Lost in translation? Insights into caribou habitat selection from forest inventory data

The research-implementation gap is an impediment to effective conservation and management of the environment (Cash et al. 2003; Weichselgartner and Kasperson 2010). Barriers to the implementation of research in applied settings include mismatch between data produced from research and the information needed by practitioners or policy-makers, difficulties interpreting and communicating scientific results, and lack of resources or feasible plans to drive implementation of scientific results (Arlettaz et al. 2010; Cook et al. 2013). Translating science into actionable management and policy requires effective communication and collaboration among scientists, practitioners, and policy-makers (i.e., translational ecology; Sutherland et al. 2011; Enquist et al. 2017; Hallett et al. 2017 or boundary science; Cook et al. 2013).

In planning conservation efforts, mitigation of human impacts, and adaptive management, ecological research is an essential means of linking habitat to species use and occurrence in a spatial framework (Margules and Pressey 2000; Pressey et al. 2007; de Groot et al. 2010). In recent years, use of satellite-derived habitat data to understand associations between species and habitat and for spatial planning has increased along with the availability of high-resolution satellite images that are accurate across large areas, updated regularly, and often freely available (e.g., Warren et al. 2014; Bleyhl et al. 2017; Hofmann et al. 2017; Tucker et al. 2018). However, despite many advantages, the satellite-derived habitat data used by scientists often do not meet the specific needs of practitioners and therefore could be a barrier between science and implementation. This mismatch in data is particularly notable in the forest sector where management is largely based on inventory data derived from a combination of aerial photo interpretation and field visits (McDermid et al. 2009; Päivinen et al. 2009). Forest inventory data provide detailed information required for harvest management, specifically, information on stand attributes and tree species that is often obscured from satellite-derived habitat data (but see Gastón et al. 2017). However, despite being tailored for use by the forest sector, forest inventory data can be challenging to integrate into ecological models due to inconsistencies between jurisdictions (but see Gillis et al. 2005; Gschwantner et al. 2016), mismatches between spatial and temporal coverage for wide-ranging species (e.g., alpine and northern areas; Johnson et al. 2003; McDermid et al. 2009; Maxie et al. 2010; Imbeau et al. 2015), and disjunction between forest resource inventory attributes and ecological function for wildlife species (McLaren and Mahoney 2001; Johnson et al. 2003; McDermid et al. 2009; Boan et al. 2013). Satellite-derived habitat data are the better choice for spatial models, particularly when the goal is a map or probability surface or where the area of interest is large or crosses jurisdictional boundaries. However, at a fine scale, integration of forest resource inventory data into ecological models represents a significant opportunity to lessen the research-implementation gap. Indeed, successful application of forest inventory data to model and predict species distribution and use would facilitate the implementation of research findings into concrete management actions and thereby serve to bridge the research-implementation gap.

Across the boreal forest of Canada, the forest sector is under increasing pressure to implement sustainable forest management that considers biodiversity and species at risk (Brown et al. 2007; Canadian Council of Forest Ministers 2014). Of particular concern in recent years are woodland caribou, Rangifer tarandus caribou, which are declining across their range and listed as threatened under the Canadian Species at Risk Act. Woodland caribou (hereafter caribou) require large tracts of intact mature forest to maintain a low spatial overlap with predators (Bergerud et al. 1990; DeCesare et al. 2010). Habitat disturbance within caribou ranges, including forest harvesting, has altered predator–prey dynamics by increasing the spatial overlap between caribou and ungulates that prefer early seral forest, which in turn has increased the spatial overlap between caribou and shared predators (DeCesare et al. 2010), increased predation risk for caribou (DeCesare et al. 2014), and resulted in widespread caribou declines (Vors and Boyce 2009; Festa-Bianchet et al. 2011; Hervieux et al. 2013). Caribou are arguably of greatest concern in the province of Alberta, and there is immediate pressure on the forest sector to adapt harvesting strategies to maintain caribou populations (Environment Canada 2012, 2014; Government of Alberta 2017). Although resource selection function probability surfaces for caribou in Alberta are available (DeCesare et al. 2012; Slater 2013), and the relationship between caribou and satellite-derived habitat data in Alberta is well understood (e.g., Neufeld 2006; DeCesare et al. 2012; Johnson et al. 2015), habitat categories such as “closed conifer” are too coarse for forest practitioners who make management plans using fine-scale stand information (e.g., height, age, and leading species) and may be impeding the ready translation of scientific results into practical solutions for caribou conservation. With the exception of one study in the northeast of the province (Latham et al. 2013) and two short-term studies in west-central Alberta (Szkorupa and Schmiegelow 2003; Saher and Schmiegelow 2005), there are no detailed multi-year assessments of caribou response to habitat described using the provincial standard for forest inventory data, the Alberta Vegetation Inventory (AVI; Alberta Sustainable Resource Development 2005), in the peer-reviewed literature. Recognizing that caribou conservation is a multi-faceted problem with social, ecologic, and economic implications, we suggest that one way to bridge the research-implementation gap in this context is to use forest resource inventory data to evaluate caribou response to habitat in a way that is easy to integrate into existing forest management planning in Alberta.

