https://cpw.cvlcollections.org/files/original/cd27c895b8b9a8fc025c19dc7d8d2470.pdf 3891ce59cd9402fab6a3e786df6edee9 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 �Landscape Ecol DOI 10.1007/s10980-017-0590-z RESEARCH ARTICLE Predation risk across a dynamic landscape: effects of anthropogenic land use, natural landscape features, and prey distribution Patrick E. Lendrum . Joseph M. Northrup . Charles R. Anderson . Glen E. Liston . Cameron L. Aldridge . Kevin R. Crooks . George Wittemyer Received: 22 April 2017 / Accepted: 25 October 2017 Ó Springer Science+Business Media B.V. 2017 Abstract Purpose Human-mediated landscape changes alter habitat configuration, which strongly structures animal distributions and interspecific interactions. The effects of anthropogenic disturbance on predator–prey relationships are fundamental to ecology, yet less well understood. We determined where predation events occurred for fawn and adult female mule deer from 2008 to 2014 in critical winter range with extensive energy development. We investigated the relationship between predation sites, energy infrastructure, and natural landscape features across contiguous areas Electronic supplementary material The online version of this article (http://doi.org/10.1007/s10980-017-0590-z) contains supplementary material, which is available to authorized users. P. E. Lendrum (&) � J. M. Northrup � K. R. Crooks � G. Wittemyer Department of Fish, Wildlife, and Conservation Biology, Colorado State University, Campus e-Delivery 1474, Fort Collins, CO 80523, USA e-mail: Patrick.lendrum@colostate.edu J. M. Northrup Department of Forest Ecosystems and Society, Oregon State University, 321 Richardson Hall, Corvallis, OR 97331, USA experiencing different degrees of energy extraction during periods of high and low intensity development. Methods We contrast spatial correlates of 286 mortality locations with random landscape locations and mule deer distribution estimated from 350,000 GPS locations. We estimated predation risk with resource selection functions and latent selection difference functions. Results Relative to the distribution of mule deer, predation risk was lower closer to pipelines and well pads, but higher closer to roads. Predation sites occurred more than expected relative to availability and deer distribution in deeper snow and non-forested habitats. Anthropogenic features had a greater influence on predation sites during the period of low activity than high activity, and natural landscape characteristics had weaker effects relative to G. E. Liston Cooperative Institute for Research in The Atmosphere, Colorado State University, 1375 Campus Delivery, Fort Collins, CO 80523, USA C. L. Aldridge Department of Ecosystem Science and Sustainability and NREL, Colorado State University, 1499 Campus Delivery, Fort Collins, CO 80523, USA C. R. Anderson Mammal Research Section, Colorado Parks and Wildlife, 317 W. Prospect, Fort Collins, CO 80526, USA 123 �Landscape Ecol anthropogenic features throughout the study. Though canids accounted for the majority of predation events, felids exhibited stronger landscape associations, driving the observed spatial patterns in predation risk to mule deer. Conclusions The emergence of varied interactions between predation and landscape features across contexts and years highlights the complexity of interspecific interactions in highly modified landscapes. Keywords Carnivores � Disturbance ecology � Energy development � Habitat fragmentation � Mule deer � Predation risk � Resource selection Introduction Natural and anthropogenic disturbances affect community assembly through alterations in habitat conditions (Larsen and Ormerod 2014). Humans are the primary drivers of contemporary habitat loss and degradation, with strong effects on community structuring and interactions (Wilcove et al. 1998; Chapin et al. 2000). The effects of habitat loss can be direct, via the destruction or alteration of habitat, or indirect, for example via behaviorally driven avoidance of human activities (Frid and Dill 2002) resulting in functional habitat loss (Aldridge and Boyce 2007). The potential for indirect impacts is emerging as a major concern given increasing anthropogenic development globally and potential effects on wildlife populations (Leu et al. 2008). Habitat loss affects 40 percent of the world’s mammals (Schipper et al. 2008). Of these species, carnivores are thought to be particularly vulnerable to habitat alteration because of their relatively large ranges, low numbers, and direct persecution by humans (Crooks 2002). In addition to direct displacement by disturbance, prey abundance and distribution are also key drivers of demography and distribution of large carnivores (Carbone and Gittleman 2002; Karanth et al. 2004). Ungulates are the primary prey base for many large carnivores throughout the world, including the western United States (Hurley et al. 2011; Elbroch et al. 2013). Prey species must make trade-offs between resource acquisition and risk of mortality, whether the risk is real or perceived (Frid 123 and Dill 2002), and may do so by altering their spatiotemporal patterns of habitat use (Laberee et al. 2014). A growing number of studies have observed avoidance of human-caused disturbance by ungulates at large spatial scales (Northrup and Wittemyer 2013), while others have detected selection for areas of disturbance at finer scales (Berger 2007; Rogala et al. 2011). Variable responses of prey species to human disturbance could drive complex interactions between carnivores and disturbance. Because the effects of habitat change on carnivores may be mediated through the response of their prey (Burton et al. 2012), a better understanding of how predator–prey interactions are structured in human-altered landscapes is warranted. Traditionally, researchers have examined the location of predation events relative to the availability of landscape features (Husseman et al. 2003; Elbroch et al. 2013) or how prey distributions shape predation risk (Hebblewhite et al. 2005; Courbin et al. 2013). However, less is known about the combined influence of habitat characteristics and prey distribution on predation risk in multi-predator communities, despite the importance of such interactions in structuring ecological communities (Ford et al. 2014; Moll et al. 2017). Across much of the western United States, sagebrush ecotones provide critical winter range habitat for mule deer, a principal big game species that has decreased across much of its range (Unsworth et al. 1999), in part from habitat loss and degradation resulting in reduced mule deer recruitment (Johnson et al. 2017). These landscapes have been extensively developed for energy extraction (McDonald et al. 2009; Northrup et al. 2015), which influences mule deer distribution and habitat selection (Sawyer et al. 2006; Lendrum et al. 2012). The Piceance Basin in northwest Colorado, USA, contains the second largest natural-gas reserve in the country (Hawkins et al. 2016). The ongoing development continues to fragment the landscape with well pads, pipelines, roads, and industrial facilities, which once supported one of the largest migratory mule deer herds across their range (Lendrum et al. 2014). We compiled a six-year data set of mortality events and mule deer space use across critical winter range habitat of the Piceance Basin to better understand how anthropogenic disturbance interacts with landscape characteristics and prey distribution to influence predation risk of mule deer. �Landscape Ecol Our primary objectives were to: (1) investigate mortality locations of radio-collared mule deer to determine cause-specific mortality and identify when and where predation occurred; and (2) evaluate the influence of anthropogenic disturbance on predation risk. For the latter objective, we examined the effects of landscape features on predation using resource selection functions (RSF; Boyce 2006) and the interaction between prey distribution and landscape features on predation using latent selection difference functions (LSDF; Erickson et al. 2014). We compared two contiguous areas with markedly different degrees of energy extraction that included a low intensity natural-gas development area (‘‘undeveloped’’) and