ASSESSING CUMULATIVE IMPACTS OF FOREST DEVELOPMENT ON THE ABUNDANCE AND DISTRIBUTION OF FURBEARERS by Michael C. Bridger B.Sc., University of Northern British Columbia, 2012 THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE IN NATURAL RESOURCES AND ENVIRONMENTAL STUDIES (BIOLOGY) UNIVERSITY OF NORTHERN BRITISH COLUMBIA April 2015 © Michael C. Bridger, 2015 UMI Number: 1526490 All rights reserved INFORMATION TO ALL USERS The quality of this reproduction is dependent upon the quality of the copy submitted. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if material had to be removed, a note will indicate the deletion. Di!ss0?t&iori P iiblist’Mlg UMI 1526490 Published by ProQuest LLC 2015. Copyright in the Dissertation held by the Author. Microform Edition © ProQuest LLC. All rights reserved. This work is protected against unauthorized copying under Title 17, United States Code. ProQuest LLC 789 East Eisenhower Parkway P.O. Box 1346 Ann Arbor, Ml 48106-1346 Abstract Furbearer populations across the central-interior of British Columbia, Canada, are exposed to the cumulative impacts o f landscape change, particularly as a result o f forest harvesting. I elicited knowledge from furbearer experts to develop habitat models for three furbearer species: fisher (Pekania pennanti), Canada lynx (Lynx canadensis), and American marten (Martes americana), and applied the models to reference landscapes to quantify changes in habitat availability and quality from 1990 to 2013. Where forest harvesting was extensive, the models predicted substantial declines in habitat for each focal species. I used trapping records and negative binomial count models to investigate the relationship between habitat change and population abundance of lynx and marten. The top-ranked count models identified combinations o f trapping effort, trapline area, and habitat availability and quality as having significantly positive effects on capture success. These results demonstrate the utility o f expert knowledge for studying cumulative impacts o f landscape change on furbearers. Table of Contents Abstract............................................................................................................................................ii List of Tables................................................................................................................................. vi List of Figures................................................................................................................................ ix Acknowledgements..........................................................................................................................1 Chapter 1: General Introduction.................................................................................................... 2 Background................................................................................................................................. 2 Research Objectives.................................................................................................................... 5 Study Area....................................................................................................................................7 Chapter 2: Assessing the cumulative impacts of forest development on the distribution o f furbearers using an expert-based habitat modeling approach.................................................... 10 Introduction................................................................................................................................11 Methods...................................................................................................................................... 14 Identification of Experts........................................................................................................14 Elicitation o f Expert Knowledge.......................................................................................... 18 Mapping Habitat....................................................................................................................21 Map Variation and Validation..............................................................................................24 Results........................................................................................................................................24 Identification of Focal Species.............................................................................................24 Identification of Habitat Variables...................................................................................... 25 Evaluation of Habitat Variables...........................................................................................27 Quantification o f Habitat Change.........................................................................................27 Map Validation.................................. f . ................................................................................. 30 Discussion..................................................................................................................................32 Habitat Change...................................................................................................................... 36 Management Implications.................................................................................................... 38 Conclusion................................................................................................................................. 41 Chapter 3: Assessing cumulative impacts of forest development on the abundance of furbearers using harvest records.................................................................................................. 43 Introduction............................................................................................................................... 44 Methods..................................................................................................................................... 47 Data Collection......................................................................................................................47 Model Development............................................................................................................. 48 Model Selection.....................................................................................................................50 Model Prediction...................................................................................................................51 Catch per Unit Effort............................................................................................................ 51 Results........................................................................................................................................52 Catch per Unit Effort............................................................................................................ 52 Count M odels........................................................................................................................ 52 Model F it............................................................................................................................... 57 Discussion................................................................................................................................. 58 Influences on Capture Success.............................................................................................60 Catch per Unit Effort............................................................................................................ 63 Habitat Change and Population Dynamics o f Furbearers..................................................64 Conclusion................................................................................................................................. 66 Chapter 4: General Summary and Management Considerations...............................................68 Summary....................................................................................................................................68 Modeling and Predicting Habitat Change........................................................................... 69 Population Modeling.............................................................................................................71 Management Concerns and Recommendations......................................................................72 Fisher......................................................................................................................................73 Lynx........................................................................................................................................76 Marten.....................................................................................................................................78 Furbearers in General............................................................................................................81 Research Conclusions...............................................................................................................84 References................................................................................................... 86 Appendix A: Survey conducted by candidate trappers to assess their suitability as experts for the subsequent development of expert-based habitat models..............................................94 Appendix B: Examples o f surveys conducted by biologist and trapper experts for the puipose of identifying the focal species and habitat variables, and evaluating habitat variables for the development of expert-based habitat models.................................................. 97 Appendix C: Description of the classification of habitat variables used to construct expert-based habitat models for each focal species...................................................................112 Appendix D: Mean eigenvector scores (representing relative importance) resulting from the expert evaluation of fisher, lynx, and marten habitat variables..........................................116 Appendix E: Examples of expert-based habitat maps, providing a spatial representation of change in the availability and quality of habitat from 1990 to 2013................................... 120 Appendix F: Candidate a priori model selection for predicting lynx and marten captures in the West and East Study Areas across central-interior BC, Canada.................................... 124 Appendix G: Coefficients and statistical parameters generated from the top ranked negative binomial regression models for die prediction o f lynx and marten captures in the West and East Study Areas across central-interior BC, Canada...............................................132 Appendix H: Difference in observed from predicted fur harvest records generated using negative binomial count models for lynx and marten from the West and East Study Areas across central-interior BC, Canada. A value o f zero suggests perfect prediction while negative values suggest over-prediction and positive values suggest under prediction. 138 List of Tables Table 2.1. Criteria for the recruitment o f suitable biologist and trappers experts for the development o f expert-based habitat models. These criteria served as a guide during the peer-referral phase o f expert identification............................................... 16 Table 2.2. Example of an AHP matrix that evaluates preferred topographic elevation by a theoretical species. For example, <500 m compared to itself is of equal importance, represented by a score o f ‘ 1’, 1000 m -1500 m is very strongly more important than <500 m, represented by a score of ‘7’, and >2000 m is moderately less important than <500 m, represented by a score of ‘ 1/3’. All shaded columns are the inverse of their respective scores. The outputs o f each AHP matrix are eigenvector scores, representing the relative value o f each habitat class.....................................................................................................................21 Table 2.3. Scoring scheme used by experts for the pairwise comparisons of habitat variables......................................................................................................................... .21 Table 2.4. List of habitat variables and spatial data sources used for the development of expert-based habitat models.......................................................................................... 22 Table 2.5. Change in availability of habitat for fisher, lynx, and marten from 1990 to 2013 in the West and East Study Area across central-interior BC, Canada. The percentages represent the composition of habitat across the study area in 1990 and 2013. The upper and lower 95% values represent the variation around the mean habitat change when the maps were constructed using the upper and lower 95th percentile eigenvector scores................................................................................ 29 Table 3.1. Variables used for the development of count models for predicting capture success of lynx and marten across central-interior BC, Canada................................49 Table 3.2. Summary o f model selection statistics for the candidate count models to predict lynx captures in the West Study Area (Capture events; N = 75) across centralinterior BC, Canada. The models were developed from six a priori categories of explanatory hypotheses. The top models from each category were selected and then ranked against each other............................................................................... 54 Table 3.3. Summary of model selection statistics for the candidate count models to predict lynx captures in the East Study Area (Capture events; N = 27) across centralinterior BC, Canada. The models were developed from six a priori categories of explanatory hypotheses. The top models from each category were selected and then ranked against each other...................................................................................... 55 Table 3.4. Summary o f model selection statistics for the candidate count models to predict marten captures in the West Study Area (Capture events; N = 86) across centralinterior BC, Canada. The models were developed from six a priori categories of explanatory hypotheses. The top models from each category were selected and then ranked against each other..................................................................................... 57 Table 3.5. Summary o f model selection statistics for the candidate count models to predict marten captures in the East Study Area (Capture events; N = 47) across centralinterior BC, Canada. The models were developed from six a priori categories of explanatory hypotheses. The top models from each category were then ranked against each other.......................................................................................................... 58 Table 4.1. Recommendations by biologist and trapper experts for maintaining habitat and numbers of fisher in the central-interior of BC, Canada. Recommendations were obtained through semi-structured interviews and surveys.......................................... 74 Table 4.2. Recommendations by biologist and trapper experts for maintaining habitat and numbers of lynx in the central-interior of BC, Canada. Recommendations were obtained through semi-structured interviews and surveys......................................... 77 Table 4.3. Recommendations by biologist and trapper experts for maintaining habitat and numbers of marten in the central-interior of BC, Canada. Recommendations were obtained through semi-structured interviews and surveys.......................................... 79 Table 4.4. Recommendations by biologist and trapper experts for maintaining habitat and numbers of the broader group of furbearers in the central-interior of BC, Canada. Recommendations were obtained through semi-structured interviews and surveys.............................................................................................................................82 Table C .l. Descriptions of the subclasses, levels, or categories o f habitat variables included in the habitat models for fisher, lynx, and marten across central-interior BC, Canada................................................................................................................... 113 Table F.l. Candidate a priori models used to select the most parsimonious count model for understanding captures of lynx and marten by trappers in the West and East Study Areas across central-interior BC, Canada........................................................ 125 Table G .l. Coefficients and statistical parameters generated from the top ranked negative binomial regression models for the prediction of lynx captures in the West Study Area across central-interior BC, Canada..........................................................135 Table G.2. Coefficients and statistical parameters generated from the top ranked negative binomial regression models for the prediction o f lynx captures in the East Study Area across central-interior BC, Canada....................................................................... 136 vii Table G.3. Coefficients and statistical parameters generated from the top ranked negative binomial regression models for the prediction o f marten captures in the West Study Area across central-interior BC, Canada............................................................137 Table G.4. Coefficients and statistical parameters generated from the top ranked negative binomial regression models for the prediction o f marten captures in the East Study Area across central-interior BC, Canada...................................................................... 137 List of Figures Figure 1.1. Location of reference landscapes (i.e., registered traplines) for the application o f habitat models and population analysis for fisher, lynx, and marten across central-interior BC, Canada. Cumulative impacts in the form o f forestry development were extensive on the dark-shaded traplines (West Study Area) during the study period, while the hash-marked traplines (East Study Area) had much less harvesting........................................................................................................ 7 Figure 2.1. The peer-referral method used for identifying biologist experts for the development of expert-based habitat models. The first round o f peer-referral required the identification of seed experts, followed by two rounds o f further nominations of experts................................................................................................... 17 Figure 2.2. The peer-referral method used for identifying trapper experts for the development of expert-based habitat models. Seed experts identified candidates that met specific criteria and were selected based on their suitability for the study.................................................................................................................................18 Figure 2.3. Combined scores (mean and standard error) from biologists and trapper experts to include variables for habitat models for fisher, lynx, and marten. The inclusion threshold was set at a score o f 2................................................................................... .26 Figure 2.4. Predicted habitat for lynx on one example trapline in the West Study Area, central-interior BC, Canada, during 1990 and 2013................................................... .28 Figure 2.5 Change in the area (hectares) of four habitat classifications (1990-2013) for expert-based habitat models for fisher, lynx, and marten developed by biologist, trapper, and combined expert models applied to the West and East Study Areas across central-interior BC, Canada............................................................................... 31 Figure 2.6. Mean accuracy (± SE) scores for the habitat maps for each focal species in the West (unshaded) and East (shaded) Study Areas. Trappers evaluated the distribution and area of ranked habitat for their traplines and provided scores representing perceived accuracy from 0-10.................................................................32 Figure 3.1. Scatterplots and linear regressions displaying catch per unit effort (CPUE) and trapping effort for lynx captures in the West Study Area (a) and marten captures in the West (b) and East Study Area (c) across central-interior BC, Canada. The CPUE represents the number o f lynx or marten captured by each trapper in a given year divided by their effort level.........................................................................53 Figure 3.2. Model coefficients and 95% confidence intervals o f top-ranked models (where A AICc was < 2.0) representing influences on capture success o f lynx in the West (a) and East (b) Study Areas across central-interior BC, Canada...............................56 Figure 3.3. Model coefficients and 95% confidence intervals o f top-ranked models (where A AICc was < 2.0) representing influences on capture success o f marten in the West (a) and East (b) Study Areas across central-interior BC, Canada..................................59 Figure D .l. Mean (95% confidence intervals) eigenvector scores (representing relative importance) resulting from the expert evaluation o f variables describing attributes o f fisher habitat............................................................................................ 117 Figure D.2. Mean (95% confidence intervals) eigenvector scores (representing relative importance) resulting from the expert evaluation o f variables describing attributes o f lynx habitat.............................................................................................. 117 Figure D.3. Mean (95% confidence intervals) eigenvector scores (representing relative importance) resulting from the expert evaluation o f variables describing attributes o f marten habitat.......................................................................................... 119 Figure H .l. Difference in observed from predicted fur harvest records generated using the top ranked negative binomial count models E+TA+GVGH+EMT (a) and E+TA+GVGH (b) for lynx from the West Study Area across central-interior BC, Canada. A value o f zero suggests perfect prediction while negative values suggest over-prediction and positive values suggest under-prediction.................... 139 Figure H.2. Difference in observed from predicted fur harvest records generated using the top ranked negative binomial count models E+TA+EMT (a) and E+TA+MMT (b) for lynx from the East Study Area across central-interior BC, Canada. A value of zero suggests perfect prediction while negative values suggest over-prediction and positive values suggest under-prediction....................................................................... 140 Figure H.3. Difference in observed from predicted fur harvest records generated using the top ranked negative binomial count models E+TA+SH+EMT (a), E+TA+SH+MMT (b), and E+TA+SH (c) for marten from the West Study Area across central-interior BC, Canada. A value of zero suggests perfect prediction while negative values suggest over-prediction and positive values suggest underprediction..........................................................................................................................141 Figure H.4. Difference in observed from predicted fur harvest records generated using the top ranked negative binomial count model E+TA+GVGH for marten from the East Study Area across central-interior BC, Canada. A value o f zero suggests perfect prediction while negative values suggest over-prediction and positive values suggest under-prediction Acknowledgements Funding for this project was provided by the Habitat Conservation Trust Foundation. The original concept for this research was proposed by two o f my supervisory committee members, Dr. Chris Johnson and Dr. Mike Gillingham. They were instrumental in the development and implementation of this research project. I would also like to thank the third member of my supervisory committee, Eric Lofroth. His strong background in furbearer ecology was very valuable to this research, and he provided great input, enthusiasm, and time commitments throughout the study. Chris Johnson’s unwavering determination to achieve a successful research product was most greatly appreciated. As the supervisor for this study, Chris provided countless hours of support to ensure its success. From aiding in the development of appropriate research methodologies to the numerous hours of editing, Chris’ contributions to this project were invaluable. His role as a mentor and confidant was equally important, and for that I am very grateful. It was an honour and pleasure to work with Mike Gillingham throughout the study. Mike has been a great mentor to me throughout my time at UNBC. His ability to think critically was instrumental in identifying research strengths and weaknesses overlooked by others. I feel very fortunate to have been able to work alongside both Chris and Mike. I would like to thank Mike Morris and the BC Trappers Association for bringing the issue of cumulative impacts on furbearers to the attention of UNBC. The successful completion o f this research could not have been achieved without the help o f the amazing expert participants. I would like to thank the following experts for their input, time commitments, and patience during the tedious surveys: Biologist experts - C. Apps, S. Crowley, L. Davis, D. Hatler, D. Hodder, T. Kinley, R. Klafki, G. Mowat, K. Poole, D. Steventon, R. Weir, and Trapper experts - D. Bell, B. Frederick, B. McKay, B. Monroe, M. Richards, W. Sharpe, G. Smith, G. Tencarre, E. Voelk, D. Wilkins. My time spent at UNBC made the completion of my research thesis that much more enjoyable. The relationships developed with friends and faculty at the university went a long way towards achieving a successful project. I will be forever grateful for the support o f my family, friends and office mates. They have all been a huge support system for me, and I am thankful for their ongoing enthusiasm and encouragement. l Chapter 1: General Introduction Background Natural resource managers must consider the importance o f the cumulative impacts of anthropogenic and natural landscape change when managing wildlife habitat and populations (Schneider et al. 2003, Johnson 2011). Cumulative impacts can be complex and difficult to understand when compared to acute, immediate changes to habitat. They can be a result o f natural or anthropogenic activities at a range of temporal and spatial scales and they can be interactive, additive, or synergistic in nature (Nitschke 2008, Johnson 2011). Rapid resource extraction can result in cumulative impacts that affect economic, cultural, and ecological resources including the distribution and abundance o f wildlife (Nitschke 2008). Species that are dependent on continuous tracts o f late-seral habitat may be particularly susceptible to habitat loss and fragmentation associated with cumulative landscape change (Schneider et al. 2003, Nitschke 2008). Conversely, generalist species may increase as a result o f landscape changes that create openings, young forests, and edge habitat. Furbearer populations found across the central-interior of British Columbia (BC), Canada (Figure 1.1), are exposed to the cumulative impacts of landscape change. This region has been subjected to unprecedented levels o f timber harvesting following the rapid development of the forestry sector. Levels o f timber harvest were historically high, but have accelerated over the past ten years in response to a mountain pine beetle (Dendroctonus ponderosae) epidemic that has killed 53% of merchantable pine stands province-wide (BC Ministry of Forests, Mines, and Lands 2010). Natural disturbance, including fire, insects, disease outbreak, and wind events may also reduce the availability of habitat for furbearers. The