complex but containing degraded old field (Table 11), and Oliver (Figure 2), a high-quality prairie but very small and near Muralt Bluff, which also was also relatively small (Table 11). Thus, these sites did not register higher abundance values that appear attributable to regional climatic patterns.
Furthermore, nearly all Wisconsin regal populations known before 1990 declined to apparently consistent non-findability of the populations: Muralt Bluff and Oliver (Figure 2), and Spring Green (Figure 9). The exception is one low-density population with regal fritillaries still recorded as present at the end of this study: Pine Island site 1 dog training area and site 4 east of this ( Figure 5). Spring Green's population was last seen in 1990 [53]. Then individuals were next observed here in 2012 by multiple observers (Figure 9; Table 3), but not in following years. This suggests that 2012 observations here represented transient dispersal from somewhere else rather than re-detection or re-establishment of this population. Thus, even though these historical populations were in conserved sites, which did not result in a beneficial long-term outcome for regal populations at most of these sites.
With increased concern about the regal fritillary and more search effort by many observers, as summarized in [32], more populations were discovered in the 1990s. This included Hogback, Thomson complex and additional areas in Iowa County in the vicinity of Thomson, and Buena Vista Grassland Swengel et al. [36] reported 3/8 positive trends to any degree and 5/8 negative to any degree for Wisconsin regal populations, which was not a significant skewing of trend in one direction or the other. Since that analysis, the extant populations have exhibited a wide range of population trends (Table 5), including a number of significant declines. By contrast, Hogback has the most favorable trend (significant increase). Fort McCoy has also increased from consistent non-observation to consistent detection. Two portions of the Thomson complex also have positive trends (significant at original, non-significant at Thousand's II), while the third portion has a significant decline. Buena Vista also has variability in trend among different areas within this site complex (Table 7), including apparent stability. At this last site, the lack of positive trends (compared to some Southern Wisconsin sites) may be due to the lack of data from earlier in the 1990s. The Buena Vista time series started right before or during the peak period of median abundance for Wisconsin populations collectively ( Figure 10).
In the 1990s, Wisconsin implemented species-specific habitat management protocols for rare butterfly conservation [36,55]. The regal fritillary protocol specified allowable fire regimes and encouraged mowing, permanent non-fire refugia, and monitoring the butterfly. This appeared to confer benefit on regal trend, compared to generalized ecosystem or habitat conservation approaches in Iowa and historic Wisconsin regal sites from the 1970s and 1980s. Whether these protocols are sufficient, and sufficiently complied with, to maintain regal populations long-term at these more recently discovered sites remains to be seen in the coming decades. That is, has the overwhelming decline for the populations known in the 1970s and 1980s been adequately turned around at sites discovered since?

Comparisons of Site Characteristics
Regal fritillary trends are a combination of regional climatic and site-specific habitat factors including vegetative composition, land use/management, and the landscape context [13,15,26,33,34,55,79]. Additionally, isolated population pathology may be an influence, as well as greater variability at range edge [80,81]. Each population area is its own unique instance of site and landscape factors (Table 11) and climatic conditions. It is difficult to parse these factors statistically in long-term population time series because of the distance among population areas, leading to variation in climate. Thus, broad-scale surveying and analysis across many sites in many states have been invaluable in identifying factors relevant to regal incidence and abundance [5,26,34]. Habitat and land management factors have been analyzed extensively [13,15,16,31,[39][40][41]55,76,79,[81][82][83][84] but we are not aware of any analysis of regal data related to climatic variation in the Midwest or elsewhere.
A synergy (positive or negative) between length of land ownership is also evident in Wisconsin's regal fritillary outcomes. On the one hand, there is a lag time of years or decades between when management implicated as adverse for the regal fritillary begins and when the regal population declines to non-findability (Table 11). On the other hand, Fort McCoy has an outstanding butterfly fauna besides regal fritillary [67][68][69][70], and represents the single longest landowner and land use of any known regal population in the state. Fire frequency varies greatly among parts of this military reservation, from frequent to moderate to infrequent to none very long-term. Other land use activities, both deliberately for vegetative management (brush cutting, timber sales, exotic plant control) and inadvertent consequences of military exercises (bivouacking, tank driving, and so on), also occur here. Vegetative structures vary from grassland (both short and taller turfs) to savanna to forest, all of these in varying patch sizes.
The second longest consistent ownership (>60 years) and land use approach for a Wisconsin regal population is Buena Vista Grassland, managed by the Wisconsin Department of Natural Resources for Greater Prairie-Chicken (Tympanuchus cupido pinnatus) [51]. Both of these site complexes hint that long-term consistency of compatible land uses in large patches of habitat is important for regal population persistence.

Conclusions
The wide range of population outcomes documented in this study illustrates both the need and the challenge of anticipating future effects of climatic variation and site-specific habitat characteristics and their landscape context. Our study provides evidence of increasing regal dispersal and northward population expansion, possibly in response to climate. Despite this, the regal fritillary remains very localized to known population areas, indicating the unsuitability of the wider landscape as regal habitat. Very large fluctuations in abundance exhibited by populations with stable or even increasing trends indicates that extreme weather can be adverse for regal abundance, even though this butterfly has a wide climatic tolerance based on its large range. The relatively large number of significantly declining or no longer detectable populations in Wisconsin indicates an ever more adverse landscape for this species. In such a circumstance, it is reasonable to expect that a site will need to have habitat characteristics that are ever more optimal, and optimal for a wide range of climatic conditions, for a regal population to persist viably.