Figures
Abstract
Development of hydrocarbon resources across northwest Canada has spurred economic prosperity and generated concerns over impacts to biodiversity. To balance these interests, numerous jurisdictions have adopted management thresholds that allow for limited energy development but minimize undesirable impacts to wildlife. Used for exploration, seismic lines are the most abundant linear feature in the boreal forest and exist at a variety of widths and recovery states. We used American marten (Martes americana) as a model species to measure how line attributes influence species’ response to seismic lines, and asked whether responses to individual lines trigger population impacts. Marten response to seismic lines was strongly influenced by line width and recovery state. Compared to forest interiors, marten used open seismic lines ≥ 3 m wide less often, but used open lines ≤ 2 m wide and partially recovered lines ≥ 6 m wide similarly. Marten response to individual line types appeared to trigger population impacts. The probability of occurrence at the home range scale declined with increasing seismic line density, and the inclusion of behavioral response to line density calculations improved model fit. In our top performing model, we excluded seismic lines ≤ 2 m from our calculation of line density, and the probability of occurrence declined > 80% between home ranges with the lowest and highest line densities. Models that excluded seismic lines did not strongly explain occurrence. We show how wildlife-derived metrics can inform regulatory guidelines to increase the likelihood those guidelines meet intended management objectives. With respect to marten, not all seismic lines constitute disturbances, but avoidance of certain line types scales to population impacts. This approach provides the ecological context required to understand cause and effect relationships among socio-economic and ecological conservation goals.
Citation: Tigner J, Bayne EM, Boutin S (2015) American Marten Respond to Seismic Lines in Northern Canada at Two Spatial Scales. PLoS ONE 10(3): e0118720. https://doi.org/10.1371/journal.pone.0118720
Academic Editor: Mónica V. Cunha, INIAV, I.P.- National Institute of Agriculture and Veterinary Research, PORTUGAL
Received: May 1, 2014; Accepted: January 6, 2015; Published: March 13, 2015
Copyright: © 2015 Tigner et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited
Data Availability: All relevant data are within the paper and its Supporting Information files.
Funding: Funding from Environmental Studies Research Fund, National Science and Engineering Research Council (NSERC) Integrated Landscape Management Chair, Alberta Sports, Recreation, Parks, and Wildlife Foundation DIP Program, Canadian Circumpolar Institute C/Bar Program, Canadian Association of Geophysical Contractors, British Columbia Ministry of Environment, Government of the Northwest Territories, IHS Energy, Ducks Unlimited Canada, Canadian Wildlife Service, Aboriginal Affairs and Northern Development Canada, Acho Dene Koe and the Dehcho First Nations, Apache Canada, Canadian Natural, EOG Canada, Nexen, Nabors International, and the University of Alberta. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors received funding and in-kind support for this research from a number of sources across a broad spectrum of stakeholder groups including commercial interests, government agencies, and First Nations groups. Commercial sources of funding and in-kind support came from professional societies, oil companies, and oil field service providers including Canadian Association of Geophysical Contractors, IHS Energy, Apache Canada, Canadian Natural, EOG Canada, Nexen, and Nabors International. Neither this commercial support, nor support from other non-commercial interests influenced the design, execution, or reporting of this study. This support does not alter the authors’ adherence to PLOS ONE policies on sharing data and materials. No funders had any role in this research outside of providing financial or in-kind support. Funders did not influence study design, data collection and analysis, decision to publish, or preparation of the manuscript in any way.
Introduction
The exploration and production of hydrocarbon resources (energy development) has increased dramatically worldwide in recent decades spurring both economic opportunities and conservation challenges. Balancing opportunities and challenges is often stymied because an understanding of the impacts to wildlife and the efficacy of mitigation options often lags behind the pace of development [1]. Across northwest boreal Canada, energy development involves the creation of linear disturbances including seismic lines, roads, and pipelines [2]. As such, a threshold-based management approach, aimed at limiting the density of linear features, has been widely suggested across the region [3, 4]. Few data exist to inform threshold calculation or density targets (but see [5]) resulting in targets ranging from 0.6 km of linear disturbance per square kilometer (km/km2) to 2.4 km/km2, in some cases calculated using all linear features ever created and in others using only certain features [6–8]. Critics contend that inconsistent density targets and a failure to account for differences in linear features confound threshold calculations and ultimately, management objectives [9].
Particularly contentious are seismic lines. Used for resource exploration, seismic lines are the narrowest, but most numerous linear disturbances associated with the energy sector in boreal Canada [3, 10]. The impact of seismic lines on wildlife, including changes to habitat use [11], home range configuration [5], and movement [12] have been clearly documented, however a detailed understanding of impacts is lacking because most studies categorize all seismic lines into a single disturbance class [11, 13, 14] (but see [15, 16]). Prior to the mid-1990s seismic lines were constructed between 6 and 10 m wide (conventional lines) [17]. In response to numerous concerns, the energy sector began constructing narrower and meandering lines collectively called low-impact seismic (LIS); currently LIS lines range in width from ≤ 2 to 5.5 m. Whether these narrower seismic lines actually mitigate impacts to wildlife is poorly understood [18] (but see [12, 15, 19]).
Until recently, the energy sector invested little effort in promoting the active recovering of vegetation along lines because seismic lines were considered temporary disturbances that would recover naturally after seismic data were acquired. However, on lines, recovery is highly variable [20]: some lines recover to heavy shrub or sapling growth over time [21, 22], but others remain in open or semi-open states for decades [17, 23, 24]. Anecdotal evidence suggests that wildlife respond to older or reclaimed seismic lines differently than open ones [25, 26], but limited research suggests the response is weak [19, 27] and poorly understood for most species.
Behavioral responses of wildlife to anthropogenic habitat disturbances (i.e., whether and how disturbances are used by individuals) play an important role in the development of management actions [28, 29]. A number of studies on wildlife response to energy development have interpreted any changes in organism behavior as inherently negative (e.g. [11, 30, 31]), and such interpretations have played a crucial role in the development of linear feature thresholds in boreal Canada. Whether statistically significant changes in behavior are indicative of biologically significant changes that impact the fitness of individuals [32], or translate into population level effects is poorly understood for most species [28, 33]. Without a clear link between behavioral and population responses, a threshold system may not achieve its intended management objectives [34], ultimately misappropriating mitigation efforts [1].
The American marten (Martes americana) is a medium-sized mustelid that inhabits conifer dominated forest types primarily in northern North America. Across boreal Canada marten are ubiquitous, and are a valued furbearer for many northern communities [35]. Well studied across their range, marten habitat relationships and responses to habitat disturbances triggered by forestry are well understood at multiple spatial and to a lesser degree temporal scales. At local spatial scales marten select habitat patches that provide overhead cover [36, 37] and lateral complexity [38–40]. In contrast, marten avoid naturally open stands [41, 42], and stands opened or simplified by timber harvest [36, 39, 42, 43]. Marten also use areas close to roads less than areas farther away [44]. Given sufficient recovery of overhead and structural attributes, marten will re-use regenerated habitats [45–49].