cumulative impacts o f anthropogenic and natural disturbance are especially concerning 2 for some species of furbearers thought to be sensitive to human disturbance and the loss of late successional forests. Furbearers play important roles within the ecosystem and have significant cultural and economic importance for the fur-trapping industry throughout North America (Webb and Boyce 2009). Local trapping organizations have reported declines in the abundance o f a number o f furbearer species across central-interior BC (M. Bridger unpub. data), yet there is a lack of monitoring and management in response to those observations (Webb et al. 2008). There is an immediate need to investigate the long-term impacts of cumulative landscape change on the habitat and abundance of furbearer populations. Many furbearer species act as indicators o f healthy ecosystems (Buskirk 1992, Wiebe et al. 2013). American marten (Martes americana) and fisher (Pekania pennanti), for example, are often associated with intact, late-successional forests (Thompson and Colgan 1994, Payer and Harrison 2003, Weir and Harestad 2003, Proulx 2006, Proulx 2011). These forests provide structural complexity and other attributes that are vital to the persistence of furbearers and other old-growth dependent species (Buskirk 1992, Payer and Harrison 2003, Proulx 2006). Other furbearer species, like Canada lynx (Lynx canadensis), may thrive across landscapes where different stages of successional forests are present; old-growth forests may provide habitat for denning and resting, while regenerating forests support prey (Hoving et al. 2004). The habitat requirements and disturbance responses of furbearers to landscape change are not well quantified, presenting challenges to natural resource professionals. Biologists and forest managers, however, recognize that many furbearer species avoid forest openings 3 associated with industrial activity and habitat disturbance (Thompson and Colgan 1994, Hargis et al. 1999, Potvin et al. 2000, Fuller and Harrison 2005, Proulx 2009). Furthermore, fragmentation o f habitat at larger spatial scales may affect population productivity (Payer and Harrison 2003). The reduction and fragmentation of old forest across landscapes may have profound long-term effects on the abundance and persistence o f furbearer populations (Thompson 1994, Proulx 2000). Expert-based wildlife studies may serve as an alternative to empirical-based research, particularly when the cryptic behaviour o f many furbearer species makes them difficult to study (Ruette et al. 2003). Where empirical data are lacking, expert knowledge can provide useful insights on cumulative impacts and their influence on the distribution and abundance of furbearers. Expert-based studies have faced skepticism in the past due to the inherent difficulties in assessing the variability, uncertainty, and accuracy of expert knowledge (Drescher et al. 2013). Expert-based wildlife studies, however, can achieve scientific credibility through the application of rigorous and repeatable methods. Thus, there has been an increase in the use and acceptance of expert knowledge in ecological studies over the past 20 years (Drescher et al. 2013). When elicited effectively, expert knowledge can be used to parameterise predictive habitat models (Store and Kangas 2001, Johnson and Gillingham 2004, O ’Neill et al. 2008, Burgman et al. 201 lb, Johnson et al. 2012). Additionally, expert-based approaches are effective for collecting and evaluating harvest data, such as trapping records, which can be used to quantify and explore population change over time (Erickson 1982, Raphael 1994). For example, consultations with fur trappers can allow researchers to control for factors known to influence capture success, such as trapping effort. By accounting for such factors, 4 trapping data can be used to determine the impacts of habitat change on population abundance. Research Objectives The primary goal of my research was to understand of the influence of cumulative landscape change on the availability of habitat and, ultimately, the abundance o f furbearer populations found across the central-interior of BC. To meet that goal, I developed and addressed the following three research objectives: 1) Develop expert-based habitat models for fisher, lynx, and marten, and map changes in habitat availability and quality from 1990 to 2013 (Chapter 2). I elicited expert knowledge from furbearer biologists and trappers that described the habitat relationships of the three focal species. I used that knowledge to develop habitat models that were applied to ten registered traplines that served as reference landscapes. The experts provided input throughout the study, including the selection of focal species and the identification and evaluation o f habitat variables that were included in the models. I applied the habitat models to the reference landscapes at four time intervals (i.e., 1990, 2000, 2005, and 2013) to quantify temporal changes in habitat availability and quality. I tested the utility of the expert-based modeling approach by quantifying uncertainty and variation in the expert responses. 2) Use trapping records to quantify the harvest, and ultimately population changes, of two furbearer species following changes in the availability and quality o f habitat across the reference landscapes (Chapter 3). I collected trapping records, dating back to 1990, from ten trappers for lynx and marten. I used negative binomial count 5 models to investigate the factors hypothesised to influence capture success, including habitat availability and quality, trapline size, trapping effort, and climatic variables. By controlling for confounding variables, I was able to investigate the effect of habitat change on the population abundance o f lynx and marten, while examining the utility of trapping records as a measure o f abundance. 3) Identify important issues and concerns, as expressed by trappers and biologists, in the management of furbearer habitat and populations, and develop recommendations to address the impacts o f cumulative habitat change for those species (Chapter 4). I conducted semi-structured interviews with furbearer biologists and trappers to discuss the management of fisher, lynx, marten, and the broader group of furbearers within BC. The interviews guided experts to identify key concerns regarding the management o f furbearer habitat and populations, and subsequently to provide recommendations in response to those concerns. Taken together, results of this study provide a broader understanding o f the habitat ecology of furbearers in BC, including their response to landscape change. This understanding is a prerequisite for sustaining, restoring or enhancing furbearer habitat. The study process increased the participation of stakeholders in both the extension o f research results and the management o f furbearer habitat. Finally, I investigated the utility o f expertbased methods for understanding habitat and population change. A transparent and rigorous process, continuous expert engagement, and relatively low uncertainty in model results suggested that the methods applied throughout this study were robust and likely suitable for rapid studies o f cumulative habitat change for other species. 6 Study Area Brit sn Colum bia t James' I ra se r^ -^ d e rh o o f Pnnce Gteorge T raplines with High F orestry Developm ent 100 km T raplines with Low F orestry D evelopm ent Figure 1.1. Location of reference landscapes (i.e., registered traplines) for the application of habitat models and population analysis for fisher, lynx, and marten across central-interior BC, Canada. Cumulative impacts in the form o f forestry development were extensive on the dark-shaded traplines (West Study Area) during the study period, while the hash-marked traplines (East Study Area) had much less harvesting. The application of habitat models and the population assessments for the focal species occurred across ten reference landscapes consisting of 17 registered traplines belonging to ten trappers. The reference landscapes were subjected to varying levels o f industrial development: high levels o f forest harvesting (up to 75% of the trapline area harvested in the past 40 years; hereafter referred to as the ‘West Study Area’) and minimal levels of harvesting (up to 11% of the trapline area harvested in the past 40 years; hereafter referred to as the ‘East Study Area’). Intensive salvage logging that occurred in the West Study Area was in response to a mountain pine beetle epidemic that began during the early 2000’s. Increased harvesting across that area resulted in reduced forestry activity in tenures that did 7 not contain a large component of lodgepole pine (Pinus contorta); such areas were found in the East Study Area. There were ten traplines in the West Study Area, registered to six trappers and encompassing an area o f 244,923 ha. There were seven traplines in the East Study Area registered to four trappers; these reference landscapes encompassed a total area of 357,767 ha. The reference landscapes were approximately centered on the city o f Prince George within the central-interior o f BC, Canada (Figure 1.1). In the West Study Area, landscapes occurred primarily within the Sub-Boreal Spruce (SBS) biogeoclimatic zone (Meidinger and Pojar 1991). The climate o f the SBS zone is characterized by severe, snowy winters with mean monthly temperatures below 0° C for the months o f November to February. Summers are warm, moist and short with mean monthly temperatures above 10° C. Mean annual precipitation ranges from 440-900 mm, 25-50% o f which is in the form o f snow. The SBS zone is generally between 1100-1300 m, and is dominated by upland coniferous forests consisting o f hybrid spruce (Picea engelmannii x glauca), subalpine fir (Abies lasiocarpa), lodgepole pine, or Douglas-fir (Pseudotsuga menziesii) on dry, warm sites. Trembling aspen (Populus tremuloides) and paper birch (Betula papyrifera) are common deciduous species. The extensive timber harvesting in this area has resulted in a serai distribution that is skewed towards younger age classes. The SBS zone contains ideal habitat for a variety of furbearers, including marten, fisher, lynx, wolverine {Gulo gulo), and beaver (Castor canadensis), resulting in some of the province’s highest fur-harvest levels. The East Study Area includes portions of the SBS and the Engelmann SpruceSubalpine Fir (ESSF) biogeoclimatic zone (Meidinger and Pojar 1991). The ESSF zone generally occurs above the SBS zone at elevations ranging from 900-1700 m. Low 8 temperatures are common with mean monthly temperatures below 0° C for the months of November to April. Mean annual precipitation may range from 500-2200 mm, with snow accounting for 50-70% o f the total precipitation. Engelmann spruce (Picea engelmannii) and subalpine fir are the dominant climax species, but lodgepole pine is also present. Small areas of interior cedar-hemlock forests are also found within the East Study Area. 9 Chapter 2: Assessing the cumulative impacts of forest development on the distribution of furbearers using an expert-based habitat modeling approach Abstract: Cumulative impacts of anthropogenic landscape change must be considered when managing and conserving wildlife habitat. Across the central-interior of BC, Canada, industrial activities are altering the habitat of furbearer species. This region has witnessed unprecedented levels o f anthropogenic landscape change following rapid development in a number o f resource sectors, particularly forestry. Our1 objective was to create expert-based habitat models for three furbearer species: fisher (Pekania pennanti), Canada lynx {Lynx canadensis), and American marten (Martes americana) and quantify habitat change for those species. We recruited ten biologist and ten trapper experts and then used the analytical hierarchy process to elicit expert knowledge o f habitat variables important to each species. We applied the models to reference landscapes (i.e., registered traplines) in two distinct study areas and then quantified the change in habitat availability from 1990 to 2013. There was strong agreement between expert groups in the choice o f habitat variables and associated scores. Where anthropogenic impacts had increased considerably over the study period, the habitat models showed substantial declines in habitat availability for each focal species (78% decline in optimal fisher habitat; 83% decline in optimal lynx habitat; and 79% decline in optimal marten habitat). For those traplines with relatively little forest harvesting, the habitat models showed no substantial change in the availability o f habitat over time. These results suggest that habitat for these three furbearer species declined significantly as a result of the cumulative impacts of forest harvesting. Results of this study illustrate the utility o f expert knowledge for understanding large-scale patterns of habitat change over long time periods. 11 used first-person plural to acknowledge co-authorship for the publication of thesis Chapters 2 and 3. 10 Introduction Furbearers have significant cultural and economic importance for fur-trapping communities (Hamilton et al. 1998, Webb and Boyce 2009). Furthermore, many furbearer species act as indicators of healthy ecosystems (Buskirk 1992, Wiebe et al. 2013). For example, American marten (Martes americana) and fisher (Pekania pennanti) are often associated with intact, late-successional forests (Thompson and Colgan 1994, Payer and Harrison 2003, Weir and Harestad 2003, Proulx 2006, Proulx 2011). Other species, like Canada lynx (Lynx canadensis), may thrive where different stages o f successional forests are present; old-growth forests may provide habitat for denning and resting, while regenerating forests support prey species (Hoving et al. 2004). Cumulative landscape change can have profound effects for furbearer habitat and populations, yet in many jurisdictions there is little to no monitoring and research to document these impacts. Furbearers are thought to be susceptible to habitat loss and fragmentation (Soutiere 1979, Thompson and Colgan 1994, Hargis et al. 1999, Potvin et al. 2000, Proulx 2000, Fuller and Harrison 2005, Weir and Almuedo 2010, Weir and Corbould 2010). Across the centralinterior of British Columbia (BC), Canada, levels o f timber harvest were historically high, but have accelerated over the past ten years in response to a mountain pine beetle (Dendroctonus ponderosae) epidemic which has killed 53% of merchantable pine stands (BC Ministry of Forests, Mines, and Lands 2010). The magnitude o f these impacts, relative to the distribution and abundance o f furbearers, is unknown. Although many past studies have investigated fine-scale changes in habitat, few studies have attempted to address such questions at large spatial or temporal scales. 11 Anthropogenic activities, such as forest harvesting, may result in the immediate loss or fragmentation of habitat. Alternatively, the impacts of many developments accumulate over a long timeframe. Compared to acute, immediate changes to habitat, cumulative impacts can be complex and difficult to understand (Johnson 2011). These impacts are the result of natural or anthropogenic processes and events that may accumulate due to changes in environmental and socio-economic systems at varying temporal and spatial scales (Nitschke 2008). In addition to being additive, cumulative impacts can be interactive or synergistic in nature (Nitschke 2008, Johnson 2011). Cumulative landscape change may have profound impacts on the quality and distribution of furbearer habitat that persist for many years (Thompson 1994, Proulx 2000). Species-distribution models can be an important tool for quantifying cumulative changes in the availability or quality of wildlife habitat. These models are typically empirical, relating field observations of a species’ occurrence to environmental variables hypothesised to influence species distribution (Guisan and Zimmerman 2000, Guisan and Thuiller 2005, Johnson et al. 2012). When empirical data are unavailable, expert knowledge can be used to parameterise such models. Experts can formulate model structure, including the identification of important variables, and provide quantitative scores denoting the importance o f each predictor. When elicited effectively, expert knowledge can be a valuable source of information and data for developing and parameterising predictive habitat models and subsequent maps (Store and Kangas 2001, Johnson and Gillingham 2004, O’Neill et al. 2008, Burgman et al. 201 lb, Johnson et al. 2012). Such maps can quantify habitat change over time and space, and aid in management and planning decisions (Johnson et al. 2012). 12 There has been an increase in the use of expert knowledge in ecological studies in the last 20 years (Drescher et al. 2013). Expert knowledge, however, may still face skepticism when compared to empirical research (McBride and Burgman 2012). When developed or implemented poorly, such studies may be biased by the methods used to acquire the expert knowledge (Martin et al. 2011, McBride and Burgman 2012). Furthermore, there are inherent difficulties when assessing the variability, uncertainty, and accuracy o f expert knowledge (Drescher et al. 2013). Wildlife studies incorporating expert knowledge can achieve scientific credibility by adopting rigorous methods that include an unbiased sample of experts, transparent and repeatable elicitation of knowledge, and the quantification of uncertainty (Burgman et al. 201 la, Johnson et al. 2012, McBride and Burgman 2012, Drescher et al. 2013). When conducted effectively, expert knowledge can serve as an excellent source o f information for studying cumulative impacts at a large spatial scale, particularly when empirical data are limited. The objective of this study was to use expert knowledge to quantify the cumulative impacts of landscape change relative to the availability and quality of furbearer habitat. We elicited expert knowledge and parameterised species-distribution models representing the habitat o f fisher, lynx, and marten. The models were used to develop a chronology o f maps displaying habitat change since 1990 across traplines subjected to high levels o f forestry development (hereafter referred to as the ‘West Study Area’) and traplines subjected to minimal levels of forestry development (hereafter referred to as the ‘East Study Area’; Figure 1.1). We evaluated the utility and consistency o f knowledge from two groups o f experts with different domains o f expertise: trappers with intimate knowledge of their trapping areas and furbearer biologists with a potentially broader perspective on the habitat ecology o f the focal 13 species. We hypothesised that the distribution and quality of furbearer habitat would decline following rapid and extensive cumulative landscape changes. Methods We used expert-based habitat models to map and quantify habitat change for three focal furbearer species. We recruited experts from two distinct groups and elicited expert knowledge that would aid in the development o f habitat models for each species. The models were applied spatially to ten trapline areas that served as reference landscapes and at four time intervals (i.e., 1990, 2000, 2005, and 2013), which represented increasing levels of forest harvesting. Using this approach, we quantified cumulative habitat change over time as well as the uncertainty in those predictions. Identification of Experts We recruited 21 experts from two general categories: professional experts in the form o f furbearer biologists, and expert practitioners in the form o f furbearer trappers. Although there are no specific criteria to identify the appropriate number o f experts for such studies, it is important that the participants represent the knowledge bounds relative to the study objectives and that the sample is large enough to prevent significant bias or error from any one expert (McBride and Burgman 2012). We used peer-referral techniques to identify a collection o f potential candidates from both categories o f experts. Where a population of experts may be difficult to enumerate, peer-referral techniques provided a practical method for identifying appropriate individuals to represent the expert community. 14 Identification o f Biologist Experts We used three rounds o f peer-referral to identify three seed experts, eight subsequent experts, and an advisory expert. The role of the seed experts was to recommend additional experts, who in turn, were asked to nominate further experts. The role o f the advisory expert was to review the suitability and appropriateness o f the research surveys, and provide feedback or possible revisions before submission to the expert groups. We conducted a thorough literature review of furbearer research in BC to identify a group o f candidate biologist experts. We identified one seed expert from each o f the following disciplines: government biologists, academic researchers, and private-consulting biologists. They were selected based on their expertise in the field of furbearer ecology, including their knowledge of, and access to, further experts. We established a set of criteria to ensure candidate experts were credible and qualified (Table 2.1). Drawing from these criteria, the seed experts were asked to provide a list o f furbearer biologists currently working within BC, and subsequently rank the candidates in terms of suitability for the study. We then contacted the top four ranked candidates and invited them to participate in the study. They were then asked to submit a further list and ranking of candidate furbearer biologists (using the set o f criteria for guidance). We recruited the top four ranked experts in the second round o f referral, providing us with a total of ten professional experts (including two o f the seed experts; Figure 2.1). We then identified the advisory expert from the list o f candidates. 15 Table 2.1. Criteria for the recruitment of suitable biologist and trappers experts for the development of expert-based habitat models. These criteria served as a guide during the peer-referral phase o f expert identification. General Criteria Thresholds Explanation o f Criteria Biologist Experts Years of direct experience with furbearer ecology >5 years General research experience with furbearer species including habitat use, behaviour, and/or landscape change. Number of relevant publications/reports >3 publications Peer-reviewed or relevant grey literature. In-depth knowledge of specific species biology >1 species Focused research on one or more furbearer species. Location o f past research BC Knowledge specific to British Columbia; ideally, knowledgeable of furbearer habitat and populations found in central-interior BC. >15 years Total years o f trapping furbearers native to North America. Number of continuous years trapping on current trapline >10 years Specific local knowledge of trapping landscape, habitat features, habitat use, habitat change, etc. Location o f current trapline Centralinterior BC Current trapline located in north-central BC. Extent of personal trapping records >10 years Detailed records o f captures, effort level, etc. Levels o f landscape change on trapline Casedependent Forms and extent o f past or present landscape change; may range from very little to extensive change across trapline. T rapper Experts Number of years trapping 16 Id e n tify one s e e d e x p e rt fro m e a c h o f th e fo llo w in g d iscip lin es: g o v e rn m e n t b io lo g ists, acad em ic researc h ers, a n d p riv a te co n su ltin g b io lo g ists Round 1 G o v ern m en t b io lo g is t se e d ex p e rt A cad em ic re se a rc h e r se e d ex p e rt P riv ate c o n s u lta n t b io lo g is t ex p e rt E a ch se e d e x p e rt s u b m itte d a list a n d ran k in g o f c a n d id ate ex p e rts. T o p fo u r ca n d id ate s w e re se le c te d RoundZ T o p ra n k e d e x p e rt 1 T o p ra n k e d e x p e rt 3 T o p ra n k e d ex p e rt 2 T o p ra n k e d ex p e rt 4 E a ch o f th e fo u r to p ra n k e d ex p e rts s u b m itte d a lis t a n d ran k in g o f fu rth e r c a n d id a te ex p erts. T o p fo u r ca n d id ate s w e re sele cted Round3 T o p ra n k e d e x p e rt 5 T o p ra n k e d ex p ert 6 T o p ra n k e d e x p e rt 7 T o p ra n k e d ex p e rt 8 T o tal o f te n p ro fe ssio n a l ex p e rts; an a d d itio n a l c a n d id ate w a s se le c te d a s a c o n su ltin g e x p e rt Figure 2.1. The peer-referral method used for identifying biologist experts for the development of expert-based habitat models. The first round of peer-referral required the identification of seed experts, followed by two rounds of further nominations o f experts. Identification o f Trapper Experts We sought guidance from the former president of the British Columbia Trappers Association who identified two seed experts that had knowledge of, and access to, many suitable candidate experts, in this case local experts. As with the biologists, we established a set o f criteria (Table 2.1) to be used as a reference when selecting trappers. Applying these criteria, the seed experts submitted a list of qualified trappers to participate in the study. Because traplines were to be used as reference landscapes for modeling habitat, it was critical to identify trappers with traplines that were located within the proposed study area. We conducted a short survey (Appendix A) and ranked each trapper in terms o f their ability to 17 meet the study objective. O f the eight top-ranked candidates, all agreed to participate (Figure 2 .2 ). Two seed experts were identified Seed experts provided a list of candidate trappers based on established criteria Candidates were contacted and completed a survey to assess whether they met the established criteria. Candidates were subsequently ranked based on potential suitability Consulted with seed experts to verify the suitability of the candidates for the study Recruitment of eight top-ranked trappers for a total of ten experts (including two seed experts) Figure 2.2. The peer-referral method used for identifying trapper experts for the development of expert-based habitat models. Seed experts identified candidates that met specific criteria and were selected based on their suitability for the study. Elicitation of Expert Knowledge Prior to, and throughout the elicitation process experts were made aware of the time commitments expected, the type o f information to be elicited, and how that information would be used. During the elicitation, we met with individual trapper experts in person or via telephone to conduct surveys. Due to travel and time constraints, all surveys with biologists were conducted via email and portable document format (PDF) forms. Prior to conducting a survey, we submitted the set of proposed questions and topics to the advisory expert for review. 18 Identification o f Focal Species The first step in the elicitation process required the experts to complete a survey identifying three focal furbearer species for the study (Appendix B). The experts were asked to identify those species that were most sensitive to landscape change resulting from anthropogenic activities. We tallied the number of selections for each species from the biologist experts and the trapper experts separately to observe the agreement within and between groups. The expert responses were then combined and totaled and we identified the three species with the greatest number of selections. Identification o f Habitat Variables We conducted a thorough literature review and identified 18 candidate habitat and disturbance variables (Appendix B) hypothesised to influence the distribution o f the focal species. Experts were asked to provide a score from 0-4 for each variable, representing its relative importance in contributing to habitat for each focal species (Appendix B). Experts were also asked to provide a confidence score on a scale from 1-10, representing the confidence that the expert had in assessing each variable. We used a Wilcoxon rank-sum test to measure the significance of differences in scores between expert groups (Stata, ver. 12.1, StataCorp, 2011). Few differences occurred, thus the expert scores were combined. We established a model inclusion threshold o f ‘2’; any variable meeting this threshold was considered to be at least moderately important and was included in the respective habitat model. All habitat variables that were included in the models for each species were further classified into subclasses, levels, or categories (Appendix C). The classifications were based on previous literature, or on-the-ground measurements. 