a relatively high intensity development area (‘‘developed’’), across two time periods representing different levels of active development, 2009-2011 (‘‘high activity’’) and 2012–2014 (‘‘low activity’’). We hypothesized that: (1) during the high activity period, predation risk would be reduced in proximity to anthropogenic features with high levels of human activity (well pads and industrial facilities) because predators tend to be more adversely affected by human disturbance than prey (Berger 2007; Ripple et al. 2014), and that these effects would be reduced during the low activity period when human activity was reduced; and (2) predation risk would be increased in proximity to linear features (roads and pipelines) because linear features can facilitate the movement of predators (Leblond et al. 2013) and create edge habitat known to increase the risk of predation (Elbroch et al. 2013). Materials and methods Study area We monitored mule deer across varying levels of natural-gas development within the Piceance Basin (Fig. 1): a developed area comprised of two subsections (141 km2, 0.6 well pads/km2 and 83 km2, 0.8 well pads/km2) and an undeveloped subsection (79 km2, 0.1 well pads/km2; Lendrum et al. 2012, 2013). The climate of the region was typified by warm dry summers and cold winters, with most annual moisture in the form of winter snow and monsoonal spring rainstorms. The study area was topographically variable, with elevation ranging from 1675 to 2285 m. Pinyon pine (Pinus edulis) and Utah juniper (Juniperus osteosperma) were the dominant overstory species; common shrubs included big sagebrush (Artemisia tridentata), Utah serviceberry (Amelanchier utahensis), bitterbrush (Purshia tridentata), and rabbitbrush (Crysothamnus spp.; Lendrum et al. 2014). Species of large carnivores included coyotes (Canis latrans), cougars (Puma concolor), bobcats (Lynx rufus), and black bears (Ursus americanus). In addition to mule deer, the area contained other potential prey items including North American elk (Cervus elaphus), cottontail rabbits (Sylvilagus spp), as well as smaller rodents and birds. Data collection From 2008 to 2014, mule deer were net-gunned from helicopters (Krausman et al. 1985). Three hundredninety adult female mule deer were captured and equipped with GPS collars (GPS-4400S, Lotek Wireless, Newmarket, Ontario, Canada, on 14 individuals the first year; G2110D, Advanced Telemetry Systems, Isanti, Minnesota, USA, thereafter). During the same time period, we also fit mule deer fawns (December captures annually; 6 months old of either sex) with VHF collars to monitor cause-specific mortality. Because carnivores are typically most active during crepuscular and nighttime hours in disturbed landscapes (Lendrum et al. 2017), and therefore prey are most likely to be killed during these times (Anderson and Lindzey 2003), we only retained GPS locations that occurred between 18:00 and 06:00 in our analysis (see below). We stratified the dataset into 63,804 locations during the high activity period (2012–2014) and 135,620 locations during the low activity period (2009–2011) in the undeveloped area, and 78,066 locations during the high activity period and 127,255 locations during the low activity period in the developed area. GPS collars were store-on-board and programmed to attempt a fix once every 5 h, with a subset of collars programed to obtained 1 h fixes (Northrup et al. 2015). Collars were either programmed to drop off during April of the year following deployment (i.e., 16 months post capture) to allow collar retrieval and download of stored GPS locations, or January of the next year (i.e., 13 months post capture). Fawn collars were spliced and fitted with rubber surgical tubing to allow for growth and for collars to drop off between mid-summer and early autumn. 123 �Landscape Ecol Fig. 1 Locations of mule deer predation sites and natural-gas development infrastructure in the Piceance Basin, CO, USA during the period of high development, 2008–2011, across undeveloped and developed study sites All collars were equipped with mortality sensors that transmitted a signal after 8 h of inactivity. A 123 trained technician scanned for mortality beacons daily and investigated the site to determine cause of death �Landscape Ecol after each mortality detection. Located carcasses were examined for hemorrhaging or peticiations to verify that a predation event took place rather than scavenging after a non-predation mortality (Stonehouse et al. 2016). To determine the predator involved, the width of the canine punctures, tracks present, and style in which carcass remains were distributed (i.e., cached or scattered) were recorded. If the predator was indiscernible, the predation event was marked as unknown predator, and if there was uncertainty if a predation event took place, the mortality was excluded from this analysis (Table 1). We compared predator-specific predation events with a non-parametric cumulative incidence function estimator for cause-specific rates of mortality by age class of mule deer (Heisey and Patterson 2006). Descriptive statistics comparing predation sites, GPS, and random locations can be viewed in Appendix 1. At each predation, GPS, and random location, we sampled four natural landscape variables known to influence predation risk (Hebblewhite 2005; Elbroch et al. 2013) and six metrics of anthropogenic disturbance previously identified to be influential in mule deer habitat selection (Sawyer et al. 2006; Lendrum et al. 2013; Northrup et al. 2015). Natural landscape variables included terrain ruggedness, distance to ecotone edge (forest and shrublands), concealment cover, and snow depth. Anthropogenic landscape features included distance to nearest well pad (producing and drilling), road (primary and secondary), pipeline, and industrial facility (Fig. 1). Industrial facilities include compressor stations and operation centers of frequent human activity. Drilling well pads and industrial facilities occurred with such low frequency in the undeveloped site that these metrics of anthropogenic disturbance were retained in the model to control for variation, rather than make biological inference. A detailed description of how variables were calculated is available in Appendix 2. Table 1 Annual numbers of radio-tracked mule deer, cumulative predation rate, number of mule deer preyed upon by each predator (including unknown), cumulative precipitation (cm), average temperature, and the number of well pads in the development phase, during the winter in the Piceance Basin, CO, USA, 2008–2014 Collared deera Cumulative predation rate 08–09 187 0.13 8 09–10 250 0.11 9 Year Data analysis The landscape context of predation sites was examined using a generalized resource selection function approach (RSF; Manly et al. 2002; Boyce 2006). Natural (topographic, environmental, and habitat characteristics) and anthropogenic (proximity to energy development) features associated with predation sites were compared to an available sample drawn from within a winter range area characterized by merging 1.2-km radius buffers around every predation location (approximately equivalent to the 4.5 km2 average home range of deer in the system as described in Northrup et al. 2016). The relationship between predation sites and mule deer distribution (assessed from GPS telemetry data) was examined using a latent selection difference function (LSDF; Erickson et al. 2014), an analytical framework similar to resource selection functions used to provide quantitative estimates of differences in selection behavior between two datasets comprised of animal locations, in this case predation sites and GPS telemetered locations. All GPS locations within the buffer from the corresponding time frame were included in the LSDF analyses, and the sample of random locations employed in the RSF was set to the same number of Cause of predation Coyote Cougar Cumulative precipitation (cm) Temp (°C) Drilling well pads Bobcat Bear Unknown 2 5 0 10 15.53 0.00 30 2 5 0 11 14.25 - 0.24 24 10–11 267 0.33 24 14 8 1 41 21.52 - 0.64 11 11–12 295 0.25 21 9 3 2 39 13.04 1.80 3 12–13 316 0.14 15 6 2 0 20 12.92 - 1.56 3 13–14 307 0.09 3 4 1 1 20 19.96 1.00 1 a Includes recaptures from the previous year