Regular issues
Special Issues

Prairie Naturalist
    PRNA Home
    Aim and Scope
    Board of Editors
    Editorial Workflow
    Publication Charges

Other Eagle Hill Journals
    Northeastern Naturalist
    Southeastern Naturalist
    Caribbean Naturalist
    Neotropical Naturalist
    Urban Naturalist
    Eastern Paleontologist
    Journal of the North Atlantic

Eagle Hill Institute Home

Subecoregional Influence on Swift Fox Habitat Suitability

Sarah K. Olimb1*, Donelle L. Schwalm2, and Kristy L.S. Bly1

1Northern Great Plains Program, World Wildlife Fund, 13 S. Willson, Suite 1, Bozeman, MT 59715, USA. 2Department of Biology, University of Maine-Farmington, 173 High Street, Farmington, ME 04938, USA. *Corresponding author.

Praire Naturalist, Volume 53 (2021):1–15

Grassland-dependent Vulpes velox Say (Swift Foxes) occupy only a portion of their historical range in the North American Great Plains and remaining subpopulations are functionally disconnected due to habitat fragmentation and other barriers. Modeling habitat suitability is critical to identifying sites for habitat conservation and reintroduction and increasing subpopulation connectivity. We used mixed-effects modeling to simultaneously evaluate Swift Fox presence against habitat variables and subecoregional location. Our results show that habitat suitability is dependent on geographic location. Individual subecoregional models were each influenced by land cover and soil composition but varied by dominant soil component and correlation with surrounding land uses. These findings prioritize localized data for species management and predict how changing landscape composition may impact Swift Fox distribution.

pdf iconDownload Full-text pdf



Site by Bennett Web & Design Co.
Prairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 1 2021 PRAIRIE NATURALIST 53:1–15 Subecoregional Influence on Swift Fox Habitat Suitability Sarah K. Olimb1*, Donelle L. Schwalm2, and Kristy L.S. Bly1 Abstract - Grassland-dependent Vulpes velox Say (Swift Foxes) occupy only a portion of their historical range in the North American Great Plains and remaining subpopulations are functionally disconnected due to habitat fragmentation and other barriers. Modeling habitat suitability is critical to identifying sites for habitat conservation and reintroduction and increasing subpopulation connectivity. We used mixed-effects modeling to simultaneously evaluate Swift Fox presence against habitat variables and subecoregional location. Our results show that habitat suitability is dependent on geographic location. Individual subecoregional models were each influenced by land cover and soil composition but varied by dominant soil component and correlation with surrounding land uses. These findings prioritize localized data for species management and predict how changing landscape composition may impact Swift Fox distribution. Introduction Once abundant across the North American Great Plains, Vulpes velox Say (Swift Fox) populations dwindled in the early 1900s to near extinction due to habitat loss, rodent extirpation campaigns aimed at prey species such as Cynomys spp. (Prairie Dogs) and Marmota spp. (Ground Squirrels), and predator eradication policies (Allardyce and Sovada 2003, Cutter 1958, Egoscue 1979, Kilgore 1969). Reintroduction efforts have helped re-establish Swift Foxes to some areas where they were extirpated, although some reestablished populations remain structurally and functionally disconnected from each other and the core distribution of the species due to habitat fragmentation and other barriers to connectivity and range expansion (Alexander et al. 2016, Schwalm et al. 2014). In the northern portion of their range, distinct populations exist in southern Saskatchewan, Canada, and northern Montana, the western Dakotas, and Wyoming; connecting these populations likely will require coordination among government and private stakeholders (Alexander et al. 2016). It will also require site specific habitat assessments to ascertain suitability and address limitations to potential movement corridors. Multiple efforts have been made to identify the best habitat in which to restore Swift Fox populations and to otherwise facilitate connectivity between populations in the northern extent of the species’ historical range. Montana Fish, Wildlife and Parks (2011) used Maximum Entropy (MaxEnt) modeling to predict likely core habitat based on physical (habitat) characteristics and life-history information, such as minimum breeding population size. Olimb et al. (2009) used an expert-opinion based Analytical Hierarchy Process (AHP) to relate physical habitat factors (e.g., soil characteristics, distance to water, land cover) and constraints (e.g., slope, distance to roads, and crop density) to likely occupied habitat. In their southern Alberta and Saskatchewan study area, Moehrenschlager et al. (2006) used live-trap data and a suite of habitat variables in a multi-scale analysis to predict Swift Fox habitat suitability. The habitat variables found to be significant were later used by Olimb et al. (2010) and Alexander et al. (2016) to extrapolate the habitat suitability index (HSI) to eastern Montana and portions of surrounding states (North Dakota, South Dakota, 1Northern Great Plains Program, World Wildlife Fund, 13 S. Willson, Suite 1, Bozeman, MT 59715, USA. 2Department of Biology, University of Maine-Farmington, 173 High Street, Farmington, ME 04938, USA. *Corresponding author: Manuscript Editor: M. Colter Chitwood Prairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 2 and Wyoming). Most recently, the Montana Natural Heritage Program (2016) used both inductive and deductive modeling techniques to predict likely distribution and identify likely habitat, respectively, for Swift Foxes in Montana. In their inductive approach, they used MaxEnt with 19 state-wide biotic and abiotic layers as inputs, evaluating final outputs with an absolute validation index (AVI) and deviance. The deductive approach used location data to quantify occurrence by type of land cover classification (e.g., mixed grass prairie) to demonstrate which habitats were most often associated with Swift Fox presence. In previous Swift Fox habitat modeling, the focus has either been a restricted, localized area (e.g., Moehrenschlager et. al. 2006) or at the state-wide scale where habitat characteristics