University of Montana ScholarWorks at University of Montana
Graduate Student Theses, Dissertations, & Professional Papers Graduate School
2010
Resource selection and abundance estimation of moose: Implications for caribou recovery in a human altered landscape
Wibke Peters The University of Montana
Follow this and additional works at: https://scholarworks.umt.edu/etd Let us know how access to this document benefits ou.y
Recommended Citation Peters, Wibke, "Resource selection and abundance estimation of moose: Implications for caribou recovery in a human altered landscape" (2010). Graduate Student Theses, Dissertations, & Professional Papers. 830. https://scholarworks.umt.edu/etd/830
This Thesis is brought to you for free and open access by the Graduate School at ScholarWorks at University of Montana. It has been accepted for inclusion in Graduate Student Theses, Dissertations, & Professional Papers by an authorized administrator of ScholarWorks at University of Montana. For more information, please contact [email protected]. RESOURCE SELECTION AND ABUNDANCE ESTIMATION OF MOOSE:
IMPLICATIONS FOR CARIBOU RECOVERY IN A HUMAN ALTERED
LANDSCAPE
By
WIBKE ERIKA BRIGITTA PETERS
Bachelor of Science, University of Applied Sciences Eberswalde, Germany, 2006
Thesis
presented in partial fulfillment of the requirements for the degree of
Master of Science in Wildlife Biology
The University of Montana Missoula, MT
December 2010
Approved by:
Perry Brown, Associate Provost for Graduate Education Graduate School
Mark Hebblewhite, Chair Wildlife Biology Program, Department of Ecosystem and Conservation Sciences
Joel Berger, Committee Member Wildlife Biology Program, Division of Biological Sciences
Paul R. Krausman, Committee Member Wildlife Biology Program, Department of Ecosystem and Conservation Sciences
Marco Musiani, Committee Member Faculty of Environmental Design, University of Calgary, Calgary, AB, Canada
ABSTRACT
Peters, Wibke E.B., M.S., December 2010, Wildlife Biology Resource selection and abundance estimation of moose: Implications for caribou recovery in a human altered landscape Chairperson: Mark Hebblewhite, Ph.D. Committee: Joel Berger Ph.D., Paul R. Krausman Ph.D., Marco Musiani Ph.D.
Woodland caribou ( Rangifer tarandus caribou ) are threatened across Canada due to human disturbance altering predator prey dynamics. The niche specialization of caribou enables them to survive in nutrient poor habitats spatially separated from other ungulates and their shared predators. The conversion of old growth forests to young seral stands is hypothesized to increase the abundance of moose ( Alces alces ), the dominant prey for wolves (Canis lupus ), resulting in apparent competition. We first examined habitat selection of moose in 2 regions with differing intensities of human disturbance in west central Alberta and east central British Columbia to assess how human disturbance affects the spatial separation of moose and caribou. We built resource selection functions with data from global positioning system (GPS) collars deployed on 17 moose (8 in a region with high and 9 in a region with low human disturbance) at 2 spatial scales. Our results indicated that moose in our study area make forage risk tradeoffs in a hierarchical fashion similar to caribou, potentially eroding spatial separation in human disturbed landscapes. We also evaluated the spatial partitioning of resources by comparing resource use with GPS locations from 17 moose and 17 paired caribou using logistic regression. As expected, human disturbance decreased the resource partitioning between moose and caribou. Thus, systematic moose management and monitoring will be essential for caribou conservation. Currently, a Stratified Random Block (SRB) survey design is widely used to estimate moose populations, but these surveys are expensive and often result in imprecise population estimates when not corrected for sightability bias. We evaluated the application of distance sampling as an alternative to SRB surveys, especially for use in caribou ranges. To correct for moose missed on the transect line, where a detection rate of 100% is critical, we developed a sightability model using 21 radio collared moose. After correcting for sightability, distance sampling was more precise and efficient than SRB surveys. In this way, more efficient distance sampling methodology can be an important tool for caribou conservation. Combined, our results showed the importance of moose management in caribou ranges due to decreased spatial separation between both ungulate species in disturbed landscapes.
ii
ACKNOWLEDGEMENTS
The completion of this thesis has had significant support and contributions of many people and organizations. I am indebted to my advisor Mark Hebblewhite for his continuous encouragement, patience and for shaping my greater understanding of holistic approaches to conservation biology, academic writing and statistics. I am grateful he gave me the opportunity to participate in this large scale research project and be part of an outstanding research team. Additionally, I greatly appreciate the advice and support of my committee members Joel Berger and Paul Krausman of the University of Montana and Marco Musiani of the University of Calgary. I thank numerous staff of Alberta Fish and Wildlife Division (ABFW) and Alberta Conservation Association (ACA) for their important inputs and help. Especially, I thank Kirby Smith and Dave Hobson (ABFW) for sharing their broad knowledge on the mountains and wildlife of west central Alberta, their outstanding humor and significant contributions to many aspects of this thesis. Nate and Shevenell Webb (ABFW and ACA) helped to make the aerial survey component of this thesis possible, were of immense help with logistics, and I feel fortunate to call them my friends. I am deeply thankful for Saakje Hazenberg’s excellent field assistance, coordination of a huge project, and appreciation for good food, wine and loud music. I am also grateful for the warm welcome I received from Layla Neufeld and Lucas Habib into their home in Jasper. Thanks to the McDermid lab at the University of Calgary and all involved in the creation of the GIS datasets I used. This project would not have been feasible without Clay Wilson from Bighorn Helicopters and also Brad Culling of Diversified Environmental Services who captured animals safely and quickly. I can’t thank John Saunders and Steve Wotton from Peregrine Helicopters and Newman Subritzky from Qwest helicopters enough for safe flying under often difficult conditions. Dave Maier from Silvertip Aviation provided very accurate moose GPS locations without which sightability estimation would have been impossible. I am grateful to the many observers who assisted in spotting moose including (in alphabetical order) Randy Bedell, Lane Doucette, Mike Ewald, Erin Fredlund, Bernie Goski, Dave Hobson, Kirby Smith, Harry Prosser, Conrad Thiessen and Nathan and Shevenell Webb. The Hebblewhite lab at the University of Montana provided me with an unforgettable “American” experience, friendship and encouragement. In particular, I thank Clay for his endless humor, Hugh for always calming me down, Jean for great friendship and proof reading many, many documents, Josh for being a statistics genius, Lacey for her positive philosophy, Nick for sharing his knowledge and priceless GIS assistance, and Shawn for always helping in any way possible. Jeanne Franz has contributed essential administrative assistance and the often needed sugar fix.
iii
I thank my family for their unwavering support and letting me go my own way to explore the world. I am especially grateful to my mom, Kerstin Peters, who sparked my admiration for wildlife and forests and is my strongest advocate. Florian Noll has been my personal cheerleader over past years and I am grateful for his constant encouragement, companionship and tolerance for my work. My Missoula friends greatly influenced my quality of life throughout graduate school, were welcoming and always let me know that I can lean on them; you know who you are. Thanks to long term friends scattered around the world for keeping in touch during busy times. Lastly, thanks to “my” moose – my appreciation for these strong, yet fragile, animals has grown enormously over the duration of my M.S. research. I would like to dedicate this thesis to the memory of my hunting mentor and friend, Willy Lemke, who deepened my respect for wildlife and nature, and put so many things into perspective. Weidmann’s Heil!
I am deeply thankful to the following organizations who provided funding, equipment, logistical support, and personal funding: Alberta Conservation Association (Grant Eligible Conservation Fund), Alberta Sustainable Resource Development, Environment Canada and World Wildlife Fund (Endangered Species Recovery Fund), Parks Canada, Petroleum Technology Alliance Canada/Canadian Association of Petroleum Producers, Royal Dutch Shell, Fulbright Commission, George & Mildred Cirica Scholarship, German Academic Exchange Service, Philanthropic Educational Organization, and the University of Montana.
iv
TABLE OF CONTENTS
Abstract ...... ii Acknowledgements ...... iii Table of Contents ...... v List of Tables ...... vii List of Figures ...... xi Chapter 1: Introduction ...... 1 Literature Cited ...... 4 Chapter 2: Implications of Moose Habitat Selection for Resource Partitioning by Caribou in Human Altered Landscapes ...... 8 Introduction ...... 8 Methods ...... 12 Study Area ...... 12 Animal Capture ...... 13 Moose Resource Selection Function Modeling ...... 14 Moose Caribou Resource Partitioning...... 17 Habitat Covariates...... 19 Results ...... 21 Regional Habitat Availability ...... 21 Landscape Scale Moose Resource Selection ...... 22 Home range scale Moose Resource Selection ...... 23 Moose Caribou Resource Partitioning...... 24 Discussion ...... 26 Management Recommendations ...... 34 Literature Cited ...... 35 Chapter 3: Moose Population Estimation Using Distance Sampling in West Central Alberta ...... 64 Introduction ...... 64 Methods ...... 67 Study Area ...... 67
v
Simple Random Block Surveys ...... 68 Distance Sampling ...... 69 Sightability Trials ...... 71 Correcting for Sightability Bias at g(0) ...... 72 Comparison of Distance Sampling and SRB Surveys ...... 73 Results ...... 73 Simple Random Block Surveys ...... 73 Distance Sampling ...... 74 Sightability Trials ...... 75 Correcting for Sightability Bias at g(0) ...... 75 Comparison of Distance Sampling and SRB Surveys ...... 76 Discussion ...... 77 Management Recommendations ...... 81 Literature Cited ...... 82 Appendix A. Additional Information Chapter 2 ...... 97 Appendix B. Additional Information Chapter 3...... 102 Appendix B 1. Calculation of the moose population estimate and confidence interval following Gasaway et al. (1986) ...... 102 Appendix B 2. Multiple Covariate Distance Sampling ...... 103
vi
LIST OF TABLES
Chapter 2. Table 2 1. The spatial separation hypothesis predicts that woodland caribou spatially separate from other ungulates and their shared predators through their niche specialization. We tested different hypotheses in two landscapes with high and low human landscape disturbance and derived predictions based on the spatial separation hypothesis...... 49 Table 2 2. Description of covariates used in RSF models to estimate moose habitat selection and moose at multiple temporal and spatial scales (data collected 2008 – 2010) and to determine differences in habitat use between moose and woodland caribou (data collected 2007 – 2010) in west central Alberta and east central British Columbia, Canada...... 50 Table 2 3. Selection coefficients (β) and standard errors (SE) from the most parsimonious generalized linear mixed models with a random intercept describing moose resource selection during summer (22 May – 16 Nov) at the landscape and home range scales in the foothills of west central Alberta and east central British Columbia, Canada, from 2008 2010. Coefficients in bold indicate significant variables in the models at an α level of 0.05. Closed conifer was the reference category for landcover types...... 51 Table 2 4. Selection coefficients (β) and standard errors (SE) from the most parsimonious generalized linear mixed models with a random intercept describing moose resource selection during winter (17 Nov – 21 May) at the landscape and home range scales in the foothills of west central Alberta and east central British Columbia, Canada, from 2008 2010. Coefficients in bold indicate significant variables in the models at an α level of 0.05. Closed conifer was the reference category for landcover types...... 52 Table 2 5. Selection coefficients (β) and standard errors (SE) for the most parsimonious generalized linear mixed models with a random intercept describing moose resource selection during summer (22 May – 16 Nov) at the landscape and home range scales in the mountain region in west central Alberta and east central
vii
British Columbia, Canada, from 2008 2010. Coefficients in bold indicate significant variables in the models at an α level of 0.05. Closed conifer was the reference category for landcover types...... 53 Table 2 6. Selection coefficients (β) and standard errors (SE) from the most parsimonious generalized linear mixed models with a random intercept describing moose resource selection during winter (17 Nov – 21 May) at the landscape and home range scales in the mountains of west central Alberta and east central British Columbia, Canada, from 2008 2010. Coefficients in bold indicate significant variables in the models at an α level of 0.05. Closed conifer was the reference category for landcover types...... 54 Table 2 7. Coefficients (β), standard errors (SE) and z values for the most parsimonious generalized linear mixed models with a random intercept describing differences in habitat use by woodland caribou (dependent variable = 1) and moose (dependent variable = 0) in the foothills of west central Alberta and east central British Columbia, Canada. Habitat use was compared during winter (17 Nov – 21 May) and summer (22 May – 16 Nov) from 2007 2009. Closed conifer was the reference category for land cover types. All variables were significant at an α level of 0.05...... 55 Table 2 8. Coefficients (β), standard errors (SE) and z values for the most parsimonious generalized linear mixed models with a random intercept describing differences in habitat use by woodland caribou (dependent variable = 1) and moose (dependent variable = 0) in the mountains of west central Alberta and east central British Columbia, Canada. Habitat use was compared during winter (17 Nov 21 May) and summer (22 May 16 Nov) from 2007 2009. Closed conifer was the reference category for land cover types. All variables were significant at an α level of 0.05...... 56
Chapter 3. Table 3 1. Covariates recorded during sightability surveys for detected (dependent variable = 1) and missed (dependent variable = 0) radio collared moose groups in west central Alberta and east central British Columbia, Canada, conducted in winters 2008/2009 and 2009/2010...... 88 viii
2 Table 3 2. Moose population density estimates ( ˖/km ) in Wildlife Management Unit (WMU) 353 in west central Alberta, 2002 2010, obtained via helicopter based stratified random block (SRB) surveys and distance sampling. For each method and year, the density, survey area, coefficient of variation CV( ˖) at a 90% confidence interval, and survey flight effort in hours of helicopter aircraft time and hours/10km 2 are reported. The distance sampling results are given for an assumed complete probability of detection on the transect line ( g(0) = 1 ) as well as the estimated probability of detection on the transect line ( g(0) = 0.621); SRB results are uncorrected for sightability...... 89 Table 3 3. A priori candidate distance sampling detection functions used to estimate moose density in program DISTANCE 6.2 in winter 2009/2010 in Wildlife Management Unit 353 in west central Alberta, Canada. Ranking was based on the difference in Akaike’s Information Criterion corrected for small sample sizes
( AIC c). k is the number of parameters, χ² GOF is the p value of the χ² goodness of fit test, ˜ is the estimated average detection probability and CV( ˜ ) its coefficient of variation at a 90% confidence interval (CI) , ˖ is the estimated moose density for the study area and CV( ˖) is its coefficient of variation at a 90% CI...... 90 Table 3 4. Mean (SE) distance of moose groups observed or missed during sightability surveys in west central Alberta and east central British Columbia, Canada, in winters 2008/2009 and 2009/2010 and numbers of groups in each group size category (one, two, > two), activity class (bedded, standing/moving), canopy closure category (low, medium, high; in intervals of 33%), terrain class (flat, uneven) and light intensity category on the side the moose was detected or missed (bright, flat)...... 91 Table 3 5. Candidate models for predicting moose sightability in west central Alberta and east central British Columbia, Canada, using radio collared moose. Sightability surveys were conducted in winters 2008/2009 and 2009/2010. Sample size ( n), Log likelihood ( LL ), number of parameters ( k) and the difference of Akaike Information Criterion values for small sample sizes from the model with the
ix
lowest value ( AIC c) for all candidate models with significant parameters at α=0.05 predicting moose detection probability...... 92
Appendix A. Table A 1. Summary of all moose captured between winters 2007/2008 and 2009/2010. Table includes the collar type (Global Positioning System (GPS) or Very High Frequency (VHF) collar), the capture date and the end date or fate of the collar if applicable, the caribou range in or near which the moose has been captured (LSM = Little Smoky, RPC = Redrock Prairie Creek, NAR = Narraway, ALP = A La Peche, JNP = Jasper National Park), sex, number of GPS and VHF locations, and fix rates from individual (moose ID) moose in west central Alberta and east central British Columbia. GPS collars displayed in bold were used for moose resource selection modeling and moose and caribou resource use comparison. .. 97 Table A 2. Proportional availability of categorical landcover classes and means and standard errors (SE) of continuous covariates used to model resource selection by moose at multiple scales in the foothills and mountains of west central Alberta and east central British Columbia, Canada. Availability was defined by a buffer around summer (22 May – 16 Nov) and winter (17 Nov – 21 May) 99% fixed kernel home ranges estimated with global positioning system (GPS) collar data. Buffer size was equal to the seasonal maximum distance of GPS locations for each individual moose. GPS data were collected with 8 moose in the foothills and 9 moose in the mountains between winters of 2007/2008 and 2009/2010...... 99 Table A 3. Mean and standard deviation of continuous covariates measured at used moose and woodland caribou global positioning system (GPS) locations during summer (22 May – 16 Nov) and winter (17 Nov – 21 May) in the foothills (FH) and mountains (MT) of west central Alberta and east central British Columbia, Canada. Data were collected with 17 moose and 17 caribou GPS collars between winters of 2007/2008 and 2009/2010...... 100
x
LIST OF FIGURES
Chapter 2. Figure 2 1. Study area based on the maximum extent of available locations for moose resource selection function modeling within the foothills (northeast) and mountains (southwest) of west central Alberta and east central British Columbia, Canada. Seventeen Global Positioning System (GPS) were deployed on moose between winters 2007/2008 and 2009/2010 (see 99% home range kernels) within or adjacent to caribou herd home ranges (see 95% home range kernels). Data (collected between winters 2006/2007 and 2009/2010) from 17 GPS collars deployed on caribou of the displayed herds were used for comparison of resource use between caribou and moose...... 57 Figure 2 2. Selection coefficients (β) with standard error bars for young and old cutblocks and burns, and shrubs from the most parsimonious generalized linear mixed models with a random intercept estimating moose resource selection at the landscape and home range scale during summer (22 May 16 Nov) and winter (17 Nov 21 May). Data were collected with 17 global position in system collars in the foothills (a; n=8) and mountains (b; n=9) of west central Alberta and east central British Columbia, Canada, between winters of 2007/2008 and 2009/2010. All estimates are in comparison to the categorical land cover variables subsumed in the model specific intercept (of top models)...... 58 Figure 2 3. Selection coefficients (β) for road density (km/km 2) from the most parsimonious generalized linear mixed models with a random intercept describing moose resource selection at the landscape and home range scales during summer (22 May –16 Nov) and winter (17 Nov – 21 May). Significant coefficients are marked with a star. Data were collected with 17 moose global positioning system collars in the foothills (a; n=8) and mountains (b; n=9) of west central Alberta and east central British Columbia, Canada, between winters of 2007/2008 and 2009/2010...... 59 Figure 2 4. Selection coefficients for quadratic functions describing density of linear features (seismic exploration lines, pipe lines, railways and human trials
xi
combined; km/km 2) from the most parsimonious generalized linear mixed models with a random intercept estimating moose resource selection at the landscape and home range scales during summer (22 May – 16 Nov) and winter (17 Nov – 21 May). Data were collected with 17 moose GPS collars in the foothills (a, b; n=8) and mountains (c, d; n=9) of west central Alberta and east central British Columbia, Canada, between winters of 2007/2008 and 2009/2010...... 60 Figure 2 5. Relative probability of use by woodland caribou and moose from the most parsimonious generalized linear mixed models with a random intercept during summer (22 May – 16 Nov) and winter (17 Nov – 21 May) in the foothills (a, b) and mountains (c, d) of west central Alberta and east central British Columbia, Canada. Values closer to 0 indicate high relative probability of use by moose and conversely, values closer to 1 indicate high relative probability of use by caribou. Areas with high overlap of both species indicate low resource partitioning. Models were built with data from global positioning collars (GPS) deployed on 17 moose and 17 caribou between winters 2007/2008 and 2009/2010...... 61 Figure 2 6. Moose and caribou niche overlap based on predicted probabilities of resource use from most parsimonious generalized linear mixed models with a random intercept describing resource partitioning of both species in the foothills (FH) and mountains (MT) of west central Alberta and east central British Columbia, Canada, during summer (su; 22 May – 16 Nov) and winter (wi; 17 Nov – 21 May). Models were built with data collected with global positioning collars deployed on 17 moose and 17 caribou between winters 2007/2008 and 2009/2010. Niche overlap indices were calculated following MacArthur and Levins (M&L; 1967) and Pianka (1973)...... 62 Figure 2 7. Caribou mortalities versus predicted probabilities of resource use by woodland caribou relative to moose in the foothills (FH) and mountains (MT) of west central Alberta and east central British Columbia, Canada, during summer (su; 22 May 16 Nov) and winter (wi; 17 Nov – 21 May). Values of 0.1 and 0.2 indicate the highest relative probability of use by moose and conversely, values of 0.9 and 1 indicate the highest relative probability of use by caribou. Predictive logistic regression models were built with data collected with global positioning
xii
collars (GPS) deployed on 17 moose and 17 caribou between winters 2007/2008 and 2009/2010. We used 45 predator caused caribou mortalities recorded by Alberta Fish and Wildlife Division and Parks Canada between 1999 and 2009. . 63
Chapter 3. Figure 3 1. Conceptual example of line transect sampling with decreased probability of detection on the transect line ( g(0) ≠ 1 ). The probability of sighting an animal decreases within 8 distance classes with increasing distance from the transect line (g(0) ). Detection function A assumes that no animals are left undetected along the transect line ( g(0) = 1 ). Detection function B shows the rescaled detection function if 6 out of 14 animals remained undetected along the transect line. Rescaling the detection function shifts the intercept of the function up, which corrects for the bias of decreased detection along the transect line...... 93 Figure 3 2. Study area located in west central Alberta and east central British Columbia, Canada. Stratified random block (SRB) surveys to estimate moose population size were conducted in winters 2001/2002, 2006/2007 and 2008/2009 in wildlife management unit (WMU) 353 (in 2008/2009 only shaded area was surveyed). Furthermore, a distance sampling survey to estimate moose density was conducted in winter 2009/2010 in WMU 353, following a pilot survey in WMU 440 in winter 2008/2009. Lastly, sightability data were collected using radio collared moose in winters 2008/2009 and 2009/2010 to develop a sightability model...... 94 Figure 3 3. Estimated detection probability function of moose groups in wildlife management unit 353 in west central Alberta, Canada, modeled in program DISTANCE 6.2. Distance sampling survey data were collected in winter 2009/2010. Seventy one moose groups were detected on 33 transect lines (total of 777.8 km).The model is a half normal key function with no adjustment terms, and did not correct for g(0) < 1...... 95 Figure 3 4. Required helicopter flight effort in hours per 10km 2 versus the estimated moose densities per km 2 from aerial surveys using a simple random block (SRB) survey design and distance sampling within wildlife management unit 353 in Alberta, Canada. Values next to symbols indicate the year the survey was xiii
conducted and the coefficient of variation (CV) of the moose density estimate. Distance sampling density estimates were corrected for visibility bias, while SRB surveys are shown uncorrected. Thus, density estimates and CVs of SRB are likely underestimated. The distance sampling “Scenario” is an estimate of the survey time required to achieve an anticipated CV of ± 20% at a 90% CI of the distance sampling survey conducted in winter 2009/2010. Effort does not include fixed wing survey flight time required for pre stratification flights in SRB designs...... 96
Appendix A. Figure A 1. Average daily travel speed (km/hr) of 17 global positioning system (GPS) collared moose (data collected between winters 2007/2008 and 2009/2010) and 212 woodland caribou in 7 herds (data collected between winters 1999/2000 and 2008/2009; N. DeCesare, University of Montana, unpublished data) in west central Alberta and east central British Columbia, Canada, to delineate seasons for resource selection function modeling. We identified the start of summer using the timing of calving as 22 May when movements of female moose and caribou decreased. The start of the winter season was identified by the end of caribou and moose fall migration, which we estimated to be approximately 17 November when movement rates decreased...... 101
xiv
CHAPTER 1: INTRODUCTION
Population declines of endangered species may be attributed to proximate or ultimate causes (Caughley 1994, Sinclair and Byrom 2006). Proximate causes such as predation and invasive species often result from ultimate causes, like human induced habitat loss and climate change, which affect ecosystems at large scales (Simberloff 1986, Sinclair and Byrom 2006). Proximate and ultimate causes act concurrently and their combination can drive vulnerable populations towards extinction (Duinker and Greig 2006, Mills 2007). For example, the decline of northern spotted owls (Strix occidentalis caurina ) has been partially attributed to competition and inbreeding with barred owls ( Strix varia; proximate cause of decline) that have been increasing in spotted owl habitat benefiting from large scale landscape development by humans (ultimate cause of decline; Peterson and Robins 2003, Haig et al. 2004). This example shows that to conserve endangered species it is essential to understand the relationships and interactions between threatened species and the communities they live in. Furthermore, though abiotic and biotic factors affecting the dynamics of individual populations are complex on spatial and temporal scales (Sinclair and Byrom 2006), they must be considered when studying population declines and developing conservation strategies. In 2000, the Committee on the Status of Endangered Wildlife in Canada (COSEWIC) listed boreal and southern mountain woodland caribou (Rangifer tarandus caribou ) as federally threatened under the Species at Risk Act (Government of Canada 2002). While the ultimate cause of caribou declines is landscape alteration by forestry and energy development in Alberta, the proximal mechanisms are mediated by changes in predator prey dynamics (James et al. 2004). The conversion of old growth forests to young seral stands is hypothesized to increase the abundance of moose ( Alces alces ), the dominant prey for wolves (Canis lupus ) throughout the boreal forest, and thereby might support larger wolf and other predator populations (McNicol and Gilbert 1980, Kunkel and Pletscher 2000, Wittmer et al. 2005, Stotyn 2008). Changes in species composition and distribution may exacerbate population declines and extinction of threatened populations through apparent competition, the process by which two prey species can affect each other’s fitness through their numerical
1 response of a shared predator (Holt 1977, Holt and Lawton 1994). While moose and caribou are sympatric throughout the boreal coniferous biome, they are hypothesized to coexist through resource partitioning (Boer 2007, Hirzel and Le Lay 2008). The diet of woodland caribou is comprised of terrestrial and arboreal lichens, especially during winter (Klein 1982, McLoughlin et al. 2006). Moose are generalist browsers, mainly feeding on shrubs (Renecker and Schwartz 2007) and they select early succession forest stages (e.g., after fire, forest harvesting, insect outbreaks, windfall) that provide improved forage (McNicol and Gilbert 1980, Peek 2007). Prey species are able to adopt antipredator strategies including the use of refuges to reduce predation risk (Sih 1987). The spatial separation hypothesis suggests that the niche specialization by caribou enables them to survive in nutrient poor habitats spatially separated from other ungulates and their predators, reducing the negative effects of apparent competition by avoidance of wolves and thereby increasing survival and persistence (Bergerud 1974, Fuller and Keith 1981, Bergerud and Page 1987, Seip 1992). However, human landscape disturbance leads to an increase in younger forests selected by moose and spatial fragmentation of older forests that caribou select, limiting the ability of caribou to spatially separate from moose and leading to higher predator numbers searching per unit area (Bergerud et al. 1984, James et al. 2004). Thus, understanding resource partitioning by moose and caribou in the context of spatial separation is key to evaluating the foundation of apparent competition. Moreover, knowledge of moose density is important for caribou conservation and recovery planning if caribou declines are a result of apparent competition with moose, because predator numbers are often regulated by their prey base (Wegge and Storaas 2009). For example, modeling studies by Weclaw and Hudson (2004), Lessard et al. (2005), and Courtois and Quellet (2007) reported wolf reduction without concurrent moose reduction will fail to recover caribou. Wolf populations will quickly recuperate, unless moose density is also reduced (Hayes et al. 2003) due to rapid increases of moose following predator control and subsequent attraction of wolves from adjacent areas. Therefore, the Alberta Woodland Caribou Recovery Plan recommends active management of predators in combination with controlling moose densities in caribou ranges (Alberta Woodland Caribou Recovery Team 2005). Contrarily, wildlife managers are also under pressure to increase moose populations for subsistence and recreational
2 hunting (James et al. 2004).To simultaneously consider these conflicting goals, management of moose populations needs to be monitored effectively to promote caribou recovery and to achieve harvest goals. Despite these recommendations, little is known about moose abundance and population trend in west central Alberta and monitoring efforts by Alberta Fish and Wildlife (ABFW) are hampered by limited financial resources. Aerial surveys have become an invaluable tool to estimate population size of wildlife (Caughley 1974, Samuel et al. 1987, Anderson and Lindzey 1996). In North America, moose populations are commonly surveyed using a stratified random block (SRB) sampling design (Gasaway et al. 1986). Due to high time and cost requirements (Ward et al. 2000), this block based method is appropriate for dense moose populations in small survey areas and open habitats (Buckland et al. 2001, Nielson et al. 2006). Yet, woodland caribou often occur in closed habitats over large areas, making the expensive SRB survey approach ineffective to monitor moose population size and trend especially in caribou habitat. My thesis focuses on two themes of moose ecology: habitat selection and abundance, within the framework of understanding apparent competition between moose and caribou. In chapter 2, I first examined habitat selection of moose with respect to specific predictions of the spatial separation hypothesis in two regions with differing intensities of human landscape disturbance using resource selection functions (Manly et al. 2002). After assessing what habitat components moose select at coarser and finer scales during summer and winter, I compared moose and caribou habitat use to examine resource partitioning of both species. Next, in chapter 3, I assessed the applicability of aerial distance sampling (Buckland et al. 2001) for moose population estimation in my study area and compared my results to more traditional SRB survey designs. Further, I tested for sightability bias when conducting aerial moose surveys in west central Alberta and east central British Columbia and developed a moose sightability model. Ultimately, understanding the distribution and abundance of moose will allow development of conservation strategies that depend on the nature of the relationship between resource selection and density. For example, if moose densities are highest in selected habitats, then this has different implications for caribou conservation than if occurrence and abundance are not tightly linked. The two components of my thesis, chapter 2 (resource
3 selection) and chapter 3 (abundance) will allow an evaluation of the relationship between abundance and distribution in the future. Chapter 2 and 3 are anticipated for scientific publication and are co authored by Mark Hebblewhite (Chapter 2 and 3), Nick DeCesare (Chapter 2), Kirby Smith (Chapter 3) and Shevenell M. Webb (Chapter 3). Due to the significant contributions of the large group of people who have assisted with this project, I refer to the first person plural (“we”) instead of the first person singular (“I”) throughout this thesis.
