SEARCH

SEARCH BY CITATION

Keywords:

  • Bayesian hierarchical modeling;
  • Cervus elaphus;
  • discrete choice modeling;
  • energy development;
  • human activity;
  • natural gas field;
  • resource selection function;
  • rocky mountain elk

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Acknowledgments
  9. References
  10. Supporting Information

Managing wildlife populations in areas subject to human activity is an increasingly prominent challenge. Estimating resource selection functions for species of conservation concern and developing spatially explicit maps predicting animal use across landscapes is a powerful tool for minimizing negative impacts and enhancing positive influences of human activities. However, if animals modify their selection of resources in response to humans, application of spatially explicit conservation tools based on resource selection among animals exposed to high levels of human activity risks uncertainty in the performance of such tools. This could lead to ineffective conservation action and wasted conservation dollars. To evaluate the magnitude of differences between spatial predictions based on animals exposed to different levels of human activity and develop reliable conservation tools, we used the treatment/control concept and Bayesian hierarchical discrete choice methods to model day time resource selection by female elk in a natural gas field and in areas adjacent to the gas field during winter. We found that female elk showed strong variation in resource selection patterns among years, tended to avoid roads and natural gas wells and consistently showed stronger selection for security cover, steeper slopes and greater distance to edge habitats within the gas field relative to outside of the gas field. Predictive probability of use maps based on ‘within gas field’ models classified probability of use differently in 10–55% of grid cells relative to outside of the gas field models depending on year. Conservation research and applications should consider that models based on resource selection data collected from animals subjected to human activity may not elucidate innate resource selection patterns and therefore may result in reduced effectiveness of management actions.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Acknowledgments
  9. References
  10. Supporting Information

Estimating the selection of resources by animals can be a powerful tool in balancing animal conservation with expanding human footprints. However, conservation actions may be misapplied when resource selection behavior is modeled using animals influenced by human activity. Apparent selection for some resources may be partially or largely a function of avoidance of human activity. While most ecologists would agree that human activity can modify resource selection by animals, a quantitative evaluation of the extent to which human activity can be a confounding factor in research intended to support conservation and management does not appear in the published literature. Application of resource selection modeling to quantify and predict occurrence patterns has become a dominant paradigm in research intended to support management and conservation (Johnson, Seip & Boyce, 2004). Failure to account for human modification of animal resource selection could significantly impact research results and the management actions derived from such research.

As human activity continues to expand across landscapes, conservation planning will increasingly rely on identifying preferred habitats within a matrix of varying levels of human use (Margules & Pressey, 2000). The use of rigorous before-after-control-impact studies to measure impacts of disturbance and identify preferred habitats is difficult because of pervasive human influence in most ecosystems. Therefore many resource selection studies explicitly expect that human activity may alter resource selection and incorporate disturbance features as either a primary (e.g. Dyer et al., 2002; Sawyer et al., 2006; Doherty et al., 2008) or secondary concern (e.g. Hebblewhite, Merrill & McDonald, 2005; Klar et al., 2008). While this approach can be very useful for quantifying some disturbance impacts (Sawyer et al., 2006), it may have limitations for identifying preferred habitats within disturbed areas. Conservation based on such models may have diminished effectiveness because these models are not based on preferred (and presumably optimal) selection of resources (Cooper & Millspaugh, 1999; Garshelis, 2000).

Our goal was to determine whether animals in fact do alter selection of environmental resources in the presence of intense human activity and, if so, to quantify how altered selection impacts predictive resource selection maps. We used a treatment-control design to investigate resource selection by free-ranging female elk Cervus elaphus within and around a natural gas field in Colorado, USA. We focused on winter resource selection because winter is often energetically demanding for elk populations due to limited forage availability and increased energy expenditure associated with thermoregulation (Christianson & Creel, 2007). We also focused on daytime resource selection because, in this study area, the primary types of human activity (ranching and maintenance of producing natural gas wells) were largely restricted to the day time and companion studies suggested that daytime activity associated with these land uses had a larger impact on elk resource selection than associated infrastructure (Dzialak et al., 2011). This study therefore addresses periods (day and winter) when elk would be most vulnerable to human-induced change. Our specific objectives were to: (1) model daytime resource selection by female elk on winter range within and adjacent to a natural gas field; (2) assess responses to roads and natural gas wells; (3) compare resource selection patterns among winters; (4) develop and validate maps predicting relative probability of female elk use; (5) assess differences in classification rate of predictive maps based on models of resource selection by elk within versus adjacent to the natural gas field.