Avoidance of disturbances at fine scales can translate to changes in habitat use at broader scales, and ultimately to changes in population demographics. Compared to undisturbed landscapes, marten move longer distances [50–52] and more tortuous routes in disturbed landscapes to avoid disturbances [53]. To compensate for fine scale habitat loss, marten adopt larger home ranges in disturbed landscapes [43, 50, 54] and several studies show precipitous declines in marten use, occurrence, and occupancy of landscapes once 25%- 49% of the total area has been disturbed [39, 55–57]. Such changes are linked to lower survival [51, 52, 58, 59], reduced reproduction [59, 60], diminished body condition [52, 61] and population declines [54, 57, 59, 62] in areas with more disturbances.
To date, marten response to oil and gas development is unstudied. We measured the response of American marten to seismic lines at 2 spatial scales: the individual line and the home range. First, we measured marten response to individual seismic line types as the odds of a marten using a seismic line relative to the odds of using the forest interior. We compared use at a series of seismic line treatments that varied by width and vegetation recovery state. We expected that lines perceived by marten as disturbances would be used less frequently than forest interior, but that lines not perceived as disturbances would be used comparably to forest interior. Second, we measured marten response to seismic line density as the probability of occurrence at the home range scale across a continuum of line density. We expected that if there was a link between behavioral and population responses to seismic lines, marten avoidance of individual line types would trigger a decline in home range occurrence relative to increasing line density. Lastly, we compared the efficiency of an all-seismic line density metric to explain marten occurrence at the home range scale with other common metrics of linear feature density used in threshold calculations. We expected model fit to improve if other metrics better explained variation in marten occurrence at the home range scale.
Materials and Methods
Study area
Our study area encompassed 200,000 km2 of northern boreal forest in northwest Alberta (AB), northeast British Columbia (BC), and southwest Northwest Territories (NWT) between 61°48′ and 58°48′ latitude and 122°41′ and 117°39′ longitude in 2008 and 2009 (Fig. 1). Energy development was the primary industrial land—use in the study area and was widespread south of the 60th parallel (in AB and BC), but uncommon to the north (Fig. 1). Forestry occurs in parts of AB and BC, but we excluded these areas from our study to avoid confounding effects. Trapping for marten occurs throughout the area, but mainly where there is little or no energy development.
The study area in northeast Alberta, northwest British Columbia, and southwest Northwest Territories (A). Territorial and Provincial jurisdictions in western Canada are subject to divergent energy policies as evident from the extent of the disturbance footprint (shown in light grey, but excluding seismic lines for clarity) (B).
Forest types in this region were broadly classified into upland or lowland types [63]. Lowland forests were typified by open black spruce (Picea mariana) dominated muskeg and peatlands with shallow surface water, organic soils, and sparse shrub cover. By contrast, upland forests were dry and characterized by mixed stands of white spruce (Picea glauca) and aspen (Populus tremuoidis and balsamifera) (but sometimes pure conifer or deciduous) with dense shrub cover.
Site Selection
In a Geographic Information System (GIS; ArcGIS 9.3, ESRI, Redlands, California), we reclassified Earth Observation for Sustainable Development (EOSD) [64] and Ducks Unlimited Canada Earth Cover Classification data into upland and lowland forests, and several non-forested land covers. We compiled the energy sector disturbance footprint from government, industry, and First Nations data sources and generated individual layers for seismic lines, roads, pipelines, well-pads, and oil field facilities accurate to the time of field sampling.
To isolate marten response to seismic lines and line density, we controlled for dominant forest type a priori using our GIS layers. Within a 5 km2 moving window, (average marten home range in our region [46, 65, 66]), we stratified the study area into upland and lowland forest types (≥50% of habitat within 5 km2), and we measured seismic line density as a continuous variable. We generated a set of 250 random candidate sites (125 each in upland and lowland forest types) across a continuum of seismic line density ranging from <0.1 km/km2 to >26 km/km2 within each forest type. Candidate sites were a minimum spacing of 5 km from each other or a human settlement, but within 15 km of road, ATV trail, or navigable water for access. In the field, if a candidate site was inaccessible or incorrectly categorized, we chose the next closest suitable site.
Camera Trap Protocol
At each sampled site, we used a cluster of 6 remote cameras (Bushnell Scout Model 119833, Bushnell Corp. Overland Park, KS; and Reconyx RapidFire P85, Reconyx, Inc., Holmen, WI) to measure marten response to seismic line type and to seismic line density simultaneously (Government of the Northwest Territories Wildlife Research Permit No. WL-005752). All field methods were approved by the University of Alberta Animal Care and Use Committee under Protocol No. 476705. To measure marten use of seismic line type, we set up pairs of cameras with 1 on and 1 off of a seismic line (line and interior cameras, respectively). To measure marten occurrence at the home range scale we grouped 3 camera pairs together along a single seismic line to form a camera cluster. We spaced line and interior cameras by 450 m, and we spaced camera pairs within a cluster by 900 m. All cameras were left to sample for 10 trap nights (i.e. 240 hours). We considered individual cameras as the sample unit when evaluating use of seismic line types and camera clusters as the sample unit when evaluating the probability of occurrence.
We programmed all cameras to collect data 24 hours/day and to capture a photograph when a combination of motion and a change in heat (infrared energy) was detected (both camera models employed a passive-infrared sensor). We baited all cameras with 150 g of canned dog food and 50 g of tinned sardines once at camera set-up. To standardize the zone of detection for camera locations, we set baits 3–4 m from the camera at 0.5–1 m from the ground. Reported trigger speed for both camera models was < 1 second.
In 2008 we sampled using Bushnell cameras between May and September and in 2009 we used Bushnell and Reconyx cameras between May and October. Survey timing was selected to coincide with when marten are most likely to occupy established home ranges in our study area. During this time detections are more likely representative of residents’ behaviors (i.e., after mating, but before natal dispersal when marten incur longer and transient movements that may not be representative of typical habitat use or selection [52]). Camera models were not mixed within a cluster, but both models were used in upland and lowland forest types across the study area and sampled across the continuum of line density.
Data Collection and Analysis
For our purposes use and occurrence were defined as a single photographic detection at the line and cluster level, respectively.
Use of Seismic Line Type
We evaluated marten use of seismic line types as the odds of an observation on a line relative to an observation in the forest interior (marten photographed at a camera = 1 vs. not photographed = 0). We pointed line cameras along a seismic line in 1 of the following treatments: 1) line ≥ 6 m wide and open (open conventional); 2) line ≥ 6 m wide and partially recovered (partial conventional); 3) line ≥ 6 m and recovered (closed conventional); 4) line open and ≤ 2 m wide (open 2 m); 5) line open and 3–4 m wide (open 3–4 m); 6) line open and 5 m wide (open 5 m) (Fig. 2). In the field we quantified the structural recovery state of conventional seismic lines by measuring horizontal cover (i.e., visual obstruction), shrub stem density, overhead cover, mode canopy height, coarse woody debris, and whether or not lines supported trees (woody stem ≥ 8 cm at dbh). See [20] for the full protocol description. We obtained equivalent measures in undisturbed forest plots adjacent lines for reference.