so totals do not equate to the total number of individuals collared 123 �Landscape Ecol GPS locations used in the LSDF to ensure comparability (Northrup and Wittemyer 2013). We first fit a macro-scale model that included only known canid (coyote) and felid (bobcat and cougar) predation locations across all study sites and years to determine how different hunting strategies (cursorial and ambush) might influence predation risk. We then pooled all kill sites regardless of predator species (providing insight to general predation risk) and fit separate models for spatial areas representing the undeveloped and developed sites, as well as across the two time periods of high and low energy development activity, resulting in 4 models (undeveloped low activity, undeveloped high activity, developed low activity and developed high activity). We conducted separate RSF and LSDF models using a Bayesian logistic regression (Gelman and Hill 2006) framework run in the R statistical software (R Development Core Team 2015) for each period of activity and level of development. The first 50,000 iterations of the Markov chain Monte Carlo (MCMC) algorithm written in R were discarded, and 500,000 samples were saved to build posterior distributions. To facilitate convergence and to allow for comparison of the magnitude of the effects of regression coefficients, we standardized continuous predictor variables by subtracting their means and dividing by their standard deviations, but did not transform binary predictors (Gelman and Hill 2006). Standardization was also conducted using the mean and standard deviation calculated for all datasets combined to allow direct comparison of coefficients across time periods and study areas. Prior to fitting models, we tested for pairwise correlations among covariates to ensure that no covariates were highly correlated (|r| [ 0.7) and calculated condition numbers to test for multicollinearity as suggested by Lazaridis (2007) to ensure that none were over 5.4. We ran each MCMC algorithm twice for each model, using starting values that were expected to be overdispersed relative to the posterior distributions. Model convergence was evaluated by visual inspection of trace plots and by Gelman–Rubin diagnostic (mean values \ 1.1 indicate convergence; Gelman and Rubin 1992). With both analytical methods, the influence of a given feature on the location of a predation site can be interpreted by the directionality and magnitude of the median estimate of the coefficient. These relationships are multifaceted and vary 123 depending on the variable being measured and therefore must be interpreted on a case-by-case basis (Appendix 3). If C 90% of the posterior distributions for the standardized b coefficients did not overlap zero, we concluded that there was strong evidence of an effect of the predictor variable. The magnitude of the coefficients allowed us to make inference about the relative influence of the given variable on the probability of a predation event occurring (Hobbs and Hooten 2015). Results We radio-collared and monitored 1357 individual mule deer including 126 adult female mule deer in the undeveloped study area and 264 in the developed area, and 316 fawns in the undeveloped study area and 651 in the developed area. We documented 313 mortality events across designated winter range habitat (Table 1). Of these mortalities, we identified 286 as predation events, 22 as road kill, 1 hunter harvest, and 4 attributed to malnutrition or disease. One hundred fifty-three (53.50%) of the predation events occurred in the undeveloped study area and 133 (46.50%) occurred in the developed area. Two hundred twenty-four predation events were of C 6 month old fawns, constituting the majority (78.32%) of the depredated individuals, and 62 predation events were of adult females. Median overwinter predation rate of fawns and adult collared females was 0.18 (SE = 0.06) and 0.10 (SE = 0.01), respectively. Combined overwinter predation rates of collared mule deer ranged from an apparent low estimate of 0.09 during the winter of 2013–2014 to a high of 0.33 during the winter of 2010–2011; median overwinter predation rate across the six years was 0.13 (Table 1). Of the predation events where a predator was identified, coyotes preyed on significantly more mule deer than the other predators, primarily targeting fawns (CIF = 5.05, p = 0.02, all other p values [ 0.50). Cumulative predation rates across all years were 0.05 for coyote predation, 0.02 for cougar predation, 0.01 for bobcat predation, 0.002 for bear predation, and 0.09 for unknown predation (Table 1). The proportions of fawn predation by cause were 0.47 unknown, 0.31 coyote, 0.11 cougar, 0.10 bobcat, and 0.01 bear. Adult female mule deer predation by cause �Landscape Ecol were 0.58 unknown, 0.19 cougar, 0.18 coyote, 0.3 bobcat, and 0.2 bear. Predator-specific predation risk closer to drilling wells (b = - 0.12, b = - 0.26, respectively; Table 2), however the median predation distance to a drilling well pad was 4.9 km, suggesting this result may not be biologically relevant (SE = 1028.13 m; Appendix 1). Canid predation events were randomly distributed relative to environmental characteristics measured; terrain ruggedness, snow depth, habitat type, and distance to ecotone edge (Table 2). Conversely, RSF and LSDF models indicated felid predation of mule deer occurred preferentially in deeper snow relative to availability across the landscape (b = 0.24) and to prey distribution (b = 0.19; Table 2; Fig. 2). Felid predation events were more strongly influenced by anthropogenic features compared to canid predations; felid predation locations were further from pipelines relative to availability across the landscape (b = 0.35) and to prey distribution (b = 0.72), and were closer to primary (b = -0.29) and secondary (b = -0.19) roads than expected relative to mule deer distributions (Table 2; Fig. 2). RSF models indicated that canids and felids killed further from producing well pads than expected relative to availability across the landscape (b = 0.21, b = 0.27), while LSDF models indicated that only canids killed further from well pads relative to prey distributions (b = 0.23). RSF and LSDF models also indicated that canids preyed on mule deer During the period of high energy development (2008–2011), RSF models indicated that predation sites occurred further from producing well pads than expected based on availability in the developed area (b = 0.24; Table 3, Appendix 1). Predation sites occurred in proportion to availability for all other anthropogenic features regardless of level of development. RSF models also indicated that predation locations occurred more often than expected in areas of shrub cover in the undeveloped area (b = 0.62) and in deeper snow in the developed area (b = 0.12; Table 3; Fig. 3). Terrain ruggedness and distance to ecotone edge between trees and shrubs were not strong predictors of predation site locations. Relative to mule deer distributions, LSDF models indicated predation sites occurred further from pipelines in both the undeveloped (b = 0.74) and developed (b = 0.19) areas, and closer to primary roads only in the undeveloped area (b = - 0.27). Distance to Table 2 Median standardized parameter estimates of predation sites of radio-collared mule deer by canid (coyote) and felid (cougar and bobcat) predators from resource selection functions (RSF) and latent selection difference functions (LSDF) during the winter in the Piceance Basin, CO, USA, 2008–2014 Covariate Predation during high development activity Canid Felid RSF LSDF RSF Ruggedness - 0.208 - 0.185 Snow depth 0.068 0.091 Open habitat 0.355 0.418 - 0.355 - 0.309 Shrub habitat 0.351 0.310 - 0.022 0.078 0.809 0.235* LSDF 0.795 0.191* Distance to Habitat edge Drilling wells Producing wells Pipelines 0.044 - 0.117* 0.207* - 0.010 - 0.015 - 0.018 - 0.257* - 0.084 - 0.132 0.233* 0.265* 0.188 - 0.150 0.046 0.353* 0.722* Primary roads 0.123 - 0.080 - 0.031 - 0.292* Secondary roads 0.051 0.004 - 0.148 - 0.188* Facilities 0.060 - 0.020 0.450* 0.225 *Indicates C 90% of the posterior distributions for the standardized b coefficients did not overlap zero 123 �Landscape Ecol Fig. 2 Median standardized parameter estimates (solid lines) and 