Habitat Suitability Modeling of the Federally Endangered Poweshiek Skipperling in Michigan

In 2014 the U.S. Fish and Wildlife Service (USFWS) and the Committee on the Status of Endangered Wildlife in Canada listed the Poweshiek skipperling butterfly Oarisma poweshiek (Lepidoptera: Hesperiidae) as an endangered species under the U.S. Endangered Species Act of 1973 and the 2002 Species at Risk Act (SARA), respectively (ESA 1973, as amended; SARA 2002; COSEWIC 2014; USFWS 2014). Surveyors verified Poweshiek skipperlings at only five U.S. localities in summer 2015: four in Michigan prairie fens, totaling 84 recorded individuals (D. Cuthrell, unpublished data), and one in a Wisconsin mesic prairie with only two individuals (Swengel and Swengel 2015). In 2015, surveyors recorded 34 individual Poweshiek skipperling within approximately 2,300 ha of tallgrass prairie preserve in Manitoba (R. Westwood, unpublished data).

The Poweshiek skipperling is univoltine, having a single period per year during which pupae emerge as adults. This period typically occurs between mid-June and late July and lasts for approximately 2 wk. Female Poweshiek skipperlings oviposit on thin, thread-like graminoids, on or near suspected larval host plants (McAlpine 1972; Cuthrell and Slaughter 2012). Suspected larval host plants include prairie dropseed (Sporobolus heterolepis; Michigan special concern), mat muhly (Muhlenbergia richardsonis; Michigan threatened), little bluestem (Schizachyrium scoparium), and one dubious report of golden-seeded spikerush (Eleocharis elliptica; Holzman 1972; Borkin 1995; Selby 2005; Cuthrell and Slaughter 2012). Borkin (1995) reported in situ larval feeding on prairie dropseed and little bluestem in Wisconsin populations, and several oviposition events were documented D.L.C. on mat muhly at a site in Michigan during 2015 (D. Cuthrell, unpublished data). However, researchers need to collect more data to confirm in situ larval host plants in Michigan and the degree of dependence upon these species (Selby 2005; Cuthrell and Slaughter 2012). After adults emerge, they nectar on a suite of flowers that vary throughout their historic distribution. These plants include black-eyed Susan Rudbeckia hirta, shrubby cinquefoil Dasiphora frutcosa, sticky tofieldia Triantha glutinosa, northern bedstraw Galium boreale, pale spike lobelia Lobelia spicata, tickseed Coreopsis palmata, purple coneflower Echinacea angustifolia, and yellow ox-eye Heliopsis helianthoides (Selby 2005, Cuthrell and Slaughter 2012; Figure 1). Poweshiek skipperlings in Michigan are vulnerable to stochastic events (e.g., wildfire, severe weather, flooding), because of their ephemeral flight and possible reliance on plant species of conservation concern as larval hosts.

All documentation of Poweshiek skipperling in Michigan has been within prairie fens (Michigan Natural Features Inventory 2014). In other locales, Poweshiek skipperling populations did or currently do occupy mesic prairies (Selby 2005). Prairie fens are calcareous groundwater-fed natural communities found throughout the upper Midwest. Notably, prairie fens have a tallgrass prairie component, which helps distinguish them from other fens (Kost et al. 2007). The presence of moist soils, plant species associated with tallgrass prairies, and historical juxtaposition with upland prairies or savannas likely produced habitat for Poweshiek skipperling similar to other natural communities throughout their historic range. These similarities have likely left prairie fens as the only natural community in Michigan in which Poweshiek skipperlings can survive post-European settlement (Landis et al. 2012). Prairie fens are at-risk communities threatened by surrounding land cover, invasive species, nutrient-loading, and hydrological alterations (Spieles et al. 1999).

A variety of factors are known to affect insect populations and have been offered by researchers as possible reasons for the decline in Poweshiek skipperlings. These factors include habitat destruction and fragmentation (Ricketts 2001; Selby 2005; Donald and Evans 2006; Davis et al. 2007; Weibull et al. 2008; Cuthrell and Slaughter 2012), pesticides (Dover et al. 1990; Selby 2005; Godfray et al. 2014), invasive species (Kost et al. 2007), imprudent burning practices (Selby 2005; Swengel et al. 2010), pathogens (Selby 2005; Werren et al. 2008; Nice et al. 2009; Poweshiek Working Group 2011), and climate change (Selby 2005; Hoving et al. 2013). Many of these stressors are directly or indirectly related to environmental factors, such as surrounding land cover and road density.