In the field we identified and sampled along the following 6 types of seismic lines: (Top row, from upper left) 1) ≥ 6 m wide and open (open conventional); 2) ≥ 6 m wide and partially recovered (partial conventional); 3) ≥ 6 m and recovered (closed conventional); (Bottom row, from left) 4) open and ≤ 2 m wide (open 2 m); 5) open and 3–4 m wide (open 3–4 m); 6) open and 5 m wide (open 5 m). Photo credits: J. Tigner.
At each camera location, we classified the surrounding forest stand as upland (and noted stands as deciduous, mixed-wood, or conifer) or lowland; and we estimated stand age on a rank scale from 1 to 4, corresponding to initiation, stem exclusion, mature, and old-growth, respectively [67]. We also searched for evidence of trapping activity at camera locations, along sampled seismic lines, and within clusters.
To test for significance of seismic line use relative to the forest interior we used a population averaged generalized estimating equation (GEE) [68] with a logit link and binomial error family in Stata 11.1 IC (Stata-Corp, College Station, Texas). We used data only from those clusters where a marten was detected in this analysis to evaluate differential use of seismic line types over a 10-day exposure period. We assumed that within a cluster where marten were known to occur, detection at a camera relative to others reflected a behavioural decision [69, 70]; cameras that did not detect a marten were in areas used less than areas where a camera detected a marten. We treated camera clusters as panels and assumed an exchangeable correlation structure among cameras within each cluster. We used a semi-robust estimator of variance to generate standard errors robust to misspecification.
We assumed the likelihood of detection at a given camera remained constant over the sample period. In some cases cameras failed prior to the full 10-night sample period. We removed cameras that failed within 24 hours of set up from the analysis. For remaining cameras, we included the natural log of the total sampling duration (in minutes) per camera as an offset in our model to account for the opportunity time for each camera to capture a photograph.
We included seismic line type, forest type and age at each camera as identified in the field, occurrence of trapping, the number of other species photographed at each camera, and Julian date at camera set up in a global model. Prior to fitting the global model we tested for collinearity among variables using a threshold of |r| ≥ 0.7. We then removed non-significant variables using a backward stepwise procedure and retained variables at P < 0.15 in the final model, but assessed significance at P < 0.1 [71]. We also tested for an interaction between line type and cumulative seismic line density, cumulative disturbance footprint, and forest type at the cluster level obtained from the GIS to measure whether marten response to seismic line types was constant or influenced by those surrounding variables. We report results as an odds-ratio (OR) using the forest interior as the reference condition to show the direction and magnitude of difference in use between lines and the forest interior [72].
Occurrence at the Home Range
We evaluated marten occurrence as the probability of detecting marten within a camera cluster along a continuum of seismic line density (photograph of marten in a home range = 1 vs. no photograph of marten = 0). We measured occurrence rather than occupancy because the assumption of closure is crucial to properly interpret occupancy estimates, but that assumption could not be ensured with our design [73]. Our survey duration and camera spacing were designed to garner a high likelihood of detecting marten if present [74–76] and we have no strong ecological basis with which to differentiate independent sub-surveys from our cameras into repeated surveys. We do not assume that detection probabilities are 1, and concede that the detection probability may decrease with decreasing marten density [77]. While this may underestimate an estimate of true occupancy [78], numerous studies on marten show that detection probability is high and consistent even at low densities [56, 57, 79–82], and is sufficient to directly compare occupied and non-occupied home ranges [57, 83, 84]. We assume that with our survey effort non-detections accurately reflect a reasonable estimate of true absence, and thus occurrence accurately reflects marten response to line density.
To test for significance between occurrence and line density we used the same GEE procedure and parameters described above. Here we grouped camera clusters by ecological district, a Canada-wide land cover classification (Ecological Stratification Working Group) to control for minor variations in forest structure and composition related to latitude and longitude across the study area. Clusters of mostly upland and mostly lowland forest occurred within each ecological district. We controlled for dominant forest type at the cluster scale as obtained from the GIS and for Julian date at camera set up during analysis. Significance was assessed at P < 0.1.
We then compared the relative strength of the all-seismic line density metric to explain occurrence at the home range by comparing model fit of other common metrics of linear feature density to the home range occurrence data. Specifically, we compared the all-seismic line metric (cumulative seismic density) to 1) seismic lines only, but excluding those ≤ 2 m wide (corrected seismic density); 2) roads and pipelines only (roads and pipelines); 3) roads only (roads); 4) all roads, pipelines, and seismic lines (cumulative line density); and 5) cumulative line density, but excluding seismic lines ≤ 2 m wide (corrected cumulative line density). Seismic line widths were identified from metadata attached to shape files, visually using satellite and Light Detection and Ranging (LiDAR) imagery, as measured in the field, and from discussions with energy companies active in the area. We compared the fit of each model to our data using a quasilikelihood information criterion (QIC) [85]. In all models we controlled for dominant forest type and Julian date by retaining those variables in each model run.
Results
We deployed cameras at 1035 unique locations across 173 clusters between May and September in 2008 and May and October in 2009. Sixty-three cameras failed in < 24 hours and were removed from analyses; camera failure occurred randomly across line treatments and political jurisdiction. Twenty-two camera clusters were established differently than described in the above protocol to attain sufficient line treatment replication; these clusters were not used in occurrence analyses, but were used to group camera availability for line use analyses. We considered the remaining 151 to be home range clusters and used those for occurrence analyses. Marten were detected at 143 out of 972 (14.7%) remaining locations, and in 66 of 151 (43.7%) home range clusters (75 of 173 total clusters).
Use of Seismic Line Type
We used data from 423 unique camera locations across 75 total clusters (open conventional: n = 39, partial conventional: n = 49, closed conventional: n = 46, open 2 m: n = 23, open 3–4 m: n = 15, open 5 m: n = 22, forest interior: n = 229). Marten response to seismic lines was significantly affected by the type of line encountered (Table 1). Line treatments with an odds-ratio > 1 were used more frequently than forest interior; those with an odds-ratio < 1 were used less frequently. Marten used open lines ≥ 3 m wide up to 90% less than forest interiors, and the odds of use of those line types relative to the odds of use of forest interiors was significant. In contrast, the odds marten used open lines ≤ 2 m wide or conventional lines supporting at least some regeneration of woody vegetation relative to the odds marten used forest interiors did not differ statistically.