are combined across populations that are geographically distinct. Though the Montana Natural Heritage Program’s (2016) deductive analysis used Level 3 ecological systems from the Montana Land Use/Land Cover Data (which assigns a category to each unique land cover within the state) to inform habitat suitability, no additional ecosystem location variables were used in conjunction with the land cover classification. To our knowledge, no study has investigated whether or not the influence of habitat is variable based on geographic location. Pooling individuals across disconnected populations assumes that available resources, and the average response of individuals among populations to those resources, are constant (Gillies et al. 2006). Thus, we pose the question: Is it appropriate to use the same variables to predict suitable habitat for geographically separated populations of Swift Foxes? Existing evidence indicates that Swift Foxes exhibit variable habitat associations across their distribution. For example, Kamler et al. (2003) and Finley et al. (2005) found that Swift Fox occurrence was negatively associated with agricultural development, while Sovada et al. (2001), Matlack et al. (2000), and Kilgore (1969) observed frequent use of agricultural fields by Swift Foxes. Similarly, in the Northern Great Plains, Swift Foxes are commonly associated with Prairie Dog colonies (Kotliar et al. 1999), whereas evidence of negative association has been documented in the southern Great Plains (Nicholson et al. 2006) and Swift Foxes are known to persist in areas outside of the historical distribution of Prairie Dogs (e.g., Blackfeet Nation; Ausband and Foresman 2007a). While Swift Foxes are known to occupy sagebrush steppe in Wyoming (Olson and Lindzey 2002a) and Montana (Moehrenschager et al. 2006), the proportion of sagebrush in the landscape is negatively associated with inter-population connectivity (Schwalm 2012). In general, habitat qualities that appear to be universal across the species distribution regardless of vegetation type include height less than 30 cm and low to gently rolling topography (Kilgore 1969, Meyer 2009), as this facilitates Swift Fox detection of predators such as Canis latrans Say (Coyotes; Sovada et al. 1998). Incorporating site- or region-specific habitat variables for geographically separated populations of a species has been effectively demonstrated for other species including Centrocercus urophasianus Bonaparte (Greater Sage-grouse; Doherty et al. 2016), Lynx canadensis Kerr (Canada Lynx; Hornseth et al. 2014, Vashon et al. 2008), and Lemur catta Linnaeus (Ring-tailed Lemur; Cameron 2010). We hypothesized that disconnected populations of Swift Foxes would also respond to varying habitat characteristics or that certain variables would have a greater or lesser influence in habitat selection in separated geographic areas. Our objectives through this process were two-fold: 1) to evaluate the role of subecoregional location on Swift Fox habitat preferences using mixed-effects modeling (MEM); and 2) to use this information to apply a model (or models) of appropriate and accurate geographic scale to predict Swift Fox habitat, thereby improving future management, reintroduction, and connectivity efforts. Prairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 3 Materials and Methods Study Area The study encompassed approximately 400,615 km2, including 38 counties in Montana, 13 counties in North Dakota, seven counties in South Dakota, and five counties in Wyoming (Fig. 1). The area was primarily within the Level II West Central Semi-Arid Prairies as defined by the Commission for Environmental Cooperation (CEC 2013). This area is a portion of the historical distribution of the Swift Foxes in the Northern Great Plains where population recovery has been limited and where Swift Fox populations are small and separated from each other by large expanses of habitat of unknown suitability (Fig. 1). We divided the study area into northern and southern subecoregion parts based on North American Terrestrial Ecoregions—Level III data from the CEC (Wiken et al. 2011; Fig. 1). The northern subecoregion was relatively dry (250–550 mm annual precipitation) and defined by cold winters and warm summers (mean annual temperature 2.5–7° C; CEC 2013). Figure 1. Vulpes velox Say (Swift Fox) habitat suitability study area for the central Northern Great Plains divided into northern and southern subecoregional subsections. Presence points used to inform the mixedeffects and individual subecoregional habitat suitability models are shown. Background data (shaded area) show the historical distribution of Swift Foxes (Sovada et al. 2009). Inset map shows the relationship of the study area to the full Northern Great Plains ecoregion. Hill, Blaine, and Phillips Counties, Montana, which emerged from the model as areas of high habitat suitability, are highlighted in both full and inset maps. Prairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 4 Vegetation included native grasses (e.g., Bouteloua gracilis (Kunth) Lag. Ex Griffiths [Blue Grama]) and shrubs, as well as species introduced for forage (e.g., Festuca arundinacea Schreb [Tall Fescue], and Agropyron cristatum (L.) Gaertn. [Crested Wheatgrass]) and soil stabilization (e.g., Bromus inermis Leyss [Smooth Brome]). The southern subecoregion was similarly dry (250–510 mm annual precipitation) and was marked by cold winters and hot summers (mean annual temperature 5–8.5° C; CEC 2013). Vegetation composition included a similar suite of grass and shrub species; however, this area included a greater percentage of sagebrush (e.g., Artemisia tridentata Nutt [Wyoming Big Sagebrush]) and isolated pockets of Pinus ponderosa Douglas ex P. Lawson & C. Lawson (Ponderosa Pine) and Juniperus scopulorum Sarg. (Rocky Mountain Juniper). The land ownership of both subecoregions included a mixture of private, federal, state, and tribal holdings with limited cropland; livestock grazing ranked as the most common land use activity (W iken et al. 2011). Modeling We compiled Swift Fox observation records within the study area from the relevant state agencies (i.e., Montana Fish, Wildlife and Parks; Montana Natural Heritage Program; North Dakota Game and Fish Department; South Dakota Department of Game, Fish and Parks; Wyoming Game and Fish) tasked with monitoring Swift Fox distribution