https://cpw.cvlcollections.org/files/original/68a4337747e84c82b0d833d930b70bd9.pdf 2d20ec29f0f251e15e8887c716fe6049 PDF Text Text The research in this publication was partially or fully funded by Colorado Parks and Wildlife. Dan Prenzlow, Director, Colorado Parks and Wildlife • Parks and Wildlife Commission: Marvin McDaniel, Chair • Carrie Besnette Hauser, Vice-Chair Marie Haskett, Secretary • Taishya Adams • Betsy Blecha • Charles Garcia • Dallas May • Duke Phillips, IV • Luke B. Schafer • James Jay Tutchton • Eden Vardy �Applied Geography 86 (2017) 66e91 Contents lists available at ScienceDirect Applied Geography journal homepage: www.elsevier.com/locate/apgeog Modeling large-scale winter recreation terrain selection with implications for recreation management and wildlife Lucretia E. Olson a, *, John R. Squires a, Elizabeth K. Roberts b, Aubrey D. Miller b, 1, Jacob S. Ivan c, Mark Hebblewhite d a Rocky Mountain Research Station, United States Forest Service, Missoula, MT 59801, USA White River National Forest, United States Forest Service, Glenwood Springs, CO, USA Colorado Parks and Wildlife, Fort Collins, CO 80526, USA d Wildlife Biology Program, Department of Ecosystem and Conservation Sciences, College of Forestry and Conservation, University of Montana, Missoula, MT 59812, USA b c a r t i c l e i n f o a b s t r a c t Article history: Received 15 December 2016 Received in revised form 13 June 2017 Accepted 19 June 2017 Available online 3 July 2017 Winter recreation is a rapidly growing activity, and advances in technology make it possible for increasing numbers of people to access remote backcountry terrain. Increased winter recreation may lead to more frequent conﬂict between recreationists, as well as greater potential disturbance to wildlife. To better understand the environmental characteristics favored by winter recreationists, and thus predict areas of potential conﬂict or disturbance, we modeled terrain selection of motorized and non-motorized recreationists, including snowmobile, backcountry ski, and snowmobile-assisted hybrid ski. We used sports recorder Global Positioning System (GPS) devices carried by recreationists at two study areas in Colorado, USA, (Vail Pass and the San Juan Mountains), to record detailed tracks of each recreation type. For each recreation activity, we modeled selection of remotely-sensed environmental characteristics, including topography, vegetation, climate, and road access. We then created spatial maps depicting areas that recreation activities were predicted to select and combined these maps to show areas of potential ecological disturbance or interpersonal conﬂict between motorized and non-motorized activities. Model results indicate that motorized and non-motorized activities select different environmental characteristics, while still exhibiting some similarities, such as selection for ease of access, reﬂected in proximity to highways and densities of open forest roads. Areas predicted to have only motorized recreation were more likely to occur further from highways, with greater forest road densities, lower canopy cover, and smoother, less steep terrain, while areas with only non-motorized recreation were closer to highways, with lower forest road densities, more canopy cover and steeper terrain. Our work provides spatially detailed insights into terrain characteristics favored by recreationists, allowing managers to maintain winter recreation opportunities while reducing interpersonal conﬂict or ecological impacts to sensitive wildlife. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http:// creativecommons.org/licenses/by-nc-nd/4.0/). Keywords: GPS tracking Habitat models Interpersonal conﬂict Motorized recreation Non-motorized recreation Recreation planning 1. Introduction The ecological impact of human recreation on the landscape is a rapidly growing concern for land-use managers, as centers of human population spread out into previously sparsely populated areas (Theobald, 2004). Winter recreation, including backcountry and downhill skiing, snowshoeing, and snowmobiling, is a popular * Corresponding author. E-mail address: lucretiaolson@fs.fed.us (L.E. Olson). 1 School of Surveying, University of Otago, Dunedin 9054, New Zealand. use of public lands, as well as a primary economic driver to communities throughout the western United States (Bowker et al., 2012). Technological advances in motorized winter recreation, such as heliskiing, snow biking, more powerful snowmobiles, and snowmobile-assisted (hybrid) skiing, means that recreationists access increasingly remote areas. With greater numbers of recreationists seeking their own recreation experience on a shared landscape, ecological impacts of recreation as well as encounters between non-motorized and motorized recreationists are likely to increase (Gramann, 1982; Manning & Valliere, 2001). Increases in the number of people using a recreation area or in http://dx.doi.org/10.1016/j.apgeog.2017.06.023 0143-6228/Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). �L.E. Olson et al. / Applied Geography 86 (2017) 66e91 the spatial extent of recreation can have negative ecological consequences, such as increased disturbance to wildlife. For instance, large-scale displacement of animal populations to areas of poorer habitat has been demonstrated in moose (Alces alces) due to disturbance from snowmobiles (Harris, Nielson, Rinaldi, & Lohuis, 2014) and mountain goats (Oreamnos americanus) due to the ^ te �, 2016). The challenge of presence of ski areas (Richard & Co managing recreation to both allow human use of public lands while also conserving ecosystems is intensiﬁed by a lack of detailed knowledge of the spatial and environmental characteristics of human recreation. In addition to ecological implications, increased recreation also has the potential to exacerbate conﬂict or safety issues between different recreation user groups (Miller, Vaske, Squires, Olson, & Roberts, 2016; Thapa & Graefe, 2004; Vaske, Carothers, Donnelly, & Baird, 2000). Interpersonal conﬂict, in which direct or indirect contact between different types of recreationists aggravates users (Jacob & Schreyer, 1980; Vaske, Needham, & Cline Jr., 2007), is likely to depend on the environmental preferences of each type of activity, and the degree to which these preferences overlap. Vaske, Donnelly, Wittmann, and Laidlaw (1995) found low interpersonal conﬂict between hunters and non-hunters in a Colorado study due to their natural separation by topography, as well as management regulations that prevented interaction. To predict areas more likely to engender interpersonal conﬂict among recreation types, a better understanding of the terrain characteristics favored by different types of recreationists is needed (Kliskey, 2000; Snyder, Whitmore, Schneider, & Becker, 2008). Most recreation studies rely heavily on the recreationist to selfreport details about his/her movements and interactions with other recreationists (Brown & Raymond, 2014; D'Antonio et al., 2010; Tomczyk, 2011). This provides neither an objective nor complete depiction of the spatial and temporal movement patterns of a recreationist through a landscape (Cole & Daniel, 2003; Hallo, Manning, Valliere, & Budruck, 2004). In addition, self-reported interactions or conﬂicts with other users may be unconsciously biased by user perception, which may differ from realized interpersonal conﬂict. For instance, hikers in New Zealand who did not encounter mountain bikers had a more negative opinion of them than those that did (Cessford, 2003). A difference in perception versus realization of conﬂict could lead to inappropriate management practices in an attempt to reduce conﬂict where none exists. One way to overcome these methodological issues is to use Global Positioning System (GPS) devices to collect high-resolution spatial data, which can provide an objective depiction of recreationist movements (Beeco & Brown, 2013; Hallo et al., 2012; Lai, Li, Chan, & Kwong, 2007) and interactions. We use GPS locations collected by recreationists in two locations in western Colorado, USA to model landscape-level recreation patterns. Like many areas in western USA, western Colorado is experiencing rapidly growing winter recreation on public lands, and also has a number of sensitive wildlife species that may be negatively affected by increasing recreation, including threatened Canada lynx (Lynx canadensis). We apply resource selection functions (RSFs) and step-selection functions (SSFs) to quantify the importance of a given set of environmental covariates to each recreation activity, as well as to provide a spatial depiction of predicted areas of use (Boyce, Vernier, Nielsen, & Schmiegelow, 2002; Manly, McDonald, Thomas, McDonald, & Erickson, 2002). Both types of models are frequently used in wildlife studies to quantify habitat selection, which is characterized by the environmental conditions at sites used by individuals compared to those same conditions at a set of randomly available locations (Manly et al., 2002). Here we use RSFs and SSFs in a novel way: to determine which environmental characteristics are selected by people 67 taking part in different recreation activities. We quantify selection over the entire recreation study area using RSF models, and employ SSFs to determine selection at a ﬁner scale, as each recreationist moves through the landscape. The goals of our research were to: 1) use GPS technology to measure movement characteristics of motorized (snowmobile, hybrid ski) and non-motorized (backcountry ski), winter recreationists 2) use spatially-explicit models to predict environmental characteristics and spatial landscapes likely sought by winter recreationists, and 3) use these modeled understandings to determine characteristics of potential interpersonal conﬂict or ecological impact. Results from our analyses can be used to identify areas selected by different recreation activities to inform management decisions on recreation zoning or education programs to limit interpersonal conﬂict or reduce wildlife disturbance. 2. Methods 2.1. Study area Our study area consisted of two broad locations in the Colorado Rocky Mountains, USA (Fig. 1). The Vail Pass site covers an area in the northern Sawatch and Mosquito Ranges, southern Gore Range and western Front Range (approximate centroid coordinates 106.30� W, 39.45oN) near the towns of Vail, Leadville, and Frisco, CO. Data were collected on public lands administered by the White River National Forest and the San Isabel National Forest. The San Juan site covers a large area in southwest Colorado in the San Juan mountain range near the towns of Silverton and Telluride (approximate centroid 107.88oW, 37.82oN). Data were collected on public lands administered by the San Juan National Forest, the Uncompahgre National Forest, and the Bureau of Land Management. Both sites experienced winter recreation between the end of December and early April in the sub-alpine and alpine zones with elevations between 2380 m and 4340 m and annual snowfall typically between 380 cm and 1000 cm (National Weather Service, 2017). Both sites had some level of recreation zoning, where motorized recreation was prohibited in certain designated areas. The sites differed in terms of terrain and accessibility. Recreation in the Vail Pass site was largely inﬂuenced by proximity to major population centers, which are within a 1e2 h drive. Winter recreation was concentrated along Interstate 70 between Copper Mountain and Vail, CO in the fee-operated Vail Pass Winter Recreation Area (VPWRA) managed by the White River National Forest, as well as along Highway 6 over Loveland pass (non-motorized use only). Motorized recreation was heavily concentrated along a network of 50 miles of established groomed routes in the VPWRA. Non-motorized access to backcountry huts in the VPWRA also attract recreation to the area. The VPWRA sees roughly 35,000 feepaying visitors per winter season, of whom approximately 11,000 are hut visitors (U.S.D.A. Forest Service, 2015). Hybrid use has increased sharply on the VPWRA, where backcountry skiers and snowboarders use snowmobiles or snow coaches to access terrain that would otherwise be inaccessible in a single day-trip. The majority of data collected was motorized or hybrid-use in the Vail Pass site. Winter recreation in the San Juan site was more dispersed, with a greater number of access portals spread over a larger spatial extent than Vail. Access was highly dependent on the network of maintained roads, especially along U.S. Highway 550 and C.O. Highway 145 (see Fig. 1A), and there was no recreation fee area. The San Juan site was more isolated from major population centers (none within 2e3 h drive). While the majority of winter recreation in the Vail Pass study site was concentrated in fewer than 10 access portals, recreation in the San Juan site occurred from over 50 access �68 L.E. Olson et al. / Applied Geography 86 (2017) 66e91 portals, and included over 250 km of established groomed routes. Due to steep terrain, motorized recreation in the San Juan site was more concentrated compared to non-motorized recreation. Taken together, the two study sites effectively capture the spectrum of winter recreation in the Colorado Rocky Mountains and thus provide a broad sample of recreation terrain in western Colorado. 