Stand type at the camera was not retained in the final model indicating marten detection was not influenced by the forest type at the seismic line scale. Rank age of forest stands at the camera affected marten use (odds ratio = 1.473, P < 0.01); marten were more likely to use older stands (rank ages of 3 and 4) relative to younger ones (rank ages of 1 and 2). The number of other species photographed at a camera was retained in the final model (odds ratio = 0.749, P = 0.13) and may indicate that bait loss due to other species may influence marten finding cameras. Including or excluding any of these variables did not change the observed patterns or significance of use of seismic line types by marten. Julian date was not retained in the final model as we saw no change in marten observation with date. Evidence of trapping activity at camera locations, along sampled seismic lines, or within clusters was not retained in the final model. No tested interactions were significant indicating marten response was consistent relative to habitat type and disturbance at broader scales.
The recovery states of open, partial, and closed conventional seismic lines are shown in Table 2.
Occurrence at the Home Range
The probability of marten occurrence was significantly affected by seismic line density (cumulative seismic line density) at the home range scale (Fig. 3). The probability of occurrence per sampled home range ranged from a high of 0.76 in the least impacted home range (0 km/km2 line density) to 0.04 in the most impacted (26.4 km/km2). The mean probability fell from almost 60% in home ranges where seismic line density was low to approximately 20% in home ranges with the highest density of seismic lines (β = -0.083, P < 0.01) (Fig. 3). Interestingly, however, no marten where detected in home ranges beyond a seismic line density of 20.8 km/km2.
The probability that American marten (Martes americana) occur at the home range is significantly influenced by seismic line density in northwest Canada. As seismic line density increases, the mean probability of occurrence declines.
Univariate analyses showed both lowland and upland forest types affected marten occurrence at the home range scale, but that the negative effect of lowland forest was stronger than the positive effect of uplands (lowland: β = -0.487, P < 0.01, upland: β = 0.361, P < 0.01). As such, we controlled for lowland forest as the dominant forest type in our model. Dominant forest type did not have a significant effect on marten occurrence in the cumulative seismic density model, but was significant in models where seismic lines ≤ 2 m were removed from the density metrics (corrected seismic density and corrected cumulative line density). Julian date was not significant in any models.
The metric of linear feature density that best fit our data was corrected seismic density which included all constructed seismic lines, except those ≤ 2 m wide (Table 3). However, all metrics of linear feature density that included seismic lines carry similar QIC weights suggesting all plausibly explain declining marten occurrence at the home range scale [86] (Table 3). Despite similar model weights, the shape of the relationship between line density and the mean probability of occurrence differed across models. For example, corrected seismic line density predicts a steeper and larger decline in marten occurrence than cumulative line density (Fig. 4). Corrected seismic line density estimated a decline in the mean probability of marten occurrence at the home range from almost 60% to less than 10%, representing a reduction of over 80% across the sampled continuum of seismic line density (Fig. 4).
The relationship between the probability of American marten (Martes americana) occurrence and linear feature density at the home range in northwest Canada is sensitive to line density calculations. Although both metrics of seismic line density show a clear decline, where narrow seismic lines (≤ 2 m) are removed from the calculation (i.e., corrected seismic density), decline in mean probability of occurrence is almost 80%.
All metrics including seismic lines were significant in model runs (corrected seismic density: β = -0.116, P < 0.01; cumulative line density: β = -0.073; P < 0.01; corrected cumulative line density: β = -0.100, P < 0.01). Models that excluded seismic lines ≤ 2 m wide performed better than models that did not differentiate between seismic line types. Where seismic lines ≤ 2 m wide were removed from density metrics, average linear feature density fell from 9.436 km/km2 to 6.531 km/km2 (seismic only models), and from 10.031 km/km2 to 7.369 km/km2 (cumulative line models). The line density metrics based on roads and on roads and pipelines performed poorly and were not significant (P = 0.16 and P = 0.52, respectively).
Within group (ecological district) correlation of model residuals for the corrected seismic density model were extremely low (0.006) suggesting any spatial autocorrelation between occupied home ranges was accounted for in the model. Correlation among corrected cumulative line density residuals was similarly low (0.007). Mean probability of home range occurrence was 0.437, but that differed significantly between jurisdictions. In the NWT development is minimal and occurrence was 0.571. In BC development is widespread but employs industry best practices and occurrence was 0.436. In Alberta development is widespread, has been continuous for > 5 decades, and often uses older development methods. Occurrence in Alberta was 0.115. Relative to the NWT, occurrence was not statistically different in BC (OR = 0.754, P = 0.47), but was in Alberta (OR = 0.282, P < 0.01).
Discussion
The goal of this study was to measure the response of American marten to seismic lines that varied by width and recovery states, to test for a link between marten response to individual lines and occurrence at a home range scale along a continuum of line density, and to compare the efficacy of common metrics of linear feature density to explain home range occurrence. Marten response to seismic lines was influenced by line width and line recovery. While marten avoided open seismic lines ≥ 3 m wide (i.e., both conventional and some types of LIS lines) compared to forest interiors, use of open lines ≤ 2 m wide and conventional lines (≥ 6 m wide) with at least partial recovery of woody vegetation was similar to use of forest interior.
Uncovering the mechanisms driving marten response to seismic lines and linear feature density was beyond the scope of this study. Still, interpreting our results within the context of marten ecology in a boreal system is useful in recommending reasonable mitigation strategies. We suspect that marten response to individual seismic lines is driven by a real or perceived risk of predation. Numerous studies show marten avoid open habitats [36, 39, 41–43], but reuse disturbed habitats with sufficient overhead and lateral cover [36, 45–49]. Marten are predated most frequently in openings [50, 87] and predated on by a variety of mammalian and avian predators [88], some of which use seismic lines as travel corridors [12, 19]. Even with high prey availability in open areas, marten do not travel far from cover [40, 56]. Despite observed avoidance of open seismic lines in this study, open linear features in the Northwest Territories support ample food resources for marten [89]. Renewed use of lines occurred with even moderate increases in woody vegetation, likely well before such increases would trigger structural shifts on lines to support forest-dwelling prey items [90–92], favored by marten [93, 94]. It is implausible trapping influenced marten use of seismic lines because we tested the response to lines only within home ranges where marten occurred. Avoidance would require marten to comprehend both a trapping risk and have some knowledge of specific trapping locations along lines.
Marten occurrence at the home range scale declined precipitously relative to increasing seismic line density. In our best model, the mean probability of occurrence declined by over 80% between the least and most impacted home ranges. This shows a clear link between a mean behavioral response of individual marten to avoid certain seismic line types and a population impact. Numerous studies show that declines in marten occurrence at a landscape scale translate to population declines [57, 60, 62], lower population density [51, 54, 77], changes in population structure [59], and reduced fitness of remaining marten [52, 61]. Other studies that link marten avoidance of fine scale disturbances to landscape abandonment typically report a threshold response (e.g. a sudden dramatic decline) beyond disturbance to 25%- 49% of study landscapes [39, 55–57]. By contrast, we observed a decline in occurrence at low levels of total habitat disturbance, and no discernable threshold response. Only ~15% of the total land base within our most heavily disturbed home range was directly impacted, and decline was consistent along the continuum of line density we sampled.