Our objectives were to monitor known Poweshiek skipperling sites and to identify potentially suitable new locations. The reduction in Poweshiek skipperling populations over the last 20 y, recent status change to Federally Endangered, ongoing conversion of habitat, and decreasing survey counts, call for assessments of Poweshiek skipperling status, including an evaluation of potential habitat locations and environmental factors affecting the few remaining populations. Using our recent survey data, our goal was to map other potential sites with similar habitat conditions to help assist with future monitoring and conservation efforts. This is the first study we are aware of to examine potential suitable habitat for the Poweshiek skipperling in Michigan.

We completed modified Pollard-Yates (Pollard and Yates 1993) surveys targeting Poweshiek skipperling in prairie fens from 2009 through 2015. We used results from these surveys to determine site status and locations of Poweshiek skipperlings (Michigan Natural Features Inventory 2015). We included the nine sites with documented occupancy within the past 20 y as our survey locations for all years. In 2014 and 2015 we expanded the number of sites surveyed to include prairie fens with characteristics presumed to be beneficial to Poweshiek skipperling success (e.g., presence of prairie dropseed or mat muhly, proximity to other recently occupied locales). These additional sites had never been surveyed for Poweshiek skipperling to our knowledge. We surveyed 17 sites in 2014 and 15 sites in 2015. We surveyed from June 30 to July 11 during the 2014 flight and from June 25 to July 14 during the 2015 flight. At each site surveyed in 2014 and 2015, we conducted surveys between 1000 and 1800 hours Eastern Daylight Time when temperature was greater than 15°C, there was no precipitation, and winds were less than 25 km/h. When temperatures were 15–21°C, we conducted surveys if cloud cover was less than or equal to 50% of the sky. There was no cloud cover restriction if the temperature was above 21° C. If weather conditions deteriorated during a survey, we terminated the survey and resurveyed the entire site on a suitable day. These restrictions helped provide consistent surveys during peak lepidopteran activity. Surveys consisted of a series of transects paralleling the outer boundary of the prairie fen (or suitable fen patch within a large wetland complex). We conducted surveys with two observers, spaced 10 m apart, in all areas that contained blooming nectar sources (e.g., Rudbeckia hirta, Dasiphora fruticosa, Galium boreale), <50% invasive species cover, and an open canopy (<50% tall [>1.5 m] shrub cover; <25% mature tree cover). While surveying, we recorded survey tracks and waypoints of each identified individual Poweshiek skipperling within 5 m to either side and 10 m in front of the surveyors. For each Poweshiek skipperling detected, the identifying surveyor would verbally indicate that they had recorded it to ensure that no individuals were counted more than once. We completed surveys twice at each location and the maximum count of the two visits was recorded as the survey number for that location.

These surveys helped define the following mutually exclusive categories of occupancy for prairie fen locations: verified extant, recently occupied, historic, and no record of presence (Figure 2). Verified extant areas had Poweshiek skipperling detections in 2015. Recently occupied areas had positive identification of Poweshiek skipperling by a species expert within the past 20 y but did not have detections in 2015. Historical sites have documented Poweshiek skipperling occurrence, but no detections within the last 20 y. Prairie fens with no record of presence have had no detections of Poweshiek skipperling and varied amounts of survey effort, ranging from none to many surveys. It is important to note that in addition to the recently occupied sites, one site had an unverified report of the species from 1996, a second site had six reports of the species but only one documentation that quantified the individuals observed (n = 1 Poweshiek skipperling), and a third had a single documented individual. The two sites with confirmed observations do not have associated location data. Due to the lack of location data for these sites, we chose to not include them in our models.