95% credible intervals (dashed lines) of predation sites of radio-collared mule by canids (black) and felids (grey) in the Piceance Basin, CO, USA: a RSF output indicates the relative risk of predation by felids increased in deeper snow than expected based on habitat availability. b LSDF output indicates the relative risk of predation increased by canids and felids further from well pads than expected based on mule deer distribution; c LSDF output indicates the relative risk of predation by felids increased further from pipelines than expected based on deer distribution; d LSDF output indicates the relative risk of predation by felids increased closer to primary roads than expected based on mule deer distribution pipelines had the greatest influence of all the measured variables during the high activity period (Table 3, Appendix 1). With respect to natural landscape features, locations of predation sites were similar to those assessed with the RSF. In the undeveloped and developed area, predation sites occurred more than expected based on mule deer distribution in areas of increased shrub cover relative to tree cover (b = 0.57 undeveloped, b = 0.30 developed) and in deeper snow in both areas (b = 0.19, b = 0.20, Table 3). Predation during low development activity 123 During the period of low energy development (2012–2014), RSF models for the undeveloped and developed areas indicated that predation locations occurred closer to drilling well pads (b = - 1.4 undeveloped, b = - 0.16 developed) and further from producing well pads (b = 0.84, b = 0.24) than expected relative to the availability of that feature (Table 3, Appendix 1). Predation sites occurred in proportion to the availability of pipelines and roads. With respect to natural landscape variables, predation sites occurred more than expected in trees than shrubs �Landscape Ecol Table 3 Median standardized parameter estimates of predation sites of radio-collared mule deer from resource selection functions (RSF) and latent selection difference functions Covariate (LSDF) in periods of high and low energy development in undeveloped and developed study areas during the winter in the Piceance Basin, CO, USA, 2008–2014 High activity Low activity Undeveloped RSF Developed LSDF RSF Undeveloped LSDF RSF Developed LSDF RSF LSDF Ruggedness 0.57 0.33 0.68 0.47 0.02 - 0.06 - 0.68 - 0.75 Snow depth 0.13 0.19* 0.12* 0.20* 0.07 0.03 0.09 0.10 0.48* 0.20 0.56* 0.40* Open habitat Shrub habitat - 0.14 0.62* 0.14 0.57* - 0.13 0.25 - 0.25 0.30* 0.54 - 0.49* 1.01* - 0.45 Distance to Habitat edge - 0.16 - 0.17 0.09 0.11 - 0.05 - 0.04 Drilling wells - 0.05 - 0.02 0.03 0.12 - 1.4* - 0.51* - 0.16* Producing wells 0.14 - 0.08 0.24* 0.23* 0.84* 0.55* 0.24* 0.24* Pipelines 0.21 0.01 0.19* 0.07 0.32* 0.10 0.40* - 0.31* Primary roads Secondary roads Facilities 0.01 0.74* 0.02 0.00 - 0.3* - 0.27* 0.03 - 0.06 - 0.08 0.03 - 0.33* - 0.13 - 0.13 0.07 0.07 0.12 0.10 - 0.12 - 0.22* 0.14 - 0.03 0.01 0.00 0.30 0.51* - 0.06 - 0.11* *Indicates C 90% of the posterior distributions for the standardized b coefficients did not overlap zero in the undeveloped area (b = - 0.49), whereas predation sites occurred in areas of less tree cover in the developed area (b = 0.48, Table 3). Neither terrain ruggedness, snow depth, nor distance to ecotone edge were strong predictors of predation sites in the low activity period. LSDF models for the undeveloped and developed areas indicated predation locations occurred closer to drilling well pads (b = - 0.51 undeveloped, b = - 0.30 developed), further from producing well pads (b = 0.55, b = 0.24), further from pipelines (b = 0.32, b = 0.40), and closer to primary roads (b = - 0.31, b = - 0.33) relative to the distribution of mule deer (Table 3, Fig. 3, Appendix 1). Additionally, in the developed area, predation sites also occurred closer to secondary roads (b = - 0.22) and industrial facilities (b = - 0.11, Table 3), but predation sites were further from industrial facilities that bordered the undeveloped area (b = 0.51, Table 3, Appendix 1). Predation sites were more likely to occur in non-forested habitats than expected based on mule deer distribution in both the undeveloped (b = 1.01) and developed (b = 0.56) areas (Table 3). As with the RSF results, terrain ruggedness, snow depth and distance to ecotone edge were not strong predictors of predation sites. Discussion Human-caused habitat change is the primary disturbance influencing carnivore populations globally (Crooks et al. 2011; Ripple et al. 2014), with subsequent effects on interactions with prey species (Estes et al. 2011). While the impacts of this anthropogenic stressor on wildlife behavior has been well documented, interspecific interactions are a critical, but underserved area for investigation (Northrup and Wittemyer 2013). We recorded coyote, cougar, bobcat, and black bear predation on C 6 month old mule deer in our study area, all of which are documented predators of mule deer in the intermountain West (Unsworth et al. 1999), though black bears were hibernating during most of the study period examined. Coyotes were the primary documented predator of wintering deer within our sample, accounting for over half of the identified predation events, while cougars constituted the majority of the remaining identifiable predations at slightly under one-third of the events. Coyotes were the predominant identified predator of fawns, and cougars were the primary identified predator of adult females, consistent with prior studies (Unsworth et al. 1999; Bishop et al. 2005; Hurley et al. 2011). Predation is most often the major proximate 123 �Landscape Ecol Fig. 3 Posterior distribution of predicted mule deer predation site selection in the Piceance Basin, CO, USA: a RSF output indicates the relative risk of predation increased in deeper snow than expected based on habitat availability in the developed area during the period of high activity; b LSDF output indicates the relative risk of predation increased further from well pads than expected based on mule deer distribution; c LSDF output indicates the relative risk of predation increased further from pipelines than expected based on deer distribution; d LSDF output indicates the relative risk of predation increased closer to primary roads than expected based on mule deer distribution cause of mule deer mortality and generally considered compensatory (Bartmann et al. 1992; Bishop et al. 2005). In our study, we did not assess the impact of predation on mule deer population trends, though it is notable that the population increased during the study and that survival of adult females and fawns [ 6 months was similar between study areas (Northrup et al. in revision). Predicting and describing patterns of predation is difficult, and is further complicated in ecosystems with varying levels of anthropogenic disturbance over wide spatiotemporal extents. Our approach allowed us to examine how anthropogenic disturbance and natural landscape features structure spatial patterns of ‘true’ predation risk (i.e., the probability of mortality by predation; Moll et al. 2017), while accounting for prey distribution. Predator hunting strategies also influence predation risk, with felid predators that commonly exhibit a sit and pursue ambush strategy evoking stronger risk effects than active, or cursorial, predators such as canids (Preisser et al. 2007). Accordingly, we also observed stronger relationships between felid kill sites and landscape structure than among canid kill sites, ultimately influencing many of the detected spatial patterns in predation risk to mule deer. By employing resource selection functions (RSF) and latent selection difference functions (LSDF), we were able to interpret the different roles of the spatial 123 �Landscape Ecol arrangements of landscape features and the distribution of mule deer on predation risk. Had we considered only habitat characteristics and not accounted for prey distribution, we would not have been able to evaluate adequately the factors contributing to spatial predation risk (Moll et al. 2017). Anthropogenic effects on predation Linear features (i.e., pipelines and roads) had the most consistent effect on predation site locations in this study, though in a complex