2.2. Data collection From January to March of 2010e2013, we stationed technicians at recreation access portals to distribute GPS units (Qstarz International Co., Ltd., model BT-Q1300, Position accuracy < 10 m). Technicians sampled recreationists by walking through a parking area and selecting every 4th vehicle (Vail Pass) or driving between access portals and approaching recreationists still at their vehicle (San Juans). For the latter approach, technicians began driving between access portals at approximately 10:00 h, and checked all known access portals (~50 portals) for recreationists at least once per day; technicians spent between 15 min and 1 h at each location, depending on the number of recreationists present, and did not vary the order in which they checked sites. Technicians gave a brief explanation of the project goals, informed recreationists that no personally identiﬁable information would be collected, and offered a map of the track made by the recreationist as an incentive for carrying the GPS unit. Participants then dropped the GPS unit into a collection bin at the end of the day, or returned it by mail. Technicians recorded the type of recreation activity engaged in and number of people in the group. If > 1 person was in the group, only one GPS unit was given to the group as a whole. While technicians did not sample the same people multiple times per day, it is possible that some recreationists carried a GPS unit more than once during the study. Given the large number (>35,000) of recreationists in our study areas, however, we do not believe that this happened frequently and thus assume independence of recreation tracks, which we deﬁne as a single user's, or group of users', daily travel pattern. We recorded snowmobile, backcountry ski or snowboard (hereafter backcountry ski), and hybrid recreation. Snowmobile included any motorized use, including snow-cats and motorized bikes. Hybrid use occurred when skiers or snowboarders were transported by a snowmobile or snow-cat, usually to a peak or ridge, and then skied down the slope. We visually screened all recorded recreation tracks for erroneous points using ArcGIS (Environmental Systems Research Institute 2011, ArcGIS Desktop: Release 10. Redlands, CA). When screening data, we deleted points that were in areas where recreation was not taking place, such as in parking lots or on highways, as well as outliers that were obviously erroneous based on large distances between a given point and the points directly before and after it. Additionally, points were more prone to error immediately after GPS units were turned on, as the units searched for sufﬁcient satellites to collect data; we closely examined the start of each recreation track and removed all inaccurate points until the locations visually became more consistent (Beeco & Hallo, 2014). For analysis, we divided the GPS points recorded by snowmobilers into those that occurred on groomed routes and those that took place in non-groomed areas (henceforth on- and off-trail, respectively) and hybrid GPS points into ski (non-motorized) and snowmobile (motorized) segments, since we expected terrain selection to differ between these categories. We used road and trail GIS layers provided by the U. S. Forest Service (White River NF, Uncompahgre NF, San Juan NF travel management GIS layer) and considered snowmobile tracks < 15 m to either side of a road or trail as “on-trail” and points > than 15 m as “off-trail” to account for spatial resolution of GPS data. We classiﬁed motorized hybrid data when the average speed was greater than or equal to 30 miles per hour (48 km/h) and the track was gaining elevation, or the point fell within 15 m of a trail, and non-motorized hybrid data otherwise. GPS location data were recorded at 5 s intervals; if GPS units remained stationary, however, no location was collected until the device detected movement. Since recreation activities occurred at different speeds, this resulted in locations that ranged from 1 m to 40 m apart. To best assess conﬂict potential between recreation activities, we standardized spatial scales by sub-sampling recreation activities at approximately 140 m between points (20 s interval for snowmobiles, 25 s for hybrid snowmobiles, 60 s for hybrid skiers, 120 s for backcountry skiers). This represented a ﬁne-scale perception distance at which both motorized and non-motorized recreationists might make movement decisions. We also used magnetic and infra-red trail counters as an independent assessment of recreation intensity and distribution throughout our study areas to verify the efﬁcacy of our GPS sampling. Trail counters recorded the number of people passing by constricted trail segments used by various recreation activities. We visually compared the counts from trail counters to GPS recreation tracks to locate any areas that had recreation but were not being adequately sampled by GPS methods, and adjusted our sampling efforts accordingly. We also used trail counters to identify intense periods of use during the day and week to better inform our sampling effort. We summarized trail counter data to mean counter hits per day of week and hour of day at each study area. 2.3. Environmental variables We considered 12 environmental covariates as potential predictors of recreation selection. Covariates were chosen based on factors that we believed were important to recreationists: topography, vegetation, climate, and access (Table 1). To account for the possibility that recreationists might consider these environmental covariates at different spatial scales when making land use decisions, we considered all variables at four spatial scales. We used ArcGIS to calculate the average of each covariate within 125 m, 500 m, 1250 m, and 2500 m radii, chosen to span both small and large-scale movements based on observed recreation travel distances. We standardized all covariates by subtracting the mean and dividing by the standard deviation to allow direct comparison between estimated model coefﬁcients and for ease of model ﬁtting. 2.4. Statistical analyses To measure movement characteristics of recreation tracks, we calculated the total number of points recorded for each track, the total distance covered, the average movement speed, the length of time spent moving, and the minimum and maximum elevation reached along each track. We calculated the time and distance between two consecutive GPS points and used these to calculate average movement speed and length of time spent moving. We considered a point to be ‘moving’ if the speed was greater than 1 km/h. Total distance covered was calculated by summing the distance between consecutive GPS points. We used a digital elevation model (DEM; USGS National Elevation dataset) to determine difference between the points in each track with the minimum and maximum elevation. Once these characteristics were calculated for each track, they were summarized by taking the median over all tracks within each recreation activity. To summarize the environmental conditions that were available to each recreation type, as well as the conditions that each recreation type actually used (as compared to what they select, which is measured below and may differ from use), we also calculated the mean of all ‘used’ and ‘available’ points for each recreation activity for each of the 12 environmental covariates. �L.E. Olson et al. / Applied Geography 86 (2017) 66e91 69 We used both resource selection function (RSF) and stepselection function (SSF) models to characterize environmental selection of snowmobiles, hybrid skiers, and backcountry skiers. Both RSF and SSF functions compare environmental characteristics at actual GPS locations (‘used’ locations) to those same characteristics at locations randomly selected across a study area (‘available’ locations); environmental characteristics that are used disproportionately more than what is available are said to be selected. The area considered as available in the models was deﬁned as a minimum convex polygon around all recreation locations at each study site (Fig. 1); this insured that inferences made from each model would be comparable for all recreation types. Within this boundary, we removed privately owned land not available to recreationists. For motorized models only, we also removed areas administratively closed to motorized recreation, such as wilderness or designated non-motorized areas (Fig. 1B&C). We used a general linear mixed-effects model with a logit link function (logistic regression; Hosmer, Lemeshow, & Sturdivant, 2013) and individual recreation track ID as a random intercept to control for non-independence between points within a single track (Gillies et al., 2006) to estimate separate relative probability RSFs for backcountry ski, hybrid ski, hybrid snowmobile, snowmobile on-trail, and snowmobile off-trail recreation activities. We randomly generated ‘available’ points within the available areas deﬁned above for a given recreation activity at a ratio of 1 ‘used’ point to 5 ‘available’ points so that available environmental characteristics were adequately sampled. Correlations among covariates within small (125 m and 500 m radii) and large (1250 m and 2500 m radii) spatial scales were often high. Thus, we initially ﬁt univariate models with only one covariate at a given scale at a time to discard any covariates with poorer ﬁt than a null model based on Akaike Information Criteria (AIC; Akaike, 1974), and to select one large and one small spatial scale per covariate. We included quadratic forms of covariates to investigate non-linear relationships if supported by AIC. We then used the selected covariates to construct all subsets of candidate models for multivariate analysis using the ‘lme4’ (Bates, Maechler, Bolker, & Walker, 2014) and ‘MuMIn’ packages (Barton, 2015) in R (R Core Team, 2015); covariates correlated at jrj > 0.6 were not allowed in the same model. We ranked multivariate models using AIC. The SSF models that evaluated ﬁne-scale selection by winter recreationists used conditional logistic regression to estimate relative probability of selection (Fortin et al., 2005; Thurfjell, Ciuti, & Boyce, 2014). At each ‘used’ GPS location, we compared 5 ‘available’ GPS locations that were selected based on the known distribution of step lengths (straight-line distance from one GPS point to the next) and turn angles estimated from actual recreation data. Thus, each used point was compared directly to a set of available points that the recreationist could have chosen as they moved from point A to point B on a track. We used the same covariates as in the RSF, but limited scales to only 125 m and 250 m since the purpose of the SSF model was to evaluate selection decisions at a ﬁne-scale as recreationists traverse landscapes. Variable selection and model ﬁtting were performed as in the RSF models, except that models were ﬁtted using the R package ‘survival’ (Therneau, 2015) to estimate conditional regression models. To provide managers with a map that could be used to inform management decisions on recreation zoning or to identify areas selected by different recreation activities, we created maps of predicted relative probability of selection for each recreation type across western Colorado within an elevation zone delineated by Fig. 1. Spatial extent of recreation used in this study at two locations in western CO, USA: the more northerly Vail Pass and the southerly San Juan Mountains (A); inset shows the position of Colorado within the USA. Panels B and C show areas that were closed to motorized recreation (gray wilderness areas and horizontally striped zoned areas) within the Vail (B) and San Juan (C) study areas. �70 L.E. Olson et al. / Applied Geography 86 (2017) 66e91 Table 1 Variable names, native resolution, source and description for all covariates used to model selection of environmental characteristics by recreationists in Colorado, USA. Name Resolution Source Highway Vector/ 30 m Elevation 30 m Canopy 30 m Evergreen 30 m North 30 m Precipitation 800 m Slope 30 m Temperature 800 m Roughness 30 m TPI 30 m Road Density Forest Edge Vector/ 30 m 30 m Euclidean distance to nearest highway (m) Elevation (m) Percent tree canopy cover Percent conifer forest Index of north-facing aspect Average annual precipitation (mm) Slope in degrees Mean annual temperature (oC) Index of terrain variability; 3D area divided by 2D area Topographic position index, measure of landscape concavity or convexity National Forest travel management road layer, including only forest roads, Non-drivable forest roads that can be used as travel corridors; length not highways of road per unit area, varying scales National Land Cover Database 2011 Landcover Type: Deciduous, Index of forest connectivity; length of forest/non-forest edge per unit Evergreen, and Mixed Forest (Homer et al., 2015) area, varying scales minimum and maximum elevation from all recreation data combined. We used the top-performing RSF model from each recreation type to predict relative probability of selection based on the environmental covariates across western Colorado. The used-available study design employed here produces a relative probability of selection since the number of sampled available points is arbitrary (Keating & Cherry, 2004). Thus, we used the equation wðxÞ ¼ Description Colorado Department of Transportation Online Transportation Information System United States Geological Survey National Elevation Dataset National Land Cover Database (2011) Tree Canopy (Homer et al., 2015) National Land Cover Database (2011) Land Cover (Homer et al., 2015) ArcGIS Aspect Tool, Cosine transformation PRISM 1980e2010 Precipitation normals ArcGIS Slope Tool PRISM 1980e2010 Mean temperature normals DEM Surface Tools (Jenness, 2013) Land Facet Corridor Tools (Jenness, Brost, & Beier, 2013) expðb1 x 1 þ b2 x 2 þ … þ bk x k Þ ð1 þ expðb1 x 1 þ b2 x 2 þ … þ bk x k ÞÞ where b is an estimated model coefﬁcient and x is the value of k covariates, to estimate relative probability of selection rescaled from 0 to 1 (Manly et al., 2002). values of the withheld 5th group; each group was withheld in turn. We predicted RSF values at all ‘available’ locations and binned these values into 10 quantiles. Predictions from ‘used’ locations were then grouped based on these quantiles, and the number of predicted used locations in each quantile was counted. We compared the predicted count of used locations to the quantile rank using a Spearman rank correlation (Boyce et al., 2002). Good model ﬁt is indicated by a strong correlation between predicted values and quantile number. In addition, for RSF models, we performed a second independent validation using 100,000 withheld GPS points from each recreation type. The RSF value was predicted at each of these withheld points and then binned according to Boyce et al. (2002). 