It is possible this decline is indicative of a functional habitat loss, wherein areas adjacent to, but not directly impacted by seismic lines were avoided. Functional habitat loss relative to seismic lines has also been shown in woodland caribou [11], and relative to well pads in mule deer [30, 95]. However, marten use cut block edges for travel and hunting at least proportionally to availability [53, 96] elsewhere.
Perhaps more likely is a cumulative effect of linear feature development and increased trapping pressure. Used to facilitate exploration, seismic lines are often the first or only access to remote areas and can facilitate new or increased access for trapping [4]. Where seismic exploration is promising, increased development activity ultimately results in other disturbances including well pads, roads, and pipelines. Although trapping typically wanes where extensive energy development occurs, initial access via seismic lines may have triggered over-trapping [97, 98] or trapping from which marten could not readily rebound given additional habitat disturbance from additional development activity. Regardless of whether seismic line density or trapping is the specific mechanism, we show a strong negative correlation between seismic line density and marten occurrence at the home range scale.
Several common metrics of linear feature density explained declines in home range occurrence, but others did not. QIC scores showed metrics that included seismic lines performed better than metrics that did not, and that metrics corrected for marten response to seismic line types (i.e., exclusion of seismic lines ≤ 2 m wide based on behavioral analyses) performed best [86]. This is important because we clearly show it is necessary to account for linear feature type when calculating threshold metrics. Without differentiating between a mapped and an ecologically relevant disturbance features, we actually underestimated the impact of linear feature density on marten occurrence. Compared to cumulative metrics, occurrence declined more steeply and to a lower overall probability using an accurate interpretation of line density. Of greater concern is the potential to exclude ecologically relevant disturbances from line density metrics. For example, we identified many home ranges where roads were absent or road density was low, but where seismic line density was high. If threshold metrics were based on road density alone, management would fail to prevent the significant impacts on marten related to seismic lines.
Although corrected metrics performed best, improvement was not substantial given the clear behavioral responses to seismic line types. In home ranges containing LIS lines ≤ 2 m wide, the average change between cumulative and corrected line densities was only 2.73 km/km2 (9.439 km/km2 and 6.709 km/km2, respectively). It is possible that line density was sufficiently high to affect occurrence regardless of how density metrics were calculated. It is also possible, that a further correction of density metrics wherein recovered lines were excluded may have improved model fit. Unfortunately, this was not possible because line recovery is poorly tied to line age [17, 20, 24, 99], nor were spatial data available for our study area to estimate current recovery states of seismic lines.
Our intention with this research was not to suggest that seismic lines are the most influential component of an energy sector disturbance footprint on marten. As discussed, seismic lines often lead to additional disturbances, all of which have been shown to impact species in different ways. Instead this work makes an important contribution to understanding the impacts of seismic lines on a species and in developing mitigation strategies based on ecologically-derived metrics.
By comparing marten use along seismic lines to use at forest interior within a population-averaged framework, we showed how individual marten would respond, on average, across a 200,000 km2 region to a given seismic line relative to expected use at a reference category [19, 72]. Our findings are in stark contrast to current guidelines that suggest any LIS lines are capable of mitigating behavioral impacts for wildlife, or that seismic lines constitute permanent disturbance features. Based on our study, marten make no distinction between seismic lines between 3 m and ≥ 6 m wide when open, but do distinguish open seismic lines ≥ 6 m wide as different from partially recovered and closed lines. This provides the ecological context from which to make informed management decisions on seismic lines in the boreal forest. In other words, for LIS line types to meet management expectations for marten, lines should be cut to a maximum width of ≤ 2 m wide. However, once a seismic line is cut it likely recovers to functional habitat for marten at levels well below what would be considered recovered from a forestry perspective. Clearly, the maximum line width or minimum recovery state required to mitigate the impacts of seismic lines on other species may differ [12, 15, 19]. Given a large number of wide seismic lines across boreal Canada today, additional work is needed to better understand region-specific recovery trajectories and to better identify recovery states using remote sensing technologies.
Similarly, by comparing marten occurrence at the home range along a continuum of linear feature density, we showed the expected probability of marten occurrence along that continuum [100, 101]. This provides a clear way to contextualize the ramifications of varying linear feature densities on marten in a way that can easily inform decision-making processes for setting specific thresholds. For example, in our study area, proposed regulatory thresholds for line density range from 0.6 km/km2 to 2.4 km/km2, but we observed very little change in the probability of marten occurrence between those densities. Based on our data, it seems likely that even the most liberal threshold limit in our study area is sufficient to maintain marten on a landscape subject to energy development. Whether a threshold limit of 2.4 km/km2 is the appropriate for northwest Canada is dependent on a myriad of socio-economic values and considerations. Regardless, we show that for marten, we can expect a 2.4 km/km2 threshold to incur very little cost in terms of a reduced probability of marten occurrence or a population impact relative to undisturbed landscapes. However, seismic lines are often clumped in space. Since regulatory thresholds are strongly influenced by the spatial extent at which they are applied, regulators need to be aware that portions of planning units may become unsuitable for marten if a regionally-applied threshold of 2.4 km/km2 is achieved without regard for the spatial patterning of seismic lines.
Lastly, we showed that a complete understanding of the cumulative impact of linear disturbances to marten at a population scale is predicated on a thorough understanding of behavioral responses to individual disturbance types. Attributing declines in species populations to a cumulative disturbance footprint based behavioral responses to individual disturbances [3, 14, 102–105] may overemphasize the impacts of disturbances thereby precluding efficient mitigation strategies. Our findings clearly show that all seismic lines do not constitute equal disturbances and that different lines may warrant different management considerations. This is mirrored in other, similarly in depth studies of species response to disturbance features that vary by attribute [32, 95, 106–108]. Too narrow a focus on mitigating the impacts of individual lines may overlook potentially substantial ecological consequences of accumulating seismic line densities at broad spatial scales.
Supporting Information
S1 Dataset. Data used to measure American marten response to seismic lines and seismic line density.
https://doi.org/10.1371/journal.pone.0118720.s001
(XLSX)
Acknowledgments
Thanks to our hard working field assistants: M. Wilson, D. MacNearney, E. Hamblen, C. Chang, J. Masserone, S. O’Donovan, J. Hayden, D. Huntingzer, M. Minic, J. Samuel, and K. Knysh. Thanks also to J. Goshen, R. Powell, A. Shafer, J. Northrup, M. Cunha and 2 anonymous reviewers for suggestions that improved this manuscript.
Author Contributions
Conceived and designed the experiments: JT EMB SB. Performed the experiments: JT. Analyzed the data: JT EMB. Contributed reagents/materials/analysis tools: JT EMB SB. Wrote the paper: JT EMB SB.
References
- 1.
Northrup JM, Wittemyer G. Characterising the impacts of emerging energy development on wildlife, with an eye towards mitigation. Ecol Lett. 2012; https://doi.org/10.1111/ele.12009
- 2.
