https://cpw.cvlcollections.org/files/original/6d4aee6981872e05c6f19b9f8674665e.pdf 5a816af15d773053bb41ab36e8a426bc 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 �Global Change Biology (2015) 21, 3961–3970, doi: 10.1111/gcb.13037 Quantifying spatial habitat loss from hydrocarbon development through assessing habitat selection patterns of mule deer J O S E P H M . N O R T H R U P 1 , C H A R L E S R . A N D E R S O N J R . 2 and G E O R G E W I T T E M Y E R 1 , 3 1 Department of Fish, Wildlife and Conservation Biology, Colorado State University, Fort Collins, CO, USA, 2Mammals Research Section, Colorado Parks and Wildlife, Fort Collins, CO, USA, 3Graduate Degree Program in Ecology, Colorado State University, Fort Collins, CO, USA Abstract Extraction of oil and natural gas (hydrocarbons) from shale is increasing rapidly in North America, with documented impacts to native species and ecosystems. With shale oil and gas resources on nearly every continent, this development is set to become a major driver of global land-use change. It is increasingly critical to quantify spatial habitat loss driven by this development to implement effective mitigation strategies and develop habitat offsets. Habitat selection is a fundamental ecological process, influencing both individual fitness and population-level distribution on the landscape. Examinations of habitat selection provide a natural means for understanding spatial impacts. We examined the impact of natural gas development on habitat selection patterns of mule deer on their winter range in Colorado. We fit resource selection functions in a Bayesian hierarchical framework, with habitat availability defined using a movement-based modeling approach. Energy development drove considerable alterations to deer habitat selection patterns, with the most substantial impacts manifested as avoidance of well pads with active drilling to a distance of at least 800 m. Deer displayed more nuanced responses to other infrastructure, avoiding pads with active production and roads to a greater degree during the day than night. In aggregate, these responses equate to alteration of behavior by human development in over 50% of the critical winter range in our study area during the day and over 25% at night. Compared to other regions, the topographic and vegetative diversity in the study area appear to provide refugia that allow deer to behaviorally mediate some of the impacts of development. This study, and the methods we employed, provides a template for quantifying spatial take by industrial activities in natural areas and the results offer guidance for policy makers, mangers, and industry when attempting to mitigate habitat loss due to energy development. Keywords: animal movement, Bayesian hierarchical model, energy development, habitat selection, movement ecology, mule deer, natural gas, resource selection function Received 2 April 2015 and accepted 30 June 2015 Introduction Since the early 2000s, the exploration and production (hereafter development) of hydrocarbons has increased rapidly in North America (United States Energy Information Administration [USEIA] 2012). The landscapelevel disturbance resulting from this development has had a number of negative impacts on wildlife, including driving population declines and causing large-scale spatial displacement (Northrup & Wittemyer, 2013). This recent energy boom has been driven primarily by the development of shale resources (USEIA, 2012). *Correspondence: Joseph M. Northrup, tel: +970 491 6598, fax: +970 491 5091, e-mail: joe.northrup@gmail.com [Correction added on 30 September, 2015, after first online publication: The value pertaining to the Density of natural gas well pads across the study area for the year 2010 under the section “Materials and Methods” was changed from “20” to “0.20”] Shale resources are proven to exist on every continent save Antarctica and their development is projected to continue to increase (USEIA, 2013). Thus, this sector is poised to become a major driver of global land-use change and impacts to biodiversity. Given its projected global footprint, there is a pressing need for robust quantification of habitat loss driven by hydrocarbon development. Such information can be used to aid in development planning, assess cumulative impacts, develop mitigation measures, and quantify the size of habitat offsets. However, understanding the impact of hydrocarbon development and subsequent mitigation measures is complex as the associated disturbances are spatially variable and temporally dynamic and their cumulative effects not well understood, which can obfuscate animal responses. In light of this complexity, there is a need for more complete information on the ways in which animals respond to © 2015 The Authors. Global Change Biology Published by John Wiley & Sons Ltd. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes. 3961 �3962 J . M . N O R T H R U P et al. development. Detailed understanding of the distance at which different types of development elicit responses from different species will be particularly important for quantifying habitat impacts and identifying effective mitigation strategies as this industry becomes a major driver of global land-use change. Examinations of the habitat selection patterns of wildlife provide a natural means for understanding the spatial impacts of development. Habitat selection is a fundamental ecological process by which animals distribute themselves across landscapes by selecting habitats that maximize their fitness (Fretwell & Lucas, 1969). Examinations of habitat selection provide insight into individual-based ecological processes (e.g., drivers of site fidelity: Creel et al., 2005; and tradeoffs between foraging and predation risk: Switzer, 1997), but also to larger scale factors that influence population distribution and abundance (e.g., population dynamics: Pulliam & Danielson, 1991; speciation: Rice, 1987; and dispersal: Shafer et al., 2012). Human disturbance can alter habitat selection patterns of animals (e.g., Sawyer et al., 2006), but the nature of this response and the subsequent ramifications for different species are complex. Humans directly convert habitat, but their activities also lead to functional habitat loss disproportionately greater than the area that is directly disturbed (e.g., Sawyer et al., 2006). Responses also can be more nuanced, with humans being perceived as akin to predators, driving behavioral shifts reflecting tradeoffs between security and other demands such as foraging or reproduction (Frid & Dill, 2002; Hebblewhite & Merrill, 2008). Alternatively, animals can be attracted to human developments due to associated resources, or as protection against predation (Berger, 2007). This attraction can positively impact animals, but can also lead to greater potential for negative encounters with humans (Johnson et al., 2004) and the formation of ecological traps (e.g., Northrup et al., 2012b). In light of the array of complex responses of animals to human disturbance, research on the mechanisms driving changes in wildlife behavior are critical for developing appropriate mitigation measures. In many areas of the western United States hydrocarbon development has taken place on mule deer (Odocoileus hemionus Rafinesque) winter range, where the species faces acute welfare issues related to decreased access to high quality forage (Parker et al., 1984). Mule deer have experienced major population declines across their range (Unsworth et al., 1999) and recent studies have shown deer to experience alterations of habitat selection patterns and large scale displacement in response to hydrocarbon development (Sawyer et al., 2006, 2009). Obtaining information on the impact of development on deer habitat selection