manner. LSDF analyses demonstrated that, relative to deer distribution, predation sites occurred further from pipelines in all study sites and periods and closer to primary roads except for when human disturbance was highest. Conversely, RSFs indicated predation took place in proportion to the availability of these features across the landscape. These patterns appeared predominantly to reflect felid predation, which was closer to roads and further from pipelines compared to kills by coyotes. Linear corridors can alter movement patterns and species interactions depending on whether the feature is perceived as a travel corridor or barrier (Brittingham et al. 2014). When humans are not immediately present, carnivores may increase their use of human infrastructure as travel routes (Kertson et al. 2011; Knopff et al. 2014), which can increase predation risk for prey near these features (Whittington et al. 2011). Mule deer in the Piceance Basin have been observed to select for habitats closer to roads at night (Northrup et al. 2015). The combined effects of selection by prey for areas closer to roads, restricted movement or escape abilities which has been observed in other systems (Dyer et al. 2002), and the use of roads with low human activity by carnivores, could account for the observed increased predation risk near roads. While an interaction between snow and pipelines reduced the benefit of pipelines as travel corridors to coursing predators in the boreal forests of Canada (Dickie et al. 2017), pipelines inhibited predation in the Piceance irrespective of snow depth. Other aspects of pipelines may be reducing predation risk in our system, for example increased visibility of predators by mule deer. Results from the RSF and LSDF analyses consistently indicated that predation sites occurred further from producing wells in the developed site regardless of activity level. Producing well pads were the only anthropogenic landscape feature avoided in predation site selection relative to both habitat availability and mule deer distribution. These combined results suggest producing well pads may serve as a predatory shield (Berger 2007) and might explain the documented selection for areas close to producing well pads by deer in this area (Northrup et al. 2015). The closest distance a predation site occurred to the edge of a producing well pad was * 140 m during the period of high activity and * 70 m during the period of low activity, while mule deer locations occurred directly on the well pads in both periods. Well pads are relatively large swaths of land (average well pad = 3.4 ha) that have been cleared of vegetation, reducing concealment cover for predators to ambush attack prey, and, in turn, providing prey with increased detection abilities. Predation events were closer to industrial facilities and drilling pads than expected based on mule deer distributions during the low activity period. However, the median distance of kill sites and mule deer locations were greater than 4 km from drilling pads and 1.5 km from industrial facilities, distances that are unlikely to have biological meaning given the topographic diversity of the study area. Thus, we do not think much can be gleaned from these results. Effects of natural landscape features on predation We observed a strong response to snow depth in the high activity period and vegetation cover across both periods, indicating these natural landscape features enhanced predation risk. It is well established that snow depth influences the locomotion of ungulates (Telfer and Kelsall 1984; Parker et al. 1984) and carnivores (Murray and Boutin 1991; Crête and Larivière 2003). The effect of snow, in turn, influences predation risk (Telfer and Kelsall 1984; Husseman et al. 2003), with the advantage often given to predator, which have lower foot-loads (ratio of body mass to foot area) than their ungulate prey. This appeared to be particularly the case for felid predators. Accordingly, mule deer were more likely to be preyed upon in deeper snow relative to their use in both study sites, and in deeper snow relative to availability in the developed site. Cumulative snowfall was lower during the period of low relative to high activity when averaged across years, though the winter of 2010-11 had the greatest yearly snowfall. We speculate that, on 123 �Landscape Ecol average, snow was below a depth that influences deer locomotion (Parker et al. 1984) which may, in part, explain why snow depth was less influential during the low activity period. Cougars will often kill prey in steeper terrain with greater cover while coyotes prefer more open habitat (Bishop et al. 2005), though not always (Elbroch et al. 2013). Contrary to expectations, we did not detect a strong influence of terrain ruggedness on predation site locations. We also did not observe a response in predation site selection to ecotone edge between forested and shrub habitats, which has been observed to facilitate effective predation in similar systems (Laundré and Loxterman 2007; Elbroch et al. 2013). The highly fragmented landscape of the Piceance Basin may provide so much edge habitat that any selection may not be observable. In the Piceance Basin, mule deer select shrub habitats over trees, particularly during the night (Lendrum et al. 2012; Northrup et al. 2015), and predation was found to occur in shrub lands more than treed areas. In combination with greater deer availability, dense shrub stands may provide greater concealment cover for predators than the relatively open understory of the pinyon-juniper woodland. The removal and disturbance of sagebrush habitat for expansion of energy infrastructure (Brittingham et al. 2014) has the potential to limit hunting habitat for carnivores. Conclusions Animals alter their behavior in association with natural and anthropogenic habitat alterations through the spatial selection or avoidance of an area (Laberee et al. 2014), which influences interspecific interactions. We observed that landscape features associated with energy development altered predation risk in this system. Specifically, we note that some features (pipelines and well pads) appeared to inhibit predation, while others (namely roads) were affiliated with predation, making a simplified assessment of the impact of development on predator–prey dynamics difficult. In areas where predation of mule deer is a concern, a reduction in road development may merit consideration by managers. In predator–prey interactions, the benefit to one guild is often a detriment to the other and therefore must be considered when making informed management decisions, especially during a time of 123 unprecedented human-induced landscape alteration. Disentangling the intricacies of interspecific interactions in landscapes altered by human activities is challenging and it is even more difficult to relate these metrics to demographic effects. When data are available, combining resource selection functions and latent selection difference functions resolves some of the spatial complexity and can serve as a template to further our understanding of predation risk in anthropogenically altered landscapes. Acknowledgements Funding and support came from Federal Aid in Wildlife Restoration, Colorado Mule Deer Association, Colorado Mule Deer Foundation, Colorado Oil and Gas Conservation Commission, Colorado State University, Williams Production LMT CO., Chevron Corporation, EnCana Corp., ExxonMobil Production Co., Shell Petroleum, Marathon Oil Corp. and Colorado Parks and Wildlife (CPW). We thank C. Bishop, D. Freddy, M. Michaels and the personnel at Little Hills State Wildlife Area for support. We thank L. Gepfert and L. Coulter for fixed-wing aircraft support, and L. Wolfe, M. Fisher, C. Bishop, and D. Finley of CPW for assistance during capture efforts. We thank B. Walker and E. Bergman for reviewing previous versions of this manuscript. Finally, we thank the White River Bureau of Land Management and the U.S. Forest Service, along with numerous private landowners for their cooperation. References Aldridge CL, Boyce MS (2007) Linking occurrence and fitness to persistence: a habitat-based approach for Greater SageGrouse. Ecol Appl 17:508–526 Anderson CR Jr, Lindzey FG (2003) Estimating