in their jurisdiction, along with recent and current studies (Mitchell 2018, Schwalm et al. 2014). We favored high integrity data and removed points with low location (>1 km) and temporal (>18 years, when reported) precision due to the observed practice of “estimating” imprecise locations and dates at these values or greater. This resulted in 416 points (300 in northern subecoregion, 116 in southern subecoregion). We created a used/available dataset by augmenting observation points with an equal number of naïve points randomly generated using ArcGIS, excluding uninhabitable areas such as open water (ESRI 2015) and area within 1 km of a used point. Input habitat variables for regression analysis were derived from variables in Alexander et al. (2016) and additional variables identified as important by the relevant literature (variable identity and sources listed in Table 1). All variables, except for topographic complexity variables, which were derived from 1–km source data and downscaled to 90 m, had resolutions of 100 m or finer; hence, data were generalized to 100-m pixel size, a scale justified for Swift Fox habitat analysis by Russell (2006) and Alexander et al. (2016). We checked for collinearity of variables using Variance Inflation Factors (R Development Core Team 2013, package HH). Percent grass and percent crop were collinear; therefore, we ran individual regressions for these two variables as well as the two measures of topographic complexity (Terrain Roughness and Surface Relief Ratio [SRR]), which had known correlation issues (D. Schwalm, University of Maine-Farmington, Farmington, ME, unpubl. data). Percent grass and SRR were selected as the more suitable inputs for the regression analysis based on greater R2 values, and percent crop and terrain roughness were excluded from potential models. To test for the significance of location as a grouping factor, we first built individual logistic regression models, one with and one without subecoregion as a random effect, to evaluate the effects of habitat variables on Swift Fox presence. We scaled the numeric parameters (n = 14) to account for differences in the order of magnitude of the input variables (R Development Core Team 2013). For the model including subecoregion, we used a mixed-effects logistic regression model in R packages lme4 and lmerTest (Bates et al. 2010, Kuznetsova 2017, R Development Core Team 2013). Because the mixed-effects model showed that subecoregion did affect model performance, we built individual models to represent the northern and southern geographies (Fig. 1) using Prairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 5 Table 1. Input variables for Vulpes velox Say (Swift Fox) habitat suitability indices considered in central Northern Great Plains mixed-effects model and individual subecoregional (northern and southern) logistic regressions. Expanded from Alexander et al. (2016) using variables deemed important to Swift Fox distribution in relevant literature. Variable Description Source Reference Brightnessa Measure of soil reflectance 2010 ESRI Tasseled Cap 2010 Global Land Survey (GLS) datasets Crop density Percentage of crop cover in 80 surrounding 100 m cells 2015 Plowprint Gage et al. 2016 Distance to nearest black-tailed Prairie Dog (Cynomys ludovicianus) colony Euclidean distance to nearest colony edge State/provincial agencies; federal agencies; natural heritage programs Montana Natural Heritage Program (2014); North Dakota Game and Fish (2013); Parks Canada (2009); South Dakota Game, Fish and Parks (2011); US Bureau of Indian Affairs (2012); Wyoming Natural Diversity Database (2011) Greennessa Measure of vegetation 2010 ESRI Tasseled Cap 2010 Global Land Survey (GLS) datasets Land Capability Classification (LCC) Classification of soils based on ability to support agriculture and other vegetation STATSGO2 General Soil Map NRCS Soil Survey (2013) Road density Kilometers of roads in a 1-km search radius National roads networks Canada Road Network 2016; TIGER 2016 Soil composition: sand, silt, clay Percentage of each soil component STATSGO2 General Soil Map NRCS Soil Survey (2013) Surface Relief Ratio (SRR) b Measure of topographic complexity, calculated at 1-km resolution and downscaled to 90 m 2012 Digital Elevation Model The National Mapc Surrounding landcover composition: grass, forest, scrub, cropb, d Percent of each land cover type in surrounding 1-km circular window 2011 National Landcover Database Homer et al. 2015 Terrain roughnessb, d Measure of topographic complexity, calculated at 1 km resolution and downscaled to 90 m 2012 Digital Elevation Model The National Mapc Wetnessa Measure of moisture content in cover vegetation and soil 2010 ESRI Tasseled Cap 2010 Global Land Survey (GLS) datasets a Physical characteristics of the landscape derived via tasseled cap transformation of remote sensing data used widely in ecology and agriculture applications, e.g., Sheng et al. (2011). bCalculated using the Geomorphometry and Gradient Metrics ArcToolbox add on toolbox (Evans et al. 2014). cAvailable online at . dPercent crop and Terrain Roughness excluded from models due to collinearity with other variables. Prairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 6 the Classification and Regression Training (caret) package in R (Kuhn 2018, R Development Core Team 2013). For each geography, we randomly split the data into training (80%) and test (20%) subsets. The test subset was reserved for post-analysis evaluation. Using the training data, we developed global logistic regression models for each subecoregion using used/ available points as the dependent variable and the suite of habitat variables as the independent variables (Table 1). We used R package glmulti to derive an exhaustive set of additive models (i.e., all independent variable combinations were considered) and used AIC values to select the best model for each geography (Calcagno and de Mazancourt 2010, R Development Core Team 2013). Though it is common in these analyses to develop an a priori model representing hypothesized relationships between presence and predictors, it was unnecessary in this scenario because our variables were based on a previously published model of Swift Fox habitat preference (i.e., Alexander et al. 2016). We used McFadden’s R2, a common pseudo-R2 value used to assess logistic regression models, to evaluate the fit