Schneider RR. Alternative futures: Alberta’s boreal forest at the crossroads. Edmonton: Federation of Alberta Naturalists; 2002. https://doi.org/10.2527/jas.2014-7375 pmid:25074450
- 3. Sorensen T, McLoughlin PD, Hervieux D, Dzus E, Nolan J, Wynes B, et al. Sustainable levels of cumulative effects for boreal caribou. J Wildl Manage. 2008; 72: 900–905.
- 4.
Environment Canada. Recovery strategy for the woodland caribou, boreal population (Rangifer tarandus caribou) in Canada. Ottawa: Environment Canada; 2012.
- 5. Bayne EM, Van Wilgenburg SL, Boutin S, Hobson KA. Modeling and field-testing of ovenbird (Seiurus aurocapillus) responses to boreal forest dissection by the energy sector development at multiple spatial scales. Landsc Ecol. 2005a; 20: 203–216.
- 6.
Salmo Consulting, Diversified Environmental Services, Gaia Consultants, Forem Technologies, Axys Environmental Consulting. Cumulative effects indicators, thresholds, and case studies. Victoria: BC Oil and Gas Commission; 2003.
- 7.
Salmo Consulting, Axys Environmental Consulting, Forem Technologies. Deh Cho cumulative effects study, Phase I: management indicators and thresholds. Calgary: Salmo Consulting; 2004.
- 8.
Dehcho Land Use Planning Committee. Respect for the land: the Dehcho land use plan. Fort Providence: Dehcho First Nation; 2006.
- 9.
Kennett S. From science-based thresholds to regulatory limits: implementation issues for cumulative effects management. Calgary: Canadian Institute of Resources Law; 2006.
- 10. Komers PE, Stanojevic Z. Rates of disturbance vary by data resolution: implications for conservation schedules using the Alberta boreal forest as a case study. Glob Chang Biol. 2013; 19: 2916–2928. pmid:23703708
- 11. Dyer SJ, O’Neil JP, Wasel JM, Boutin S. Avoidance of industrial development by woodland caribou. J Wildl Manage. 2001; 65: 531–542.
- 12. Latham ADM, Latham MC, Boyce MS, Boutin S. Movement responses by wolves to industrial linear features and their effect on woodland caribou in northeast Alberta. Ecol Appl. 2011; 21: 2854–2865.
- 13. James ARC, Stuart-Smith AK. Distribution of caribou and wolves in relation to linear corridors. J Wildl Manage. 2000; 64: 154–159.
- 14. Dyer SJ, O’Neil JP, Wasel JM, Boutin S. Quantifying barrier effects of roads and seismic lines on movements of female woodland caribou in northeastern Alberta. Can J Zool. 2002; 80: 839–845.
- 15. Bayne EM, Boutin S, Tracz B, Charest K. Functional and numerical responses of ovenbirds (Seiurus aurocapilla) to changing seismic exploration practices in Alberta’s boreal forest. Ecoscience 2005b; 12: 216–222.
- 16. Ashenhurst AR, Hannon SJ. Effects of seismic lines on the abundance of breeding birds in the Kendall Island Bird Sanctuary, Northwest Territories, Canada. Arctic 2008; 51: 190–198.
- 17. Lee P, Boutin S. Persistence and developmental transition of wide seismic lines in the western Boreal Plains of Canada. J Environ Manage. 2006; 78: 240–250. pmid:16112795
- 18. Weclaw P, Hudson RJ. Simulation of conservation and management of woodland caribou. Ecol Modell 2004; 177: 75–94.
- 19. Tigner J, Bayne EM, Boutin S. Black bear use of seismic lines in northern Canada. J Wildl Manage. 2014; 78: 282–292.
- 20.
Bayne E, Lankau H, Tigner J. Ecologically-based criteria to assess the impact and recovery of seismic lines: the importance of width, regeneration, and seismic density. Calgary: Environmental Studies Research Fund; 2011.
- 21.
Revel RD, Dougherty TD, Downing DJ. Forest growth and regeneration along seismic lines. Calgary: University of Calgary Press; 1984.
- 22. Machtans CS. Songbird response to seismic lines in the western boreal forest: a manipulative experiment. Can J Zool. 2006; 84: 1421–1430.
- 23.
MacFarlane AK. Vegetation response to seismic lines: edge effects and on-line succession. M.Sc. Thesis, University of Alberta. 2003.
- 24. Kemper JT, Macdonald SE. Directional change in upland tundra plant communities 20–30 years after seismic exploration in the Canadian low-arctic. J Veg Sci. 2009; 20: 557–567.
- 25.
Oberg PR. Responses of mountain caribou to linear features in a west-central Alberta landscape. M.Sc. Thesis, University of Alberta. 2001. Available: http://hdl.handle.net/10402/era.21699
- 26.
Neufeld LM. Spatial dynamics of wolves and woodland caribou in an industrial forest landscape in west-central Alberta. M.Sc. Thesis, University of Alberta. 2006.
- 27. Lankau HE, Bayne EM, Machtans CS. Ovenbird (Seiurus auocapilla) territory placement near seismic lines is influenced by forest regeneration and conspecific density. Avian Conservation and Ecology 2013; 8(1): 5.
- 28. Caro T. The behaviour-conservation interface. Trends Ecol Evol. 1999; 14: 66–369.
- 29.
Gill JA, Sutherland WJ. The role of behavioural decision-making in predicting the consequences of human disturbance. In Gosling LM, Sutherland WJ, editors. Behaviour and conservation. Cambridge: Cambridge University Press; 2000. pp. 51–64.
- 30. Sawyer H, Nielson RM, Lindzey F, McDonald LL. Winter habitat selection of mule deer before and during development of a natural gas field. J Wildl Manage. 2006; 70: 396–403.
- 31. Doherty KE, Naugle DE, Walker BL, Graham JM. Greater sage-grouse winter habitat selection and energy development. J Wildl Manage. 2008; 72: 187–195.
- 32. Habib L, Bayne EM, Boutin S. Chronic industrial noise affects pairing success and age structure of ovenbirds Seiurus aurocapilla. J Appl Ecol. 2007; 44: 176–184.
- 33. Gill JA, Norris K, Sutherland WJ. Why behavioural responses may not reflect the population consequences of human disturbance. Biol Conserv. 2001; 97: 265–268.
- 34. Griffin SC, Valois T, Taper ML, Mills LS. Effects of tourists on behavior and demography of Olympic marmots. Conserv Biol. 2007; 21: 1070–1081. pmid:17650256
- 35.
Indian and Northern Affairs Canada. A preliminary state of knowledge report of valued components. Yellowknife: NWT Cumulative Impact Monitoring Program; 2007.
- 36. Payer DC, Harrison DJ. Influence of forest structure on habitat use by American marten in an industrial forest. For Ecol Manage. 2003; 179: 145–156.
- 37. Godbout G, Ouellet J. Fine-scale habitat selection of American marten at the southern fringe of the boreal forest. Ecoscience 2010; 17: 175–185.