cougar predation rates from GPS location clusters. J Wildl Manag 67:307–316 Bartmann RM, White GC, Carpenter LH (1992) Compensatory mortality in a Colorado mule deer population. Wildl Monogr 121:3–39 Berger J (2007) Fear, human shields and the redistribution of prey and predators in protected areas. Biol Lett 3:620–623 Bishop CJ, Unsworth JW, Garton EO (2005) Mule deer survival among adjacent populations in southwest Idaho. J Wildl Manag 69:311–321 Boyce MS (2006) Scale for resource selection functions. Divers Distrib 12:269–276 Brittingham MC, Maloney KO, Farag AM, Harper DD, Bowen ZH (2014) Ecological risks of shale oil and gas development to wildlife, aquatic resources and their habitats. Environ Sci Technol 48:11034–11047 Burton AC, Sam MK, Balangtaa C, Bashares JS (2012) Hierarchical multi-species modeling of carnivore responses to hunting, habitat and prey in a West African protected area. PLoS ONE 7:e38007 Carbone C, Gittleman JL (2002) A common rule for the scaling of carnivore density. Science 295:2273–2276 �Landscape Ecol Chapin FS III, Zavaleta ES, Eviner VT, Naylor RL, Vitousek PM, Reynolds HL, Hooper DU, Lavorel S, Sala OE, Hobbie SE, Mack MC, Dı́az S (2000) Consequences of changing biodiversity. Nature 405:234–242 Courbin N, Fortin D, Dussault C, Fargeot V, Courtois R (2013) Multi-trophic resource selection function enlightens the behavioural game between wolves and their prey. J Anim Ecol 82:1062–1071 Crête M, Larivière S (2003) Estimating the costs of locomotion in snow for coyotes. Can J Zool 81:1808–1814 Crooks KR (2002) Relative sensitivities of mammalian carnivores to habitat fragmentation. Conserv Biol 16:488–502 Crooks KR, Burdett CL, Theobald DM, Rondinini C, Boitani L (2011) Global patterns of fragmentation and connectivity of mammalian carnivore habitat. Philos Trans R Soc B 366:2642–2651 Dickie M, Serrouya R, McNay RS, Boutin S (2017) Faster and farther: wolf movement on linear features and implications for hunting behavior. J Appl Ecol 54:253–263 Dyer SJ, O’Neill JP, Wasel SM, Boutin S (2002) Quantifying barrier effects of roads and seismic lines on movements of female woodland caribou in northeastern Alberta. Can J Zool 80:839–845 Elbroch LM, Lendrum PE, Newby J, Quigley H, Craighead D (2013) Seasonal foraging ecology of non-migratory cougars in a system with migrating prey. PLoS ONE 8:e83375 Erickson ME, Found-Jackson C, Boyce MS (2014) Using latent selection difference to model persistence in a declining population. PLoS ONE 9:e98126 Estes JA, Terborgh J, Brashares JS, Power ME, Berger J, Bond WJ, Carpenter SR, Essington TE, Holt RD, Jackson JBC, Marquis RJ, Oksanen L, Oksanen T, Paine RT, Pikitch EK, Ripple WJ, Sandin SA, Scheffer M, Schoener TW, Shurin JB, Sinclair ARE, Soulé ME, Virtanen R, Wardle DA (2011) Trophic downgrading of planet earth. Science 333:301–306 Ford AT, Goheen JR, Otieno TO, Bidner L, Isbell LA, Palmer TM, Ward D, Woodroffe R, Pringle RM (2014) Large carnivores make savanna tree communities less thorny. Science 346:346–349 Frid A, Dill L (2002) Human-caused disturbance as a form of predation risk. Conserv Ecol 6:11 Gelman A, Hill J (2006) Data analysis using regression and multilevel/hierarchical models. Cambridge University Press, New York Gelman A, Rubin DB (1992) A single series from the Gibbs sampler provides a false sense of security. Bayesian Stat 4:625–631 Hawkins SJ, Charpentier RR, Schenk CJ, Leathers-Miller HM, Klett TR, Brownfield ME, Finn TM, Gaswirth SB, Marra KR, Le PA, Mercier TJ, Pitman JK, Tennyson ME (2016) Assessment of continuous (unconventional) oil and gas resources in the Late Cretaceous Mancos Shale of the Piceance Basin, Uinta-Piceance Province, Colorado and Utah, 2016: US Geological Survey Fact Sheet 2016. http:// dx.doi.org/103133/fs20163030 Hebblewhite M (2005) Predation interacts with the North Pacific Oscillation (NPO) to influence western North American elk population dynamics. J Anim Ecol 74:226–233 Hebblewhite M, Merrill EH, McDonald TL (2005) Spatial decomposition of predation risk using resource selection functions: an example in a wolf–elk predator–prey system. Oikos 111:101–111 Heisey DM, Patterson BR (2006) A review of methods to estimate cause-specific mortality in presence of competing risks. J Wildl Manag 70:1544–1555 Hobbs NT, Hooten MB (2015) Bayesian models: a statistical primer for ecologists. Princeton University Press, New Jersey Hurley MA, Unsworth JW, Zager P, Hebblewhite M, Garton EO, Montgomery DM, Skalski JR, Maycock CL (2011) Demographic response of mule deer to experimental reduction of coyotes and mountain lions in southeastern Idaho. Wildl Monogr 178:1–33 Husseman JS, Murray DL, Power G, Mack C, Wenger CR, Quigley H (2003) Assessing differential prey selection patterns between two sympatric large carnivores. Oikos 101:591–601 Johnson HE, Sushinsky JR, Holland A, Bergman EJ, Balzer T, Garner J, Reed SE (2017) Increases in residential and energy development are associated with reductions in recruitment for a large ungulate. Glob Chang Biol 23:578–591 Karanth KU, Nichols JD, Kumar NS, Link WA, Hines JE (2004) Tigers and their prey: predicting carnivore densities from prey abundance. Proc Natl Acad Sci USA 101:4854–4858 Kertson BN, Spencer RD, Marzluff JM, Hepinstall-Cymerman J, Grue CE (2011) Cougar space use and movements in the wildland–urban landscape of western Washington. Ecol Appl 21:2866–2881 Knopff AA, Knopff KH, Boyce MS, St. Clair CC (2014) Flexible habitat selection by cougars in response to anthropogenic development. Biol Conserv 178:136–145 Krausman PR, Hervert JJ, Ordway LL (1985) Capturing deer and mountain sheep with a netgun. Wildl Soc B 13:71–73 Laberee K, Nelson TA, Stewart BP, McKay T, Stenhouse GB (2014) Oil and gas infrastructure and the spatial pattern of grizzly bear habitat selection in Alberta, Canada. Can Geogr 58:79–94 Larsen S, Ormerod SJ (2014) Anthropogenic modification disrupts species co-occurrence in stream invertebrates. Glob Chang Biol 20:51–60 Laundré JW, Loxterman J (2007) Impact of edge habitat on summer home range size in female pumas. Am Midl Nat 157:221–229 Lazaridis A (2007) A note regarding the condition number: the case of spurious and latent multicollinearity. Qual Quant 41:123–135 Leblond M, Dussault C, Ouellet J-P (2013) Impacts of human disturbance on large prey species: do behavioral reactions translate to fitness consequences? PLoS ONE 8:e73695 Lendrum PE, Anderson CR Jr, Long RA, Kie JG, Bowyer RT (2012) Habitat selection by mule deer during migration: effects of landscape structure and natural-gas development. Ecosphere 3(9):1–19 Lendrum PE, Anderson CR Jr, Monteith KL, Jenks, JA, Bowyer RT (2013) Migrating mule deer: effects of anthropogenically altered landscapes. PLoS ONE 8:e64548 Lendrum PE, Anderson CR Jr, Monteith KL, Jenks, JA, Bowyer RT (2014) Relating the movement of a rapidly migrating 123 �Landscape Ecol ungulate to spatiotemporal patterns of forage quality. Mammal Biol 79:369–375 Lendrum PE, Crooks KR, Wittemyer G (2017) Changes in circadian activity patterns of a wildlife community post high-intensity energy development. J Mammal 98:1265–1271 Leu M, Hanser SE, Knick ST (2008) The human footprint in the west: a large-scale analysis of anthropogenic impacts. Ecol Appl 18:1119–1139 Manly BFJ, McDonald LL, Thomas DL, McDonald TL, Erickson WP (2002) Resource selection by animals, statistical design and analysis for field studies, 2nd edn. Kluwer Academic, Dordrecht McDonald RI, Fargione J, Kiesecker J, Miller WM, Powell J (2009) Energy sprawl or energy efficiency: climate policy impacts on natural habitat for the United States of America. PLoS ONE 4:e6802 Moll RJ, Redilla KM, Mudumba T, Muneza AB, Gray SM, Abade L, Hayward MW, Millspaugh JJ, Montgomery RA (2017) The many faces of fear: a synthesis of the methodological variation in characterizing predation risk. J Anim Ecol 86:749–765 Murray DL, Boutin S (1991) The influence of snow on lynx and coyote movements: does morphology affect behavior? Oecologia 88:463–469 Northrup JM, Anderson CR Jr, Wittemyer G (2015) Quantifying spatial habitat loss from hydrocarbon development through assessing habitat selection patterns of mule deer. Glob Chang Biol 21:3961–3970 Northrup JM, Anderson CR Jr, Wittemyer G (2016) Environmental dynamics