2.5. Recreation overlap analysis 3. Results To determine what environmental conditions are present at areas of predicted spatial overlap between motorized and nonmotorized forms of recreation, and thus what conditions may favor conﬂict between user groups, we performed the following analysis. We ﬁrst created a binary depiction of each recreation type from each continuous relative probability surface generated above based on the maximum sum of sensitivity (true positives) and speciﬁcity (true negatives; Freeman & Moisen, 2008). This threshold optimizes the number of ‘used’ recreation locations correctly assigned into ‘recreation area’ and the number of ‘available’ locations correctly assigned into ‘non-recreation area’. We then used the binary surfaces to identify areas of motorized activities only (snowmobile and hybrid-snowmobile), non-motorized activities only (backcountry ski and hybrid-ski), and areas with both motorized and non-motorized recreation. To generate a summary of environmental characteristics at these areas of predicted overlap compared to areas with only one predicted type of recreation, we averaged each of our 12 environmental variables (Table 1) across each of these areas. We also determined the degree to which each predicted continuous surface was similar to the others, using a Pearson correlation, to determine which types of recreation were more likely to select similar environmental characteristics. 3.1. Recreation summary 2.6. Model validation We used 5-fold cross validation to determine goodness of model ﬁt. Recreation tracks were split into 5 equal sized groups, the model was re-estimated on 4 of the groups and used to predict the RSF In January to March of 2010e2013, we recorded 2143 recreation tracks. We collected an average of 1306 (SD ¼ 435) GPS points per track (Table 2; Fig. 2). The most tracks in our dataset came from backcountry skiing or snowboarding (52%), followed by snowmobile (32%). Snowmobiles traveling on trails or groomed routes traveled the fastest, with a median speed of 30.6 km/h, while backcountry ski was slowest, at a median 4.3 km/h (Table 3). Hybrid recreationists traveled greatest distances, with median track length 41.0 km, while back-country skiers traveled shortest, 5.2 km. Within hybrid recreation tracks, approximately 4.8 km, or 12% of total distance, was spent skiing. Snowmobiles averaged 35.2 km tracks, of which a median 4.9 km (approximately 13%) were spent off-trail (Table 3). The duration of trips was similar among hybrid, backcountry skiers, and snowmobiles, at approximately 4 h. Of this time, each recreation type also spent approximately 2.5 h in active movement. Snowmobilers had the biggest change from minimum Table 2 Summary of the number of tracks collected for each winter recreation activity in Colorado, 2010e2013. The total number of GPS points originally recorded at 5 s intervals, as well as the average and standard deviation of GPS points per track, are given. Recreation Mode # Tracks Total # of points Mean pts/track SD Snowmobile Hybrid Backcountry Ski Total 686 346 1111 2143 889,674 604,223 973,163 2,467,060 1297 1746 876 1306 827 1203 921 435 �L.E. Olson et al. / Applied Geography 86 (2017) 66e91 71 Fig. 2. Examples of recreation tracks recorded with GPS units during the study in western Colorado, 2010e2013. Panel A) snowmobile tracks primarily on trails in the Vail study area, B) hybrid skiing in the Vail study area; thick lines near the bottom of the picture show snowmobile travel, while thinner dispersed lines further back show skiing, C) backcountry ski recreation in the San Juans study area, and D) a combination of all three recreation types at the Vail study area, showing areas of overlap as well areas used primarily by one recreation type. Image credit: Google, DigitalGlobe. Table 3 Median movement characteristics for all snowmobiles (Snmb), snowmobiles on trails (Snmb on-tr), snowmobiles off trails (Snmb off-tr), all hybrid (Hybrid), hybrid snowmobile (Hyb snmb), hybrid ski (Hyb ski), and backcountry ski (BC ski) recreation types studied in western CO, 2010e2013. The median and bootstrapped 95% lower conﬁdence interval (LCI) and upper conﬁdence interval (UCI) for movement speed (km/hr), total track distance (km), time spent actively moving (hr), total recorded trip time (hr), and total elevation change (m) is given. Movement Speed (km/h) Track Distance (km) Active move time (hr) Total trip time (hr) Total Elevation Change (m) Median 95% LCI 95% UCI Median 95% LCI 95% UCI Median 95% LCI 95% UCI Median 95% LCI 95% UCI Median 95% LCI 95% UCI Snmb Snmb on-tr Snmb off-tr Hybrid Hybr snmb Hyb ski BC ski 24.5 24.0 25.2 35.2 32.9 37.0 2.4 2.3 2.5 3.8 3.6 4.0 660.0 557.0 715.0 30.6 29.7 31.4 33.2 31.3 35.2 1.8 1.7 1.9 2.5 2.4 2.6 557.0 538.0 643.0 22.4 21.9 22.9 4.9 4.1 5.6 0.4 0.4 0.5 0.7 0.6 0.8 321.5 293.0 345.0 27.6 26.8 28.4 41.0 38.4 44.1 2.5 2.2 2.7 4.6 4.3 4.8 498.0 489.0 516.0 28.3 27.6 28.8 35.5 33.6 37.3 2.6 2.3 2.8 3.5 3.5 3.7 490.0 482.0 501.5 14.0 13.0 14.8 4.8 4.4 5.5 0.7 0.6 0.7 1.0 0.9 1.1 375.0 369.0 386.0 4.3 4.2 4.4 5.2 5.0 5.4 2.0 1.9 2.1 3.6 3.5 3.8 382.0 371.0 395.0 to maximum elevation within tracks, with a median difference of 660 m. Back-country ski had the least elevation change, of 382 m (Table 3). Based on the mean of used GPS points, the covariates that indexed distance to highway, road density, percent canopy cover, and slope showed the greatest differences between winterrecreation types (Appendix A: Table A.1, Fig A.1). Hybrid skiers used areas that were farthest from highways (as averaged over all used GPS points; 4.61 km), followed by hybrid snowmobiles (4.05 km); snowmobiles on-trail (3.41 km) and off-trail (3.38 km) were next and did not differ from each other, and backcountry skiers remained nearest to major roads (2.46 km; Appendix A: Table A.1). On-trail snowmobiles and hybrid snowmobiles used areas with greater forest road density (1.19 km/km2 and 0.92 km/ km2, respectively), while off-trail snowmobiles and backcountry skiers used the least (0.65 km/km2 and 0.62 km/km2, respectively; �72 L.E. Olson et al. / Applied Geography 86 (2017) 66e91 Appendix A: Table A.1). Snowmobilers both on- and off-trail had the greatest mean percent canopy cover at used GPS locations (37.88% and 35.25%, respectively), followed by hybrid snowmobiles (33.88%) and backcountry skiers (31.26%). Backcountry skiers and hybrid skiers used steeper slopes than other recreationists (18.31� and 17.26� , respectively), while off-trail snowmobiles used the shallowest (14.7� ; Appendix A: Table A.1, Fig A.1). We deployed 140 trail counters at 95 locations over both study areas from 2010 to 2013. Trail counters conﬁrmed higher concentrated levels of use in the Vail area than in the more dispersed San Juan Mountains, with average seasonal counts approximately 5 times greater (average Vail 2010e2011: 73,967; average San Juans 2011e2013: 14,994 counter hits per year). Counter data also indicated greater recreation intensity during weekends (Saturday and Sunday, Vail: 55.9, SD ¼ 84.1; San Juans: 25.0, SD ¼ 34.9 counter hits per day) then during weekdays (Vail: 33.0, SD ¼ 46.6; San Juans: 13.2, SD ¼ 20.6 counter hits per day), a pattern consistent across study areas (Fig. 3). Hourly counts indicated virtually no recreation took place after dark: 96% of trail counter hits occurred between 0800 and 1700 h. Peak use occurred between 1000 and 1500 h, with an average of 5.3 (SD ¼ 11.1) hits per hour during this time in Vail Pass and 2.3 (SD ¼ 5.5) hits per hour in the San Juans (Fig. 3). 3.2. Reponses of winter recreationists to environmental features Top performing RSF models for all winter recreation activities indicated the importance of topography, access, and climate when making landscape-scale selection choices. All top models were >D4 AIC better than the next performing model (Appendix B: Tables B.1B.5). Based on coefﬁcient conﬁdence interval overlap with 0, all parameters in the top model for each recreation type were significant predictors of selection (except canopy cover for hybrid snowmobiles and backcountry ski, which did overlap 0). For brevity, we mention the top three contributing covariates for each model here, based on the strength of standardized beta coefﬁcients, but all contributing covariates are presented in Table 4. Snowmobiles on trails selected areas that had greater forest road density, moderate annual precipitation, and lower terrain variability (Table 4; Fig. 4). Off-trail snowmobiles selected moderate levels of snow, shallow slopes, and higher elevation (Table 4). Hybrid recreationists selected shallow slopes, intermediate distances from highways, and greater annual precipitation while on snowmobiles (Table 4), and moderate north-facing slopes with greater precipitation while on skis (Table 4; Fig. 4). Backcountry skiers selected areas that were closer to highways, had greater annual precipitation, and higher forest road density (Table 4; Fig. 4). Maps of predicted probabilities of landscape selection generated from topperforming RSF models for each type of recreation across western Colorado are shown in Appendix C: Figs C.1-C.5. At a ﬁne-scale, winter recreationists were sensitive to access, topography and vegetation when making movement decisions, again as determined by the size of standardized coefﬁcients in topperforming SSF models. There was some SSF model uncertainty, with between one and four models within >D4 AIC of the topperforming model (Appendix D: Tables D.1-D.5). However, models within >D4 AIC differed from the top-performing model by only one term, indicating that the extra parameters were noninformative, and thus we took the top-ranked, most parsimonious, model. All parameters in the top model for each recreation type were signiﬁcant predictors of selection, based on coefﬁcient conﬁdence interval overlap with 0; for brevity, we mention the top three contributing covariates for each model here, but all contributing covariates are presented in Table 5. Snowmobiles, while on trails, selected movement paths with moderate forest road density, moderate canopy cover, and higher elevation, while off-trail, they selected movement paths closer to the highway with moderate canopy cover and low terrain variability (Table 5). Hybrid recreationists, while snowmobiling, selected movement paths with moderate canopy cover, greater annual precipitation, and greater distances from highways, while on skis they selected warmer temperatures and greater annual precipitation, and avoided level terrain (Table 5). Backcountry skiers selectively moved through areas that were intermediate distances from highways, at middle elevations, and with greater forest road density (Table 5). 