patterns is thus a major management priority, as extraction is projected to continue to increase over the next several decades (USEIA, 2014). We fit resource selection functions (RSFs) in a hierarchical Bayesian framework to understand responses of a mule deer population to hydrocarbon development on winter range. Resource selection functions are the most commonly used approach to examine the habitat selection process, but a major methodological and conceptual hurdle to their application is the sensitivity of results to definitions of habitat availability (Johnson, 1980; Hooten et al., 2013; Lele et al., 2013; Northrup et al., 2013). With technological advances in global positioning system (GPS) radio collars, animal location data are being collected at increasingly fine scales revealing complex temporal autocorrelation structures (Wittemyer et al., 2008; Boyce et al., 2010) that can compound methodological issues related to defining availability in RSF analyses. Although methods exist for potentially managing this autocorrelation (see Fieberg et al., 2010 for a review), approaches for addressing autocorrelation at the scale of the availability sample are limited. Using methods developed in the animal movement literature, Hooten et al. (2013) propose a dynamic movement-based method for determining availability on an individual animal and location-specific basis. We apply a similar methodology to address three questions: (1) how does hydrocarbon development (roads and well pads) influence deer habitat selection?; (2) do deer respond to energy development differently at night than during the day?; and (3) at what spatial scale do mule deer most strongly respond to different development features? Our results provide insights into the spatial and temporal factors influencing mule deer habitat selection and the influence of energy development on this behavior, while offering a template for assessing spatial habitat loss and guidance for the mitigation of development impacts on wildlife. Materials and methods Study area We examined mule deer habitat selection on winter range in the Piceance Basin in Northwestern Colorado, USA, (39.954°N, 108.356°W; Fig. 1), during a time of ongoing development of natural gas. Deer in this area migrate from high elevations during the summer to low elevations during the winter, with winter range occupancy generally occurring between October and May (Lendrum et al., 2013; Northrup et al., 2014b). The area is topographically diverse and dominated by sagebrush (Artemisia tridentata Nutt.) and a pinyon pine (Pinus edulis Engelm.) and Utah Juniper (Juniperus osteosperma Torr.) shrubland complex. The vegetation of the area is © 2015 The Authors. Global Change Biology Published by John Wiley & Sons Ltd., 21, 3961–3970 �H A B I T A T L O S S F R O M H Y D R O C A R B O N D E V E L O P M E N T 3963 Fig. 1 Study area location in the United States, closet human settlement (Meeker, CO), and outline of winter range study area used by adult female mule deer in the Piceance Basin of Northwest Colorado. described in detail by Bartmann & Steinert (1981) and Bartmann et al. (1992). The dominant human activity in the area is natural gas development, with winter cattle grazing occurring primarily in the valley bottoms. Density of natural gas well pads across the study area were approximately 0.18 well pads km�2 in 2008, 0.20 well pads km�2 in 2009 and 0.20 well pads km�2 in 2010, although local densities ranged from 0–6 pads km�2. The area is popular for hunting during the fall, and experiences warm, dry summers with monsoonal precipitation and cold winters, with the majority of moisture resulting from snow melt in the spring. Mule deer data We monitored adult (>1 year old) female mule deer on their winter range between January 2008 and December 2010. Adult females were studied as they are the age and sex class known to be the dominant driver of population dynamics. Deer were captured using helicopter net gunning during December and March of each year and were fit with store-on-board global positioning system (GPS) radio collars (G2110D, Advance Telemetry Systems, Istanti, MN, USA and model 4400, Lotek Wireless, Newmarket, ON, Canada) programmed to attempt a relocation once every 5 hours. All procedures were approved by the Colorado State University (protocol ID: 10-2350A) and Colorado Parks and Wildlife (protocol ID: 15-2008) Animal Care and Use Committees. Collars were equipped with timed release mechanisms set to release after 16 months. At this time collars were recovered, and data were downloaded. Due to the potential behavioral impacts of capture on mule deer (Northrup et al., 2014a), we censored all data for one week following capture. Deer in this area are migratory, so we only included data occurring between the termination of fall migration and the initiation of spring migration. Migration termination and initiation were estimated visually in ArcMap 10 (Environmental Systems Research Institute, Redlands, CA, USA). We removed all locations for which the positional dilution of precision (PDOP) was >10 (<1% of locations: D’eon & Delparte, 2005; Lewis et al., 2007). We calculated the percent of successful GPS fixes for each individual by dividing the number of total locations by the number of attempted fixes. Overall fix success rate was 93%, which exceeds the threshold commonly used to indicate the need for habitat-bias corrections in habitat modeling (Frair et al., 2004; Hebblewhite et al., 2007). Lastly, we divided locations into night and day, with night classified as the time between sunset and sunrise (http://aa.usno. navy.mil/data/docs/RS_One Year.php). Predictor variables We chose an approach for RSF modeling that maximized our understanding of the impacts of development. We first chose a set of environmental covariates for RSF modeling that we hypothesized to be important predictors of deer resource selection based on previous studies (Pierce et al., 2004; Sawyer et al., 2006, 2009; Stewart et al., 2010). These covariates represented our best understanding of how mule deer selected © 2015 The Authors. Global Change Biology Published by John Wiley & Sons Ltd., 21, 3961–3970 �3964 J . M . N O R T H R U P et al. habitat. We then added to these covariates different representations of natural gas development to understand deer response to development while also accounting for environmental characteristics that they are known to respond to. The environmental covariates included the terrain variables slope (slope), and elevation (elev), calculated from a digital elevation model. In addition, we obtained land cover data from the Colorado Vegetation Classification Project (http:// ndis.nrel.colostate.edu/coveg/). This land cover database has 69 classes, however our study area is dominated by two classes (44% sagebrush and 39% pinyon-juniper). While numerous land cover classes existed in our study area, all classes other than pinyon-juniper and sagebrush were relatively rare, and not all deer interacted with all of the other categories. Thus, we combined all categories into two classes that all deer interacted with. These classes consisted of treed or open (tree). Lastly we calculated the distance to treed edges (d_edge). We also digitized all roads in the study area from aerial imagery from the National Agricultural Imagery Program (NAIP) and calculated the distance to the nearest road from each location (d_rds), and also included a quadratic term for the distance to roads. To obtain information on development, we downloaded the location of all oil and natural gas wells in the study area from the Colorado Oil and Gas Conservation Commission website (cogcc.state.co.us), which maintains a daily updated database of the locations, drilling onset date and drilling completion