- 38. Chapin TG, Harrison DJ, Phillips DM. Seasonal habitat selection by marten in an untrapped forest preserve. J Wildl Manage. 1997; 61: 707–717.
- 39. Potvin F, Belanger L, Lowell K. Marten habitat selection in a clearcut boreal landscape. Conserv Biol. 2002; 14: 844–857.
- 40. Andruskiw M, Fryxell JM, Thompson ID, Baker JA. Habitat-mediated variation in predation risk by the American marten. Ecology. 2008; 89: 2273–2280. pmid:18724737
- 41. Smith AC, Schaefer JA. Home-range size and habitat selection by American marten (Martes americana) in Labrador. Can J Zool. 2002; 80: 1602–1609.
- 42. Godbout G, Ouellet J. Habitat selection of American marten in a logged landscape at the southern fringe of the boreal forest. Ecoscience 2008; 15: 332–342.
- 43. Fuller AK, Harrison DJ. Influence of partial timber harvesting on American martens in north-central Maine. J Wildl Manage. 2005; 69: 710–722.
- 44. Robitaille JF, Aubry K. Occurrence and activity of American martens (Martes americana) in relation to roads and other routes. Acta Theriol. 2000; 45: 137–143.
- 45. Payer DC, Harrison DJ. Structural differences between forests regenerating following spruce budworm defoliation and clear-cut harvesting: implications for marten. Can J For Res. 2000; 30: 1965–1972.
- 46. Poole KG, Porter AD, de Vries A, Maundrell C, Grindal SD, St Clair CC. Suitability of a young deciduous-dominated forest for American marten and the effects of forest removal. Can J Zool. 2004; 82: 423–435.
- 47. Porter AD, St. Clair CC, de Vries A. Fine-scale selection by marten during winter in a young deciduous forest. Can J For Res. 2005; 35: 901–909.
- 48. Thompson ID, Baker JA, Jastrebski C, Dacosta J, Fryxell J, Corbett D. Effects of post-harvest silviculture on use of boreal forest stands by amphibians and marten in Ontario. Forestry Chronicle 2008; 84: 741–747.
- 49. Hearn BJ, Harrison DJ, Fuller AK, Lundrigan CG, Curran WJ. Paradigm shifts in habitat ecology of threatened Newfoundland martens. J Wildl Manage. 2010; 74: 719–728.
- 50. Thompson ID, Colgan PW. Marten activity in uncut and logged boreal forests in Ontario. J Wildl Manage. 1994; 58: 280–288.
- 51.
Potvin F, Brenton L. Short-term effects of clearcutting on martens and their prey in the boreal forest of western Quebec. In: Proulx G, Bryant HN, Woodard PM, editors. Martes: taxonomy, ecology, techniques and management. Edmonton: Provincial Museum of Alberta; 1997. pp. 452–474.
- 52. Johnson CA, Fryxell JM, Thompson ID, Baker JA. Mortality risk increases with natal dispersal distance in American martens. Proc R Soc Lond B Biol Sci. 2009; 276: 3361–3367.
- 53. Cushman SA, Raphael MG, Ruggiero LF, Shirk AS, Wasserman TN, O’Doherty EC. Limiting factors and landscape connectivity: the American marten in the Rocky Mountains. Landsc Ecol. 2011; 26: 1137–1149.
- 54. Soutiere EC. Effects of timber harvesting on marten in Maine. J Wildl Manage. 1979; 43: 850–860.
- 55. Chapin TG, Harrison DJ, Katnik DD. Influence of landscape pattern on habitat use by American marten in an industrial forest. Conserv Biol. 1998; 12: 1327–1337.
- 56. Hargis CD, Bissonette JA, Turner DL. The influence of forest fragmentation and landscape pattern on American marten. J Appl Ecol. 1999; 36:157–172.
- 57. Moriarty KM, Zielinski WJ, Forsman ED. Decline in American marten occupancy rates at Sagehen Experimental Forest, California. J Wildl Manage. 2011; 75: 1774–1787.
- 58.
Fortin C, Cantin M. The effects of trapping on a newly exploited American marten population. In: Buskirk SW, Harestad AS, Raphael MG, Powell RA, editors. Martens, sables, and fishers: biology and conservation. Ithaca: Cornell University Press; 1994. Pp. 179–191.
- 59. Thompson ID. Marten populations in uncut and logged boreal forests in Ontario. J Wildl Manage. 1994; 58: 272–280.
- 60.
Payer DC. Influence of timber harvesting and trapping on habitat selection and demographic characteristics of marten. Ph.D. Dissertation, University of Maine. 1999.
- 61. Cheveau M, Imbeau L, Drapeau P, Belanger L. Marten space use and habitat selection in managed coniferous boreal forests of eastern Canada. J Wildl Manage. 2013; 77: 749–760.
- 62. Snyder JE, Bissonette JA. Marten use of clear-cuttings and residential forest stands in western Newfoundland. Can J Zool. 1987; 65: 169–174.
- 63.
Ecosystem Classification Group. Ecological regions of the Northwest Territories: taiga plains. Yellowknife: Government of the Northwest Territories; 2007.
- 64. Wulder MA, Dechka JA, Gillis MA, Luther JE, Hall RJ, Beaudoin A, et al. 2003. Operational mapping of the land cover of the forested area of Canada with Landsat data: EOSD land cover program. Forestry Chronicles 2003; 79: 1075–1083.
- 65. Buskirk SW, McDonald LL. Analysis of variability in home-range sizes of the American marten. J Wildl Manage. 1989; 53: 997–1004.
- 66.
Powell RA. Structure and spacing of Martes populations. In: Buskirk SW, Harestad AS, Raphael MG, Powell RA, editors. Martens, sables, and fishers: biology and conservation. Ithaca: Cornell University Press; 1994. Pp. 101–121.
- 67. Chen HYH, Popadiouk RV. Dynamics of North American boreal mixedwoods. Environmental Reviews 2002; 10: 137–166.
- 68.
Hardin JW, Hilbe LM. Generalized estimating equations. Boca Raton: Chapman & Hall / CRC; 2003.
- 69. Aarts G, MacKenzie M, McConnell B, Fedak M, Matthiopoulos J. Estimating space-use and habitat preference from wildlife telemetry data. Ecography 2008; 31: 140–160.
- 70. Beyer HL, Haydon DT, Morales JM, Frair JL, Hebblewhite M, Mitchell M, et al. The interpretation of habitat preference metrics under use-availability designs. Philos Trans R Soc Lond B Biol Sci 2010; 365: 2245–2254. pmid:20566501
- 71.
Hosmer DW, Lemeshow S. 2000. Applied logistic regression. 2nd ed. Hoboken: John Wiley and Sons Inc.; 2000.
- 72. Nielsen SE, Bayne EM, Schieck J, Herbers J, Boutin S. A new method to estimate species and biodiversity intactness using empirically derived reference conditions. Biol Conserv. 2007; 137: 403–414.