of the models. Here, we present the results of the top subecoregion models (northern and southern) along with the parameters of the reduced mixed-effects model (including only variables with P < 0.05) for comparison (Table 2). To display the models spatially, we scaled input variables to match the data used in the statistical models and applied the models derived for the northern and southern subecoregions. Both models were rescaled to a 0 (low quality or low preference habitat) to 255 (high quality or high preference habitat) scale. We used Jenk’s Natural Breaks to divide each model into five classes ranging from low to very high suitability and used these classes to evaluate classification accuracy of the test data points reserved for mod el evaluation (ESRI 2015). Results A comparison of the mixed-effects model including subecoregion as a random factor and the null model without subecoregion indicated that geographic location is important in determining habitat suitability (χ2 1 = 29.42, P ≤ 0.001). Thus, after splitting the data into subecoregion subsets, our analysis produced two distinct habitat suitability models, one for the northern and one for the southern subecoregion of the study area. The model for the northern subecoregion was driven by the following variables (positive or negative association denoted in parenthesis): Land Capability Classification (LCC) (-), percentage of clay (+) in soil, percentage of grass (+), road density (+), brightness (+) and wetness (-). The model for the southern subecoregion of the study included distance to Prairie Dog colony (-), percentage of sand in soil (+), percent forest (-), road density (+), brightness (+) and wetness (-) (Table 2). While McFadden’s R2 showed that our northern model had greater explanatory power (0.53) compared to the southern model (0.26), values for both models were indicative of high model fit. Unlike standard R2 values, McFadden’s R2 values tend to be much lower; values of 0.2–0.4 are considered an excellent fit and have been compared to values of 0.7 to 0.9 for a standard measurement of a linear function (Louviere et al 2000, McFadden 1974, McFadden 1979). Evaluation of test points showed that both models performed very well, with 98.3% (n = 58) and 100% (n = 16) of positive occurrence test points (i.e., actual Swift Fox locations) for the northern and southern models, respectively, in the highest suitability class. Both models included predicted habitat quality ranging from high to low. In the northern subecoregion, highest quality habitat was concentrated in Hill, Blaine, and Phillips counties of Montana (Figs. 1 and 2). Lowest quality habitat was predicted on the western edge of the northern subecoregion where the terrain shifts to the foothills of the northern Rocky Mountains and surrounding areas of open water such as the Missouri River (Fig. 2). The range of habitat quality in the northern subecoregion was more distinct with representaPrairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 7 tion of both extremes (i.e., very high quality and very low quality habitat). Contrarily, the southern subecoregion, though including large contiguous areas of suitable habitat, lacked the concentrated areas of very high quality habitat found in the north (Fig. 2). Similar to the north, the southern subecoregion also had areas of very low quality habitat in areas of hilly terrain and forest cover (e.g., the Black Hills of South Dakota and the Bighorn National Forest in north-central Wyoming; Fig. 2). Discussion The northern subecoregion model in this analysis had greater explanatory power; however, it was roughly half the size of the southern study area with a similar number of sample points. The predicted area of highest quality habitat was concentrated in the central part of the region Table 2. Output of Vulpes velox Say (Swift Fox) habitat suitability analysis for northern and southern subecoregional (logistic regression models) and full central Northern Great Plains (mixed-effects model) locations. Statistics for covariates are shown for variables included in the reduced subecoregional models. Numeric parameters have been scaled to account for differences in the order of magnitude of the input variables. Only coefficients with significance at P < 0.05 (for at least one model) are included. Habitat variable Northern Southern Mixed-effects model β (SE) χ2 P β (SE) χ2 P β (SE) P Intercept -0.01 (0.31) - 0.99 -2.12 (0.41) - <0.01 -0.28 (0.49) 0.57 Brightness 1.71 (0.36) 23.06 <0.01 0.53 (0.17) 10.21 <0.01 0.80 (0.13) <0.01 Distance to nearest blacktailed Prairie Dog (Cynomys ludovicianus) colony - - - -1.82 (0.52) 12.32 <0.01 - - Land Capability Classification (LCC) -1.51 (0.36) 17.81 <0.01 -0.53 (0.13) <0.01 Road density 0.84 (0.35) 5.80 0.02 0.27 (0.12) 4.71 0.03 0.67 (0.13) <0.01 Soil composition: sand - - - 0.29 (0.11) 6.41 0.01 0.40 (0.13) <0.01 Soil composition: clay 0.68 (0.30) 5.02 0.03 - - - 0.24 (0.14) 0.07 Surrounding landcover composition: grass 1.66 (0.25) 45.08 <0.01 0.94 (0.13) <0.01 Surrounding landcover composition: forest - - - -1.26 (0.52) 5.80 0.02 -1.10 (0.51) 0.03 Surrounding landcover composition: scrub - - - - - - 0.39 (0.13) <0.01 Wetness -0.67 (0.31) 4.73 0.03 -0.77 (0.20) 14.82 <0.001 -0.63 (0.14) <0.01 Overall model fit McFadden’s R2 0.53 0.26 - Prairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 8 in the area with the most Swift Fox location data. Thus, the predicted high quality of this section likely was due to a combination of actual suitability and a saturation of data points. Brightness, wetness, and road density were the only variables that occurred in both individual subecoregion models. Brightness is a measure of surface reflectance; increasing values indicate reduced vegetation cover or more exposed soils (including agricultural areas). Wetness values reflect the amount of moisture on the surface and in the soil; thus, increasing values indicate wetter conditions. Both models had a positive relationship to brightness and negative relationship to wetness, indicating that Swift Foxes in both geographies show preference to areas with sparse vegetation cover and relatively drier soils. This supports prior range-wide research showing Swift Foxes prefer areas of less dense or shorter vegetation and tend to den on well-drained soils (Meyer 2009). Both models also showed a positive association