date of oil and natural gas wells throughout the state. We classified each well in our study area into one of three classes: (1) wells actively being drilled (wells in this stage generally see continuous 24/7 human activity); (2) wells that were actively producing natural gas with no drilling activity; and (3) wells that were abandoned (see Appendix S1 for further details). We created a series of time-specific spatial layers representing the status of each well accurate to the day. These layers were generated for the entire time period during which collared deer were active on winter range in the study area (Oct–May of each year). We grouped individual wells by pad visually using a layer for well pads digitized from the NAIP imagery. We then classified each pad as a drilling, producing, or abandoned pad for every day of the study period. If a pad had any wells that were being actively drilled, the entire pad was classified as drilling. Likewise, if the pad had both abandoned and producing wells, it was classified as producing. Our ultimate unit of replication was the well pad, as pads can contain multiple wells that can be in different stages. Using the resulting data, we created different covariates to represent active natural gas development. Our approach consisted of fitting a single model structure with nested concentric buffers around well pads. Including concentric buffers in the models allows us to identify the distance at which deer ceased to respond to well pads. We created eight covariates for this model: the number of well pads within 400 m (measured to the edge of the well pad; drill_400 and prod_400), the number of pads between 400–600 m (drill_600 and prod_600), the number of pads between 600–800 m (drill_800 and prod_800) and the number of pads between 800–1000 m (drill_1000 and prod_1000). The smallest buffer distance assessed (i.e., 400 m) corresponded to the approximate mean distance moved between successful deer relocations spaced 5 h apart. We initially attempted to assess responses to the number of pads within 200 m but convergence failed for both night and day models that included these covariates after more than 2 million iterations (traceplots showed poor mixing). On closer examination, this appeared to result from few deer locations within 200 m of well pads classified as being drilled [23 locations during the night (0.17% of night time locations) and 17 locations during the day (0.11% of daytime locations)]. We excluded abandoned pads from analysis as there was no extraction activity associated with these features. Model formulation We estimated RSFs separately for night and day locations using hierarchical conditional logistic regression (sensu Duchesne et al., 2010), in a Bayesian framework where all coefficients varied by individual. In this framework, each used location is paired with a set of random locations drawn from an area deemed to be immediately available to the animal at that time (Boyce, 2006). Following Revelt & Train (1998), and Duchesne et al. (2010), the probability that an animal (n) chooses a resource unit (y) represented by a suite of habitat covariates (xy) from a set of available alternative resource units (J), represented by suite of habitat covariates (xj) at time t can be written as follows: expðx0ytn bn Þ ½ytn jbn � ¼ PJ 0 j¼1 expðxjtn bn Þ Using this probability mass function, we can estimate coefficients for each individual and the population as a whole by placing the model in a Bayesian hierarchical framework as follows: bn � normal ðlb ; r2b IÞ lb � normal ð0; 1000000IÞ logðr2b Þ � normal ð0; 1000000Þ Characterizing availability In a RSF model using conditional logistic regression, each used location is paired with random locations sampled within a distance of the used location presumed to be immediately available to the animal (Boyce, 2006). There is no standard approach for determining this distance for drawing availability, although methods in the literature include using the average distance moved between successive GPS locations (Boyce et al., 2003), or drawing from empirical step length and turn angle distributions (Fortin et al., 2005). Although such methods clearly have biological underpinnings, few definitions of availability have accounted for the dynamic movement behavior of animals. With the proliferation of studies using GPS radio collar data, there currently exists an array of methods, developed in © 2015 The Authors. Global Change Biology Published by John Wiley & Sons Ltd., 21, 3961–3970 �H A B I T A T L O S S F R O M H Y D R O C A R B O N D E V E L O P M E N T 3965 the animal movement literature, that model the dynamic movement behavior of animals (e.g., Hooten et al., 2013). We used the continuous-time correlated random walk (CTCRW) model described by Johnson et al. (2008) to categorize availability (sensu Hooten et al., 2013). The CTCRW model describes movement as an Ornstein-Uhlenbeck process, where the velocity of an animal at the current time step is dependent on its previous velocity, an autocorrelation parameter, and an error term scaled by the time between known locations (Johnson et al., 2008). Hooten et al. (2013) use the results of the CTCRW model to characterize resource availability as the predictor distribution for the location and velocity of an animal at any time, which is a description of the uncertainty in the location at the current time given all preceding data. The location of this predictor distribution varies with the location and speed of the animal and thus its position relative to the used location is dynamic and dependent on the current behavior of the animal. We fit the a CTCRW model for each individual animal using the ‘crawl’ package (Johnson et al., 2008) in the R statistical software (R Core Team, 2013). The coordinates of a set of random locations were drawn from the predictor distribution specific to each used location. To ensure a sufficiently large availability sample (Northrup et al., 2013), we explored the stability of coefficient estimates from models fit to varying availability sample sizes (5, 25, 50, 100, 250, 500 and 1000 random locations per used location). Drawing from a set of 10 000 random locations per observed location, we ran 25 models at each availability sample size to examine variation in coefficient estimates as a function of the availability sample. Once the sample size that provided stable covariate estimates had been determined, we drew that number of random locations for each used location for each individual for the hierarchical model described above. Model fitting Using the model formulation and data described above, we fit models to deer locations across all years. We first standardized all continuous predictor covariates ðx � �xÞ=r. We tested for correlations among covariates that appeared in the same model (Appendix S2) to ensure that no covariates were highly correlated (|r| > 0.7). Using the Bayesian hierarchical framework described above, we fit RSFs using a Markov-Chain Monte Carlo (MCMC) procedure written in the R programming language. We ran two parallel chains for each model for 1 000 000 iterations, discarding the first 100 000 as burn-in. We selected starting values for each parameter chain that were expected to be overdispersed relative to the posterior distributions and monitored convergence to the posterior distribution by examining traceplots of MCMC samples against iterations to determine if there was proper mixing, and by calculating the Gelman-Rubin diagnostic (mean values <1.1 indicate convergence; Gelman & Rubin, 1992). In addition to fitting the single model structure discussed above, we also fit a set of models each with a single covariate representing the number of well pads within overlapping buffer distances (see Appendix S2 