Study area

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Acknowledgments
  9. References
  10. Supporting Information

The study area was located in the northern portion of the Raton Basin in south-central Colorado, USA (Fig. 1). Ranching (both within and outside of the gas field) and energy development (only within the gas field) were the two predominant land use practices. The core of the study area encompassed historic and ongoing coal-bed methane gas development, which contained 2421 well pads (1.77 well pads/km2) and 2933 wells (2.14 wells/km2) as of October 2008. Potential predators of elk on the study area included black bears Ursus americanus, mountain lions Puma concolor and coyotes Canis latrans. Topography ranges from rolling forested ridges and grassy valleys to steep slopes and cliffs. Vegetation included conifer forest, montane shrub and grassland. Dominant species included ponderosa pine Pinus ponderosa, one-seed juniper Juniperus monosperma, two-needle pinyon Pinus edulis, Gambel oak Quercus gambelii, which commonly forms shrub-thickets on southern aspects, antelope bitterbrush Purshia tridentata, skunkbush sumac Rhus trilobata and willow (Salix spp.) in riparian areas. We deployed seven weather stations located across the study area at elevations ranging from 1983 to 2841 m. January and July minimum and maximum temperatures were −26.3 and 12.4 °C, and 3.2 and 26.4 °C, respectively, at the highest elevation weather station. At the lowest elevation weather station, January and July minimum and maximum temperatures were −25.2 and 20.9 °C, and 7.1 and 33.8 °C, respectively.

image

Figure 1.  Study area for investigating day time resource selection by female elk on winter range in Raton Basin, CO, USA, 2006–2009.

Download figure to PowerPoint

We defined winter as the period between November 1 and March 31. The extent of the study area was governed by both elk movement and the extent of available aerial imagery. To account for between-year construction of new well pads and roads, the boundary differentiating inside versus outside of the gas field was re-defined annually using high-resolution annual aerial photography. We defined the gas field as anywhere within 1 km of a natural gas well; 1 km seemed a reasonable distance at which elk no longer responded to the presence of infrastructure and activity associated with the gas field. All elk locations >1 km from a natural gas well were considered outside of the gas field. Elk locations that occurred within 1 km of the aerial photography boundary and those outside of the boundary were excluded from analysis because of uncertainty regarding the amount and presence of disturbance features in areas outside of the aerial photographs. The geographic extent of aerial photos varied between years, so for our final predictive maps we constrained the boundary to the geographic area that was covered by aerial photos in all years. This predictive area was 1702.91 km2 with 64.68% of this area being within 1 km of a natural gas well.

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Acknowledgments
  9. References
  10. Supporting Information

Capture and handling

We captured female adult elk using a helicopter and either a dart-gun or net-gun (Leading Edge Aviation, LLC, Lewiston, ID, USA) annually during February and March 2006–2009. We captured elk from across the study area, both within and outside of the gas field. Animals captured using the net-gun were not immobilized; rather, animals were manually restrained with hobbles and fitted with blindfolds to reduce stress. Darted elk were anesthesized using either carfentanil or a synthetic narcotic thiafentanil (A-3080, Wildlife Pharmaceuticals Inc., Fort Collins, CO, USA). Sedated elk were also restrained with hobbles and fitted with blindfolds. Naltrexone was used as an antagonist to both carfentanil and thiafentanil. Elk were fitted with either a VHF or GPS collar (TGW-3590, Telonics Inc., Mesa, AZ, USA) and released at site of capture after reversal. Only data from GPS-collared elk were included in this analysis. GPS collars were programmed to record one location every 3 h (i.e. eight locations per day) beginning at 02:00 h, mountain standard time. We only used the 14:00 h GPS locations for this analysis. Location data were remotely downloaded from fixed-wing aircraft every 2–4 weeks at c. 1500–3000 feet above ground level. Animal capture and handling protocols were approved by the Colorado Division of Wildlife (Permit No. 06TR1083, 07TR1083, 08TR1083 and 09TR1083A001).