- 73. Mackenzie DI, Nichols JD, Lachman GB, Dorege S, Royle JA, Langtimm CA. Estimating site occupancy rates when detection probabilities are less than one. Ecology 2002; 83: 2248–2255.
- 74. Otis DL, Burnham KP, White GW, Anderson DR. Statistical inference from capture data on closed animal populations. Wildlife Monographs 1987; 62.
- 75. Roughton RD, Sweeny MD. Refinements in scent-station methodology for assessing trends in carnivore populations. J Wildl Manage. 1982; 46: 217–229.
- 76. Mackenzie DI, Royle JA. Designing occupancy studies: general advice and allocating survey effort. J Appl Ecol. 2005; 42: 1105–1114.
- 77. Smith JB, Jenkins JA, Klaver RW. Evaluating detection probabilities for American marten in the Black Hills, South Dakota. J Wildl Manage. 2007; 71: 2412–2416.
- 78. Gu WD, Swihart RK. Absent or undetected? Effects of non-detection of species occurrence on wildlife-habitat models. Biol Conserv. 2004; 116: 195–203.
- 79. Slauson KM, Zielinski WJ, Hayes JP. Habitat selection by American martens in coastal California. J Wildl Manage. 2007; 71: 458–468.
- 80. Baldwin RA, Bender LC. Distribution, occupancy, and habitat correlates of American martens (Martes americana) in Rocky Mountain Nation Park, Colorado. J Mammal 2008; 89: 419–427.
- 81. Zielinski WJ, Slauson KM, Bowles A. Effects of off-highway vehicle use on the American marten. J Wildl Manage. 2008; 72: 1558–1571.
- 82. Flynn RW, Schumacher TV. Temporal changes in population dynamics of American martens. J Wildl Manage. 2009; 73: 1269–1281.
- 83. Zielinski WJ, Stauffer B. Monitoring Martes populations in California: survey design and power analysis. Ecol Appl. 1996; 6: 1254–1267.
- 84. Kirk TA, Zielinski WJ. Developing and testing a landscape habitat suitability model for the American marten (Martes americana) in the Cascade mountains of California. Landsc Ecol. 2009; 24: 759–773.
- 85. Pan W. Akaike’s information criterion in generalized estimating equations. Biometrics 2001; 57: 120–125. pmid:11252586
- 86.
Burnham KP, Anderson DR. Model selection and multimodel inference. A practical information-theoretic approach. 2nd ed. New York: Springer-Verlag; 2002.
- 87.
Ruggiero LF, Aubry KB, Buskrik SW, Lyon JL, Zielinski WJ. The scientific basis for conserving forest carnivores: American marten, fisher, lynx, and wolverine in the western United States. Washington D. C.: United States Forest Service; 1994.
- 88. McCann NP, Zollner PA, Gilbert JH. Survival of adult martens in northern Wisconsin. J Wildl Manage. 2010; 74: 1502–1507.
- 89.
Darling AF. Linear feature effects on small mammal abundance and resources in the boreal forest. M. Sc. Thesis. University of Alberta. 2009.
- 90. St-Laurent M-H, Ferron J, Hache S, Gagon R. Planning timber harvest of residual forest stands without compromising bird and small mammal communities in boreal landscapes. For Ecol Manage. 2008; 254: 261–275.
- 91. Zwolak R. A meta-analysis of the effects of wildfire, clearcutting, and partial harvest on the abundance of North American small mammals. For Ecol Manage. 2009; 258: 539–545.
- 92. Vanderwel MC, Malcolm JR, Caspersen JP, Newman MA. Fine-scale habitat association of red-backed voles in boreal mixedwood stands. J Wildl Manage. 2010; 74: 1492–1501.
- 93. Douglass RJ, Fisher LG, Mair M. Habitat selection and food-habits of marten, Martes americana, in the Northwest Territories. Can Field Nat 1983; 97: 71–74.
- 94. Ben-David MR, Flynn W, Schell DM. Annual and seasonal changes in diets of martens: evidence from stable isotope analysis. Oecologia 1997; 111: 280–291.
- 95. Sawyer H, Kauffman MJ, Nielson RM. Influence of well pad activity on winter habitat selection patterns of mule deer. J Wildl Manage. 2009; 73: 1052–1061.
- 96. Vigeant-Langlois C, Desrochers A. Movements of wintering American marten (Martes americana): relative influences of prey activity and forest stand age. Can J For Res. 2011; 41: 2202–2208.
- 97. Hodgman TP, Harrison DJ, Katnik DD, Elowe KD. Survival in an intensively trapped marten population in Maine. J Wildl Manage. 1994; 58: 593–600.
- 98. Wiebe PA, Fryxell JM, Thompson ID, Börger L, Baker JA. Do trappers understand marten habitat? J Wildl Manage. 2013; 77: 379–391.
- 99. Jorgenson JC, Ver Hoef JM, Jorgenson MT. Long-term recovery patterns of arctic tundra after winter seismic exploration. Ecol Appl. 2010; 20: 205–221. pmid:20349841
- 100. Schmiegelow FKA, Mönkkönen M. Habitat loss and fragmentation in dynamic landscapes: avian perspectives from the boreal forest. Ecol Appl. 2002; 12: 375–389.
- 101. Schneider RR, Stelfox JB, Boutin S, Wasel S. Managing the cumulative impacts of land uses in the Western Canadian Sedimentary Basin: a modeling approach. Conservation Ecology 2003; 7: 8.
- 102. Walker BL, Naugle DE, Doherty KE. Greater sage-grouse population response to energy development and habitat loss. J Wildl Manage. 2007; 71: 2644–2654.
- 103. Harju SM, Dzialak MR, Taylor RC, Hayden-Wing LD, Winstead JB. Thresholds and time lags in effects of energy development on greater sage-grouse populations. J Wildl Manage. 2010; 74: 437–448.
- 104. Gilbert MM, Chalfoun AD. Energy development affects populations of sagebrush songbirds in Wyoming. J Wildl Manage. 2011; 75: 816–824.
- 105. Strimbu B, Innes J. An analytical platform for cumulative impact assessment based on multiple futures: the impact of petroleum drilling and forest harvesting on moose (Alces alces) and marten (Martes americana) habitats in northeastern British Columbia. J Environ Manage. 2011; 92: 1740–1752. pmid:21396768
- 106. Bayne EM, Habib L, Boutin S. Impacts of chronic anthropogenic noise from energy-sector activity on abundance of songbirds in the boreal forest. Conserv Biol. 2008; 22: 1186–1193. pmid:18616740
- 107. Francis CD, Paritsis J, Ortega CP, Cruz A. Landscape patterns of avian habitat use and nest success are affected by chronic gas well compressor noise. Landsc Ecol. 2011; 26: 1269–1280.
- 108. Wasser SK, Keim JL, Taper ML, Lele SR. The influences of wolf predation, habitat loss, and human activity on caribou and moose in the Alberta oil sands. Front Ecol Environ 2011; 9: 546–551.