3.3. Recreation overlap Fig. 3. Mean hourly (A) and daily (B) count of recreationists from 140 magnetic and infrared trail counters deployed in Vail Pass (light gray) and San Juan (dark gray) study areas, western CO, USA. The minimum and maximum elevation from all recreation points combined was 2300 me4250 m; thus, we created predicted binary surfaces of winter recreation within this zone across western Colorado, a total area of 3123 km2. Using the binary motorized and non-motorized recreation maps we predicted that at least one type of recreation would occur on 590 km2 (18.9%). In areas with at least one type of recreation, motorized-only was predicted to occur on 35.2%, non-motorized recreation on 27.3%, and both activities were predicted to occur on 37.5% of this area (Fig. 5). Areas predicted to have both types of recreation were characterized by closer proximity to highways, high forest road density, high elevation, greater annual precipitation, and patchier forest, as well as intermediate levels of canopy cover, slope, TPI, and roughness, as compared to motorized or non-motorized only areas (Fig. 6). Winter recreationists with highest potential conﬂict based on predicted selection probabilities were backcountry skiers and hybrid skiers, with a �L.E. Olson et al. / Applied Geography 86 (2017) 66e91 73 Table 4 Model coefﬁcients and standard errors, as well as the scale of the covariate (m), from general linear mixed models (resource selection functions) of landscape-scale recreation terrain selection in western CO, USA; variance of the random effect (individual track) is also given. All covariates (except canopy cover for hybrid snowmobile and backcountry ski) were signiﬁcant predictors of recreation selection, based on conﬁdence interval overlap with 0. A superscript 2 indicates covariates that were ﬁtted as a quadratic function. Covariate Snowmobile On-Trail Scale b SE Scale b SE Scale b SE Scale b SE Scale b SE Highway Highway2 Elevation Elevation2 Forest Edge Canopy Canopy2 Evergreen Evergreen2 North Precipitation Precipitation2 Road Density Slope Slope2 Roughness Temperature Temperature2 TPI Random effect 2500 �0.87 0.01 2500 �0.85 0.02 1250 1250 1.11 �1.72 0.02 0.02 1250 1250 1.25 �1.26 0.06 0.05 2500 �1.73 0.02 125 125 2500 2500 2500 125 125 2500 0.03 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.03 0.02 0.01 0.02 125 2500 2500 125 �0.13 �0.02 �0.55 �1.29 0.01 0.02 0.01 0.01 2500 125 0.64 �0.96 0.05 0.04 2500 �1.2 1.32 0.02 0.01 2500 �2.12 1.82 0.13 0.06 2500 2500 2500 500 500 500 0.64 �0.02 �0.16 0.08 �0.49 �0.16 1.12 0.01 0.02 0.01 0.01 0.01 0.01 0.02 125 125 1.57 �0.11 0.46 1.49 �0.29 �0.68 �0.65 0.34 2.36 �0.61 0.35 �1.6 125 1250 1.03 �1.97 0.01 0.01 0.01 0.02 2500 �0.58 2.16 0.02 1.47 2500 0.02 0.02 0.01 0.01 0.89 125 500 �1.01 0.18 �0.95 �0.3 0.79 0.04 0.08 0.07 0.04 0.07 0.05 0.04 1.08 0.9 �0.87 1.15 1.96 �2.22 �0.53 �0.22 �1.28 �0.11 1.16 125 2500 1250 125 125 2500 �0.74 �0.78 �0.62 0.14 0.72 0.01 0.02 0.01 0.01 0.85 125 125 125 2500 2500 500 125 1250 1250 500 500 0.16 �0.5 0.05 0.32 �0.65 �0.12 1.22 �0.47 1.84 �0.79 �0.27 �0.9 0.29 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 �0.54 0.5 0.01 0.71 Snowmobile Off-Trail Hybrid Snowmobile Hybrid Ski 2500 Backcountry Ski 125 Table 5 Coefﬁcients and standard errors from conditional logistic regression (step selection function) models of ﬁne-scale recreation terrain selection in western CO, USA. All covariates were signiﬁcant predictors of recreation selection, based on conﬁdence interval overlap with 0. A superscript 2 indicates covariates that were ﬁtted as a quadratic function. Covariate Snmb On-Trail b Highway Highway 2 Elevation Elevation2 Forest Edge Canopy Canopy2 Evergreen Evergreen2 North Precipitation Road Density Road Density2 Slope Roughness Roughness2 Temperature TPI TPI2 Snmb Off-Trail SE Hybrid Snmb Hybrid Ski b SE b SE �0.53 0.17 0.49 0.06 b Backcountry Ski SE 0.38 0.05 0.28 0.09 0.02 �0.40 �0.10 0.01 0.01 0.01 0.05 �0.39 �0.20 0.01 0.02 0.01 �0.23 �0.68 �0.28 0.01 0.01 0.01 �0.09 0.08 �0.32 0.02 0.03 0.03 0.01 0.00 �0.03 0.01 0.77 �0.13 �0.08 �0.26 0.01 0.00 0.03 0.01 0.24 0.01 0.03 0.18 0.04 0.04 0.01 �0.23 0.01 �0.23 0.02 0.01 0.07 0.01 0.00 0.02 0.01 0.00 0.04 0.01 0.01 0.14 1.01 �0.08 �0.14 �0.31 0.12 0.50 0.34 �0.04 �0.37 �0.48 0.04 �0.34 0.03 0.31 �0.27 �0.45 0.13 1.87 �0.49 0.23 0.05 0.03 0.01 0.09 0.03 0.02 Pearson correlation coefﬁcient of 0.25. Recreationists with the least potential conﬂict were hybrid snowmobiling and off-trail snowmobiles with a correlation of 0.07 (Appendix E). 3.4. Validation Cross-validation indicated excellent RSF model ﬁt for all recreation types, with Spearman rank correlations of 0.98 for off-trail snowmobile, on-trail snowmobile, hybrid ski, and hybrid snowmobile, and 1.0 for backcountry ski. Our independent validation of withheld points also indicated strong model performance, with Spearman rank correlations of 0.99 for off-trail snowmobile, 1.0 for on-trail snowmobile, 0.95 for hybrid ski, 0.99 for hybrid snowmobile, and 1.0 for backcountry ski. Good predictive ability is indicated when independent recreation data have high predicted RSF values and Spearman correlations are closer to 1. b SE �0.64 0.13 �0.59 �0.22 0.06 0.21 �0.24 �0.18 �0.10 0.07 0.03 0.05 0.02 0.02 0.03 0.02 0.01 0.01 0.37 0.01 �0.31 0.01 0.08 0.07 0.01 0.01 4. Discussion This study provides a measure of winter recreation at a spatial scale and magnitude of data collection which has not, to our knowledge, been previously accomplished in the literature. We recorded approximately 2100 unique GPS tracks of recreationists and demonstrated the efﬁcacy of resource selection models to better understand winter recreation. Our analysis is unique in its application of a modeled understanding of environmental selection to winter recreation, using the actual locations of recreationists rather than metrics inferred by surveys, parking lot counts, or track evidence. We found differences in modeled terrain selection between motorized and non-motorized forms of recreation: areas predicted to be selected only by motorized users were farther from highways, with greater forest road densities, more open canopy, and shallower slopes, while areas predicted to be used only by non- �74 L.E. Olson et al. / Applied Geography 86 (2017) 66e91 Fig. 4. Example of spatial predictions from top-performing RSF recreation models at the Vail study area in Colorado, USA. Warm colors indicate greater probability of selection by each recreation type; white tracks are actual GPS locations from recreationists. All panels show same spatial extent in the area of Vail Pass Winter Recreation Area; panel A is on-trail snowmobile, B shows an aerial image of the actual terrain, C is hybrid ski, and D is backcountry ski. Image credit: Esri software. (For interpretation of the references to colour in this ﬁgure legend, the reader is referred to the web version of this article.) motorized users tended to be closer to highways, in denser canopy cover, with more terrain variability and steeper slopes (Fig. 6). These results can help identify areas where interpersonal recreation conﬂict between different user groups is likely to occur as well as ecologically sensitive areas that may be more susceptible to disturbance from a given type of recreation. 4.1. Environmental characteristics of recreation Few studies have similarly examined the land use patterns of winter recreationists. Braunisch, Patthey, and Arlettaz (2011) used snow track data and found a preference by skiers for smooth terrain, though the study was conducted only on areas near ski resorts and ski-lifts in Switzerland. In a study using surveys in British Columbia, Canada, Kliskey (2000) found preferences of snowmobilers for low canopy closure and less steep slopes. Rupf et al. (2011) sampled 303 individuals with GPS data loggers and found a tendency for skiers and snowboarders to be peak-oriented, although their study was focused on wildlife and not recreation. While we found differences in the selection of environmental characteristics for each type of recreation, in general, certain environmental characteristics were consistently important to all types of winter recreation at a landscape scale. Access to recreation areas was important to both motorized and non-motorized recreationists; snowmobilers and skiers selected areas that were close to highways and all recreation types selected greater density of forest roads, indicating the importance of accessibility over other environmental characteristics. A key ﬁnding from this study is the importance of roads to all types of winter recreation. The presence of paved highways enables recreationists to quickly reach areas open to recreation, while the presence of forest roads allows them to permeate forested backcountry areas more easily. Recreation is predicted to increase with increases in the extent of highways or the density and extent of forest roads, supporting the idea that recreation is an emergent property of roads on the landscape. Westcott and Andrew (2015) similarly showed that road proximity was one of the most important predictors when modeling the environmental associations of �L.E. Olson et al. / Applied Geography 86 (2017) 66e91 75 recreationists are likely to use forest roads to access the backcountry even if these roads are closed to vehicles (Havlick, 2002). Through the creation of forest roads, whether through logging operations, as part of ﬁre reduction or suppression activities, or for access to human developments, recreation is likely to show a corresponding increase as well. Differences in the results of the RSF and SSF models provide information on the importance of environmental characteristics to recreationists when ﬁrst selecting where to recreate, and then deciding how to move through the landscape once there. Topographic features, such as low to moderate slope, low terrain variability, and selection for drainages (except for skiers who selected ridges), were consistent predictors of recreation selection at a landscape scale, while vegetation characteristics were generally not among the top contributing covariates. Fine-scale movement models, conversely, were most strongly inﬂuenced by access and vegetation characteristics, and were more variable between different types of recreation. A stronger response to vegetation covariates at a small scale suggests that recreationists select areas in which to recreate at a hierarchical scale, with road access and large topographic features dictating an initial area selection, and ﬁner scale features such as forest density determining where to move within this area. The greater inﬂuence of vegetation at a small spatial scale may be related to the differences in movement speed and maneuverability of the different recreation types, since nonmotorized recreationists may be better able to safely move through dense trees, while motorized recreationists may select open areas for play and fast travel. Temporally, recreationists exhibited clear patterns of use with respect to time of day and day of the week. Nearly all recreation occurred during daylight hours, and dropped off to almost nothing after dark. Recreation was also markedly higher on weekends, particularly Saturdays, as compared to the rest of the week (Fig. 3). Thus, the ecological impact of winter recreation may decrease for species that are crepuscular or nocturnal, which will be active in times when little or no recreation is present. Similarly, weekdays may have a lower ecological impact than weekends, so that if management were undertaken to reduce or cap the number of users in an area, it may only need implementation during weekends. 4.2. Conﬂict and ecological implications Fig. 5. The distribution of predicted areas of potential overlap between motorized and non-motorized recreation activities within the Vail (A) and San Juan (B) study areas (thick black line denotes study area boundary). Green indicates areas predicted to be selected by both types of recreation, yellow is non-motorized only, and blue indicates motorized recreation. Background image credit: Esri software. (For interpretation of the references to colour in this ﬁgure legend, the reader is referred to the web version of this article.) off-road vehicle recreation. Indeed, the preferences of recreationists for certain environmental characteristics may be outweighed in practice by accessibility, with areas considered less suitable receiving more actual use due to the presence of ample parking areas and road access (Beeco, Hallo, & Brownlee, 2014; Brabyn & Sutton, 2013). Our models showed that areas greater than 11 km from a highway were predicted to have virtually no recreation at all, while areas predicted to have the highest recreation use, both motorized and non-motorized, were nearest highways. This has implications for forest and recreation management, since The predictions from our landscape scale selection models made possible a spatially resolute depiction of areas which motorized and non-motorized recreation were likely to select, and thus where interpersonal conﬂict may be more likely (Miller, 2016; Vaske et al., 2000). In a related survey study focused only on the Vail Pass area, Miller et al. (2016) found greater interpersonal conﬂict in areas of shared-use. Managers often employ spatial or temporal closures of areas to motorized or non-motorized activities in an attempt to limit shared-use and minimize conﬂict (Albritton & Stein, 2011; Leung & Marion, 1999). This is often an asymmetrical solution, however, with non-motorized users reporting increased satisfaction while motorized users are dissatisﬁed with increased restrictions (Jackson, Haider, & Elliot, 2003). Our model indicates that while zoning is a useful tool in some areas, it may be unnecessary in others. The environmental characteristics at areas predicted to have both types of recreation tended to differ from areas with either type alone (Fig. 6). Areas of overlap were closer to roads, had moderate slopes, and were in areas of patchier or more fragmented forest. This pattern may result from the use by both motorized and nonmotorized recreation of areas that are logistically necessary but not preferred, such as areas near parking lots and large groomed travel corridors. Thus, managers may be able to limit zoning to �76 L.E. Olson et al. / Applied Geography 86 (2017) 66e91 Fig. 6. Mean of environmental characteristics summarized in areas predicted across western Colorado to be selected by either motorized (Moto, circle) or non-motorized (Nonmoto, square) winter recreation only, or both (triangle) or neither (diamond). these areas of forced co-occurrence, while allowing recreationists more liberty outside these areas, where terrain selection should diverge. Outside of overlap areas, motorized and non-motorized forms of recreation show distinct separation in many environmental traits. Motorized recreationists tend to select drainages with low slope and low terrain