Conceptual organization

We used discrete choice modeling to allow our definition of resources available to an individual elk to vary spatiotemporally and be defined by the individual's movement or, under certain criteria, by the movement behavior of the sample population of female elk. We generated straight lines (i.e. steplengths) between successive 3 h locations as the basis for our definition of availability. The area of availability for a given point l1 was a circular buffer with a radius equal to the steplength between l0 and l1, centered on l0 (Fig. 2). We generated five random locations within the area of availability for each used location and considered these points as a single stratum. Thus, our discrete choice model compared a choice made by an individual female elk (i.e. used location) with five alternative choices that also were available temporally and spatially but were not chosen (i.e. random locations). Based on our definition of availability, we measured patch-level or third-order selection (Johnson, 1980). We only analyzed GPS locations from 14:00 h (thus representing resource selection between 11:00 and 14:00 h) to (1) allow for a reasonable assumption of independence among locations within each individual; (2) measure resource selection during the middle of the day to avoid confounding effects of crepuscular activity of elk and crepuscular variations in human activity. We set a minimum steplength equal to the median steplength across all elk for the time period 11:00–14:00 h within each respective winter to acknowledge there was a minimum area available even if an individual did not move during that time period on a given day. The minimum steplengths enforced for each winter were 67 m for the winter of 2006–2007, 101 m for 2007–2008 and 235 m for 2008–2009. We excluded from analysis used points whose previous steplength was in the largest 1% of all steplengths to remove long distance movements (i.e. movements not related to patch-level selection) and equipment failures (i.e. missing GPS locations before recording the used location). We used ArcGIS 9.2 (ESRI, Redlands, CA, USA) for all spatial analyses.

image

Figure 2.  Example of definition of availability associated with elk location l1, defined by steplength distance between l0 and l1, centered on l0. Open circles represent random locations.

Download figure to PowerPoint

We assigned GPS locations as within or outside of the gas field rather than assigning individual elk as either within or outside of the gas field. Some individual elk spent time both within and outside of gas field within each year. To evaluate the assumption that resource selection was influenced by location relative to the gas field rather than individual selection patterns, we investigated intra-individual differences in selection ratios for security cover as a function of location relative to the gas field. We calculated selection ratios as wji=exp(βil) for the model-derived β estimate for selection for security cover for each m elk where l equals 0 or 1 depending on whether the β estimates represents that individuals selection for security cover outside versus within the gas field, respectively (McDonald et al., 2006). Most of the transition elk wandered back and forth across the boundary delineating the gas field over the course of the winter.

Validation

A key issue when developing resource selection models lies in their predictive validity (Boyce et al., 2002; Wiens et al., 2008). We validated the outside of the gas field models solely to determine their suitability for applied use within the study area. We did not validate the within gas field models because a priori we assumed that if there were differences in resource selection outside versus within the gas field, resource selection outside of the gas field was a truer approximation of innate resource preferences of elk. To validate our models, we withheld ≥10% of randomly selected individual elk (and therefore c. 10% of locations) from each treatment group (within vs. outside the gas field) within each year. Locations from the withheld individuals were then pooled and plotted on the final three-category raw and smoothed multi-year maps. The number of independent locations that fell within each relative predicted probability of use category was compared with the number of locations expected if elk resource selection was random with respect to the developed RSF and predictive multi-year map.

Covariate data

We chose to model female elk resource selection as a function of four environmental variables (i.e. cover, slope, convexity and distance to habitat edge) both within and outside of the gas field. We incorporated two additional variables (i.e. distance to roads and well pads) to represent human activity within the gas field. See supporting information Appendix S1 for details on selection and calculation of covariates.