LITERATURE CITED
Alberta Woodland Caribou Recovery Team. 2005. Alberta woodland caribou recovery plan 2004/05 2013/14. Edmonton, AB. Anderson, C. R., and F. G. Lindzey. 1996. Moose sightability model developed from helicopter surveys. Wildlife Society Bulletin 24:247 259. Bergerud, A. T. 1974. Decline of caribou in North America following settlement. Journal of Wildlife Management 38:757 770. Bergerud, A. T., H. E. Butler, and D. R. Miller. 1984. Antipredator tactics of calving caribou dispersion in mountains. Canadian Journal of Zoology 62:1566 1575. Bergerud, A. T., and R. E. Page. 1987. Deplacement and dispersion of parturient caribou at calving as antipredator tactics. Canadian Journal of Zoology 65:1597 1606. Boer, A. H. 2007. Interspecific Relationships. Pages 337 349 in A. W. Franzmann, and C. C. Schwartz, editors. Ecology and Management of the North Amerian Moose. The University Press of Colorado, Boulder, Colorado, USA. Buckland, S. T., D. R. Anderson, K. P. Burnham, J. L. Laake, D. L. Borchers, and L. Thomas. 2001. Introduction to Distance Sampling, Estimating abundance of biological populations. Oxford University Press, New York. Caughley, G. 1974. Bias in aerial survey. Journal of Wildlife Management 38:921 933. _____. 1994. Directions in conservation biology. Journal of Animal Ecology 63:215 244. Courtois, R., and J. P. Ouellet. 2007. Modeling the impact of moose and wolf management on persistence of woodland caribou. Alces 43:13 27.
4
Duinker, P. N., and L. A. Greig. 2006. The impotence of cumulative effects assessment in Canada: Ailments and ideas for redeployment. Environmental Management 37:153 161. Fuller, T. K., and L. B. Keith. 1981. Woodland caribou population dynamics in Northeastern Alberta. Journal of Wildlife Management 45:197 213. Gasaway, W. C., S. D. DuBois, D. J. Reed, and S. J. Harbo. 1986. Estimating moose population parameters from aerial surveys. Biological Papers of the University of Alaska, Fairbanks. Government of Canada. 2002. Species at Risk Act: An act representing the protection of wildlife species at risk in Canada. in c. Statutes of Canada, vol. Bill C 5, editor. Haig, S. M., T. D. Mullins, E. D. Forsman, P. W. Trail, and L. Wennerberg. 2004. Genetic identification of Spotted Owls, Barred Owls, and their hybrids: Legal implications of hybrid identity. Conservation Biology 18:1347 1357. Hayes, R. D., R. Farnell, R. M. P. Ward, J. Carey, M. Dehn, G. W. Kuzyk, A. M. Baer, C. L. Gardner, and M. O'Donoghue. 2003. Experimental reduction of wolves in the Yukon: Ungulate responses and management implications. Wildlife Monographs:1 35. Hirzel, A. H., and G. Le Lay. 2008. Habitat suitability modelling and niche theory. Journal of Applied Ecology 45:1372 1381. Holt, R. D. 1977. Predation, apparent competition, and structure of prey communities. Theoretical Population Biology 12:197 229. Holt, R. D., and J. H. Lawton. 1994. The ecological consequences of shared natural enemies. Annual Review of Ecology and Systematics 25:495 520. James, A. R. C., S. Boutin, D. M. Hebert, and A. B. Rippin. 2004. Spatial separation of caribou from moose and its relation to predation by wolves. Journal of Wildlife Management 68:799 809. Klein, D. R. 1982. Fire, lichens, and caribou. Journal of Range Management 35:390 395. Kunkel, K. E., and D. H. Pletscher. 2000. Habitat factors affecting vulnerability of moose to predation by wolves in southeastern British Columbia. Canadian Journal of Zoology 78:150 157.
5
Lessard, R. B., S. J. D. Martell, C. J. Walters, T. E. Essington, and J. F. Kitchell. 2005. Should ecosystem management involve active control of species abundances? Ecology and Society 10 (2). Manly, B. F. J., L. L. McDonald, D. L. Thomas, T. L. McDonald, and W. P. Erickson. 2002. Resource selection by animals: statistical design and analysis for field studies. 2nd edn. Kluwer Academic Publishers, Dordrecht, Netherlands. McLoughlin, P. D., M. S. Boyce, T. Coulson, and T. Clutton Brock. 2006. Lifetime reproductive success and density dependent, multi variable resource selection. Proceedings of the Royal Society B Biological Sciences 273:1449 1454. McNicol, J. G., and F. F. Gilbert. 1980. Late winter use of upland cutovers by moose. Journal of Wildlife Management 44:363 371. Mills, L. S. 2007. Conservation of Wildlife Populations; demography, genetics, and management. 2nd edition. Blackwell Publishing Ltd, Malden, USA. Nielson, R. M., L. L. McDonald, and S. D. Kovach. 2006. Aerial line transect survey protocols and data analysis methods to monitor moose (Alces alces) abundance as applied on the Innoko National Wildlife Refuge, Alaska. Technical report prepared for US Fish and Wildlife Service Peek, J. M. 2007. Habitat relationships. Pages 351 375 in A. W. Franzmann, and C. C. Schwartz, editors. Ecology and management of the North American moose. University Press Colorado, Boulder, Colorado. Peterson, A. T., and C. R. Robins. 2003. Using ecological niche modeling to predict Barred Owl invasions with implications for Spotted Owl conservation. Conservation Biology 17:1161 1165. Renecker, L. A., and C. C. Schwartz. 2007. Food habits and feeding behavior. Pages 403 439 in A. W. Franzmann, and C. C. Schwartz, editors. Ecology and management of the North American moose University Press of Colorado Boulder, Colorado, USA. Samuel, M. D., E. O. Garton, M. W. Schlegel, and R. G. Carson. 1987. Visibility bias in aerial surveys of elk in North central Idaho. Journal of Wildlife Management 51:622 630.
6
Seip, D. R. 1992. Factors limiting woodland caribou populations and their interrelationships with wolves and moose in Southeastern British Columbia. Canadian Journal of Zoology 70:1494 1503. Sih, A. 1987. Prey refuges and predator prey stability. Theoretical Population Biology 31:1 12. Simberloff, D. 1986. The proximate causes of extinction. Pages 259 276 in D. M. Raup, and D. Jablonski, editors. Patterns and processes in the history of life Springer Verlag Berlin. Sinclair, A. R. E., and A. E. Byrom. 2006. Understanding ecosystem dynamics for conservation of biota. Journal of Animal Ecology 75:64 79. Stotyn, S. A. 2008. Ecological interactions of mountain caribou, wolves and moose in the North Columbia Mountains, British Columbia. Master of Science Thesis, University of Alberta, Edmonton. Ward, M. P., W. C. Gasaway, and M. M. Dehn. 2000. Precision of moose density estimates derived from stratification survey data. Alces 36:197 203. Weclaw, P., and R. J. Hudson. 2004. Simulation of conservation and management of woodland caribou. Ecological Modeling 177:75 94. Wegge, P., and T. Storaas. 2009. Sampling tiger ungulate prey by the distance method: lessons learned in Bardia National Park, Nepal. Animal Conservation 12:78 84. Wittmer, H. U., A. R. E. Sinclair, and B. N. McLellan. 2005. The role of predation in the decline and extirpation of woodland caribou. Oecologia 144:257 267.
7
CHAPTER 2: IMPLICATIONS OF MOOSE HABITAT SELECTION FOR RESOURCE
PARTITIONING BY CARIBOU IN HUMAN ALTERED LANDSCAPES
INTRODUCTION
Biologists need to understand the mechanisms leading to population declines and potential interactions among those to manage and conserve species. Woodland caribou (Rangifer tarandus caribou ) are listed as threatened under the Alberta Wildlife Act (2000) and throughout Canada under the Species at Risk Act (Government of Canada 2002). Throughout their distribution, widespread human disturbance (i.e., energy and forestry exploitation and associated road and seismic line implementation) is the ultimate cause for caribou declines (COSEWIC 2002, McLoughlin et al. 2003). However, direct habitat loss is unlikely to limit forage for woodland caribou because most populations are hypothesized to be well below the forage carrying capacity (Bergerud 1974, McLoughlin et al. 2003, Wittmer et al. 2005a). Instead, the leading proximate cause for caribou declines is hypothesized to be increased predation by wolves ( Canis lupus ) due to altered predator prey interactions following habitat disturbance by humans (James et al. 2004). Holt (1977) coined the term “apparent competition”, which occurs when two species indirectly compete via a shared predator. Higher primary prey density can support the numeric response of a shared predator, resulting in higher predation rates that can drive alternative prey extinct (Sinclair et al. 1998, Wittmer et al. 2005b). For woodland caribou, the conversion of old growth forests to young seral stands likely increases the abundance of moose ( Alces alces ), the dominant prey for wolves throughout caribou distribution (McNicol and Gilbert 1980, Wittmer et al. 2005a, Stotyn 2008). Increasing moose density is assumed to contribute to higher wolf density, resulting in higher predation rates on caribou (James and Stuart Smith 2000, McLoughlin et al. 2003, James et al. 2004). Additionally, linear features, such as roads and seismic lines, can facilitate the efficiency of wolf predation on caribou (James and Stuart Smith 2000). Where human activities promote apparent competition, it is important to understand the temporal and spatial distribution of the competing species (Holt and Lawton 1994). According to ecological niche theory, different species may exploit different habitat components, resulting in fitness differences between species in geographic space
8
(MacArthur and Levins 1967, Hirzel and Le Lay 2008). Habitat can be defined as the resources and conditions present in an area producing occupancy and determining the survival and reproduction of a certain organism (Hall et al. 1997). Individual habitat selection, the multi scale process by which an animal chooses resources (Johnson 1980), can lead to spatial and temporal segregation or differing diet preferences (Stelfox and Taber 1969) and thereby facilitate the coexistence of sympatric species. Moose and caribou are sympatric throughout much of their ranges and they are hypothesized to coexist through resource partitioning (Boer 2007). Caribou diet is comprised of terrestrial and arboreal lichens, especially during winter, and lichen biomass is highest in low productivity and older forest stands (Servheen and Lyon 1989, Thomas et al. 1996, McLoughlin et al. 2006). Moose are generalist browsers, mainly feeding on shrubs (Renecker and Schwartz 2007), and select early succession forest stages (e.g., following fire or forest harvesting) that provide improved forage (McNicol and Gilbert 1980, Wittmer et al. 2005a, Peek 2007). The spatial separation hypothesis suggests that the niche specialization of caribou should spatially separate them from moose and their predators (Bergerud and Page 1987, Seip 1992, James et al. 2004). Thus, understanding habitat selection and resource partitioning by moose and caribou is key for evaluating the foundations of apparent competition. Habitat selection is influenced by a variety of covariates, such as nutrition, behavior, competition, predation, and the scale of selection (Manly et al. 2002, Hirzel and Le Lay 2008), but is assumed, albeit often without rigorous tests, to be related to fitness (Boyce and McDonald 1999, Pearce and Ferrier 2001, Railsback et al. 2003, Nielsen et al. 2005). Due to the high conservation concern for caribou, habitat selection patterns have been analyzed for many populations across Canada (Rettie and Messier 2000, Apps et al. 2001, Johnson et al. 2004, Gustine et al. 2006). In general, these studies suggest that caribou select older forests in areas with reduced human disturbance, isolating themselves from moose and predators (Stuart Smith et al. 1997, James et al. 2004, McLoughlin et al. 2005). In particular, in boreal forests, caribou select large contiguous patches of low productivity, older seral stands and occur in low densities (Euler 1981, McLoughlin et al. 2005), whereas in mountains, caribou generally select higher elevations and exhibit distinct migratory behavior (Seip 1992, Stotyn 2008). These mechanisms support the
9 general predictions of the spatial separation hypothesis, assuming that moose prosper in early succession vegetation communities due to a higher quantity and quality of forage at lower elevations (Timmermann and McNicol 1988, Forbes and Theberge 1993). However, because human development leads to an increase in younger forests and fragments mature forests, human disturbance may limit the ability of caribou to spatially separate from moose and hence, predation risk (Seip 1992, James et al. 2004). For ungulates, the trade off between forage and predation is a key driver of resource selection (Fryxell et al. 1988, Rettie and Messier 2000, Dussault et al. 2005). High forage biomass can also be coupled with increased exposure to predation risk and lack of shelter and hiding cover (Dussault et al. 2004, Dussault et al. 2005). For example, forestry practices increase forage availability (Poole and Serrouya 2003), but also adds linear features (roads), which promote e.g., human and wolf predation efficiency on moose and caribou (Bergerud 1974, James and Stuart Smith 2000, Neufeld 2006). Further, wolves often select for landcover types with high forage biomass (e.g., shrub communities, burns, logged areas) presumably to increase encounter rates with prey (Hebblewhite et al. 2005). Black (Ursus americanus ) and grizzly bears ( Ursus arctos ) have also been reported to select human disturbed habitats due to increased forage quality and quantity for these omnivores (Mosnier et al. 2008, Roever et al. 2008). Thus, moose, like caribou, must make trade offs between food availability and exposure to other limiting factors (Dussault et al. 2005). It is generally hypothesized that ungulates respond to risk forage trade offs in a hierarchical fashion (Senft et al. 1987), and may select habitats that reduce risk of predation at coarser scales and maximize forage intake at smaller scales (Rettie and Messier 2000, Johnson et al. 2001, Hebblewhite and Merrill 2009). However, local habitat selection can vary substantially (Peek 2007) and moose ecologists caution against making generalizations about moose habitat selection (Timmermann and McNicol 1988, Osko et al. 2004). Unfortunately, direct comparisons of moose and caribou habitat selection are rare, and the connection between human activity and multi scale moose habitat selection is required to understand the mechanisms of apparent competition. Our broad objectives were to determine the relationship between habitat alteration by humans and moose resource selection, and to understand the spatial separation
10 between moose and caribou. Our first specific objective was to assess seasonal habitat selection using location data from global positioning system (GPS) collared moose at multiple spatial scales across a boreal foothills mountain gradient with differing intensities of human disturbance (Table 2 1). According to the spatial separation hypothesis, moose are predicted to select for browse rich habitats (i.e., shrubs, recent burns and recently logged areas) often associated with human resource extraction activities (Seip 1992, McLoughlin et al. 2005). However, if moose select forage (and thus cutblocks, which are synonymous with clear cuts in our study area, in human disturbed landscapes) at coarse scales as generally assumed under the spatial separation hypothesis (Forbes and Theberge 1993), we might expect moose to avoid predation risk (including human hunting) at finer spatial scales (Dussault et al. 2005, Berger 2007, Poole et al. 2007, Winnie and Creel 2007). Overall, moose habitat selection itself could be a driver of reduced spatial separation between caribou and moose depending on the similarity of scale dependent avoidance of human disturbance by both species. We tested whether moose habitat selection at coarse scales is driven by selection for improved forage (e.g., cutblocks, burns) as predicted by the spatial separation hypothesis (P1a in Table 2 1). Next, we examined whether spatial separation between caribou and moose might be diluted in human disturbed landscapes, in part due to such changes in moose resource selection for human developments. We predicted that at fine scales, moose would avoid predation risk by avoiding roads and other linear features and cutblocks (P1b in Table 2 1; Eason 1989, Courtois et al. 2002). In this way, finer scale resource selection by moose may erode spatial separation with caribou, resulting in increased niche overlap. Our second major objective was to understand the relationship between moose and caribou resource partitioning and resource overlap (Table 2 1). We predicted moose should use areas disturbed by humans more than caribou at lower elevations, in younger forests structures (i.e., cutblocks, burns) and in areas with high browse availability (P2a in Table 2 1; Seip 1992, McLoughlin et al. 2005). The diet of moose and caribou can overlap in summer when both species consume forbs and deciduous vegetation at higher elevations (Servheen and Lyon 1989, Boer 2007). Thus, we predicted that spatial separation would be lower in summer due to resource overlap. We evaluated the degree of spatial overlap in adjacent boreal foothills and mountain landscapes with varying
11 human impacts using data from GPS collared sympatric moose and caribou. Habitat selection studies model the realized niche and thus, can be used to reflect resource use and overlap between species (Hirzel and Le Lay 2008). Niche overlap characterizes the use of shared resources by multiple species (Abrams 1980), in our study by moose and caribou. We measured resource use (niche) overlap using two indices following MacArthur and Levins (1967) and Pianka (1973). We predicted that overlap would be 1) higher during summer due to forage overlap (P3a in Table 2 1), and 2) higher in more heavily developed lower elevation boreal foothills (P3b in Table 2 1). In terms of niche theory, a prediction of apparent competition is asymmetric overlap in favor of moose; namely, that moose will overlap more with caribou niches than caribou with moose niches (DeCesare et al. 2010). Finally, the spatial separation hypothesis predicts increased risk of mortality in areas of higher overlap between moose and caribou (McLoughlin et al. 2005). We tested whether most predation caused mortalities of caribou occurred with the highest probabilities of overlap of moose and caribou habitat use (P4a in Table 2 1).
METHODS
STUDY AREA We studied resource selection of moose and caribou across the foothills and mountains of west central Alberta (AB) and east central British Columbia (BC) within the ranges of 5 woodland caribou herds (Figure 2 1). We divided our study area into a foothill (16,750 km 2) and a mountain region (37,306 km 2) based on the ecological regions of North America (Commission for Environmental Cooperation 2009). Moose were then assigned a region based on where the majority (i.e., > 50%) of GPS locations occurred. The mountain region was characterized by low human disturbance and a high proportion of protected areas (approximately 50%), including Jasper and Banff National Parks, the Wilmore Wilderness, Kakwa Wildland Park (AB) and Kakwa Provincial Park (BC; Figure 2 1). The foothills region had a higher proportion of provincial lands managed primarily for resource extraction, with correspondingly higher human disturbance in the form of forest harvest cutblocks and linear developments (e.g., roads, pipelines, seismic lines). Elevations ranged from about 500m in the foothills to more than 3,000m in the
12 mountains. The foothills region was characterized by mixed wood forests, comprised mainly of trembling aspen ( Populus tremuloides ), lodgepole pine ( Pinus contorta ), white spruce ( Picea glauca ), and black spruce ( Picea mariana ); while the western forests in the mountain region, were dominated by lodgepole pine and engelman spruce ( Picea engelmanii ). Moose and white tailed deer ( Odocoileus virginianus ) comprised the majority of the ungulate population, whereas elk (Cervus canadensis ), mule deer (Odocoileus hemionus ) and woodland caribou were less common. Bighorn sheep ( Ovis canadensis ) and mountain goats ( Oreamnos americanus ) inhabited the mountain region. In addition to wolves, other large predator species included black bear, grizzly bear, coyote ( Canis latrans ), cougar ( Puma concolor ), wolverine ( Gulo gulo ) and lynx ( Lynx canadensis ). Human hunting of moose by Treaty First Nations (i.e., year round, unregulated) and by licensed hunting in late fall, occurred throughout the study area (except for most protected areas), but was highest in the foothills region. Moose harvest varied from 6% of the estimated population in the mountains to 30 40% of the estimated population in the foothills (AB Fish and Wildlife Division, unpublished data). Licensed caribou hunting in the study area has been banned in the study area since the 1980s, and although a small amount of Treaty First Nation harvest may occur, caribou harvest is likely very low or absent.