19 We used the analytical hierarchy process (AHP; Saaty 1977) to develop the expertbased habitat models (Appendix B). Analytical hierarchy process is a decision making process in which experts provide pairwise comparisons of the relative importance o f two habitat criteria. A form of multi-criteria evaluation, AHP provides structure to the elicitation process so that each variable is evaluated consistently across all participating experts (Johnson et al. 2012). Experts provided pairwise scores, on a nine-point scale, o f the relative importance of every possible combination of habitat variables and associated subclasses of those variables (Tables 2.2, 2.3). Before scoring the variables, experts were provided with detailed instruction on the AHP protocols and examples of AHP matrices. We also provided a handout containing photographic examples of different classifications o f most habitat variables (Appendix B). This was used to aid the experts in visualizing the features during the elicitation process, and increase consistency across expert responses. Experts also provided confidence scores for each completed matrix. After the elicitation process, we calculated eigenvector values and consistency ratio scores (Microsoft Excel, ver. 14.0, Microsoft Corporation, 2010). The eigenvector values represented the relative importance o f each classification of habitat variables (Saaty 1977). Consistency ratios, which tested the probability that the matrices were randomly generated (Saaty 1977), were used to assess the matrices for operative errors. Initially, we kept the biologist and trapper responses separate and conducted a Wilcoxon rank-sum test to examine the differences in eigenvector scores between expert groups. We then combined the expert scores for both groups and calculated the mean eigenvector scores, along with the standard error and 95% confidence intervals. We used the mean eigenvector scores (representing the relative value of each habitat variable) to build the expert-based habitat models. 20 Table 2.2. Example of an AHP matrix that evaluates preferred topographic elevation by a theoretical species. For example, <500 m compared to itself is of equal importance, represented by a score of ‘ 1’, 1000 m -1500 m is very strongly more important than <500 m, represented by a score of ‘7’, and >2000 m is moderately less important than <500 m, represented by a score of ‘ 1/3’. All shaded columns are the inverse o f their respective scores. The outputs o f each AHP matrix are eigenvector scores, representing the relative value of each habitat class. <500 m <500 m 500 m -< 1000 m 1000 m-<1500 m 1500 m -<2000m >2000 m 500 m <1000m 1000 m <1500m 1500 m <2000 m 1 5 3 1/5 1 1/3 1/7 1 1/5 1 3 7 5 1/3 >2000 m 1 Table 2.3. Scoring scheme used by experts for the pairwise comparisons o f habitat variables. Negative Values Positive Values 1 = Equal importance 3 = Moderately more important 5 = Strongly more important 7 = Very strongly more important 9 = Extremely strongly more important 1 = Equal importance 1/3 = Moderately less important 1/5 = Strongly less important 1/7 = Very strongly less important 1/9 = Extremely strongly less important M apping H abitat We used Geographic Information Systems (GIS) to develop a chronology o f maps showing habitat change across the ten reference landscapes/traplines at four time intervals: 1990, 2000, 2005, and 2013 (ArcGIS ver. 10.1, ESRI Inc., 2012). Habitat variables were represented by a number o f spatial data sources (Table 2.4). Primarily, we used a broadscale, forest-cover layer, the BC Vegetation Resource Inventory (VRI; DataBC Distribution Service), which contained numerous forest attributes used for the elicitation process (Appendix C). The VRI data were unavailable for 10% of one trapline and 25% o f another; 21 these areas were removed from the analysis. All large water-bodies were also removed from the analysis. Table 2.4. List of habitat variables and spatial data sources used for the development of Habitat Variable Spatial Data Source Canopy Cover, Coarse Woody Debris, Forest Stand Age, Forest Stand Density, Ground Shrub Cover, Leading Tree Species, Structural Complexity Vegetation Resource Inventory - DataBC Distribution Service (http://www.data.gov.bc.ca/) Cutblock Age Vegetation Resource Inventory and Forest Tenure Cutblock Polygons - DataBC Distribution Service (http://www.data.gov.bc.ca/) Forest Fire Age Fire Perimeters - Historical - DataBC Distribution Service (http://www.data.gov.bc.ca/) Habitat Connectivity GIS-derived variable Proportion o f Landscape Harvested GIS-derived variable We used multiple data sources to represent the history o f forest harvesting across the study landscape. We retrieved the cutblock (logging) polygons from the VRI layer (DataBC Distribution Service) and the Forest Tenure Cutblock layer (DataBC Distribution Service). All polygons within the cutblock layer that had a projected age >40 years old (relative to the map date), basal area >10 m2/ha, and canopy cover >10%, were removed from the cutblock layer and merged with the VRI layer; these polygons were considered reforested. Habitat connectivity was calculated according to the adjacency o f mature forest stands with harvested or disturbed patches (i.e., cutblocks). High values for connectivity (based on expert scores) were applied to forest polygons adjacent to one another, while forest polygons adjacent to, or 22 intersected by, disturbances or openings were given lower values. Finally, forest fire age was derived from the Fire Perimeters - Historical layer (DataBC Distribution Service). All fire polygons >40 years old were removed from the layer and considered reforested. There were no corporate data representing the historical growth and composition of BC’s forests. Thus, for the historical landscapes (1990, 2000, and 2005), stand attributes were back-dated to account for change over time. Because the contemporary VRI layer contained cutblock polygons that were harvested after the 1990, 2000, and 2005 time periods, those ‘future cutblocks’ required designation of forest attributes that best represented the historical forest stand. To do this, we selected for all forest-cover polygons directly adjacent to the ‘future cutblock’ polygons and used VRI data to determine the most likely leading tree species, and the average canopy cover, basal area, shrub cover, and forest stand age. In order to back-date the canopy cover and basal area, we used VRI data to determine the natural rate of change based on forest stand age. We assumed that the leading tree species for each harvested block was constant over the study period. All spatial data were rasterized (25x25-m cell) and the mean eigenvector scores for each habitat variable were applied to their respective raster layer (IDRISI Selva ver. 17, Clark Labs, 2011). Raster layers were then combined additively to generate one habitat map for each focal species on each reference landscape at four time intervals. For each historical map we classified the habitat into four classes (i.e., Poor, Moderate, Good, and Very Good). The categorical break points were calculated as the quartile values of the habitat scores for the year-2000 map. This allowed for the consistent comparison o f the change in habitat area over time (i.e., the cumulative impacts o f landscape change). These maps also illustrated the spatial configuration o f habitat across the landscape. 23 Map Variation and Validation Expert-based habitat models can be sensitive to variation in the knowledge elicited from the experts (Johnson and Gillingham 2004). Thus, we calculated the upper and lower 95% confidence intervals for the combined eigenvector scores for the biologist and trapper experts and recreated the habitat maps. We recalculated the area of each habitat class and used the difference between the upper and lower 95th percentile maps to represent the variation around the mean predicted area o f habitat. Additionally, we recreated habitat maps based on the eigenvector scores of the biologists and trappers separately. We quantified habitat change according to the recreated maps and compared these values to the combined expert maps in order to determine variation between expert groups. As one form of validation, we asked each trapper to evaluate the distribution and area o f ranked habitat for their traplines. Trappers located both good and poor trapping areas on generic maps o f their traplines, and then compared those locations to the expert-based habitat maps for each focal species. For each species, trappers were asked to provide a score from 0-10 representing the accuracy of the maps o f predicted habitat. Results Identification of Focal Species Experts identified three species that were most suitable for this study, based on their perceived ecological and economic importance, and susceptibility to landscape change. The biologists identified marten, lynx, and fisher while the trappers identified marten, lynx, and beaver (Castor canadensis). Trappers also suggested wolverine (Gulo gulo) and red squirrel (Tamiasciurus hudsonicus), while biologists identified ermine (Mustela erminea), mink 24 (Neovison vison), and otter (Lontra canadensis) as other potential candidates. When combined, marten, lynx, and fisher were selected as the most suitable focal species. Identification of Habitat Variables We provided the experts with a list of 18 candidate habitat variables for each focal species. Eleven variables were voted into the habitat models for fisher and marten, while ten variables were voted into the lynx model (Figure 2.3). The combined scores for the inclusion o f habitat variables for fisher was highest for structural complexity, while cutblock age was highest for lynx, and coarse woody debris was highest for marten. Uncertainty in the selection o f habitat variables by biologists was lowest for fisher (SE = 0.283) and highest for lynx (SE = 0.312). For trappers, uncertainty was lowest for marten (SE = 0.279) and highest for fisher habitat variables (SE = 0.333). Confidence scores were highest for biologists and trappers when voting for lynx habitat variables (x confidence scores = 7.12 and 8.57, respectively) and lowest when biologists voted for marten (x = 6.94) and when trappers voted for habitat variables for fisher (x = 8 . 10). Overall, there was high consistency in the selection o f variables by the two expert groups. A Wilcoxon rank-sum test revealed significant differences in scores between expert groups for Canopy Cover (z = 1.994, P = 0.046) and Forest Stand Age (z = 2.743, P = 0.006) in the marten model. There were no significant differences in voting between expert groups for the fisher (All z <1.764; all P >0.078) and lynx habitat variables (All z <1.864; all P >0.062). 25 Expert Voting □ Fisher ■ Marten Figure 2.3. Combined scores (mean and standard error) from biologists and trapper experts to include variables for habitat models for fisher, lynx, and marten. The inclusion threshold (horizontal reference line) was set at a score of 2. Evaluation of Habitat Variables We used consistency ratios to test for randomness in the weighting o f habitat variables by the two expert groups. No ratios surpassed the threshold of 0.1, suggesting that there were few or minor operative errors during the survey process (Saaty 1977). Wilcoxon rank-sum tests revealed that six variables in the fisher model had significantly different eigenvector scores when comparing expert groups. Trappers provided significantly higher scores for Open Canopy Cover, Other Conifers, Young Deciduous, and Mid-Age Deciduous, while biologists provided significantly higher scores for Cottonwood and Old Deciduous (all z < 2.741; all P < 0.041; Appendix D). In the lynx model, biologists provided significantly higher scores for Lodgepole Pine (z = 2.001, P = 0.045; Appendix D). The marten model had seven variables with significantly different eigenvector scores when comparing expert groups. Trappers provided significantly higher scores for Moderate Structural Complexity, Minimal Canopy Cover, High Forest Stand Density, Young Deciduous, and Mid-Age Deciduous, while biologists provided significantly higher scores for Moderate Canopy Cover, and Old Deciduous (all z < 2.551; all P < 0.041; Appendix D). Quantification of Habitat Change Habitat models for fisher, lynx, and marten were applied to the ten reference landscapes providing a measure o f change in habitat availability and quality over time (Figure 2.4 and Appendix E). In the West, the combined expert-based habitat model revealed a 52% and 79% decrease in ‘Good’ and ‘Very Good’ fisher habitat, respectively (Table 2.5). Similar trends in habitat change were observed for lynx and marten habitat. In the East, and minimally deforested study area, the availability of habitat changed very little since 1990 (Table 2.5). 27 Lynx Habitat 1990 m Poor - 3382 ha m Moderate - 4982 ha I | Good - 771 ha I Very Good - 8624 ha Waterbodies Lynx Habitat 2013 m Poor - 9629 ha N 2.5 I_ km A Waterbodies ■ I Moderate - 4471 ha } | Good - 97 ha I I Very Good - 3562 ha N 2.5 I_ km A Figure 2.4. Predicted habitat for lynx on one example trapline in the West Study Area, central-interior BC, Canada, during 1990 and 2013. Table 2.5. Change in availability o f habitat for fisher, lynx, and marten from 1990 to 2013 in the West and East Study Area across central-interior BC, Canada. The percentages represent the composition o f habitat across the study area in 1990 and 2013. The upper and lower 95% values represent the variation around the mean habitat change when the maps were constructed using the upper and lower 95th percentile eigenvector scores. East Study Area West Study Area 1990 (%) 2013 (%) Net Change (ha) Lower and Upper 95% (ha) 1990 (%) 2013 (%) Net Change (ha) Poor Moderate Good Very Good 18.3 25.1 34.1 22.5 52.2 26.8 16.3 4.8 77916 3776 -40961 -40734 72006 -11635 -57105 -49959 83826 19187 -24817 -31509 26.0 27.5 23.3 23.2 25.8 28.7 19.6 25.9 -540 4539 -13385 9387 5124 12605 -8057 17118 -6204 -3527 -18713 1656 Lynx Habitat Poor Moderate Good Very Good 17.6 21.7 28.8 31.9 50.9 35.7 8.0 5.4 76537 32344 -47944 -60936 75374 -24167 -80279 -74123 77700 88855 -15609 47749 26.8 24.8 22.3 26.1 25.7 24.4 18.8 31.1 -3983 -1604 -12472 18058 -1500 6163 -5304 30761 -6466 -9371 -19640 5355 Marten Habitat Poor Moderate Good Very Good 18.1 25.4 34.5 22.0 52.3 24.3 17.4 6.0 78616 -2500 -39343 -36786 69470 -10709 -50077 -42471 87762 5709 -28609 -31101 26.2 27.0 23.1 23.8 25.9 28.5 18.3 27.4 -983 5305 -17215 12896 6327 10552 -11611 20557 -8293 58 -22819 5235 Lower and Upper 95% (ha) Fisher Habitat There was a 16% decrease and 11% increase in ‘Good’ and ‘Very Good’ fisher habitat, respectively. Similar trends in habitat change were observed for lynx and marten habitat. ‘Very Good’ habitat increased by 19% for lynx and by 15% for marten in the East Study Area. We used the mean eigenvector scores from the biologists and trappers separately to build habitat models for fisher, lynx, and marten in the West and East Study areas (Figure 2.5). In the West, scores from both expert groups suggested declines in fisher, lynx, and marten habitat. The biologist model predicted an 81% and 71% decrease in ‘Very Good’ fisher and marten habitat, while the trapper model predicted an 80% and 73% decrease. Similarly, the biologist model predicted a 70% decrease in ‘Very Good’ lynx habitat, while the trapper model predicted an 85% decrease. In the East, there was variation in the predictions o f habitat change for fisher. The biologist model predicted a 1 and 11% increase in ‘Very Good’ fisher and marten habitat, while the trapper model predicted a 7 and 8% decrease. Similarly, the biologist model predicted a 3% decrease in ‘Very Good’ lynx habitat, while the trapper model predicted a 33% decrease. Map Validation The trappers were asked to evaluate the accuracy of the habitat maps for each focal species based on their perception of the species’ distribution on their individual traplines (Figure 2.6). In the West and East Study Areas, the marten habitat maps scored the highest with averages of 8.7 (SE = 0.30) and 8.6 (SE = 0.32), respectively. The fisher habitat maps in the West and East Study Areas averaged 7.6 (SE = 0.46) and 8.2 (SE = 0.89), respectively. The lynx maps had the lowest scores in both the West and East Study Areas, with averages of 5.5 (SE = 0.91) and 6.7 (SE = 1.96), respectively. 30 West Study Area East Study Area ■ Combined □ Biologists ■ Trappers 80000 60000 40000 20000 H 0 -20000 20000 0 -20000 -40000 -60000 -80000 -40000 -60000 -80000 Poor Moderate Good Habitat Classification a Combined □ Biologists ■ Trappers 80000 60000 40000 Very Good Poor Moderate Good Habitat Classification Poor Moderate Good Habitat Classification Very Good (Fisher) j2 80000 60000 S 40000 «§ 20000 a 0 5 -20000 8) -40000 5 -60000 .c -80000 u 80000 60000 40000 20000 0 -20000 -40000 -60000 -80000 Poor Moderate Good Habitat Classification Very Good Very Good (Lynx) 80000 60000 40000 20000 80000 60000 40000 20000 0 -20000 -40000 -60000 -80000 0 -20000 -40000 -60000 -80000 Poor Moderate Good Habitat Classification Very Good Poor Moderate Good Habitat Classification Very Good (M arten) Figure 2.5. Change in the area (ha) o f four habitat classifications (1990 to 2013) for expert-based habitat models for fisher, lynx, and marten developed by biologist, trapper, and combined expert models applied to the West and East Study Areas across central-interior BC, Canada. Marten Fisher Figure 2.6. Mean accuracy (± SE) scores for the habitat maps for each focal species in the West (unshaded) and East (shaded) Study Areas. Trappers evaluated the distribution and area o f ranked habitat for their traplines and provided scores representing perceived accuracy from 0-10. Discussion Expert-Based Habitat Modeling The utility of expert knowledge for predictive modeling has been acknowledged in previous wildlife research (Store and Kangas 2001, Yamada et al. 2003, Doswald et al. 2007, O’Neill et al. 2008, Hurley et al. 2009). With traditional empirical studies, the collection of data is often limited by financial and logistical constraints. For this study, we acquired large amounts o f data over a short time span, which could then be applied to habitat models that were applicable across a large study area (602,690 ha). The consistency in the selection of species and AHP scores within and across expert groups, and the strong validation of the 32 final maps suggested that expert-based approaches were appropriate for the species and study areas modeled in this project. The selection of experts is a critical step in the elicitation and application of expertbased knowledge (O’Neill et al. 2008). Although more robust than ad hoc approaches, the peer-referral process can result in selection bias and misrepresentation o f the spectrum of knowledge, as peer nomination can lead to the referral o f likeminded people (Drescher et al. 2013). To reduce selection bias we identified a diverse range of experts by recruiting seed experts from three sub-categories of biologist experts (government, academic, and consultant). Throughout the elicitation process, we maintained rigorous, transparent, and repeatable methods; such rigour is necessary to conduct effective expert-based habitat modeling (Johnson et al. 2012). The experts were given full opportunity to provide input throughout the study, from the identification of focal species and habitat variables to the subsequent evaluation of those variables. The relatively high consistency in expert scores and lack of operative errors in the AHP suggested that the research design was appropriate for the suite of experts involved. The experts agreed that marten and lynx were ideal species for the study, while trappers suggested beaver as the third species and biologists suggested fisher. Overall, fisher had more votes than beaver. Fisher, lynx, and marten are sensitive to habitat change, and thus fit the objectives of this research (Soutiere 1979, Thompson and Colgan 1994, Buskirk et al. 2000, Proulx 2000, Mowat and Slough 2003, Poole 2003, Hoving et al. 2004, Fuller and Harrison 2005, Weir and Almuedo 2010). Although similar in their general ecology, the 33 three focal species use a range o f habitats that may be affected differently by broad-scale, forest harvesting. Marten occupy small home-range sizes, averaging 2-3 km2 for females and 5 km2 for males, and are heavily dependent on forest structures associated with oldgrowth stands (Buskirk 1992, Thompson and Colgan 1994, Chapin et al. 1998, Payer and Harrison 2003, Carroll 2007). Although fisher also depend on old-growth forests, they are more likely to use a range of different serai stages of forest (Proulx 2006, Weir and Almuedo 2010). The home ranges of fisher in BC are typically >100 km2 for males and >25 km2 for females (Weir and Almuedo 2010). Lynx also have large home ranges, averaging around 220 km2 depending on habitat features and prey abundance (Hatler and Beal 2003). Lynx are believed to be adaptable to a range o f serai stages and may benefit from landscape changes that promote habitat for prey (Mowat and Slough 2003, Poole 2003, Hoving et al. 2004). Experts were consistent in their choice o f habitat variables for each species. Biologists provided their largest confidence scores when selecting habitat variables for lynx; however, as a group, there was high variability between individual biologists possibly due to the propensity o f lynx to use a variety o f habitat types. These experts were most certain when scoring habitat for fisher. In contrast, trappers displayed their lowest confidence scores and the highest variation when voting for habitat variables for fisher. Low capture rates of fisher on most traplines in the study area (likely corresponding to low fisher densities) may equate to less knowledge of fisher habitat by trappers. Habitat classifications were relatively consistent within and between expert groups, and occurred without operator error (according to the consistency ratios of the AHP matrices). Trappers reported higher average confidence scores than biologists; however, the variation around the actual eigenvector scores was lower for biologists than trappers. The 34 trappers’ specific knowledge o f furbearer habitat may be limited to, or skewed by, the habitat types that are present on their respective traplines, possibly resulting in discrepancies among the experts in the group. The agreement among biologists may be a product of exposure to similar research studies and literature, or perhaps they possess a broader, general knowledge o f the focal species’ habitat. Doswald et al. (2007) and Hurley et al. (2009) found similar variation when comparing the evaluations o f habitat variables obtained from two distinct expert groups. Failure to consider uncertainty when interpreting habitat models and subsequent maps can lead to inaccurate representations o f habitat, potentially biasing future management decisions (Johnson and Gillingham 2004, Johnson et al. 2012). A relatively high consistency in the scoring of variables by the two expert groups resulted in no substantive changes to the conclusions o f the study. The underlying error in the GIS data was an additional source of uncertainty that may have had a differential effect across the three species. Although we had no means to ground truth or assess the accuracy of the spatial data, the VRI has been verified in previous studies focused on furbearers (Proulx 2006, Proulx et al. 2006). Variation in expert scores was generally greatest for lynx and least for marten and resulted in some differences in the predictions of habitat change. The relatively wide range of scores in the lynx model may be because this species is a habitat generalist, occurring across landscapes with a diversity o f serai stages and habitat types (Mowat and Slough 2003, Poole 2003, Hoving et al. 2004). This inherent variability in habitat use could have led to less certainty by individual experts and greater differences amongst experts when parameterising the lynx model. Additionally, there has been little to no previous research of 35 lynx in the ecological zones found across the central-interior of BC. Trappers assessed the lynx maps as the least accurate o f the three focal species. In the West Study Area, biologists and trappers showed similar predictions of habitat change over time for the three focal species. In the East, the consistency in scores between expert groups was lower. This raises concerns regarding the application of these habitat models across ecosystems. The experts were initially asked to evaluate habitat according to their knowledge of the Sub-Boreal Spruce BEC zone, which comprises the majority of the West Study Area. The models may not have been compatible with the East Study Area, which includes a greater proportion o f the Engelmann Spruce-Subalpine Fir zone. Habitat Change The expert-based habitat models suggested significant declines in predicted habitat for furbearers in the West Study Area, while the East remained relatively stable. The West Study Area has been subjected to unprecedented levels o f timber harvesting during the study period in response to a mountain pine beetle outbreak. In order to salvage merchantable timber, the allowable annual cut in the interior of BC had increased significantly over the past decade, peaking at over 60 million m3 per year (BC Ministry o f Forests, Mines, and Lands 2010). Large-scale clear-cut logging was the principal method o f timber removal that also included mature spruce and fir species. Although the rate o f timber harvesting is now decreasing, the rapid extraction of timber has resulted in younger and less diverse forest types across much o f the interior o f BC. Forestry is the driving force for cumulative impacts in the region and the reduction in habitat for the three focal species. A lack of mature and complex forest stands, widespread openings resulting from clear-cut logging, and habitat fragmentation may be limiting the distribution of furbearers. 