Data analysis

We conducted our resource selection modeling in a Bayesian hierarchical framework to: (1) model all covariates as random effects; (2) allow probabilistic inference on individual and population-level selection coefficients (Link et al., 2002); (3) allow population-level inference while explicitly estimating individual heterogeneity in the selection for resources (Thomas, Johnson & Griffith, 2006). We used WinBUGS software for analysis of statistical models (Lunn et al., 2000; see http://www.mrc-bsu.cam.ac.uk/bugs/). We present 95% credible intervals for parameter estimates, but caution against interpretation of confidence intervals as differentiating a parameter estimate as either statistically significant or not statistically significant (Yoccoz, 1991; Johnson, 1999). Rather, we interpret credible intervals as providing probabilistic information on the true value of the parameter, such that when credible intervals broadly overlap zero there is little confidence in the sign of the parameter estimate and when credible intervals barely or do not overlap zero there is high confidence in the sign, and potentially magnitude, of the parameter estimate. See supporting information Appendix S2 for model development, specification and WinBUGS code.

Predicted probability of use

Next, we used the population-level β estimates from the hierarchical discrete choice model to develop study area-wide predictive maps of relative probability of daytime use following our statistical model and standard methodology. See supporting information Appendix S3 for details on how we developed within-year and multi-year predictive maps to quantify misclassification rates of maps based on resource selection within the gas field.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Acknowledgments
  9. References
  10. Supporting Information

The main difference between areas within and outside of the gas field was the level of human-related activity. Average distance from a grid cell to the nearest road was three times greater outside of the gas field (480.6 m, SD 464.5) than within the gas field (160.1 m, SD 147.9) and the total density of roads was 2.2 times greater inside (2.4 km/km2) versus outside (1.1 km/km2) of the gas field. No natural gas well occurred outside of the gas field, as this was our definition of the gas field. Environmental variables generated from remote-sensed data were equivalent within and outside of the gas field. Average grid cell values for slope were comparable within (12.15 deg, SD 6.81) and outside (12.79 deg, sd 7.56) of the gas field, as were the proportion of cells classified as security cover (within: 0.80; outside: 0.83) and average convexity values (within: 8.82 m, sd 223.33; outside: 16.11, sd 230.51). Average distance to edge habitat was 42.2% higher (40.3 m further) outside than within the gas field, suggesting that habitat configuration (patches of cells with open or closed canopy) was more contiguous outside of the gas field than within. Because the extent of the study area was largely restricted to the center of the basin, elevation was similar within (2267.29 m, sd 159.39) and outside (2344.91 m, sd 308.12) of the gas field.

Sample sizes reflected relatively high numbers of individuals within years both within and outside of the gas field (Table 1). Total number of female elk fitted with a GPS collar for the winters of 2006–2007, 2007–2008 and 2008–2009 was 48, 72, 83, respectively. High inter-annual variability limited broad statements regarding elk selection for or against resources (Fig. 3). Coefficient estimates in selection for slope were generally small and had 95% credible intervals that broadly overlapped zero both within and outside of the gas field in the winter of 2006–2007 and outside of the gas field in 2007–2008 (Fig. 3a). Inside of the gas field in 2007–2008 elk exhibited moderate selection for steeper slopes with a 95% credible interval that barely overlapped zero. However in the winter of 2008–2009, female elk both within and outside of the gas field exhibited comparatively strong avoidance of steep slopes (i.e. strong selection for flat areas).

Table 1.  Sample sizes used in analysis of resource selection by female elk during daytime on winter range in Raton Basin, CO, USA, 2006–2009
YearWithin/outsideaModel-buildingValidation
No. of elkNo. of locationsAvg.bMin.Max.No. of elkNo. of locationsAvg.bMin.Max.
  • a

    Models within or outside of the natural gas field.

  • b

    Average number of locations per elk.