ANIMAL CAPTURE We captured and radio collared moose via net gunning (Barrett et al. 1982, Carpenter and Innes 1995) in winters of 2007/2008 and 2008/2009. We used data from Global Positioning System (GPS) collars (ATS G2000 GPS collars; Advanced Telemetry Systems, Isanti, MN, USA) deployed on 10 female and 7 male moose within and adjacent to caribou ranges (Figure 2 1; Appendix A, Table A 1). We radio collared female and male moose to evaluate population level habitat selection and moose population overlap with female caribou. For threatened caribou populations, female caribou are the most relevant sex to study, because adult female caribou survival drives caribou population growth rates (Gaillard et al. 2000, DeCesare et al., in press). Therefore, we used GPS collar (GPS 3300, 4400, LOTEK Engineering Ltd., Newmarket, ON, Canada) data from 17 female caribou, captured using the same methods as described above for moose. Net 13 gunning protocols were approved by the University of Montana Animal Care and Use Protocol 056 56MHECS 010207 and 059 09MHWB 122109, Alberta Sustainable Resource Development collection licenses #21803, #27086, #27088, #27090 and Parks Canada research and collection permit JNP 2007 952. Our moose and caribou sample sizes were comparable to recent similar GPS wildlife studies (Hebblewhite and Haydon 2010), but lower than former moose studies using GPS collar technology, in part because of the unfortunate failure of 8 moose GPS collars (see Appendix A, Table A 1). Both moose and caribou GPS collars collected locations every 2 to 4 hours, which we re sampled to a consistent 4 hour relocation schedule. The majority of moose GPS collars were deployed for approximately 1 year, but caribou GPS data often spanned a longer time frame. Therefore, we limited caribou location data to one calendar year as well. Fix rate success of less than 90% can cause habitat induced bias in resource selection studies (Frair et al. 2004). In our study, fix rates for moose and caribou were 92.4% and 90.3% respectively. As a result, we did not correct for habitat induced fix rate bias.
MOOSE RESOURCE SELECTION FUNCTION MODELING Resource Selection Functions (RSFs) provide quantitative, spatially explicit, predictive models of animal occurrence (Manly et al. 2002). To estimate moose resource selection, we used a logistic regression framework to compare used resources relative to available resources (Manly et al. 2002). Johnson (1980) described resource selection as hierarchical across spatial scales and defined the following orders of selection: geographic range (first order), between home ranges within the landscape (second order), within the home ranges (third order), and fine scale elements (e.g., food, den sites; fourth order). We modeled RSFs at second (i.e., landscape) and third order (i.e., home range) scales during winter (18 November – 21 May) and summer (22 May – 17 November). We delineated seasons using movement behavior to reflect animal responses to seasonal shifts in resource selection (Vander Wal and Rodgers 2009). We estimated the average daily movement rate of individual moose and 7 caribou herds (N. DeCesare, University of Montana, unpublished data) and stratified GPS data into two seasons, summer and winter (Appendix A, Figure A 1). We identified the start of summer using the timing of calving following the spring migration as 22 May, when movements of female moose and caribou decreased. The start of the winter season was identified by the end of caribou and 14 moose fall migration, which we visually estimated to be approximately 17 November (Appendix A, Figure A 1). We estimated seasonal RSFs at the landscape scale by sampling availability using 1,000 random used locations within each moose’s home range and 1,000 random available locations within a buffer around each home range (Katnik and Wielgus 2005). We determined the buffer size based on the seasonal maximum distance between GPS locations for each individual moose (Osko et al. 2004). We used ArcGIS ® 9.3.1 (Environmental Systems Research Institute, Inc., Redlands, CA, USA) and the Home Range Tools Extension (HRT; Rodgers et al. 2007) to estimate the 99% fixed kernel (Kernohan et al. 2001) summer and winter home ranges for each moose using a reference smoothing factor ( href, Worton 1995) of 0.7x. This reference smoothing factor is appropriate for short interval GPS data with many locations per moose (Hemson et al. 2005, Robinson 2007). We used the logit link function (Hosmer and Lemeshow 2000) in a generalized linear mixed effects models (GLMM) framework (Gillies et al. 2006) to estimate the coefficients for the exponential approximation to the logistic discriminant function, which yields a relative probability of selection (Johnson et al. 2006). To account for each individual moose as the sample unit, we included a random intercept (β0 + γ0j) for each animal in GLMM’s (Hebblewhite and Merrill 2008, Bolker et al. 2009). Generalized linear mixed effects models help to account for unbalanced sample sizes between individuals and non independence of GPS locations by partitioning the total variation into a subject specific random intercept (Rabe Hesketh et al. 2004). The linear form of the generalized logistic mixed effects model that we used to predict the relative probability w*(x) of use as a function of covariates x1… n was:
w*(x) ij = β 0 + γ 0j + β 1 x1ij + … + β n xnij + є ij (1) where i is the used location 1…n, j is the individual moose 1…n, γ 0j is the random intercept for moose j, β1…n are the selection coefficients estimated from fixed effects logistic regression, and є ij is the unexplained residual variation (Manly et al. 2002). Positive coefficients (>0) indicated that moose selected for a habitat covariate and negative coefficients (<0) indicate avoidance, relative to availability. At the home range scale (i.e., third order) we employed a conditional (i.e., matched case control) logistic regression modeling design (Compton et al. 2002) to
15 estimate the relative probability of moose resource selection from one time step to the next. We generated available points from the bearing and empirical step length and turning angle distribution of all used moose movement pathways (four hour sampling interval of used locations; Compton et al. 2002). Thus, each used moose location was compared to a specific control point rather than the overall distribution of available points using conditional likelihood (Whittington et al. 2005). Inferences about the intercept (β0) cannot be made in conditional logistic regression (Hosmer and Lemeshow 2000), which can make implementation of mixed effects conditional logistic regression difficult (Duchesne et al. 2010). Instead of using a mixed effects approach at the home range scale, we accounted for unbalanced sample sizes between individual moose by weighting each animal using the inverse probability that an individual moose location was included in the sample (Alldredge et al. 1998, Ferrier et al. 2002). We employed a manual stepwise model building process at both scales described by Hosmer and Lemeshow (2000) to create the most parsimonious models for moose resource selection. We considered candidate covariates (see descriptions below; Table 2 1) that were previously reported to influence moose and caribou resource selection. To test our hypotheses about human influences on spatial separation between moose and caribou, we constrained models to contain key covariates including human disturbance (densities of roads and linear features, recent and old cutblocks), burns (recent and old), shrub landcover because of its importance for moose forage (Renecker and Schwartz 1998), and elevation. All covariates were screened for collinearity using the Pearson’s correlation coefficient threshold of | r | > 0.6 for covariate removal (Hosmer and Lemeshow 2000). We retained the covariate with the lower log likelihood, highest coefficient of determination (pseudo R 2) and lowest P values (Boyce et al. 2002) except in the case of human disturbance covariates, which we always retained. We first conducted univariate logistic regression analysis, using a P < 0.25 on a Wald chi2 statistic as a cut off for the inclusion in model building. To test whether coefficients were nonlinear we explored covariates using semi parametric Generalized Additive Models (GAMs; Hastie and Tibshirani 1990), and either transformed coefficients or used quadratics to describe non linear patterns in GLMM models (Hosmer and Lemeshow 2000). Covariates were also screened for relevant interaction terms. Retained covariates
16 entered the multivariate logistic regression modeling process to build a small subset of biologically sensible candidate models (Hosmer and Lemeshow 2000). We selected the top model using Akaike’s information criterion ( AIC; Manly et al. 2002). Statistical analyses were carried out in STATA 11.0 (StataCorp 2007). To assess the predictive capabilities of moose RSF models, we conducted k fold cross validation, withholding one fifth of the data from each training model set and using the withheld data as evaluation data sets (Boyce et al. 2002). To assess the fit of these predicted values we used a Spearman rank test statistic to compare the frequency of the predicted values of the test data set within one of 10 bins to the bin’s respective RSF score rank (Boyce et al. 2002). The frequency of moose locations should increase in higher habitat ranks if a RSF model has high predictive power, indicated by a Spearman’s rank correlation (rho) of > 0.64 (Fielding and Bell 1997, Boyce et al. 2002).
MOOSE CARIBOU RESOURCE PARTITIONING To evaluate resource partitioning of moose and caribou, we paired each moose with one caribou in the respective caribou herd home range (99% fixed kernel) in or near which the moose was captured (Figure 2 1) to control for equal availability of resources to both species. Because of moose collar failure in the Red Rock Prairie Creek caribou herd (Appendix A, Table A 1), we compared one caribou from this herd to the closest moose available . We used logistic regression to model differences in the resource use of moose and caribou, where caribou used locations were coded as 1 and moose used locations as 0. This analysis determined which covariates predicted spatial separation between moose and caribou resource use, measured by the estimated β coefficients from logistic regression. We predicted that caribou and moose would spatially separate by caribou selecting higher elevations and lower human disturbance than moose (P2a in Table 2 1). To account for differences in sample sizes and autocorrelation of GPS locations of individual animals we used generalized logistic mixed effects models with a random intercept for each pairing of a caribou and moose (Gillies et al. 2006). Covariate selection and model building was conducted in the same manner as outlined above. In cases of complete separation of categorical covariates (e.g., complete avoidance of cutblocks by caribou in the foothills), we substituted one value of the most abundant category (i.e., closed conifer) to avoid zero cells and associated large standard errors. 17
One measure of resource overlap is the degree to which logistic models could successfully predict whether the location was used by caribou or moose. The area under the relative operating curve (ROC) index measures the discrimination capacity of logistic regression models and ranges from 0.5, indicating no discrimination ability of the model, to 1 for models with perfect discrimination (Pearce and Ferrier 2000). Generally, ROC scores of > 0.7 are considered acceptable, and ROC > 0.8 indicate excellent discrimination (Hosmer and Lemeshow 2000). We also calculated the percentage of correct predictions for each species (sensitivity and specificity). For our study design (caribou = 1, moose = 0), sensitivity measures positive classification for caribou locations, while specificity measures negative classification success for moose (Pearce and Ferrier 2000). Under the spatial separation hypothesis, we expected the most ‘confusion’ or misclassification between moose and caribou in the region with higher human disturbance levels, especially during summer when forage overlap might occur (P3a and 3b in Table 2 1). We also estimated overlap between moose and caribou resource use by calculating the MacArthur and Levins index (Mmc or Mcm; 1967), which measures the extent of which one species overlaps the niche of the other species, and also Pianka’s index (O; 1973), which is a symmetrical measure of niche overlap. We measured resource (niche) overlap using spatial distributions for each species predicted from the top mixed effects caribou moose logistic regression model. First, we created spatial predictive maps of resource use from the top model using ArcGIS 9.3.1 raster calculator and classified each map into ten equal sized ordinal bins (0.1 1). These maps provided a relative index of caribou and moose resource use, where values closer to 0.1 indicate the highest relative probability of use by moose and conversely, values closer to 1 indicate the highest relative probability of use by caribou. We then calculated niche overlap in these 10 ranked resource use categories for moose and caribou telemetry locations using both overlap indices. Both indices range from 0 (no resource use in common) to 1 (complete overlap). Overlap indices do not provide statistical inferences about ecological relationships between species, but they can measure ecological similarity (Jenkins and Wright 1988). While interpretation of resource (niche) overlap indices as measures of competition between species has been subject to debate (Abrams 1980, Gotelli and
18
Graves 1996), we used niche overlap indices rather as a measure for the spatial distribution of moose and caribou. Finally, we tested whether predator caused caribou mortalities occurred with higher frequency in areas of high overlap between moose and caribou as expected under the spatial separation hypothesis (McLoughlin et al. 2005). We defined areas of high overlap as areas that both species were predicted to have intermediate relative probability of use (i.e., use of bins 0.3 0.8). We overlaid spatial location data of predator caused caribou mortalities (wolf = 32, grizzly/black bear = 5, unknown predator = 8), from the past decade (since 1999) on our predicted probability of caribou/moose resource use maps and assessed in which relative probability of use bins mortalities fell. Long term mortality data were compiled by Alberta Fish and Wildlife Division and Parks Canada based on radio collared (VHF and GPS) caribou. Radio collared caribou were located at least every 3 months from a fixed wing aircraft. Sensors of radio collars indicated mortalities by altering their beacon frequency when the caribou was immobile for more than 8 hours. Animal mortalities were investigated on the ground as soon as possible to determine cause of death. The most likely cause of death was recorded based on signs of struggle, carcass use, scat, hair, tracks, subcutaneous hemorrhaging at wound sites and blood signs (Kunkel and Pletscher 1999). When signs of multiple predator species were present the caribou mortality cause was classified as unknown predator caused. Data indicative of physical condition (blood samples, weights, measurements, status of bone marrow, physical appearance) of caribou were also collected to further exclude natural death due to male nutrition or old age. We tested the hypothesis that more predator caused caribou mortalities occurred in overlap habitat than in moose (bins 0.1 and 0.2) or caribou (bins 0.9 and 1) habitat using ANOVA (P4a in Table 2 1).
HABITAT COVARIATES We characterized moose and caribou habitat using a variety of spatial geographic information system (GIS) covariates of topography, vegetation, and human disturbance (Table 2 2). All habitat covariates were developed at 30m resolution raster layers, unless otherwise specified. We used a digital elevation model (DEM) to estimate elevation (m) and slope (degrees). Vegetation was characterized by 12 categorical landcover layers, which were calculated on the basis of landcover type (10 classes; e.g., Upland Trees, 19
Shrubs, Snow/Ice), forest canopy closure and tree species composition (Table 2 2). These 3 layers were produced with Landsat 5 and 7 TM sensors (McDermid et al. 2005). Closed conifer was used as the reference category in habitat models and thus always subsumed into the intercept. Alpine landcover was delineated by estimating tree line, which was modeled through a curvilinear relationship between latitude and tree line along the north south study area gradient following Paulsen and Körner (2001). Two landcover types, burns and cutblocks, were produced based on combined data from BC Ministry of Forests and Range Data Models (British Columbia Ministry of Sustainable Resource Management 2010), BC Forest Vegetation Composite Polygons and Rank 1 data (British Columbia Ministry of Sustainable Resource Management 2009), data from the Foothills Research Institute Grizzly Bear Program (FRIGBP) and the AB Sustainable Resource Development. We determined whether burns and cutblocks were 0 20 years of age (young) or 21 40 years (old) using a stand age layer from these forest inventory datasets. Landsat data of forest canopy closure was also used to estimate the distance (m) to the nearest open canopy, were values of 0 were locations in open canopy. Normalized Differential Vegetation Index (NDVI) can be used as an index of vegetation productivity (greenness) to characterize green forage biomass (Pettorelli et al. 2005, Hebblewhite et al. 2008).We estimated NDVI during the growing season using 16 day composites derived from NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS, MOD13Q1; Huete et al. 2002). Hebblewhite et al. (2008) estimated the mean growing season from 3 May (Julian day 123) to 9 October (Julian day 282) near Banff National Park. Because the growing season decreases with increasing latitude, we estimated average NDVI based on these growing season dates, but used the closest day after 3 May and before 1 October (Julian day 129 and 273) at a 250m resolution for which MODIS data were available for our calculations. Percent snow cover was estimated from 8 day composites of maximum snow extent maps at a 250m resolution produced by MODIS satellites (MOD10A2; Hall et al. 2000). The number of days snow occupied a cell was divided by the number of days in the seasonal period to derive spatial models of percent snow cover. Season start and end dates were the same as for logistic regression models. For landscape scale moose models we used NDVI and snow data from winters of 2008/2009 and summer 2009. For home range scale moose models and
20 moose/caribou comparisons we calculated NDVI and snow cover values at the time of the animal’s use and available location separately for each location by using the layer of the corresponding year. Spatial predictions were made using NDVI and percent snow layers from 2009. Besides cutblocks, human disturbance was further estimated from a variety of vector geodatabases of roads, seismic exploration lines, railways, pipelines and human trails (Alberta SRD –Resource Information Management Branch and digitized 2004 SPOT imagery and 1:250 000 NTS maps). We calculated density layers for roads and linear features (km/km 2) in the Spatial Analyst extension for ArcGIS ® Desktop 9.3.1 software.
RESULTS
REGIONAL HABITAT AVAILABILITY We observed some differences in habitat availability (as defined within the seasonal range of radio collared moose) between the foothills and mountains (Appendix A, Table A 2). Mean foothills elevations ranged from 1,248m (SE = 2.5) during winter to 1,263m (SE = 2.8) during summer, while mean elevations in the mountains ranged from 1,761m (SE = 4.2) in winter to 1,776m (SE = 4.2) in summer. The mountains were 10% steeper than the foothills. While we did not assess significance of differences of composition of landcover types, we observed that there was less closed conifer (19.6% less in summer and 16.8% less in winter) and a higher proportion of alpine (27.8% more in summer and 29% more in winter) habitat available in the mountains compared to the foothills. Shrub landcover was approximately twice as common in the foothills as in the mountains (6.0% versus 3.5% in summer and 6.1% versus 3.5% in winter), as was muskeg (6.1% more in summer and in winter). Burns of both age classes comprised only small proportions of available landcover types from a low of 0.27% of old burns in the foothills during summer to a high of 0.39% in the foothills in winter. Young burns were rarest in the foothills during summer (0.36%) and most common (1.59%) in the mountains during summer. The foothills were more heavily impacted by human disturbance with 6.2% and 6.4% higher availability of young cutblocks during summer and winter respectively.
21
Older cutblocks were rare in both regions (1.0% and 0.9% in the foothill during summer and winter and 0.4% and 0.5% in the mountains during summer and winter). Furthermore, the foothills also had higher human disturbance in the form of roads with mean densities of 0.35 km/km 2 (SE = 0.006) in both winter and summer in the foothills, compared to the mountains with 0.08 km/km 2 (SE = 0.003) in summer and 0.11 km/km 2 (SE = 0.004) in winter. The foothills linear feature density was also higher year round (1.62 km/km 2, SE = 0.016 in summer and 1.69 km/km 2, SE = 0.016 in winter) compared to the mountains (0.44 km/km 2, SE = 0.009 in summer and 0.50 km/km 2, SE = 0.010, in winter) in winter and summer (Appendix A, Table A 2).
LANDSCAPE SCALE MOOSE RESOURCE SELECTION We categorized 8 moose (5 f and 3 m) as foothill with high human landscape disturbance and 9 moose (5 f and 4 m) as mountain with low human landscape disturbance. Average home range size of moose in the mountains was 289.5 km 2 (SE = 187.74) in summer and 169.7 km 2 (SE = 96.3) in winter, while the average home range size of moose in the foothills was slightly smaller at 237.9km 2 (SE = 101.1) in summer and 110.7km 2 (SE = 36.91) in winter. Overall, standard errors of home range sizes between moose in the foothills and mountains were large and overlapped each other. On average, the maximum seasonal distances a moose traveled, and thus average buffer size around moose home range kernel, were 19.5km (SE = 3.19) in the mountains and 20.8km (SE = 6.41) in the foothills during winter, and 30.0 km (SE = 7.01) in the mountains and 17.4km (SE = 4.04) in the foothills during summer. Landscape scale resource selection models cross validated very well in k folds, confirming their predictive capacity with average Spearman’s rho of 0.89 ( P < 0.001) and 0.99 ( P < 0.001) for the foothills summer and winter models and 0.97 ( P < 0.001) and 0.99 ( P < 0.001) for the mountain summer and winter models respectively. During winter at the landscape scale, mountain and foothill moose selected for similar intermediate (quadratic) elevations of about 1,269m and 1,263m. Conversely, in the summer, foothills moose selected elevations of 1,405m, while mountain moose selected elevations of about 1,874m (Tables 2 3 – 2 6). Young burns (< 20 years) were selected year round by moose in the mountains at the landscape scale (β su = 2.18, β wi =
1.27). Moose avoided rare older burns ( > 20 years) in the mountains (β wi = 1.50). 22
Overall, most available burns occurred in the mountains, which made evaluation of selection for burns in the foothills difficult, but we observed strong avoidance of older burns during summer (foothill: β su = 2.12). Moose resource selection for cutblocks was almost opposite between regions compared to burns (Figure 2 2). Foothill moose avoided young cutblocks during both seasons at the landscape scale (β su = 0.43, β wi = 0.01), as did mountain moose (β su = 1.74, β wi = 4.35). Old cutblocks (> 20 years) were avoided by moose in the foothills, however, old cutblocks were so rare in the mountains that evaluating selection by mountain moose was unfeasible. In comparison to burns and cutblocks, moose consistently selected for shrubs in both regions and season (Figure 2 2; Tables 2 3 – 2 6). Moose in both regions strongly avoided high road densities at landscape scale. In the foothills moose showed similar, but statistically weak, avoidance of road densities during both seasons (β su = 0.51, β wi = 0.47), while moose in the mountains exhibited a statistically much stronger avoidance (β su = 3.10, β wi = 1.65; Figure 2 3). Moose selected for intermediate (quadratic) densities of linear features in both regions at the landscape scale, but higher densities in the foothills (approximately 2.2 km/km 2 in both seasons) compared to the mountains (0.90 1.17 km/km 2; Figure 2 4). Finally, moose avoided areas more frequently covered by snow at the landscape scale in both regions during winter and in the mountains during summer (foothills: β wi = 4.18; mountains: β su
= 1.12, β wi = 1.973).
HOME RANGE SCALE MOOSE RESOURCE SELECTION We used 17,144 moose GPS locations in the mountain region and 16,914 GPS locations in the foothill region to develop home range scale habitat selection models. The average number of locations per moose was 2,143 (SE = 117.2) in the mountains and 1,879 (SE = 74.5) in the foothills. At the home range scale, resource selection models had much lower or even insignificant predictive capacity with Spearman rho values of 0.45 ( P = 0.204) and 0.72 (P = 0.021) for the summer and winter foothills models and 0.75 ( P = 0.029) and 0.71 ( P = 0.030) for the summer and winter mountain models respectively. Moose avoided high elevations in both regions and showed stronger avoidance for steep slopes than at the landscape scale (Tables 2 3 – 2 6). In contrast to the landscape scale, within home ranges moose in the mountains avoided young burns during winter, 23 but selected young burns during summer (β su = 0.59, β wi = 0.56; Figure 2 2). Burns were too rare to estimate selection coefficients for moose in the foothills, and young cutblocks were similarly rare in the mountains. However, young cutblocks were selected by foothill moose especially in winter (β su = 0.16 (not significant), β wi = 0.51). Again, moose responded positively to shrubs in both regions and during both seasons (Tables 2 3 – 2 6;
Figure 2 2). While moose in the mountains avoided high road densities in winter (β wi =
1.41), moose in the foothills avoided high road densities especially during summer (β su = 0.48), and neither selected nor avoided road densities during summer in the mountains and during winter in the foothills (i.e., coefficients not significant; Figure 2 3). Resource Selection Functions indicated that moose avoided linear features during summer in the foothills and mountains, and selected linear feature densities during winter in both regions, but selection coefficients were not significant (Tables 2 3 – 2 6). Finally, moose avoided areas more frequently covered by snow during winter in the mountains and positively selected for snow covered areas during winter in the foothills at the home range scale.
MOOSE CARIBOU RESOURCE PARTITIONING We evaluated differences in resource use with 16,907 caribou and 15,779 moose GPS locations during summer and 17,322 caribou and 18,273 moose locations during winter from 17 individuals of each species. The average number of locations per caribou and per moose in summer was 995 (SE = 45.8) and 928 (SE = 23.4), and in winter 1,019 (SE = 57.4) and 1,075 (SE = 40.3) respectively. Resource use by caribou and moose and the degree of spatial separation differed greatly among seasons and regions (Figure 2 5). Caribou used significantly higher elevations than moose (the highest z values; Tables 2 7 – 2 8) during each season and in each region, although spatial separation due to elevation was weakest during summer in the mountains. Caribou and moose partitioned space by their dissimilar use of areas with human disturbance (roads and linear features), but the pattern was opposite in the foothills and mountains. In the foothills, caribou used areas with lower densities of roads and linear features (summer: βRoad = 0.66, β LFeat. = 0.35; winter: βRoad = 0.25, β LFeat. = 0.23) than moose, while in the mountains, caribou used areas with higher densities of roads and linear features compared to moose (summer;
βRoad = 5.03; winter: βRoad = 3.42, β L.feat. = 0.68). Caribou avoided young cutblocks in the 24 foothills (βsu = 0.75, βwi = 1.94) as well as young burns in the mountains (βsu = 4.15, β wi = 4.45). In contrast, caribou selected for young burns in the foothills stronger than moose during summer (β = 5.23). Caribou avoided shrub landcover in the mountains (summer: β = 0.71; winter: β = 0.76), and selected for lower NDVI values compared to moose during summer in both regions. Furthermore, caribou consistently partitioned resources during both seasons and in both regions by their differential use of mixed forests, herbaceous and open conifer landcover classes. While caribou used the first two categories less compared to moose, they occurred more often in open conifer. Caribou in mountains used barren areas less than moose during both seasons and strongly positively responded to muskeg during summer (based on high z value of 14.08; Tables 2 7 – 2 8). As predicted under the spatial separation hypothesis, the overall proportion of caribou and moose locations correctly classified by our models and ROC scores were higher in the mountains than the foothills. Our models were able to predict caribou (sensitivity) locations 69% in summer and 71% in winter in the foothills and 76% in summer and 79% in winter in the mountains (Tables 2 7 – 2 8). Moose locations (specificity) were correctly classified 72% and 68% in summer and winter respectively in the foothills and 76% and 75% in summer and winter in the mountains. For foothills models ROC scores varied between 0.81 in summer and 0.78 in winter. The discrimination ability of mountain models was higher with ROC scores of 0.83 in summer and 0.86 in winter. Similarly, Pianka’s niche overlap index revealed higher overlap between moose and caribou in the foothills than the mountains in winter ( OFH =
0.629, OMT = 0.351). However, in summer Pianka’s overlap index indicated almost no difference between foothills and mountain niche overlap index values ( OFH = 0.422, OMT = 0.410; Figure 2 6). Asymmetric overlap was highest for both species during winter in the foothills when moose and caribou niches overlapped each other with approximately equal extends ( Mcm = 0.624, Mmc = 0.635). Overall, moose overlapped more with caribou than vice versa, supporting our prediction from apparent competition of asymmetric overlap in favor of the more abundant prey species (Figure 2 6). The only time when the moose niche extend was overlapped more by the caribou niche extend was during the winter in the mountains when Mcm was 0.401 and Mmc was 0.356.