We developed a habitat suitability model to understand if and to what magnitude different environmental variables are associated with the presence of Poweshiek skipperling, and then used these relationships to identify other similar locations that could be suitable for the species. We used the modeling software Maxent with the Poweshiek skipperling occurrence data and select spatio-environmental characteristics (prairie fen area, road density within 1,000 m, and land cover within 1,000 m) to develop a model that geographically maps locations for habitat suitability (Phillips et al. 2004). Maxent uses presence-only occurrence data and geographic environmental data to generate species–environmental relationship curves that represent the environmental niche of the species, then the niche is geographically mapped to identify other locations with similar environmental variable values. Although many habitat modeling techniques exist, Maxent is well suited for rare species because it does not require absence data, has been shown to be among the best models (Elith et al. 2006; Peterson et al. 2007), and performs well with rare and cryptic species having few occurrence points (Ortega-Huerta and Peterson 2008). Maxent is based on the principle of maximum entropy, which is used to approximate an unknown probability distribution with the fewest constraints given the knowledge of the system. The Maxent software uses a deterministic machine learning algorithm to generate the solution. Features, or presence data, are compared against this distribution and are fit using linear, quadratic, product, threshold, or hinge functions, or a combination of these functions. A more in-depth view of the mathematical theory of maximum entropy regarding Maxent can be found in Phillips et al. (2006).

The data consist of eight verified prairie fens with Poweshiek skipperling populations based on 1,712 observations from 2009 to 2015 and three geographic–environmental factors: prairie fen area (hectares; Michigan Natural Features Inventory 2014), land cover within 1,000 m (hectares; Jin et al. 2013), and road density within 1,000 m (km/km²; TIGER 2014; Table 1). The number of observations used ranged from 147 to 236 depending on training dataset. Prairie fen area is an indicator of overall system health and biodiversity in prairie fens (Hackett 2013), as well as an indicator of resource availability. Land cover within 1,000 m characterizes the landscape matrix and has been demonstrated as an indicator of butterfly habitat quality (Davis et al. 2007). Road density within 1,000 m is an indicator of anthropogenic disturbance and development (Forman 2000).

We scaled all environmental layers to a 30-m resolution using ArcMap 10.2.2 (ESRI 2014) and clipped the data to the same extent. This accommodates the Maxent requirement for consistent spatial attributes among environmental data. We masked non–prairie fen locations by applying a NoData value to these pixels to exclude unsuitable environments. We converted prairie fen polygons to rasters so that each pixel within a particular prairie fen was assigned the value of the total fen area. These data are not provided publicly in an effort to safeguard these vulnerable natural communities. We reclassified land cover into four categories: agricultural, developed, natural, and open water. Each land cover type was represented in raster form, in which each pixel value represents hectares of a land cover type within the 314-ha neighborhood. We selected surrounding land cover with a 1,000-m radius to adequately address the impact of the encompassing matrix without the influence of spatial autocorrelation, while also retaining a local neighborhood. To calculate road density within a 1-km radius of the center of each pixel, we calculated pixel values using the Line Density function in ArcMap 10.2.l (ESRI 2014).

The Maxent modeling software provides many parameters for model creation and assessment. Maxent allows for different function types to fit response curves, including linear, quadratic, product, threshold, and hinge features, or any combination thereof. It has been demonstrated that including both threshold and hinge features can result in overfitted response curves in the model that do not accurately represent the biology of the system; therefore, we excluded threshold features from our response curve development (Heumann et al. 2013). We employed a jackknife approach to evaluate our model due to the small number of sites where Poweshiek skipperlings were documented (Shcheglovita and Anderson 2013). We grouped location data by fen and used seven of the groups for training data, whereas we used data from an eighth prairie fen for validation. This allowed us to examine how well the model could predict known locations of occupied Poweshiek skipperling habitat that were independent of the training dataset. We completed this eight times so that each prairie fen was used as validation data once. Training datasets ranged from 147 to 236 nonoverlapping occurrence points across our eight datasets. We set the maximum number of background points at the default (10,000), used “Crossvalidate” as our replicated run type, and ran 10 replicates per training set. We set the maximum iteration to the default (500) and selected “.asc” as our output file type.