36 The East Study Area has been exposed to relatively low levels of timber harvesting. In the mid-1980s, this area experienced an outbreak o f spruce budworm (Choristoneura biennis), which initiated some salvage logging. Since that outbreak, new logging activity has been limited. This has allowed forest stands to mature and may be the primary reason for the slight increase in optimal habitat for all three focal species. Thus, observed differences in the level of forest harvesting across the East and West Study Areas are consistent with the structure and ultimately the predictions o f the expert-based habitat models. The combination o f late-successional and regenerating forests in the East may provide an ideal mix o f habitat for fisher (although they may be limited by high snow depths in this area). The abundance of old-growth conifer forests may be ideal for marten. According to the expert-based models, habitat availability does not appear to be a limiting factor for fisher or marten in that area. The distribution of both species may be restricted by other factors such as elevation, overhead cover in recently logged areas, snow accumulation (Weir and Harestad 2003), or the availability of elemental habitat features like denning trees (Weir et al. 2012). Competition with fisher may also limit marten in areas where they overlap (Carroll 2007). The expert-based habitat model predicted significant decreases in the availability and quality o f lynx habitat across the West, but not the East study area. Again, the discrepancy is likely a result of vastly different levels of anthropogenic impacts. Although the habitat model predicted declines in lynx habitat in the West Study Area, the trapping community report relatively high numbers o f lynx (M. Bridger unpub. data). This suggests that lynx are not necessarily tied to what experts perceive as quality habitat, but rather are dependent on prey availability or other factors. The presence o f widespread regenerating pine forests may 37 currently be benefiting snowshoe hare (Lepus americanus) populations, which appear to be peaking (M. Bridger unpub. data). Some experts reported that the lynx habitat model may have been overly influenced by the attributes of old-forest stands, possibly over-predicting habitat quality in areas o f late-successional forests. Although lynx do use old forest stands for certain life requisites (Paragi et al. 1997), perhaps not enough emphasis was placed on the importance o f younger, regenerating forest stands that may provide abundant prey habitat. Management Implications Fisher In BC, fishers are most prevalent across intact landscapes containing habitat features associated with late-successional forests (Proulx 2006, Weir and Almuedo 2010, Weir and Corbould 2010). These features are particularly important for denning, resting, and providing cover from snow accumulation (Weir and Harestad 2003). Given the large home ranges and low density o f fishers in BC (Weir and Corbould 2006), the maintenance o f landscape-level habitat connectivity may be critical. Habitat fragmentation is a significant concern given the fishers’ propensity to avoid open areas associated with industrial activities (Weir and Harestad 2003, Weir and Almuedo 2010). Although the models depicted fisher habitat at a landscape level, the experts recognized the importance o f elemental habitat features throughout the process of model development. Certain life requisites are associated with elements found in late-successional forests, but fishers are not necessarily old-growth specialists. They are known to use earlyseral forests, mixed forest stands, and edge habitat (Weir and Almuedo 2010). The structure of the forest itself may be more important than stand age or type. Their primary prey species, 38 the snowshoe hare, and other small mammalian prey are often associated with young, regenerating forests (Weir and Harestad 2003, Weir and Almuedo 2010). These prey species, and thus, fisher, still require abundant coarse woody debris (CWD), ground cover, and structurally complex forest floors (Weir and Harestad 2003). In BC, the rearing of young fishers occurs exclusively in tree cavities, primarily deciduous tree species (Lofroth et al. 2010). Thus, during denning, female fishers are dependent on large, decaying trees, primarily deciduous or conifers such as Douglas fir and pine that have heart rot (Weir et al. 2012). Managers must recognize the importance of maintaining or promoting den sites, but these features are not directly represented in largescale landscape models (McCann et al. 2014), such as those presented in this work. Lynx Anthropogenic disturbance may influence lynx habitat in both positive and negative ways. In general, lynx distribution is highly correlated with their primary prey species, snowshoe hare (Apps 2000, Poole 2003, Hoving et al. 2004, Simons-Legaard et al. 2013). Habitat for snowshoe hare is often associated with densely vegetated, regenerating forests that occur following logging or wildfire (Mowat et al. 2000, Poole 2003, Simons-Legaard et al. 2013). Consequently, both snowshoe hare and lynx may benefit from landscape disturbance (Mowat and Slough 2003, Hoving et al. 2004). Habitat changes stemming from forestry activities may not be immediately beneficial, as lynx have been found to avoid recently disturbed habitats (Poole 2003, Hoving et al. 2004). Furthermore, an increase in openings associated with timber harvesting may fragment lynx habitat. Given the large annual range o f lynx, habitat fragmentation may be a concern (Buskirk et al. 2000, Mowat and Slough 2003). 39 Although most foraging habitat is associated with early-successional forests and edge habitat, lynx do require mature forest stands to meet certain life requisites. Mature conifer or mixed forests provide valuable habitat for denning, resting, and cover from extreme climatic conditions (Paragi et al. 1997). It is widely accepted that lynx use mature forest stands, although few studies have reported selection for these habitats (Mowat et al. 2000, Poole 2003, Hoving et al. 2004). Late-successional forests may be of particular importance to lynx during low periods o f the hare cycle, as lynx often switch to alternate prey found in these areas, such as red squirrels (Paragi et al. 1997, Mowat et al. 2000). Marten The availability o f marten habitat must be managed at a landscape level, as they are sensitive to habitat fragmentation (Soutiere 1979, Thompson and Colgan 1994, Hargis et al. 1999, Fuller and Harrison 2005). Home-range sizes o f marten living on fragmented landscapes are much larger compared to those inhabiting intact forests (Soutiere 1979, Thompson and Colgan 1994, Fuller and Harrison 2005). Marten avoid openings associated with anthropogenic disturbances, such as cutblocks, likely due to predation risk (Potvin et al. 2000, Carroll 2007, Cheveau et al. 2013). Previous studies have shown that marten do not tolerate landscapes that contain greater than 25-30% unsuitable habitat, including natural openings and cutblocks (Hargis et al. 1999, Potvin et al. 2000, Cheveau et al. 2013). Marten are widely associated with late-successional, coniferous forest stands (Payer and Harrison 2003, Proulx et al. 2006, Carroll 2007, Webb and Boyce 2009, Cheveau et al. 2013). These forests contain elemental features that provide thermal cover during winter, cover from predators, foraging habitat, and resting and denning areas (Thompson and Colgan 1994, Potvin et al. 2000, Carroll 2007, Cheveau et al. 2013). Unlike lynx, and in some cases 40 fisher, marten avoid regenerating forests (Soutiere 1979, Potvin et al. 2000, Fuller and Harrison 2005); although, they may hunt along edges provided there is adequate overhead cover (Chapin et al. 1998). Researchers have emphasized the importance o f maintaining forest stands with basal areas greater than 18-20 m2/ha and 30-50% canopy cover (Soutiere 1979, Fuller and Harrison 2005, Proulx et al. 2006). The maintenance o f CWD in forest stands is also critical (Payer and Harrison 2003). Basal area, crown closure, and CWD were recognized as important by both expert groups and, thus, were parameterised at the scale o f the supporting GIS data (Figure 2.3). Conclusion Habitat models can be important tools for developing and implementing management plans for wildlife species. Although empirical models have value for predicting and mapping wildlife habitat (Guisan and Thuiller 2005), a need to develop models quickly and effectively may lead researchers to consider expert-based approaches. Where the collection o f empirical data is limited by time and financial constraints, expert knowledge provides an efficient and rapid alternative. This may be particularly important for cryptic species that are difficult to study. The performance of expert-based models has received mixed reviews (Store and Kangas 2001, Clevenger et al. 2002, Doswald et al. 2007). The results of this study, however, suggest considerable utility in the use o f expert knowledge for mapping habitat and subsequently examining habitat change over time. There are advantages to involving multiple groups of experts that have unique, but complementary domains of expertise. For this study, trapper experts can assist with the on­ ground validation o f model results and link habitat change to population abundance in the form of trapping records (Smith et al. 1984, Ruette et al. 2003). Professional experts may 41 provide a more general perspective on the habitat requirements o f individual species. For the three focal species presented in this study, model structure was remarkably similar among expert groups. There are a variety o f techniques available for eliciting expert knowledge that can be applied to the development o f habitat models. These depend on the study objectives, the skill o f the research team, the focal species, and the availability o f supporting GIS data. The identification of suitable experts, application of rigorous and repeatable methods, and quantification of uncertainty, however, are key components o f any method (Burgman et al. 201 la, Johnson et al. 2012, McBride and Burgman 2012, Drescher et al. 2013). This research demonstrates strict adherence to those principles. The purposeful and incremental assessment of the elicitation process - including consistency ratios, measures o f confidence, parameter variance and validation o f predictions - provides evidence of rigour and a measure o f the reliability o f resulting models and predictions of habitat change. Ultimately, however, understanding o f the spatiotemporal change in habitat for these three species would be impossible or much delayed if there was a strict requirement for empirical data. These study findings are transparent and defensible and can be used to influence habitat management and strategic decisions relative to past and future cumulative landscape change. The findings also suggest that habitat for fisher, lynx, and marten may decrease substantially where intensive forestry occurs. Forest and wildlife managers must recognize the importance of maintaining furbearer habitat on landscapes subjected to cumulative industrial impacts. 42 Chapter 3: Assessing cumulative impacts of forest development on the abundance of furbearers using harvest records Abstract: Understanding the cumulative impacts of landscape change is important when managing and conserving wildlife populations. Across the central-interior of British Columbia, Canada, furbearer populations are being subjected to the cumulative impacts of industrial development. This region has witnessed unprecedented levels o f anthropogenic landscape change, primarily in the form of increased forestry. We used trapping records to investigate the relationship between habitat change resulting from cumulative impacts of landscape change and population abundance o f Canada lynx (Lynx canadensis) and American marten (Maries americana). We applied expert-based habitat models to ten reference landscapes (i.e., traplines) in two distinct study areas to serve as measures of habitat change over the study period between 1990 and 2013. We elicited fur harvest records (1990-2013) from trapper experts and then used negative binomial count models to relate capture success to habitat change. We controlled for factors that were hypothesised to influence capture success, including effort and climatic conditions, allowing us to observe the effects o f habitat availability and quality on population abundance. Overall, the top-ranked count models identified combinations of habitat availability and quality, trapping effort, and trapline area as factors positively ipfluencing the capture success of lynx and marten. These results suggest that reduction in high-quality habitat may have a direct and negative effect on the abundance of lynx and marten in the study area. Results o f this study also illustrate the utility of fur-harvest records for investigating population abundance o f furbearer species. A precise measure o f trapping effort, however, is necessary to relate environmental covariates, including habitat change, to harvest at the scale o f individual traplines. 43 Introduction Furbearer species have significant cultural and economic importance for fur-trapping industries across North America (Hamilton et al. 1998, Webb and Boyce 2009). Between the years 2000-2012, the fur trapping industry in British Columbia (BC), Canada, grossed nearly $17 million in revenue (BC Fur Returns unpub. data). Recent increases in industrial activity, however, have led to concerns about the distribution and abundance o f furbearer populations. Across the central-interior of BC, forestry has had a notable impact on the landscape, largely in response to a mountain pine beetle (Dendroctonus ponderosae) epidemic that has killed 53% of merchantable pine stands (BC Ministry of Forests, Lands, and Mines 2010). Given the current understanding of furbearer biology, the large-scale loss o f forests and resulting salvage harvest is certain to have negative impacts on the habitat of furbearer populations; however, the magnitude of these impacts is unknown. The high rate and large area of forest harvesting occurring across the central-interior of BC may result in cumulative impacts for wildlife species that are dependent on old forests. Cumulative impacts can be interactive, additive, or synergistic, and can alter the environment at a number o f temporal and spatial scales (Nitschke 2008, Johnson 2011). Compared to acute, immediate changes to habitat, cumulative impacts can be complex and difficult to understand (Johnson 2011). Cumulative landscape change may have substantial and long­ term negative impacts on the habitats and ultimately the distribution and abundance o f furbearer populations (Thompson 1994, Proulx 2000, Webb and Boyce 2009). Canada lynx (Lynx canadensis) and American marten (Martes americana) use a range o f habitats that may be affected differently by the cumulative impacts o f landscape 44 change. Lynx have large home ranges and are adaptable to varying serai stages o f forested habitats (Mowat and Slough 2003, Poole 2003, Hoving et al. 2004). Lynx populations may ultimately benefit from some landscape disturbances that promote habitat for prey, provided other life-history requirements remain available. In contrast, marten have small home ranges and are considered old-forest specialists, dependent on forest structures, such as coarse woody debris, associated with late successional forests (Buskirk 1992, Thompson and Colgan 1994, Chapin et al. 1998, Payer and Harrison 2003, Carroll 2007). The loss o f contiguous habitat to timber harvesting is likely to have negative impacts on marten populations, a concern which has been voiced by trappers (M. Bridger unpub. data; Chapter 2). There are approximately 2600 registered traplines in BC, and 900 licenses issued annually. Trappers are required to document their capture totals when selling furs to market, which are entered into a provincial database. Additionally, many trappers consider themselves stewards o f populations on their traplines, and thus, keep personal records of trapping activity and harvests. Trapping records have been used to monitor abundance and population trends of furbearers (McDonald and Harris 1999, Ruette et al. 2003, Webb and Boyce 2009) and they can be particularly useful when the behavior o f furbearer species makes typical monitoring difficult (Ruette et al. 2003). Also, trapping records represent a significant amount of data that can be collected for large geographic areas at a relatively low cost (Ruette et al. 2003). Although trapping records may not be useful for detecting population changes over small spatial and temporal scales, they have utility for identifying long-term population trends across regional areas (Smith et al. 1984, Raphael 1994, Poole and Mowat 2001, Ruette 45 et al. 2003). Such data require careful interpretation, however, and researchers have cautioned against the use of databases that report harvest totals only. In particular, a failure to report or control for factors that influence harvest dynamics can make interpretation of harvest records difficult (McDonald and Harris 1999, Poole and Mowat 2001). The pitfalls associated with broad-scale harvest data may be avoided by using site-specific harvest records from individual traplines that cover large geographical areas over long timeframes (Erickson 1982, Raphael 1994). Numerous factors may influence trapping success thus biasing capture records or data representing fur sales. O f greatest importance may be trapping effort that can be influenced by population cycles, quota changes, trap-type restrictions, access, weather, fur prices, and socio-economic conditions (Raphael 1994, McDonald and Harris 1999, Poole and Mowat 2001, Cattadori et al. 2003, Ruette et al. 2003, Webb and Boyce 2009). Failing to account for such factors will limit any inference to the underlying population dynamics o f the trapped species (McDonald and Harris 1999). We developed an expert-based approach to quantify variation in the harvest and ultimately the population abundance of two species of furbearers following rapid change in the quality and availability o f habitat across landscapes. Where previous studies have used harvest datasets collected from large geographical areas, we used the personal records of trappers specific to their registered traplines to account for both effort and success. We related trap data for marten and lynx to the availability and quality of habitat, as modeled using expert knowledge. After controlling for trapping effort, this relationship served as an index of the population abundance o f the two species. Traplines were located within two distinct study areas subjected to high levels o f forestry development (hereafter referred to as 46 the ‘West Study Area’) and low levels of forestry development (hereafter referred to as the ‘East Study Area’; Figure 1.1). We hypothesised that the capture success o f lynx and marten would be related to trapping effort and habitat availability and quality. Methods Data Collection We used trapping records (corresponding to registered traplines) and negative binomial count models to investigate factors that influenced capture success of lynx and marten. We collected harvest data specific to the traplines of ten trappers. We met with trappers in person and documented annual catch statistics for each species. Records were in the form of personal journals or fur return records (mandatory for the commercial sale o f furs in BC). We collected records dating back to 1990 for lynx and marten; records for fisher (Pekania pennanti) were also collected, but the capture rate was too infrequent to be used for these analyses. Few trappers had complete records from 1990-2013. We established a number o f measures o f trapping effort for each set of harvest data, depending on the information possessed by the trappers. In one instance, a trapper kept catch per unit effort (CPUE) harvest records for each year o f trapping. Three trappers recorded only the number of days spent trapping per season, while three others were able to approximate the number o f traps set per season. The three remaining trappers did not record data that could be used to quantify effort. In this case, the trappers used their recollection of trapping activity to assign a value of effort from 0-10 for each year. These trappers identified particular years when effort was highest and assigned a score of 10; scores for the remaining years were assessed relative to those years o f highest effort. All other measures of 47 effort were then standardised on a scale from 0-10 to be used as a covariate in the negative binomial count models. We conducted a brief survey with the trappers to identify other variables that they felt influenced trapping effort and success. We also discussed changes in their trapping methods over time, as well as the spatial distribution o f effort across the trapline. Trappers provided map locations of productive trapping areas and noted changes in trapping locations due to habitat alteration. Model Development We used the literature and consulted with expert participants to identify a set of predictor variables that we hypothesised would explain capture success o f lynx and marten (Table 3.1). Those variables included measures of effort and habitat value, trapline area, fur prices, and climatic conditions. Determining the influence o f cumulative landscape change on capture success was o f particular interest. Therefore, we tested three habitat-related variables derived from habitat models. Those models were developed through an expertbased approach, where biologists and trappers identified and evaluated habitat variables for lynx and marten. The expert-based habitat models were used to quantify change in habitat availability and quality for lynx and marten on each trapline at four time intervals: 1990, 2000, 2005, and 2013 (Chapter 2). We extrapolated habitat values (Table 3.1) for all missing years between 1990 and 2013 by determining the rate o f change between time intervals; this assumed a linear rate o f habitat change. 48 Table 3.1. Variables used for the development of count models for predicting capture success o f lynx and marten across central-interior BC, Canada. Parameter Abbreviation Description Effort E Measure of trapping effort on a scale from 0-10 Standardised Effort SE Measure of trapping effort relative to trapline area, where ‘Effort’ was multiplied by ‘Trapline Area’ Trapline Area TA Trapline area in hectares Habitat Value HV Sum total of the raster habitat values on each respective trapline, according to expertbased habitat maps Standardised Habitat Value SH Habitat relative to trapline area, where ‘Habitat Value’ was divided by ‘Trapline Area’ GVGH Percent of the respective trapline area composed of ‘Good’ or ‘Very Good’ habitat, according to expert-based habitat maps Fur Price FP The average fur price from the previous year’s trapping season Mean Minimum Temperature MMT Average daily minimum air temperature recorded at Prince George Airport weather station (Nov-Jan for marten; Dec-Feb for lynx) Extreme Minimum Temperature EMT Average monthly extreme minimum air temperature recorded at Prince George Airport weather station (Nov-Jan for marten; Dec-Feb for lynx) Snow Depth Sum SMS Cumulative snow depth recorded at Prince George Airport weather station (Nov-Jan for marten; Dec-Feb for lynx) % Good and Very Good Habitat 49 We acquired historic fur prices through the provincial fur return database, and weather data from the Environment Canada weather station at Prince George, BC (Environment Canada 2014). We acquired weather data for the peak trapping months for lynx (December-February) and marten (November-January). Trappers identified low temperatures and snow cover as important factors influencing capture success, thus we calculated the mean minimum and extreme minimum air temperatures and snow depth for those months. We used negative binomial regression models (NBRM) to examine factors that influenced capture success (Stata, ver. 12.1, StataCorp, 2011). The harvest data from individual traplines for each year were used as the dependent variable for the count models. We developed sets o f models for lynx and marten in the West and East Study Areas separately. Due to a relatively high number o f zero capture events of lynx in the East Study Area, I attempted to fit a zero-inflated negative binomial model (ZINB) to the data (Vuong tests; Vuong 1989). However, the ZINB model did not conform well to the data, thus I used a NBRM. Model Selection Empirical data explaining variation in capture success were lacking, thus we used categories o f explanatory hypotheses to guide the model-selection process: trapping effort; variation in habitat over time; effort and habitat; effort and weather; habitat and weather; and effort, habitat, and weather (Appendix F). In some cases, low sample sizes dictated relatively few model parameters, where we followed a rule o f approximately one covariate for every 10 records (Vittinghoff and McCulloch 2007). We used an information-theoretic approach and Akaike’s Information Criteria (AICc) for small sample sizes to identify the most 50 parsimonious model from each set o f explanatory hypotheses (Burnham and Anderson 2002). The best model in the set had the lowest AICCI and the highest AICc weight (w,); where model separation was uncertain, we selected all models that differed by < 2 A, AIC points. For the best models in the set, we generated P-coefficients for each parameter, representing the positive or negative direction of the effect; we considered a parameter statistically significant at a <0.05. We clustered data on trapline to correct the variance for repeated sampling across years on each trapline (Dormann et al. 2007). We used the variance inflation factor to test the best-fit models for multicollinearity; no covariates surpassed the variance inflation threshold of 10 (Chatteijee and Hadi 2006). Model Prediction Information theoretic approaches provide only a relative measure of model parsimony, not an absolute measure of model fit. Thus, we used cross-validation to assess the predictive accuracy of the most parsimonious models. We used a bootstrapping-type method, where each capture record was withheld sequentially from the model fitting process, and the resulting model (N -l) and the withheld record was used to predict an independent harvest count. We then calculated the unstandardised residuals (difference between observed and predicted counts); a mean of zero suggested perfect prediction, a negative value suggested over-prediction, and a positive value suggested under-prediction. We used Wilcoxon rank sum tests to statistically compare the predicted and observed captures. Catch per Unit Effort As a second index of lynx and marten population abundance, we assessed the trends in CPUE and trapping effort over time (Microsoft Excel, ver.14.0, Microsoft Corporation, 51 2010). Catch per unit effort was derived from the annual number o f captures by each trapper divided by their effort level. Both trapping effort and CPUE were standardised by trapline area. We calculated best-fit linear trend lines to the resulting scatterplots. Results Catch per Unit Effort Trends in CPUE and trapping effort for lynx and marten varied over time. In the West Study Area, CPUE (F 1/73 = 5.56, P = 0.021) and trapping effort (Fi,73 = 17.22, P <0.001) for lynx increased over time (Figure 3.1a), suggesting that trappers were capturing more lynx as they increased their effort levels. We did not analyse the CPUE and trapping effort for lynx in the East Study Area, due to low capture numbers. In the West, CPUE (F 1,84 = 6.80, P = 0.011) for marten decreased and trapping effort (Fi,g4 = 9.14, P = 0.003) increased with time, suggesting that trappers applied more effort over time, but captured fewer marten (Figure 3.1b). In the East, there was no significant relationship between CPUE or trapping effort and time (Fi,45 = 1.90, P = 0.175 and Fi,45 = 0.25, P = 0.616, respectively; Figure 3.1c). Count Models Lynx The NBRM was the best model for lynx in the West Study. While the ZINB model appeared to be a better fit for lynx captures in the East (Vuong = 2.42, P = 0.008), the ZINB model did not conform to the data, thus I used a NBRM. The top-ranked model for lynx in the West Study Area (AICc w, = 0.520) included the parameters for ‘Effort’, ‘Trapline Area’, ‘Good and Very Good Habitat’, and ‘Extreme Minimum Temperature’ (Table 3.2). 52 (a) 1.6 1-2 0) (0 ® 1 < /) 3 (0 ^ ®0.8 Mo.6 q. O 0.4 x CPUE • Effort :e 70 £ 1.4 40 ■o a) ro a) 30 ” « R2 = 0.0707 R2 =.0.1909 I * * 0.2 £ 20 < A .-x - * • • ! 8 *• M * Linear (CPUE) Linear (Effort) 10 | x X x0 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 Year T - x -x - -X X (b) CPUE a ~u ® 2 5 -I I I 5 S i 4 111 S ’ 3 z> J= o. 2 50 - • • • * * R2 = 0.0749 * • * -a ""! • *x • * * -arlf * * 30 « % Linear (CPUE) - Linear (Effort) O r ; R ^ = 0.0981* x i ■ M y » j j * x x * * x * , | | * 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 Year (c) ac> "a xCPUE • Effort £ 8 § « I I 2 UJ ra ■c it UJ 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 Year Figure 3.1. Scatterplots and linear regressions displaying catch per unit effort (CPUE) and trapping effort for lynx captures in the West Study Area (a) and marten captures in the West (b) and East Study Area (c) across central-interior BC, Canada. The CPUE represents the number of lynx or marten captured by each trapper in a given year divided by their effort level. 