25
Finally, as predicted by the spatial separation hypothesis, a one way analysis of variance (ANOVA) indicated significant differences among the mean number of predator caused mortalities falling into the caribou, moose and overlap resource use categories, indicating that the greatest proportion of caribou killed by predators occurred where resource overlap between moose and caribou was highest compared to the categories where resource partitioning by caribou and moose was greatest ( F2,37 = 3.37 , P = 0.045; Figure 2 7).
DISCUSSION
Our results broadly support the hypothesis that human disturbance reduces the spatial separation between moose and caribou. By examining moose resource selection and comparing moose and caribou habitat use we found general support for reduced resource partitioning between the two species in the highly developed foothills region. First, we found that moose habitat selection varied between scales, seasons and intensity of human disturbance. Interestingly, our results demonstrate a more complex pattern of moose resource selection under human landscape disturbance than often assumed under the spatial separation hypothesis. Our research also suggested that moose habitat selection is a function of the availability of resources (Osko et al. 2004, Peek 2007) and can vary depending on the intensity of human disturbance. Second, we found that spatial separation of caribou and moose was driven largely by caribou use of higher elevations and avoidance of human landscape disturbance (i.e., cutblocks, densities of linear features and roads). We also found that spatial separation varied between seasons and the two study regions. The contrasting patterns of moose and caribou resource use generally resulted in spatial separation in the mountains. In comparison, differing landscape scale availabilities resulted in increased niche overlap in the foothills region, likely due to increased human disturbance. These results were consistent with predictions of the spatial separation hypothesis. Previous studies of the spatial separation between moose and caribou have assumed that moose generally respond positively to human disturbance and have rarely considered moose resource selection at multiple spatial scales (Weclaw and Hudson 2004, McLoughlin et al. 2005). In contrast, we employed methods that explicitly
26 examined these assumptions and tested our predictions at two spatial scales. Our results suggested that moose indeed make important trade offs between forage and risk associated with human disturbance in a hierarchical fashion (Boyce et al. 2003, Dussault et al. 2005). Interestingly, this hierarchal response could further erode spatial separation (James et al. 2004) in high human disturbance landscapes if moose avoid predation risk similarly to caribou. Our observation that moose avoided human disturbance (cutblocks, roads, linear features) at the landscape scale, but selected them at the home range scale is contrary to scale independent predictions of the spatial separation hypothesis (Table 2 1). We interpret our results as suggestive for scale dependent trade offs between risk and forage. For example, in contrast to avoidance of cutblocks at the landscape scale in the foothills, we found that moose selected young burns in the mountains at this coarser scale. While cutblocks of up to 20 years are a surrogate for shrubs (Dyrness 1973), predation risk is usually elevated due to roads associated with each cutblock (Rempel et al. 1997, Frair et al. 2008). Contrarily, burns also provide abundant forage for moose (Sachro et al. 2005), but without roads. Thus, moose responded to burns (Robinson et al. 2010) without having to trade off against increased risks as suggested in the foothills for cutblocks. Overall, hierarchical avoidance of human disturbance by moose in the foothills could lead to higher spatial overlap between moose and caribou. Moose RSF models also suggested that moose make trade offs between forage and differently during different seasons. For example, foothills moose showed no significant avoidance of roads during winter and selected strongly for surrounding cutblocks (correlation coefficient between road density and cutblocks = 0.52), suggesting that during this forage sparse season, moose selected forage and thereby exposed themselves to higher risk (Godvik et al. 2009). In contrast, during summer moose avoided roads perhaps because forage is much more abundant in other less risky habitats. Our results support this conclusion since moose showed strong selection for herbaceous (β = 0.78), barren (β = 0.66), muskeg (β = 0.31) and mixed forests (β = 0.31) opposed to insignificant selection for young cutblocks in summer (Table 2 3). In the foothills, avoidance of human disturbance by moose and potential for overlapping forage selection of moose and caribou in summer (Servheen and Lyon 1989, Apps et al. 2001, Boer 2007) could put moose and caribou closer together. Overall, we found that moose may make
27 selection decisions in a hierarchical fashion with a stronger sensitivity to increased human disturbance and predation risk at coarser scales and select habitat to meet forage requirements at finer scales (Rettie and Messier 2000). Lastly, selection for human disturbance seemed to be conditional on the relative availability of human disturbance (Mysterud and Ims 1998). Resource selection studies are influenced by how available resources are sampled or varying behavioral mechanisms of selection (Aebischer et al. 1993, Garshelis 2000). Interestingly, our results differed from the moose specific predictions derived from the spatial separation hypothesis (Table 2 1). Avoidance of human disturbed habitats at the landscape scale does not mean that moose never occurred in these areas, but rather that disturbed habitats were used less than expected based on their availability. Although we tried to control for potential bias arising from defining availability at the landscape scale by estimating individual specific buffers for each moose, it could be possible that moose home ranges are a result of daily decisions rather than exploration of large landscapes prior to home range establishment (Courtois et al. 2002). Furthermore, establishment of moose home ranges within the landscape could be driven by behavioral mechanisms, such as differing dispersal strategies between sexes or site fidelity (Cederlund et al. 1987, Labonte et al. 1998, Peek 2007) rather than habitat selection at large spatial scales per se. In general, animals should select habitats that permit avoidance of the most dominant limiting factors at larger spatial scales (Rettie and Messier 2000). Our study was limited by the availability of data characterizing other limiting factors potentially influencing fitness of individual moose in a population, such as snow depth, canopy closure or habitat configuration matrices (Dussault et al. 2005, Peek 2007, Renecker and Schwartz 2007). Also, we relied on using the density of roads and other linear features as surrogate for predation risk (Eason 1989, Rempel et al. 1997, Frair et al. 2008), but spatial distribution data of predators may have been more variable than road and line density. Overall, the divergent moose habitat selection pattern in the foothills and mountains and equivocal results with respect to predictions based on the spatial separation hypothesis lead us to focus on the consequences of moose resource selection for the spatial separation of moose and caribou in each region.
28
Results from comparing resource use by moose and caribou were consistent with previous studies that demonstrated that caribou and moose spatially separate themselves by resource partitioning (Boonstra and Sinclair 1984, Seip 1992, Stotyn 2008). In particular, we found significant differences in caribou use of elevation and moose forage rich habitats, such as cutblocks, shrubs and burns, which was congruent with the predictions of the spatial separation hypothesis. Our results were consisten t with the general consensus that caribou strongly avoid human disturbance especially in the foothills, presumably to decrease predation risk. For example, several studies have suggested that caribou avoid roads and seismic lines (Dyer et al. 2001, Nellemann et al. 2001, Cameron et al. 2005, Neufeld 2006). Smith et al. (2000) reported long term avoidance of cutblocks by caribou where caribou locations were on average 11.1 km from cutblocks in west central Alberta. Cutblocks in Ontario have significantly displaced caribou from harvested areas (Darby and Duquette 1986). Our results indicated that caribou avoidance of human disturbance was stronger than moose avoidance. Interestingly, caribou used young burns more than moose during summer in the foothills. In particular, these fires occurred within two years prior to caribou use. We assume that caribou use could be attributed to site fidelity to their annual and seasonal home ranges that were established before disturbance (Rettie and Messier 2001, Dalerum et al. 2007). However, caribou use of burns could also indicate reduced spatial separation of moose and caribou during summer (Robinson et al. 2010) if caribou use burns to access abundant herbaceous green up during summer (Sachro et al. 2005) similarly to moose. In the mountains, moose avoided roads and linear features more than caribou. However, both species were substantially separated from human disturbance due to the low road density in this region. For example, moose used areas with average densities of linear features and roads of 0.64 km/km 2 and 0.02 km/km 2 respectively during winter, while caribou used areas with average line and road densities of 0.74 km/km 2 and 0.05 km/km 2, both of which were still very low densities in comparison to the foothills (Appendix A, Table A 3). Thus, landscape gradients in human development between both regions appeared to drive differences in moose and caribou resource use similar to the effects of seasonal variation in habitat composition. While individual habitat components were usually available to animals in both regions the relative use of habitat
29 components was dissimilar, suggesting that landscape configuration could define resource partitioning patterns (Osko et al. 2004). Consistent with the hypothesized effect of human activities on resource partitioning, we found increased general resource overlap of moose and caribou in the foothills. Both, Pianka’s niche overlap index and ROC scores were lowest in this region, indicating the lowest resource partitioning (Krebs 1999, Pearce and Ferrier 2000) . Caribou likely used alternate habitats that were increasingly occupied by moose, because their otherwise selected habitat (unfragmented old growth forests) became less available due to human disturbance (Szkorupa and Schmiegelow 2003, Wittmer et al. 2005a). Furthermore, we observed contrasting seasonal niche overlap, with highest overlap during winter in the foothills, and only slightly increased overlap in the foothills compared to the mountain region during summer. We suggest that this results from the partially migratory behavior of caribou (only some individuals migrate), which was difficult to incorporate into our study design. Only one herd in our study area is sedentary in the foothills (Little Smoky herd), while the other 4 herds are partially migratory (Alberta Sustainable Resource Development and Alberta Conservation Association 2010). Some caribou likely migrated to higher elevations during summer, a mechanism of spatial separation (Seip et al. 1992) that caused the unexpected low niche overlap in the foothills during summer. Therefore, further investigation of exclusively sedentary caribou and moose would be necessary to determine niche overlap during summer in the foothills. Resource overlap (niche) overlap in the mountains was greater during summer than winter, which is consistent with observations from other studies. For example, Stotyn (2008) suggested that spatial separation of caribou and moose in the mountains of British Columbia was highest in late winter and less pronounced during summer due to potential overlapping forage and elevation preferences of the two species during summer (Boertje et al. 1988, Seip 1992, Robinson et al. 2010). In general, moose overlapped caribou to a greater extent, except during winter in the mountains. Moose are a generalist species with a broad niche (Peek 2007), but are hindered in their movements by snow depths 60 70cm (Telfer 1970, Peek 2007, Renecker and Schwartz 2007). In contrast, the niche specialization of caribou allows them to live at higher elevations in deeper snow where they access to arboreal lichen in old Engelmann spruce and subalpine fir stands
30
(Apps et al. 2001) that are spatially separated from moose (Klein 1982, Brown and Mallory 2007). Thus, caribou are better adapted to harsh winter conditions in the mountain regions than moose, while moose are able to exploit a broader range of resources than caribou during summer when they are not constrained by snow. Spatial overlap of sympatric prey species can result in concurrent occurrence of exploitative (shared resources consumption) and apparent competition (shared predators; Holt and Lawton 1994). Although we did not specifically assess diet composition and foraging by moose and caribou, habitat use of caribou seemed to indicate that exploitative competition between the two species is unlikely, at least during winter. In a stable isotope diet study conducted by Ben David et al. (2001) in Alaska, moose and caribou stable isotope ratios were significantly different from each other in late summer autumn and winter. Similarly, Mysterud (2000) found that diet of moose and reindeer ( Rangifer tarandus ) only overlapped by 0.6% in winter. While we cannot exclude forage overlap in summer (Boer 2007), we assume that forage overlap did not cause the increase in resource overlap in the foothills during winter, but rather the limited availability of undisturbed caribou habitat as previously suggested. In fact, destruction of unfragmented habitat in which caribou can space out in low densities has been a major concern in caribou conservation (Hervieux et al. 1996, Smith et al. 2000, Neufeld 2006). For example, human landscape disturbance in winter ranges has been hypothesized to lead to changes in migratory behavior of caribou herds observed in recent decades (e.g., Smith et al. 2000). The 2010 status report for woodland caribou in Alberta details that between 23% and 38% of the winter or permanent ranges of the 5 caribou herds we studied were altered by forestry based on satellite imagery. Consequently, caribou populations may choose suboptimal long term winter range, which reduces the suitability of caribou habitat and negatively affects population viability (Alberta Sustainable Resource Development and Alberta Conservation Association 2010). Caribou survival and population growth were significantly reduced in regions with increased disturbance (Smith 2004) and increased spatial overlap between caribou, primary prey and wolves (James and Stuart Smith 2000, McLoughlin et al. 2005). Similarly, caribou killed by predators had lower proportions of old forests in their home range compared to surviving caribou in a study by Wittmer et al. (2007). Consistent with these population level
31 conclusions, we reported higher risk of mortality for caribou in resource use categories where moose and caribou use were likely to overlap (McLoughlin et al. 2005; Figure 2 7). Consequently, decreased availability of undisturbed caribou habitat might increase the mortality risk for caribou due to reduced spatial separation between moose and caribou in these disturbed landscapes. Although we feel confident in our conclusion that spatial separation between moose and caribou is decreased in landscapes with high human disturbance, some characteristics of our study design held the potential to affect our results and require further research. Resource selection studies should be interpreted cautiously because of the common assumption that resource selection is directly linked to fitness (Aebischer et al. 1993, Boyce and McDonald 1999, Garshelis 2000). While this assumption of a direct relationship between animal density and resource quality has been tested by other researchers for different species (Beard et al. 1999, Corsi et al. 1999), it cannot be generalized (Van Horne 1983). Obtaining data on population demography is particularly important when animals could experience ecological traps that decrease survival and are especially common in human modified environments where an evolutionary lag occurs between habitat quality and the species’ adaptation to novel mortality risks (Delibes et al. 2001, Donovan and Thompson 2001, Schlaepfer et al. 2002). Furthermore, while high levels of overlap in resource use are often used to infer competition (Jenkins and Wright 1988), it is essentially the ratio of the density of consumer individuals (i.e., moose and caribou) relative to the resource base (i.e., habitat) that determines the strength of competitive interactions (Abrams 1980). Caribou populations experience negative growth rates in landscapes altered by humans (James et al. 2004). For example, finite annual rates of population increase (lambda; λ) have been below 1 (indicating population declines) due to high calf and moderate to high cow mortalities for herds considered in this study (Alberta Sustainable Resource Development and Alberta Conservation Association 2010). Population growth rates were also significantly lower in caribou ranges with more human disturbance. While low caribou population viability confirms our main conclusion that the spatial separation between caribou and moose is deterred by human disturbance, we were unable to link our resource selection function and resource overlap measures to moose
32 demography. Habitat selection is generally assumed to be related to fitness (Boyce and McDonald 1999, Pearce and Ferrier 2001, Railsback et al. 2003, Nielsen et al. 2005, McLoughlin et al. 2006), but the occurrence of a species may not always be a good predictor of habitat quality (Van Horne 1983). Additionally, moose density itself could be a driver for moose habitat selection. For example, high densities could lead to increased use of marginal habitats with increased mortality risk (Sinclair and Arcese 1995). Moose density also affects the numerical response of predators (Messier 1994) and therefore predation risk for caribou. High costs for traditional aerial survey methods, the most practical tool for moose population estimation, imprecise survey results, and difficult survey conditions in mountainous regions (Timmermann and Buss 2007) constrained our efforts to estimate moose density across caribou ranges. However, we tested an alternative aerial survey technique to traditionally used methods (Chapter 3) which shows great promise to improve moose population estimation that will also allow us to link moose resource selection to population fitness parameters (Garshelis 2000). Furthermore, we observed only one natural death of a total sample of 33 radio collared moose that were monitored for at least one year each and all female captured moose were pregnant at the time of capture determined by blood serum progesterone levels when blood samples were available (n = 17; Appendix A, Table A 1; Haigh et al. 1993), suggesting high moose population viability in stark contrast to low caribou survival and population declines. Thus, despite the untested assumption about moose density relating to highly selected moose habitats, our results are indicative for higher moose density in caribou ranges as a result of increased human disturbance. Our improved distance sampling methods (Chapter 3) can be used in the future to test this prediction. Overall, our results supported the hypothesis that with the encroachment of human disturbance on the landscape caribou refugia from moose, and hence predators like wolves, are compromised and their spatial separation strategy may no longer be effective. This could potentially result in destabilizing the relationship between predators and prey as predicted by the spatial separation hypothesis. Recognition that moose selection for human features varies at multiple spatial and temporal scales is an important step in establishing insight into the complex predator prey relationships determining caribou persistence. We also found evidence that increased overlap between moose and
33 caribou elevates predation risk as most predator killed caribou occurred in areas with high probability of overlap between resource use by moose and caribou. If moose and caribou selected for similar habitats in human disturbed landscapes, caribou are more likely to be encountered by predators (Lessard et al. 2005). Predators, such as wolves, that numerically respond to moose can lead to extirpation of a secondary prey species, like caribou, co existing in the same landscape (Messier 1994, Sinclair et al. 1998, Messier and Joly 2000).
MANAGEMENT RECOMMENDATIONS Woodland caribou populations require large undisturbed ranges of old coniferous forest that reduce predation by allowing caribou to avoid areas selected by primary prey species, and thus predation risk (Bergerud et al. 1984, Bergerud and Page 1987, Seip 1992, Brown et al. 2003, James et al. 2004). The most important management recommendation from our research is to maximize the integrity of caribou refugia (old coniferous forest) and the connectivity between these refuges in already compromised caribou ranges and to avoid any further forest harvest activity in caribou ranges if the goal is to recover threatened woodland caribou populations. Forest harvesting and other industrial activities should be concentrated in spatially restricted areas rather than spread out over extensive regions avoiding caribou habitat fragmentation and interspersion of moose and caribou habitat. Ideally, because previous researchers reported that overlap between moose, caribou and their predators (especially wolves) can occur at relatively large spatial scales (Rettie and Messier 2000, Alberta Woodland Caribou Recovery Team 2005, McLoughlin et al. 2005, Vistnes and Nellemann 2008), a conservative approach to future resource extraction should also be applied within some distance buffer surrounding caribou ranges to maintain separation between caribou, moose and their predators. Based on our results, that showed that moose and caribou resource overlap is strongest in the region with high human disturbance, we suggest a potential minimum buffer distance could be the average summer moose home range size (i.e., 8.7 km based on the radius of the average summer 99% kernel home range of moose living in the foothills, i.e., 237.9km 2), although moose home ranges can vary greatly (Hundertmark 2007). Furthermore, at a finer spatial scale, our results suggested that moose select for browse rich habitats. In particular, the amount of shrub cover and other deciduous vegetation in 34 young cutblocks could be reduced through high density planting of coniferous trees immediately following forest harvest and control of regenerating deciduous browse vegetation to make these areas less attractive for moose. Population responses of moose, but also other potential primary prey species, such as deer and elk, should be closely monitored in relation to landscape changes and cause and effect relationships between human landscape disturbance and ungulate population responses identified. In our study, caribou used areas with lower densities of roads and other linear features more compared to moose in the region with high human disturbance, suggesting that limiting the amount of linear features that potentially increase predation efficiency by predators (James et al. 2004) in caribou ranges is important for caribou recovery. Predation risk for caribou can also be reduced indirectly through management of the primary prey species upon which predators depend. Thus far, increased moose harvest and wolf control has led to short term improvement of caribou population viability for herds considered in this study (Alberta Sustainable Resource Development and Alberta Conservation Association 2010), but failure to address the ultimate habitat based causes of caribou declines will likely result in continuous long term caribou population decrease. Sufficient amounts of unfragmented woodland caribou habitat (older coniferous forests) over long term periods will be required to secure caribou populations in west central Alberta and east central British Columbia.
LITERATURE CITED
Abrams, P. 1980. Some comments on measuring niche overlap. Ecology (Washington, D.C.) 61:44 49. Aebischer, N. J., P. A. Robertson, and R. E. Kenward. 1993. Compositional analysis of habitat use from animal radio tracking data. Ecology 74:1313 1325. Alberta Sustainable Resource Development and Alberta Conservation Association. 2010. Status of the Woodland Caribou (Rangifer tarandus caribou) in Alberta: Update 2010. Alberta Sustainable Resource Development. Wildlife Status Report No. 30 (Update 2010), Edmonton, AB. Alberta Woodland Caribou Recovery Team. 2005. Alberta woodland caribou recovery plan 2004/05 2013/14. Edmonton, AB.
35
Alldredge, J. R., D. L. Thomas, and L. L. McDonald. 1998. Survey and comparison of methods for study of resource selection. Journal of Agricultural Biological and Environmental Statistics 3:237 253. Apps, C. D., B. N. McLellan, T. A. Kinley, and J. P. Flaa. 2001. Scale dependent habitat selection by mountain caribou, Columbia Mountains, British Columbia. Journal of Wildlife Management 65:65 77. Barrett, M. W., J. W. Nolan, and L. D. Roy. 1982. Evaluation of a hand held net gun to capture large mammals Wildlife Society Bulletin 10:108 114. Beard, K. H., N. Hengartner, and D. K. Skelly. 1999. Effectiveness of predicting breeding bird distributions using probabilistic models. Conservation Biology 13:1108 1116. Berger, J. 2007. Fear, human shields and the redistribution of prey and predators in protected areas. Biology Letters 3:620 623. Bergerud, A. T. 1974. Decline of caribou in North America following settlement. Journal of Wildlife Management 38:757 770. Bergerud, A. T., H. E. Butler, and D. R. Miller. 1984. Antipredator tactics of calving caribou dispersion in mountains. Canadian Journal of Zoology 62:1566 1575. Bergerud, A. T., and R. E. Page. 1987. Deplacement and dispersion of parturient caribou at calving as antipredator tactics. Canadian Journal of Zoology 65:1597 1606. Boer, A. H. 2007. Interspecific Relationships. Pages 337 349 in A. W. Franzmann, and C. C. Schwartz, editors. Ecology and Management of the North Amerian Moose. The University Press of Colorado, Boulder, Colorado, USA. Boertje, R. D., W. C. Gasaway, D. V. Grangaard, and D. G. Kelleyhouse. 1988. Predation on moose and caribou by radio collared grizzly bears in east central Alaska. Canadian Journal of Zoology 66:2492 2499. Bolker, B. M., M. E. Brooks, C. J. Clark, S. W. Geange, J. R. Poulsen, M. H. H. Stevens, and J. S. White. 2009. Generalized linear mixed models: a practical guide for ecology and evolution. Trends in Ecology & Evolution. Boonstra, R., and A. R. E. Sinclair. 1984. Distribution and habitat use of woodland caribou, rangifer tarandus caribou, and moose, alces alces andersoni, in the spatzi plateau wilderness area, British Columbia. Canadian Field Naturalist 98:12 21.
36
Boyce, M. S., J. S. Mao, E. H. Merrill, D. Fortin, M. G. Turner, J. Fryxell, and P. Turchin. 2003. Scale and heterogeneity in habitat selection by elk in Yellowstone National Park. Ecoscience 10:421 431. Boyce, M. S., and L. L. McDonald. 1999. Relating populations to habitats using resource selection functions. Trends in Ecology & Evolution 14:268 272. Boyce, M. S., P. R. Vernier, S. E. Nielsen, and F. K. A. Schmiegelow. 2002. Evaluating resource selection functions. Ecological Modelling 157:281 300. British Columbia Ministry of Sustainable Resource Management. 2009. Vegetation Resource Inventry Forest Vegetation Composite Polygons and Rank 1 Layer. Resources Inventory Committee. BCGOV FOR Forest Analysis and Inventory Branch, Victoria, British Columbia, Canada. Accessed 05/20/2010. Available at: http://www.for.gov.bc.ca/hts/vridata/standards/index.html _____. 2010. RESULTS Openings. Resources Inventory Committee. BCGOV FOR FS Division Forest Practices Branch. Accessed: 05/20/2010. Available at: http://www.for.gov.bc.ca/his/datadmin/models/models.htm#models Brown, G., and F. Mallory. 2007. A review of ungulate nutrition and the role of top down and bottom up forces in woodland caribou population dynamics. NCASI Technical Bulletin 934:i xiii, 1 94. Brown, G. S., F. F. Mallory, and W. J. Rettie. 2003. Range size and seasonal movement for female woodland caribou in the boreal forest of Ontario. Rangifer 14:227 233. Cameron, R. D., W. T. Smith, R. G. White, and B. Griffith. 2005. Central Arctic Caribou and petroleum development: Distributional, nutritional, and reproductive implications. Arctic 58:1 9. Carpenter, L. H., and J. I. Innes. 1995. Helicopter netgunning: a successful moose capture technique. Alces 31:181 184. Cederlund, G., F. Sandegren, and K. Larsson. 1987. Summer movements of female moose and dispersal of their offspring. Journal of Wildlife Management 51:342 352. Commission for Environmental Cooperation. 2009. Ecological Regions of North America in Commission for Environmental Cooperation, Montréal, Québec, Canada.