with road density. This may be an artifact of data collection techniques. Though many of the data points were collected in studies designed to take into account location bias, not all data contain detailed collection information, and 25% or more of points from natural heritage programs and state agencies are reported sightings, either roadkill or during travel, potentially biasing the estimate of road importance. Conversely, many studies have demonstrated a positive link between roads and Swift Fox presence (Harrison 2003, Hines and Case 1991, Olson 2000, Pruss 1999, Russell Figure 2. Habitat Suitability Index (HSI) models for Vulpes velox Say (Swift Fox) derived for northern and southern subecoregions of our central Northern Great Plains study area, shown separated by dashed line. Habitat quality scale ranges from high quality habitat (255; black) to low quality habitat (0; white). Prairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 9 2006, Sasmal et al. 2011). Proposed explanations for this relationship include reduced risk of Coyote predation (Pruss 1999, Russell 2006, Sasmal et al. 2011), movement corridors (Hines and Case 1991, Pruss 1999), and food subsidies in the form of vehicle-killed carcasses and increased small mammal densities in roadside ditches (Hines and Case 1991, Klausz 1997). Thus, although it is possible that the connection between Swift Fox presence and transportation routes is overemphasized in our models, it is also likely the positive relationship observed represents an ecological response. Geographic Variability The models also indicate the importance of geographic variation in the factors that drive Swift Fox habitat suitability in the two subecoregions. For example, the northern model showed a positive association with percentage of clay in the soil and amount of grassland habitat and a negative association with Land Capability Classification (LCC). The LCC scale is an assessment of physical land characteristics in respect to agricultural suitability: Lower values are prime for row crop agriculture, while greater values are limiting to cultivation (Klingebiel and Montgomery 1961). It is logical that the same requirements for agriculture can also apply to Swift Fox habitat selection: The lower classes include level or gently sloping topography with little standing water and deeper soils appropriate for denning. The greater classes often include steep slopes, high probability of erosion, thin soils, and standing water, all of which are characteristics generally not associated with prime Swift Fox habitat. This association implies that there is reasonable potential for future agricultural development to further influence availability of suitable habitat for Swift Foxes. The positive association between Swift Fox presence and percent of grass in the landscape implies that agricultural development may reduce habitat suitability in the northern subecoregion. Given that Swift Fox response to agricultural development varies across their distribution (Finley et al. 2005, Kamler et al. 2003, Matlack et al. 2000), improving understanding of the effect of cropland on Swift Foxes in this region is necessary to prepare for potential future agricultural expansion in the Northern Great Plains (Sohl et al. 2012). The southern model also included habitat and soil factors, but they indicated a negative association with forests and positive association with percentage of sand in the soil. Additionally, the southern model included a negative association with increasing distance to Prairie Dog colonies, resulting in a positive association between Swift Foxes and Prairie Dog colony proximity. Swift Foxes often are associated with Prairie Dog colonies in South Dakota and Oklahoma (Lomolino and Smith 2003, Sharps and Uresk 1990), and Sasmal et al. (2011) reported that female Swift Foxes at Badlands National Park in South Dakota used Prairie Dog colonies in proportion to their availability. Prairie dogs (and other small mammals) serve as prey for Swift Foxes, their burrows provide shelter and dens in which to raise young, and their grazing maintains short-statured vegetation—the preferred habitat of Swift Foxes (Carbyn 1998, Kilgore 1969, Uresk and Sharps 1986). The reduction in Prairie Dog colony abundance and distribution from sylvatic plague outbreaks, recreational Prairie Dog shooting, and poisoning may limit the model’s ability to reflect the importance of Prairie Dogs to Swift Foxes. Conversely, there have been multiple studies confirming a negative relationship between the Swift Foxes and the presence of Prairie Dog colonies, more commonly in the southern Great Plains. For example, a three-year study in Texas found significantly fewer Swift Foxes near Prairie Dog colonies (Nicholson et al. 2006). Also, in Oklahoma, Swift Foxes were detected less frequently in sites near Prairie Dog colonies (Shaughnessy and Cifelli 2004). It has been suggested that poisoning of Prairie Dogs and the presence of additional carnivores, such as coyotes, may influence Swift Fox occurrence Prairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 10 and density in these areas (Stephens and Anderson 2005). Potentially, differences in the relationship between Swift Foxes and Prairie Dog colonies in the Northern Great Plains (+) and the southern Great Plains (-) may also be driven by variation in precipitation and vegetation height in the dominant grassland community (mixed grass prairie and shortgrass prairie, respectively), although this relationship has not been formally explored. Also, it is worth noting that the northern ecoregion represents the northern bound of the black-tailed Prairie Dog range. In some areas within this ecoregion, the species is naturally absent and colonial Urocitellus richardsonii Sabine (Richardson’s Ground Squirrels) may serve as an important surrogate for Swift Foxes. Unfortunately, we are unaware of any Ground Squirrel dataset of the appropriate scale that could be included in the analyses presented here and suggest that this is an area where further research is warranted. Next Steps Toward Improved Understanding of Swift Fox Habitat Suitability Much of the current understanding of Swift Fox habitat is derived from studies in the central and southern portion of the species distribution (e.g., Cutter 1958; Finley et al. 2005; Kamler et al. 2003; Kilgore 1969; Matlack et al. 