53 Table 3.2. Summary o f model selection statistics for the candidate count models to predict lynx captures in the West Study Area (Capture events; N = 75) across central-interior BC, Canada. The models were developed from six a priori categories o f explanatory hypotheses. The top models from each category were selected and then ranked against each other. Model Category Rank AICc, AlCcWi A AICc E + TA + GVGH + EMT E+TA+GVGH E + TA E + EMT GVGH + TA SH + EMT SH + SDS SH + SDS + EMT SH + SDS + MMT Effort + Habitat + Weather Effort + Habitat Effort Effort + Weather Habitat Habitat + Weather Habitat + Weather Habitat + Weather Habitat + Weather 1 2 3 4 5 6 7 8 9 397.97 398.14 412.83 428.93 468.41 470.39 471.49 471.78 472.34 0.520 0.479 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.000 0.165 14.853 30.953 70.433 72.413 73.513 73.805 74.365 The second-ranked model differed by <2 points, but was a subset of the top model. The top two models accounted for 99.9% of the AICcw,. There were only 27 capture events for lynx in the East Study Area. The low sample size required the inclusion of few parameters in the models. The top-ranked model (AICcw, = 0.223) included the parameters for ‘Effort’, ‘Trapline Area’, and ‘Mean Minimum Temperature’ (Table 3.3). The top two models accounted for only 58.1% o f the AICcw,. The coefficients generated from the best NBRM suggested that lynx captures in the West Study Area were positively influenced by ‘Effort’, ‘Trapline Area’, and ‘Good and Very Good Habitat’ (Figure 3.2a). ‘Extreme Minimum Temperature’ did not have a significant effect on capture success. In the East, the coefficients generated from the top-two NBRMs suggested that ‘Effort’ and ‘Trapline Area’ had a significantly positive influence on capture success o f lynx, while ‘Mean Minimum Temperature’ and ‘Extreme Minimum Temperature’ had a significant negative effect (Figure 3.2b; Appendix G). 54 Table 3.3. Summary o f model selection statistics for the candidate count models to predict lynx captures in the East Study Area (Capture events; N = 27) across central-interior BC, Canada. The models were developed from six a priori categories o f explanatory hypotheses. The top models from each category were selected and then ranked against each other. Model Category Rank AICc A IC cM>, E + TA + MMT E+TA +EM T E + TA E + SH E + TA + SH E + SH + SDS E + TA + GVGH E + GVGH E + SH + EMT E + SH + MMT E + GVGH + EMT HV HV + SDS HV + EMT Effort + Weather Effort + Weather Effort Effort + Habitat Effort + Habitat Effort + Habitat + Weather Effort + Habitat Effort + Habitat Effort + Habitat + Weather Effort + Habitat + Weather Effort + Habitat + Weather Habitat Habitat + Weather Habitat + Weather 1 2 3 4 5 6 7 8 9 10 11 12 13 14 117.86 118.06 122.76 122.98 120.86 122.28 122.34 124.92 122.52 122.78 123.54 166.52 165.20 166.14 0.305 0.276 0.094 0.084 0.068 0.034 0.033 0.032 0.030 0.026 0.018 <0.001 <0.001 <0.001 A, AICc 0.000 0.200 2.357 2.577 3.000 4.420 4.480 4.517 4.660 4.920 5.680 43.777 44.797 45.737 Marten The top-ranked negative binomial regression model explaining captures of marten in the West Study Area (AICc w, = 0.471) included parameters for ‘Effort’, ‘Trapline Area’, ‘Standardised Habitat’, and ‘Extreme Minimum Temperature’ (Table 3.4). The top three models differed by <2 points and accounted for 99.9% o f the AICcw,. In the East Study Area, the top model accounted for 51.0% o f the AICcw, and included parameters for ‘Effort’, ‘Trapline Area’, and ‘Good and Very Good Habitat’ (Table 3.5). The subsequent three top-ranked models contained the same parameters, but included climatic variables. The top-ranked model was the most parsimonious and differed from the second-ranked model by >2 points. 55 (a) 0.45 - ♦ E+TA+GVGH+EMT 0.40 SE+TA+GVGH 0.35 o 0.30 V ° 0s IX ) O) 0.25 c 0.20 a> 1 015 O 0.10 0.05 -0 - 0.00 -0.05 Effort Trapline Area Good/Very Good Habitat Extreme Min. Temp. (b) ♦ E+TA+EMT ■ E+TA+MMT in - 0.1 - 0.2 Effort Trapline Area Extreme Min. Temp. Mean Min. Temp. Figure 3.2. Model coefficients and 95% confidence intervals of top-ranked models (where A AICc was < 2 .0 ) representing influences on capture success of lynx in the West (a) and East (b) Study Areas across central-interior BC, Canada. 56 Table 3.4. Summary o f model selection statistics for the candidate count models to predict marten captures in the West Study Area (Capture events; N = 86) across central-interior BC, Canada. The models were developed from six a priori categories o f explanatory hypotheses. The top models from each category were selected and then ranked against each other. Model Category Rank AlCrf AICcw/ A A IG E +T A +SH +EM T E + TA+ SH + MMT E + TA+SH E + TA E + EMT E + SDS + EMT E + SDS HV HV + EMT SH + TA HV + TA GVGH+TA HV + MMT GVGH HV + SDS Effort + Habitat + Weather Effort + Habitat + Weather Effort + Habitat Effort Effort + Weather Effort + Weather Effort + Weather Habitat Habitat + Weather Habitat Habitat Habitat Habitat + Weather Habitat Habitat + Weather 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 660.83 661.41 662.79 676.34 680.78 681.83 682.52 745.75 746.38 746.56 746.76 746.96 746.98 747.01 747.44 0.471 0.352 0.177 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.000 0.582 1.963 15.515 19.955 21.003 21.695 84.918 85.555 85.735 85.935 86.135 86.155 86.178 86.615 The coefficients generated from the best NBRMs suggested that marten captures in the West Study Area were positively influenced by ‘Effort’, ‘Trapline Area’, and ‘Standardised Habitat’, while ‘Extreme Minimum Temperature’ and ‘Mean Minimum Temperature’ had a negative effect (Figure 3.3a; Appendix G). The coefficients generated from the best NBRM for marten captures in the East suggested that ‘Effort’, ‘Trapline Area’, and ‘Good and Very Good Habitat’ had a significantly positive influence on the capture success o f marten (Figure 3.3b; Appendix G). Model Fit We used Wilcoxon rank sum tests to compare the predicted and observed captures from all top models for lynx and marten in the West and East Study Areas. There were significant differences in observed and predicted captures of lynx in both study areas (Appendix H; all H-statistics >24.09, all P <0.039). There were also significant differences 57 Table 3.5. Summary o f model selection statistics for the candidate count models to predict marten captures in the East Study Area (Capture events; N = 47) across central-interior BC, Canada. The models were developed from six a priori categories o f explanatory hypotheses. The top models from each category were then ranked against each other. Model Category Rank AlCd AIC ch>, A, AICc E+TA+GVGH E + TA +GVGH + SDS E + TA +GVGH + EMT E + TA + GVGH + MMT GVGH + TA GVGH + EMT GVGH + SDS SE SE + SDS SE + FP SE + TA SE + EMT Effort + Habitat Effort + Habitat + Weather Effort + Habitat + Weather Effort + Habitat + Weather Effort Effort + Weather Effort + Weather Habitat Habitat + Weather Habitat Habitat Habitat 1 2 3 4 5 6 7 8 9 10 11 12 429.84 431.96 432.22 432.23 439.31 466.25 466.41 468.23 470.15 470.17 470.19 470.31 0.510 0.177 0.155 0.154 0.004 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.000 2.117 2.383 2.393 9.475 36.415 36.575 38.391 40.315 40.335 40.355 40.475 in observed and predicted captures of marten in the West Study Area (Appendix H; all Hstatistics >72.23, all P <0.022). There was no significant difference between observed and predicted captures of marten in the East Study Area (Appendix H; H = 38.89, P = 0.385). Discussion Our results demonstrate the utility of trapping records for investigating furbearer abundance in relation to cumulative habitat change. Many studies have attempted to use trapping records as a proxy for abundance and for long-term population monitoring (Smith et al. 1984, Raphael 1994, McDonald and Harris 1999, Poole and Mowat 2001, Ruette et al. 2003, Webb and Boyce 2009). Few studies, however, have used trapping records to relate population abundance to habitat change (Webb and Boyce 2009). Such records provide a large amount of data that can be collected relatively quickly and inexpensively (Ruette et al. 2003). This may be particularly important for cryptic and inconspicuous species like 58 (a) 0.35 ♦ E+TA+SH+EMT 0.30 BE+TA+SH+MMT 0.25 AE+TA+SH 0.20 o o' in 0.15 a> c 0.10 4 0) jg 0.05 -0 - O 0.00 -0.05 - 0.10 H -t- -0.15 Effort Trapline Area Standardized Habitat Extreme Min. Temp. Mean Min. Temp. (b) 0.35 n ♦ E+TA+GVGH 0.30 0.25 - O vP 0.20 - in ro i - 0.15 .92 e(D o.io ° 0.05 - 0.00 - -0.05 Trapline Area Effort Good/Very Good Habitat Figure 3.3. Model coefficients and 95% confidence intervals o f top-ranked models (where A AICc was < 2 .0 ) representing influences on capture success o f marten in the West (a) and East (b) Study Areas across central-interior BC, Canada 59 furbearers, where other forms of population monitoring may be difficult (Ruette et al. 2003). Although other studies have used large-scale harvest databases with varying success, we implemented a novel approach that was dependent on the knowledge and personal furharvest records o f trappers. This allowed us to account for a number of factors that might influence inter-annual variation in harvest success, including habitat change and trapping effort, important considerations that have been difficult to address in other studies (Smith et al. 1984, Raphael 1994, McDonald and Harris 1999, Poole and Mowat 2001). Erickson (1982) and Raphael (1994) suggested that site-specific harvest records may be useful for identifying relative abundance and population trends, contingent on the quantification of annual trapping effort. An ideal measure of trapping effort would include the number of traps set multiplied by the number o f trap days (McDonald and Harris 1999). We attempted to quantify trapping effort through a variety of measures, including the number o f traps set per season or per day, or the number of days spent trapping during the season. When quantifiable measures of effort were not available, trappers provided self-evaluations o f effort for each year of trapping. These elicited data were a reasonable measure o f trapping effort as indicated by the consistent performance o f the effort variables in the count models representing annual variation in harvest of each species. Influences on Capture Success The most parsimonious count models suggested that changes in lynx and marten harvest were influenced by trapping effort and habitat, and in some cases, weather parameters were also informative. Trapping effort was included in each model, with a positive influence on capture success for both lynx and marten. Size of the trapline also had a positive influence on capture success in all cases. For marten in the West Study Area, low 60 air temperatures increased capture success. Many trappers suggested that marten captures increase with cold temperatures (M. Bridger unpub. data), as marten forage more frequently to meet high energetic demands. This is supported by empirical studies o f the winter activity o f marten (Buskirk et al. 1988). Similarly, colder temperatures appeared to contribute to increased captures o f lynx in the East Study Area. The top-models for lynx in the West and marten in both study areas included a measure of habitat change: capture success on the traplines were positively related to habitat availability and quality. This supported our hypothesis that capture success, and ultimately population abundance, would vary with cumulative impacts of landscape change. A covariate for habitat availability and quality, however, was not present in the top two models for lynx captures on traplines located in the East Study Area. Measures of habitat availability and quality were derived from expert-based habitat maps that represented cumulative landscape change for each trapline over time (1990 to 2013; Chapter 2). A previous study by Webb and Boyce (2009) used harvest records and measures of landscape disturbance to model the relationship between inferred habitat change and the abundance of marten. Our results were similar to those found by Webb and Boyce (2009), where the trapping o f marten was positively related to forest cover and inversely related to measures o f disturbance; however, those authors did not account for trapping effort. Our approach explicitly represented habitat change, rather than measures of disturbance and we tested a relatively full set o f factors that might influence trapping success. The methods employed in this study reduced many o f the biases and limitations associated with large data sets of harvest records, typically maintained by management 61 agencies. Working with individual trappers at the scale of the trapline, however, is time consuming, resulting in fewer data. Although the models identified statistically significant predictors of capture success, uncertainty in the model selection and limited predictive power may have been a result of these small sample sizes. Model selection was particularly uncertain for lynx in the East Study Area, where only 27 capture records were obtained. Future studies should include more trapper participants, increasing the sample size of trapping records as well as the spatial scale o f the study. Validation o f the habitat models with empirical data would also be beneficial (Kliskey et al. 1999). Our results suggested that the availability o f habitat influenced trapping success, likely representing changes in the distribution or abundance of the trapped populations following intensive forestry activity across the West Study Area. The spatial distribution o f trapping effort should be considered when attempting to understand the impacts o f habitat change. Over the short-term, habitat loss and population decline may not necessarily be represented by low capture success if trappers focus their efforts in large patches of remnant habitat; this may result in a lag in reduced capture rate and an apparent lag in population decline or collapse (Raphael 1994). We worked closely with the trappers to understand variation in effort across time and spatially across their respective traplines. Most trappers reported that on a yearly basis they trapped the same general locations, unless significant habitat disturbance occurred, in which case they would abandon those locations. Additionally, there was uncertainty as to whether trappers were focusing their efforts in locations that were representative of the highest quality habitat for marten and lynx. Wiebe et al. (2013) reported that trappers generally established marten sets in habitat that was selected for by marten; however, their study found that 62 trapping locations were highly influenced by road or trail access. During our consultations with trappers, we found that many trapping locations aligned with areas identified as highquality marten habitat on the expert-based habitat maps (see Chapter 2). For lynx, trappers appeared to be applying most o f their effort near or within regenerating forest stands; these locations were generally considered moderate lynx habitat, according to the habitat maps. Aside from effort, there are many other factors that may influence capture success, and ultimately the inferences that can be made to population abundance. Such factors include season length, quota changes, weather, access, trapper skills and motivation, trapping methods, activity on neighbouring traplines, and furbearer population dynamics (McDonald and Harris 1999, Poole and Mowat 2001, Ruette et al. 2003). Thus, one should be cautious when using trapping records as an index of population abundance. We attempted to identify and control for the major confounding factors by interviewing individual trappers. Most trappers reported that their capture success was affected by population abundance, effort, weather, and habitat availability. Catch per Unit Effort The count models suggested that capture success of lynx and marten, and ultimately population abundance, varied with habitat availability and quality. This conclusion is further demonstrated when evaluating CPUE and trapping effort for marten in the West Study Area. Since 1990, trappers increased their effort to capture marten in the West, yet CPUE decreased. Over this same time period, the expert-based habitat models predicted substantial declines in the availability and quality of marten habitat. Decreases in CPUE may suggest overharvest, but in this case, provides further evidence that marten populations may be declining in response to habitat loss. This relationship was not found for marten captures in 63 the East Study Area, where habitat had remained relatively stable, even increasing on certain traplines (see Chapter 2). For lynx captures in the West, however, CPUE increased over time, as did trapping effort. This result suggests that lynx populations in the West were not declining, or that increased trapping effort or efficiency masked any potential declines that may have occurred. Habitat Change and Population Dynamics of Furbearers Although many researchers would agree that lynx populations depend heavily on the availability of prey (Apps 2000, Poole 2003, Hoving et al. 2004, Simons-Legaard et al. 2013), the link between lynx populations and the availability and quality of habitat is not well quantified. Habitat for their primary prey species, snowshoe hare (Lepus americanus), is often associated with regenerating forests that occur following logging or wildfire (Mowat et al. 2000, Poole 2003, Simons-Legaard et al. 2013). Consequently, both snowshoe hare and lynx populations may benefit from landscape disturbance (Mowat and Slough 2003, Hoving et al. 2004). Less is known, however, about the dependence o f lynx populations on habitat associated with mature forest stands. Mature conifer or mixed forests provide valuable habitat for denning, resting, and cover from climatic conditions (Paragi et al. 1997). The loss o f such forests due to timber harvesting could have detrimental impacts to lynx at a population level. Lynx captures appeared to be high in the West Study Area despite a considerable reduction in old forests and apparent loss o f habitat. This suggests that lynx populations are not necessarily dependent on habitat that is associated with old forests; rather they are regulated by other factors, like the abundance o f snowshoe hare. Accounts from trappers suggest that both snowshoe hare and lynx populations are currently high in many portions of 64 the West Study Area. Conversely, lynx captures have been historically low in the East despite a relatively slow rate of forest harvesting; further evidence that lynx populations may not be limited by what experts perceive as quality habitat (see Chapter 2). Many studies have investigated the impacts o f industrial development on marten habitat (Thompson 1994, Chapin et al. 1998, Hargis et al. 1999, Potvin et al. 2000, Payer and Harrison 2003, Fuller and Harrison 2005, Steventon and Daust 2009, Cheveau et al. 2013); however, few have linked habitat change to population abundance. Marten are widely associated with old-growth coniferous stands (Payer and Harrison 2003, Proulx et al. 2006, Carroll 2007, Webb and Boyce 2009, Cheveau et al. 2013). These forests contain structural elements and canopy conditions that ameliorate extreme weather conditions and provide cover from predators, foraging habitat, and resting and denning sites (Thompson and Colgan 1994, Potvin et al. 2000, Carroll 2007, Cheveau et al. 2013). Extensive losses o f these latesuccessional forests due to timber harvesting may have direct impacts on marten populations (Steventon and Daust 2009, Webb and Boyce 2009). For example, Thompson (1994) found that marten had lower mean ages, lower reproductive success, and higher natural and trapping mortality across areas with a high level o f forest harvesting. Both the habitat and count models developed in this study provide further evidence that the loss of habitat may result in declines in the abundance o f marten populations. Trappers in the recently and extensively logged West Study Area reported that they were having greater difficulty capturing marten; when interviewed, they suggested that habitat loss was the driving force for low capture rates. The empirical results o f this study support those statements. Marten captures, and likely the density o f marten populations, varied with the availability o f habitat. The trappers in the East Study Area, where timber 65 harvesting has been limited, did not report reductions in capture success. Since 1990, marten captures were stable or increasing (Figure 3.1c). Conclusion The results of this study suggest that the capture records maintained by trappers have considerable utility for empirically documenting variation in furbearer abundance relative to cumulative habitat change. Focusing on both capture records and trapping effort at the scale o f the trapline avoids many of the pitfalls associated with the use of large-scale harvest databases typically maintained by management agencies (Smith et al. 1984, McDonald and Harris 1999, Poole and Mowat 2001). When trappers are engaged in the research or management process, records can be acquired at a low cost and over a relatively short time span. These data can be particularly important when studying cryptic species like furbearers (Ruette et al. 2003). Harvest records may be poor indicators o f short-term population trends (Erickson 1982, Poole and Mowat 2001); however, such records may be ideal for monitoring long-term trends provided researchers apply rigorous data management and control for influential factors, such as trapping effort. Count models suggested that after controlling for effort, habitat change did influence capture success. This is some of the first empirical evidence indicating that rapid and largescale forest harvesting can result in a decline in the abundance o f marten populations. This is supported both by theory and the expert knowledge of trappers. Where the rapid extraction of timber results in large cutblocks, and the loss of forest complexity and habitat corridors, marten populations are likely at risk. Wildlife and forest managers must consider the cumulative impacts of forest harvesting and perhaps alternative silviculture practices when 66 attempting to maintain large and widely distributed populations of marten (Payer and Harrison 2003). The influence of forest harvesting on the habitat and population dynamics of lynx is unclear. Although the empirical evidence indicates that lynx abundance varies with habitat availability, expert knowledge of the trappers suggests that lynx populations are primarily influenced by prey populations. Availability o f old-forest habitat may not necessarily promote lynx populations if prey species are not supported. Forest managers must consider promoting or enhancing habitat that benefits prey populations, often in the form of early serai forests, while maintaining adequate availability o f old-forest habitat. Management of lynx populations would benefit from further research investigating the relationships between lynx, forest structure, and prey abundance. 67 Chapter 4: General Summary and Management Considerations Summary I documented the cumulative impacts of landscape change on habitat availability and quality for three focal species (see Chapter 2). I developed expert-based habitat models to quantify habitat change in the central-interior of BC for fisher (Pekania pennanti), Canada lynx (Lynx canadensis), and American marten (Maries americana) since 1990. Throughout this process, I examined the utility o f expert-based habitat modeling, including the quantification of uncertainty. Across the West Study Area (see Figure 1.1), where recent forest harvesting was extensive, the availability and quality o f habitat for all three species decreased dramatically. This result contrasted with the East Study Area where there was relatively little forest harvesting and habitat availability and quality remained stable over time. These results suggest that intensive forestry negatively effects the habitat o f these three furbearer species in an immediate and cumulative manner. As a secondary outcome, this study illustrates the utility of expert knowledge for investigating the response of furbearers to cumulative landscape change. I hypothesised that the extensive loss of habitat would have population implications, thus, I used trapping records to investigate the relationship between habitat change resulting from cumulative impacts and population abundance of lynx and marten (see Chapter 3). I used count models to relate capture success to habitat change, while controlling for other influential factors, including trapping effort. In both study areas, habitat availability and quality, along with trapping effort and trapline area, were found to positively influence 68 capture success of lynx and marten. These results suggest that habitat change may directly affect the abundance o f lynx and marten in the study areas. These results also illustrate the utility of trapping records for investigating population dynamics of furbearers; however, a measure o f trapping effort is required to relate environmental covariates, including habitat change, to fur harvest at the scale o f individual traplines. Modeling and Predicting Habitat Change Expert-based habitat modeling required rigorous and defensible methods. The initial selection o f experts was a critical step. I developed explicit criteria for identifying experts and then applied a peer-referral technique to select biologist and trapper experts for the study. By identifying seed experts from three unique disciplines of biologists, I was able to recruit ten biologists from different backgrounds and with broad knowledge of furbearers. Seed experts from the local trapping community were also used to aid in the identification of ten suitable trapper experts that met the requirements for this study. This provided me with two distinct groups of experts that offered unique perspectives during the development o f the habitat models. This approach also allowed me to test expert agreement and uncertainty within and between expert groups during the model development stages. The total number of experts was logistically manageable, allowing for personal instruction in the elicitation method, as well as direct and timely feedback on study progress and findings, while providing a wide breadth of knowledge of furbearer-habitat relationships. I assumed that the relatively large sample of experts prevented bias and influential error from any one individual (McBride and Burgman 2012). My study design allowed the experts to participate and guide all stages o f the development of the habitat models. Initially, they voted for three focal species that were 69 considered to be economically important to the fur trapping industry and were hypothesised to be sensitive to landscape change associated with natural or anthropogenic disturbance. Fisher, lynx, and marten were selected as the focal species. Although similar in their general biology, these species use different habitats, and thus may be affected differently by cumulative landscape change. The experts also selected the variables to be included in the habitat models. Overall, there was high consistency between experts both within and between groups when identifying variables hypothesised to influence the distribution o f the three focal species; 11 variables were identified for the fisher and marten models, while ten variables were identified for the lynx model. Uncertainty in the selection of habitat variables by biologists was highest for lynx and lowest for fisher. For trappers, uncertainty was highest for fisher and lowest for marten. Using the analytical hierarchy process (AHP), experts then evaluated the relative importance o f each subcategory of variables for predicting the distribution of habitat. Elicitation o f scores for each variable occurred with minimal operative errors and relatively high consistency within and between experts. When compared to biologist experts, trappers reported a higher confidence in the ranking o f their scores; however, there was higher variation in the eigenvector scores generated using the variable rankings provided by the trappers (product of the AHP matrices representing the relative value of the habitat variables). I used geographic information systems (GIS) to apply the expert-based habitat models to ten reference landscapes (i.e., registered traplines). I developed a chronology o f maps showing habitat change across each reference landscapes at four time intervals (i.e., 1990, 2000, 2005, and 2013). According to the models, there were significant declines in habitat 70 for fisher, lynx, and marten in the West Study Area, where high levels o f forestry had occurred over the study period. The models predicted relatively little change in habitat for all three species in the East Study Area, where landscape changes have been minimal since 1990. I evaluated the uncertainty in expert-based models and the resulting habitat maps. I recreated the maps using the upper and lower 95th percentile eigenvector scores resulting from the AHP elicitation. I also created habitat maps based on the biologist and trapper scores separately. This assessment suggested that discrepancies between experts can influence the prediction of habitat area; however, a relatively high consistency of habitat scores between expert groups resulted in no significant changes to the conclusions o f this study. As a form of model validation, the trappers evaluated the distribution and area of ranked habitat on their individual traplines. They reported highest accuracy scores for maps of marten habitat and the lowest for lynx. The low score for the lynx maps may have resulted from variation in the prediction of lynx habitat, due to their propensity to use varying serai stages and habitat types. Additionally, the lynx habitat model may have been overly influenced by the attributes o f old-forest stands, resulting in an over-prediction o f habitat quality in areas of late-successional forests. Population Modeling I used trapping records maintained by each of the ten participating trappers to examine the relationship between lynx and marten population dynamics and habitat change. When using capture success as a proxy for population abundance one must control for trapping effort. Along with effort, I identified other factors that might explain capture success, including habitat availability and quality, trapline area, and climatic conditions. 