for more details). This approach was taken to aid our inference in relation to the concentric buffers analysis discussed above. One of the most basic assumptions of model fitting is that the model is a faithful representation of the data generating process. To test this assumption we performed posterior predictive checks (Appendix S2; see examples in Gelman & Hill, 2007). Results Model specifications We monitored 53 adult female mule deer across 3 years (18 per year, with one individual collared for consecutive years), for a total of 29 083 winter range (Oct–May) locations (�x = 548.7 locations per deer). Both 250 and 500 available locations per used location provided sufficiently accurate estimation of coefficients. Upon initiation of model fitting, 500 locations proved to be computationally infeasible on a high-performance supercomputer. Thus we included 300 available locations per used location. All parameters converged to their posterior distribution (i.e., all mean Gelman-Rubin values were <1.1). There were strong similarities to the models fit with concentric buffers and additional models fit with single covariates representing the number of pads within overlapping buffers (Appendix S2), thus we only present results of the concentric buffers analysis. Ecological and anthropogenic drivers of selection Deer selected open areas over treed areas and areas further from edges during the night, while during the day, deer selected treed areas over open areas and areas closer to edges (Fig. 2). In addition, deer selected areas closer to roads during the night than during the day (Fig. 3). Throughout the day and night, deer selected areas with steeper slopes and at higher elevations, although the strength of this selection was greater during the night (Table 1, Fig. 2). Deer responses to drilling and producing well pads varied by buffer distance (Fig. 4). During both night and day deer avoided drilling well pads at the 0–400 m buffer and the 400–600 m buffer. During the night this avoidance persisted to the 600–800 m and 800–1000 m buffers, but was relatively weak at the furthest buffer distance. Contrarily, deer showed no avoidance of the areas 600–800 m and 800– 1000 m from drilling well pads during the day (Fig. 4). During the day, deer also avoided well pads actively producing natural gas at the 0–400 m buffer and 400– 600 m buffer, while showing no avoidance of the areas between 600–1000 m from these pads (Fig. 4). During the night deer displayed weaker avoidance of producing well pads within the smallest buffer (0–400 m) than © 2015 The Authors. Global Change Biology Published by John Wiley & Sons Ltd., 21, 3961–3970 �3966 J . M . N O R T H R U P et al. Table 1 Covariate names, median posterior coefficient (coeff.) values, and proportion (prop.) of posteriors above and below 0 for resource selection function models fit to GPS data from adult female mule deer in the Piceance Basin, Colorado, USA during the night and day separately. Covariate Night coeff. Night prop. < 0 Night prop. > 0 Day coeff. Day prop. < 0 Day prop. > 0 d_edge slope elev d_rds d_rds2 tree prod_400 drill_400 prod_600 drill_600 prod_800 drill_800 prod_1000 drill_1000 0.11 0.17 0.91 �0.35 �0.43 �0.27 �0.06 �0.73 0.08 �0.40 0.12 �0.27 0.07 �0.09 0.00 0.00 0.00 1.00 1.00 1.00 0.71 0.99 0.19 0.96 0.03 0.95 0.05 0.78 1.00 1.00 1.00 0.00 0.00 0.00 0.29 0.01 0.81 0.04 0.97 0.05 0.95 0.22 �0.17 0.05 0.69 0.17 �0.30 0.08 �0.41 �0.82 �0.14 �0.28 �0.04 0.00 0.02 0.04 1.00 0.01 0.00 0.00 1.00 0.01 1.00 1.00 0.98 0.99 0.77 0.49 0.29 0.29 0.00 0.99 1.00 1.00 0.00 0.99 0.00 0.00 0.02 0.01 0.23 0.51 0.71 0.71 (a) (b) Fig. 2 Posterior distributions of population-level coefficients for RSF models during the (a) day and (b) night for 53 adult female mule deer in the Piceance Basin, Northwest Colorado. Dashed line indicates 0 selection or avoidance of the habitat features. Displayed coefficients are for non-well pad covariates only, but are taken from models including well pad covariates. ‘Edge’ refers to the distance to treed edges in meters, ‘Slope’ was measured in degrees, ‘Elev’ refers to elevation in meters, ‘Roads’ refers to the distance to roads in meters, and ‘Tree’ refers to treed land cover. during the day, while displaying selection for areas at all other buffer distances (Fig. 4). Discussion Hydrocarbon development is projected to continue to increase in the United States and elsewhere (USEIA, 2013, 2014) and thus is set to become a major driver of global land-use change. As such, there is a pressing need for assessments of the nature of impacts to wild- life and the spatial extent to which these impacts extend. The habitat selection patterns of deer in our system were strongly influenced by hydrocarbon development, with deer displaying both spatial displacement and alterations in temporal behavioral patterns relative to these features. The nature of these responses differed depending on disturbance type, time of day, and the distance from development. Our methodology, which accounted for the dynamic nature of deer behavior by allowing for the resource availability sample to change © 2015 The Authors. Global Change Biology Published by John Wiley & Sons Ltd., 21, 3961–3970 �H A B I T A T L O S S F R O M H Y D R O C A R B O N D E V E L O P M E N T 3967 (a) (b) Fig. 3 Posterior distribution of predicted selection as a function of distance to roads from resource selection function models fit to data during the (a) day and (b) night for 53 female mule deer in the Piceance Basin, Northwest Colorado. (a) (b) Fig. 4 Posterior distributions of population-level coefficients related to natural gas development for RSF models during the (a) day and (b) night for 53 female mule deer in the Piceance Basin, Northwest Colorado. Dashed line indicates 0 selection or avoidance of the habitat features. ‘Drill’ and ‘Prod’ refer to well pads where there was active drilling or not, respectively. The numbers following ‘Drill’ or ‘Prod’ represent the concentric buffer over which the number of well pads was calculated (e.g., ‘Drill 600’ is the number of well pads with active drilling between 400–600 m from the deer location). relative to current speed of the deer, and ensured the sample was conditioned by time and location, allowed us to identify the dynamic nature of the responses to development. Our results advance understanding of how animals perceive and adjust their behavior to mini- mize exposure to human disturbances, offering important insight for measures to mitigate the impacts of human land-use change associated with development. The approach we took can serve as a template for future work quantifying habitat take by land-use conversion. © 2015 The Authors. Global Change Biology Published by John Wiley & Sons Ltd., 21, 3961–3970 �3968 J . M . N O R T H R U P et al. The drilling stage of natural gas development elicited the strongest response by deer in our system. Deer strongly avoided areas within 600 m of well pads with active drilling at all times, and this avoidance persisted out to 1000 m at night (with the strongest responses within 800 m). During both day and night, the strength of avoidance of drilling well pads increased as distance decreased, with essentially no locations falling within 200 m of these pads. Sawyer et al. (2009) also documented a greater avoidance of active drilling than other energy development activities by mule deer, indicating that this activity is the predominate stressor during hydrocarbon development. Thus, measures aimed at mitigating impacts from drilling, such as seasonal drilling restrictions, sound and light barriers, and reductions in vehicle traffic, are likely to have the greatest benefit to deer. The other development infrastructure (i.e., roads and producing pads) altered deer behavior, but