2000; Nicholson et al. 2006, 2007; Schauster et al. 2002; but see Moehrenschlager et al. 2006). For Swift Foxes in the Northern Great Plains, there remain several key data needs that could improve understanding of Swift Fox habitat requirements in the region considerably. First, the addition of an Artemisia spp. (Sagebrush) habitat classification layer (Homer et al. 2009) could provide a more accurate representation of suitable Swift Fox habitat; unfortunately, a Sagebrush layer was not available for this area at the time of the analysis. Swift Foxes are known to occur in sagebrush steppe in Wyoming (Olson and Lindzey 2002a, b) but it remains unknown if Sagebrush communities in Montana can support robust Swift Fox populations. Second, although our model captures broad scale land-use patterns, there are several other anthropogenic factors that may influence Swift Fox occurrence but are not well documented in terms of frequency, intensity, or spatial distribution of impact. These human-induced factors include Coyote control, Prairie Dog poisoning and shooting, and predator trapping. We argue that understanding contemporary habitat suitability and its drivers, including anthropogenic influences, is a critical step for informing the conservation of Swift Foxes and identifying what conditions support Swift Fox occurrence in human-altered landscapes. Due to the scale at which some predictor variables were originally calculated (e.g., topography complexity variables), our model does not necessarily reflect fine-scale habitat characteristics at the sub-home range scale. Exploring fine-scale habitat selection by Swift Foxes, perhaps using Global Positioning System collars, could elucidate additional factors that influence habitat suitability. Two such factors are prey availability and the influence of sympatric canids. Several studies have demonstrated that prey availability is not positively associated with Swift Fox density or survival; rather, these factors appear to be driven by Coyote abundance (Gese and Thompson 2014, Thompson and Gese 2007). Fine-scale territorial overlap and resource partitioning between Coyotes and Swift Foxes could be critical determinants of Swift Fox occurrence in the study area. The impact of these interactions, and of intensive Coyote control in parts of the study area, are important considerations but cannot be explored with the data available. Notably, the studies by Thompson and Gese (2007) and Gese and Thompson (2014) occurred where colonial rodents, such as Blacktailed Prairie Dogs or Richardson’s Ground Squirrels were uncommon or absent. Colonial rodents, such as Prairie Dogs, offer a concentrated prey resource, shelter, and reduced vegetation structure (Hoogland 1995, Kotliar et al. 2005, Miller et al. 1994) but also increased Prairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 11 Coyote activity (Eads et al. 2015, Lomolino and Smith 2003). As discussed earlier, the relationship between Swift Foxes and colonial rodents is variable, and ultimately, poorly understood. We suggest that collecting systematic fine-scale data that includes Coyote presence and activity, Coyote control, small mammal community density and distribution, and colonial small mammal presence and control efforts is critical for informing Swift Fox habitat suitability. Management Implications Our results illustrate the complex, geographically influenced patterns of habitat suitability exhibited by Swift Foxes. These results suggest emphasizing different habitat characteristics when targeting habitat conservation or restoration for Swift Foxes between the two subecoregions. They further support the necessity of considering regional influence of habitat composition and use when identifying suitable habitats for conservation efforts of species more generally. Models for the northern and southern subecoregions predict there are large areas of suitable habitat where Swift Foxes are currently absent or undetected, implying considerable potential for Swift Fox restoration efforts. Additionally, these areas represent important targets for habitat conservation, restoration, and subsequent conservation of active Prairie Dog colonies, reintroducing disturbance and nutrient cycling regimes via prescribed fire, and grazing by livestock to facilitate short-structured vegetation. Additional work is warranted to explore why natural recovery has been delayed in these areas. Potential explanations include distance from the source population to suitable habitat, limited northward dispersal from the source population, poor survival or reproduction following dispersal, effects of interspecific competition by Coyotes or other species, site-specific variation in habitat suitability not captured by these models, or additional barriers to connectivity not captured through habitat modeling. For these reasons, we recommend ground-truthing local habitat suitability prior to considering any reintroduction effort. Acknowledgements The authors wish to thank D. Jorgensen and B. Skone for comments regarding the study design or manuscript. Funding was provided by World Wildlife Fund (WWF). Additional funding for and data from Swift Fox surveys used in this study was provided by: National Fish and Wildlife Foundation, project ID 0103.14.045477; Blackfeet Fish and Wildlife Department; Crow Fish and Game Department; Defenders of Wildlife; Department of Fisheries and Wildlife, Oregon State University; Fort Belknap Fish and Wildlife; Fort Peck Assiniboine and Sioux Tribes; Office of Natural Resources, Montana Fish, Wildlife and Parks; Northern Cheyenne Tribe, Department of Environmental Protection and Natural Resources; Emily Mitchell and South Dakota State University; and Wyoming Game and Fish Department. Literature Cited Allardyce, D., and M.A. Sovada. 2003. A review of the ecology, distribution, and status of Swift Foxes in the United States. Pp. 3–13, In M.A. Sovada and L.N. Carbyn (Eds.). The Swift Fox: Ecology and Conservation of Swift Foxes in a Changing World. Canadian Plains Research Center, Regina, Saskatchewan, Canada. 250 pp. Alexander, J.L., S.K. Olimb, L.S. Bly, and M. Restani. 2016. Use of least-cost path analysis to identify potential movement corridors of Swift Foxes in Montana. Journal of Mammalogy 97:891–898. Ausband, D.E., and K.R. Foresman. 2007a. Swift Fox reintroductions on the Blackfeet Indian Reservation, Montana, USA. Biological Conservation 136:423–430. Ausband, D.E., and K.R. Foresman. 