Driven by needs of the forest sector to understand the relationship between caribou and habitat described using forest inventory data, our goal was to leverage caribou Global Positioning System (GPS) data collected between 1998 and 2015 to describe habitat selection in west-central Alberta using the AVI. To avoid some of the drawbacks of using AVI data (i.e., decreased accuracy across larger geographic scales, combining multiple datasets of different ages, and lower classification accuracy within mixed and broadleaf stands (McDermid et al. 2009)), we focused our study within a conifer-dominated 3700 km² portion of west-central Alberta that was managed by Weyerhaeuser Co. Ltd. Grande Prairie under a forest management agreement (FMA) with the Government of Alberta, an area containing the winter ranges of two central mountain caribou herds. Our specific objective was to assess whether caribou responded differently to stands defined using AVI attributes stand origin (natural (i.e., fire) versus harvested), regeneration height, and canopy closure. Results of this research will permit forest managers to directly evaluate the impact of harvest management strategies on the probability of caribou occurrence in their FMA and, in doing so, facilitate bridging of the research-implementation gap by emphasizing habitat categories readily utilized by the forest industry.

Because of excessive multicollinearity with other variables (VIF > 3), we excluded distance to tree line and road density from candidate models fitted to data from the Narraway range. We also excluded cutblocks, previously harvested closed canopy coniferous forest, and the “other” categories from Narraway data sets because of the complete absence of “used” observations. For both seasons and herds, according to AICc, the full model including AVI, terrain, and disturbance parameters was the most parsimonious (ω_i = 1) and better at explaining caribou habitat selection than the models including AVI and terrain combined, or models including only terrain, AVI, or disturbance parameters (Tables S3 and S4). As the goal of our study was to describe caribou habitat selection using forest inventory habitat data, only responses of caribou to AVI-derived habitat are described in detail here. Complete model parameters are in Tables S5 and S6.

During early winter, in comparison to the reference category, Redrock–Prairie Creek caribou were more likely to select harvest-origin open pine stands up to 14 m in height, natural-origin closed pine stands up to 14 m in height, pine stands between 14 and 18 m in height regardless of canopy cover or stand origin, wetlands, and anthropogenic features besides cutblocks (Fig. 2). During early winter, in comparison to the reference category, Redrock–Prairie Creek caribou were less likely to select open and closed pine stands >18 m in height regardless of stand origin and were less likely to select closed conifer and mixed deciduous stands, previously harvested open conifer stands, and shrub and herb (Fig. 2). During late winter, in comparison to the reference category, Redrock–Prairie Creek caribou were less likely to select open conifer stands, anthropogenic features other than cutblocks, and open pine stands regardless of stand origin or height. During early and late winter, Redrock–Prairie Creek caribou were also more likely to select areas further from roads (early-winter OR 1.23; late winter OR 1.37), areas further from the tree line (early-winter OR 1.77; late winter OR 1.22), and areas at higher elevations, with the latter especially important during early winter (early-winter OR 4.30; late winter OR 1.62; Table S5).

During early winter, in comparison to the reference category, Narraway caribou were more likely to select natural-origin open pine stands up to 14 m in height and open pine stands between 14 and 17 m in height, but were less likely to select harvest-origin pine stands <14 m in height, closed pine stands >14 m in height, open pine stands >18 m in height, harvest-origin open conifer stands, and natural-origin closed conifer stands (Fig. 3). During late winter, in comparison to the reference category, Narraway caribou were more likely to select natural-origin open pine stands up to 14 m in height, and regardless of stand origin were more likely to select open pine stands >14 m in height and closed pine stands between 14 and 18 m in height. During late winter, Narraway caribou were less likely to select harvest-origin open pine stands <14 m in height, closed pine stands <14 m in height and >18 m in height, natural-origin closed conifer stands, and harvest-origin open conifer stands (Fig. 3). During early and late winter Narraway caribou were also less likely to select mixed deciduous stands and shrub and herb (Fig. 3). Narraway caribou were also more likely to select areas further from roads (early winter OR 1.04; late winter OR 1.56) and streams (early winter OR 1.32; late winter OR 1.04), areas with lower densities of cutblock edges (early winter OR 0.22; late winter OR 0.19), areas at high elevations during early winter (OR 1.59), and areas at lower elevations during late winter (OR 0.94; Table S6). K-fold cross-validation indicated strong predictive power of final models (Redrock–Prairie Creek early winter mean r_s 0.981, late winter mean r_s 0.982; Narraway early winter mean r_s 0.931, late winter mean r_s 0.898; Table S7).