to a lesser extent. Deer avoided the areas closest to both of these development types to some degree, but the strength and scale of the responses varied between night and day, with stronger avoidance during the day when deer also selected areas with greater vegetative cover. It appears deer temporally modulate their behaviors so as to avoid these features during the most disturbing times of day (e.g., in relation to circadian traffic pulses). Dzialak et al. (2011) documented a similar pattern for elk in a natural gas field, with animals subject to disturbance selecting ‘security cover’ more strongly during the day. This behavior might be a common response by mobile wildlife to disturbance that has any type of temporal signature (e.g., roads; Northrup et al., 2012a). Understanding the spatial scale at which wildlife behavior is impacted by human disturbance is critical for developing effective mitigation strategies and quantifying the full footprint of development. As hydrocarbon development expands globally, developing frameworks for consistently assessing the spatial habitat take from this land-use change is a critical need. Our analysis design, examining selection or avoidance of concentric buffers around development, allowed us to identify the threshold distance where avoidance ceased and can serve as a template for future assessments. Deer displayed complete avoidance of areas within 200 m of well pad edges (approximately 2% of the critical winter range used by deer in our study). This distance should be considered the minimum at which indirect habitat loss occurs. However, avoidance was demonstrated to a distance of at least 800 m around drilling pads at night, and 600 m around producing pads during the day. These distances equate to greater than 20% of the critical winter range being impacted by producing pads (area within 600 m) and 2% by drilling pads (area within 600 m; the density of drilling pads is much lower in the study area), during the day, with 6% impacted by producing pads (area within 200 m) and 6% by drilling pads (area within 800 m) during the night. In addition, 28% of the critical winter range fell within 100 m of roads (the distance at which the relative probability of selection fell to half of the peak value during the day) and 15% fell within 50 m of roads (the distance at which the relative probability of selection fell to half of the peak value during the night). Although these values do not equate to complete habitat loss, they do indicate that more than half of the critical winter range was impacted by development during the day, and more than one quarter of the range was impacted during the night. The costs of this reduction (avoidance by deer) likely include the time lost during travel or from foraging in suboptimal areas during times of high human activity (Lima & Dill, 1990; Creel & Christianson, 2008), both of which can have impacts on nutritional condition and ultimately reproductive success (Houston et al., 2012). It is important to recognize that fitness costs of range avoidance likely are compounded during the winter when deer face a negative energy balance. We note that to date the deer population in this study area has been increasing (Anderson Jr., 2014), so this avoidance at the observed development intensity has not resulted in a population decline, although it is impossible to tell if the population would be increasing at a greater rate in the absence of development. The spatial scales of reduced use relative to specific types of infrastructure as defined in this study should be considered by managers when attempting to develop mitigation strategies. In a recent published assessment of mule deer response to natural gas development, Sawyer et al. (2006) found larger-scale displacement of deer from the area around development than those reported here. Although our results show similar general behavioral responses (i.e., alteration of habitat selection patterns), the scale of displacement was less. This likely relates to differences in the landscapes between the study areas, where the Piceance system has substantially greater topographic and vegetative diversity than the open, flat areas in the Pinedale area of Wyoming where Sawyer et al. (2006) conducted their work. We hypothesize that the structural diversity of the habitat and topography provide refuge areas for deer in our system at relatively close proximity to infrastructure that allows them to behaviorally mediate impacts. Such complexity in habitat structure can provide refuge for wildlife and should be considered and maintained by managers and developers when planning projects, through spacing of roads and pads to ensure sufficient areas outside the 800 m © 2015 The Authors. Global Change Biology Published by John Wiley & Sons Ltd., 21, 3961–3970 �H A B I T A T L O S S F R O M H Y D R O C A R B O N D E V E L O P M E N T 3969 buffers around drilling pads and 50–100 m buffers around roads. Oil and gas development is projected to continue to increase in the United States (USEIA, 2014), with major shale development likely to occur globally within the next decade. Quantifying the spatial extent of development related impacts to wildlife is critical for appropriately gauging the repercussions of this activity and identifying potential mitigation measures, which are critical for sustainable development practices (Northrup & Wittemyer, 2013). Importantly, drilling is temporary, as human activity declines once drilling is complete and wells begin producing (Sawyer et al., 2009). The temporary nature of this activity provides an opportunity to either avoid drilling during the winter months, or structure development in a manner that allows refuge habitat during the most acute periods of stress. Many drilling pads in an area, as might occur with rapid development, leads to large functional losses in habitat, apparently driving abandonment of areas by deer (e.g., Sawyer et al., 2006). Where development is conducted at lower densities, or in a manner that ensures that sufficient area is left undeveloped (i.e., refuge habitat is maintained), impacts are likely to be reduced. Even where drilling occurs in a manner that provides refuge, consideration of the spatial structure of the final footprint of roads, producing wells and facilities is critical to ensure adequate space for deer to structure their behaviors in a manner that mitigates negative impacts during the late stage production phase. Coupling spatial patterning of the permanent development footprint with approaches that reduce human activity at these areas, such as remote liquid gathering systems, will reduce the amount of disturbance (e.g., Sawyer et al., 2009) and subsequently any negative impacts. Contrasting results from the Piceance Basin and Pinedale provides insight to features that may allow deer to behaviorally mediate disturbance (although this should not be construed as eliminating all negative impacts; Lima & Dill, 1990), although the exact nature of these components in different systems requires more rigorous examination. Therefore, it is critical for future studies to identify thresholds to gain better understanding of the disturbance-habitat relationship and ensure sustainable development practices in areas with sensitive wildlife. dation, the Colorado Mule Deer Association, Safari Club International, Colorado Oil and Gas Conservation Commission, and the Colorado State Severance Tax. We thank L. Wolfe, C. Bishop, D. Finley, and D. Freddy (CPW) and numerous field technicians for project coordination and field assistance. We thank Quicksilver Air, Inc. for deer captures, and L. Gepfert (CPW) and Coulter Aviation, Inc. for fixed-wing aircraft support. We thank J. Tigner, and S. Downing for assistance with interpretation of