2006Within3812033211175138282232
Outside271062391144419649298
2007Within632861451152721330351
Outside42114127195512826842
2008Within69362253113883854812112
Outside558891617178011234
image

Figure 3.  Population-level coefficient estimates (log odds) of selection for environmental variables by female elk on winter range in Raton basin, CO, USA. Open bars are outside of gas field results, grey bars are within gas field results, and error bars are 95% credible intervals. Positive and negative values indicate selection for a given resource greater or less than the availability of that resource, respectively. Subpanels a, b, c, and d represent selection for the variables slope, security cover, distance to cover/open edge, and convexity, respectively.

Download figure to PowerPoint

Female elk both within and outside of the gas field selected for security cover (Fig. 3b) and avoided edges (Fig. 3c) during midday in the first two winters of the study with strong divergence in the third winter, when elk within the gas field exhibited stronger selection for security cover but were close to edges while elk outside of the gas field exhibited stronger selection for open areas and were likewise close to edges (Fig. 3b and c). Selection for convexity was nonexistent in all winters both within and outside of the gas field, with population-level coefficient estimates close to zero and 95% credible intervals that broadly overlapped zero (Fig. 3d). Elk within the gas field avoided roads and natural gas wells more than expected compared with random locations (Fig. 4). The magnitude of their estimated avoidance of roads decreased over the 3-year study period whereas no temporal trend was evident in their avoidance of wells (Fig. 4).

image

Figure 4.  Population-level coefficient estimates (log odds) of selection for distance to the nearest road or natural gas well by female elk on winter range in Raton Basin, CO, USA, 2006–2009. Results are for elk within the gas field; models for elk locations outside of the gas field did not include distance to road or well as covariates. Dark gray bars represent winter 2006–2007, light gray bars represent winter 2007–2008 and open bars represent winter 2008–2009. Error bars are 95% credible intervals. Positive values indicate selection for greater values of distance to nearest feature (i.e. avoidance of roads or natural gas wells).

Download figure to PowerPoint

Within each winter there was a consistent difference in the strength of selection for or against environmental variables within versus outside of the gas field (Fig. 3). Elk within the gas field more strongly selected for steeper areas, security cover and distance to edge habitat than elk outside of the gas field during the first two winters. During the third winter, selection for flat areas and edge habitat was weaker in elk within the gas field than in elk outside of the gas field (i.e. they still tended to use steeper slopes within compared to outside of the gas field).

Individual heterogeneity in selection for resources (standard deviation of individuals from the population mean), after accounting for the strength of population-level selection (population-level CV), was larger for elk outside of the gas field than within the gas field in 7 of 12 variable-year comparisons (Table 2). For the three environmental variables that appeared meaningful at the population level (i.e. slope, cover and distance to edge), individual heterogeneity was stronger in elk outside of the gas field in 7 of the 9 variable-year comparisons.

Table 2.  Individual heterogeneity in selection for resources across years, and within and outside of the natural gas field for female elk on winter range during daytime in Raton Basin, CO, USA, 2006–2009
ParameterGas fieldStandard deviationaCVbPer cent differencec
200620072008200620072008200620072008
  • a

    Parameter estimates for standard deviation of individual elk from the population mean parameter estimate (see Fig. 3); for brevity, 95% credible intervals are not presented.

  • b

    See ‘Methods’ for calculation of CV.

  • c

    Per cent difference in CVinside versus CVoutside the natural gas field [per cent difference=(CVinside/CVoutside−1) × 100].

  • d

    Dist road and dist well were not calculated for elk locations outside of the natural gas field.

SlopeInside0.100.080.129.484.251.1219.17−98.40−22.23
Outside0.120.110.177.95266.071.44   
CoverInside0.380.520.780.421.450.49−67.47−95.32−48.14
Outside0.480.501.021.3130.980.95   
Dist edgeInside0.140.110.212.641.991.70−18.92−69.095.96
Outside0.150.140.213.256.431.60   
ConvexityInside0.080.060.0631.81166.04261.5011.15103.62814.03
Outside0.090.070.0628.6281.5428.61   
Dist roadInside0.450.250.220.740.96207.69d
Dist wellInside0.670.351.7111.411.478.40