and anthropogenic development alter philopatry and space-use in a North American cervid. Divers Distrib 22:547–557 Northrup JM, Wittemyer G (2013) Characterizing the impacts of emerging energy development on wildlife, with an eye towards mitigation. Ecol Lett 16:112–125 Northrup JM, Anderson CR Jr, Gerber BD, Wittemyer G (In revision) Behavioral plasticity buffers mule deer from demographic impacts of energy development. Wildlife Monogr Parker KL, Robbins CT, Hanley TA (1984) Energy expenditures for locomotion by mule deer and elk. J Wildl Manag 48:474–488 Preisser EL, Orrock JL, Schmitz OJ (2007) Predator hunting mode and habitat domain alter nonconsumptive effects in predator–prey interactions. Ecology 88:2744–2751 R Core Team (2015) R: A language and environement for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. http://www.R-project.org/ Ripple WJ, Estes JA, Beschta RL, Wilmers CC, Ritchie EG, Hebblewhite M, Berger J, Elmhagen B, Letnic M, Nelson MP, Schmitz OJ, Smith DW, Wallach AD, Wirsing AJ (2014) Status and ecological effects of the world’s largest carnivores. Science 343:123–148 Rogala JK, Hebblewhite M, Whittington J, White CA, Coleshill J, Musiani M (2011) Human activity differentially 123 View publication stats redistributes large mammals in the Canadian Rockies National Parks. Ecol Soc 16:16 Sawyer H, Nielson RM, Lindzey FG, McDonald LL (2006) Winter habitat selection of mule deer before and during development of a natural gas field. J Wildlife Manag 70:396–403 Schipper J, Chanson JS, Chiozza F, Cox NA, Hoffmann M, Katariya V, Lamoreux J, Rodrigues AS, Stuart SN, Temple HJ, Baillie J, Boitani L, Lacher TE Jr, Mittermeier RA, Smith AT, Absolon D, Aguiar JM, Amori G, Bakkour N, Baldi R, Berridge RJ, Bielby J, Black PA, Blanc JJ, Brooks TM, Burton JA, Butynski TM, Catullo G, Chapman R, Cokeliss Z, Collen B, Conroy J, Cooke JG, da Fonseca GA, Derocher AE, Dublin HT, Duckworth JW, Emmons L, Emslie RH, Festa-Bianchet M, Foster M, Foster S, Garshelis DL, Gates C, Gimenez-Dixon M, Gonzalez S, Gonzalez-Maya JF, Good TC, Hammerson G, Hammond PS, Happold D, Happold M, Hare J, Harris RB, Hawkins CE, Haywood M, Heaney LR, Hedges S, Helgen KM, Hilton-Taylor C, Hussain SA, Ishii N, Jefferson TA, Jenkins RK, Johnston CH, Keith M, Kingdon J, Knox DH, Kovacs KM, Langhammer P, Leus K, Lewison R, Lichtenstein G, Lowry LF, Macavoy Z, Mace GM, Mallon DP, Masi M, McKnight MW, Medellı́n RA, Medici P, Mills G, Moehlman PD, Molur S, Mora A, Nowell K, Oates JF, Olech W, Oliver WR, Oprea M, Patterson BD, Perrin WF, Polidoro BA, Pollock C, Powel A, Protas Y, Racey P, Ragle J, Ramani P, Rathbun G, Reeves RR, Reilly SB, Reynolds JE 3rd, Rondinini C, Rosell-Ambal RG, Rulli M, Rylands AB, Savini S, Schank CJ, Sechrest W, Self-Sullivan C, Shoemaker A, Sillero-Zubiri C, De Silva N, Smith DE, Srinivasulu C, Stephenson PJ, van Strien N, Talukdar BK, Taylor BL, Timmins R, Tirira DG, Tognelli MF, Tsytsulina K, Veiga LM, Vié JC, Williamson EA, Wyatt SA, Xie Y, Young BE (2008) The status of the world’s land and marine mammals: diversity, threat, and knowledge. Science 322:225–230 Stonehouse FK, Anderson CR Jr, Peterson ME, Collins DR (2016) Approaches to field investigations of cause–specific mortality in mule deer. Technical Publication No 48, Colorado Parks and Wildlife, Fort Collins, CO, USA Telfer ES, Kelsall JP (1984) Adaptation of some large North American mammals for survival in snow. Ecology 65:1828–1834 Unsworth JW, Pac DF, White GC, Bartmann RM (1999) Mule deer survival in Colorado, Idaho, and Montana. J Wildl Manag 63:315–326 Whittington J, Hebblewhite M, DeCesare NJ, Neufeld L, Bradley M, Wilmshurst J, Musiani M (2011) Caribou encounters with wolves increase near roads and trails: a time-to-event approach. J Appl Ecol 48:1535–1542 Wilcove DS, Rothstein D, Dubow J, Phillips A, Losos E (1998) Quantifying threats to imperiled species in the United States. Bioscience 48:607–615 � https://cpw.cvlcollections.org/files/original/fd557461ecdffc62a231db7cb923c533.pdf 4643a31498244b5ca26a10a805f07be7 PDF Text Text Appendix 1. Descriptive statistics of variables examined in predation site selection, GPS deer locations, and random locations in the undeveloped area, and developed area during the periods of high and low development: terrain ruggedness (VRM), snow depth; and distance to: ecotone edge (between forest and shrubs), drilling well pads, producing well pads, pipelines, secondary roads, primary roads, and industrial facilities during the winter in the Piceance Basin, CO, USA, 2008-2014. Variable and Site Predation sites Mean SD Median Min Max VRM Undeveloped high 0.14 0.16 0.05 0.00 0.81 Developed high 0.16 0.19 0.08 0.00 0.83 Undeveloped low 0.13 0.20 0.05 0.00 0.55 Developed low 0.11 0.15 0.05 0.00 0.77 Snow Undeveloped high 0.18 0.10 0.18 0.00 0.43 Developed high 0.13 0.11 0.12 0.00 0.41 Undeveloped low 0.13 0.14 0.14 0.00 0.32 Developed low 0.10 0.09 0.08 0.00 0.27 Ecotone edge Undeveloped high 65.17 134.93 32.31 0.45 352.61 Developed high 70.26 92.49 32.65 0.25 607.58 Undeveloped low 101.28 83.80 38.76 2.51 514.38 Developed low 83.48 87.68 57.74 1.26 333.43 Drilling Undeveloped high 6331.80 2701.54 6058.32 448.09 15771.92 Developed high 3124.40 2101.95 2735.89 323.62 14054.96 Undeveloped low 7666.85 3439.06 7262.23 3615.77 14971.30 Developed low 4010.60 2416.28 3849.84 338.27 14294.39 Producing Undeveloped high 2048.32 711.34 2107.70 371.16 3777.16 Developed high 900.09 624.52 744.45 139.13 3413.36 Undeveloped low 1910.86 907.21 1953.20 368.16 3555.12 Developed low 770.79 536.32 609.28 69.63 2628.22 Pipeline Undeveloped high 776.50 498.62 653.04 9.50 2111.87 Developed high 282.12 256.96 232.98 1.10 1269.54 Undeveloped low 608.68 545.52 415.73 7.73 1949.02 Developed low 304.95 282.73 210.03 1.67 1419.86 Major roads Undeveloped high 846.14 637.41 686.51 2.63 2434.72 Developed high 779.70 719.22 540.48 4.46 3294.76 �Undeveloped low Developed low Secondary roads Undeveloped high Developed high Undeveloped low Developed low Facility Undeveloped high Developed high Undeveloped low Developed low Variable and Site VRM Undeveloped high Developed high Undeveloped low Developed low Snow Undeveloped high Developed high Undeveloped low Developed low Ecotone edge Undeveloped high Developed high Undeveloped low Developed low Drilling Undeveloped high Developed high Undeveloped low Developed low Producing Undeveloped high Developed high Undeveloped low Developed low Pipeline 785.63 740.87 604.61 799.45 643.51 447.69 9.57 1.98 2771.02 3635.37 363.54 271.88 409.60 231.96 284.54 218.56 255.35 178.25 310.34 220.21 404.07 184.70 0.36 18.20 3.62 1.32 1059.48 1062.04 1334.09 708.40 3235.66 1403.26 3192.57 1558.03 1423.33 874.84 1349.71 1387.55 3294.92 1201.55 3037.51 1146.16 342.50 62.06 503.75 0.00 5639.30 3972.16 5703.94 9491.87 GPS locations Mean SD Max Random locations Mean SD Median Min Median M 0.12 0.14 0.10 0.14 0.16 0.17 0.15 0.17 0.05 0.07 0.04 0.06 0.00 0.00 0.00 0.00 0.97 0.98 0.96 0.98 0.12 0.13 0.12 0.13 0.16 0.17 0.16 0.16 0.05 0.06 0.05 0.05 0 0 0 0 0.14 0.10 0.14 0.09 0.13 0.12 0.11 0.09 0.11 0.05 0.14 0.07 0.00 0.00 0.00 0.00 0.56 0.66 0.48 0.51 0.15 0.11 0.12 0.09 0.15 0.13 0.11 0.09 0.11 0.05 0.12 0.07 0 0 0 0 95.38 59.78 135.05 70.69 139.72 69.55 163.96 105.77 39.13 35.14 60.05 34.24 0.00 0.00 0.00 0.00 888.22 572.29 888.26 888.26 80.95 62.43 88.54 73.92 117.20 72.73 124.70 96.18 34.42 35.90 37.01 38.75 0 0 0 0 6019.83 3099.77 7041.10 4177.49 3011.92 2043.91 3004.90 1703.10 5977.01 2687.90 6626.15 4241.07 12.70 12.70 2315.50 24.28 16956.21 13972.06 18818.39 15560.62 6948.43 3392.38 8707.96 4878.73 3778.29 2362.91 3442.48 2959.76 6368.57 2960.34 8223.15 4307.75 0 0 2 0 1919.59 811.67 1539.78 707.51 922.08 549.48 886.87 511.63 1915.77 703.10 1438.01 597.76 0.00 0.00 0.00 0.00 4665.89 3942.03 5874.71 3137.18 1997.03 875.58 1783.95 764.93 969.19 630.07 847.79 582.39 1946.73 732.44 1772.80 620.87 0 0 0 0 �Undeveloped high 572.03 495.06 422.50 0.02 2389.88 723.18 564.38 Developed high 245.86 228.83 177.88 0.00 2317.40 308.19 323.28 Undeveloped low 491.17 488.59 307.00 0.00 2215.53 622.98 507.60 Developed low 263.89 271.41 180.70 0.00 2307.43 320.80 331.53 Major roads Undeveloped high 849.64 724.96 602.85 0.01 3567.30 838.22 701.99 Developed high 818.90 780.65 