2007b. Dispersal, survival, and reproduction of wild-born, yearling Swift Foxes in a reintroduced population. Canadian Journal of Zoology 85:185–189. Prairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 12 Bates, D., M. Maechler, and B. Bolker. 2010. LME4: Linear mixed-effects models using S4 classes. Available online at Accessed 22 August 2017. Calcagno, V., and C. de Mazancourt. 2010. glmulti: An R package for easy automated model selection with (generalized) linear models. Journal of Statistical Software 34. Cameron, A. 2010. Behavioural surveys and edge-sensitivity estimates of two populations of freeranging Ringtailed Lemurs (Lemur catta) in rocky outcrop/savannah mosaic habitat at Anja Special Reserve and the Tsaranoro Valley, southcentral Madagascar. Thesis, University of Victoria, Victoria, BC, Canada. 103pp. Carbyn, L.N. 1998. Update COSEWIC status report on the Swift Fox (Vulpes velox) in Canada. Report to the Committee on the Status of Endangered Wildlife in Canada. Edmonton, Alberta, Canada. 49 pp. Commission for Environmental Cooperation (CEC). Available online at Accessed 17 December 2013. Cutter, W.L. 1958. Denning of the Swift Fox in northern Texas. Journal of Mammalogy 39:70–74. Doherty, K.E., J.S. Evans, P.S. Coates, L.M. Juliusson, and B.C. Fedy. 2016. Importance of regional variation in conservation planning: A rangewide example of the Greater Sage-Grouse. Ecosphere 7:e01462. Available online at ecs2.1462. Accessed 17 April 2018. Eads, D.A., D.E. Biggins, and T.M. Livieri. 2015. Spatial and temporal use of a Prairie Dog colony by Coyotes and rabbits: Potential indirect effects on endangered Black-footed Ferrets. Journal of Zoology 296:146–152. Egoscue, H.J. 1979. Vulpes velox. Mammalian Species 122:1–5. ESRI. 2015. ArcGIS Desktop: Release 10.4. Environmental System Research Institute, Inc. Redlands, California, USA. Evans, J.S., J. Oakleaf, S.A. Cushman, and D. Theobald. 2014. An ArcGIS toolbox for surface gradient and geomorphometric modelling. Version 2.0–0. Available online at evansspatial. Accessed 4 August 2016. Finley, D.J., G.C. White, and J.P. Fitzgerald. 2005. Estimation of Swift Fox population size and occupancy rates in eastern Colorado. Journal of Wildlife Management 69:861–873. Gage, A.M., S.K. Olimb, and J. Nelson. 2016. Plowprint: Tracking cumulative cropland expansion to target grassland conservation. Great Plains Research 26:107–116. Gese, E.M., and C.M. Thompson. 2014. Does habitat heterogeneity in a multi-use landscape influence survival rates and density of a native mesocarnivore? PLoS ONE 9: e100500. doi:10.1371/ journal.pone.0100500 Gillies, C.S., M. Hebblewhite, S.E. Nielsen, and M.A. Krawchuk. 2006. Application of random effects to the study of resource selection by animals. Journal of Animal Ecology 75:887–898. Harrison, R.L. 2003. Swift Fox demography, movements, denning and diet in New Mexico. Southwestern Naturalist 48:261–273. Hines, T.D., and R.M. Case. 1991. Diet, home range, movements and activity patterns of Swift Fox in Nebraska. Prairie Naturalist 23:131–138. Hoogland, J.L. 1995. The black-tailed Prairie Dog: Social life of a burrowing mammal. University of Chicago Press, Chicago, IL, USA. 562 pp. Homer, C.G., C.L. Aldridge, D.K. Meyer, M.J. Coan, and Z.H. Bowen. 2009. Multiscale sagebrush rangeland habitat modeling in southwest Wyoming. U.S. Geological Survey open-file report 2008_1027. Reston, VA, USA. 14 pp. Homer, C.G., Dewitz, J., and L. Yang. 2015. Completion of the 2011 National Land Cover Database for the conterminous United States–representing a decade of land cover change information. Photogrammetric Engineering and Remote Sensing 81:345–354. Hornseth, M.L., A.A. Walpole, L.R. Walton, J. Bowman, J.C. Ray, M-J. Fortin, and D.L. Murray. 2014. Habitat loss, not fragmentation, drives occurrence patterns of Canada Lynx at the southern range periphery. Plos One 9:e113511. Kamler, J.F., W.B. Ballard, E.B. Fish, P.R. Lemons, K. Mote, and C.C. Perchellet. 2003. Habitat use, home ranges, and survival of Swift Foxes in a fragmented landscape: Conservation implications. Journal of Mammalogy 84:989–995. Prairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 13 Kilgore, D.L., Jr. 1969. An ecological study of the Swift Fox (Vulpes velox) in the Oklahoma panhandle. American Midland Naturalist 81:512–534. Klausz, E.E. 1997. Small mammal winter abundance and distribution in the Canadian mixed grass prairies and implications for the Swift Fox. M.Sc. Thesis. University of Alberta, Edmonton, Alberta, Canada. 108 pp. Klingebiel, A.A. and P.H. Montgomery. 1961. Land capability classification. U.S. Department of Agriculture Handbook 210, Government Printer, Washington D., USA. 25 pp. Kotliar, N.B., B.W. Baker, A.D. Whicker, and G. Plumb. 1999. A critical review of assumptions about the Prairie Dog as a keystone species. Environmental Management 24:177–192. Kotliar, N.B., B.J. Miller, R.P. Reading, and T.W. Clark. 2005. The Prairie Dog as a keystone species. Pp. 53–64, In J. Hoogland (Ed.). Conservation of the Black-tailed Prairie Dog: Saving North America’s Western Grasslands. Island Press, Washington, DC, USA. 342 pp. Kuhn, M. 2018. Package ‘caret’ for R. Available online at Accessed 22 May 2018. Kuznetsova, A., P.B. Brockhoff, and R.H.B Christensen. 2017. lmerTest: Tests in linear mixed effects models. Available online at Accessed 22 May 2018. Lomolino. M.V., and G.A. Smith. 2003. Terrestrial vertebrate communities at Black-tailed Prairie Dog (Cynomys ludovicianus) towns. Biological Conservation 115:89–100. Louviere, J.J., D.A. Hensher, and J.D. Swait. 2000. Stated choice methods: Analysis and applications. Cambridge University Press. Cambridge, England. 420 pp. Matlack, R.S., P.S. Gipson, and D.W. Kaufman. 2000. The Swift Fox in rangeland and cropland in western Kansas: Relative abundance, mortality, and body size. Southwestern Naturalist 45:221– 225. McFadden, D. 1974. Conditional logit analysis of qualitative choice behavior. Pp. 105–142, In P. Zarembka (Ed.). Frontiers in Econometrics. Academic Press. Cambridge, MA, USA. 252 pp. McFadden, D. 1979. Quantitative methods for analyzing travel behaviour on individuals: Some recent developments. Chapter 15, In D. Hensher and P. Stopher (Eds.). Behavioural Travel Modelling. Croom Helm, London, England. 872 pp. Meyer, R. 2009. Fire effects information system: Vulpes velox. U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station, Fire Sciences Laboratory. Available online at www. Accessed 17 May 2018. Miller, B., G. Ceballos, and R. Reading. 1994. The Prairie Dog and biotic diversity. Conservation Biology 8:677–681. Mitchell, E. 2018. Distribution, ecology, disease risk, and genetic diversity of Swift Fox (Vulpes velox) in the Dakotas. M.Sc. Thesis. South Dakota State University, Brookings, SD, USA. 170 pp. Moehrenschlager, A., S. Alexander, and T. Brichieri-Colombi. 2006. Habitat suitability and population viability analysis for reintroduced Swift Foxes in Canada and northern Montana. Centre for Conservation Research Report No. 2. Calgary, Alberta, Canada. 30 pp. Montana Fish, Wildlife and Parks. 2011. Montana connectivity project: A statewide analysis. Montana Fish, Wildlife and Parks, Helena, MT. Available online at html?id=53365. Accessed 19 September 2018. Montana Natural Heritage Program. 2016. Swift Fox (Vulpes velox) predicted suitable habitat models created on November 04, 2016. Montana Natural Heritage Program, Helena, MT. Available online at Accessed 6 November 2018. Nicholson, K.L., W.B. Ballard, B.K. McGee, J. Surles, J.F. Kamler, and P.R. Lemons. 2006. Swift Fox use of Black-tailed Prairie Dog towns in northwest Texas. Journal of Wildlife Management. 70:1659–1666. Nicholson, K.L., W.B. Ballard, B.K. McGee, and H.A. Whitlaw. 2007. Dispersal and extraterritorial movements of Swift Foxes (Vulpes velox) in northwestern Texas. Western North American Naturalist. 67:102–108. Olimb, S.K., S. Forrest, and K. Bly. 2009. Swift Fox analytic hierarchy process (AHP) habitat suitability model. World Wildlife Fund, Northern Great Plains Program, Bozeman, MT, USA. 9 pp. Prairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 14 Olimb. S.K., and K. Bly. 2010. Swift Fox habitat suitability index for the Northern Great Plains ecoregion. World Wildlife Fund, Northern Great Plains Program, Bozeman, MT, USA. 9 pp. Olson, T.L. 2000. Population characteristics, habitat selection patterns, and diet of Swift Foxes in southeast Wyoming. M.Sc. Thesis. University of Wyoming, Laramie, WY, USA. 139 pp. Olson, T.L., and F.G. Lindzey. 2002a. Swift Fox (Vulpes velox) home-range dispersion patterns in southeastern Wyoming. Canadian Journal of Zoology 80:2024–2029. Olson, T.L., and F.G. Lindzey. 2002b. Swift Fox survival and production in southeastern Wyoming. Journal of Mammalogy 83:199–206. Pruss, S.D. 1999. Selection of natal dens by the Swift Fox (Vulpes velox) on the Canadian prairies. Canadian Journal of Zoology 77:646–652. R Development Core Team. 2013. R: A language and environment for statistical computing. Version 3.3.2. R Foundation for Statistical Computing, Vienna, Austria. Available online at http://www.Rproject. org/. Accessed 16 December 2016. Russell, T.A. 2006. Habitat selection by Swift Foxes in Badlands National Park and the surrounding area in South Dakota. M.Sc. Thesis. South Dakota State University, Brookings, SD, USA.119 pp. Sasmal, I., J.A. Jenks, T.W. Grovenburg, S. Datta, G. Schroeder, R.W. Klaver, and K.M. Honness. 2011. Habitat selection by female Swift Foxes (Vulpes velox) during the pup-rearing season. The Prairie Naturalist 43:29–37. Schauster, E.R., E.M. Gese, A.M. Kitchen. 2002. Population ecology of Swift Foxes (Vulpes velox) in southeastern Colorado. Canadian Journal of Zoology 80:307–319. Schwalm, D.L. 2012. Understanding functional connectivity in the shortgrass and mixed grass prairies using the Swift Fox as a model organism. Dissertation. Texas Tech University, Lubbock, TX. 181 pp. Schwalm, D.L., L.P. Waits, and W.B. Ballard. 2014. Little Fox on the prairie: Genetic structure and diversity throughout the distribution of a grassland carnivore in the United States. Conservation Genetics 15:1503–1514. Sharps, J.C., and D.W. Uresk. 1990. Ecological review of black-tailed Prairie Dogs and associated species in western South Dakota. Great Basin Naturalist 50:339–345. Shaughnessy, M.J., Jr., and R.L. Cifelli. 2004. Influence of black-tailed Prairie Dogs on carnivore distributions in the Oklahoma Panhandle. Western North American Naturalist 64:184–192. Sheng, L., J. Huang, and X. Tang. 2011. A tasseled cap transformation for CBERS-02B CCD data. Journal of Zhejiang University 12:780–786. Sohl, T.L., B.M. Sleeter, K.L. Sayler, M.A. Bouchard, and R.R. Reker. 2012. Spatially explicit landuse and land-cover scenarios for the Great Plains of the United States. Agriculture, Ecosystems and Environment 153:1–15. Soil Survey Staff, Natural Resources Conservation Service, United States Department of Agriculture. Web soil survey. Available online at Accessed 1 January 2013. Sovada, M.A., C.C. Roy, J.B. Bright, and J.R. Gillis. 1998. Causes and rates of mortality of Swift Foxes in western Kansas. Journal of Wildlife Management 62:1300–1306. Sovada, M.A., C.C. Roy, and D.J. Telesco. 2001. Seasonal food habits of Swift Fox (Vulpes velox) in cropland and rangeland landscapes in western Kansas. The American Midland Naturalist 145:101–111. Sovada, M.A., R.O. Woodward, and L.D. Igl. 2009. Historical range, current distribution, and conservation status of the Swift Fox, Vulpes velox, in North America. The Canadian Field- Naturalist 123:346–367. Stephens, R.M., and S.H. Anderson. 2005. Swift Fox (Vulpes velox): A technical conservation assessment, In Species Conservation Program/species Conservation Assessments. U.S. Department of Agriculture, Forest Service, Rocky Mountain Region, Golden, Colorado, USA. Available online at Accessed 23 January 2017 Thompson, C.M., and E. Gese. 2007. Food webs and intraguild predation: Community interactions of a native mesocarnivore. Ecology 88:334–346. Uresk, D.W., and J.C. Sharps. 1986. Denning habitat and diet of the Swift Fox in western South Dakota. Great Basin Naturalist 46:249–253. Prairie Naturalist S.K. Olimb, D.L. Schwalm, and K.L.S. Bly 2021 53:1–15 15 Vashon, J.H., A.L. Meehan, W.J. Jakubas, J.F. Organ, A.D. Vashon, C.R. McLaughlin, G.J. Matula Jr., and S.M. Crowley. 2008. Spatial ecology of a Canada lynx population in northern Maine. Journal of Wildlife Management 72:1479–1487. Wiken, E., J.N. Francisco, and G. Griffith. 2011. North American terrestrial ecoregions—level III. Commission for Environmental Cooperation, Montreal, Canada. Available online at http://www. Accessed 11 April 2011.