development data. We thank M. Hooten, B. Gerber and P. Williams for analysis advice. N.T. Hobbs, M. Hooten, H. Johnson (CPW), K. Logan (CPW), A. Maki and 2 anonymous reviewers provided comments on an earlier draft that greatly improved the manuscript. This research utilized the CSU ISTeC Cray HPC system supported by NSF Grant CNS-0923386. References Anderson CR Jr. (2014) Population performance of the Piceance basin mule deer in response to natural gas resource extravtion and mitigation efforts to address human activity and habitat degredation. (ed. Federal Aid in Wildlife Restoration) Job Progress Report. Colorado Parks and Wildlife, Fort Collins, CO. Bartmann RM, Steinert SF (1981) Distribution and Movemetns of Mule Deer in the White River Drainage, Colorado. Colorado Dibision of Wildlife, Fort Collins, Colorado, USA. Bartmann RM, White GC, Carpenter LH (1992) Compensatory mortality in a Colorado Mule Deer Population. Wildlife Monographs, 121, 1–39. Berger J (2007) Fear, human shields and the redistribution of prey and predators in protected areas. Biology Letters, 3, 620–623. Boyce MS (2006) Scale for resource selection functions. Diversity and Distributions, 12, 269–276. Boyce MS, Mao JS, Merrill EH, Fortin D, Turner MG, Fryxell J, Turchin P (2003) Scale and heterogeneity in habitat selection by elk in Yellowstone National Park. Ecoscience, 10, 421–431. Boyce MS, Pitt J, Northrup JM, Morehouse AT, Knopff KH, Cristescu B, Stenhouse GB (2010) Temporal autocorrelation functions for movement rates from global positioning system radiotelemetry data. Philosophical Transactions of the Royal Society B-Biological Sciences, 365, 2213–2219. Creel S, Christianson D (2008) Relationships between direct predation and risk effects. Trends in Ecology & Evolution, 23, 194–201. Creel S, Winnie J, Maxwell B, Hamlin K, Creel M (2005) Elk habitat selection as an antipredator response to wolves. Ecology, 86, 3387–3397. D’eon RG, Delparte D (2005) Effects of radio-collar position and orientation on GPS radio-collar performance, and the implications of PDOP in data screening. Journal of Applied Ecology, 42, 383–388. Duchesne T, Fortin D, Courbin N (2010) Mixed conditional logistic regression for habitat selection studies. Journal of Animal Ecology, 79, 548–555. Dzialak MR, Harju SM, Osborn RG, Wondzell JJ, Hayden-Wing LD, Winstead JB, Webb SL (2011) Prioritizing conservation of ungulate calving resources in multiple-use landscapes. PLoS ONE, 6, e14597. Fieberg J, Matthiopoulos J, Hebblewhite M, Boyce MS, Frair JL (2010) Correlation and studies of habitat selection: problem, red herring or opportunity? Philosophical Transactions of the Royal Society B-Biological Sciences, 365, 2233–2244. Fortin D, Beyer HL, Boyce MS, Smith DW, Duchesne T, Mao JS (2005) Wolves influence elk movements: behavior shapes a trophic cascade in Yellowstone National Park. Ecology, 86, 1320–1330. Acknowledgements Frair JL, Nielsen SE, Merrill EH et al. (2004) Removing GPS collar bias in habitat selection studies. Journal of Applied Ecology, 41, 201–212. Fretwell SD, Lucas HL (1969) On territorial behavior and other factors influencing habitat distribution in birds. Acta Biotheoretica, 19, 16–36. Frid A, Dill LM (2002) Human-caused disturbance stimuli as a form of predation risk. Ecology and Society, 6, 11. Mule deer capture and monitoring was funded and/or supported by Colorado Parks and Wildlife (CPW), White River Field Office of Bureau of Land Management, ExxonMobil Production/XTO Energy, WPX Energy, Shell Exploration and Production, EnCana Corp., Marathon Oil Corp., Federal Aid in Wildlife Restoration (W-185-R), the Colorado Mule Deer Foun- Gelman A, Hill J (2007) Data Analysis Using Regression and Multilevel/Hierarchical Models. Cambridge University Press, Cambridge; New York. Gelman A, Rubin DB (1992) Inference from iterative simulation using multiple sequences. Statistical Science, 7, 457–511. Hebblewhite M, Merrill E (2008) Modelling wildlife-human relationships for social species with mixed-effects resource selection models. Journal of Applied Ecology, 45, 834–844. © 2015 The Authors. Global Change Biology Published by John Wiley & Sons Ltd., 21, 3961–3970 �3970 J . M . N O R T H R U P et al. Hebblewhite M, Percy M, Merrill EH (2007) Are all global positioning system collars created equal? Correcting habitat-induced bias using three brands in the Central Revelt D, Train K (1998) Mixed logit with repeated choices: households’ choices of appliance efficiency level. Review of Economics and Statistics, 80, 647–657. Canadian Rockies. Journal of Wildlife Management, 71, 2026–2033. Hooten MB, Hanks EM, Johnson DS, Alldredge MW (2013) Temporal variation and scale in movement-based resource selection functions. Statistical Methodology, 17, 82–98. Houston AI, Prosser E, Sans E (2012) The cost of disturbance: a waste of time and energy? Oikos, 121, 597–604. Johnson DH (1980) The comparison of usage and availability measurements for eval- Rice W (1987) Speciation via habitat specialization: the evolution of reproductive isolation as a correlated character. Evolutionary Ecology, 1, 301–314. Sawyer H, Nielson RM, Lindzey F, Mcdonald LL (2006) Winter habitat selection of mule deer before and during development of a natural gas field. Journal of Wildlife Management, 70, 396–403. Sawyer H, Kauffman MJ, Nielson RM (2009) Influence of well pad activity on winter habitat selection patterns of mule deer. Journal of Wildlife Management, 73, 1052– uating resource preference. Ecology, 61, 65–71. Johnson CJ, Boyce MS, Schwartz CC, Haroldson MA (2004) Modeling survival: application of the Andersen-Gill model to Yellowstone Grizzly Bears. Journal of Wildlife Management, 68, 966–978. Johnson DS, London JM, Lea MA, Durban JW (2008) Continuous-time correlated random walk model for animal telemetry data. Ecology, 89, 1208–1215. 1061. Shafer BA, Northrup JM, White K, Boyce MS, C^ ot�e SD, Coltman D (2012) Habitat selection predicts genetic relatedness in an alpine ungulate. Ecology, 93, 1317–1329. Stewart KM, Bowyer RT, Kie JG, Hurley MA (2010) Spatial distributions of mule deer and North American Elk: resource partitioning in a sage-steppe environment. American Midland Naturalist, 163, 400–412. Lele SR, Merrill EH, Keim J, Boyce MS (2013) Selection, use, choice and occupancy: clarifying concepts in resource selection studies. Journal of Animal Ecology, 82, 1183–1191. Lendrum PE, Anderson CR Jr, Monteith KL, Jenks JA, Bowyer RT (2013) Migrating mule deer: effects of anthropogenically altered landscapes. PLoS ONE, 8, e64548. Lewis JS, Rachlow JL, Garton EO, Vierling LA (2007) Effects of habitat on GPS collar performance: using data screening to reduce location error. Journal of Applied Ecol- Switzer PV (1997) Past reproductive success affects future habitat selection. Behavioral Ecology and Sociobiology, 40, 307–312. United States Energey Information Administration (2012) Annual Energy Review 2011. U.S. Energy Information Administartion Office of Energy Statistics U.S. Department of Energy, Washington, DC, USA. United States Energey Information Administration (2013) Technically Recoverable Shale oil and Shale gas Resources: An Assessment of 137 Shale Formations in 41 Countries Out- ogy, 44, 663–671. Lima SL, Dill LM (1990) Behavioral decisions made under the risk of predation: a review and prospectus. Canadian Journal of Zoology, 68, 619–640. Northrup JM, Wittemyer G (2013) Characterising the impacts of emerging energy development on wildlife, with an eye towards mitigation. Ecology Letters, 16, 112–125. Northrup JM, Pitt J, Muhly TB, Stenhouse GB, Musiani M, Boyce MS (2012a) Vehicle side the United States. United States Energey Information Administration, Washington, DC, USA. United States Energey Information Administration (2014) Annual Energy Outlook 