variability, in lower elevation areas with more open canopy and less precipitation. This suite of characteristics probably favors fast, long-distance movements, which our results show are characteristic of snowmobiles. Non-motorized recreationists, alternatively, select ridges with steeper slope and greater terrain variability, at higher elevations and with less open canopy and more snow (Fig. 6), traits consistent with skiing down steep, treed slopes. Differences in environmental characteristics used by each recreation type may provide useful guidelines on determining whether to zone certain areas for motorized or non-motorized use only, while still providing each type of recreation the environmental characteristics they prefer. Areas of steep slope, for instance, may be set aside for backcountry skiers or hybrid-skiers with little effect to snowmobilers, since they prefer more ﬂat terrain. Modeled areas of overlap also have implications for conﬂict between recreation and species of conservation concern. Motorized winter recreation creates increased noise and engine emissions which can negatively impact wildlife (Shively et al., 2008; Zielinski, Slauson, & Bowles, 2008), while non-motorized forms may displace wildlife (Krebs, Lofroth, & Parﬁtt, 2007; Reimers, Eftestøl, & Colman, 2003) or contribute to habitat loss through the construction of recreation infrastructure (Sato, Wood, & Lindenmayer, 2013). Wildlife may also respond differently to motorized versus non-motorized types of winter recreation (Larson, Reed, Merenlender, & Crooks, 2016); Reimers et al. (2003) found that reindeer (Rangifer tarandus tarandus) detected snowmobiles sooner than skiers, but responded to skiers by moving greater distances than from snowmobiles, and Seip, Johnson, and Watts (2007) found threatened woodland caribou strongly avoided motorized snowmobile recreation over huge areas. The spatial depiction of relative recreation probability (Appendix C: Figs C.1-C.5) generated by our models provides detailed maps which can be used to determine the likelihood of motorized or non-motorized forms of recreation in a given area. The use of a modeled RSF allows managers to consider the relative probability of a speciﬁc type of recreation co-occurring with a given species, and thus will allow decisions to be tailored for species that differ in sensitivity to different types of recreation. 5. Conclusions The sharp increase in the extent and popularity of winter recreation presents a challenge to land managers responsible for multiple-use lands (Bowker et al., 2012), with associated concern as to its impact on wildlife and the environment (Arlettaz et al., 2015; �L.E. Olson et al. / Applied Geography 86 (2017) 66e91 Braunisch et al., 2011; Patthey, Wirthner, Signorell, & Arlettaz, 2008). Thus, managers face multiple challenges of reducing impacts to the environment and wildlife while also minimizing interpersonal conﬂict and still providing winter recreation opportunities. One way in which the likelihood of interpersonal conﬂict may be minimized is to reduce the time that motorized and nonmotorized users are funneled into a single shared-use access area or travel corridor, since our results show that the conditions that motorized and non-motorized users select are fairly distinct, and thus recreationists may self-select areas that reduce co-occurrence between the two types. Alternatively, if active zoning is required to separate users to reduce conﬂict or for safety, the conditions that each recreation type favors should be considered. Our results underscore the importance of road and road-access management in affecting the spatial footprint of winter recreation. Decisions about the placement or density of roads need careful assessment as they can inﬂuence the movements of winter recreationists relative to wildlife or each other. Management practices that lower tree density and increase forest patchiness will also inﬂuence motorized and non-motorized recreation at ﬁne spatial scales. Funding This work was supported by the U.S. Department of Agriculture, U.S. Forest Service, and White River National Forest. Additional funding and support was provided by the Rocky Mountain Research Covariate Distance to Highway (km) Elevation (m) Forest Edge (km/km2) Percent Canopy Cover Percent Evergreen Forest Average Annual Precipitation (mm) Forest Road Density (km/km2) Slope (degrees) Activity Snmb On-Tr Snmb Off-Tr Hybrid Snmb Hybrid Ski Ski Snmb On-Tr Snmb Off-Tr Hybrid Snmb Hybrid Ski Ski Snmb On-Tr Snmb Off-Tr Hybrid Snmb Hybrid Ski Ski/Board Snmb On-Tr Snmb Off-Tr Hybrid Snmb Hybrid Ski Ski/Board Snmb On-Tr Snmb Off-Tr Hybrid Snmb Hybrid Ski Ski/Board Snmb On-Tr Snmb Off-Tr Hybrid Snmb Hybrid Ski Ski/Board Snmb On-Tr Snmb Off-Tr Hybrid Snmb Hybrid Ski Ski/Board Snmb On-Tr Snmb Off-Tr Hybrid Snmb 77 Station, Vail Associates Inc., Colorado BLM state ofﬁce, U.S. Forest Service Region 2 Regional Ofﬁce Renewable Resources Department, 10th Mountain Huts, and Colorado Department of Transportation. Acknowledgements We thank W. George for valuable assistance with preliminary data analysis, the many ﬁeld technicians that distributed GPS units to recreationists, the participants who volunteered to carry the GPS units, the outﬁtters and guides who agreed to carry them, and the local FS ofﬁces for providing logistical support and information about the area. Appendix A. Table A.1 The mean and 95% conﬁdence intervals of all used and available GPS points for each environmental covariate (see Table 1 in manuscript for more covariate information) at the 2500 m scale used to model winter recreation selection in western Colorado, USA, from 2010 to 2013. Summaries for each winter recreation activity, on-trail snowmobile (Snmb On-Tr), off-trail snowmobile (Snmb Off-Tr), snowmobile segments of snowmobile-assisted hybrid skiing (Hybrid Snmb), ski segments of snowmobileassisted hybrid skiing (Hybrid Ski), and back-country ski or snowboard (Ski), are provided to allow comparison between recreation types within a given covariate. Used Points Available Points Mean 95% CI Mean 95% CI 3.41 3.38 4.05 4.61 2.46 3208.27 3395.73 3408.50 3425.94 3375.08 3.77 4.11 3.57 3.35 3.98 37.88 35.25 33.88 32.48 31.26 51.54 51.08 50.84 51.31 44.58 82.77 90.49 84.65 84.94 90.17 1.19 0.65 0.92 0.91 0.62 15.97 14.70 16.28 3.19e3.62 3.19e3.57 3.9e4.21 4.48e4.74 2.35e2.56 3190.77e3225.77 3383.01e3408.45 3399.38e3417.61 3417.01e3434.86 3366.35e3383.82 3.68e3.86 4.03e4.18 3.47e3.66 3.28e3.43 3.92e4.04 37.17e38.58 34.58e35.92 32.99e34.78 31.51e33.44 30.7e31.83 50.23e52.85 49.82e52.35 49.61e52.07 49.97e52.66 43.56e45.59 81.75e83.8 89.58e91.4 83.84e85.46 84.04e85.84 89.14e91.21 1.12e1.26 0.6e0.71 0.87e0.97 0.86e0.95 0.6e0.65 15.67e16.27 14.46e14.95 15.99e16.57 4.35 4.37 3.59 3.58 4.95 3246.61 3246.23 3278.60 3279.75 3298.73 3.22 3.22 3.26 3.26 3.16 34.12 34.03 36.34 36.31 33.50 46.92 46.73 53.33 53.30 46.01 79.00 79.18 71.93 71.98 84.04 0.52 0.52 0.63 0.63 0.54 18.00 18.03 16.71 4.05e4.65 4.07e4.67 3.28e3.9 3.27e3.89 4.7e5.19 3222.78e3270.45 3222.29e3270.18 3254.29e3302.91 3255.34e3304.17 3282.91e3314.55 3.13e3.32 3.13e3.32 3.14e3.38 3.14e3.39 3.09e3.23 33.15e35.09 33.06e35 35.11e37.58 35.08e37.54 32.75e34.25 45.12e48.73 44.92e48.53 51.03e55.63 51e55.6 44.63e47.39 77.44e80.56 77.61e80.75 70.4e73.45 70.45e73.5 82.77e85.3 0.47e0.58 0.47e0.58 0.59e0.67 0.59e0.67 0.5e0.57 17.56e18.45 17.58e18.48 16.22e17.2 (continued on next page) �78 L.E. Olson et al. / Applied Geography 86 (2017) 66e91 (continued ) Covariate Roughnessa Mean Annual Temperature (oC) Topographic Position Index (TPIb) a b Activity Hybrid Ski Ski/Board Snmb On-Tr Snmb Off-Tr Hybrid Snmb Hybrid Ski Ski/Board Snmb On-Tr Snmb Off-Tr Hybrid Snmb Hybrid Ski Ski/Board Snmb On-Tr Snmb Off-Tr Hybrid Snmb Hybrid Ski Ski/Board Used Points Available Points Mean 95% CI Mean 95% CI 17.26 18.31 1.003 1.001 1.001 1.002 1.003 1.59 0.69 0.48 0.29 1.06 �50.08 28.52 11.94 63.15 �18.06 16.99e17.53 17.98e18.64 1.002e1.003 1.001e1.002 1.001e1.002 1.002e1.002 1.003e1.003 1.5e1.68 0.61e0.77 0.4e0.55 0.21e0.36 1.01e1.11 �60.07e�40.09 21.27e35.78 �1.89e25.76 49.91e76.39 �27.13e�8.99 16.74 18.88 1.004 1.004 1.003 1.003 1.004 1.25 1.25 0.93 0.92 1.20 �9.44 �9.83 �10.32 �9.54 1.11 16.25e17.23 18.54e19.22 1.004e1.004 1.004e1.004 1.003e1.004 1.003e1.004 1.004e1.005 1.13e1.36 1.13e1.37 0.79e1.07 0.78e1.06 1.11e1.3 �20.55e1.67 �20.97e1.32 �23.97e3.33 �23.21e4.13 �7.98e10.2 Higher values represent greater terrain variability. Negative values indicate drainages, positive indicate ridges. Fig A.1. Mean and 95% CI summaries of environmental characteristics at used and random locations of each recreation activity at both study areas in Colorado, USA. Plots shown are distance to highway (km), road density (km/km2), percent canopy closure (%), and slope (degrees). �L.E. Olson et al. / Applied Geography 86 (2017) 66e91 79 Appendix B. Model selection results showing the top 10 models from resource selection functions (RSF) for each recreation type studied in western Colorado, USA from 2010 to 2013. Table B.1 Model selection table for on-trail snowmobile RSF models showing habitat selection of winter recreationists driving snowmobiles on trails. Only the top 10 models are shown. K is the number of model parameters, LL is model log likelihood. The scale at which the covariate was measured (in meters) is given in subscript numbers; covariates included as quadratics are indicated with a superscript ‘2’. Further information on environmental covariates is given in Table 1 of the manuscript. Model covariates K AIC D AIC AIC Wt LL Highway2500 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ Evergreen2500 þ Evergreen22500 þ North500 þ Precip þ Precip2 þ RdDensity125 þ Slope1250 þ Slope21250 þ Roughness500 þ Temp þ TPI500 Highway2500 þ ForestEdge125 þ Canopy125 þ Evergreen2500 þ Evergreen22500 þ North500 þ Precip þ Precip2 þ RdDensity125 þ Slope1250 þ Slope21250 þ Roughness500 þ Temp þ TPI500 Highway2500 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ Evergreen2500 þ Evergreen22500 þ North2500 þ Precip þ Precip2 þ RdDensity125 þ Slope1250 þ Slope21250 þ Roughness500 þ Temp þ TPI500 Highway2500 þ ForestEdge125 þ Canopy125 þ Evergreen2500 þ Evergreen22500 þ North2500 þ Precip þ Precip2 þ RdDensity125 þ Slope1250 þ Slope21250 þ Roughness500 þ Temp þ TPI500 Highway2500 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ Evergreen2500 þ Evergreen22500 þ Precip þ Precip2 þ RdDensity125 þ Slope1250 þ Slope21250 þ Roughness500 þ Temp þ TPI500 Highway2500 þ ForestEdge125 þ Canopy125 þ Evergreen2500 þ Evergreen22500 þ Precip þ Precip2 þ RdDensity125 þ Slope1250 þ Slope21250 þ Roughness500 þ Temp þ TPI500 Highway2500 þ Elevation2500 þ Elevation22500 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ Evergreen2500 þ Evergreen22500 þ North500 þ Precip þ Precip2 þ RdDensity125 þ Slope1250 þ Slope21250 þ Roughness500 þ TPI500 Highway2500 þ Elevation2500 þ Elevation22500 þ ForestEdge125 þ Canopy125 þ Evergreen2500 þ Evergreen22500 þ North500 þ Precip þ Precip2 þ RdDensity125 þ Slope1250 þ Slope21250 þ Roughness500 þ TPI500 Highway2500 þ Canopy125 þ Canopy2125 þ Evergreen2500 þ Evergreen22500 þ North500 þ Precip þ Precip2 þ RdDensity125 þ Slope1250 2 þ Slope1250 þ Roughness500 þ Temp þ TPI500 Highway2500 þ ForestEdge2500 þ Canopy125 þ Canopy2125 þ Evergreen2500 þ Evergreen22500 þ North500 þ Precip þ Precip2 þ RdDensity125 þ Slope1250 þ Slope21250 þ Roughness500 þ Temp þ TPI500 17 212,588.3 0 1 �106277 16 212,617.5 29.19 0 �106293 17 212,749.7 161.46 0 �106358 16 212,762.1 173.85 0 �106365 16 212,877.4 289.12 0 �106423 15 212,889.9 301.58 0 �106430 18 213,091.4 503.11 0 �106528 17 213,110.5 522.24 0 �106538 16 213,128.7 540.46 0 �106548 17 213,131.1 542.83 0 �106549 �80 L.E. Olson et al. / Applied Geography 86 (2017) 66e91 Table B.2 Model selection table for off-trail snowmobile RSF models showing habitat selection of winter recreationists driving snowmobiles on off-trail play areas. Only the top 10 models are shown. K is the number of model parameters, LL is model log likelihood. The scale at which the covariate was measured (in meters) is given in subscript numbers; covariates included as quadratics are indicated with a superscript ‘2’. Further information on environmental covariates is given in Table 1 of the manuscript. Model Covariates K AIC D AIC AIC Wt LL Highway2500 þ Elevation125 þ Elevation2125 þ ForestEdge2500 þ Canopy2500 þ Canopy22500 þ Evergreen125 þ Evergreen2125 þ North2500 þ Precip þ Precip2 þ RdDensity125 þ Slope125 þ Roughness2500 þ TPI500 Highway2500 þ Elevation125 þ ForestEdge2500 þ Canopy2500 þ Canopy22500 þ Evergreen125 þ Evergreen2125 þ North2500 þ Precip þ Precip2 þ RdDensity125 þ Slope125 þ Roughness2500 þ TPI500 Highway2500 þ Elevation125 þ Elevation2125 þ ForestEdge500 þ Canopy2500 þ Canopy22500 þ Evergreen125 þ Evergreen2125 þ North2500 þ Precip þ Precip2 þ RdDensity125 þ Slope125 þ Roughness2500 þ TPI500 Highway2500 þ Elevation125 þ Elevation2125 þ ForestEdge2500 þ Canopy2500 þ Canopy22500 þ Evergreen125 þ Evergreen2125 þ North2500 þ Precip þ Precip2 þ RdDensity2500 þ Slope125 þ Roughness2500 þ TPI500 Highway2500 þ Elevation125 þ ForestEdge500 þ Canopy2500 þ Canopy22500 þ Evergreen125 þ Evergreen2125 þ North2500 þ Precip þ Precip2 þ RdDensity125 þ Slope125 þ Roughness2500 þ TPI500 Highway2500 þ Elevation125 þ ForestEdge2500 þ Canopy2500 þ Canopy22500 þ Evergreen125 þ Evergreen2125 þ North2500 þ Precip þ Precip2 þ RdDensity2500 þ Slope125 þ Roughness2500 þ TPI500 