573.93 0.04 3966.31 772.17 726.19 Undeveloped low 958.98 816.06 652.24 0.18 3617.54 842.17 677.49 Developed low 1120.16 965.15 831.17 0.10 3973.88 768.45 775.94 Secondary roads Undeveloped high 342.96 240.44 292.67 0.08 1404.54 395.97 295.21 Developed high 255.09 203.06 205.27 0.01 1469.24 282.79 230.36 Undeveloped low 335.28 253.06 271.75 0.01 1535.11 396.14 299.88 Developed low 243.02 185.34 200.87 0.00 1285.11 267.73 212.80 Facility Undeveloped high 3277.58 1556.52 3732.28 0.00 6338.56 3251.22 1413.97 Developed high 1529.79 903.20 1401.02 0.00 4677.28 1598.26 956.65 Undeveloped low 2574.79 1394.76 2477.69 0.00 5777.28 3256.40 1355.71 Developed low 1869.25 1123.11 1683.85 0.00 9482.86 1887.25 1637.64 Appendix 2. A terrain ruggedness index was derived from a digital-elevation model (DEM) at a resolution of 30 m (http://datagateway.nrcs.usda.gov/) following the method of Sappington et al. (2007), ranging between 0 (flat) and 1 (most rugged). We reclassified the 87 vegetation classes provided by the Colorado Vegetation Classification Project (http://ndis.nrel.colostate.edu/coveg/) layer, at a resolution of 25 m, into three broad categories of concealment based on similarity of vegetation types: (1) forbs, grasslands, and barren habitat types (low concealment); (2) shrub dominant (moderate concealment); and (3) forested habitats (high concealment). Snow depth was predicted for each day of the study at a resolution of 30 m from a distributed snow evolution model (SnowModel; Liston and Elder 2006). Variable inputs required for SnowModel include temporally varying fields of precipitation, wind speed and direction, air temperature, and relative humidity obtained from meteorological stations and an atmospheric model located within or near the simulation domain; and spatially distributed fields of topography and vegetation type. Represented processes include accumulation from snow precipitation; blowing-snow redistribution and sublimation; interception, unloading, and sublimation within forest canopies; snow-density evolution; and snowpack ripening and melt (Liston and Elder 2006). Locations of well pads and industrial facilities were obtained from the Colorado Oil and Gas Conservation Commission (http://cogcc.state.co.us/) from Dec 2008 – August 2014, which designated the date and location that each well pad was in a development (actively being prepared and drilled) or production (post drilling and actively extracting natural gas) phase (see Northrup et al. 2015 for further details). A roads layer was derived by combining the TIGER/Line shape files of the U.S. Census Bureau (http://www.census.gov/geo/www/tiger/shp.html) and the Colorado Department of Transportation shape files (http://apps.coloradodot.info/dataaccess/). We considered county roads as primary roads and spur roads used for purposes of natural-gas extraction as secondary roads, but we were unable to differentiate levels of vehicle use among roads. Locations of 607.44 210.17 502.92 214.90 0 0 0 0 643.98 570.97 670.20 517.13 0 0 0 0 331.80 230.02 328.12 222.33 0 0 0 0 3366.41 1457.77 3279.07 1480.01 0 0 0 0 �pipelines were obtained from the Bureau of Land Management White River field office. The spatial and temporal information for all landscape disturbances were validated or corrected with National Agriculture Imagery Program aerial images (http://datagateway.nrcs.usda.gov/) from 2009, 2011, and 2013. Liston GE, Elder K (2006) A distributed snow-evolution modeling system (SnowModel). J Hydrometeorol 7:1259–1276 Sappington JM, Longshore KM, Thompson DB (2007) Quantifying landscape ruggedness for animal habitat analysis: a case study using bighorn sheep in the Mojave Desert. J Wildlife Manage 71:1419–1426 Appendix 3. Interpretation of resource selection function (RSF) and latent selection difference function (LSDF) for three different categories of covariates: binary (or categorical) such as habitat type; position-based such as snow depth, terrain ruggedness; proximity-based such as distance to ecotone edge or distance to nearest road Coefficient RSF LSDF Binary covariates + greater than expected at random greater than prey distribution less than expected at random less than prey distribution = at random equal to prey distribution Position based covariates + higher than expected at random higher than prey distribution lower than expected at random lower than prey distribution = at random equal to prey distribution Proximity based covariates + further than expected at random further than prey distribution closer than expected at random closer than prey distribution = at random equal to prey distribution � 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 Predation risk across a dynamic landscape: effects of anthropogenic land use, natural landscape features, and prey distribution Description An account of the resource <p class="c-article__sub-heading">Purpose</p> <p>Human-mediated landscape changes alter habitat configuration, which strongly structures animal distributions and interspecific interactions. The effects of anthropogenic disturbance on predator–prey relationships are fundamental to ecology, yet less well understood. We determined where predation events occurred for fawn and adult female mule deer from 2008 to 2014 in critical winter range with extensive energy development. We investigated the relationship between predation sites, energy infrastructure, and natural landscape features across contiguous areas experiencing different degrees of energy extraction during periods of high and low intensity development.</p> <p class="c-article__sub-heading">Methods</p> <p>We contrast spatial correlates of 286 mortality locations with random landscape locations and mule deer distribution estimated from 350,000 GPS locations. We estimated predation risk with resource selection functions and latent selection difference functions.</p> <p class="c-article__sub-heading">Results</p> <p>Relative to the distribution of mule deer, predation risk was lower closer to pipelines and well pads, but higher closer to roads. Predation sites occurred more than expected relative to availability and deer distribution in deeper snow and non-forested habitats. Anthropogenic features had a greater influence on predation sites during the period of low activity than high activity, and natural landscape characteristics had weaker effects relative to anthropogenic features throughout the study. Though canids accounted for the majority of predation events, felids exhibited stronger landscape associations, driving the observed spatial patterns in predation risk to mule deer.</p> <p class="c-article__sub-heading">Conclusions</p> <p>The emergence of varied interactions between predation and landscape features across contexts and years highlights the complexity of interspecific interactions in highly modified landscapes.</p> Bibliographic Citation A bibliographic reference for the resource. Recommended practice is to include sufficient bibliographic detail to identify the resource as unambiguously as possible. Lendrum, P. E., J. M. Northrup, C. R. Anderson, G. E. Liston, C. L. Aldridge, K. R. Crooks, and G. Wittemyer. 2017. Predation risk across a dynamic landscape: effects of anthropogenic land use, natural landscape features, and prey distribution. Landscape Ecology 33:151-170. <a href="https://doi.org/10.1007/s10980-017-0590-z" target="_blank" rel="noreferrer noopener">https://doi.org/10.1007/s10980-017-0590-z</a> Creator An entity primarily responsible for making the resource Lendrum, Patrick E. Northrup, Joseph M. Anderson Jr, Charles R. Liston, Glen E. Aldridge, Cameron L. Crooks, Kevin R. Wittemyer, George Subject The topic of the resource Carnivores Disturbance ecology Energy development Habitat fragmentation Mule deer Predation risk Resource selection Extent The size or duration of the resource. 14 pages Date Created Date of creation of the resource. 2017-10-31 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> Format The file format, physical medium, or dimensions of the resource application/pdf Language A language of the resource English Is Part Of A related resource in which the described resource is physically or logically included. Landscape Ecology Type The nature or genre of the resource Article