2014. U.S. Energy Information Administration Offic of Integrated and International Energy Analysis U.S. Department of Energy, Washintgon, DC, USA. traffic shapes grizzly bear behaviour on a multiple-use landscape. Journal of Applied Ecology, 49, 1159–1167. Northrup JM, Stenhouse GB, Boyce MS (2012b) Agricultural lands as ecological traps for grizzly bears. Animal Conservation, 15, 369–377. Northrup JM, Hooten MB, Anderson CR Jr, Wittemyer G (2013) Practical guidance on characterizing availability in resource selection functions under a use-availability design. Ecology, 94, 1456–1463. Unsworth JW, Pac DF, White GC, Bartmann RM (1999) Mule deer survival in Colorado, Idaho, and Montana. Journal of Wildlife Management, 63, 315–326. Wittemyer G, Polansky L, Douglas-Hamilton I, Getz WM (2008) Disentangling the effects of forage, social rank, and risk on movement autocorrelation of elephants using Fourier and wavelet analyses. Proceedings of the National Academy of Sciences of the United States of America, 105, 19108–19113. Northrup JM, Anderson CR Jr, Wittemyer G (2014a) Effects of helicopter capture and handling on movement behavior of mule deer. The Journal of Wildlife Management, 78, 731–738. Northrup JM, Shafer BA, Anderson CR Jr, Coltman DW, Wittemyer G (2014b) Finescale genetic correlates to condition and migration in a wild cervid. Evolutionary Applications, 7, 937–948. Parker KL, Robbins CT, Hanley TA (1984) Energy expenditures for locomotion by mule deer and Elk. The Journal of Wildlife Management, 48, 474–488. Pierce BM, Bowyer RT, Bleich VC (2004) Habitat selection by mule deer: forage benefits or risk of predation? Journal of Wildlife Management, 68, 533–541. Pulliam HR, Danielson BJ (1991) Sources, sinks, and habitat selection: a landscape perspective on population dynamics. The American Naturalist, 137, S50–S66. R Core Team (2013) R: A Language and Environment for Statistical Computing. R Foun- Supporting Information Additional Supporting Information may be found in the online version of this article: Appendix S1. Detailed description of well classifications. Appendix S2. Overlapping buffers analysis, model structures, results of all fitted models and posterior predictive checks. dation for Statistical Computing, Vienna, Austria. http://www.R-project.org/. © 2015 The Authors. Global Change Biology Published by John Wiley & Sons Ltd., 21, 3961–3970 � https://cpw.cvlcollections.org/files/original/ee05b0020be6f5352d7c586ee7c95f7c.pdf 936901a5e4f2152a99738505604d386c PDF Text Text Appendix S1: Detailed description of well classifications The Colorado Oil and Gas Conservation Commission (COGCC) data provide the location of every well drilled in the state, the current status of each well, and the dates drilling began (spud date), the date that drilling reached its deepest depth (total depth date), and the date that the well was completed (the test date). We first attempted to categorize each well into one of 3 classes for each day during which we had mule deer GPS data. Wells were classified as drilling on every day between the spud date and the test date. Wells were classified as producing on days after the test date until the well was listed as abandoned. Wells were listed as abandoned from the time their status was listed as abandoned. In several cases the status of the well could not be directly categorized as one of these three statuses, and instead had a status of temporarily abandoned, injection well (wells where fluids are injected underground), shut in (wells that have been drilled but are not producing natural gas), or waiting on completion (wells that have been drilled but not completed). These instances were infrequent (<5% of wells), and typically it was impossible to determine the date of any activity associated with the well as listed dates were prior to the onset of the study or were missing. In light of these difficulties, we categorized all of these wells as producing as the number of wells was too few to include in models as a separate covariate and we assumed the activity associated with these pads was more similar to a producing pad than a drilling pad. In addition to the above statuses, the COGCC database includes a number of records for permitted locations that were never drilled. To ensure that these classifications were accurate we overlaid the well data with aerial imagery from the National Agriculture Imagery Program (NAIP) to assess if these records were indeed abandoned locations or if there was evidence of disturbance. � https://cpw.cvlcollections.org/files/original/437347b48751c96f19096e6f8a9805c9.pdf 59733630d95d0389a2019b13ab0c3d7c PDF Text Text Appendix S2: Overlapping buffers analysis, model structures, results of all fitted models and posterior predictive checks In addition to the single model structure discussed in the main text, where the number of well pads within concentric buffers was analyzed, we also fit a set of models including covariates for the number of well pads within overlapping buffers (Table S2.1). This analysis incorporates the entire response within each buffer distance and does not assess differences in responses at different distances within the same model. This analysis was done to examine if different inference was gained through including a single buffer in each model only. For this analysis we created 8 separate covariates representing active natural gas development. We first calculated the distance to the closest well pad classified as either drilling or producing (d_drill and d_prod respectively). We next calculated the number of well pads of each type falling within buffers of different sizes (400 m; drill_400_2 and prod_400_2, 600 m; drill_600_2 and prod_600_2, and 800 m; drill_800_2 and prod_800_2). These 8 variables (continuous distance and the four buffers) represent separate hypotheses for the scale and nature of mule deer responses to well pads. Model fitting proceeded as in the main text but the total number of iterations for which models were run and the number of iterations removed as burn-in varied by model (Table S2.1). We compared models using the Watanabe-Akaike Information Criteria (Watanabe 2010; see Hooten & Hobbs 2014 for a discussion of applications in ecology). �TABLES Table S2.1. Model numbers, covariates included in each model, Watanabe-Akaike Information Criteria (WAIC), total MCMC iterations, and burn-in for resource selection functions fit to GPS radio collar data from 53 adult female mule deer in the Piceance Basin winter range, Northwest Colorado, Jan 2008—Dec 2010. Model Covariates WAIC Total iterations Burn-in M1 d_edge + slope +elev +d_rds +d_rds2 +prod_800_2 +drill_800_2 + tree 218,163.50 200,000 50,000 M2 d_edge + slope +elev + d_rds + d_rds2 +prod_600_2 +drill_600_2 + tree 219,770.30 200,000 50,000 M3 d_edge + slope +elev + d_rds + d_rds2 +prod_400_2 +drill_400_2 + 400,000 100,000 1,800,000 700,000 200,000 50,000 Night tree M4 219,601.10 d_edge + slope +elev + d_rds + d_rds2 +d_prod + d_prod2 + d_drill + d_drill2 + tree 251,666.30 d_edge + slope +elev + d_rds + d_rds2 +prod_800_2 +drill_800_2 + 227,247.40 Day M1 �tree M2 d_edge + slope +elev + d_rds + d_rds2 +prod_600_2 +drill_600_2 + tree M3 d_edge + slope +elev + d_rds + d_rds2 +prod_400_2 +drill_400_2 + tree M4 226,333.10 50,000 400,000 50,000 1,800,000 700,000 225,421.00 d_edge + slope +elev + d_rds + d_rds2 +d_prod + d_prod2 + d_drill + d_drill2 + tree 200,000 239,439.00 �Table S2.2. Covariates, median coefficient (coeff.) values, and the proportion (prop.) of the posterior falling above or below 0 for resource selection function models, fit to separate night and day GPS radio collar data from 53 adult female mule deer in the Piceance Basin winter range, Northwest Colorado, Jan 2008—Dec 2010. Covaraite Median coeff. Prop. < 0 