37
Compton, B. W., J. M. Rhymer, and M. McCollough. 2002. Habitat selection by wood turtles ( Clemmys insculpta ): An application of paired logistic regression. Ecology 83:833 843. Corsi, F., E. Dupre, and L. Boitani. 1999. A large scale model of wolf distribution in Italy for conservation planning. Conservation Biology 13:150 159. COSEWIC. 2002. COSEWIC assessment and update status report on the Woodland Caribou, Rangifer tarandus caribou , in Canada. Ottawa, Ontario, Canada. Courtois, R., C. Dussault, F. Potvin, and G. Daigle. 2002. Habitat selection by moose (Alces alces ) in clear cut landscapes Alces 38:177 192. Dalerum, F., S. Boutin, and J. S. Dunford. 2007. Wildfire effects on home range size and fidelity of boreal caribou in Alberta, Canada. Canadian Journal of Zoology 85:26 32. Darby, W. R., and L. S. Duquette. 1986. Woodland caribou in Ontario, Canada. Rangifer:87 93. DeCesare, N., M. Hebblewhite, H. Robinson, and M. Musiani. 2010. Endangered, apparently: the role of apparent competition in endangered species conservation. Animal Conservation 13:353 362. DeCesare, N., J. Whittington, H. Robinson, M. Hebblewhite, M. Bradley, L. Neufeld, and M. Musiani. 2011. The role of translocation in recovery of woodland caribou populations. Conservation Biology 25, in press . Delibes, M., P. Gaona, and P. Ferreras. 2001. Effects of an attractive sink leading into maladaptive habitat selection. American Naturalist 158:277 285. Donovan, T. M., and F. R. Thompson. 2001. Modeling the ecological trap hypothesis: A habitat and demographic analysis for migrant songbirds. Ecological Applications 11:871 882. Duchesne, T., D. Fortin, and N. Courbin. 2010. Mixed conditional logistic regression for habitat selection studies. Journal of Animal Ecology 79:548 555. Dussault, C., J. P. Ouellet, R. Courtois, J. Huot, L. Breton, and H. Jolicoeur. 2005. Linking moose habitat selection to limiting factors. Ecography 28:619 628.
38
Dussault, C., J. P. Ouellet, R. Courtois, J. Huot, L. Breton, and J. Larochelle. 2004. Behavioural responses of moose to thermal conditions in the boreal forest. Ecoscience 11:321 328. Dyer, S. J., J. P. O'Neill, S. M. Wasel, and S. Boutin. 2001. Avoidance of industrial development by woodland caribou. Journal of Wildlife Management 65:531 542. Dyrness, C. T. 1973. Early stages of plant sucession following logging and burning in the western Cascades of Oregon. Ecology 54:57 69. Eason, G. 1989. Moose response to hunting and 1 km² block cutting. Alces 17:111 125. Euler, D. 1981. A moose habitat strategy for Ontario. Alces 17:180 192. Ferrier, S., G. Watson, J. Pearce, and M. Drielsma. 2002. Extended statistical approaches to modelling spatial pattern in biodiversity in northeast New South Wales. I. Species level modelling. Biodiversity and Conservation 11:2275 2307. Fielding, A. H., and J. F. Bell. 1997. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environmental Conservation 24:38 49. Forbes, G. J., and J. B. Theberge. 1993. Multiple landscape scales and winter distribution of moose, Alces alces, in a forest ecotone. Canadian Field Naturalist 107:201 207. Frair, J. L., E. H. Merrill, H. L. Beyer, and J. M. Morales. 2008. Thresholds in landscape connectivity and mortality risks in response to growing road networks. Journal of Applied Ecology 45:1504 1513. Frair, J. L., S. E. Nielsen, E. H. Merrill, S. R. Lele, M. S. Boyce, R. H. M. Munro, G. B. Stenhouse, and H. L. Beyer. 2004. Removing GPS collar bias in habitat selection studies. Journal of Applied Ecology 41:201 212. Fryxell, J. M., J. Greever, and A. R. E. Sinclair. 1988. Why are ungulate so abundant. American Naturalist 131:781 798. Gaillard, J. M., M. Festa Bianchet, N. G. Yoccoz, A. Loison, and C. Toigo. 2000. Temporal variation in fitness components and population dynamics of large herbivores. Annual Review of Ecology and Systematics 31:367 393. Garshelis, D. L. 2000. Delusions in habitat evaluation: Measuring use, selection and importance. in L. Boitani, andT. K. Fuller, editors. Research techniques in animal
39
ecology: controversies and consequences. Columbia University, New York, New York, USA. Gillies, C., M. Hebblewhite, S. E. Nielsen, M. Krawchuk, C. Aldridge, J. Frair, C. Stevens, D. J. Saher, and C. Jerde. 2006. Application of random effects to the study of resource selection by animals. Journal of Animal Ecology 75:887 898. Godvik, I. M. R., L. E. Loe, J. O. Vik, V. Veiberg, R. Langvatn, and A. Mysterud. 2009. Temporal scales, trade offs, and functional responses in red deer habitat selection. Ecology 90:699 710. Gotelli, N. J., and G. R. Graves. 1996. Null models in ecology. Smithsonian Institution Press. Government of Canada. 2002. Species at Risk Act: An act representing the protection of wildlife species at risk in Canada. in c. Statutes of Canada, vol. Bill C 5, editor. Gustine, D. D., K. L. Parker, R. J. Lay, M. P. Gillingham, and D. C. Heard. 2006. Interpreting resource selection at different scales for woodland caribou in winter. Journal of Wildlife Management 70:1601 1614. Haigh, J. C., W. J. Dalton, C. A. Ruder, and R. G. Sasser. 1993. Diagnosis of pregnancy in moose using a bovine assey for pregnancy specific protein b. Theriogenology 40:905 911. Hall, D. K., G. A. Riggs, and V. V. Salomonson. 2000. MODIS/Terra snow cover 8 day L3. Global 500m Grid V03, February 2000 to February 2002. in National Snow and Ice Data Center. Digital media. Boulder, CO, USA. Hall, L. S., P. R. Krausman, and M. L. Morrison. 1997. The habitat concept and a plea for standard terminology. Wildlife Society Bulletin 25:173 182. Hastie, T. J., and R. Tibshirani. 1990. Generalized additive models. Chapman and Hall, London. Hebblewhite, M., and D. T. Haydon. 2010. Distinguishing technology from biology: a critical review of the use of GPS telemetry data in ecology. Philosophical Transactions of the Royal Society B: Biological Sciences 365:2303 2312. Hebblewhite, M., and E. Merrill. 2008. Modelling wildlife human relationships for social species with mixed effects resource selection models. Journal of Applied Ecology 45:834 844.
40
Hebblewhite, M., E. Merrill, and G. McDermid. 2008. A multi scale test of the forage maturation hypothesis in a partially migratory ungulate population. Ecological Monographs 78:141 166. Hebblewhite, M., and E. H. Merrill. 2009. Trade offs between wolf predation risk and forage at multiple spatial scales in a partially migratory ungulate. Ecology 26:54 59. Hebblewhite, M., E. H. Merrill, and T. L. McDonald. 2005. Spatial decomposition of predation risk using resource selection functions: an example in a wolf elk predator prey system. Oikos 111:101 111. Hemson, G., P. Johnson, A. South, R. Kenward, R. Ripley, and D. Macdonald. 2005. Are kernels the mustard? Data from global positioning system (GPS) collars suggests problems for kernel home range analyses with least squares cross validation. Journal of Animal Ecology 74:455 463. Hervieux, D., J. Edmonds, R. Bonar, and J. McCammon. 1996. Successful and unsuccessful attempts to resolve caribou management and timber harvesting issues in west central Alberta. Rangifer 16:185 190. Hirzel, A. H., and G. Le Lay. 2008. Habitat suitability modelling and niche theory. Journal of Applied Ecology 45:1372 1381. Holt, R. D. 1977. Predation, apparent competition, and structure of prey communities. Theoretical Population Biology 12:197 229. Holt, R. D., and J. H. Lawton. 1994. The ecological consequences of shared natural enemies. Annual Review of Ecology and Systematics 25:495 520. Hosmer, D. W., and S. Lemeshow, editors. 2000. Applied Logistic Regression. John Wiley and Sons, New York. Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira. 2002. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment 83:195 213. Hundertmark, K. J. 2007. Home Range, Dispersal and Migration. Pages 223 245 in A. W. Franzmann, and C. C. Schwartz, editors. Ecology and management of the North American moose. University Press of Colorado, Boulder, Colorado, USA
41
James, A. R. C., S. Boutin, D. M. Hebert, and A. B. Rippin. 2004. Spatial separation of caribou from moose and its relation to predation by wolves. Journal of Wildlife Management 68:799 809. James, A. R. C., and A. K. Stuart Smith. 2000. Distribution of caribou and wolves in relation to linear corridors. Journal of Wildlife Management 64:154 159. Jenkins, K. J., and R. G. Wright. 1988. Resource partitioning and competition among cervids in the northern Rocky Mountains. Journal of Applied Ecology 25:11 24. Johnson, C. J., K. L. Parker, and D. C. Heard. 2001. Foraging across a variable landscape: behavioral decisions made by woodland caribou at multiple spatial scales. Oecologia 127:590 602. Johnson, C. J., D. R. Seip, and M. S. Boyce. 2004. A quantitative approach to conservation planning: using resource selection functions to map the distribution of mountain caribou at multiple spatial scales. Journal of Applied Ecology 41:238 251. Johnson, D. H. 1980. The comparison of usage and availability measurements for evaluating resource preference. Ecology 61:65 71. Katnik, D. D., and R. B. Wielgus. 2005. Landscape proportions versus Monte Carlo simulated home ranges for estimating habitat availability. Journal of Wildlife Management 69:20 32. Kernohan, B. J., R. A. Gitzen, and J. J. Millspaugh. 2001. Analysis of animal space use and movements. Pages 126 166 in J. J. Millspaugh and J. M. Marzluff, editors. Radio Tracking and Animal Populations. Academic Press, Inc., San Diego, California, USA. Klein, D. R. 1982. Fire, lichens, and caribou. Journal of Range Management 35:390 395. Krebs, C. J. 1999. Ecological methodolgy. 2nd edition edition. Benjamin/Cummings, Menlo Park, CA. Kunkel, K., and D. H. Pletscher. 1999. Species specific population dynamics of cervids in a multipredator ecosystem. Journal of Wildlife Management 63:1082 1093. Labonte, J., J. P. Ouellet, R. Courtois, and F. Belisle. 1998. Moose dispersal and its role in the maintenance of harvested populations. Journal of Wildlife Management 62:225 235.
42
Lessard, R. B., S. J. D. Martell, C. J. Walters, T. E. Essington, and J. F. Kitchell. 2005. Should ecosystem management involve active control of species abundances? Ecology and Society 10 (2). MacArthur, R. H., and R. Levins. 1967. The limiting similarity, convergence and divergence of coexisting species. American Naturalist 101:377 385. Manly, B. F. J., L. L. McDonald, D. L. Thomas, T. L. McDonald, and W. P. Erickson. 2002. Resource selection by animals: statistical design and analysis for field studies. 2nd edn. Kluwer Academic Publishers, Dordrecht, Netherlands. McDermid, G. J., S. E. Franklin, and E. F. LeDrew. 2005. Remote sensing for large area habitat mapping. Progress in Physical Geography 29:449 474. McLoughlin, P. D., M. S. Boyce, T. Coulson, and T. Clutton Brock. 2006. Lifetime reproductive success and density dependent, multi variable resource selection. Proceedings of the Royal Society B Biological Sciences 273:1449 1454. McLoughlin, P. D., J. S. Dunford, and S. Boutin. 2005. Relating predation mortality to broad scale habitat selection. Journal of Animal Ecology 74:701 707. McLoughlin, P. D., E. Dzus, B. Wynes, and S. Boutin. 2003. Declines in populations of woodland caribou. Journal of Wildlife Management 67:755 761. McNicol, J. G., and F. F. Gilbert. 1980. Late winter use of upland cutovers by moose. Journal of Wildlife Management 44:363 371. Messier, F. 1994. Ungulate populatioon models with predation a case study with the North American moose. Ecology 75:478 488. Messier, F., and D. O. Joly. 2000. Comment: Regulation of moose populations by wolf predation. Canadian Journal of Zoology 78:506 510. Mosnier, A., J. P. Ouellet, and R. Courtois. 2008. Black bear adaptation to low productivity in the boreal forest. Ecoscience 15:485 497. Mysterud, A. 2000. Diet overlap among ruminants in Fennoscandia. Oecologia 124:130 137. Mysterud, A., and R. A. Ims. 1998. Functional responses in habitat use: Availability influences relative use in trade off situations. Ecology 79:1435 1441.
43
Nellemann, C., I. Vistnes, P. Jordhoy, and O. Strand. 2001. Winter distribution of wild reindeer in relation to power lines, roads and resorts. Biological Conservation 101:351 360. Neufeld, L. M. 2006. Spatial dynamics of wolves and woodland caribou in an industrial forest landscape in west central Alberta. Master of Science Thesis. University of Alberta, Edmonton, Canada. Nielsen, S. E., C. J. Johnson, D. C. Heard, and M. S. Boyce. 2005. Can models of presence absence be used to scale abundance? Two case studies considering extremes in life history. Ecography 28:197 208. Osko, T. J., M. N. Hiltz, R. J. Hudson, and S. M. Wasel. 2004. Moose habitat preferences in response to changing availability. Journal of Wildlife Management 68:576 584. Paulsen, J., and C. Körner. 2001. GIS analysis of tree line elevation in the Swiss Alps suggests no exposure effect. Journal of Vegetation Science 12:817 824. Pearce, J., and S. Ferrier. 2000. Evaluating the predictive performance of habitat models developed using logistic regression. Ecological Modelling 133:225 245. _____. 2001. The practical value of modelling relative abundance of species for regional conservation planning: a case study. Biological Conservation 98:33 43. Peek, J. M. 2007. Habitat relationships. Pages 351 375 in A. W. Franzmann, and C. C. Schwartz, editors. Ecology and management of the North American moose. University Press Colorado, Boulder, Colorado. Pettorelli, N., J. O. Vik, A. Mysterud, J. M. Gaillard, C. J. Tucker, and N. C. Stenseth. 2005. Using the satellite derived NDVI to assess ecological responses to environmental change. Trends in Ecology & Evolution 20:503 510. Pianka, E. R. 1973. The structure of lizard communities. Johnston, Richard F. (Ed.). Annual Review of Ecology and Systematics, Vol. 4. Vii+424p. Illus. Map. Annual Reviews Inc.: Palo Alto, Calif., USA:53 74. Poole, K. G., and R. Serrouya. 2003. Moose population monitoring in the Lake Revelstoke valley, 2002 2003. Prepared for: Downie Street Sawmills, Wood River Forest Inc., Joe Kozek Sawmills Ltd., Revelstoke Community Forest Corporation and Mount Revelstoke and Glacier National Parks, Revelstoke, B.C.
44
Poole, K. G., R. Serrouya, and K. Stuart Smith. 2007. Moose calving strategies in interior montane ecosystems. Journal of Mammalogy 88:139 150. Rabe Hesketh, S., A. Skrondal, and A. Pickles. 2004. Generalized multilevel structural equation modeling. Psychometrika 69:167 190. Railsback, S. F., H. B. Stauffer, and B. C. Harvey. 2003. What can habitat preference models tell us? Tests using a virtual trout population. Ecological Applications 13:1580 1594. Rempel, R. S., P. C. Elkie, A. R. Rodgers, and M. J. Gluck. 1997. Timber management and natural disturbance effects on moose habitat: Landscape evaluation. Journal of Wildlife Management 61:517 524. Renecker, L. A., and C. C. Schwartz. 2007. Food habits and feeding behavior. Pages 403 439 in A. W. Franzmann, and C. C. Schwartz, editors. Ecology and management of the North American moose. University Press of Colorado Boulder, Colorado, USA. Rettie, W. J., and F. Messier. 2000. Hierarchical habitat selection by woodland caribou: its relationship to limiting factors. Ecography 23:466 478. _____. 2001. Range use and movement rates of woodland caribou in Saskatchewan Canadian Journal of Zoology 79:1933 1940. Robinson, H. 2007. Cougar demographics and resource use in response to mule deer and white tailed deer densities: A test of the apparent competition hypothesis. Dissertation, Washington State University. Robinson, H., M. Hebblewhite, M. Musiani, N. DeCesare, W. Peters, B. Weckworth, and A. McDevitt. 2010. Modeling relationships between fire, caribou, wolves, elk and moose to aid recovery of Southern Mountain Woodland Caribou in the Canadian Rocky Mountain National Parks. Final Report submitted to Parks Canada, Banff and Jasper National Parks, Revised version, Sept 27, 2010. Rodgers, A. R., A. P. Carr, H. L. Beyer, L. Smith, and J. G. Kie. 2007. Version 1.1. Ontario Ministry of Natural Resources, Centre for Northern Forest Ecosystem Research, Thunder Bay, Ontario, Canada.
45
Roever, C. L., M. S. Boyce, and G. B. Stenhouse. 2008. Grizzly bears and forestry II: Grizzly bear habitat selection and conflicts with road placement. Forest Ecology and Management 256:1262 1269. Sachro, L. L., W. L. Strong, and C. C. Gates. 2005. Prescribed burning effects on summer elk forage availability in the subalpine zone, Banff National Park, Canada. Journal of Environmental Management 77:183 193. Schlaepfer, M. A., M. C. Runge, and P. W. Sherman. 2002. Ecological and evolutionary traps. Trends in Ecology & Evolution 17:474 480. Seip, D. R. 1992. Factors limiting woodland caribou populations and their interrelationships with wolves and moose in Southeastern British Columbia. Canadian Journal of Zoology 70:1494 1503. Senft, R. L., M. B. Coughenour, D. W. Bailey, L. R. Rittenhouse, O. E. Sala, and D. M. Swift. 1987. Large herbivore foraging and ecological hierarchies. Bioscience 37:789 795 & 798 799. Servheen, G., and L. J. Lyon. 1989. Habitat use by woodland caribou in the Selkirk Mountains. Journal of Wildlife Management 53:230 237. Sinclair, A. R. E., and P. Arcese. 1995. Population consequences of predation sensitive foraging the serengeti wildbeest. Ecology 76:882 891. Sinclair, A. R. E., R. P. Pech, C. R. Dickman, D. Hik, P. Mahon, and A. E. Newsome. 1998. Predicting effects of predation on conservation of endangered prey. Conservation Biology 12:564 575. Smith, K. G. 2004. Woodland caribou demography and presistence relative tolandscape chnage in west central Alberta. Master of Science Thesis. University of Alberta, Edmonton, AB. Smith, K. G., E. J. Ficht, D. Hobson, T. C. Sorensen, and D. Hervieux. 2000. Winter distribution of woodland caribou in relation to clear cut logging in west central Alberta. Canadian Journal of Zoology 78:1433 1440. StataCorp. 2007. StataCorp LP, College Station, TX, USA. Stelfox, J. G., and R. D. Taber. 1969. Big game in the northern Rocky Mountains coniferous forest. Pages 197 220 in R. D. Taber, editor. coniferous forests the northern Rocky Mountains University Montana Foundation, Missoula, MT, USA.
46
Stotyn, S. A. 2008. Ecological interactions of mountain caribou, wolves and moose in the North Columbia Mountains, British Columbia. Master of Science Thesis, University of Alberta, Edmonton. Stuart Smith, A. K., C. J. A. Bradshaw, S. Boutin, D. M. Hebert, and A. B. Rippin. 1997. Woodland Caribou relative to landscape patterns in northeastern Alberta. Journal of Wildlife Management 61:622 633. Szkorupa, T., and F. Schmiegelow. 2003. Multi scale habitat selection by mountain caribou in West Central Alberta. Rangifer:293 294. Telfer, E. S. 1970. Winter habitat selection by moose and white tailed deer. Journal of Wildlife Management 34:553 559. Thomas, D. L., E. J. Edmonds, and W. K. Brown. 1996. The diet of woodland caribou populatiosn in west central Alberta Rangifer Special Issue No. 9:337 342. Timmermann, H. R., and M. E. Buss. 2007. Population and Harvest Management. Pages 559 616 in A. W. Franzmann, and C. C. Schwartz, editors. Ecology and Management of the North American Moose. University Press of Colorado, Boulder. Timmermann, H. R., and J. G. McNicol. 1988. Moose habitat needs. Forestry Chronicle 64:238 245. Van Horne, B. 1983. Density as a midleading indicator of habitat quality. Journal of Wildlife Management 47:893 901. Vander Wal, E. V., and A. R. Rodgers. 2009. Designating Seasonality Using Rate of Movement. Journal of Wildlife Management 73:1189 1196. Vistnes, I., and C. Nellemann. 2008. The matter of spatial and temporal scales: a review of reindeer and caribou response to human activity. Polar Biology 31:399 407. Weclaw, P., and R. J. Hudson. 2004. Simulation of conservation and management of woodland caribou. Ecological Modeling 177:75 94. Whittington, J., C. C. St Clair, and G. Mercer. 2005. Spatial responses of wolves to roads and trails in mountain valleys. Ecological Applications 15:543 553. Winnie, J., and S. Creel. 2007. Sex specific behavioural responses of elk to spatial and temporal variation in the threat of wolf predation. Animal Behaviour 73:215 225.
47
Wittmer, H. U., B. N. McLellan, D. R. Seip, J. A. Young, T. A. Kinley, G. S. Watts, and D. Hamilton. 2005a. Population dynamics of the endangered mountain ecotype of woodland caribou ( Rangifer tarandus caribou ) in British Columbia, Canada. Canadian Journal of Zoology 83:407 418. Wittmer, H. U., B. N. McLellan, R. Serrouya, and C. D. Apps. 2007. Changes in landscape composition influence the decline of a threatened woodland caribou population. Journal of Animal Ecology 76:568 579. Wittmer, H. U., A. R. E. Sinclair, and B. N. McLellan. 2005b. The role of predation in the decline and extirpation of woodland caribou. Oecologia 144:257 267. Worton, B. J. 1995. Using monte carlo simulation to evaluate kernel based home range estimators. Journal of Wildlife Management 59:794 800.
48
Table 2 1. The spatial separation hypothesis predicts that woodland caribou spatially separate from other ungulates and their shared predators through their niche specialization. We tested different hypotheses in two landscapes with high and low human landscape disturbance and derived predictions based on the spatial separation hypothesis. Objectives Hypotheses Predictions 1) Moose H1: Moose select habitats P1a: Moose select disturbed habitats (i.e., habitat associated with human burns, cutblocks) due to increased forage selection landscape alteration, but make at large spatial scales (i.e., landscape trade offs between forage and scale). predation at multiple spatial P1b: Moose avoid predation risk by scales. spatially separating from roads and other linear features at finer spatial scales (i.e., home range scale). 2) Caribou H2: Caribou and moose P2a: Caribou use browse rich habitats and moose partition resources to maintain (i.e., shrubs, cutblocks, burns) at lower resource spatial separation as predicted elevations less compared to moose and partitioning by the spatial separation avoid high densities of linear features (i.e., hypothesis. roads, seismic lines, etc) more compared to moose. H3: The spatial separation P3a: Lower spatial separation and between caribou and moose is therefore, higher niche overlap in summer decreased by human due to increased resource overlap landscape alteration. compared to winter. P3b: Decreased resource partitioning and spatial separation leads to increased overlap of realized moose and caribou niches in human altered landscapes. P3c: Asymmetric overlap in favor of moose, i.e., moose will overlap more with caribou niches than vice versa. H4: Predator caused caribou P4a: Caribou mortality is higher in areas mortality risk is elevated with with high probability of moose and decreased spatial separation. caribou resource overlap.
49
Table 2 2. Description of covariates used in RSF models to estimate moose habitat selection and moose at multiple temporal and spatial scales (data collected 2008 – 2010) and to determine differences in habitat use between moose and woodland caribou (data collected 2007 – 2010) in west central Alberta and east central British Columbia, Canada.
Covariate Type Covariate Description Human/Natural Disturbance Road Density Continuous Road density calculated for each cell (km/km 2) based on a composite roads data layer. Density of Linear Continuous Density of seismic exploration lines, hiking trails, Features (Line railways and pipelines for each cell (km/km 2) based Density) on a composite linear features data layer. Young cutblock Categorical Cut blocks < 20 years old. Old cutblock Categorical Cut blocks ≥ 20 years and < 40 years old. Young burn Categorical Burns < 20 years old. Old burn Categorical Burns ≥ 20 years and < 40 years old. Forage Baseline Shrub Categorical Shrub communities below tree line. Topography Elevation Continuous Elevation in meters. Slope Continuous Percent slope (0 89°). Other Variables Closed conifer Categorical Closed conifer forest with ≥ 50% canopy closure and ≥ 70% coniferous. Reference category. Open conifer Categorical Open conifer forest ≤ 50% canopy closure and ≥ 70% coniferous. Mixed forest Categorical Mixed forest ≥ 30%, but < 70% coniferous. Deciduous Categorical Deciduous dominated forest < 30% coniferous. Alpine Categorical Regions above tree line, except glaciers. Herbaceous Categorical Grasslands below tree line. Barren Categorical Barren ground below tree line. Muskeg Categorical Treed and herbaceous wetlands at all elevations. Water Categorical Water at all elevations. Glacier Categorical Permanent ice. NDVI Continuous Mean of NDVI in non forested habitats for growing season (each year). Snow Continuous Seasonal average a raster cell has been covered by snow (each year). Distance to open Continuous Singe direction distance (m) to open canopy (i.e., value within open canopy is 0).