To assess the suitability of prairie fens, we followed the lowest-presence threshold (LPT) methodology of Pearson et al. (2007). To do this, we classified each prairie fen by the maximum within-fen pixel value by calculating zonal statistics in ArcMap using the Michigan Natural Features Inventory prairie fen polygon shapefile to delineate each prairie fen as a zone. We opted to use the maximum pixel value to classify the fens as some Poweshiek skipperling patches noted in the field are similar in size to a 30-m pixel. We determined the LPT by using the Extract Values to Table tool to determine the pixel value (habitat suitability index) at each Poweshiek skipperling observation in the training dataset. To prevent overestimation of model performance, we opted to use a 99% LPT by omitting the bottom 1% of pixel values.

We assessed the habitat suitability models in two ways. First, for each model run, we compared the maximum pixel value of the test fen with the 99% LPT. We considered models successful when the maximum pixel value from the test fen was greater than the 99% LPT. Second, we computed a P value using the pValueCompute software (outlined in Pearson et al. 2007) to empirically estimate the probability of obtaining the results of each model compared to random permutation. Using the pValueCompute software (Pearson et al. 2007), we performed a jackknife analysis examining the success of each of the eight individual models along with the probability of success under randomness. For this probability, we used the maximum pixel value for the test fen. The results provided a success rate and associated P value.

For the final results, we averaged the eight “.asc” models in ArcMap 10.2.2 to generate an overall model (ESRI 2014). We used the entire Poweshiek skipperling occurrence dataset and the overall model to determine the 99% LPT threshold, which we used to identify suitable habitat. We considered prairie fens with maximum pixel values greater than the 99% LPT to potentially contain suitable habitat. Using the average model data, we calculated the mean percentage of contribution of the environmental variables, the species–environment response curves, and the area under the curve of the receiver operating characteristic.

We detected Poweshiek skipperling at four sites in 2015. We failed to detect Poweshiek skipperling in five locations where they have been previously known to exist. We observed no novel Poweshiek skipperling locations (Table 2). We observed a 50% reduction in the number of extant localities since 2008. Survey results in 2015 were lower than 2014 counts at three of the four locations.

Maximum pixel values within all prairie fens in the overall Maxent model ranged from 0 to 0.756 and maximum pixel values at verified extant and recently occupied locations ranged from 0.077 to 0.756. We noted 26 prairie fen locations that have no known occurrence of Poweshiek skipperling that exhibited maximum pixel values greater than the 99% LPT (0.074; Table 3). These 26 prairie fens with potential suitable habitat ranged in size from 5.0 to 110.4 ha with a mean of 25.7 ha and a standard deviation of 25.7 ha. All verified extant locations and five recently occupied locations had a maximum pixel value greater than the 99% LPT.

The jackknife models predicted suitable habitat in the validation fen for six of the eight models. This resulted in a 0.75 success rate with a P value <0.05. The overall model predicted the greatest suitability of Poweshiek skipperling habitat in the eastern portion of the study area. Standard deviations for the area under the curve of the receiver operating characteristic of the eight models were very small (0.006 to 0.017).

Mean percentage of contribution of the environmental variables ranged from prairie fen area at 43.0% to surrounding developed land cover at 4.4% (Table 4). Based on percentage of contribution, prairie fen area, followed by surrounding natural and agricultural land cover, were the principal environmental variables contributing to the model based on our data inputs (Figure 3). Species–environment response curves show that suitability increases with prairie fen area. Similarly, surrounding natural land cover exhibited a positive association with suitability index values.

Suitability index values exceed the 99% LPT when at least 30% (92 ha) of the 1,000 m neighborhood was covered with natural land cover. Surrounding agricultural land cover exhibited a peak near 14% (43 ha) with greater agricultural land cover decreasing suitability. Surrounding open water values most associated with Poweshiek skipperling habitat suitability were between 3 and 10% (10 and 30 ha) of the surrounding neighborhood. Road density was negatively associated with habitat suitability and decreased logarithmically. Suitable habitat was found at locations with less than 3.4 km/km² roads. Surrounding developed land cover was negatively associated with habitat suitability and decreased logarithmically. Suitability less than the 99% LPT was observed when developed land was greater than 33% (105 ha) of the surrounding neighborhood.