71 Determining the influence o f cumulative landscape change on trapping success was of particular interest. Thus, I tested three variables representing change in habitat availability and quality over time, derived from the expert-based habitat models (see Chapter 2). I used NBRMs to relate harvest data from individual traplines to factors hypothesised to influence capture success. I developed sets of a priori candidate models and used AIC to identify the most parsimonious model from each set o f explanatory hypotheses. The most parsimonious models explaining lynx captures in the West Study Area and marten captures in the East and West included a measure o f trapping effort, habitat availability and quality, and trapline area; in all cases, these three predictor variables had a significantly positive influence on capture success. Weather variables were also included in the most parsimonious models for lynx captures in the East and West, and marten captures in the West, where extreme or mean minimum temperatures had a significantly negative influence on marten captures in the West Study Area and lynx captures in the East. I assessed the predictive ability of the top count models by testing for differences between observed and predicted harvest data and found that the models had relatively low predictive power. An increase in the number of participating trappers and resulting number of trapping records might improve the predictive ability of the count models. Management Concerns and Recommendations The results of this study provide insights into the influence o f cumulative impacts of landscape change on the abundance and distribution o f furbearers in central-interior BC. To help guide management recommendations that minimize future cumulative impacts on habitat and populations, I conducted surveys with the ten biologist experts and nine o f the 72 trapper experts to discuss management concerns and recommendations. Semi-structured interviews were conducted via telephone with the trappers and semi-structured surveys were sent via email to the biologists. The interviews and surveys were structured to guide discussions o f key concerns and recommendations toward habitat and population management for the focal species, as well as for the broader group o f furbearers in the province. Fisher The biologists and trapper experts identified several key themes important to the management of fisher habitat (Table 4.1). Most biologists and trappers identified old deciduous trees (i.e., cottonwood and poplar) as limiting habitat features for fisher, as the rearing of young takes place primarily in such trees (Lofroth et al. 2010). Although mature deciduous trees are not generally targeted for timber harvest, the experts reported that they are often removed incidentally, or lost due to wind-fall in exposed logged areas. At a landscape scale, the experts stressed the importance o f habitat connectivity in areas affected by extensive timber harvesting. The trappers were particularly concerned with the large size of the cutblocks. These experts may have implicitly recognized the large home range sizes and naturally low population densities of fisher (Weir and Almuedo 2010). The experts reported that extensive salvage logging of forest stands killed by mountain pine beetle (MPB) has resulted in a high density o f cutblocks, pine-dominated forests that lack structure, and a lack o f contiguous mature forest. The biologists and trappers recommended that forest managers maintain linkage zones and corridors across the landscape, limit the size of cutblocks, and reduce the rate of timber harvest. 73 Table 4.1. Recommendations by biologist and trapper experts for maintaining habitat and numbers o f fisher in the central-interior o f BC, Canada. Recommendations were obtained through semi-structured interviews and surveys. Habitat Concerns Habitat Recommendations Loss of denning trees (old deciduous) as a result o f forestry practices. Maintain and promote large, deciduous trees for denning sites; experiment with man-made structures as possible denning locations (i.e., denning boxes). High density of disturbances and large cutblock sizes across the landscape resulting in the loss of mature forests, reduced habitat connectivity, and increased patch isolation. Focus habitat management on maintaining corridors and linkages; use logging prescriptions that reduce habitat fragmentation across landscapes. High rate of timber harvest associated with MPB salvage logging and the promotion of future pinedominated forests lacking structure. Promote landscape heterogeneity, including prey habitat; reduce the rate of timber harvesting. Lack of CWD retention in cutblocks. Retain CWD and dead standing trees to promote future habitat as cutblocks regenerate. Broadcast herbicide use as a silviculture practice resulting in loss of cover and prey habitat. Reduce or eliminate the use o f aerial herbicides as a silviculture practice. Population Concerns Population Recommendations Populations likely in decline resulting from MPB salvage logging. See habitat recommendations for promoting fisher habitat. Lack of population connectivity, particularly with low densities and large home ranges; population isolation could result in complete losses of populations. See habitat recommendations for promoting habitat/population connectivity. Trapping may be an additive pressure on sensitive populations; uncertainty surrounding the sustainability of fisher harvest. Trapping management or restrictions may be required in areas with small or declining populations; management of captures on individual traplines; minimize trapper by-catch in marten and lynx traps (i.e., restriction plates on marten traps). Lack of population monitoring to observe trends. Better inventory o f population numbers and trends; increase in mark-recapture studies or analysis o f trapping records. Climate change could have a positive effect on populations if a reduction in snow cover occurs. Further research on impacts of climate change on fisher populations. 74 Relative to elevated levels of forest harvesting across the central-interior o f BC, experts were also concerned with a failure to retain sufficient amounts o f coarse woody debris (CWD) in recent cutblocks. The experts suggested that greater amounts o f CWD should be retained and that this strategy would promote future habitat as the cutblocks regenerate. Several trappers were also concerned with broadcast spraying of herbicides as a silviculture practice in regenerating cutblocks. They suggested that this practice decreases the diversity and structure o f future habitat and reduces the quality of habitat for the prey of fisher. Finally, several trappers recognized a lack of coordination of management plans between multiple industrial sectors, as well as between industry and wildlife managers. Many o f the experts suggested that fisher populations are declining in the centralinterior o f BC, particularly where populations overlap with MPB salvage logging. They reported that the decline was a function of a loss o f optimal habitat and a lack o f population connectivity across the landscape. Some experts felt that population isolation due to fragmented landscapes could result in complete losses of fisher populations. Most experts were uncertain when describing the potential impacts of trapping on fisher populations, as there is a lack o f monitoring. Most biologists suggested that trapping may be an additive source of mortality for sensitive populations; however, the trappers were generally unconcerned about the impacts o f trapping, likely due to low capture rates. Few trappers targeted fisher with most captures being incidental in marten or lynx traps. The biologists recommended better use of trapping records to monitor harvest and population trends. Also, they recommended that trapping restrictions be implemented in areas where populations are thought to be sensitive to overharvest or other anthropogenic impacts. 75 Lynx The biologists and trappers identified both positive and negative influences of landscape change on lynx habitat (Table 4.2). Most experts agreed that early serai forests resulting from logging or forest fires could potentially benefit lynx through the promotion of prey habitat; however, trappers were concerned about the length o f time required before these habitats become optimal. Most experts recommended implementing forestry practices that promote heterogeneous landscapes, and provide early serai prey habitat. Both expert groups were unsure o f the importance of old-forest stands and attributes for the productivity of lynx populations; several biologists stated a need for further research on lynx and their habitat requirements. Currently, there have been no relevant studies o f lynx habitat requirements in the central-interior of BC. Many biologists and trappers suggested that there is a lack of mature forest attributes which may provide habitat for cover, resting, and denning, and such areas must be preserved as refuge habitat. Many trappers were concerned with the aerial broadcast spraying of herbicides, suggesting that this practice significantly impacts prey habitat in regenerating forests; they recommended the elimination of herbicides as a silviculture practice. Most experts agreed that in the central-interior o f BC lynx and hare populations cycle closely. Thus, as regenerating forests continue to provide ample habitat for hare, there may be little reason to consider the cumulative impacts of industrial development relative to the broad-scale habitat needs of lynx. Multiple trappers suggested that the amplitude and period of the hare population cycle, and subsequent lynx cycle, has stabilised as a result o f increased hare habitat. Several biologists stated, however, that there is a lack of knowledge of lynx population dynamics in the northern portion o f their range. 76 Table 4.2. Recommendations by biologist and trapper experts for maintaining habitat and numbers o f lynx in the central-interior o f BC, Canada. Recommendations were obtained through semi-structured interviews and surveys. Habitat Recommendations Habitat Concerns Large-scale habitat alterations, including large cutblocks, promoting homogenous landscapes. Implement forestry practices that promote heterogeneous landscapes and natural forest dynamics (i.e., young forests for prey habitat, mixed forests for bedding habitat, and mature forests for denning and cover habitat). Distribution of lynx is heavily reliant on the availability o f prey habitat; regenerating cutblocks have slow recovery times before becoming suitable prey habitat. Implement forestry and silviculture practices (i.e., reforestation, broadcast burning, and late thinning) that rapidly promote early serai habitat for snowshoe hare populations. Uncertainty surrounding the amount of elemental attributes required to sustain lynx populations; loss of mature forests and a lack of old-forest attributes on landscape. Increased research on interactions of lynx with forestry practices; maintain large refuge areas of habitat; preserve mature forests. Broadcast use of aerial herbicides reducing prey habitat and cover. Reduce or eliminate the use of aerial herbicides on regenerating cutblocks. Population Concerns Population Recommendations Lynx populations may not be at risk where snowshoe hare populations are abundant; populations may increases in areas that have been subjected to logging or wildfires, however, it may take 20-30 years post-disturbance. See habitat recommendations for promoting prey habitat. There is uncertainty surrounding the impacts of trapping; fur prices may lead to high trapping pressure, however northern populations may not be susceptible to over-trapping, due to wide distribution and nomadic behavior. Harvest management may not be necessary in northern portions of their range; restrict harvest (trapping and access) in southern portion of range (southern BC); harvest should be reduced during low periods in the population cycle to increase recovery speed. Currently, government has a passive approach to harvest management and it is unknown whether that is acceptable; lack of population monitoring and trapping data usage. Implement better use of harvest data for population abundance and population trends. Lack of knowledge of population dynamics in northern part of range. Increased research on population dynamics of lynx and the differences between populations inhabiting boreal and mountain ecosystems. 77 They recommended better use of trapping records to monitor populations, and an overall increase in baseline population monitoring and research. Many experts suggested uncertainty surrounding the impacts o f trapping on lynx populations. Some biologists reported that trapping may be of concern to sensitive lynx populations in southern BC, but not necessarily in their northern range. They felt that lynx populations could benefit from reduced trapping pressure during low population cycles or restricted trapping of sensitive populations. In contrast, most trappers stated that lynx are insensitive to over-trapping due to their large home ranges and nomadic behavior. Some biologists suggested that there is currently a ‘hands o ff management approach to regulating lynx trapping in the northern half of the province; it is unclear whether this approach is acceptable. Most trappers recognized the importance o f population management by trappers on individual traplines, or across multiple traplines. Marten The experts identified several concerns and recommendations for the management o f marten habitat relative to cumulative landscape change (Table 4.3). At a landscape scale, the experts were concerned about extensive reductions in contiguous mature and old-growth conifer forests due to forestry, resulting in fragmented habitat and direct losses o f critical habitat features. This was o f particular concern for areas subjected to MPB salvage logging, where habitat may have been diminished to an extent that no longer supports viable marten populations. Trappers reported the negative impacts associated with large cutblocks resulting from salvage logging. They suggested a reduction in the size o f cutblocks and the rate of forest harvest. Most experts recognized the importance o f preserving contiguous tracts of old and mature forest that provides habitat connectivity. 78 Table 4.3. Recommendations by biologist and trapper experts for maintaining habitat and numbers o f marten in the central-interior o f BC, Canada. Recommendations were obtained through semi-structured interviews and surveys. Habitat Concerns Habitat Recommendations Loss of habitat connectivity due to forestry impacts; loss of large, contiguous patches of oldgrowth forest. Maintain old- and mid-age conifer forests, and habitat connectivity and corridors. Loss o f important habitat structures, including CWD; lack of CWD retention in cutblocks. Retain CWD and large-diameter standing trees within and around cutblocks to promote future habitat. Replanting of lodgepole pine resulting in homogenous forests that lack complexity. Stand-level forest practices to maintain or enhance structure; promote old-forest habitat attributes in younger stands to reduce recovery time. Large cutblock sizes associated with MPB salvage logging; timber being harvested at an unsustainable rate. Smaller cutblock sizes; reduce the rate of timber harvest. Aerial herbicide use reduces structure, diversity, and prey habitat in regenerating cutblocks. Eliminate or reduce the use of aerial herbicides. Lack of research on population-level impacts of habitat change. Further research on impacts of large-scale habitat loss, and ability o f marten to inhabit younger forests. Population Recommendations Population Concerns Likely a decline in marten populations in centralinterior BC due to extensive forest harvesting. See habitat recommendations for promoting marten habitat. Risk of population fragmentation across disturbed landscapes. See habitat recommendations for maintaining habitat/population connectivity. Populations may be at relatively low risk province-wide given high abundance and wide distribution o f marten, however, local populations may be at risk; lack of population monitoring and research investigating the impacts of trapping. Increase research and population monitoring (i.e., mark-recapture estimates, or trapper samples); implement better use of trapping records for monitoring population trends. Risk of over-trapping sensitive populations, particularly on small traplines; fur prices may drive trapping pressure. Management of harvest on individual traplines is important; trappers should employ a cautious approach. Trapping may be an additive pressure on sensitive populations Establish refugia, free from trapping pressure. 79 Both expert groups reported a heavy dependency o f marten on old-forest habitat features for denning, resting, cover, and foraging. They suggested that these features may be limiting on industrially impacted landscapes. Many experts stressed the importance of preserving old forest stands that provide important elemental habitat features, such as CWD and structural complexity. Some biologists stated that the small home ranges o f marten should allow for the successful implementation o f stand-level management. They suggested that certain forestry practices can retain or promote the necessary elemental features and structure required by marten, reducing the recovery time o f post-disturbance forests; a suggestion that is supported by previous research (Payer and Harrison 2003, Poole et al. 2004). Most biologists and trappers acknowledged that large tracts of marten habitat had been lost; therefore, there should be an emphasis on establishing practices that promote future habitat. This included the retention of CWD and dead standing trees in cutblocks. Several trappers suggested that slash piles within cutblocks should be left intact to provide sub-nivean access and prey habitat. Many trappers also suggested that aerial herbicides reduced the structural diversity, cover, and prey habitat in regenerating forest stands. Some experts cautioned against replanting cutblocks with primarily lodgepole pine, as this promotes future forests with simplified forest structure unsuitable for marten. Many biologists recognized a need for research on the ability o f marten to use younger forests, as well as further research on the impacts of large-scale habitat change. Several trappers recommended increased coordination o f management plans between industrial sectors to reduce cumulative impacts, and stated that there should be more consultation by industry with furbearer experts. 80 Both expert groups were in agreement that there is likely a decline in marten populations in the central-interior of BC where MPB salvage logging has occurred. Several biologists, however, suggested that marten populations at a broad scale are at relatively low risk, due to their overall abundance, wide distribution, availability o f residual habitat, and small home-range sizes. Most experts agreed that habitat loss was the primary concern, but several suggested that populations would recover over time as forests regenerated. Many biologists felt that local populations may be at risk of over-trapping, although there is currently a lack o f population monitoring, thus the impacts of trapping are relatively unknown. They recommended establishing refugia, free from trapping pressure. The biologists also supported increased research and population monitoring. Some trappers felt that there is a risk of over-trapping on individual traplines, particularly if harvest by trappers is mismanaged or high fur prices promotes increased trapping pressure. Conversely, several trappers suggested that marten are at low risk of over-trapping. Most trappers, however, agreed that managing marten harvest on individual traplines was important. Furbearers in General The experts also commented on management concerns and recommendations for the broader group of furbearers in BC (Table 4.4). Both expert groups recognized a lack of mandates for industry to manage, maintain, and promote furbearer habitat. Most trappers were particularly concerned with the rate o f timber harvesting, the size o f the cutblocks, and the overall loss of optimal habitat and travel corridors. They also suggested that the current minimum requirements for industry to preserve or promote habitat are inadequate. 81 Table 4.4. Recommendations by biologist and trapper experts for maintaining habitat and numbers o f the broader group o f furbearers in the central-interior o f BC, Canada. Recommendations were obtained through semi-structured interviews and surveys. Habitat Concerns Habitat Recommendations Few mandates for industry to manage, maintain, or promote furbearer habitat; current requirements may be insufficient. Habitat management is important; priority should be placed on preserving old-growth forests. Cumulative impacts from multiple, competing resource sectors. Mitigate cumulative impacts; increase coordination amongst industry and consultation with furbearer experts. Habitat loss resulting from rapid timber harvesting; large cutblocks and lack of corridors. Forest managers must maintain refuge habitats and landscape-level connectivity. Reforestation can be a slow, lengthy process. Habitat must be given sufficient time to regenerate before further alterations occur. Simplified forest structures resulting from intensive forest harvesting and silviculture practices. Modify harvest and silviculture practices to retain natural patterns across the landscape. Lack o f knowledge of habitat requirements for furbearers. Increased research pertaining to habitat requirements of furbearers and impacts of habitat change. Population Concerns Population Recommendations Populations may be affected at different spatiotemporal scales; population impacts may be greater at regional scales. Habitat management should be a priority for maintaining furbearer populations. Lack of population monitoring and research; trapping data is currently underutilized. Increase scrutiny and use of trapping data; increase mandatory trapping reporting or inspections to collect population data, particularly for species at risk. Increased trapper access; trapping pressure may increase with fur prices. Restrict trapping access and pressure on sensitive populations. Trapping may not be a concern for the persistence of ftirbearer species at the broad-scale, but may have implications on local populations. Increased public and trapper education on population management, particularly on individual traplines. 82 Furthermore, they reported that a lack o f coordination within and between industrial sectors relative to resource extraction is likely to result in cumulative impacts that reduce the extent and quality of furbearer habitat. Several biologists acknowledged that the loss o f habitat associated with industry is often rationalized by eventual reforestation; however, this process may occur slowly over a long period o f time. Additionally, many of the experts suggested that timber harvesting and subsequent silviculture practices are promoting the regeneration of simplified forest stands resulting in unproductive furbearer habitat. Most trappers reported that the use of aerial herbicides, as a silviculture practice, is detrimental to furbearer habitat. The forest industry should implement harvesting and silviculture practices that mimic natural landscape changes, while preserving critical habitats. Both expert groups emphasized the importance of protecting old-growth forests and maintaining refuge areas. Several biologists proposed a multi-species approach to habitat management, rather than managing for individual species. This may reduce the complexity o f developing and implementing multiple management plans while reducing problems associated with competing habitat requirements of single species. Forestry and other industrial activities have impacts that vary across different spatiotemporal scales. At a province-wide scale, many biologists suggested that most furbearer species are not o f conservation concern. At a regional scale, however, the experts agreed that local populations may be at risk of declines. The majority o f experts agreed that the most productive method o f supporting furbearer populations is through habitat management. 