50
Table 2 3. Selection coefficients (β) and standard errors (SE) from the most parsimonious generalized linear mixed models with a random intercept describing moose resource selection during summer (22 May – 16 Nov) at the landscape and home range scales in the foothills of west central Alberta and east central British Columbia, Canada, from 2008 2010. Coefficients in bold indicate significant variables in the models at an α level of 0.05. Closed conifer was the reference category for landcover types.
Landscape scale Home range scale Covariates β SE β SE Human/ Natural Disturbance Young cutblock 0.425 0.0690 0.164 0.0994 Old cutblock 1.332 0.2658 0.151 0.8710 Young burn Old burn 2.124 0.7577 Road density (km/km 2) 0.512 0.0378 0.476 0.1117 Line density (km/km 2) 0.181 0.0365 0.023 0.0471 Line density 2 (km/km 2) 0.042 0.006 Moose Forage Baseline Shrub 0.185 0.0663 0.172 0.0704 Topography Elevation (100m) 1.302 0.0608 0.321 0.0520 Elevation 2 (100m) 4.63E 04 2.20E 05 Slope 0.013 0.0029 0.015 0.0046 Other Significant Variables Distance to open (100m) 0.033 0.0060 0.050 0.0131 NDVI average 4.15E 04 2.40E 05 1.79E 04 4.88E 05 Barren 0.705 0.1550 0.655 0.1988 Muskeg 0.601 0.0749 0.305 0.0963 Herbaceous 0.784 0.1386 Mixed forest 0.312 0.0730 Open conifer 0.258 0.0932 Deciduous 0.255 0.1127 Model constant 5.854 0.4102 k folds rho (p value) 0.893 (<0.001) 0.451 (0.204) Notes: Young cutblocks and burns were ≤ 20 years old and old cutblocks and burns were >20 years old. Line density represents densities of seismic exploration lines, pipelines, railways and human trails. NDVI = Normalized Difference Vegetation Index.
51
Table 2 4. Selection coefficients (β) and standard errors (SE) from the most parsimonious generalized linear mixed models with a random intercept describing moose resource selection during winter (17 Nov – 21 May) at the landscape and home range scales in the foothills of west central Alberta and east central British Columbia, Canada, from 2008 2010. Coefficients in bold indicate significant variables in the models at an α level of 0.05. Closed conifer was the reference category for landcover types.
Landscape scale Home range scale Covariates β SE β SE Human/ Natural Disturbance Young cutblock 0.008 0.0622 0.505 0.0907 Old cutblock 0.170 0.1900 0.928 0.4227 Young burn Old burn Road density (km/km 2) 0.470 0.0370 0.102 0.1060 Line density (km/km 2) 0.057 0.0364 0.078 0.0470 Line density 2 (km/km 2) 0.013 0.0562 Moose Forage Baseline Shrub 0.450 0.0635 0.306 0.0600 Topography Elevation (100m) 3.915 0.1299 0.608 0.0708 Elevation 2 (100m) 0.002 0.0001 Other Significant Variables Distance to open (100m) 0.018 0.0063 0.167 0.0151 Snow 4.180 0.3159 1.675 0.6817 Barren 0.337 0.0635 Muskeg 0.585 0.0740 0.504 0.0741 Herbaceous 0.293 0.1277 0.608 0.1381 Mixed forest 0.213 0.0716 Open conifer 0.192 0.0942 Deciduous 0.249 0.1043 Model constant 20.310 0.7880 k folds rho (p value) 0.993 (<0.001) 0.723 (0.0210) Notes: Young cutblocks and burns were ≤ 20 years old and old cutblocks and burns were > 20 years old. Line density represents densities of seismic exploration lines, pipelines, railways and human trails.
52
Table 2 5. Selection coefficients (β) and standard errors (SE) for the most parsimonious generalized linear mixed models with a random intercept describing moose resource selection during summer (22 May – 16 Nov) at the landscape and home range scales in the mountain region in west central Alberta and east central British Columbia, Canada, from 2008 2010. Coefficients in bold indicate significant variables in the models at an α level of 0.05. Closed conifer was the reference category for landcover types.
Landscape scale Home range scale Covariates β SE β SE Human/ Natural Disturbance Young cutblock 1.735 0.5387 Old cutblock Young burn 2.175 0.1176 0.590 0.2500 Old burn Road density (km/km 2) 3.100 0.2131 0.654 0.9325 Line density (km/km 2) 0.430 0.0726 0.040 0.0875 Line density 2 (km/km 2) 0.239 0.0321 Moose Forage Baseline Shrub 0.331 0.0868 0.220 0.0840 Topography Elevation (100m) 0.476 0.0436 0.089 0.0368 Elevation 2 (100m) 1.27E 04 1.20E 05 Slope 0.019 0.0020 0.039 0.0038 Other Significant Variables Distance to open (100m) 0.017 0.008 0.066 0.019 Snow 1.116 0.2003 NDVI average 2.51E 04 3.15E 05 Barren 0.558 0.1642 Herbaceous 0.732 0.1848 Mixed forest 0.430 0.0968 Glacier 0.937 0.2194 Model constant 3.723 0.416 k folds rho (p value) 0.973 (<0.001) 0.754 (0.289) Notes: Young cutblocks and burns were ≤ 20 years old and old cutblocks and burns were >20years old. Line density represents densities of seismic exploration lines, pipelines, railways and human trails. NDVI = Normalized Difference Vegetation Index.
53
Table 2 6. Selection coefficients (β) and standard errors (SE) from the most parsimonious generalized linear mixed models with a random intercept describing moose resource selection during winter (17 Nov – 21 May) at the landscape and home range scales in the mountains of west central Alberta and east central British Columbia, Canada, from 2008 2010. Coefficients in bold indicate significant variables in the models at an α level of 0.05. Closed conifer was the reference category for landcover types.
Landscape scale Home range scale Covariates β SE β SE Human/ Natural Disturbance Young cutblock 4.345 1.0089 Old cutblock Young burn 1.267 0.1309 0.556 0.1612 Old burn 1.498 0.4275 Road density (km/km 2) 1.653 0.0094 1.407 0.0441 Line density (km/km 2) 0.740 0.0066 0.130 0.0839 Line density 2 (km/km 2) 0.317 0.0266 Moose Forage Baseline Shrub 0.331 0.0868 0.220 0.0840 Topography Elevation (100m) 0.401 0.053 0.004 0.0004 Elevation 2 (100m) 1.58E 04 1.48E 05 Slope 0.034 0.0039 Other Significant Variables Distance to open (100m) 0.132 0.014 Snow 1.973 0.2450 1.176 0.3432 Herbaceous 0.785 0.1803 Mixed forest 0.273 0.0912 Open conifer 0.229 0.0599 Deciduous 0.588 0.1582 Glacier 3.126 1.0091 Model constant 0.314 0.7880 k folds rho (p value) 0.990 (<0.001) 0.714 (0.030) Notes: Young cutblocks and burns were ≤ 20 years old and old cutblocks and burns were > 20 years old. Line Density represents densities of seismic exploration lines, pipelines, railways and human trails.
54
Table 2 7. Coefficients (β), standard errors (SE) and z values for the most parsimonious generalized linear mixed models with a random intercept describing differences in habitat use by woodland caribou (dependent variable = 1) and moose (dependent variable = 0) in the foothills of west central Alberta and east central British Columbia, Canada. Habitat use was compared during winter (17 Nov – 21 May) and summer (22 May – 16 Nov) from 2007 2009. Closed conifer was the reference category for land cover types. All variables were significant at an α level of 0.05.
Summer Winter Covariate β SE z value β SE z value Human/Natural Disturbance Young cutblock 0.75 0.118 5.89 1.94 0.131 14.85 Young burn 5.23 1.011 5.18 Line density (km/km 2) 0.35 0.022 15.7 0.23 0.021 11.28 Road density (km/km 2) 0.66 0.072 9.15 0.25 0.054 4.71 Moose Forage Baseline Shrub Topography Elevation (100m) 1.22 0.022 54.63 0.93 0.200 45.42 Other Significant Variables Muskeg 0.39 0.064 6.02 Mixed forest 0.61 0.096 5.38 0.92 0.090 9.53 Deciduous 1.26 0.161 7.86 Herbaceous 0.79 0.159 4.47 1.47 0.159 9.24 Open conifer 0.30 0.077 3.36 0.88 0.079 11.13 Distance to open (100m) 0.096 0.0069 13.90 NDVI 1.29E 04 3.13E 05 54.63 Model constant 14.76 0.87 17.07 11.08 0.233 24.14 ROC 0.81 0.78 Sensitivity (Caribou) 69.29 71.40 Specificity (Moose) 71.84 68.10 Notes: Young cutblocks and burns were ≤ 20 years old and old cutblocks and burns were > 20 years old. Line Density represents densities of seismic exploration lines, pipelines, railways and human trails. NDVI = Normalized Difference Vegetation Index.
55
Table 2 8. Coefficients (β), standard errors (SE) and z values for the most parsimonious generalized linear mixed models with a random intercept describing differences in habitat use by woodland caribou (dependent variable = 1) and moose (dependent variable = 0) in the mountains of west central Alberta and east central British Columbia, Canada. Habitat use was compared during winter (17 Nov 21 May) and summer (22 May 16 Nov) from 2007 2009. Closed conifer was the reference category for land cover types. All variables were significant at an α level of 0.05.
Summer Winter
Covariate β SE z value β SE z value Human/Natural Disturbance Young cutblock Young burn 4.15 0.180 25.51 4.45 1.003 4.43 Line density (km/km 2) 0.68 0.032 21.17 Road density (km/km 2) 5.03 0.466 10.79 3.42 0.156 21.83 Moose Forage Baseline Shrub 0.71 0.095 9.39 0.76 0.089 0.09 Topography Elevation (100m) 0.35 0.012 29.72 0.60 0.010 57.64 Other Significant Variables Barren 1.65 0.270 7.95 1.02 0.274 3.41 Muskeg 3.12 0.221 14.08 Mixed forest 1.43 0.127 7.74 1.27 0.155 7.65 Herbaceous 1.26 0.234 5.37 0.85 0.289 2.96 Open conifer 0.34 0.059 5.82 1.06 0.056 18.92 Distance to open (100m) 0.289 0.0088 32.94 NDVI average 0.00 0.000 29.93 Model constant 3.19 0.253 12.58 11.27 0.253 12.58 ROC 0.83 0.86
Sensitivity (Caribou) 75.98 78.96
Specificity (Moose) 75.63 74.55
Notes: Young cutblocks and burns were ≤ 20 years old and old cutblocks and burns were > 20 years old. Line Density represents densities of seismic exploration lines, pipelines, railways and human trails. NDVI = Normalized Difference Vegetation Index.
56
Figure 2 1. Study area based on the maximum extent of available locations for moose resource selection function modeling within the foothills (northeast) and mountains (southwest) of west central Alberta and east central British Columbia, Canada. Seventeen Global Positioning System (GPS) were deployed on moose between winters 2007/2008 and 2009/2010 (see 99% home range kernels) within or adjacent to caribou herd home ranges (see 95% home range kernels). Data (collected between winters 2006/2007 and 2009/2010) from 17 GPS collars deployed on caribou of the displayed herds were used for comparison of resource use between caribou and moose.
c
57
1.1 a) Foothill
0.6
0.1
0.4 Landscape SU 0.9 Home range SU
1.4 Landscape WI Home range WI 1.9 Selection Coefficient Selection
2.4
2.9 Young Old Shrub Young Burn Old Burn Cutblock Cutblock
2.2 b) Mountain
1.2
0.2
0.8 Landscape SU
1.8 Home range SU Landscape WI 2.8 Home range WI 3.8 Selection Coefficient Selection
4.8
5.8 Young Old Shrub Young Burn Old Burn Cutblock Cutblock
Figure 2 2. Selection coefficients (β) with standard error bars for young and old cutblocks and burns, and shrubs from the most parsimonious generalized linear mixed models with a random intercept estimating moose resource selection at the landscape and home range scale during summer (22 May 16 Nov) and winter (17 Nov 21 May). Data were collected with 17 global position in system collars in the foothills (a; n=8) and mountains (b; n=9) of west central Alberta and east central British Columbia, Canada, between winters of 2007/2008 and 2009/2010. All estimates are in comparison to the categorical land cover variables subsumed in the model specific intercept (of top models).
58
0.55 a) Foothill 0.5
0.45 2nd Order Summer* 0.4
3rd Order Summer* 0.35
0.3 2nd Order Winter*
Relative Probability of Moose Selection of Moose RelativeProbability 0.25 3rd Order Winter 0.2 0 0.025 0.05 0.075 0.1 0.125 0.15 0.175 0.2 0.225 0.25 0.275 0.3 Road Density (km/km2) 0.55 b) Mountain 0.5
0.45
0.4 2nd Order Summer *
0.35 3rd Order Summer
0.3 2nd Order Winter*
0.25
Relative Probability of Moose Selection of Moose RelativeProbability 3rd Order Winter* 0.2 0 0.025 0.05 0.075 0.1 0.125 0.15 0.175 0.2 0.225 0.25 0.275 0.3 Road Density (km/km2) Figure 2 3. Selection coefficients (β) for road density (km/km 2) from the most parsimonious generalized linear mixed models with a random intercept describing moose resource selection at the landscape and home range scales during summer (22 May – 16 Nov) and winter (17 Nov – 21 May). Significant coefficients are marked with a star. Data were collected with 17 moose global positioning system collars in the foothills (a; n = 8) and mountains (b; n = 9) of west central Alberta and east central British Columbia, Canada, between winters of 2007/2008 and 2009/2010.
59
a) Foothill Summer b) Foothill Winter
c) Mountain Summer d) Mountain Winter
Figure 2 4. Selection coefficients for quadratic functions describing density of linear features (seismic exploration lines, pipe lines, railways and human trials combined; km/km 2) from the most parsimonious generalized linear mixed models with a random intercept estimating moose resource selection at the landscape and home range scales during summer (22 May – 16 Nov) and winter (17 Nov – 21 May). Data were collected with 17 moose GPS collars in the foothills (a, b; n = 8) and mountains (c, d; n = 9) of west central Alberta and east central British Columbia, Canada, between winters of 2007/2008 and 2009/2010.
60
a) Foothill Summer b) Foothill Winter
c) Mountain Summer d) Mountain Winter
Figure 2 5. Relative probability of use by woodland caribou and moose from the most parsimonious generalized linear mixed models with a random intercept during summer (22 May – 16 Nov) and winter (17 Nov – 21 May) in the foothills (a, b) and mountains (c, d) of west central Alberta and east central British Columbia, Canada. Values closer to 0 indicate high relative probability of use by moose and conversely, values closer to 1 indicate high relative probability of use by caribou. Areas with high overlap of both species indicate low resource partitioning. Models were built with data from global positioning collars (GPS) deployed on 17 moose and 17 caribou between winters 2007/2008 and 2009/2010.
61
0.7 M&L Moose 0.6 M&L Caribou 0.5 Pianka 0.4
0.3
0.2
Niche OverlapValues Index Niche 0.1
0.0 FH wi FH su MT wi MT su
Figure 2 6. Moose and caribou niche overlap based on predicted probabilities of resource use from most parsimonious generalized linear mixed models with a random intercept describing resource partitioning of both species in the foothills (FH) and mountains (MT) of west central Alberta and east central British Columbia, Canada, during summer (su; 22 May – 16 Nov) and winter (wi; 17 Nov – 21 May). Models were built with data collected with global positioning collars deployed on 17 moose and 17 caribou between winters 2007/2008 and 2009/2010. Niche overlap indices were calculated following MacArthur and Levins (M&L; 1967) and Pianka (1973).
62
45 40 FH su FH wi 35
30 MT su MT wi 25 20 15 10 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Proportion Proportion of Mortalities in each Resource Bin Moose Overlap Caribou
Figure 2 7. Caribou mortalities versus predicted probabilities of resource use by woodland caribou relative to moose in the foothills (FH) and mountains (MT) of west central Alberta and east central British Columbia, Canada, during summer (su; 22 May 16 Nov) and winter (wi; 17 Nov – 21 May). Values of 0.1 and 0.2 indicate the highest relative probability of use by moose and conversely, values of 0.9 and 1 indicate the highest relative probability of use by caribou. Predictive logistic regression models were built with data collected with global positioning collars (GPS) deployed on 17 moose and 17 caribou between winters 2007/2008 and 2009/2010. We used 45 predator caused caribou mortalities recorded by Alberta Fish and Wildlife Division and Parks Canada between 1999 and 2009.
63
CHAPTER 3: MOOSE POPULATION ESTIMATION USING DISTANCE SAMPLING
IN WEST CENTRAL ALBERTA
INTRODUCTION
Abundance estimates are important for the management of wildlife species (Rivest et al. 1990, Timmermann and Buss 2007) and improve understanding of community ecology. Moose ( Alces alces ) populations, the dominant prey species for wolves ( Canis lupus ) throughout the boreal forest (Lessard et al. 2005, Wittmer et al. 2005), are hypothesized to increase following the conversion of old growth forests to young seral stands through human landscape disturbance (McNicol and Gilbert 1980, Wittmer et al. 2005). This increase in moose density is assumed to result in increased wolf populations, leading to higher predation rates on threatened woodland caribou (Rangifer tarandus caribou , COSEWIC 2002, James et al. 2004).While increased moose harvests are being implemented following Alberta Caribou Recovery Team (2005) recommendations in several caribou home ranges, wildlife managers are also under pressure to maintain moose populations for subsistence and recreational hunting (James et al. 2004). Population assessment is often viewed as one of the primary components of moose management (Ward et al. 2000) and is especially important when balancing conflicting demands between caribou recovery and harvest management. Aerial surveys are the most practical tool for estimating moose population size (LeResche and Rausch 1974, Timmermann and Buss 2007). The main problem in aerial population estimation is to account for missed animals (Caughley 1974, Pollock and Kendall 1987). The magnitude of visibility bias, the sum of the underlying causes for incomplete detection (Caughley 1974), often is especially underestimated in heterogeneous landscapes (Pollock et al. 2006). Visibility bias can be divided into perception bias and availability bias (Marsh and Sinclair 1989). Perception bias occurs when observers miss visible animals due to fatigue or other factors, while availability bias occurs when animals are not available to be detected (e.g., they may be covered by vegetation; Marsh and Sinclair 1989). Moose populations across North America are commonly surveyed using a stratified random block (SRB) design. This method is based on preliminary stratification flights to classify the survey area into survey units (SUs) 64 based on counts, which are then contemporaneously surveyed at random (Gasaway et al. 1986). The SRB survey design ideally corrects for visibility bias by resurveying smaller portions of general SUs with 2 3 times higher search effort (intensive survey). The difference between the general survey and intensive survey is then used to estimate the proportion of moose missed during the general search and develop a sightability correction factor (SCF; Gasaway et al. 1986). Unfortunately, sightability of moose can be lower than 50% due to visibility bias (the sum of availability and perception bias), especially when surveys are conducted in areas with dense vegetation cover (Anderson and Lindzey 1996). Therefore, even during a very intensive search it is unlikely that all moose will be observed (Quayle et al. 2001) and moose population estimates from SRB surveys will still be biased low. Overall, the SRB design is very flight and labor intensive, especially in study areas with lower moose densities or lower detection probabilities (i.e., sightability), and therefore is most appropriate for surveying dense populations in relatively small, open study areas (Gasaway et al. 1985, Buckland et al. 2001). The high costs of the SRB design often limit their frequency and spatial extent (Ward et al. 2000). Wildlife managers must therefore consider other cost effective survey methods such as distance sampling (Buckland et al. 2001, Nielson et al. 2006). In contrast to SRB surveys, where the probability of detecting a moose ( ˧(y) ) is assumed to be constant for all distances ( y) within a fixed transect width (e.g., in Alberta commonly 200m; Shorrocks et al. 2008), distance sampling uses the perpendicular distances to estimate detection probability as a function of distance (Laake et al. 2008). The most critical assumption for distance sampling is that animals on the transect center line are detected with 100% certainty (g(0) = 1 ; Laake et al. 2008). Unfortunately, detection probability of moose on the transect line can be much less than 1 (˧(0) ≠ 1) due to decreased visibility (Anderson and Lindzey 1996, Nielson et al. 2006), biasing estimates low (Buckland et al. 2004; Figure 3 1). Double observer approaches (i.e., two independent observers) have been used to correct for visibility bias, with mark recapture models (Manly et al. 1996, Potvin et al. 2004). While the implication of double observer distance sampling is fairly common (Pollock and Kendall 1987, Crete et al. 1991, Ridgway 2010), this method has also been criticized for overestimating sightability,
65 because it essentially corrects for perception bias (i.e., the portion of visibility bias when an animal is available to be detected), but not for availability bias (Marsh and Sinclair 1989, Laake et al. 2008). Studies also correcting for availability bias are rare, even though it may be much larger than perception bias (Buckland et al. 2004). Sightability models that correct for perception and availability bias can be developed using known (e.g., radio collared) animals by relating the probability of detecting an animal to variables that influence sightability (Anderson and Lindzey 1996, Laake et al. 2008). Moose sightability models have been developed for Wyoming (Anderson and Lindzey 1996), northern Michigan (Drummer and Aho 1998) and south central British Columbia (Quayle et al. 2001), and reported that moose sightability generally declines with increasing tree cover and decreasing group size, but have not been incorporated into distance sampling for moose. Despite the potential advantages of distance sampling, it has only been used for moose population estimation in Alaska (Nielson et al. 2001) and in northern British Columbia (BC; Thiessen 2010). Our study attempted to evaluate distance sampling for moose population estimation in a study area with low to moderate moose densities and heterogeneous canopy cover in west central Alberta, Canada. We compared survey results from distance sampling to results from SRB surveys in terms of flight time efficiency and confidence intervals (CI) within the same Wildlife Management Unit (WMU). We expected that distance sampling would provide at least as precise population estimates for moose (at a 90% CI), while requiring less survey effort. To assess whether
˧(0) = 1, we surveyed radio collared moose at known locations to estimate sightability. We hypothesized that detection probability on the transect line would be < 1.0 due to visibility bias. If ˧(0) ≠ 1, we rescaled the distance sampling detection function to the true probability of detecting moose (Figure 3 1). Based on previous studies, we expected canopy closure and group size to drive visibility of moose on the transect line (e.g., Gasaway et al. 1986, Quayle et al. 2001).
66
METHODS
STUDY AREA Distance sampling and SRB surveys were conducted during winter in WMU 353 which was representative of a broader study area in which sightability trials were conducted (Figure 3 2). Wildlife Management Unit 353 (54°N / 117°W) is approximately 4,600 km 2. The home range of the Little Smoky woodland caribou herd, which is at immediate risk of extirpation (Alberta Sustainable Resource Development and Alberta Conservation Association 2010), extends into the southern portion of WMU 353 (Figure 3 2) and as a result, the harvest of female moose has been recently increased as a caribou recovery strategy (Alberta Woodland Caribou Recovery Team 2005). Sightability trials on radio collared moose were flown in the region surrounding WMU 353 within the foothill and mountain regions of west central AB and east central BC (Figure 3 2). Climate in the study area is subarctic with short, wet, cool summers and long, dry and cool winters (Smith et al. 2000). Vegetation was characterized by pure lodgepole pine ( Pinus contorta ) or lodgepole pine/ black spruce ( Picea mariana ) forests on drier, low elevation sites, and on more mesic, higher elevation sites mixed balsam fir ( Abies lasiocarpa ), spruce ( Picea spp.), and lodgepole pine forests. Along drainages willow ( Salix spp.), birch ( Betula spp.) and some aspen ( Populus tremuloides ) are interspersed with dry grassy benches. The study area experienced substantial levels of human disturbance and was characterized by high densities of forest harvests and linear developments (e.g., roads, pipelines, seismic lines; Smith et al. 2000). In WMU 353 elevations ranged from 650 to 1,600m, similar to elevations of the whole study area, ranging from 760 to 1880m. Human hunting of moose by Treaty First Nations (i.e., year round, unregulated) and by licensed hunting in late fall occurred throughout the study area. Moose harvest rates were highest in WMU 353 with up to 40% of the estimated moose population (AB Fish and Wildlife Division, unpublished data) to support caribou recovery. Coincidently, mangers increased moose hunting licenses issued in WMU 353 in recent years and were highest in fall 2009 with 472 moose harvested (for comparison harvest rates of 2005 = 184, 2006 = 331, 2007 = 361and 2008 = 349).