Several individual elk used areas both within and outside of the gas field within each winter. During the winters of 2006–2007, 2007–2008 and 2008–2009, 17, 33 and 41 individual elk, respectively, spent time both within and outside of the gas field. Within-individual selection ratios for security cover increased when the individual was within versus outside of the gas field for 100, 78.8 and 100% of individuals during the first, second and third winter, respectively. Average change in selection ratio for security cover (as a multiplicative factor, ± 95% CI) for the first, second and third winters was 1.81 (1.65–1.96), 1.35 (1.19–1.51) and 18.30 (14.26–22.34), respectively. Therefore, for example, during the winter of 2006–2007, on average an elk that spent time both within and outside of the gas field increased its selection for security cover by 81% compared with selection for security cover when outside of the gas field.

Within-year predictive RSF maps differed in classification of relative probability of use for some grid cells depending on whether the RSF maps were based on within or outside of gas field models (Fig. 5). Strongest divergence in the predictive RSF maps was in the winter of 2008–2009 (54.9%) with milder misclassification rates in the two previous winters (Table 3). The combined multi-year predictive RSF map based on elk resource selection outside of the gas field (Fig. 5) performed well under validation. Winter daytime locations from independent elk occurred in predicted low use areas less than expected at random, equal to random for moderate use areas, and more often than random in predicted high use areas (Fig. 6).

image

Figure 5.  Predictive RSF maps of resource selection by female elk on winter range within and outside of a natural gas field in Raton Basin, CO, USA, 2006–2009. Predictive maps for each winter contain five bins of predicted probability of use; combined multi-winter maps contain three bins of predicted relative probability of use. Bins contain a roughly equal number of grid cells. See ‘Results’ for rates of disagreement of predicted probability of use for within versus outside of gas field maps for each winter.

Download figure to PowerPoint

Table 3.  Percentage of grid cells misclassified by within gas field models compared with outside of gas field models of resource selection by female elk during daytime on winter range in Raton Basin, CO, USA, 2006–2009
 Winter
2006–20072007–20082008–2009All three wintersa
  • a

    Differences in grid cell classification based on multi-year three-bin predictive maps.

  • Totals may not equal 100% due to rounding error.

Correctly classified90.0088.0645.1076.46
Total number misclassified10.0111.9454.9023.53
 Misclassified low5.634.1620.9215.35
 Misclassified high4.387.7833.988.18
image

Figure 6.  Per cent difference in number of observed validation GPS locations compared with expected predicted relative probability of use categories by female elk on daytime winter range in Raton Basin, CO, USA 2006–2009. White, gray and black bars represent differences in the number of observed versus expected validation points that fell within low, medium and high predicted probability of use areas, respectively. Fewer locations were found in predicted low use areas and more locations occurred in predicted high use areas than expected at random, indicating the predictive map performed well.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Acknowledgments
  9. References
  10. Supporting Information

Despite considerable variation in resource selection patterns among years, elk consistently modified resource selection within the gas field. Comparing selection ratios within individuals, depending on whether that individual was within or outside of the gas field, provides a strong test of the hypothesis that elk-modified resource selection within the gas field because it is robust to individual resource selection patterns. We found large shifts in resource selection patterns within individuals, whereby individual elk increased use of security cover by a factor of up to 18.3 on average when inside of the gas field. This is consistent with the findings of van Dyke & Klein (1996) and Kuck, Hompland, & Merrill (1985) who observed increased selection for security cover in the presence of energy development, and with the general notion that ungulates modify their behavior in response to perceived risk (Frid & Dill, 2002; Winnie et al., 2006).

As part of the general trend of modified resource selection inside of the gas field relative to outside, we found that individual heterogeneity in selection for landscape features was lower within versus outside of the gas field. This means that resource selection by elk within the gas field was constrained. This would be expected if areas of intense human activity are perceived as being risky and elk respond to landscape features in ways that are simplified relative to how they otherwise would respond in less-risky areas (Winnie et al., 2006). A change in resource selection in the presence of human activity is neither surprising nor necessarily cause for management intervention. Such changes may have no impact on population size or demographics (Gill, Norris & Sutherland, 2001a), or alternatively, changes in resource selection may have cascading impacts for population demography and even ecosystem processes (Fortin et al., 2005; Creel & Christianson, 2008). Regardless of impacts on population demography, a potential problem in the spatial application of RSF models for conservation planning arises if the underlying resource selection behavior is influenced by human activity.