Highway2500 þ Elevation125 þ Elevation2125 þ ForestEdge500 þ Canopy2500 þ Canopy22500 þ Evergreen125 þ Evergreen2125 þ Precip þ Precip2 þ RdDensity125 þ Slope125 þ Roughness2500 þ TPI500 Highway2500 þ Elevation125 þ Elevation2125 þ ForestEdge500 þ Canopy2500 þ Canopy22500 þ Evergreen125 þ Evergreen2125 þ North500 þ Precip þ Precip2 þ RdDensity125 þ Slope125 þ Roughness2500 þ TPI500 Highway2500 þ Elevation125 þ Elevation2125 þ ForestEdge500 þ Canopy2500 þ Canopy22500 þ Evergreen125 þ Evergreen2125 þ North2500 þ Precip þ Precip2 þ RdDensity2500 þ Slope125 þ Roughness2500 þ TPI500 Highway2500 þ Elevation125 þ ForestEdge500 þ Canopy2500 þ Canopy22500 þ Evergreen125 þ Evergreen2125 þ North2500 þ Precip þ Precip2 þ RdDensity2500 þ Slope125 þ Roughness2500 þ TPI500 17 52,437.11 0 1 �26201.6 16 52,484.34 47.23 0 �26226.2 17 52,530.61 93.49 0 �26248.3 17 52,581.75 144.64 0 �26273.9 16 52,615.75 178.64 0 �26291.9 16 52,619.33 182.22 0 �26293.7 16 52,681.12 244 0 �26324.6 17 52,681.75 244.63 0 �26323.9 17 52,685.36 248.25 0 �26325.7 16 52,758.39 321.28 0 �26363.2 �Table B.3 Model selection table for hybrid snowmobile RSF models showing habitat selection of winter recreationists driving snowmobiles while engaging in hybrid-assisted skiing. Only the top 10 models are shown. K is the number of model parameters, LL is model log likelihood. The scale at which the covariate was measured (in meters) is given in subscript numbers; covariates included as quadratics are indicated with a superscript ‘2’. Further information on environmental covariates is given in Table 1 of the manuscript. Model Covariates K þ ForestEdge125 þ Canopy2500 þ Canopy22500 AIC Wt LL 1 �47934.7 16 96,122.37 220.98 0 �48045.2 2 þ Canopy2500 þ Canopy2500 þ Evergreen125 þ North2500 þ Precip þ RdDensity125 þ Slope1250 þ Roughness2500 þ Temp þ Temp2 þ TPI2500 þ ForestEdge125 þ Canopy2500 þ Canopy22500 þ Evergreen125 þ North2500 þ Precip þ RdDensity125 þ Slope1250 þ Roughness2500 þ 15 96,129.87 228.48 0 16 96,376.94 475.55 0 �48049.9 �48172.5 þ ForestEdge2500 þ Canopy2500 þ Canopy22500 þ Evergreen125 þ North2500 þ Precip þ RdDensity125 þ Slope1250 þ Roughness2500 þ 16 96,519.77 618.38 0 �48243.9 15 15 15 14 14 �48252.3 �48264.2 �48307.5 �48314.4 �49449.4 þ ForestEdge2500 þ Canopy2500 þ þ þ þ þ þ Canopy22500 þ Evergreen125 þ North2500 þ Precip þ RdDensity125 þ Slope1250 þ Roughness2500 þ D AIC AIC Canopy22500 16 95,901.39 0 þ Evergreen125 þ North2500 þ Precip þ RdDensity125 þ Slope1250 þ Roughness2500 þ 2 Canopy2500 þ þ Evergreen125 þ North2500 þ Precip þ RdDensity125 þ Slope1250 þ Roughness2500 þ Temp þ Temp þ TPI500 ForestEdge125 þ Canopy2500 þ Canopy22500 þ Evergreen125 þ North2500 þ Precip þ RdDensity125 þ Slope1250 þ Roughness2500 þ Temp þ Temp2 2 ForestEdge2500 þ Canopy2500 þ Canopy2500 þ Evergreen125 þ North2500 þ Precip þ RdDensity125 þ Slope1250 þ Roughness2500 þ Temp þ Temp2 Canopy2500 þ Canopy22500 þ Evergreen125 þ North2500 þ Precip þ RdDensity125 þ Slope1250 þ Roughness2500 þ Temp þ Temp2 ForestEdge2500 þ Canopy2500 þ Evergreen125 þ North2500 þ Precip þ RdDensity125 þ Slope1250 þ Roughness2500 þ Temp þ Temp2 96,534.58 96,558.35 96,645.06 96,656.74 98,926.79 633.19 656.96 743.67 755.35 3025.4 0 0 0 0 0 Table B.4 Model selection table for hybrid ski RSF models showing habitat selection of winter recreationists skiing downhill while engaging in hybrid-assisted skiing. Only the top 10 models are shown. K is the number of model parameters, LL is model log likelihood. The scale at which the covariate was measured (in meters) is given in subscript numbers; covariates included as quadratics are indicated with a superscript ‘2’. Further information on environmental covariates is given in Table 1 of the manuscript. Model Covariates K AIC D AIC AIC Wt LL Highway1250 þ Highway21250 þ ForestEdge2500 þ Canopy125 þ North2500 þ Precip þ RdDensity1250 þ Slope125 þ Slope2125 þ Roughness2500 þ Temp þ Temp2 þ TPI2500 Highway1250 þ Highway21250 þ ForestEdge2500 þ Canopy125 þ North2500 þ Precip þ RdDensity1250 þ Slope125 þ Slope2125 þ Roughness2500 þ Temp þ Temp2 Highway1250 þ Highway21250 þ ForestEdge2500 þ Canopy125 þ North2500 þ Precip þ RdDensity1250 þ Slope125 þ Slope2125 þ Roughness2500 þ Temp þ Temp2 þ TPI500 Highway1250 þ Highway21250 þ ForestEdge2500 þ Evergreen125 þ North2500 þ Precip þ RdDensity1250 þ Slope125 þ Slope2125 þ Roughness2500 þ Temp þ Temp2 þ TPI500 Highway1250 þ Highway21250 þ ForestEdge500 þ Canopy125 þ North2500 þ Precip þ RdDensity1250 þ Slope125 þ Slope2125 þ Roughness2500 þ Temp þ Temp2 þ TPI500 Highway1250 þ Highway21250 þ ForestEdge2500 þ Evergreen125 þ North2500 þ Precip þ RdDensity1250 þ Slope125 þ Slope2125 þ Roughness2500 þ Temp þ Temp2 Highway1250 þ Highway21250 þ ForestEdge2500 þ Evergreen125 þ North2500 þ Precip þ RdDensity1250 þ Slope125 þ Slope2125 þ Roughness2500 þ Temp þ Temp2 þ TPI2500 Highway125 þ Highway2125 þ Highway1250 þ Highway21250 þ ForestEdge2500 þ Canopy125 þ North2500 þ Precip þ RdDensity1250 þ Slope125 þ Slope2125 þ Roughness2500 þ Temp þ Temp2 þ TPI2500 Highway1250 þ Highway21250 þ ForestEdge500 þ Canopy125 þ North2500 þ Precip þ RdDensity1250 þ Slope125 þ Slope2125 þ Roughness2500 þ Temp þ Temp2 Highway1250 þ Highway21250 þ ForestEdge500 þ Canopy125 þ North2500 þ Precip þ RdDensity1250 þ Slope125 þ Slope2125 þ Roughness2500 þ Temp þ Temp2 þ TPI2500 15 14 15 15 15 14 15 15 10,971.81 10,978.98 10,980.81 11,015.75 11,035.22 11,037.28 11,037.46 11,045.83 0.00 7.18 9.00 43.94 63.41 65.47 65.65 74.02 0.96 0.03 0.01 0.00 0.00 0.00 0.00 0.00 �5470.90 �5475.49 �5475.40 �5492.88 �5502.61 �5504.64 �5503.73 �5507.92 14 15 11,049.88 11,050.23 78.07 78.42 0.00 0.00 �5510.94 �5510.12 L.E. Olson et al. / Applied Geography 86 (2017) 66e91 Highway1250 þ Highway21250 Temp þ Temp2 þ TPI2500 Highway1250 þ Highway21250 Temp þ Temp2 þ TPI2500 2 Highway1250 þ Highway1250 Highway1250 þ Highway21250 Temp þ Temp2 þ TPI500 Highway1250 þ Highway21250 Temp þ Temp2 þ TPI500 Highway1250 þ Highway21250 Highway1250 þ Highway21250 2 Highway1250 þ Highway1250 Highway1250 þ Highway21250 Highway1250 þ Highway21250 81 �82 Table B.5 Model selection table for backcountry ski RSF models showing habitat selection of winter recreationists engaged in backcountry skiing or snowboarding. Only the top 10 models are shown. K is the number of model parameters, LL is model log likelihood. The scale at which the covariate was measured (in meters) is given in subscript numbers; covariates included as quadratics are indicated with a superscript ‘2’. Further information on environmental covariates is given in Table 1 of the manuscript. K AIC D AIC AIC Wt LL Highway2500 þ ForestEdge2500 þ Canopy2500 þ Canopy22500 þ Evergreen500 þ Evergreen2500 þ North500 þ Precip þ RdDensity125 þ Roughness125 þ Temp þ Temp2 þ TPI125 Highway2500 þ ForestEdge2500 þ Evergreen500 þ Evergreen2500 þ North500 þ Precip þ RdDensity125 þ Roughness125 þ Temp þ Temp2 þ TPI125 Highway2500 þ ForestEdge2500 þ Canopy2500 þ Evergreen500 þ Evergreen2500 þ North500 þ Precip þ RdDensity125 þ Roughness125 þ Temp þ Temp2 þ TPI125 Highway2500 þ ForestEdge2500 þ Canopy2500 þ Canopy22500 þ Evergreen500 þ Evergreen2500 þ North1250 þ Precip þ RdDensity125 þ Roughness125 þ Temp þ Temp2 þ TPI125 2 2 þ Evergreen500 þ Evergreen500 þ Precip þ Highway2500 þ ForestEdge2500 þ Canopy2500 þ Canopy2500 RdDensity125 þ Roughness125 þ Temp þ Temp2 þ TPI125 Highway2500 þ ForestEdge2500 þ Canopy2500 þ Canopy22500 þ Evergreen500 þ Evergreen2500 þ North500 þ Precip þ RdDensity125 þ Roughness125 þ Temp þ Temp2 þ TPI2500 Highway2500 þ ForestEdge2500 þ Evergreen1250 þ Evergreen21250 þ North500 þ Precip þ RdDensity125 þ Roughness125 þ Temp þ Temp2 þ TPI125 Highway2500 þ ForestEdge2500 þ Canopy2500 þ Canopy22500 þ Evergreen500 þ Evergreen2500 þ North500 þ Precip þ RdDensity125 þ Roughness125 þ Temp þ Temp2 Highway2500 þ ForestEdge2500 þ Canopy2500 þ Canopy22500 þ Evergreen500 þ Evergreen2500 þ North1250 þ Precip þ RdDensity125 þ Roughness125 þ Temp þ Temp2 þ TPI2500 Highway2500 þ ForestEdge2500 þ Canopy2500 þ Canopy22500 þ Evergreen500 þ Evergreen2500 þ North1250 þ Precip þ RdDensity125 þ Roughness125 þ Temp þ Temp2 15 90,376.84 0 1 �45173.4 13 90,522.59 145.75 0 �45248.3 14 90,523.19 146.35 0 �45247.6 15 90,525.74 148.9 0 �45247.9 14 90,573.21 196.37 0 �45272.6 15 90,626.04 249.2 0 �45298 13 90,640.83 263.98 0 �45307.4 14 90,651.89 275.05 0 �45312 15 90,764.08 387.24 0 �45367 14 90,784.79 407.95 0 �45378.4 L.E. Olson et al. / Applied Geography 86 (2017) 66e91 Model Covariates �L.E. Olson et al. / Applied Geography 86 (2017) 66e91 83 Appendix C. Mapped spatial predictions of selection for each type of winter recreation modeled with resource selection functions within the elevation range of winter recreation (2300 me4250 m) in western Colorado, 2010e2013. Figure C.1. Predicted probabilities of selection from the resource selection function model for on-trail snowmobile recreation across western Colorado. Warm colors indicate higher probability of selection, cool colors indicate an area is less likely to be selected. �84 L.E. Olson et al. / Applied Geography 86 (2017) 66e91 Figure C.2. Predicted probabilities of selection from the resource selection function model for off-trail snowmobile recreation across western Colorado. Warm colors indicate higher probability of selection, cool colors indicate an area is less likely to be selected. �L.E. Olson et al. / Applied Geography 86 (2017) 66e91 85 Figure C.3. Predicted probabilities of selection from the resource selection function model for hybrid snowmobile recreation across western Colorado. Warm colors indicate higher probability of selection, cool colors indicate an area is less likely to be selected. �86 L.E. Olson et al. / Applied Geography 86 (2017) 66e91 Figure C.4. Predicted probabilities of selection from the resource selection function model for hybrid ski recreation across western Colorado. Warm colors indicate higher probability of selection, cool colors indicate an area is less likely to be selected. �L.E. Olson et al. / Applied Geography 86 (2017) 66e91 87 Figure C.5. Predicted probabilities of selection from the resource selection function model for backcountry ski recreation across western Colorado. Warm colors indicate higher probability of selection, cool colors indicate an area is less likely to be selected. �88 L.E. Olson et al. / Applied Geography 86 (2017) 66e91 Appendix D. Model selection results showing the top 10 models from step selection functions (SSF) for each recreation type studied in western Colorado, USA from 2010 to 2013. Table D.1 Model selection table for on-trail snowmobile SSF models showing selection of movement paths by winter recreationists driving snowmobiles on trails. Only the top 10 models are shown. K is the number of model parameters, LL is model log likelihood. The scale at which the covariate was measured (in meters) is given in subscript numbers; covariates included as quadratics are indicated with a superscript ‘2’. Further information on environmental covariates is given in Table 1 of the manuscript. Model Covariates K AIC D AIC AIC Wt LL Elevation125 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ North125 þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ TPI125 Elevation125 þ Elevation2125 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ North125 þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ TPI125 Elevation125 þ Canopy125 þ Canopy2125 þ North125 þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ TPI125 2 þ North125 þ RdDensity125 þ Elevation125 þ ForestEdge500 þ Canopy125 þ Canopy125 RdDensity2125 þ Slope500 þ Roughness125 þ TPI125 Elevation125 þ Elevation2125 þ Canopy125 þ Canopy2125 þ North125 þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ TPI125 Elevation125 þ Elevation2125 þ ForestEdge500 þ Canopy125 þ Canopy2125 þ North125 þ RdDensity125 þ RdDensity2125 þ Slope500 þ 0Roughness125 þ TPI125 Elevation125 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ North125 þ RdDensity125 þ 2 RdDensity125 þ Roughness125 þ TPI125 Elevation125 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ TPI125 Elevation125 þ Elevation2125 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ North125 þ RdDensity125 þ RdDensity2125 þ Roughness125 þ TPI125 Elevation125 þ Elevation2125 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ TPI125 10 340,845.00 0.00 0.59 �170,412.50 11 340,847.00 1.99 0.22 �170,412.50 9 340,849.20 4.17 0.07 �170,415.60 10 340,849.90 4.80 0.05 �170,414.90 10 340,851.20 6.17 0.03 �170,415.60 11 340,851.80 6.80 0.02 �170,414.90 9 340,853.80 8.78 0.01 �170,417.90 9 340,853.90 8.88 0.01 �170,418.00 10 340,855.60 10.60 0.00 �170,417.80 10 340,855.90 10.90 0.00 �170,418.00 Table D.2 Model selection table for off-trail snowmobile SSF models showing selection of movement paths by winter recreationists driving snowmobiles on off-trail play areas. Only the top 10 models are shown. K is the number of model parameters, LL is model log likelihood. The scale at which the covariate was measured (in meters) is given in subscript numbers; covariates included as quadratics are indicated with a superscript ‘2’. Further information on environmental covariates is given in Table 1 of the manuscript. Model Covariates K AIC D AIC AIC Wt LL Highway500 þ Elevation125 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ North125 þ RdDensity125 þ Slope500 þ Roughness125 þ TPI125 Highway125 þ Elevation125 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ North125 þ RdDensity125 þ Slope500 þ Roughness125 þ TPI125 Highway500 þ Elevation500 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ North125 þ RdDensity125 þ Slope500 þ Roughness125 þ TPI þ Highway125 þ Elevation500 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ North125 þ RdDensity125 þ Slope500 þ Roughness125 þ TPI þ Highway500 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ North125 þ RdDensity125 þ Slope500 þ Roughness125 þ TPI þ 2 þ North125 þ RdDensity125 þ Highway125 þ ForestEdge125 þ Canopy125 þ Canopy125 Slope500 þ Roughness125 þ TPI þ Elevation125 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ North125 þ RdDensity125 þ Slope500 þ Roughness125 þ TPI þ Highway500 þ Elevation125 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ RdDensity125 þ Slope500 þ Roughness125 þ TPI Elevation500 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ North125 þ RdDensity125 þ Slope500 þ Roughness125 þ TPI þ Highway125 þ Elevation125 þ ForestEdge125 þ Canopy125 þ Canopy2125 þ RdDensity125 þ Slope500 þ Roughness125 þ TPI þ 10 102,826.30 0.00 0.42 �51403.15 10 102,826.90 0.64 0.31 �51403.46 10 102,828.50 2.24 0.14 �51404.26 10 102,829.10 2.86 0.10 �51404.57 9 102,834.00 7.70 0.01 �51408.00 9 102,834.40 8.15 0.01 �51408.22 9 102,834.50 8.25 0.01 �51408.27 9 102,836.20 9.94 0.00 �51409.12 9 102,836.30 9.98 0.00 �51409.14 9 102,836.90 10.60 0.00 �51409.43 �L.E. Olson et al. / Applied Geography 86 (2017) 66e91 89 Table D.3 Model selection table for hybrid snowmobile SSF models showing selection of movement paths by winter recreationists driving snowmobiles while engaging in hybridassisted skiing. Only the top 10 models are shown. K is the number of model parameters, LL is model log likelihood. The scale at which the covariate was measured (in meters) is given in subscript numbers; covariates included as quadratics are indicated with a superscript ‘2’. Further information on environmental covariates is given in Table 1 of the manuscript. Model Covariates K AIC D AIC AIC Wt LL Highway125 þ ForestEdge500 þ Canopy125 þ Canopy2125 þ North500 þ Precip þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 Highway500 þ ForestEdge500 þ Canopy125 þ Canopy2125 þ North500 þ Precip þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 Highway125 þ ForestEdge500 þ Canopy125 þ Canopy2125 þ North500 þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 Highway125 þ Elevation500 þ ForestEdge500 þ Canopy125 þ Canopy2125 þ North500 þ Precip þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ Roughness2125 þ TPI500 þ TPI2500 Highway125 þ Elevation500 þ Elevation2500 þ ForestEdge500 þ Canopy125 þ Canopy2125 þ North500 þ Precip þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ Roughness2125 þ TPI500 þ TPI2500 Highway125 þ ForestEdge500 þ Canopy125 þ Canopy2125 þ North500 þ Precip þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ Roughness2125 þ TPI500 þ TPI2500 ForestEdge500 þ Canopy125 þ Canopy2125 þ North500 þ Precip þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 Highway500 þ ForestEdge500 þ Canopy125 þ Canopy2125 þ North500 þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 Highway500 þ Elevation500 þ ForestEdge500 þ Canopy125 þ Canopy2125 þ North500 þ Precip þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ Roughness2125 þ TPI500 þ TPI2500 Highway500 þ ForestEdge500 þ Canopy125 þ Canopy2125 þ North500 þ Precip þ RdDensity125 þ RdDensity2125 þ Slope500 þ Roughness125 þ Roughness2125 þ TPI500 þ TPI2500 14 333,152.20 0.00 1.00 �166,562.10 14 333,169.40 17.18 0.00 �166,570.70 13 333,204.50 52.22 0.00 �166,589.20 14 333,214.40 62.15 0.00 �166,593.20 15 333,216.00 63.76 0.00 �166,593.00 13 333,216.20 63.94 0.00 �166,595.10 13 333,216.30 64.05 0.00 �166,595.10 13 333,223.30 71.03 0.00 �166,598.60 14 333,228.40 76.13 0.00 �166,600.20 13 333,230.00 77.74 0.00 �166,602.00 Table D.4 Model selection table for hybrid ski SSF models showing selection of movement paths by winter recreationists skiing downhill while engaging in hybrid-assisted skiing. Only the top 10 models are shown. K is the number of model parameters, LL is model log likelihood. The scale at which the covariate was measured (in meters) is given in subscript numbers; covariates included as quadratics are indicated with a superscript ‘2’. Further information on environmental covariates is given in Table 1 of the manuscript. Model Covariates K AIC D AIC AIC Wt LL ForestEdge125 þ Canopy125 þ Canopy2125 þ North500 þ Precip þ RdDensity500 þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 ForestEdge125 þ Canopy125 þ Canopy2125 þ North500 þ Precip þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 ForestEdge125 þ Canopy125 þ Canopy2125 þ North500 þ Precip þ RdDensity125 þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 ForestEdge500 þ Canopy125 þ Canopy2125 þ North500 þ Precip þ RdDensity500 þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 ForestEdge500 þ Canopy125 þ Canopy2125 þ North500 þ Precip þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 ForestEdge500 þ Canopy125 þ Canopy2125 þ North500 þ Precip þ RdDensity125 þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 Canopy125 þ Canopy2120 þ North500 þ Precip þ RdDensity500 þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 ForestEdge125 þ Canopy125 þ Canopy2125 þ Precip þ RdDensity500 þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 Canopy125 þ Canopy2125 þ North500 þ Precip þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 Canopy125 þ Canopy2125 þ North500 þ Precip þ RdDensity125 þ Slope500 þ Roughness125 þ Roughness2125 þ Temp þ TPI500 þ TPI2500 12 37,666.77 0.00 0.67 �18821.38 11 37,669.51 2.73 0.17 �18823.75 12 37,670.83 4.06 0.09 �18823.41 12 37,673.18 6.41 0.03 �18824.59 11 37,673.20 6.43 0.03 �18825.60 12 37,674.45 7.68 0.01 �18825.22 11 37,684.50 17.72 0.00 �18831.25 11 37,685.77 18.99 0.00 �18831.88 10 11 37,686.15 37,687.52 19.38 20.75 0.00 0.00 �18833.07 �18832.76 �90 L.E. Olson et al. / Applied Geography 86 (2017) 66e91 Table D.5 Model selection table for backcountry ski SSF models showing selection of movement paths by winter recreationists engaged in backcountry skiing. Only the top 10 models are shown. K is the number of model parameters, LL is model log likelihood. The scale at which the covariate was measured (in meters) is given in subscript numbers; covariates included as quadratics are indicated with a superscript ‘2’. Further information on environmental covariates is given in Table 1 of the manuscript. K AIC D AIC AIC Wt LL þ 14 195,195.80 0.00 0.55 �97583.89 þ 13 195,196.80 1.00 0.33 �97585.39 13 195,200.30 4.54 0.06 �97587.16 12 195,201.70 5.96 0.03 �97588.87 þ 14 195,202.30 6.50 0.02 �97587.14 þ 13 195,203.70 7.91 0.01 �97588.85 13 195,207.50 11.67 0.00 �97590.73 12 195,210.80 15.00 0.00 �97593.40 12 195,212.50 16.75 0.00 �97594.27 13 195,214.50 18.69 0.00 �97594.24 Model Covariates Highway500 þ Highway2500 þ Elevation500 þ Elevation2500 þ ForestEdge500 þ Canopy500 þ Canopy2500 Evergreen125 þ Evergreen2125 þ Precip þ RdDensity125 þ Roughness125 þ TPI500 þ TPI2500 Highway500 þ Highway2500 þ Elevation500 þ Elevation2500 þ ForestEdge500 þ Canopy500 þ Canopy2500 Evergreen125 þ Evergreen2125 þ RdDensity125 þ Roughness125 þ TPI500 þ TPI2500 Highway500 þ Highway2500 þ Elevation500 þ Elevation2500 þ Canopy500 þ Canopy2500 þ Evergreen125 þ Evergreen2125 þ Precip þ RdDensity125 þ Roughness125 þ TPI500 þ TPI2500 Highway500 þ Highway25002 þ Elevation500 þ Elevation2500 þ Canopy500 þ Canopy2500 þ Evergreen125 þ Evergreen2125 þ RdDensity125 þ Roughness125 þ TPI500 þ TPI2500 Highway500 þ Highway2500 þ Elevation500 þ Elevation2500 þ ForestEdge125 þ Canopy500 þ Canopy2500 Evergreen125 þ Evergreen2125 þ Precip þ RdDensity125 þ Roughness125 þ TPI500 þ TPI2500 Highway500 þ Highway2500 þ Elevation500 þ Elevation2500 þ ForestEdge125 þ Canopy500 þ Canopy2500 Evergreen125 þ Evergreen2125 þ RdDensity125 þ Roughness125 þ TPI500 þ TPI2500 Highway500 þ Elevation500 þ Elevation2500 þ ForestEdge500 þ Canopy 500 þ Canopy 2500 þ Evergreen125 þ Evergreen2125 þ Precip þ RdDensity125 þ Roughness125 þ TPI500 þ TPI2500 Highway500 þ Elevation500 þ Elevation2500 þ ForestEdge500 þ Canopy500 þ Canopy2500 þ Evergreen125 þ Evergreen2125 þ RdDensity125 þ Roughness125 þ TPI500 þ TPI2500 Highway500 þ Elevation00 þ Elevation2500 þ Canopy500 þ Canopy2500 þ Evergreen125 þ Evergreen2125 þ Precip þ RdDensity125 þ Roughness125 þ TPI500 þ TPI2500 Highway500 þ Elevation500 þ Elevation2500 þ ForestEdge125 þ Canopy500 þ Canopy2500 þ Evergreen125 þ Evergreen2125 þ Precip þ RdDensity125 þ Roughness125 þ TPI500 þ TPI2500 Appendix E. Pairwise similarities between the continuous predicted maps generated by the top-performing resource selection function models for each recreation type studied in western Colorado, USA 2010e2013, as measured by Pearson correlation. Pairs of recreation types with higher Pearson correlations are predicted to have greater similarity of terrain selection, and thus potentially greater interpersonal conﬂict. Table 6 Pearson correlations between predicted surfaces for each of the recreation activities. Recreation activities shown are on-trail snowmobiles (Snmb on-tr), off-trail snowmobiles (Snmb off-tr), snowmobile segments of hybrid-assisted skiing (Hybrid snmb), ski segments of hybrid-assisted skiing (Hybrid ski), and back-country ski or snowboard (BC Ski). Snmb on-rd Snmb off-rd Hybrid snmb Hybrid ski Ski/board Snmb on-tr Snmb off-tr Hybrid snmb Hybrid ski BC Ski 1.00 0.07 1.00 0.04 0.03 1.00 0.07 0.05 0.18 1.00 0.14 0.12 0.20 0.25 1.00 References Akaike, H. (1974). A new look at the statistical model identiﬁcation. IEEE Transactions on Automatic Control, 19(6), 716e723. Albritton, R., & Stein, T. V. (2011). Integrating social and natural resource information to improve planning for motorized recreation. Applied Geography, 31(1), 85e97. http://dx.doi.org/10.1016/j.apgeog.2010.02.005. Arlettaz, R., Nussle, S., Baltic, M., Vogel, P., Palme, R., Jenni-Eiermann, S., et al. (2015). Disturbance of wildlife by outdoor winter recreation: Allostatic stress response and altered activityeenergy budgets. Ecological Applications, 25(5), 1197e1212. Barton, K. (2015). MuMIn: Multi-model inference. R Package Version 1.15.1. Bates, D., Maechler, M., Bolker, B., & Walker, S. (2014). lme4: Linear mixed-effects models using Eigen and S4. R Package Version 1.1-13. Beeco, J. A., & Brown, G. (2013). Integrating space, spatial tools, and spatial analysis into the human dimensions of parks and outdoor recreation. Applied Geography, 38(1), 76e85. http://dx.doi.org/10.1016/j.apgeog.2012.11.013. Beeco, J. A., & Hallo, J. C. (2014). GPS tracking of visitor use: Factors inﬂuencing visitor spatial behavior on a complex trail system. Journal of Park and Recreation Administration, 32(2), 43e61. Beeco, J. A., Hallo, J. C., & Brownlee, M. T. J. (2014). GPS visitor tracking and recreation suitability Mapping: Tools for understanding and managing visitor use. Landscape and Urban Planning, 127, 136e145. http://dx.doi.org/10.1016/ j.landurbplan.2014.04.002. Bowker, J. M., Askew, A. E., Cordell, H. K., Betz, C. J., Zarnoch, S. J., & Seymour, L. (2012). US outdoor recreation participation projections to 2060. In Outdoor recreation participation in the United States - projections to 2060: A technical document supporting the forest Service 2010 RPA assessment (pp. 105e124). Asheville, NC: U.S. Department of Agriculture, Forest Service, Southern Research Station. Boyce, M. S., Vernier, P. R., Nielsen, S. E., & Schmiegelow, F. K. A. (2002). Evaluating resource selection functions. Ecological Modelling, 157, 281e300. Brabyn, L., & Sutton, S. (2013). A population based assessment of the geographical accessibility of outdoor recreation opportunities in New Zealand. Applied Geography, 41, 124e131. http://dx.doi.org/10.1016/j.apgeog.2013.03.013. Braunisch, V., Patthey, P., & Arlettaz, R. (2011). Spatially explicit modelling of conﬂict zones between wildlife and outdoor snow-sports: Prioritizing areas for winter refuges. Ecological Applications, 21(3), 955e967. http://dx.doi.org/10.1890/092167.1. Brown, G., & Raymond, C. M. (2014). Methods for identifying land use conﬂict potential using participatory mapping. Landscape and Urban Planning, 122, 196e208. http://dx.doi.org/10.1016/j.landurbplan.2013.11.007. Cessford, G. (2003). Perception and reality of conﬂict: Walkers and mountain bikes on the Queen Charlotte track in New Zealand. Journal for Nature Conservation, 11(4), 310e316. http://dx.doi.org/10.1078/1617-1381-00062. Cole, D. N., & Daniel, T. C. (2003). The science of visitor management in parks and protected areas: From verbal reports to simulation models. Journal for Nature Conservation, 11(4), 269e277. D'Antonio, A., Monz, C., Lawson, S., Newman, P., Pettebone, D., & Courtemanch, A. (2010). GPS-based measurements of backcountry visitors in parks and protected areas: Examples of methods and applications from three case studies. Journal of Park and Recreation Administration, 28(3), 42e60. Fortin, D., Beyer, H. L., Boyce, M. S., Smith, D. W., Duchesne, T., & Mao, J. S. (2005). Wolves inﬂuence elk movements: Behavior shapes a trophic cascade in Yellowstone National Park. Ecology, 86(5), 1320e1330. http://dx.doi.org/10.1890/ 04-0953. Freeman, E. A., & Moisen, G. G. (2008). A comparison of the performance of threshold criteria for binary classiﬁcation in terms of predicted prevalence and kappa. Ecological Modelling, 217, 48e58. http://dx.doi.org/10.1016/ j.ecolmodel.2008.05.015. Gillies, C. S., Hebblewhite, M., Nielsen, S. E., Krawchuk, M. A., Aldridge, C. L., Frair, J. L., et