To effectively bridge the research–implementation divide, practitioners need data and analyses that can be directly translated into management actions. Across the boreal forest of Canada the resource sector, including forestry, is under pressure to maintain or restore ecosystem function for caribou (Environment Canada 2012, 2014), yet few results are available that directly link caribou habitat use to fine-scale forest inventory data. By combining multi-year caribou GPS and forest inventory data we found that stand origin, tree height, tree species, and canopy cover influenced habitat use by caribou at the coarse-scale during winter, but that habitat use differed between herds, and changed across the winter season. Generally, caribou preferred conifer stands, specifically open canopy stands (canopy cover < 50%), and during early winter shorter, open and closed pine stands were preferred over open conifer stands. Our results, combined with previous research (Leblond et al. 2011; DeCesare et al. 2012), suggest that caribou select a mosaic of stand types during the winter season, which may be mediated by trade-offs by caribou to enhance access to winter forage, and reduce spatial overlap with predators (McLoughlin et al. 2010; DeCesare et al. 2014). However, our study is also unique, because by using forest inventory data combined with 17 years of caribou GPS data, our analysis revealed additional insights into the stand attributes selected by caribou, which were not apparent from previous work in our study area (DeCesare et al. 2012; Johnson et al. 2015).

During winter, in accordance with previous research, we found that caribou generally selected older and moderate to open canopy cover stands, and they avoided mixed stands and early seral habitat (e.g., Fortin et al. 2008; Leblond et al. 2011; DeCesare et al. 2012; Losier et al. 2015), likely to reduce their predation risk (Courbin et al. 2009; Ehlers et al. 2016) and maximize their access to terrestrial and arboreal lichens, important winter forage (Vitt et al. 1988; Coxson and Marsh 2001; Szkorupa and Schmiegelow 2003). However, we also found caribou selected pine stands regardless of stand age or canopy cover. It is possible that by selecting open and closed pine stands throughout the winter, caribou are mitigating their exposure to predation by wolves, which may also select open lichen forest (Leblond et al. 2011). Caribou may also be altering their habitat use to maximize access to forage in response to fine-scale weather patterns (i.e., snowfall) and increasing snow accumulation throughout the winter as denser canopy stands intercept more snow than open canopy stands (Telfer 1970; Schaefer and Pruitt 1991), and snow impedes caribou movement and cratering ability (Johnson et al. 2001; Szkorupa 2002).

In addition to canopy cover, we also found that caribou response to forest stands was dependent on stand origin (i.e., natural (fire) versus harvested) and stand height, and differed across herds and between seasons. For example, during early winter, we found that caribou from both herds selected open pine stands 0–14 m in height but Redrock–Prairie Creek caribou selected harvest-origin stands, whereas Narraway caribou selected natural-origin stands. Yet with increasing canopy height, caribou in both herds selected open pine stands regardless of stand origin. For Redrock–Prairie Creek caribou, these selection patterns are consistent with findings describing availability of terrestrial lichen in fire-origin versus harvest-origin stands. Initially, after stand disturbance, winter-harvested stands have more terrestrial lichen cover than fire-origin stands (Webb 1998; Coxson and Marsh 2001; Lafleur et al. 2016), but with time terrestrial lichens increase linearly after fire, whereas in harvested stands lichens may decrease approximately 2 years after harvesting (Waterhouse et al. 2011), and abundance can remain low in harvested stands until they reach 30 years old, with peaks in stands between 50 and 100 years old (Coxson and Marsh 2001). Why this pattern was not apparent for Narraway caribou is perhaps because of different availabilities of each habitat type (Table S2), which could influence the availability of terrestrial lichens within each herd.

Although pine stands were generally preferred by Redrock–Prairie Creek caribou during early winter, during late winter we found that Redrock–Prairie Creek caribou preferred natural-origin conifer (i.e., dominated by fir and spruce) stands and avoided open pine stands. In contrast, during late winter Narraway caribou generally selected similar stand types to those selected during early winter. Selection of open conifer (rather than pine) stands during late winter by Redrock–Prairie Creek may be explained by caribou selecting areas with more arboreal lichens when the snow is deep and difficult to crater (Telfer 1970; Schaefer and Pruitt 1991; Szkorupa and Schmiegelow 2003). For Narraway caribou that select spruce stands throughout winter (Saher 2005), fine-scale shifts in foraging and habitat selection between early and late winter may not occur. Nonetheless, our findings using 17 years of data were consistent with previous work using the earliest portion of the GPS dataset (1998–2001; Szkorupa and Schmiegelow 2003; Saher 2005) and further demonstrate the fine-scale behavioural and seasonal differences in habitat use among two herds that can occur within the same designated unit (COSEWIC 2011) and within a small geographic area (e.g., DeCesare et al. 2012; Johnson et al. 2015; Hornseth and Rempel 2016). Moreover, our findings that caribou avoided roads, cutblock edges, and early seral forest stands are also consistent with previous findings and provide further evidence that caribou avoid human disturbance at coarse and fine scales (Schaefer and Mahoney 2007; Polfus et al. 2011; DeCesare et al. 2012; Johnson et al. 2015).