83 Although there was uncertainty amongst experts on the impacts of trapping on furbearer populations, most agreed that trapping at a broad scale has minimal impacts. Many experts, however, stated that the combination of habitat loss and trapping at a regional scale could be detrimental to local populations. Several biologists suggested restricting trapping opportunities for sensitive furbearer populations. There was a plea from most experts for an increase in research and population monitoring of furbearers in BC; further development of the use o f trapping records and trapper-kills would be beneficial. Research Conclusions The results o f this study suggest that the cumulative effects o f forest harvesting can have considerable impacts on the abundance and distribution o f furbearer species. There was strong empirical evidence that the recent and rapid loss o f old forest across the West Study Area resulted in declines in furbearer habitat, and subsequently declines in population abundance, particularly for marten. Reports from trappers of high lynx numbers (also evident in the trapping records), however, suggested that lynx populations are stable or increasing despite a loss of mature-forest habitat. This finding suggested that lynx may be food-limited, rather than habitat-limited. In the East Study Area, where forest harvesting has been minimal, the results suggested that stable or increasing habitat availability and quality is promoting abundant furbearer populations. This finding was supported by high capture success for marten over time. Conversely, the trappers reported low numbers o f lynx in the East Study Area despite limited forest harvesting and plentiful old-forest habitat. Again, this suggested that lynx were not necessarily dependent on old-forest habitat, but were influenced by other factors, including the availability of prey and prey habitat (early successional 84 forests). The influence o f forest harvesting on the habitat and population abundance o f lynx remains unclear and warrants further research. This study contributes to a growing body of ecological literature that validates the use and advantages o f expert-based studies (Burgman et al. 201 la, McBride and Burgman 2012, Drescher et al. 2013). In this case, experts allowed the rapid and inexpensive development of models for predicting cumulative change in the availability o f habitat for three furbearer species. Additionally, after controlling for trapping effort, capture records maintained by trappers had considerable utility for documenting the numerical response o f furbearer populations to habitat change. As demonstrated by others, expert-based habitat modeling can serve as an efficient and rapid method of documenting species distribution, particularly for cryptic species that are difficult to study (Store and Kangas 2001, Yamada et al. 2003, O ’Neill et al. 2008). Involving multiple groups of experts can provide unique, but complementary domains of expertise that minimizes bias and provides a fuller description o f species-habitat relationships. The habitat models developed in this study had remarkably high consistency, resulting in model structure that was very similar among expert groups. The consistent results were likely the product of the development o f a rigorous study design that included the identification o f suitable experts, the application of an easily understood elicitation process, and the full and instructed involvement o f the experts throughout the study. The use of trapping records at the scale of the trapline appeared to be a reasonable method for documenting population dynamics, avoiding many of the pitfalls associated with large-scale harvest databases often generated by management agencies. Quantifying trapping 85 effort was an important step to determining the influence o f environmental covariates on capture success, and ultimately population abundance. Although harvest records may be poor indicators of short-term population dynamics, the use of such records may be ideal for monitoring long-term population trends of furbearers. For this study, I correlated harvest records with a time series o f habitat change that occurred since 1990. The application of rigorous and repeatable methods was essential for meeting the study objectives that were primarily focused on the quantification of habitat change and resulting population responses o f fisher, lynx, and marten. Application of the study design was meant to generate confidence in the results, but also stimulate discussion and perhaps improve management o f furbearers distributed across rapidly changing landscapes. Ultimately, these study findings are transparent and defensible and can be used by wildlife and forest managers to guide habitat and population management and strategic decisions relative to cumulative landscape change. References Apps, C.D. 2000. Space-Use, Diet, Demographics, and Topographic Associations of Lynx in the Southern Canadian Rocky Mountains: A Study. Pages 351-372 in L.F. Ruggiero, K.B. Aubry, S.W. Buskirk, G.M. Koehler, C.J. Krebs, K.S. McKelvey, J.R. Squires, editors. Ecology and conservation of lynx in the United States. University Press o f Colorado, Denver, CO. B.C. Ministry of Forests, Mines and Lands. 2010. The state of British Columbia’s forests, 3rd ed. Forest Practices and Investment Branch, Victoria, BC. (www.for.gov.bc.ca/hfp/sof/index.htm# 2010_report; accessed January 2013). Burgman, M.A., A. Carr, L. Godden, R. Gregory, M. McBride, L. Flander, and L. Maguire. 201 la. Redefining expertise and improving ecological judgment. Conservation Letters 4:81-87. 86 Burgman, M.A., M. McBride, R. Ashton, A. Speirs-Bridge, L. Flander, B. Wintle, F. Fidler, L. RumpfF, and C. Twardy. 201 lb. Expert status and performance. Public Library of Science ONE 6: e22998. doi: 10.1371/joumal.pone.0022998 Burnham, K.P., and D.R. Anderson. 2002. Model selection and multimodel inference: a practical information-theoretic approach. 2nd ed. Springer-Verlag, New York, NY. Buskirk, S.W. 1992. Conserving circumpolar forests for martens and fishers. Conservation Biology 6:318-320. Buskirk, S.W., H.J. Harlow, and S.C. Forrest. 1988. Temperature regulation in American marten (Martes americana) in winter. National Geographic Research 4:208-218. Buskirk, S.W., L.F. Ruggiero, and C.J. Krebs. 2000. Habitat Fragmentation and Interspecific Competition: Implications for Lynx Conservation. Pages 83-100 in L.F. Ruggiero, K.B. Aubry, S.W. Buskirk, G.M. Koehler, C.J. Krebs, K.S. McKelvey, J.R. Squires, editors. Ecology and conservation of lynx in the United States. University Press of Colorado, Denver, CO. Carroll, C. 2007. Interacting effects o f climate change, landscape conversion, and harvest on carnivore populations at the range margin: marten and lynx in the northern Appalachians. Conservation Biology 21:1092-1104. Cattadori, I.M., D.T. Haydon, S.J. Thirgood, and P.J. Hudson. 2003. Are indirect measures o f abundance a useful index of population density? The case of red grouse harvesting. Oikos 100:439-446. Chapin, T.G., D.J. Harrison, and D.D. Katnik. 1998. Influence o f landscape pattern on habitat use by American marten in an industrial forest. Conservation Biology 12:1327-1337. Chatteijee, S., and A.S. Hadi. 2006. Regression Analysis by Example, 5th ed. John Wiley and Sons, Hoboken, NJ. Cheveau, M., L. Imbeau, P. Drapeau, and L. Belanger. 2013. Marten space use and habitat selection in managed coniferous boreal forests of eastern Canada. Journal o f Wildlife Management 77:749-760. Clark Labs. 2011. IDRISI Selva 17.02. Clark Labs, Clark University, Worcester, MA. Clark, D.F., D.D. Kneeshaw, P.J. Burton, and J.A. Antos. 1998. Coarse woody debris in subboreal spruce forests of west-central British Columbia. Canadian Journal of Forest Research 28:284-290. 87 Clevenger, A.P., J. Wierzchowski, B. Chruszcz, and K. Gunson. 2002. GIS-generated, expert-based models for identifying wildlife habitat linkages and planning mitigation passages. Conservation Biology 16:503-514. Dormann, C.F, J.M. McPherson, M.B. Araujo, R. Bivand, J. Bolliger, G. Carl, R.G. Davies, A. Hirzel, W. Jetz, W.D. Kissling, I. Kuhn, R. Ohlemuller, P.R. Peres-Neto, B. Reineking, B. Schroder, F.M. Schurr, and R. Wilson. 2007. Methods to account for spatial autocorrelation in the analysis of species distributional data: a review. Ecography 30:609-628. Doswald, N., F. Zimmerman, and U. Breitenmoser. 2007. Testing expert groups for a habitat suitability model for lynx Lynx lynx in the Swiss Alps. Wildlife Biology 13:430-446. Drescher, M., A.H. Perera, C.J. Johnson, L.J. Buse, C.A. Drew, and M.A. Burgman. 2013. Toward rigorous use o f expert knowledge in ecological research. Ecosphere 4: 83. http://dx.doi.Org/10.1890/ES12-00415.l Environment Canada. 2014. Historical Climate Data, (http://climate.weather.gc.ca/; accessed July 2014). Environmental Systems Research Institute (ESRI). 2012. ArcGIS 10.1. Environmental Systems Research Institute Inc., Redlands, CA. Erickson, D.W. 1982. Estimating and using furbearer harvest information. Pages 54-65 in G.C. Sanderson, editor. Midwest furbearer management. Proceedings o f a symposium held at the 43rd Midwest Fish and Wildlife Conference, Wichita, KS. Fuller, A.K., and D.J. Harrison. 2005. Influence of partial timber harvesting on American martens in north-central Maine. Journal o f Wildlife Management 69:710-722. Guisan, A., and W. Thuiller. 2005. Predicting species distribution: offering more than simple habitat models. Ecology Letters 8:993-1009. Guisan, A., and N.K. Zimmerman. 2000. Predictive habitat distribution models in ecology. Ecological Modelling 135:147-186. Hamilton, D.A., B. Roberts, G. Linscombe, N.R. Jotham, H. Noseworthy, and J.L. Stone. 1998. The European Union’s wild fur regulation: a battle of politics, cultures, animal rights, international trade and North America’s wildlife policy. Transactions of the North American Wildlife and Natural Resources Conference 63:572-588. Hargis, C.D., J.A. Bissonette, and D.L. Turner. 1999. The influence o f forest fragmentation and landscape pattern on American martens. Journal of Applied Ecology 36:157-172. 88 Hatler, D.F., and A.M.M. Beal. 2003. Furbearer management guidelines: Lynx Lynx canadensis. Province o f BC and Habitat Conservation Trust Foundation. (http://www.env.gov.bc.ca/fw/wildlife/trapping/docs/lynx.pdf; accessed January 2013). Hoving, C.L., D.J. Harrison, W.B. Krohn, W J. Jakubas, and M.A. McCollough. 2004. Canada lynx Lynx canadensis habitat and forest succession in northern Maine, USA. Wildlife Biology 10:285-294. Hurley, M.V., E.K. Rapaport, and C.J. Johnson. 2009. Utility o f expert-based knowledge for predicting wildlife-vehicle collisions. Journal o f Wildlife Management 73:278-286. Johnson, C.J. 2011. Regulating and Planning for Cumulative Effects: The Canadian Experience. Pages 29-46 in P.R. Krausman and L.K. Harris, editors. Cumulative effects in wildlife management: impact mitigation. CRC Press, Boca Raton, FL. Johnson, C.J., and M.P. Gillingham. 2004. Mapping uncertainty: sensitivity of wildlife habitat ratings to expert opinion. Journal o f Applied Ecology 41:1032-1041. Johnson, C.J., M. Hurley, E. Rapaport, and M. Pullinger. 2012. Using Expert Knowledge Effectively: Lessons from Species Distribution Models for Wildlife Conservation Management. Pages 153-171 in A.H. Perera, C.A. Drew, and C.J. Johnson, editors. Expert knowledge and its application in landscape ecology. Springer, New York, NY. Kliskey, A.D., E.C. Loffoth, W.A. Thompson, S. Brown, and H. Schreier. 1999. Simulating and evaluating alternative resource-use strategies using GIS-based habitat suitability indices. Landscape and Urban Planning 45:163-175. Lofroth, E.C., C.M. Raley, J.M. Higley, R.L. Truex, J.S. Yaeger, J.C. Lewis, P.J. Happe, L.L. Finley, R.H. Naney, L.J. Hale, A.L. Krause, S.A. Livingston, A.M. Myers, and R.N. Brown. 2010. Conservation o f fishers (Maries pennanti) in south-central British Columbia, western Washington, western Oregon, and California Volume I: Conservation Assessment. USDI Bureau o f Land Management, Denver, CO. Martin, T.G., M.A. Burgman, F. Fidler, P.M. Kuhnert, S. Low-Choy, M. McBride, and K. Megersen. 2011. Eliciting expert knowledge in conservation science. Conservation Biology 26:29-38. McBride, M.F., and M.A. Burgman. 2012. What Is Expert Knowledge, How Is Such Knowledge Gathered, And How Do We Use It To Address Questions In Landscape Ecology? Pages 11-32 in A.H. Perera, C.A. Drew, and C.J. Johnson, editors. Expert knowledge and its application in landscape ecology. Springer, New York, NY. 89 McCann, N.P., P.A. Zollner, and J.H Gilbert. 2014. Bias in the use o f broadscale vegetation data in the analysis o f habitat selection. Journal o f Mammalogy 95:369-381. McDonald, R.A., and S. Harris. 1999. The use o f trapping records to monitor populations of stoats Mustela ermine and weasels M. nivalis: The importance o f trapping effort. Journal of Applied Ecology 36:679-688. Meidinger, D.V., and J. Pojar. 1991. Ecosystems o f British Columbia. Ministry o f Forests, Research Branch, Victoria, BC. Microsoft Corporation. 2010. Microsoft Excel 14.0. Microsoft Corporation, Redmond, WA. Mowat, G., K.G. Poole, and M. O’Donoghue. 2000. Ecology of Lynx in Northern Canada and Alaska. Pages 265-306 in L.F. Ruggiero, K.B. Aubry, S.W. Buskirk, G.M. Koehler, C.J. Krebs, K.S. McKelvey, J.R. Squires, editors. Ecology and conservation of lynx in the United States. University Press o f Colorado, Denver, CO. Mowat, G., and B. Slough. 2003. Habitat preference o f Canada lynx through a cycle in snowshoe hare abundance. Canadian Journal o f Zoology 81:1736-1745. Nitschke, C.R. 2008. The cumulative effects o f resource development on biodiversity and ecological integrity in Peace-Moberly region of northeast British Columbia, Canada. Biodiversity and Conservation 17:1715-1740. O’Neill, S.J., T.J. Osbom, M. Hulme, I. Lorenzoni, and A.R. Watkinson. 2008. Using expert knowledge to assess uncertainties in future polar bear populations under climate change. Journal o f Applied Ecology 45:1649-1659. Paragi, T.F., W.N. Johnson, D.D. Katnik, and A.J. Magoun. 1997. Selection of post-fires seres by lynx and snowshoe hares in the Alaskan taiga. Northwestern Naturalist 78:7786 . Payer, D.C., and D.J. Harrison. 2003. Influence o f forest structure on habitat use by American marten in an industrial forest. Forest Ecology and Management 179:145— 156. Poole, K.G. 2003. A review of Canada lynx, Lynx canadensis, in Canada. The Canadian Field-Naturalist 117:360-376. Poole, K.G., and G. Mowat. 2001. Alberta furbearer harvest data analysis. Alberta Sustainable Resource Development, Fish and Wildlife Division, Alberta Species at Risk Report No. 31, Edmonton, AB. 90 Poole, K.G., A.D. Porter, A. de Vries, C. Maundrell, S.D. Grindal, and C.C. St. Clair. 2004. Suitability o f a young deciduous-dominated forest for American marten and the effects of forest removal. Canadian Journal of Zoology 82:423-435. Potvin, F., L. Belanger, and K. Lowell. 2000. Marten habitat selection in a clearcut boreal landscape. Conservation Biology 14:844-857. Proulx, G. 2000. The Impact o f Human Activities on North America Mustelids. Pages 53-75 in H.I. Griffiths, editor. Mustelids in a modem world: management and conservation aspects o f small carnivore: human interactions. Backhuys Publishers, Leiden, The Netherlands. Proulx, G. 2006. Using forest inventory data to predict winter habitat use by fisher Martes pennanti in British Columbia, Canada. Acta Theriologica 51:275-282. Proulx, G. 2009. Conserving American marten Martes americana winter habitat in subboreal spruce forests affected by mountain pine beetle Dendroctonus ponderosae infestations and logging in British Columbia, Canada. Small Carnivore Conservation 41:51-57. Proulx, G. 2011. Verification of a forest rating system to predict fisher, Martes pennanti, winter distribution in sub-boreal forests o f British Columbia, Canada. The Canadian Field-Naturalist 125:7-11. Proulx, G., R. Kariz, and J. Farkvam. 2006. Using Forest Inventory Data to Predict the Distribution of Potential Winter Habitats for American Martens. Pages 77-88 in M. Santos-Reis, J.D.S. Birks, E.C. O’Doherty, and G. Proulx, editors. Martes in carnivore communities. Alpha Wildlife Publications, Sherwood Park, AB. Raphael, M.G. 1994. Techniques for Monitoring Populations o f Fishers and American Martens. Pages 224-240 in S.W. Buskirk, A.S. Harestad, M.G. Raphael, and R.A. Powell, editors. Martens, sables, and fishers: biology and conservation. Cornell University Press, Ithaca NY. Ruette, S., P. Stahl, and M. Albaret. 2003. Factors affecting trapping success o f red fox Vulpes vulpes, stone marten Martes foina and pine marten M. martes in France. Wildlife Biology 9:11-19. Saaty, T.L. 1977. A scaling method for priorities in hierarchical structures. Journal of Mathematical Psychology 15:234-281. Schneider, R.R., J.B. Stelfox, S. Boutin, and S. Wasel. 2003. Managing cumulative impacts o f land uses in the Western Canadian Sedimentary Basin: A modelling approach. Conservation Ecology 7:8. 91 Simons-Legaard, E.M., D.J. Harrison, W.B. Krohn, and J.H. Vashon. 2013. Canada lynx occurrence and forest management in the Acadian forest. Journal of Wildlife Management 77:567-578. Smith, L.M., and I.L. Brisbin Jr., and G.C. White. 1984. An evaluation of total trapline captures as estimates of furbearer abundance. Journal o f Wildlife Management 48:1452-1455. Soutiere, E.C. 1979. Effects o f timber harvesting on Marten in Maine. Journal o f Wildlife Management 43:850-860. StataCorp. 2011. Intercooled Stata 12.1 for Windows. StataCorp LP, College Station, TX. Steventon, D.J., and D.K. Daust. 2009. Management strategies for a large-scale mountain pine beetle outbreak: modelling impacts on American martens. Forest Ecology and Management 257:1976-1985. Store, R., and J. Kangas. 2001. Integrating spatial multi-criteria evaluation and expert knowledge for GIS-based habitat suitability modelling. Landscape and Urban Planning 55:79-93. Thompson, I. D. 1994. Marten populations in uncut and logged boreal forests in Ontario. Journal o f Wildlife Management 58:272-280. Thompson, I.D., and P.W. Colgan. 1994. Marten activity in uncut and logged boreal forests in Ontario. Journal of Wildlife Management 58:280-288. Vittinghoff, E., and C.E. McCulloch. 2007. Relaxing the rule of ten events per variable in logistic and cox regression. American Journal of Epidemiology 165:710-718. Vuong, Q.H. 1989. Likelihood ratio tests for model selection and non-nested hypotheses. Econometrica 57:307-333. Webb, S.M., and M.S. Boyce. 2009. Marten fur harvests and landscape change in westcentral Alberta. Journal o f Wildlife Management 73:894-903. Webb, S.M., D.J. Davidson, and M.S. Boyce. 2008. Trapper attitudes and industrial development on registered traplines in west-central Alberta. Human Dimensions of Wildlife: An International Journal 13:115-126. Weir, R.D., and P.L. Almuedo. 2010. British Columbia’s interior fisher wildlife habitat decision aid. BC Journal o f Ecosystems and Management 10:1-8. Weir, R.D., and F.B. Corbould. 2006. Density o f fishers in the sub-boreal spruce biogeoclimatic zone of British Columbia. Northwestern Naturalist 87:118-127. 92 Weir, R.D., and F.B. Corbould. 2010. Factors affecting landscape occupancy by fishers in north-central British Columbia. Journal of Wildlife Management 74:405-410. Weir, R.D., and A.S. Harestad. 2003. Scale-dependent habitat selectivity by fishers in southcentral British Columbia. Journal of Wildlife Management 67:73-82. Weir, R.D., M. Phinney, and E.C. Lofroth. 2012. Big, sick, and rotting: why tree size, damage, and decay are important to fisher reproductive habitat. Forest Ecology and Management 265:230-240. Wiebe, P.A., J.M. Fryxell, I.D. Thompson, L. Borger, and J.A. Baker. 2013. Do trappers understand marten habitat? Journal o f Wildlife Management 77:379-391. Yamada, K., J. Elith, M. McCarthy, and A. Zerger. 2003. Eliciting and integrating expert knowledge for wildlife habitat modelling. Ecological Modelling 165:251-264. 93 Appendix A. Survey conducted by candidate trappers to assess their suitability as experts for the subsequent development o f expert-based habitat models. 94 Assessment of Trapping Activity T he following questionnaire h a s b een developed in order to gain an understanding of your trapping experience and trapline activity. P le a se an sw er the following questions by typing directly into the s p a c e s provided, or by selecting the appropriate resp o n ses. 1) How m any y ears have you been trapping furbearers in C an ad a? 2) How m any y ears have you b een trapping on your current trapline? 3) W here is your current trapline located? 4) W hat is the registration num ber for your current trapline? 5) Do you keep personal records of your trapping activity? Y esQ N oO 6) If you answ ered 'Y es' to Q uestion 3, p le ase explain th e ty p es and extent of records that you keep (i.e. effort level, num ber of captures, location of captures, etc.). 7) Which sp ec ie s do you targ et on your trapline in m ost y ears? □ B eaver □ B obcat □ C o y o te □ F is h e r □ Lynx □ Marten □ Muskrat □ Otter □ Wolf □ Wolverine □ O ther | | 95 8) W hat types of landscape ch an g e is currently occurring on your trapline, or h a s occurred in the past? □ Forestry/Timber Harvesting U Forest Fires □ Mining □ Oil and G as Q Pine Beetle Kill □ Powerlines □ R oads □ O ther 9) D escribe any levels of landscape change that are currently occurring on your trapline, or have occurred in the past____________________________ 96 Appendix B. Examples of surveys conducted by biologist and trapper experts for the purpose o f identifying the focal species and habitat variables, and evaluating habitat variables for the development of expert-based habitat models. 97 Recommendation of Three Focal Species The following survey h a s been developed to allow participating experts the opportunity to recom m end three furbearers to serve a s th e focal sp ecies for the duration of the study. Keep in mind, the goal of this study is to understand the effects of landscape ch an g e on furbearer habitat and population trends. The study a re a is central-interior British Columbia. P le a se answ er the following questions by typing directly into th e sp a c e s provided, or by selecting the appropriate resp o n ses. 1) Identify 3 of the following species that you feel would be most important to focus research towards during this study. Three species will be the focus of habitat modeling to investigate the effects of landscape change on habitat quality and availability. They will also be used to relate landscape change to population abundance. The study area is central-interior BC. Ideal focal species for this study will have ecological and economic importance, and will also be sensitive to, or affected by, various forms of landscape change: □ B eaver □ B obcat □ Coyote □ Fisher □ □ Marten □ Mink □ M uskrat □ Otter □ Wolf □ Lynx Wolverine |~ | O ther | 2) a. P le a se provide a brief rationale a s to why you selected sp ec ie s 1? b. Why did you select sp ecies 2? 98 c. Why did you select sp ecies 3? Do you have any other sug g estio n s or concerns regarding th e selection of ' three sp ec ie s to serve a s the focal sp ecies during this study? 99 C a n d i d a t e H a b ita t a n d D is tu r b a n c e V a r ia b le s fo r Lynx (L y n x c a n a d e n s i s ) W in te r H a b ita t M o d e l in C e n tra l-In te rio r B ritish C o lu m b ia The following survey h as been developed to aid in the identification of general habitat variables thought to influence distribution of lynx across winter range. P lease score the following variables in term s of their importance for identifying lynx habitat. In other words, which of the following variables should be included in a lynx winter habitat model in central-interior BC? Keep in mind we are simply identifying the general categories of variables that will be included in the models. During a subsequent survey, I will then ask you to rank the specific habitat variables in term s of importance. P lease answ er the following questions by typing directly into the sp a ce s provided, or by selecting the appropriate responses. W hen completed, save the file and submit via e-mail to mbridger4@gmail.com If you have any questions, please contact me at 250-961-5869, or by e-mail at mbridger4@gmail.com Please rank the following habitat variables in terms of their importance for identifying lynx habitat, where 4 is very important; 3 is important; 2 is moderate; 1 is low importance; and 0 is unimportant. Ahabitat variable could be important whether it has a positive or negative influence on lynx distribution. Ahabitat variable would be unimportant if it had no influence on lynx distribution. C o m V ita o ^ IM rfi G r o u n d S fr u to C o v e r S t a i d u n * C o w y t a t t y ( O w e r ta o r /) C a n o p y C o v e r ( C ro w n C lo s u r e ) O w % M t a a e T t a a t ( H — ta i f t r a i ) D o m in a n t ( L o w in g ) T r e e S p e c i e * F tim a tS ta n d A g e D is ta n c e to V fe te r B o d ie s S to p K S ta q p n fltt) A s p e c t ( N ,S ,E , o r W ) C mPHHKS*S e r a l S t a g e (A g e ) o f O e a r - C u t a P x * o rto n a fa a * ^ C u te « m e « fc * r fta n g s D is ta n c e t o R o o d s P ita y B tta ta m iW ifta iB B P re s e n c e o f S a a rrs c /G a s U n e s P w s e n c e rfa M ta F o fl H a b ita t F r a g m e n ta tio n 0 1 2 3 4 O o o o o o o o o o o o o o o o o o o o 0 o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o o Q o o o o o o o o o o o o o o o o o o o o o o o o o 0 o o o 100 Confidence Scores for Habitat Variables: Please score your confidence in the answers you provided above for all habitat variables, where 10 is extremely confident. In other words, how certain are you in the answers you provided for each habitat and disturbance variable when considering their importance for influencing the distribution of lynx across their winter range. 1 cmnmmUfQ O G ro u n d S h ru b C o v er Q tw E a - f l f l m y h n t g y C a n o p y C o y e r (C r o w n C lo tu r e ) P e n t f r o f m m im t f l i t * * r e t ) D o m in a n t ( L e t t i n g ) T r e e S p e t i t a P itta n c e toW a te r B o d ie s S k 3 p t(S ta a p n e M ) A s p e c t (N ,S ,E , o r W ) C B M B ti:' S e r a i S t a g e (A g e ) o f C le a r - C tM n w p o f t m o f C le a r C g W e o o e e Wrter Raige D is ta n c e t o R o a d s D a n t a 1e l f t a ^ i a n t l t e P r e s e n c e o T S e ta n v c /G ta U n e a P r e a a n c to f B u rn t F b re tf H a b i a t F r a g m e n ta tio n ' o o o o o o o o o o o o o o o o o 2 3 4 5 6 7 o o o o o o 0 o 0 o o o o o o o o o o 0 o o o o 0 o 0 o o 0 o o o o o 0 o o o o o o o o o o o o o o o o o o o o o o o o o o 0 o o o o o o o o o o o o o o o o o o o o o o o o o o o o o 0 o o o o o o o 101 6 o o o o o o 0 o o o o o o o Q O o o o o o o o o o o 9 10 o o o 0 o o 0 o o o o o o o o o o o o o o o 0 o 0 o 0 o o o o o Q 0 o o P le a se su g g est any additional information or additional habitat variables that you feel are important to consider w hen determining the distribution of lynx acro ss their winter range in central-interior BC.___________________________ 102 M a r t e n H a b i t a t A s s e s s m e n t The following is an assessment of the importance of specific habitat variables for the distribution of marten (Martes americana) across winter landscapes in the central-interior of British Columbia. The Analytical Hierarchy Process (AHP) is being used to obtain scores in the form of pairwise comparisons for every combination o f habitat variables. The scoring scheme for the AHP is provided below. *Important* You will be providing scores for the variables in the column on the left side o f the AHP matrices compared to the variables in the row at the top of the matrices. Please keep in mind that all scores are comparative (e.g., A very important variable may be given a high score of ‘7’ or ‘9’ when compared to a low-importance variable, but may receive a lower score o f ‘ 1’ compared to a variable that is equally as important). An example of an AHP matrix is provided below. You are also asked to provide confidence scores for each of the matrices that you completed. In other words, on a scale from 1-10, how confident are you in the comparative scores you provided? (‘ 1’ being very low confidence and ‘10’ being very high) Participants are asked to type their scores directly into the AHP comparison matrices provided. 