Our survey findings are consistent with the sharp declines in abundance and distribution that have been documented across the range of the species over the past 20 y (Selby 2005; Poweshiek Skipperling Working Group 2011). Habitat modeling can be a useful tool in the initial prioritization of locations to include in inventory and monitoring efforts (Elith et al. 2011). We identified 26 prairie fens with no record of Poweshiek skipperling presence with a maximum suitability index value greater than the 99% LPT (0.074). Recent survey efforts have detected four extant localities in Michigan. However, if more occupied sites were identified, researchers could incorporate these sites into ongoing studies to better understand what factors are affecting Poweshiek skipperling. Repeated surveys of suitable unoccupied habitat would provide data for other, absence data-dependent, analyses and possibly identify localities for introduction. If novel populations were discovered, then we could obtain a more complete understanding of the habitat requirements for this species and we could refine the habitat model.

Our habitat model suggests that habitat suitability increases with prairie fen area. There is a notable spike around 10 ha, which is likely due to one small fen that still sustains Poweshiek. The decline in habitat suitability near the prairie fen area maximum is possibly the result of the largest fen in the state having no record of presence. The response curve for natural land cover when incorporated with all other variables in the model reveals some overfitting due to the small number of populations, resulting in oscillation patterns. When examining the response curves built using natural land cover as the only predictor variable there is a generally constant positive association in the model, which suggests that Poweshiek skipperling populations have persisted in more natural landscapes. The relationship between agricultural land cover and Poweshiek skipperling presence exhibits some overfitting of the model, but when examining the response curve for surrounding agriculture as the only predictor variable, there is generally a negative association. However, there is an increase in suitability initially as surrounding agricultural land cover increases from zero (Figure 3). This portion of the relationship is likely not causal, but instead is the result of Michigan's agricultural–forest–wetland matrix landscape (Davis et al. 2007). The location of this species in the matrix landscape possibly increases exposure to pesticides from nearby fields. Recent evidence suggests that neonicotinoid pesticides have contributed to declines in bees and birds (Godfray et al. 2014). These highly effective, nonselective pesticides could be a factor contributing to the precipitous decline of this species. The relationship between Poweshiek skipperling habitat suitability and surrounding open water is possibly due to the associated lakes adjacent to or within many prairie fens. However, this relationship diminishes as the area of open water reaches approximately 0.8 ha. The correlation with surrounding open water suggests that prairie fens with associated lakes might be more suitable for Poweshiek skipperling habitat, which could possibly be correlated with specific hydrogeology. Prairie fen type (e.g., mound fen, slope fen, lake-edge fen) should be further explored by researchers to understand this possible relationship. Road density is negatively associated with suitability index for Poweshiek skipperling. Road density serves as a proxy of not only development, but also potential hydrologic alterations that can affect prairie fen communities. This suggests that increased anthropogenic disturbance decreases the suitability for this species, and is commensurate with the aforementioned relationship between surrounding natural land cover and suitability index. Surrounding developed land cover was also negatively associated with suitability index. This is logical and follows with the previously stated relationships regarding amount of surrounding natural land cover and road density (Figure 3).

Our model suggests some landscape-level environmental factors could influence Poweshiek skipperling habitat, and these relationships support a more rigorous exploration of local and landscape-level factors using data collected in the field. Based on our findings, we suggest that future surveys include prairie fen locations identified as potentially having suitable habitat. These surveys may find new populations or provide further information about the habitat requirements of the Poweshiek skipperling.

The funding for this project was provided by the USFWS, Central Michigan University Biology Department, and Sigma Xi. We would like to especially acknowledge Phil Delphey with USFWS for his active interest and support of this project. We would like to thank Central Michigan University student Riley A. Zionce for his invaluable assistance with field work. We also thank our colleagues and partners who assisted with surveys, monitoring, data analysis, project oversight and support, and manuscript preparation, including Rachel Hackett (Central Michigan University), Chris May (The Nature Conservancy), Barbara Hosler and Tameka Dandridge (USFWS East Lansing Field Office), and Pete Badra, Selena Creed, and Henry Pointon (Michigan Natural Features Inventory). Additionally, we express deep gratitude to the Associate Editor, A. Townsend Peterson, Robert Dana, and the anonymous reviewers for their constructive comments. Finally, we would like to thank the landowners for allowing us access to their properties for surveys and monitoring activities, including, but not limited to, the Michigan chapter of The Nature Conservancy; Michigan Nature Association; Oakland County, Springfield Township; the State of Michigan, Department of Natural Resources; and numerous private landowners. This paper is Contribution Number 69 of the Central Michigan University Institute for Great Lakes Research.

Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding authors for the article.

Reference S1. Borkin SS. 1995. 1994 Ecological studies of the Poweshiek skipper (Oarisma poweshiek) in Wisconsin. Milwaukee Public Museum, Milwaukee, Wisconsin.

Found at DOI: http://dx.doi.org/10.3996/052015-JFWM-049.S1 (1.4 MB PDF).

Reference S2. Cuthrell DL, Slaughter SL. 2012. Special animal abstract for Oarisma poweshiek (Poweshiek skipperling). Michigan Natural Features Inventory, Lansing, Michigan.

Found at DOI: http://dx.doi.org/10.3996/052015-JFWM-049.S2; also available at http://mnfi.anr.msu.edu/abstracts/zoology/Oarisma_poweshiek.pdf (776 KB PDF).

Reference S3. Hackett, RA. 2013. Prairie fen plant biodiversity: the influence of site- and landscape- level factors on plant diversity. Master's thesis. Mount Pleasant: Central Michigan University.

Found at DOI: http://dx.doi.org/10.3996/052015-JFWM-049.S3 (2.3 MB PDF).

Reference S4. Hoving CL, Lee YM, Badra PJ, Klatt BJ. 2013. Changing climate, changing wildlife - a vulnerability assessment of 400 species of greatest conservation need and game species in Michigan. Michigan Department of Natural Resources Wildlife Division Report No. 3564.

Found at DOI: http://dx.doi.org/10.3996/052015-JFWM-049.S4; also available at https://www.michigan.gov/documents/dnr/3564_Climate_Vulnerability_Division_Report_4.24.13_418644_7.pdf (1.2 MB PDF).

Reference S5. Kost MA, Albert DA, Cohen JG, Slaughter BS, Schillo RK, Weber CR, Chapman KA. 2007. Natural communities of Michigan: classification and eescription. Michigan Natural Features Inventory, Report No. 2007-21, Lansing, Michigan.

Found at DOI: http://dx.doi.org/10.3996/052015-JFWM-049.S5; also available at http://mnfi.anr.msu.edu/reports/2007-21_Natural_Communites_of_Michigan_Classification_and_Description.pdf (1.9 MB PDF).

Reference S6. Poweshiek Working Group. 2011. Minutes from the Poweshiek skipperling workshop, March 24 and 25, Winnipeg, Manitoba. Edited by J. Dupont.

Found at DOI: http://dx.doi.org/10.3996/052015-JFWM-049.S6; also available at http://www.poweshiekskipper.org/project/Images/Final_POSK_Workshop_2011.pdf (1.9 MB PDF).

Reference S7. [SARA] Species at Risk Act, Revised Statutes of Canada (2002, c. 29)

Found at DOI: http://dx.doi.org/10.3996/052015-JFWM-049.S7; also available at http://laws-lois.justice.gc.ca/PDF/S-15.3.pdf (939 KB PDF).

Reference S8. Selby G. 2005. Status assessment and conservation guidelines: poweshiek skipperling (Oarisma poweshiek (Parker) (Lepidoptera: Hesperiidae). Prepared for Twin Cities Field Office, U.S. Fish and Wildlife Service, Bloomington, Minnesota.

Found at DOI: http://dx.doi.org/10.3996/052015-JFWM-049.S8; also available at http://www.fws.gov/midwest/es/soc/insects/posk_sa.pdf (6 MB PDF).

Reference S9. Spieles JB, Comer PJ, Albert DA, Kost MA. 1999. Natural community abstract for prairie fen. Michigan Natural Features Inventory, Lansing, Michigan. (Found at DOI: http://dx.doi.org/10.3996/052015-JFWM-049.S9; also available at http://mnfi.anr.msu.edu/abstracts/ecology/prairie_fen.pdf (2.9 MB PDF).