By linking caribou habitat use to forest inventory data, our analysis has provided new and direct links between caribou habitat use and forest stands with different stand origins, species, and canopy cover. Because our motivation for this study was to assess caribou response to stand characteristics described using forest inventory data, we conducted a second-order resource selection function analysis, as previous research found caribou response to forest management was strongest at this scale (DeCesare et al. 2012). Future work assessing caribou habitat use relative to forest inventory data at finer scales (i.e., third order (within the individual home range) and fourth order (patches within the individual home range; Johnson 1980)), perhaps in combination with finer-resolution ecosite data, could provide additional insights into specific stand attributes preferred by caribou and thereby further guide forest management within caribou ranges. Building on previous work with these same herds (Szkorupa 2002; Saher 2005), field data collection at stands used by caribou could also help tease out why particular stands are preferred by caribou over others (sensu Serrouya et al. 2007). Finally, we assessed caribou habitat use during the winter season only, as the seasonal ranges of Redrock–Prairie Creek and Narraway caribou overlap most with forests being managed for timber supply during this period. Caribou alter their habitat use across seasons (Polfus et al. 2011; DeCesare et al. 2012) and dependent on their reproductive status (Leblond et al. 2016). In other study areas where caribou inhabit managed forests throughout their annual range, assessing caribou habitat use relative to forest inventory data across multiple seasons, and relative to calving status, will provide a more complete picture of the associations between forest stand attributes and caribou habitat use.

To translate research into management and conservation action, scientists need to provide practitioners with results that are easy to interpret and implement (Arlettaz et al. 2010; Cook et al. 2013). Our results demonstrating that open and closed pine stands and open canopy coniferous stands are selected by caribou throughout winter should be considered by forest managers when planning the timing and location of harvesting in caribou ranges. In addition, as we found that caribou responded differently to natural-origin versus harvest-origin stands of the same height and canopy cover, it is unclear whether harvesting strategies that emulate natural disturbances are perceived by caribou in the same way as natural disturbances. Further work to gauge caribou response to variation in silviculture treatments, cutblock shape, and size will help to evaluate the conservation value of different ecosystem-based management approaches for caribou.

Although previous work found that forest inventory data were relatively poor at predicting wildlife habitat use or attributes associated with quality habitats (McDermid et al. 2009; Boan et al. 2013), results of model selection suggest that combining AVI data with complementary covariates measured at different scales serves to better explain caribou selection behaviour than AVI alone, and that resulting habitat models exhibit excellent predictive power. We are not suggesting that forest inventory data replace satellite-derived habitat data in species models. Indeed, satellite-derived habitat data feature several advantages, especially over large areas, across jurisdictional boundaries, or in protected areas. Nevertheless, since forest inventory data remain a practical and cost-effective means of describing stand structure, forest practitioners are likely to continue to use forest inventory data for planning and management decisions in the near future. To ease interpretation of research results by forest practitioners, scientists should consider building additional models that include forest inventory data or perhaps summarize their final probability surfaces in terms of forest inventory data post-hoc. As new sources become readily available (e.g., LiDAR), the gap between the data used by researchers and forest practitioners may close. In the meantime, however, tailoring results so they are accessible to the end user would help to better facilitate the translation of research into practice.

The study was funded by Weyerhaeuser Co. Ltd., the Foothills Landscape Management Forum (FLMF), and fRI Research. Weyerhaeuser Co. Ltd. and the FLMF provided funds and helped to conceive the project but had no role in study design, data analysis or interpretation, decision to publish, or preparation of this manuscript. fRI Research provided financial and in-kind support for the study design, data analysis, interpretation, and manuscript preparation. We would like to thank Dan Wismer (GIS Program, fRI Research) for help preparing Fig. 1, Alberta Environment and Parks for their continued support of long-term monitoring of Redrock–Prairie Creek and Narraway caribou herds, and Wendy Crosina (Weyerhaeuser Co. Ltd.) and Wayne Thorp (FLMF) for their support of this project. We would also like to thank the anonymous reviewer whose suggestions helped improve this manuscript.