103 The scoring scheme for the AHP pairwise comparisons of habitat variables 1 = Equal importance 3 = Moderately more important 5 = Strongly more important 7 = Very strongly more important 9 = Extremely strongly more important 2,4,6,8, = Intermediate values 1 = Equal importance -3 = Moderately less important -5 = Strongly less important -7 = Very strongly less important -9 = Extremely strongly less important -2, -4, -6, -8 = Intermediate values E xam ple: This matrix is com paring the im portance o f tree species for a given w ild life species. For this exam ple, w e start by com paring the variables in the left colum n to the first variable in the top row (Pine). B y default, Pine com pared to itself is o f equal im portance, represented by a score o f ‘ 1 ’. N ext. Spruce is m oderately m ore important than Pine, represented b y a score o f ‘3 '. Fir is very strongly m ore im portant than Pine, represented by a score o f ‘7 ’. H em lock is sn on gly m ore im portant than Pine, represented by a score o f ‘ 5 ’. Finally, D eciduous is m oderately less im portant than Pine, represented by a score o f ‘- 3 ’. A ll shaded cells are the opposite o f their respective scores, and do not need to be filled out. Pine Pine Spruce Fir H em lock D eciduous Spruce Fir H em lock D eciduous 1 3 7 5 -3 VV'w t*." *^ i 5 1 3 -3 -7 -5 104 ‘ *>f 1 -5 1 A H P C o m p a r is o n M a t r ic e s f o r M a r t e n H a b it a t V a r ia b le s Please fill in the blank cells o f the following AHP matrices. Give comparative scores (using the AHP scoring scheme) for the following habitat variables in terms of importance for marten winter habitat. Please refer to Habitat Variable Handout PDF for examples. ** Score the variables in the column on the left relative to the variables on die top row- * * Coarse Woody Debris Load: Patch/Stand Spatial Scale Medium Moderate Confidence Score: ■ jq Ground Shrub Cover (Understory Density): Patch/Stand Spatial Scale Moderate (30-60%caver) LOW (<30% cover) H i fih (>60% cover) LOW’ (<30% cover) 1 Moderate (30-60% cover) High (>60% cover) Confidence Score Structural Complexity (Overstory): Patch/Stand Spatial Scale Moderate Moderate Confidence Score: Canopy Cover (Crown Closure): Patch/Stand Spatial Scale Open (<10%) Minimal (10-40%) Qpen(<10%) Minimal (10-40%) Moderate (40-70%) High (>70%) Confidence Score 105 Moderate (40 - 70%) High (>70%) Forest Stand Density of Mature Trees (Basal Area): Patch/Stand Spatial Seale Moderate Low (<2Qm‘/ha) Moderate (20 - 40m"/ha) High (>40m /ha) Confidence Score: Leading Tree Species (Forest Stand-Type): Patch/Stand Spatial Scale Spruce (W h.Ea Hy) Lodgepole Pine Black Spruce Other Conifer (SubAlpine. Balsam Douglas) Deciduous Mixed (Conifer Deciduous) Spruce (WluteEngl/Hybnd) Lodgepole Ptne Black Spruce Other Conifer (Sub-Alpine/ Balsam.' Douglas) Deciduous Mixed (Coniferous. Deciduous) Confidence Score: -jq Coniferous Stand Age: Patch/Stand Spatial Scale Young (<20 years) Mid-Age (2050 years) Mature (50-80 years) Old (>80 years) Young (<20 years) Mid-Age (20-50 years) Mature (50-80 years) Old (>80 years) Confidence Score: iq Deciduous Stand Age: Patcli/Stand Spatial Scale Young (<10 years) | Mid-Age (10-30 Young (<10 years) M id -A se (10-30 years) Mature-Old (>30 years) Confidence Score: 106 Mature-Old (>30 Clear-Cut Serai Stage (Age) Recently cut (<5 years) Recently cut (<5 years) 5 - 1 0 years 10- 20 years > 20 years 1 5-10 years 10 - 20 years >20 years Confidence Score: -jq Proportion of Clear-Cuts on Landscape: Landscape Spatial Scale High level (>40%) | Medium Level (10High Level Cut C>40%) Moderate Level Cut (10-40%) Low Level Cut (<10%) Confidence Score: 1Q 1 Burned Forest Stand Age 5-20 vears 20 vears <5 years 5-20 years >20 years Confidence Score: -jq Habitat/Patch Connectivity: Landscape Spatial Scale Moderate Low Connectivity Moderate Connectivity High Connectivity Confidence Score 107 Habitat Variable Classification Handout Coarse Wood? Debris Low CWD Lead Moderate CWD Load H ■ H 1 1 High CWD Load High CWD Load Gronnd Sbrub Cover ' * Moderate G rand Sbnb C ew r f o - High G rand Shrub C om High Crow d Shrab Cover 108 Structural Compleiity (Overstory) M bdeiaieSttnctacalCoaipkxtty High Sttactanl Canptesitjr Canopy Cover (Crown Closure: Mnamai Canopy C o m ( 10-40%) HighCaaopy C a m (>70%) M odenle Canopy C ov h (40-70%) 109 5-10 years >20)«an Hi^propoiti(»ofckaitntf(>40%) Hi^ proportionofckarcrt*(>40%) Habitat/Patch Connectivity Across Landscape L ow C taneetm tf M odcntt Conoectraty Hi^h Connectivity 111 Appendix C. Description o f the classification o f habitat variables used to construct expertbased habitat models for each focal species. 112 TABLE C l. Descriptions o f the subclasses, levels, or categories o f habitat variables included in the habitat models for fisher, lynx, and marten. Canopy Cover Low = - Crown closure <10% Classifications adapted from Fuller and Harrison 2005, Proulx et al. 2006, Proulx 2009 Minimal = - Crown closure >10% and <40% Moderate = - Crown closure >40% and <70% High = - Crown closure >70% Coarse Woody Debris Classifications adapted from Clark et al. 1998 Low = - Forest age <50 years, or forest age >50 and <150 years if leading tree species is lodgepole pine - If fire present, forest age <50 years if outside fire polygon, or forest age >50 and <150 years if leading tree species is lodgepole pine - All cutblocks Moderate = - Forest age >50 and < 200, or forest age > 150 if leading tree species is lodgepole pine - If fire present, forest age <50 if within fire polygon, or forest age >50 and £200 years, or forest age >150 years if leading tree species is lodgepole pine High = - Forest age >200 years and leading tree species is not lodgepole pine Coniferous Stand Age Classifications adapted from Proulx et al. 2006, Proulx 2009 Young = - Coniferous leading tree species and forest age <20 years Mid-Age = - Leading tree species and forest age >20 and <50 years Mature = - Coniferous leading tree species and forest age >50 and <80 years Old = - Coniferous leading tree species and forest age >80 years Cutblock Age Recent = <5 years since logging Classifications based on field observations by M.Bridger 2013 Young = 5 —10 years since logging Moderate = 10 —20 years since logging Old = >20 years since logging Deciduous Stand Age Classifications based on field observations by M.Bridger 2013 Young = - Deciduous leading tree species and forest age <10 years Mid-Age = - Deciduous leading tree species and forest age >10 and <30 years Mature - Old = - Deciduous leading tree species and forest age >30 years Forest Fire Age Recent = - Fire age < 5 years Classifications adapted from Paragi et al. 1997 Moderate = - Fire age >5 and <20 years Old = - Fire age >20 years 113 Forest Stand Density Low = - Basal area <20 Classifications adapted from Fuller and Harrison 2005, Proulx et al. 2006, Proulx 2009 Moderate = Ground Shrub Cover Classifications adapted from Proulx 2009 - Basal area >20 and <40 High = - Basal area >40 Low = - Shrub crown closure <30% and soil type not mesic or hygric - All cutblocks Moderate = - Shrub crown closure <30% and soil type mesic or hygric, or shrub crown closure >30% and <60% High = - Shrub crown closure >60% Habitat Connectivity Classifications based on GIS landscape metrics, M.Bridger 2013 Low = - All forest polygons that intersect cutblocks - High elevation, alpine regions - All cutblocks Moderate = - All forest polygons that do not intersect cutblocks and ‘proportion of landscape logged’ variable is equal to ‘high’ High = - All forest polygons that do not intersect cutblocks and ‘proportion of landscape logged’ variable is equal to ‘low’ or ‘moderate’ Leading Tree Species for Marten and Lynx Classifications based on dominant leading species found in the SBS and ESSF BEC Zones, Meidinger and Pojar 1991 Spruce = - Forest polygons >60% Engelmann spruce, white spruce, or hybrid spruce, and secondary species is coniferous Lodgepole Pine = - Forest polygons >60% lodgepole pine, and secondary species is coniferous Black Spruce = - Forest polygons >60% black spruce, and secondary species is coniferous Other Conifers = - Forest polygons >60% Douglas fir, subalpine fir, balsam (true) fir, hemlock, or western cedar, and secondary species is coniferous Deciduous = - Forest polygons >60% birch, aspen, cottonwood, or poplar Mixed = - Forest polygons <60% deciduous - Or, forest polygons <60% coniferous and secondary species is deciduous 114 Leading Tree Species for Fisher Spruce = - Forest polygons >60% Engelmann spruce, white spruce, or hybrid spruce, and secondary species is coniferous Classifications based on dominant leading species found in the SBS and ESSF BEC Zones, Meidinger and Pojar 1991 Douglas Fir = - Forest polygons >60% Douglas fir, and secondary species is coniferous Lodgepole Pine = - Forest polygons >60% lodgepole pine, and secondary species is coniferous Black Spruce = - Forest polygons >60% black spruce, and secondary species is coniferous Other Conifers = - Forest polygons >60% subalpine fir, balsam (true) fir, hemlock, or western cedar, and secondary species is coniferous Cottonwood = - Forest polygons >60% cottonwood Deciduous - - Forest polygons >60% birch, aspen, or poplar Mixed = - Forest polygons <60% deciduous - Or, forest polygons <60% coniferous and secondary species is deciduous Proportion of Landscape Logged/Harvested Low Level = - Trapline area divided by total cutblock area equals <10% Moderate Level = Classifications adapted from Hargis et al. 1999, Cheveau et al. 2013 - Trapline area divided by total cutblock area equals >10% and <40% High Level = - Trapline area divided by total cutblock area equals >40% Structural Complexity Low = Classifications adapted from Proulx et al. 2006 - Forest age <80 - All cutblocks Moderate = - Forest Age >80 and < 150, or forest age >150 if leading tree species is lodgepole pine High = - Forest age >150 if leading tree species is not lodgepole pine 115 Appendix D. Mean eigenvector scores (representing relative importance) resulting from the expert evaluation o f fisher, lynx, and marten habitat variables. 116 Trappers■ B i o lo g is ts I 0.800 P 0.800 O £ 0.600 O | 0.600 0.400 0.400 l l .1 0.200 ui 0.000 Low Moderate CWD Level 0200 0.000 0.800 1 High 0.800 .I L i S 0600 | Ji ■ Low Moderate Ground Shrub Cover High 0.400 I 0.200 0.000 Low Moderate Structural Complexity 0.600 0.400 0.000 Open High 0.400 0.300 0.800 0.200 0.100 | °-400 i i I 0200 0.000 Low Moderate 0.000 b iii.il ill i / / / / / S °° V Forest Stand Density High Minimal Moderate Canopy Cover S High i i 0200 Leading Tree Species 0.800 . 0.800 1 0.400 10.200 0.000 ■ . m ... i Young Mid-age Mature Coniferous Stand Age i 0.600 0.400 0200 0.000 : f t - Old g 0.800 0.800 <8 0.600 0.600 5 I 0.400 6 0.200 IB 0.000 Recent Young Mid-age Cutblock Age I Mature-Old 0.400 i l 0.200 0.000 High levtl cut Moderate iavil cut Old Low t o i l cut Proportion of Landscape Cut 0.800 0.800 §8 0.600 (O 0.600 I 0.400 0.400 i 0200 0.200 0.000 0.000 Young It Young Mid-age Deciduous Stand Age Low Mid-age Bum Age H Moderate i High Connectivity FIG. D l. Mean eigenvector scores (representing relative importance) resulting from the expert evaluation of fisher habitat variables. The data represents the mean scores and 95% confidence intervals. 117 Trappers■ B i o lo g is ts ! 0> 0.800 0.800 | 0.600 0.600 1 0.400 0.400 0.200 0.200 UJ 0.000 0.000 Low Moderate Ground Shrub Cover High High I t l l 0.600 : 2 0.600 o £ ts0 0.400 o.4oo | y 1 0200 Open Minimal Moderate Canopy Cover Low High ® 0.600 0.500 0.400 I 0.000 i i .i i ,m ;ife,k ,ll 1^ ijjr®6 0.300 | 0200 / / / 0.100 / 0.000 Young Mid-age Mature Coniferous Stand Age 0.800 0.600 0.600 $ 0 ,4 0 0 w 0.000 0.400 4 1 ■ 0.200 0.000 Young Mid-age Mature-Old Deciduous Stand Age Recent 0.800 0.800 0.600 0.600 0.400 0.400 i 0.200 0.000 High 0200 Leading Tree Species S 0200 Moderate Forest Stand Density | 0.400 ' | l:i I I 0.000 i w i l Low Moderate Structural Complexity < lii 0200 0.000 High level cut Moderate Low level cut level cut Proportion of Landscape Cut Young i i Young Mid-age Cutblock Age Old ii Mid-age Bum Age Old 0.800 S 0.600 0.400 8#0200 O Q 0.000 Low Ii I Moderate Connectivity High FIG. D2. Mean eigenvector scores (representing relative importance) resulting from the expert evaluation of lynx habitat variables. The data represents the mean scores and 95% confidence intervals. 118 T rappers■ B iolo gists I , 0.800 0.800 5 # 0.600 6 0.600 0.400 0.400 8>0.200 0.200 | iij Low 0.800 o i f t High | 0.600 | # 0 .6 0 0 6 0.400 | | 0.400 1 0.200 iO 0.000 Low Moderate Structural Complexity 1 0.000 I Open High # 0.600 0200 1 0.400 1,0200 0.000 ... m m Lcm Moderate Forest Stand Density i 0.000 /i f / o f / / / / High ® 0.800 0.600 fe < / c / Leading Tree Species «w 0.600 0.400 i 0.000 Young Mid-age Mature Coniferous Stand Age lari 0.200 0.000 Young Old 0.800 0.800 0.600 0.600 0.400 0.400 Mid-age Mature-Old Deciduous Stand Age I 0200 0200 0.000 0.000 Recent High l i ii.!.i* I 0.100 0.800 0 1 0.400 I $>0,200 E Minimal Moderate C an o p y C o v er 0.400 0.300 6 NN 0200 ] e 0.800 High level cut Moderate Low level cut level cut Proportion of Landscape Cut Young Mid-age Cutblock Age o> 0.800 0.800 0 o w 0.600 | r Lew Moderate Ground Shrub Cover Moderate CWO Level „ 0.800 tZi mm 0.000 0.000 M Jft 0.600 0.400 0.400 8.0200 uj 0.200 0.000 Young Mid-age 1 0.000 O ld Low Bum Age Moderate Connectivity High FIG. D3. Mean eigenvector scores (representing relative importance) resulting from the expert evaluation of marten habitat variables. The data represents the mean scores and 95% confidence intervals. 119 Appendix E. Examples o f expert-based habitat maps, providing a spatial representation of change in the availability and quality o f habitat from 1990 to 2013. 120 M arten H ab itat 1990 M m H Waterbodies M arten H ab itat 2 0 1 3 Poor - 7642 ha H Poor - 24452 ha Moderate - 8386 ha I H Moderate - 1 I172 ha Waterbodies " " I G ood- 12621 ha Very Good -12696 ha Example o f the expert-based habitat model for marten applied to a trapline in the West Study Area. The maps show the availability and quality o f habitat in 1990 and 2013. Fisher Habitat 1990 m i Poor -11156 ha H I Moderate - 25475 ha L Watertoodies Fisher Habitat 2013 ■ H I Poor-38615 ha Moderate - 21210 ha Good - 24469 ha j Good - 10900 ha j Very Good -10744 ha [___ ] Very Good - 918 ha 122 Example o f the expert-based habitat model for fisher applied to a trapline in the West Study Area. The maps show the availability and quality o f habitat in 1990 and 2013. Marten Habitat 1990 | Waterbodies Marten Habitat 2013 H I Poor - 24234 ha K Poor - 17109 ha H I Moderate - 40783 ha H | Moderate - 52363 ha [ ~ ~ n Good - 28966 ha Good - 23798 ha f Very G o o d -21319 ha j Very Good - 20606 ha Waterbodies Example o f the expert-based habitat model for marten applied to a trapline in the East Study Area. The maps show the availability and quality o f habitat in 1990 and 2013. Appendix F. Candidate a priori model selection for predicting lynx and marten captures in the West and East Study Areas across central-interior BC, Canada. 124 TABLE F I. Candidate a priori models used to select the most parsimonious count model for understanding captures of lynx and marten by trappers in the West and East Study Areas across central-interior BC, Canada Parameter Definitions P aram eter Abbreviation Description Effort E Measure o f trapping effort on a scale from 0 - 1 0 Standardised Effort SE Measure o f trapping effort relative to trapline area, where ‘Effort’ was multiplied by ‘Trapline Area’ Trapline Area TA Trapline area in hectares Habitat Value HV Sum total of the raster habitat values on each respective trapline, according to expert-based habitat maps Standardised Habitat Value SH Habitat relative to trapline area, where ‘Habitat Value’ was divided by ‘Trapline Area’ GVGH Percent o f the respective trapline area composed of ‘Good’ or ‘Very Good’ habitat, according to expertbased habitat maps % Good and Very Good Habitat Fur Price FP The average fur price from the previous year’s trapping season Mean Minimum Temperature MMT Average daily minimum air temperature recorded at Prince George Airport weather station (Nov - Jan for marten; Dec - Feb for lynx) Extreme Minimum Temperature EMT Average monthly extreme minimum air temperature recorded at Prince George Airport weather station (Nov - Jan for marten; Dec - Feb for lynx) Snow Depth Sum SMS Cumulative snow depth recorded at Prince George Airport weather station (Nov - Jan for marten; Dec Feb for lynx) 125 Candidate a priori count models for predicting lynx captures in the West Study Area. Effort Model E + TA E E + FP SE + TA SE + FP SE R ank 1 2 3 4 5 6 AICn 412.83 430.79 431.47 434.63 451.77 455.01 AIC cm>, 1.000 <0.001 <0.001 <0.001 <0.001 <0.001 Aj AICc 0.000 17.968 18.640 21.800 38.940 42.188 Model R ank 1 2 3 4 5 6 AIC„ 468.41 470.83 472.45 474.77 476.87 489.99 AIC chv 0.673 0.200 0.089 0.028 0.010 <0.001 A,AICf 0.000 2.428 4.040 6.368 8.460 21.588 R ank 1 2 3 4 5 6 AICc<398.14 414.50 416.71 432.91 435.50 444.43 AICc»v, 1.000 <0.001 <0.001 <0.001 <0.001 <0.001 A, AICc 0.000 16.360 18.569 34.769 37.360 46.289 R ank 1 2 3 4 5 6 AICc, 428.93 431.02 432.89 456.55 456.61 457.34 AIC cm'/ 0.671 0.236 0.093 <0.001 <0.001 <0.001 A, AICc 0.000 2.091 3.960 27.620 27.680 28.411 Habitat GVGH + TA SH SH + TA HV HV + TA GVGH Effort and Habitat Model E + TA + GVGH E + TA + SH E + SH E + GVGH SE + TA + HV SE + HV Effort and Weather Model E + EMT E + SDS + EMT E + SDS SE + SDS SE + MMT SE + SDS + EMT 126 Habitat and Weather Model SH + EMT SH + SDS SH + SDS + EMT SH + SDS + MMT HV + EMT HV + SDS HV + SDS + EMT GVGH + MMT GVGH + SDS GVGH + SDS + MMT R ank 1 2 3 4 5 6 7 8 9 10 AICc/ 470.39 471.49 471.78 472.34 476.01 476.75 478.16 491.79 492.07 493.96 AICc)v, 0.388 0.224 0.194 0.146 0.023 0.016 0.008 <0.001 <0.001 <0.001 A, AICc 0.000 1.100 1.391 1.951 5.620 6.360 7.771 21.400 21.680 23.571 Effort, Habitat, and Weather Model E + TA + GVGH + EMT E + TA + GVGH + SDS E + TA + GVGH + MMT E + TA + SH + EMT E + TA + SH + MMT E + TA + SH + SDS SE + TA + HV + EMT SE + TA + HV + MMT SE + TA + HV + SDS R ank 1 2 3 4 5 6 7 8 9 AICC( 397.97 400.29 400.32 412.44 414.61 416.33 435.77 437.14 437.15 127 AIC cHV 0.615 0.194 0.191 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 A, AICc 0.000 2.312 2.344 14.462 16.636 18.353 37.792 39.168 39.178 Candidate a priori count models for predicting lynx captures in the East Study Area. Effort Model E + TA E SE + TA SE E + FP SE + FP R ank 1 2 3 4 5 6 AIC„ 127.26 134.96 135.82 136.84 137.00 137.90 A1Cchv 0.947 0.020 0.013 0.008 0.007 0.005 A, AICc 0.000 7.700 8.560 9.580 9.740 10.640 Model R ank 1 2 3 4 AICc, 168.52 170.72 173.50 174.30 Al CcH’, 0.662 0.239 0.059 0.040 A, A ICc 0.000 0.204 4.820 5.620 R ank 1 2 3 4 5 6 AICc, 127.48 127.90 129.38 129.42 136.26 138.20 AICfH’, 0.386 0.312 0.149 0.147 0.005 0.002 A, AICe 0.000 0.423 1.903 1.940 8.780 10.723 R ank 1 2 3 4 5 6 AICc, 124.90 125.10 135.18 137.26 137.68 138.90 AICcW»; 0.522 0.472 0.003 0.001 0.001 0.001 A, AICe 0.000 0.200 10.277 12.357 12.780 13.997 Habitat HV HV + TA SH GVGH Effort and Habitat Model E + SH E + TA + SH E + TA + GVGH E + GVGH SE + HV SE + TA + HV Effort and Weather Model E + TA + MMT E + TA + EMT SE + MMT E + SDS SE + SDS + MMT SE + SDS 128 Habitat and Weather Model HV + SDS HV + EMT SH + SDS HV + SDS + EMT GVGH + SDS SH + SDS + MMT GVGH + EMT SH + MMT GVGH + SDS + MMT GVGH + SDS + EMT R ank 1 2 3 4 5 6 7 8 9 10 AIC„ 169.70 170.64 172.14 172.22 173.80 174.60 175.54 175.66 175.92 176.12 AICcw, 0.383 0.240 0.113 0.109 0.049 0.033 0.021 0.019 0.017 0.015 A/ AICc 0.000 0.940 2.440 2.523 4.100 4.903 5.840 5.960 6.223 6.423 Effort, Habitat, and Weather Model E + SH + SDS E + SH + EMT E + SH + MMT E + GVGH + EMT E + GVGH + MMT E + GVGH + SDS SE + HV + MMT SE + HV + EMT SE + HV + SDS R ank 1 2 3 4 5 6 7 8 9 AICCI 129.32 129.56 129.82 130.58 130.98 131.86 133.50 135.20 138.44 129 AICcw, 0.244 0.216 0.190 0.130 0.106 0.068 0.030 0.013 0.003 A, AICc 0.000 0.240 0.500 1.260 1.660 2.540 4.180 5.880 9.120 Candidate a priori count models for predicting marten captures in the West Study Area. Effort Model E + TA E + FP E SE + TA SE SE + FP R ank 1 2 3 4 5 6 AICCI 676.34 678.78 680.43 718.66 738.79 740.88 AICcW,0.702 0.207 0.091 <0.001 <0.001 <0.001 A,- AICc 0.000 2.440 4.083 42.320 62.443 64.540 Model R ank 1 2 3 4 5 6 AIC„ 745.75 746.56 746.76 746.96 747.01 749.79 AICcH’f 0.288 0.191 0.173 0.157 0.153 0.038 A/ AICc 0.000 0.817 1.017 1.217 1.260 4.040 R ank 1 2 3 4 5 6 AICc/662.79 673.89 680.68 682.38 694.95 736.78 AICcw/ 0.996 0.004 <0.001 <0.001 <0.001 <0.001 A/ AICc 0.000 11.100 17.892 19.592 32.160 73.992 R ank 1 2 3 4 5 6 AICc, 680.78 681.83 682.52 740.20 740.82 742.35 AIC cH', 0.497 0.294 0.208 <0.001 <0.001 <0.001 A, AICc 0 1.048 1.740 59.420 60.040 61.568 Habitat HV SH + TA HV + TA GVGH + TA GVGH SH Effort and Habitat Model E + TA + SH E + TA + GVGH E + GVGH E + SH SE + TA + HV SE + HV Effort and Weather Model E + EMT E + SDS + EMT E + SDS SE + MMT SE + SDS SE + SDS + MMT 130 Habitat and Weather Model H V +EM T HV + MMT HV + SDS GVGH + MMT GVGH + SDS HV + SDS + MMT SH + EMT GVGH + SDS + MMT SH + SDS SH + SDS + EMT R ank 1 2 3 4 5 6 7 8 9 10 AIC„ 746.38 746.98 747.44 748.38 749.00 749.07 750.26 750.53 751.64 752.31 A IO , 0.276 0.205 0.163 0.102 0.075 0.072 0.040 0.035 0.020 0.014 A,- AICc 0.000 0.600 1.060 2.000 2.620 2.688 3.880 4.148 5.260 5.928 Effort, Habitat, and Weather Model E + TA + SH + EMT E + TA + SH + MMT E + TA + SH + SDS E + TA + GVGH + EMT E + TA + GVGH + MMT E + TA + GVGH + SDS SE + HV + MMT SE + HV + EMT SE + HV + SDS R ank 1 2 3 4 5 6 7 8 9 A IC c, 660.83 661.41 663.20 674.37 674.69 676.10 692.43 692.79 693.50 131 A IO , 0.486 0.364 0.149 0.001 <0.001 <0.001 <0.001 <0.001 <0.001 A,- A IC c 0.000 0.582 2.367 13.537 13.861 15.272 31.599 31.956 32.673 Candidate a priori count models for predicting marten captures in the East Study Area. Effort SE SE + FP SE + TA E + TA E E + FP R ank 1 2 3 4 5 6 AICc, 468.23 470.17 470.19 470.91 471.21 473.21 AIC ch>,0.431 0.163 0.161 0.113 0.097 0.036 A,- AICc 0.000 1.944 1.964 2.684 2.980 4.984 Model GVGH + TA GVGH HV SH + TA HV + TA SH R ank 1 2 3 4 5 6 AIC„ 439.31 464.37 469.11 470.47 470.57 474.85 AICcw,1.000 <0.001 <0.001 <0.001 <0.001 <0.001 A,- AICc 0.000 25.056 29.796 31.160 31.260 35.536 R ank 1 2 3 4 5 6 AICc, 429.84 451.81 465.90 467.72 468.43 469.67 AIC ch’, 1.000 <0.001 <0.001 <0.001 <0.001 <0.001 A, AICc 0.000 21.975 36.060 37.880 38.595 39.835 R ank 1 2 3 4 5 6 AICc, 470.15 470.31 472.38 473.13 473.21 475.38 AICcW, 0.361 0.334 0.119 0.081 0.078 0.027 A, AICc 0.000 0.160 2.225 2.980 3.060 5.225 Model Habitat Effort and Habitat Model E + TA + GVGH E + GVGH SE + TA + HV E + TA + SH SE + SH SE + HV Effort and Weather Model SE + SDS SE + EMT SE + SDS + EMT E + SDS E + MMT SE + SDS + MMT 132 Habitat and Weather Model GVGH + EMT GVGH + SDS GVGH + SDS + EMT HV + MMT HV + SDS HV + SDS + MMT SH + EMT SH + SDS SH + SDS + EMT R ank 1 2 3 4 5 6 7 9 10 AICc/ 466.25 466.41 468.54 471.23 471.27 473.36 476.91 476.99 479.18 AICcw,0.409 0.377 0.130 0.034 0.033 0.012 0.002 0.002 0.001 A, AICc 0.000 0.160 2.285 4.980 5.020 7.105 10.660 10.740 12.925 Effort, Habitat, and Weather Model E + TA + GVGH + SDS E + TA + GVGH + EMT E + TA + GVGH + MMT SE + TA + HV + SDS SE + TA + HV + EMT SE + TA + HV + MMT E + TA + SH + SDS E + TA + SH + EMT E + TA + SH + MMT R ank 1 2 3 4 5 6 7 8 9 AIC„ 431.96 432.22 432.23 467.80 468.09 468.22 469.63 469.84 470.01 133 AIC ch>, 0.364 0.319 0.317 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 A; AICc 0.000 0.266 0.276 35.849 36.132 36.261 37.680 37.886 38.059 Appendix G. Coefficients and statistical parameters generated from the top ranked negative binomial regression models for the prediction o f lynx and marten captures in the West and East Study Areas across central-interior BC, Canada. 134 TABLE G l. Coefficients and statistical parameters generated from the top ranked negative binomial regression models for the prediction o f lynx captures in the West Study Area across central-interior BC, Canada. Param eter P Standard E rro r Z 0.391 0.003 0.017 0.029 -2.488 0.037 <0.001 0.004 0.019 0.633 10.60 6.93 4.22 1.54 -2.71 0.391 0.003 0.018 -2.488 0.037 <0.001 0.004 0.399 10.50 7.10 4.38 -6.24 P 95% C l Lower Upper <0.001 <0.001 <0.001 0.123 0.007 0.319 0.002 0.009 -0.008 -2.952 0.464 0.004 0.025 0.066 -0.472 <0.001 <0.001 <0.001 <0.001 0.318 0.002 0.010 -3.270 0.464 0.004 0.025 -1.706 AICc R ank #1 Effort Trapline Area GVG Habitat Extreme Min. Temp. Constant A IC c R ank #2 Effort Trapline Area GVG Habitat Constant 135 TABLE G2. Coefficients and statistical parameters generated from the top ranked negative binomial regression models for the prediction o f lynx captures in the East Study Area across central-interior BC, Canada. Param eter P Standard E rro r Z 0.422 0.002 -0.099 -3.396 0.045 <0.001 0.044 0.726 9.37 3.73 -2.25 -4.67 0.443 0.002 -0.050 -3.720 0.048 <0.001 0.023 0.870 9.30 3.63 -2.19 -4.27 P 95% C l Lower Upper <0.001 <0.001 0.024 <0.001 0.334 0.001 -0.186 -4.821 0.510 0.002 -0.013 -1.971 <0.001 <0.001 0.029 <0.001 0.349 0.001 -0.094 -5.426 0.537 0.002 -0.005 -2.014 AIC c R ank #2 Effort Trapline Area Mean Min. Temp. Constant A IC c R ank #2 Effort Trapline Area Extreme Min. Temp. Constant 136 TABLE G3. Coefficients and statistical parameters generated from the top ranked negative binomial regression models for the prediction o f marten captures in the West Study Area across central-interior BC, Canada. Parameter P Standard Error Z 0.278 0.002 0.068 -0.025 -1.412 0.023 <0.001 0.015 0.012 0.654 12.05 5.12 4.47 -2.04 -2.92 0.280 0.002 0.067 -0.048 -2.071 0.023 <0.001 0.015 0.025 0.738 0.279 0.002 0.062 -1.412 0.023 <0.001 0.015 0.654 P 95% C l Lower Upper <0.001 <0.001 <0.001 0.041 0.003 0.233 0.001 0.038 -0.048 -3.760 0.323 0.002 0.098 -0.001 -0.740 12.08 5.09 4.40 -1.90 -2.81 <0.001 <0.001 <0.001 0.058 0.005 0.235 0.001 0.037 -0.098 -3.517 0.326 0.002 0.097 0.002 -0.626 11.94 4.87 4.08 -2.16 <0.001 <0.001 <0.001 0.031 0.233 0.001 0.032 -2.694 0.325 0.002 0.091 -0.130 A IC c Rank #1 Effort Trapline Area Standardized Habitat Extreme Min. Temp. Constant AIC c Rank #2 Effort Trapline Area Standardized Habitat Mean Min. Temp. Constant AIC c Rank #3 Effort Trapline Area Standardized Habitat Constant TABLE G4. Coefficients and statistical parameters generated from the top ranked negative binomial regression models for the prediction o f marten captures in the East Study Area across central-interior BC, Canada. Parameter P Standard Error Z 0.032 <0.001 0.011 0.649 3.63 5.51 8.71 -3.48 P 95% C l Lower Upper 0.054 0.001 0.073 -0.990 0.179 0.002 0.116 -0.990 AICc Rank #1 Effort Trapline Area GVG Habitat Constant 0.116 0.002 0.095 -2.262 137 <0.001 <0.001 <0.001 <0.001 Appendix H. Difference in observed from predicted fur harvest records generated using negative binomial count models for lynx and marten from the West and East Study Areas across central-interior BC, Canada. A value o f zero suggests perfect prediction while negative values mean over-prediction and positive values mean under prediction. 138 (a) (b) 40 30 - -10 - §5 - 2 0 - 0 -3 0 -40 J 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 -40 -50 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 139 FIGURE H I. Difference in observed from predicted fur harvest records generated using the top ranked negative binomial count models E+TA+GVGH+EMT (a) and E+TA+GVGH (b) for lynx from the West Study Area across central-interior BC, Canada. A value of zero suggests perfect prediction while negative values suggest over-prediction and positive values suggest under-prediction (a) (b) 20 i -10 - -30 -40 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 g -10 -15 -20 -25 -30 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 FIGURE H2. Difference in observed from predicted fur harvest records generated using the top ranked negative binomial count models E+TA+EMT (a) and E+TA+MMT (b) for lynx from the East Study Area across central-interior BC, Canada. A value o f zero suggests perfect prediction while negative values suggest over-prediction and positive values suggest under-prediction. (b) (a) 80 X c 3o 60 O 40 TJ 0) o 20 2 0 X X X JL X X -20 -40 -60 X X X X X X X Xx; X 60 X X $ ^ x xx x x * x p x X x x 20 X X$ -20 X -40 x X -60 X -80 * x x X XX X X - X ^ x X 0 x l -p - i XX X X 40 X X 80 x *x X X i^ x x X X £ * % ue X. W xX X X xX X \L/ X X -80 J 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 5 60 T> 20 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 FIGURE H3. Difference in observed from predicted fur harvest records generated using the top ranked negative binomial count models E+TA+SH+EMT (a), E+TA+SH+MMT (b), and E+TA+SH (c) for marten from the West Study Area across central-interior BC, Canada. A value of zero suggests perfect prediction while negative values mean over-prediction and positive values mean under­ prediction. Observed - Predicted Count 60 - X X X X -20 - -40 -60 -80 - -100 -120 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 FIGURE H4. Difference in observed from predicted fur harvest records generated using the top ranked negative binomial count models E+TA+GVGH for marten from the East Study Area across central-interior BC, Canada. A value o f zero suggests perfect prediction while negative values mean over-prediction and positive values mean under-prediction.