al. (2006). Application of random effects to the study of resource selection by animals. Journal of Animal Ecology, 75(4), 887e898. http:// dx.doi.org/10.1111/j.1365-2656.2006.01106.x. Gramann, J. H. (1982). Toward a behavioral theory of crowding in outdoor recreation: An evaluation and synthesis of research. Leisure Sciences, 5(2), 109e126. Hallo, J. C., Beeco, J. A., Goetcheus, C., McGee, J., McGehee, N. G., & Norman, W. C. (2012). GPS as a method for assessing spatial and temporal use distributions of nature-based tourists. Journal of Travel Research, 51(5), 591e606. http:// dx.doi.org/10.1177/0047287511431325. Hallo, J. C., Manning, R. E., Valliere, W., & Budruck, M. (2004). A case study �L.E. Olson et al. / Applied Geography 86 (2017) 66e91 comparison of visitor self-reported and GPS recorded travel routes. In Proceedings of the 2004 Northeastern recreation research symposium (pp. 172e177). Harris, G., Nielson, R. M., Rinaldi, T., & Lohuis, T. (2014). Effects of winter recreation on northern ungulates with focus on moose (Alces alces) and snowmobiles. European Journal of Wildlife Research, 60(1), 45e58. http://dx.doi.org/10.1007/ s10344-013-0749-0. Havlick, D. (2002). No place distant: Roads and motorized recreation on America's public lands. Washington, D.C.: Island Press. Homer, C. G., Dewitz, J. A., Yang, L., Jin, S., Danielson, P., Xian, G., et al. (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(5), 345e354. Hosmer, D. W., Jr., Lemeshow, S., & Sturdivant, R. X. (2013). Applied logistic regression (3rd ed.). Hoboken, New Jersey: John Wiley & Sons. Jackson, S. A., Haider, W., & Elliot, T. (2003). Resolving inter-group conﬂict in winter recreation: Chilkoot trail national historic site, British Columbia. Journal for Nature Conservation, 11(4), 317e323. http://dx.doi.org/10.1078/1617-138100063. Jacob, G. R., & Schreyer, R. (1980). Conﬂict in outdoor recreation: A theoretical perspective. Journal of Leisure Research, 12(4), 368e380. Jenness, J. (2013). DEM Surface Tools for ArcGIS (surface_area.exe). Retrieved from: http://www.jennessent.com/arcgis/surface_area.htm. Jenness, J., Brost, B., & Beier, P. (2013). Land facet corridor Designer: Extension for ArcGIS. Retrieved from: http://www.jennessent.com/arcgis/land_facets.htm. Keating, K. A., & Cherry, S. (2004). Use and interpretation of logistic regression in habitat-selection studies. Journal of Wildlife Management, 64(4), 744e789. Kliskey, A. D. (2000). Recreation terrain suitability mapping: A spatially explicit methodology for determining recreation potential for resource use assessment. Landscape and Urban Planning, 52(1), 33e43. http://dx.doi.org/10.1016/S01692046(00)00111-0. Krebs, J., Lofroth, E., & Parﬁtt, I. (2007). Multiscale habitat use by wolverines in British Columbia, Canada. Journal of Wildlife Management, 71(7), 2180e2192. http://dx.doi.org/10.2193/2007-099. Lai, P. C., Li, C. L., Chan, K. W., & Kwong, K. H. (2007). An assessment of GPS and GIS in recreational tracking. Journal of Park and Recreation Administration, 25(1), 128e139. Larson, C. L., Reed, S. E., Merenlender, A. M., & Crooks, K. R. (2016). Effects of recreation on animals revealed as widespread through a global systematic review. PLoS One, 1e21. http://dx.doi.org/10.1371/journal.pone.0167259. Leung, Y., & Marion, J. L. (1999). Spatial strategies for managing visitor impacts in national parks. Journal of Park and Recreation Administration, 17(4), 20e38. Manly, B. F. J., McDonald, L. L., Thomas, D. L., McDonald, T. L., & Erickson, W. P. (2002). Resource selection by animals: Statistical design and analysis for ﬁeld studies (2nd ed.). New York: Kluwer Academic Publishers. Manning, R. E., & Valliere, W. A. (2001). Coping in outdoor recreation: Causes and consequences of crowding and conﬂict among community residents. Journal of Leisure Research, 33(4), 410e426. Miller, A. D. (2016). Recreation conﬂict and management options in the Vail Pass winter recreation area, Colorado, USA. Colorado State University. Miller, A. D., Vaske, J. J., Squires, J. R., Olson, L. E., & Roberts, E. K. (2016). Does zoning winter recreationists reduce recreation conﬂict? Environmental Management, 1e18. http://dx.doi.org/10.1007/s00267-016-0777-0. National Weather Service, National Oceanic and Atmospheric Administration. (2017). Colorado snowpack. Retrieved from: https://www.weather.gov/bou/co_ snowpack. Patthey, P., Wirthner, S., Signorell, N., & Arlettaz, R. (2008). Impact of outdoor winter sports on the abundance of a key indicator species of alpine ecosystems. Journal of Applied Ecology, 45, 1704e1711. http://dx.doi.org/10.1111/j.1365- 91 2664.2008.01547.x. R Core Team. (2015). R: A language and environment for statistical computing. Retrieved July 1, 2015, from: https://www.r-project.org/. Reimers, E., Eftestøl, S., & Colman, J. E. (2003). Behavior responses of wild reindeer to direct provocation by a snowmobile or skier. Journal of Wildlife Management, 67(4), 747e754. ^te �, S. D. (2016). Space use analyses suggest avoidance of a ski area Richard, J. H., & Co by mountain goats. The Journal of Wildlife Management, 80(3), 387e395. http:// dx.doi.org/10.1002/jwmg.1028. €chli, D., Hediger, M., Lauber, S., Ochsner, P., et al. (2011). Rupf, R., Wyttenbach, M., Ko Assessing the spatio-temporal pattern of winter sports activities to minimize disturbance in capercaillie habitats. Eco.mont, 3(2), 23e32. http://dx.doi.org/ 10.1553/eco.mont-3-2s23. Sato, C. F., Wood, J. T., & Lindenmayer, D. B. (2013). The effects of winter recreation on alpine and subalpine fauna: A systematic review and meta-analysis. PLoS One, 8(5), e64282. http://dx.doi.org/10.1371/journal.pone.0064282. Seip, D. R., Johnson, C. J., & Watts, G. S. (2007). Displacement of mountain caribou from winter habitat by snowmobiles. Journal of Wildlife Management, 71(5), 1539e1544. http://dx.doi.org/10.2193/2006-387. Shively, D. D., Pape, B. M. C., Mower, R. N., Zhou, Y., Russo, R., & Sive, B. C. (2008). Blowing smoke in Yellowstone: Air quality impacts of oversnow motorized recreation in the park. Environmental Management, 41(2), 183e199. http:// dx.doi.org/10.1007/s00267-007-9036-8. Snyder, S. A., Whitmore, J. H., Schneider, I. E., & Becker, D. R. (2008). Ecological criteria, participant preferences and location models: A GIS approach toward ATV trail planning. Applied Geography, 28(4), 248e258. http://dx.doi.org/ 10.1016/j.apgeog.2008.07.001. Thapa, B., & Graefe, A. (2004). Recreation conﬂict and tolerance among skiers and snowboarders. Journal of Park and Recreation Administration, 22(1), 37e52. http://dx.doi.org/10.1080/01490409950202311. Theobald, D. M. (2004). Placing exurban land-use change in a human modiﬁcation framework. Frontiers in Ecology and the Environment, 2(3), 139e144. Therneau, T. (2015). A package for survival analysis in S. Retrieved from: http:// CRAN.R-project.org/package¼survival. Thurfjell, H., Ciuti, S., & Boyce, M. S. (2014). Applications of step-selection functions in ecology and conservation. Movement Ecology, 2(4), 12. http://dx.doi.org/ 10.1186/2051-3933-2-4. Tomczyk, A. M. (2011). A GIS assessment and modelling of environmental sensitivity of recreational trails: The case of Gorce National Park, Poland. Applied Geography, 31(1), 339e351. http://dx.doi.org/10.1016/j.apgeog.2010.07.006. U.S.D.A. Forest Service. (2015). White River national forest ofﬁcial webpage on Vail Pass winter recreation area. Retrieved July 30, 2015, from: http://www.fs.usda. gov/recarea/whiteriver/recreation/recarea/?recid¼41445&actid¼92. Vaske, J. J., Carothers, P., Donnelly, M. P., & Baird, B. (2000). Recreation conﬂict among skiers and snowboarders. Leisure Sciences, 22(4), 297e313. http:// dx.doi.org/10.1080/01490409950202311. Vaske, J. J., Donnelly, M. P., Wittmann, K., & Laidlaw, S. (1995). Interpersonal versus social-values conﬂict. Leisure Sciences, 17(3), 205e222. http://dx.doi.org/ 10.1080/01490409509513257. Vaske, J. J., Needham, M. D., & Cline, R. C., Jr. (2007). Clarifying interpersonal and social values conﬂict among recreationists. Journal of Leisure Research, 39(1), 182e195. Westcott, F., & Andrew, M. E. (2015). Spatial and environmental patterns of off-road vehicle recreation in a semi-arid woodland. Applied Geography, 62, 97e106. http://dx.doi.org/10.1016/j.apgeog.2015.04.011. Zielinski, W. J., Slauson, K. M., & Bowles, A. E. (2008). Effects of off highway vehicle use on the American Marten. Journal of Wildlife Management, 72(7), 1558e1571. http://dx.doi.org/10.2193/2007-397. � Dublin Core The Dublin Core metadata element set is common to all Omeka records, including items, files, and collections. For more information see, http://dublincore.org/documents/dces/. Title A name given to the resource Journal Articles Description An account of the resource CPW peer-reviewed journal publications Text A resource consisting primarily of words for reading. Examples include books, letters, dissertations, poems, newspapers, articles, archives of mailing lists. Note that facsimiles or images of texts are still of the genre Text. Dublin Core The Dublin Core metadata element set is common to all Omeka records, including items, files, and collections. For more information see, http://dublincore.org/documents/dces/. Title A name given to the resource Modeling large-scale winter recreation terrain selection with implications for recreation management and wildlife Description An account of the resource <span>Winter recreation is a rapidly growing activity, and advances in technology make it possible for increasing numbers of people to access remote backcountry terrain. Increased winter recreation may lead to more frequent conflict between recreationists, as well as greater potential disturbance to wildlife. To better understand the environmental characteristics favored by winter recreationists, and thus predict areas of potential conflict or disturbance, we modeled terrain selection of motorized and non-motorized recreationists, including snowmobile, backcountry ski, and snowmobile-assisted hybrid ski. We used sports recorder </span><a href="https://www.sciencedirect.com/topics/earth-and-planetary-sciences/global-positioning-system" title="Learn more about Global Positioning System from ScienceDirect's AI-generated Topic Pages" class="topic-link" target="_blank" rel="noreferrer noopener">Global Positioning System</a><span> (GPS) devices carried by recreationists at two study areas in Colorado, USA, (Vail Pass and the San Juan Mountains), to record detailed tracks of each recreation type. For each recreation activity, we modeled selection of remotely-sensed environmental characteristics, including topography, vegetation, climate, and road access. We then created spatial maps depicting areas that recreation activities were predicted to select and combined these maps to show areas of potential ecological disturbance or interpersonal conflict between motorized and non-motorized activities. Model results indicate that motorized and non-motorized activities select different environmental characteristics, while still exhibiting some similarities, such as selection for ease of access, reflected in proximity to highways and densities of open forest roads. Areas predicted to have only motorized recreation were more likely to occur further from highways, with greater forest road densities, lower canopy cover, and smoother, less steep terrain, while areas with only non-motorized recreation were closer to highways, with lower forest road densities, more canopy cover and steeper terrain. Our work provides spatially detailed insights into terrain characteristics favored by recreationists, allowing managers to maintain winter recreation opportunities while reducing interpersonal conflict or ecological impacts to sensitive wildlife.</span> Bibliographic Citation A bibliographic reference for the resource. Recommended practice is to include sufficient bibliographic detail to identify the resource as unambiguously as possible. Olson, L. E., J. R. Squires, E. K. Roberts, A. D. Miller, J. S. Ivan, and M. Hebblewhite. 2017. Modeling large-scale winter recreation terrain selection with implications for recreation management and wildlife. Applied Geography 86:66-91. <a href="https://doi.org/10.1016/j.apgeog.2017.06.023" target="_blank" rel="noreferrer noopener">https://doi.org/10.1016/j.apgeog.2017.06.023</a> Creator An entity primarily responsible for making the resource Olson, Lucretia E. Squires, John R. Roberts, Elizabeth K. Miller, Aubrey D. Ivan, Jacob S. Hebblewhite, Mark Subject The topic of the resource GPS tracking Habitat models Interpersonal conflict Motorized recreation Non-motorized recreation Recreation planning Extent The size or duration of the resource. 26 pages Date Created Date of creation of the resource. 2017-09 Rights Information about rights held in and over the resource <a href="http://rightsstatements.org/vocab/InC-NC/1.0/" target="_blank" rel="noreferrer noopener">In Copyright - Non-Commercial Use Permitted</a> <a href="https://creativecommons.org/licenses/by-nc-nd/4.0/" target="_blank" rel="noreferrer noopener">Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0)</a> Language A language of the resource English Is Part Of A related resource in which the described resource is physically or logically included. Applied Geography Format The file format, physical medium, or dimensions of the resource application/pdf Type The nature or genre of the resource Article