67
SIMPLE RANDOM BLOCK SURVEYS Wildlife Management Unit 353 was surveyed in winters 2000/2001, 2006/2007, and partially in 2008/2009 (about 0.75 of the WMU).We conducted SRB surveys following methodology described originally by Gasaway et al. (1986), and modified by Alberta Fish and Wildlife Division (ABFW) following Lynch and Shumaker (1995) and Lynch (1997). To stratify the survey area into low, medium and high SUs based on the frequency distribution of detected moose in each SU, straight east west transect lines at ~160km/hr at ~90m above ground level (AGL) were flown with a fixed wing aircraft (Cessna185 or 206) at every minute of latitude (Gasaway et al. 1986). Survey units were 5 minute latitude by 5 minute longitude in size and a minimum of 5 SU’s per stratum were randomly chosen (sampling without replacement) and re surveyed with a Bell 206 Jet Ranger helicopter at ~80 140km/hr. Transect line spacing was 400m and the survey altitude varied between at 60 100m AGL. Flight crews consisted of one pilot and 3 experienced ABFW or Alberta Conservation Association (ACA) observers. General surveys followed the stratification flights immediately to avoid changes of moose distribution. Data recorded during the surveys were group size, sex (based on presence of vulvae patches or antler scars) and age class (calf or adult). Population estimates for the observable stratum ( ˠ) were calculated, by dividing the total number of moose seen in all surveyed SUs of the same stratum by the total surface area of all surveyed SUs (km 2) of the same stratum. Further, variance of each stratum population estimate ˢ ˠ , the observable population estimate ( ˠ ), the sampling variance ˢ ˠ" for the total survey area, and 90% confidence intervals were calculated following Gasaway et al. (1986; see Appendix B 1 for details). In Alberta, 20% CVs at 90% CIs are usually the desired goal of precision for estimates of moose abundance. If after flying 5 SUs of each stratum confidence intervals were much > 20%, additional survey units were flown of those strata with high variance. Gasaway et al. (1986) recommend estimating a sightability correction factor (SCF) to correct for sightability bias during SRB surveys by performing high intensity searches at an outset of a sample of SUs. However, moose densities in west central AB may be < 0.39 moose/km 2 in many areas and thus, below the recommended threshold where estimating a sightability correction factor is economically feasible (Gasaway et al.
68
1986). Thus, usually no SCF was estimated because of the high costs required. As a result, SRB surveys should be considered as minimum estimates and direct comparisons of surveys should be done cautiously.
DISTANCE SAMPLING We conducted one preliminary survey in WMU 440 immediately south west of WMU 353 (Figure 3 2) in winter 2008/2009 to provide pilot data to guide subsequent survey design in WMU 353, and to test assumptions of distance sampling. We estimated the total transect length necessary for WMU 353 to produce population estimates with a CV of less than 20% at a 90% CI as a function of moose density and variance in moose group size from the preliminary survey in WMU 440 following Buckland et al. (2001). Distance sampling surveys were conducted in WMU 353 on 5 days from January – March 2010. We conducted surveys during high visibility weather and complete snow coverage (Timmermann and Buss 2007) with a Bell 206 Jet Ranger helicopter with bubble windows. We established systematic transects every 3 minutes of latitude, and flew transects following a Global Positioning System (GPS, Garmin GPS76; Garmin International, Olathe, KS, USA) at 70 100m AGL and 80 140 km/hr. Surveys were conducted by 4 experienced observers, including the pilot as the front right observer. The front left observer was responsible for detecting moose near the transect line through the foot window of the helicopter and the rear observers recorded moose on each side. The rear right observer recorded locations of moose with an independent GPS to measure perpendicular distance from the transect following Marques et al. (2006) in ArcGIS 9.3.1 (Environmental Systems Research Institute, Inc., Redlands, CA, USA). Once the helicopter was perpendicular to the observed moose, it went ‘‘off effort’’ to record the moose location. Moose groups were the unit of observation to ensure independence (Buckland et al. 2001) and included cow calf pairs or moose that were spatially closely aggregated (i.e., < 50m apart). We did not record observations that were detected upon leaving the transect line to record a GPS location. The rear left observer recorded covariates known to influence detection probability including composition (male, female, cow calf), moose activity (bedded, standing or moving), light intensity (flat or bright), and topography (flat, moderate, steep; e.g. LeResche and Rausch 1974, Gasaway et al.
69
1986, Anderson and Lindzey 1996; Table 2 1). We classified canopy closure in 3 categories at 33% intervals based on figures by Unsworth et al. (1994). Distance sampling data were analyzed in program DISTANCE v. 6.0. release 2 (Thomas et al. 2010). We conducted exploratory analysis to determine a suitable truncation distance to improve model fit of the detection functions (Buckland et al. 2001). Modeling the detection function followed a two stage modeling approach, where first a key function was selected and then a series expansion (adjustment term) was added to improve the fit of the model to the distance data (Buckland et al. 2001, Southwell 2006). We considered robust combinations of key functions and up to 3 adjustments terms following recommendations of Buckland et al. (2001). Our a priori candidate models were a half normal key function with the option of hermite adjustment terms, a uniform key function with the option of cosine or polynomial adjustments and a hazard rate key function with cosine adjustments. The best detection function was determined using Akaike’s Information Criterion with small sample size correction (AICc, Burnham and
Anderson 2002), where the model with the lowest AICc value is considered the most parsimonious (Anderson et al. 1998). We examined results from Goodness of fit tests (χ 2 GOF) and qq plots, especially at g(0) , to detect potential violations to the assumptions of distance sampling (Buckland et al. 2001). Larger moose groups might be observed further from the transect line than smaller groups (Drummer and McDonald 1987), potentially inducing a size bias which could lead to overestimation of density (Buckland et al. 2001). We used a size bias regression estimator to obtain an unbiased estimate of the expected group size ( ˗(s) ) in program DISTANCE by regressing the log of moose group size against the probability of detection at distance x (g(x) ). This method estimates ˗(s) on the transect line, where size bias should be largely negligible (Buckland et al. 2001). Thus, ˗(s) was used to estimate population density rather than the mean group size (Buckland et al. 2001). The encounter rate variance was calculated empirically treating the replicate lines as sampling unit. Density of moose within the surveyed WMU was estimated by program DISTANCE as:
ӂ" , (1) ˖ $
70 where L is the sum of all transect lengths, n denotes the number of detected moose groups and ˦ӂ(0) is the probability density function of observed perpendicular distances, evaluated at zero distance. ˦ӂ(0) is a function of 3 model components, the estimated detection probability, the encounter rate and the group size (Buckland et al. 2001). The variance of the density estimate was estimated analytically by combining the individual variance of the model components using the Delta method in program DISTANCE (Buckland et al. 2001). We also compared variance estimates from 3,000 bootstrap samples (Buckland et al. 1997) and analytical estimates to assess potential model selection uncertainty and model goodness of fit. If all three model components (i.e., the estimated detection probability, the encounter rate and the group size) were independent, bootstrapped and analytical variances should be similar (Buckland et al. 2001).
SIGHTABILITY TRIALS Moose were captured and radio collared via helicopter net gunning (Carpenter and Innes 1995) during winters 2007/2008 and 2008/2009. Net gunning protocols followed guidelines developed by ABFW (2005) and were approved by the University of Montana Animal Care and Use Protocol 056 56MHECS 010207. Seven VHF (LMRT 4; Lotek Wireless Inc., Newmarket, ON, Canada) and 14 GPS radio collars (G2000L; Advanced Telemetry Systems, Inc., Isanti, MN, USA) were used for sightability trials. Radio collars were distributed across a range of sightability conditions (Figure 3 2). Sightability surveys were conducted during February 2009 and February and March 2010 on 3 consecutive days under conditions which aerial moose surveys would be conducted in AB. Radio collared moose were first located from a fixed wing aircraft (Cessna 336/7 Skymaster) and a randomly chosen sampling block of 1.6 km by 1.6 km (1 mile by 1 mile, buffered by 300m to avoid edge effects) was projected over the moose. Sampling blocks were then surveyed in accordance to SRB surveys and distance sampling methods described above (e.g., in terms of survey speed, height AGL). Transects were spaced 400m apart, and every moose detected was recorded (regardless of whether collared) and the same covariates as for distance sampling (see above) were recorded. A missed radio collared moose was relocated immediately with radio telemetry equipment mounted to the helicopter and the same covariates as during distance sampling were recorded (Table 3 1). We discarded trials if missed moose were moving once 71 relocated with telemetry, because it was impossible to determine the initial location of a missed moose. Radio collared moose were resampled more than once, but time between survey trials was > 1 day. To reduce observer expectancy bias, we flew 5 survey blocks, which did not contain radio collared moose (i.e., “dummy plots”; Anderson and Lindzey 1996). To explore the relationship between the independent categorical covariates and the probability of detection we initially conducted univariate tests using χ2 contingency tables at an alpha level of 0.05 (Zar 1999). The effects of distance on whether a moose group was observed (1) or not (0) was examined using univariate logistic regression and non linear transformations of distance (Hosmer and Lemeshow 2000). Following univariate analyses, an a priori set of multiple logistic regression (Hosmer and Lemeshow 2000) candidate models was developed based on previous moose sightability models (Anderson and Lindzey 1996, Drummer and Aho 1998, Qualye et al. 2001). Categorical covariates were estimated using reference cell coding (Hosmer and Lemeshow 2000). We screened all candidate covariates for collinearity based on a Pearson’s correlation threshold of │r│> 0.6 and included the variable with the lowest log likelihood and smallest P value in the model (Boyce et al. 2002). We also only included significant predictor variables from univariate analysis and ignored models with insignificant variables (at an alpha level of 0.05; Arnold 2010).The logistic regression model predicting moose sightability (Y), can be written as: Y= exp U/1+ exp U, (2) where U= β 0+β 1x1+… β kxk is the linear equation of the model including the predictor covariates (x 1,…x k) influencing moose sightability (Hosmer and Lemenshow 2000). We selected the top model using AIC c as for distance sampling models above. We evaluated model fit using the Pseudo R 2, Hosmer Lemeshow’s C statistic, classification tables, the area under the receiver operating characteristic (ROC) curve (Hosmer and Lemeshow 2000). Data analysis was conducted using STATA v.10.1 (StataCorpLP, TX, USA).
CORRECTING FOR SIGHTABILITY BIAS AT g(0) Upon testing whether g(0) = 1 (see sightability trials below), and after estimating the shape of the detection function in WMU 353 with g(0) = 1, we corrected the detection
72 function for g(0) ≠1 to shift the intercept accordingly (Figure 3 1). We used the estimated g(0) for moose groups within 0 25m and its standard error, SE , estimated with the Delta method (Seber 1982) and included both as a multiplier in program DISTANCE (Buckland et al. 2004).
COMPARISON OF DISTANCE SAMPLING AND SRB SURVEYS Survey methods can be compared in terms of accuracy of the estimate (i.e., moose density), precision (90% CIs) and survey effort or cost. Accuracy of sampling methods can only be quantified when the true abundance is known, but complete census data are rarely available (Shorrocks et al. 2008) and comparisons are often made to the best available alternative method (Hounsome et al. 2005). In WMU 353, SRB surveys were the best available alternative option for comparison of distance sampling surveys, despite the fact that SRB surveys in AB are not adjusted for sightability (Lynch 1997). Unfortunately, SRB estimates were obtained in different years (2007/2008, 2008/2009) than distance sampling (2009/2010), further complicating comparison of density estimates. Therefore, we focused on comparing precision based on 90% CIs of estimates and survey effort. Because SRB surveys were not corrected for sightability, we compared precision for both unadjusted and adjusted (including our g(0) correction factor) distance sampling. We differentiated between “uncorrected” estimates for SRB surveys and “unadjusted” for distance surveys, because distance sampling methodology inherently corrects for decreased sightability with increasing distance (assuming g(0) = 1) while SRB survey estimates assume complete detection probability within 200m from the transect line (Buckland et al. 2001). We included SRB surveys conducted in WMU 353 since 2002 (winters 2002/2003, 2006/2007, 2008/2009) for our comparison. Overall, we only compared helicopter survey efforts as distance sampling did not require stratification, an additional expense of SRB designs.
RESULTS
SIMPLE RANDOM BLOCK SURVEYS Moose density estimates from SRB surveys varied between 0.28 moose/km 2 (CV = 0.25) in winter 2002/2003 and 0.51 moose/km 2 (CV = 0.17) in winter 2006/2007 (Table 3
2).The estimated moose density ( ˖ Ŵ ŷŶ Ŷź ź# from winter 2008/2009 was 73 calculated only for the surveyed portion of WMU353. The observed calf ratio per 100 cows was 38.0 in 2002/2003, 42.8 in 2007/2008 and 34.9 in 2008/2009. The bull ratio per 100 cows was 56.6 in 2002/2003, 42.0 in 2007/2008 and 41.4 in 2008/2009. Helicopter survey effort varied between 0.07 hrs/ km 2 and 0.09 hrs/ km 2 (Table 3 2).
DISTANCE SAMPLING We flew 33 transects in WMU 353 for a transect length of 777.9 km and observed 124 moose in 76 groups. Initially sighted moose in a group were bedded on 50 occasions, 24 were standing, and 2 were moving (grouped with standing). Only 6 moose were detected in steep terrain, while the remainder were found in flat areas. We detected 53 moose in open, 22 in medium and 1 in dense canopy closure. Moose groups consisted of 22 males, 76 females and 25 calves, which equals a calf ratio per 100 cows of 32.8 and a male ratio per 100 cows of 30.0. The observed group size varied between one and four moose ( =
1.606; SE = 0.089). Because the estimated expected group size of ˗$ ŵ Ÿŵŷ ( Ŵ ŴŻŷ was suggestive of size bias, we used the expected group size to estimate moose density rather than the mean group size (Buckland et al. 2001). The encounter rate of moose groups ( n/L ) was 0.09 moose/km. We selected a truncation distance (using ungrouped data) of 368m which represented the 95 th percentile of all distances recorded, corresponding to the distance at which the probabilty of detection was ~15% as recommended by Buckand et al. (2001). This removed 5 data points > 368m, leaving 71 moose groups for dection function modeling.
Based on the lowest AIC c and model fit close to the transect line, a half normal model with no adjustment terms was selected as the best detection function (Table 3 3, Figure 3 3). We observed high model selection uncertainty ( AIC < 2; Burnham and Anderson 1998) between the top model and other detection functions (Table 3 3). However, all competing models showed good fit with P values from χ2 GOF tests ; between 0.959 and 0.981 and yielded similar detection probabilities ( ˜* between 0.57 and 0.65) and density estimates (between 0.282 and 0.305) with overlapping CIs (Table 3 3, Buckland et al. 2001). To account for model selection uncertainty we model averaged by generating 3,000 bootstrap resample data sets and the model averaged ˜* was 0.61 (CV = 0.079) and the model averaged ˖ was 0.293 (CV = 0.131), supporting
74 our decision to base comparisons on only the top model. Lastly, we bootstrapped our top model using 3,000 replications and obtained bootstrap moose density estimates of 0.301 with a CV of 0.19, close to parametric estimates (Table 3 3).
SIGHTABILITY TRIALS We flew 7 sightability trials in winter 2008/2009 and 34 in winter 2009/2010, with each moose surveyed 1 to 3 times. During 41 valid sightability trials, 20 radio collared moose (51%) were missed within the 200m strips on either side of the helicopter and there was no difference in sightability by gender (χ² = 0.93, P = 0.628). Univariate analysis (Table 3 4) indicated that group size, canopy closure and terrain significantly affected sightability of moose, whereas distance, activity and light intensity were not significant. Building from these univariate relationships, the best fitting multiple logistic regression model from our candidate model set also was a function of group size, terrain, and canopy closure with all predictor covariates being significant at an alpha level of 0.05
(Table 3 5). This model was the best model by AIC c = 2.73 units, thus, we did not model average (Hosmer and Lemeshow 2000). Moose sightability decreased for single moose (Group1, β = 3.71, SE = 1.271) and increased in flat topography (Flat, β = 3.47, SE =1.385) and open canopy closure (Canopy1 (0 33%), β = 2.187, SE = 1.035). The categories Canopy2 (34 66% canopy closure) Canopy3 (67 100% canopy closure), Group2 (2 moose), Group3 (≥ 3 moose) and uneven terrain (Uneven) were subsumed into the intercept ( β0 = 1.66, SE = 1.073). The model predicted moose sightability very well according to the Hosmer and Lemeshow χ 2 statistic (χ 2 = 0.85, df = 7, P = 0.97). Classification success was high (overall 85.4% at a cut point probability of 0.5), with high classification of both detections (i.e., sensitivity = 90.5%) and missed (i.e., specificity = 80.0%). The model validated well showing ROC value of 0.93, demonstrating outstanding discrimination (Hosmer and Lemeshow 2000).
CORRECTING FOR SIGHTABILITY BIAS AT g(0) The average probability of detection within 0 25m of observations ( n = 10) on each side of the transect line was 0.62 and the standard error was 0.122 (df = 9). Using these values as a multiplier in program DISTANCE changed our density estimate for moose and the associated coefficient of variation to ˖ Ŵ Ÿ+ and CV = 0.262 (Table 3 6), an increase 75 of 39.6% and 34.6%, respectively. The correction factor at g(0) contributed the most (54.4%) towards the total variance of the density estimate. The second highest component variance was the encounter rate (36.4%), followed by the detection probability (7.4%) and finally the group size (1.8%).
COMPARISON OF DISTANCE SAMPLING AND SRB SURVEYS Density estimates from the unadjusted distance sampling survey were similar to those from 2 of the 3 SRB surveys conducted in previous years. Confidence intervals of the unadjusted distance sampling survey (CI = 0.24 0.34) overlapped density estimates from SRB surveys conducted in 2002/2003 ( ˖ Ŵ Ŷ+ ) and 2008/2009 ( ˖ Ŵ ŷŶ ; Table 3 2). The density estimate from the unadjusted distance sampling survey had a lower CV at a 90%CI (CV = 0.17) than 2 of the 3 SRB surveys conducted in previous years (Table 3 2). Density estimates of the SRB survey from 2007/2008 had the same CV of 0.170 as corrected distance sampling estimates. The CV at a 90%CI of the corrected distance sampling estimate (CV = 0.26) was similar to the CVs of density estimates from the previous year (winter 2008/2009, CV = 0.26) in a subset of WMU 353 and from winter 2002/2003 (CV = 0.25). Ratios of calves/100 cows were only slightly lower for distance sampling and SRB survey results from previous years (i.e., 2.1, 5.2 and 10 calves/100 cows for winters 2002/2003, 2007/2008 and 2008/2009 respectively), but the estimated number of bulls/100 cows differed more from SRB surveys (26.6, 12 and 11.4 bulls/100cows for winters 2002/2003, 2007/2008 and 2008/2009 respectively). Given that we did not achieve our anticipated CV of 0.2 at a 90% CI when correcting for decreased sightability at g(0) we estimated the amount of additional survey effort required to achieve this goal. We calculated the ratio of the known CV 2 and the anticipated CV 2 (CV(known) 2/ CV(anticipated) 2 = 1.69) and multiplied this estimate by the total transect length (777.9 km) of the initial distance sampling survey from winter 2009/2010, assuming the same encounter rate of moose groups (Buckland et al. 2001, Seddon et al. 2003). The resulting increase of transect line length by 536.8km would lead to an increase in required survey effort of 10.6 helicopter hours or 35% (Table 3 2). Still, the required survey effort would be below the survey efforts required for SRB surveys from previous years (Figure 3 4).
76
DISCUSSION
We present the first evaluation of distance sampling in a study area with low to moderate moose densities and heterogeneous forest cover. Estimated moose densities derived from distance sampling had comparable precision and required substantially less survey effort than traditional SRB survey methods. Thus, our results suggest that distance sampling can be an effective alternate survey method to the more expensive SRB designs under similar conditions as our study area. Furthermore, we report convincing evidence that moose are not detected with certainty at g(0) , the main assumptions of both distance sampling and SRB surveys (as conducted in AB). Detection probability of moose at g(0) was 0.691 during our study in west central AB. Hence, unadjusted distance sampling and SRB survey results are biased low and can only be treated as relative population indices (Williams et al. 2002). We addressed this bias for our distance sampling survey based on a sightability model applied to moose detections on the transect line. Even correcting for g(0) < 1, distance sampling surveys could still achieve comparable precision for a fraction of the cost. Despite the potential value of distance sampling, it has been rarely used to assess moose population size. This may be because of uncertainty about meeting the main assumptions of distance sampling (Nielson et al. 2006). We reported that in forested habitats the most critical assumption (i.e., detection probability on the transect line is 1), is likely always violated. However, we also reported that addressing this bias was possible with a correction factor derived from sightability trials. Using distance sampling without a sightability bias correction factor at g(0) would only be appropriate in open habitats (Trenkel et al. 1997). The second main assumption, that transects should be randomly distributed with respect to moose, can be easily accommodated with helicopters in the foothills. Furthermore, random allocation of transect lines in mountainous regions would be feasible using a systematic cluster sampling algorithm described by Thomas et al. (2007) or sampling elevation contour transects (Becker and Quang 2009). The third assumption, that moose do not move in response to helicopter noise, appeared to be met as our data did not reflect detection of more moose groups at further distances (Nielson et al. 2006). Lastly, we were able to ensure that perpendicular distance estimates were exact following Marques et al. (2006) with little additional flight time, because we flew off
77 transect to record sex regardless. An additional difficulty for application of distance sampling suggested by Giunchi et al. (2007) may be the statistical background required for analysis. We feel that the two comprehensive manuals on distance sampling provided by Buckland et al. (2001) and Buckland et al. (2004) are detailed resources for development of appropriate protocols for survey design and statistic methodology. Also, the user friendly interface of program DISTANCE should prevent major obstacles upon training as long as study design and data collection follow recommendations (Cassey and McArdle 1999). Our study showed that a sightability correction factor should be incorporated into aerial survey counts of moose in the boreal plains and foothills, regardless of the survey method. Our observed detection rate during sightability trials (51%) was similar to detection rates from other moose helicopter sightability models. For example, in southern interior BC sightability was 49% (Quayle et al. 2001), western Wyoming, 59% (Anderson and Lindzey 1996), or northern Ontario, 57% (Thompson 1979). Despite these similarities in sighting probability predictor covariates for sightability differed across studies. While some researchers suggested that canopy closure is the main driver of moose sightability (Anderson and Lindzey 1996; Quayle et al. 2001), the two other predictor variables (group size, terrain) included in our sightability model were also supported by previous studies. Gasaway et al. (1985) suggested that group size, activity and habitat type would be the main factors influencing moose detection rates. However, other sightability models for ungulates included vegetation cover (Samuel et al. 1987), group size (Cook and Jacobson 1979, Samuel and Pollock 1981), and terrain (LeResche and Rausch 1974). Drummer and Aho (1998) found sightability decreased when moose were bedded or moose group size was < 3 in upper Michigan. Overall, the variables included in our moose sightability model (group size, canopy closure and terrain) were supported by sightability surveys for moose and other large ungulates. Other moose sightability models correctly classified 82.7% (Anderson and Lindzey 1996) or 79% (Quayle et al. 2001) of all moose observations as missed or detected. The classification success of our model (85.4%) and the sensitivity (detected moose, 90.5%) and specificity (missed moose, 80.0%) indicated that the model was able to overall correctly classify moose sightability very well, supporting the use of our sightability model in foothills and
78 boreal plains of west central AB and east central BC. Furthermore, because we only used experienced observers and sightability survey blocks were small enough to avoid observer fatigue, we assume that perception bias likely was a small component of visibility bias. Therefore, availability bias would be the main determinant of the proportion of missed moose in our study. This suggests that double observer approaches would be insufficient to correct for visibility bias because such methods only correct for perception bias (Laake et al. 2008). Simple Random Block surveys conducted in Alberta make the assumption of complete detection probability within the 400m survey strip. Gasaway et al. (1986) observed detection probabilities between 80% and 97% for moose in Alaska. However, their surveys were conducted in predominantly open habitats, where high detection probabilities are likely. Sightability of moose may even be less than 50% when surveys are conducted in areas with higher vegetation cover (Anderson and Lindzey 1996, Quayle et al. 2001, Serrouya and Poole 2007; this study), and as such, should be considered gross underestimates of moose density. In contrast, our CVs of unadjusted distance sampling estimates compared well to the CVs of uncorrected SRB surveys. However, SRB likely underestimated variance due to un modeled detection probability heterogeneity (Laake et al. 2008). While we were not able to correct the SRB survey density estimates and CVs for detection bias post survey, we did adjust our distance sampling estimates with a sightability model at g(0) . Accounting for the sightability correction factor on the transect line in distance sampling resulted in a CV that was greater than the ABFW target of 0.2. Increasing the total transect length to achieve a CV of 0.2 would still require less total survey effort (flight time/km 2) than the average survey effort required during past moose SRB surveys in WMU 353 (Table 2 6, Figure 2 4). It is possible to improve the precision of distance sampling estimates in a multiple covariate distance sampling framework, which essentially can affect shape and scale of the detection function for different covariate values (but still assumes g(0) = 1; Buckland et al. 2004). While we considered including covariates, such as terrain or canopy closure, into our distance sampling modeling process, we achieved unstable estimates of variance due to sample size limitations (see Appendix B 2). Furthermore, pooling data over years or similar survey areas to develop a more precise global detection function for moose will improve density
79 estimates, assuming that detection probability of moose remains similar over years and strata (Buckland et al. 2001). SRB surveys have been suggested to be appropriate for smaller study areas with high visibility (Buckland et al. 2001, Wegge and Storaas 2009), but they are expensive at low densities as they require high flight effort (Ward et al. 2000, Borchers et al. 2002). In contrast, we were able to show that distance sampling can be efficient for estimating population density of relatively low density populations in large survey regions (Trenkel et al. 1997, Olson et al. 2005), but it should be noted that it is sensitive to sample size. Precision of density estimates from line transect sampling depends on the number of moose groups encountered. Thus, survey cost will increase with decreasing moose densities. A minimum sample size of 60 observations (even more when attempting to use covariates to model detection probability; see Appendix B 2) is recommended to generate population estimates with acceptable precision (Buckland et al. 2001). This may be difficult to achieve when moose densities are sparse unless survey area increases. However, based on our results moose densities would need to be below approximately 0.1 moose/km 2 for distance sampling to be more expensive than SRB surveys (following Buckland et al. 2001). If high moose densities are of concern for caribou recovery, knowing that moose densities are low (although estimates are less precise) is valuable on its own when the major management concern is caribou recovery (Alberta Woodland Caribou Recovery Team 2005) and therefore, distance sampling shows promise for use in caribou recovery due to its higher survey efficiency. Due to the required pre stratification of the survey area, SRB surveys essentially estimate density at smaller scales than line transects methodology does. Researchers can get a better idea of the spatial variability of high, medium and low units within the survey area, which can be beneficial for management planning. Density surface modeling (DSM) in program DISTANCE also allows researchers to produce similar spatial maps of moose density at an even finer scale. In addition, because distance sampling requires accurate moose locations, location data can also be used for habitat selection and subsequent survey stratification (Nielson et al. 2006). One limitation of SRB survey design is that stratification of the survey area, usually from a fixed wing aircraft, is required, adding an additional cost factor. For example, the 3 SRB surveys under
80 consideration in this study required fixed wing stratification between 0.002 hrs/km 2 (6.3 hrs to stratify 2,772 km 2) in winter 2008/2009 and 0.005 hrs/km 2 (26.6 hrs to stratify 4,580 km 2) in winter 2002/2003, adding substantial additional survey effort and costs to the SRB survey. Stratification is required just prior to the survey and delays between initiating and completing the survey (e.g., due to bad weather) may result in invalidating the stratification (Nielson et al. 2006).