In our analysis of elk, a highly mobile habitat generalist, resource selection was modified by human activity. We assumed that elk resource selection in disturbed areas represents less optimal selection than does resource selection in less-disturbed areas, and thus that measured selection of environmental resources by disturbed elk incorrectly identifies preferred resources. By following the path of translating RSFs into spatially explicit maps of predicted use (e.g. Johnson, Seip & Boyce, 2004), we would have misclassified probability of use for 10–55% of grid cells relative to spatial predictions based on how elk respond to features in the absence of intense human activity. If conservation plans within human-modified landscapes include identifying areas that are intended to buffer against human activity and promote population stability, information on resource selection patterns in existing areas of low human activity would be expected to offer guidance in prioritizing the establishment of new conservation areas in modified landscapes. Establishing conservation areas based on resource selection that reflects responses to human activity risks uncertainty in the effectiveness of such areas once human disturbance pressure is released allowing animals to respond to landscape features as they otherwise would have (Winnie et al., 2006). Further, human-influenced resource selection may have limited utility for informing development of other landscapes with relatively little human activity (Margules & Pressey, 2000). We acknowledge that the degree to which altered resource selection in human-modified landscapes poses problems for conservation planning is likely partially a function of the ecology and status of the population of interest. When human activity is below levels at which vital population rates begin to change, and when individuals are able to adapt to some level of human-modification of the landscape (sensuGill, Norris & Sutherland, 2001b), misidentifying probability of use could be of little consequence. However, for species of conservation concern or for high priority management species such as many ungulates and threatened or sensitive species, a reliable assessment of the spatial structure of occurrence probability is critical for efficient allocation of conservation dollars. In fact, the observation of nonrandom selection of resources in any context implicitly suggests some benefit to the animal, that human-induced modification of resource selection would lead to a reduction in benefits or an increase in costs (e.g. less forage or increased predation risk; Creel & Christianson, 2008) and, ultimately, that the consequences of misclassifying probability of use could be substantial.

One obvious solution to this issue would be modeling resource selection in adjacent areas subjected to substantially lower levels of human activity, and thus to provide resource selection functions closer to behavior less influenced by human activity while retaining the effect of functional responses to local environmental conditions (Mysterud & Ims, 1998). Conservation actions in areas subject to human activity will likely be more effective if they are based on resource selection patterns of animals not subjected to human activity. However, such proximal undisturbed places may not exist. In this case interpretation of human-influenced resource selection models should explicitly consider the ecology of the organism when discussing the importance of specific environmental resources. Selection for some resources may be inflexible in the presence of human activity (e.g. water), whereas other resources may be moderately exchangeable (e.g. security cover vs. open areas). These results suggest that explicit consideration of how human activity affects resource selection is important for maximizing the positive effect of applied management and conservation actions.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Acknowledgments
  9. References
  10. Supporting Information

Pioneer Natural Resources supported this study. J. Wondzell provided invaluable assistance in the field and constructive comments on the manuscript. S.L. Webb, C. Hedley, R. Schindler, M.B. Rice, V.J. Dreitz, C.R. Anderson, P.M. Lukacs and C.J. Bishop, and two anonymous reviewers provided insightful comments on this manuscript. This study would not have been possible without the cooperation and support of many private landowners.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Acknowledgments
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study area
  5. Methods
  6. Results
  7. Discussion
  8. Acknowledgments
  9. References
  10. Supporting Information

Appendix S1. Selection and calculation of covariates

Appendix S2. Model development and specification.

Appendix S3. Development of predictive maps.

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer-reviewed and may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.

FilenameFormatSizeDescription
acv_456_sm_appendix-1.pdf10KSupporting info item
acv_456_sm_appendix-2.pdf92KSupporting info item
acv_456_sm_appendix-3.pdf82KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.