Prop. > 0 Night M1 d_edge 0.11 0.00 1.00 slope 0.18 0.00 1.00 elev 0.90 0.00 1.00 d_rds -0.36 1.00 0.00 d_rds2 -0.45 1.00 0.00 prod_800_2 0.07 0.14 0.86 drill_800_2 -0.36 0.99 0.01 tree -0.29 1.00 0.00 d_edge 0.11 0.00 1.00 slope 0.17 0.00 1.00 elev 0.85 0.00 1.00 d_rds -0.38 1.00 0.00 d_rds2 -0.47 1.00 0.00 prod_600_2 -0.05 0.77 0.23 M2 �drill_600_2 -0.58 1.00 0.00 tree -0.28 1.00 0.00 d_edge 0.11 0.00 1.00 slope 0.17 0.00 1.00 elev 0.81 0.00 1.00 d_rds -0.40 1.00 0.00 d_rds2 -0.47 1.00 0.00 prod_400_2 -0.21 0.99 0.01 drill_400_2 -0.78 1.00 0.00 tree -0.28 1.00 0.00 d_edge 0.11 0.00 1.00 slope 0.17 0.00 1.00 elev 1.07 0.00 1.00 d_rds -0.37 1.00 0.00 d_rds2 -0.45 1.00 0.00 d_prod -0.67 1.00 0.00 d_prod2 -0.63 1.00 0.00 d_drill -1.51 1.00 0.00 d_drill2 -1.37 1.00 0.00 M3 M4 �-0.29 1.00 0.00 -0.18 1.00 0.00 slope 0.06 0.00 1.00 elev 0.65 0.00 1.00 d_rds 0.19 0.00 1.00 d_rds2 -0.30 1.00 0.00 prod_800_2 -0.12 0.98 0.02 drill_800_2 -0.18 0.99 0.01 0.08 0.01 0.99 -0.18 1.00 0.00 slope 0.05 0.01 0.99 elev 0.63 0.00 1.00 d_rds 0.16 0.00 1.00 d_rds2 -0.32 1.00 0.00 prod_600_2 -0.23 1.00 0.00 drill_600_2 -0.50 1.00 0.00 0.08 0.01 0.99 tree Day M1 d_edge tree M2 d_edge tree �M3 -0.18 1.00 0.00 slope 0.05 0.01 0.99 elev 0.61 0.00 1.00 d_rds 0.16 0.00 1.00 d_rds2 -0.30 1.00 0.00 prod_400_2 -0.36 1.00 0.00 drill_400_2 -0.84 1.00 0.00 0.09 0.00 1.00 -0.17 1.00 0.00 slope 0.05 0.01 0.99 elev 0.73 0.00 1.00 d_rds 0.18 0.00 1.00 d_rds2 -0.27 1.00 0.00 d_prod -0.21 0.90 0.10 d_prod2 -0.55 1.00 0.00 d_drill 0.13 0.33 0.67 d_drill2 -0.88 1.00 0.00 0.08 0.01 0.99 d_edge tree M4 d_edge tree �Figure S2.1. Posterior predicted relative probability of selection as a function of distance to well pads actively producing natural gas in meters from resource selection function models fit to 53 adult female mule deer in the Piceance Basin, Colorado, USA. The left panel �is for the model from the day time and the right panel for the model from the night. Solid lines represent median posterior predicted values and dashed lines represent the 95% credible intervals. �Figure S2.2. Posterior predicted relative probability of selection as a function of distance to well pads with active drilling in meters from resource selection function models fit to 53 adult female mule deer in the Piceance Basin, Colorado, USA. The left panel is for the model from the day time and the right panel for the model from the night. Solid lines represent median posterior predicted values and dashed lines represent the 95% credible intervals. �Posterior predictive check To conduct a posterior predictive check we first calculated a posterior distribution of the probability of each available location associated with each used location being selected by the deer. We then calculated the proportion of available locations that were predicted to be selected at a higher probability than the used location to which they were associated. If the model was accurately representing the data generating process then the used location would be predicted to be selected at a higher probability than the majority of the available locations. This was the case in all assessed models. ��Figure S2.3. Results of posterior predictive check on the day time RSF model with concentric buffers fit to winter range GPS data from 53 female mule deer. X-axis represents the proportion of available locations that were predicted to be selected at a lower probability than the used locations. ��Figure S2.4. Results of posterior predictive check on the night time RSF model with concentric buffers fit to winter range GPS data from 53 female mule deer. X-axis represents the proportion of available locations that were predicted to be selected at a lower probability than the used locations. �LITERATURE CITED Hooten, M.B. & Hobbs, N.T. (2014) A Guide to Bayesian Model Selection for Ecologists. Ecological Monographs. Watanabe, S. (2010) Asymptotic Equivalence of Bayes Cross Validation and Widely Applicable Information Criterion in Singular Learning Theory. J. Mach. Learn. Res., 11, 3571-3594. � 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 Quantifying spatial habitat loss from hydrocarbon development through assessing habitat selection patterns of mule deer Rights Information about rights held in and over the resource <a href="http://rightsstatements.org/vocab/InC-NC/1.0/">In Copyright - Non-Commercial Use Permitted</a> <a href="https://creativecommons.org/licenses/by-nc/4.0/" target="_blank" rel="noreferrer noopener">Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)</a> Date Created Date of creation of the resource. 2015-08-12 Subject The topic of the resource Animal movement Bayesian hierarchical model Energy development Habitat selection Movement ecology Mule deer Natural gas Resource selection function Description An account of the resource Extraction of oil and natural gas (hydrocarbons) from shale is increasing rapidly in North America, with documented impacts to native species and ecosystems. With shale oil and gas resources on nearly every continent, this development is set to become a major driver of global land-use change. It is increasingly critical to quantify spatial habitat loss driven by this development to implement effective mitigation strategies and develop habitat offsets. Habitat selection is a fundamental ecological process, influencing both individual fitness and population-level distribution on the landscape. Examinations of habitat selection provide a natural means for understanding spatial impacts. We examined the impact of natural gas development on habitat selection patterns of mule deer on their winter range in Colorado. We fit resource selection functions in a Bayesian hierarchical framework, with habitat availability defined using a movement-based modeling approach. Energy development drove considerable alterations to deer habitat selection patterns, with the most substantial impacts manifested as avoidance of well pads with active drilling to a distance of at least 800 m. Deer displayed more nuanced responses to other infrastructure, avoiding pads with active production and roads to a greater degree during the day than night. In aggregate, these responses equate to alteration of behavior by human development in over 50% of the critical winter range in our study area during the day and over 25% at night. Compared to other regions, the topographic and vegetative diversity in the study area appear to provide refugia that allow deer to behaviorally mediate some of the impacts of development. This study, and the methods we employed, provides a template for quantifying spatial take by industrial activities in natural areas and the results offer guidance for policy makers, mangers, and industry when attempting to mitigate habitat loss due to energy development. Creator An entity primarily responsible for making the resource Northrup, Joseph M. Anderson Jr, Charles R. Wittemyer, George Language A language of the resource English Is Part Of A related resource in which the described resource is physically or logically included. Global Change Biology Extent The size or duration of the resource. 10 pages Bibliographic Citation A bibliographic reference for the resource. Recommended practice is to include sufficient bibliographic detail to identify the resource as unambiguously as possible. Northrup, J. M., C. R. Anderson Jr., and G. Wittemyer. 2015. Quantifying spatial habitat loss from hydrocarbon development through assessing habitat selection patterns of mule deer. Global Change Biology 21:3961–3970. <a href="https://doi.org/10.1111/gcb.13037" target="_blank" rel="noreferrer noopener">https://doi.org/10.1111/gcb.13037</a> Format The file format, physical medium, or dimensions of the resource application/pdf Type The nature or genre of the resource Article