MANAGEMENT RECOMMENDATIONS Aerial survey techniques used to estimate animal densities for management should be accurate, precise, cost effective and repeatable (Gasaway et al. 1986, Anderson and Lindzey 1996). We used distance sampling guidelines recommended by Buckland et al. (2001, 2004) and refined them by incorporating field techniques based on ABFW wildlife sampling protocol for moose population estimation to provide a survey method that can be repeated by management personnel in future surveys with minimal training. Distance sampling was clearly more efficient in terms of survey effort (even ignoring effort of stratification flights) and provided comparable population estimates with higher precision due to the estimation of a sightability correction factor on the transect line. However, we did not perform tests of accuracy in a controlled experimental design. While moose populations likely have been rather decreasing than increasing due to the high harvest pressure to augment caribou recovery in our study area, comparing results from different surveys methods that were not conducted during the same time is a significant problem as standardization of survey conditions is difficult to evaluate (Southwell 2006). Thus, a direct comparison of corrected SRB counts and distance sampling surveys conducted at the same time is the next logical step. Already, Alberta wildlife managers collect coarse distance data when stratification flights for SRB surveys area conducted with helicopters (i.e., usually fixed wing; M. Russell, ABFW, personal communication). With minimal additional effort, such as helicopter strut marks, precise distance data could be collected simultaneously, allowing evaluation of SRB and distance sampling methods. We further recommend testing for the validity of pooling datasets across future distance sampling surveys to increase sample sizes resulting in a more accurate global distance sampling detection probability function. Also, based on an initial analysis (Appendix B 2) multiple covariate distance sampling holds promise when larger data sets become available. Thus, 81 collecting covariates that have been shown to affect moose detection probability (i.e., terrain, group size and canopy closure in our study area) during future distance surveys and analysis of data in the multiple covariate distance sampling engine in program DISTANCE is highly commended. Lastly, the collection of additional sightability data would also contribute to higher precision of the sightability model we applied to adjust the detection function at g(0) and thereby decrease the component percentages of the variance of the density estimate. Predicting sightability of a comparable data set (e.g., data set used to devolve a sightability correction by Quayle et al. (2001), would be helpful to assess validation of the model with out of sample data.
LITERATURE CITED
Alberta Fish and Wildlife Division. 2005. Ungulate Capture by Net Gunning, Handling, and Release. Alberta Wildlife Animal Care Committe, Alberta Sustainable Resource Devlopment Edmonton, Canada. Alberta Sustainable Resource Development and Alberta Conservation Association. 2010. Status of the Woodland Caribou ( Rangifer tarandus caribou ) in Alberta: Update 2010. Alberta Sustainable Resource Development. Wildlife Status Report No. 30 (Update 2010), Edmonton, AB. Alberta Woodland Caribou Recovery Team. 2005. Alberta woodland caribou recovery plan 2004/05 2013/14. Edmonton, AB. Anderson, C. R., and F. G. Lindzey. 1996. Moose sightability model developed from helicopter surveys. Wildlife Society Bulletin 24:247 259. Anderson, D. R., K. P. Burnham, and G. C. White. 1998. Comparison of Akaike information criterion and consistent Akaike information criterion for model selection and statistical inference from capture recapture studies. Journal of Applied Statistics 25:263 282. Arnold, T. W. 2010. Uninformative Parameters and Model Selection Using Akaike's Information Criterion. Journal of Wildlife Management 74:1175 1178. Becker, E. F., and P. X. Quang. 2009. A Gamma Shaped Detection Function for Line Transect Surveys With Mark Recapture and Covariate Data. Journal of Agricultural Biological and Environmental Statistics 14:207 223.
82
Borchers, D. L., S. T. Buckland, and W. Zucchini. 2002. Estimating Animal Abundance, Closed Populations. Springer Verlag, London, Berlin, Heidelberg. Boyce, M. S., P. R. Vernier, S. E. Nielsen, and F. K. A. Schmiegelow. 2002. Evaluating resource selection functions. Ecological Modelling 157:281 300. Buckland, S. T., C. R. Anderson, K. P. Burnham, J. L. Laake, D. L. Borchers, and D. C. Thomas. 2004. Advanced distance sampling, estimating abundance of biological populations. Oxford University Press, Oxford. Buckland, S. T., D. R. Anderson, K. P. Burnham, J. L. Laake, D. L. Borchers, and L. Thomas. 2001. Introduction to Distance Sampling, Estimating abundance of biological populations. Oxford University Press, New York. Buckland, S. T., K. P. Burnham, and N. H. Augustin. 1997. Model selection: An integral part of inference. Biometrics 53:603 618. Burnham, K. P., and C. R. Anderson. 2002. Model selection and inference: A practical information theoretic approach Volume second edition.Wiley. Carpenter, L. H., and J. I. Innes. 1995. Helicopter netgunning: a successful moose capture technique Alces 31:181 184. Cassey, P., and B. H. McArdle. 1999. An assessment of distance sampling techniques for estimating animal abundance. Environmetrics 10:261 278. Caughley, G. 1974. Bias in aerial survey. Journal of Wildlife Management 38:921 933. Cook, R. D., and J. O. Jacobson. 1979. A design for estimating visibilty bias in aerial survey. Biometrics 35:735 742. COSEWIC. 2002. COSEWIC assessment and update status report on the Woodland Caribou, Rangifer tarandus caribou, in Canada. Ottawa, Ontario, Canada. Crete, M., D. Vandal, L. P. Rivest, and F. Potvin. 1991. Double counts in aerial surveys to estimate polar bear numbers during ice free priod. Arctic 44:275 278. Drummer, T. D., and W. R. Aho. 1998. A Sightability Model for Moose in Upper Michigan. Alces 34:15 19. Drummer, T. D., and L. L. McDonald. 1987. Size bias in line transect sampling. Biometrics 43:13 21. Gasaway, W. C., S. D. Dubois, and S. J. Harbo. 1985. Biases in aerial transect surveys for moose during May and June. Journal of Wildlife Management 49:777 784.
83
Gasaway, W. C., S. D. DuBois, D. J. Reed, and S. J. Harbo. 1986. Estimating moose population parameters from aerial surveys. Biological Papers of the University of Alaska, Fairbanks. Giunchi, D., V. Gaggini, and N. E. Baldaccini. 2007. Distance sampling as an effective method for monitoring feral pigeon ( Columba livia f. domestica ) urban populations. Urban Ecosystems 10:397 412. Hosmer, D. W., and S. Lemeshow, editors. 2000. Applied Logistic Regression. John Wiley and Sons, New York. Hounsome, T. D., R. P. Young, J. Davison, R. W. Yarnell, I. D. Trewby, B. T. Garnett, R. J. Delahay, and G. J. Wilson. 2005. An evaluation of distance sampling to estimate badger ( Meles meles ) abundance. Journal of Zoology 266:81 87. James, A. R. C., S. Boutin, D. M. Hebert, and A. B. Rippin. 2004. Spatial separation of caribou from moose and its relation to predation by wolves. Journal of Wildlife Management 68:799 809. Laake, J., M. J. Dawson, and J. Hone. 2008. Visibility bias in aerial survey: mark recapture, line transect or both? Wildlife Research 35:299 309. LeResche, R. E., and R. A. Rausch. 1974. Accuracy and precision of aerial moose censusing. Journal of Wildlife Management 38:175 182. Lessard, R. B., S. J. D. Martell, C. J. Walters, T. E. Essington, and J. F. Kitchell. 2005. Should ecosystem management involve active control of species abundances? Ecology and Society 10 (2). Lynch, G. 1997. Nothern moose program, moose survey field manual. Unpublished report by Wildlife Management Consulting. Lynch, G. M., and G. E. Shumaker. 1995. GPS and GIS assisted moose surveys. Alces 31:145 151. Manly, B. F. J., L. L. McDonald, and G. G. Garner. 1996. Maximum likelihood estimation for the double count method with independent observers. Journal of Agricultural, Biological, Environmental Statistics 1:170 189. Marques, T. A., M. Andersen, S. Christensen Dalsgaard, S. Belikov, A. Boltunov, O. Wiig, S. T. Buckland, and J. Aars. 2006. The use of Global Positioning Systems
84
to record distances in a helicopter line transect survey. Wildlife Society Bulletin 34:759 763. Marsh, H., and D. F. Sinclair. 1989. Correction for visibility bias in strip transect aerial surveys of aquatic fauna. Journal of Wildlife Management 53:1017 1024. McNicol, J. G., and F. F. Gilbert. 1980. Late winter use of upland cutovers by moose. Journal of Wildlife Management 44:363 371. Nielson, R. M., L. L. McDonald, and S. D. Kovach. 2006. Aerial line transect survey protocols and data analysis methods to monitor moose (Alces alces) abundance as applied on the Innoko National Wildlife Refuge, Alaska. Technical report prepared for US Fish and Wildlife Service Olson, K. A., T. K. Fuller, G. B. Schaller, D. Odonkhuu, and M. G. Murray. 2005. Estimating the population density of Mongolian gazelles Procapra gutturosa by driving long distance transects. Oryx 39:164 169. Pollock, K. H., and W. L. Kendall. 1987. Visibility bias in aerial surveys a review of estimation procedures. Journal of Wildlife Management 51:502 510. Pollock, K. H., H. D. Marsh, I. R. Lawler, and M. W. Alldredge. 2006. Estimating animal abundance in heterogeneous environments: An application to aerial surveys for Dugongs. Journal of Wildlife Management 70:255 262. Potvin, F., L. Breton, and L. P. Rivest. 2004. Aerial surveys for white tailed deer with the double count technique in Quebec: two 5 year plans completed. Wildlife Society Bulletin 32:1099 1107. Quayle, J. F., A. G. MacHutchon, and D. N. Jury. 2001. Modelling moose sightability in south central Bristish Columbia. Alces 37:43 54. Ridgway, M. S. 2010. Line transect distance sampling in aerial surveys for double crested cormorants in coastal regions of Lake Huron. Journal of Great Lakes Research 36:403 410. Rivest, L. P., H. Crépeau, and M. Crête. 1990. A two phase sampling plan for the estimation of the size of a moose population. Biometrics 46:163 176. Samuel, M. D., E. O. Garton, M. W. Schlegel, and R. G. Carson. 1987. Visibility bias in aerial surveys of elk in North central Idaho. Journal of Wildlife Management 51:622 630.
85
Samuel, M. D., and K. H. Pollock. 1981. Correction of visibilty bias in aerial surevys where animals occur in groups Journal of Wildlife Management 45:993 997. Seber, G. A. F., editor. 1982. The estimation of animal abundance and related parameters. Griffin, C.W., London, United Kingdom. Seddon, P. J., K. Ismail, M. Shobrak, S. Ostrowski, and C. Magin. 2003. A comparison of derived population estimate, mark resighting and distance sampling methods to determine the population size of a desert ungulate, the Arabian oryx. Oryx 37:286 294. Serrouya, R., and K. G. Poole. 2007. Moose population monitoring in the Lake Ravelstroke (Management Units 4 38 and 4 39) and North Thompson (MUs 3 43 and 3 44) valleys, January 2006 and 2007. Auroroa Wildlife Research. Shorrocks, B., B. Cristescu, and S. Magane. 2008. Estimating density of Kirk's dik dik (Madoqua kirkii Gunther ), impala ( Aepyceros melampus Lichtenstein ) and common zebra ( Equus burchelli Gray ) at Mpala, Laikipia District, Kenya. African Journal of Ecology 46:612 619. Smith, K. G., E. J. Ficht, D. Hobson, T. C. Sorensen, and D. Hervieux. 2000. Winter distribution of woodland caribou in relation to clear cut logging in west central Alberta. Canadian Journal of Zoology 78:1433 1440. Southwell, C. 2006. Estimation of population size and density when clusters are incomplete. in D. E. Eds, F. R. Wilson, J. D. Cole, R. Nichols, R. Rudran, and M. S. Foster, editors. Measuring and Monitoring Biological Density. Standard Methods for Mammals. Smithsonian Institution Press Washington. StataCorp. 2007. StataCorp LP, College Station, TX, USA. Thiessen, C. 2010. Horn River Basin Moose Inventory January/February 2010. BC Ministry of Environment, Fort St. John, BC, Canada Thomas, L., S. T. Buckland, E. A. Rexstad, J. L. Laake, S. Strindberg, S. L. Hedley, J. R. B. Bishop, F. F. C. Marques, and K. P. Burnham. 2010. Distance software: design and analysis of distance sampling surveys for estimating population size. Journal of Applied Ecology 47:5 14. Thomas, L., R. Williams, and D. Sandilands. 2007. Designing line transect surveys for complex survey regions. Journal of Cetacean Research and Management 9:1 13.
86
Thompson, I. D. 1979. A method of correcting population sex and age estimates from aerial surveys for moose. Alces 15:148 168. Timmermann, H. R., and M. E. Buss. 2007. Population and Harvest Management. Pages 559 616 in A. W. Franzmann, and C. C. Schwartz, editors. Ecology and Management of the North American Moose. University Press of Colorado, Boulder. Trenkel, V. M., S. T. Buckland, C. McLean, and D. A. Elston. 1997. Evaluation of aerial line transect methodology for estimating red deer (Cervus elaphus ) abundance in Scotland. Journal of Environmental Management 50:39 50. Unsworth, J. A., F. A. Leban, D. J. Leptich, E. O. Garton, and P. Zager. 1994. Aerial Survey: User's Manual. Ward, M. P., W. C. Gasaway, and M. M. Dehn. 2000. Precision of moose density estimates derived from stratification survey data. Alces 36:197 203. Wegge, P., and T. Storaas. 2009. Sampling tiger ungulate prey by the distance method: lessons learned in Bardia National Park, Nepal. Animal Conservation 12:78 84. Williams, B. K., J. D. Nichols, and M. J. Conroy. 2002. Analysis and management of animal populations. Academic Press, New York, USA. Wittmer, H. U., B. N. McLellan, D. R. Seip, J. A. Young, T. A. Kinley, G. S. Watts, and D. Hamilton. 2005. Population dynamics of the endangered mountain ecotype of woodland caribou ( Rangifer tarandus caribou ) in British Columbia, Canada. Canadian Journal of Zoology 83:407 418. Zar, J. H. 1999. Biostatistical analysis. Fouth edition. Prentice Hall Inc., NJ, USA.
87
Table 3 1. Covariates recorded during sightability surveys for detected (dependent variable = 1) and missed (dependent variable = 0) radio collared moose groups in west central Alberta and east central British Columbia, Canada, conducted in winters 2008/2009 and 2009/2010. Covariates Description Continuous Distance (m) Perpendicular distance from the transect line to the center of the group. Categorical Canopy Cover Vegetative cover that blocked the observer's view of the moose, based on the proportion of the ground hidden in a 10m diameter around the moose group. We initially used figures developed by Unsworth et al. (1991) as baseline for the estimates and then grouped Unsworth et al. (1991) 10 classes into 3: 0 33% (Canopy1), 34 66% (Canopy2), 67 100% (Canopy3). Activity Bedded, Standing, Moving ; observations of Standing and Moving moose were grouped into one category post survey due to uncertainty whether a moose was standing or moving when missed Group size All moose associated with the target animal. One (Group1), two (Group2) or more than two moose (Group3). Topography Flat, Uneven, Steep; later grouped into Flat and Uneven due to sample size limitation for Steep Light Intensity Bright or Flat; Estimated on the side of the helicopter where the moose was detected or missed.
88
2 Table 3 2. Moose population density estimates ( ˖/km ) in Wildlife Management Unit (WMU) 353 in west central Alberta, 2002 2010, obtained via helicopter based stratified random block (SRB) surveys and distance sampling. For each method and year, the density, survey area, coefficient of variation CV( ˖) at a 90% confidence interval, and survey flight effort in hours of helicopter aircraft time and hours/10km 2 are reported. The distance sampling results are given for an assumed complete probability of detection on the transect line ( g(0) = 1) as well as the estimated probability of detection on the transect line ( g(0) = 0.621); SRB results are uncorrected for sightability.
Survey Helicopter Helicopter 2 2 Method Survey Year Area (km ) ͦ CV( ͦ) hrs hrs/ 10km SRB 2002/2003 4,606.8 0.28 0.249 38.0 0.08 SRB 2006/2007 4,579.5 0.51 0.170 41.6 0.09 SRB 2008/2009 2,772.0 0.32 0.256 20.3* 0.07* Distance, 2009/2010 4,906.3 0.29 0.172 16.4 0.04 (g(0) = 1) Distance, 2009/2010 4,906.3 0.48 0.259 16.4 0.04 (g(0) = 0.621) *This moose survey was conducted within a caribou home range across boundaries of 2 WMUs and the flight hours for the portion within WMU 353 were not separated during the survey. Thus, we estimated the proportionate survey effort for the survey region within WMU353 posterior based on the survey area located within WMU353 (60%) of the total survey area. The combined survey area was 4630.2km 2 with total RW survey effort 33.9 hrs. .
89
Table 3 3. A priori candidate distance sampling detection functions used to estimate moose density in program DISTANCE 6.2 in winter 2009/2010 in Wildlife Management Unit 353 in west central Alberta, Canada. Ranking was based on the difference in
Akaike’s Information Criterion corrected for small sample sizes ( AIC c). k is the number of parameters, χ² GOF is the p value of the χ² goodness of fit test, ˜* is the estimated average detection probability and CV( ˜*) its coefficient of variation at a 90% confidence interval (CI) , ˖ is the estimated moose density for the study area and CV( ˖) is its coefficient of variation at a 90% CI. χ² GOF CV( ) CV( ) Model Key Adjustment Term k AIC c -ͷ ˜* ͦ ͦ Half normal None 1 0.00 0.974 0.59 0.110 0.30 0.172 Uniform Cosine 1 0.28 0.961 0.57 0.080 0.31 0.156 Uniform Simple Poly. 1 0.28 0.981 0.65 0.040 0.30 0.141 Uniform Cosine 2 1.52 0.969 0.65 0.160 0.28 0.211 Uniform Simple Poly. 2 1.60 0.972 0.61 0.110 0.29 0.172 Half normal Hermite Poly. 2 1.64 0.967 0.64 0.181 0.28 0.226 Half normal Cosine 2 1.80 0.959 0.64 0.190 0.28 0.234 Hazard rate Simple Poly. 2 2.10 0.931 0.68 0.090 0.27 0.165 Half normal Cosine 3 3.39 0.970 0.59 0.214 0.31 0.253 Uniform Cosine 4 3.47 0.963 0.61 0.200 0.29 0.239 Uniform Hermite Poly. 4 3.53 0.946 0.65 0.160 0.28 0.207 Half normal Hermite Poly. 3 3.76 0.971 0.63 0.186 0.29 0.231 Hazard rate Simple Poly. 3 3.99 0.916 0.67 0.120 0.27 0.180 Hazard rate Simple Poly. 4 5.30 0.881 0.67 0.160 0.28 0.213 Half normal Cosine 4 5.59 0.938 0.57 0.240 0.31 0.274 Half normal Hermite Poly. 4 No convergence achieved Notes: Poly. = Polynomial
90
Table 3 4. Mean (SE) distance of moose groups observed or missed during sightability surveys in west central Alberta and east central British Columbia, Canada, in winters 2008/2009 and 2009/2010 and numbers of groups in each group size category (one, two, > two), activity class (bedded, standing/moving), canopy closure category (low, medium, high; in intervals of 33%), terrain class (flat, uneven) and light intensity category on the side the moose was detected or missed (bright, flat). Covariate Seen Missed P Continuous Distance (m; (sd)) 72.5 (59.34) 95.3 (50.21) 0.191 Categorical Group size 1 4 15
2 11 4 >0.001
> 2 6 1
Canopy Closure Low 15 5 Medium 5 7 >0.001
High 1 8
Activity Bedded 10 13 0.262 Standing/Moving 11 7
Terrain Flat 19 10 0.004 Uneven 2 10
Light Flat 17 15 0.899 Bright 4 5
91
Table 3 5. Candidate models for predicting moose sightability in west central Alberta and east central British Columbia, Canada, using radio collared moose. Sightability surveys were conducted in winters 2008/2009 and 2009/2010. Sample size ( n), Log likelihood (LL ), number of parameters ( k) and the difference of Akaike Information Criterion values for small sample sizes from the model with the lowest value ( AIC c) for all candidate models with significant parameters at α=0.05 predicting moose detection probability. Model n LL k AICc
Group Size 1, Flat, Canopy 1 41 13.068 4 0
Flat, Canopy 2 41 15.665 3 2.732
Group Size 1, Canopy 2, Canopy 3 41 16.028 4 5.9201
Group Size 1, Group Size 2, Canopy 2, Canopy 3 41 15.992 5 8.4511
Group Size 1 41 21.570 2 12.208
Group Size 1, Group Size 2 41 21.348 3 14.098
Canopy 2, Canopy 3 41 22.537 3 16.475
Flat 41 24.088 2 17.245
92
1414
1212 Bias 1010 B 8 A 6 No. of animals detected animals of No.
4 # Detected Animals Detected # 2
0 11 2 3 3 4 4 5 6 7 8 8 Perpendicular DistanceDistance Category Category Figure 3 1. Conceptual example of line transect sampling with decreased probability of detection on the transect line ( g(0) ≠ 1). The probability of sighting an animal decreases within 8 distance classes with increasing distance from the transect line ( g(0) ). Detection function A assumes that no animals are left undetected along the transect line ( g(0) = 1). Detection function B shows the rescaled detection function if 6 out of 14 animals remained undetected along the transect line. Rescaling the detection function shifts the intercept of the function up, which corrects for the bias of decreased detection along the transect line.
93
Figure 3 2. Study area located in west central Alberta and east central British Columbia, Canada. Stratified random block (SRB) surveys to estimate moose population size were conducted in winters 2001/2002, 2006/2007 and 2008/2009 in wildlife management unit (WMU) 353 (in 2008/2009 only shaded area was surveyed). Furthermore, a distance sampling survey to estimate moose density was conducted in winter 2009/2010 in WMU 353, following a pilot survey in WMU 440 in winter 2008/2009. Lastly, sightability data were collected using radio collared moose in winters 2008/2009 and 2009/2010 to develop a sightability model.
94
1.0
0.9
GOF test 0.8 Total χ² value = 1.6514, d.f. = 6.00
Probability of a greater χ² value = 0.94881 0.7
0.6
0.5
0.4
Detection Probability Detection 0.3
0.2
0.1
0.0 0 50 100 150 200 250 300 350 400 Perpendicular Distance in Meters Figure 3 3. Estimated detection probability function of moose groups in wildlife management unit 353 in west central Alberta, Canada, modeled in program DISTANCE 6.2. Distance sampling survey data were collected in winter 2009/2010. Seventy one moose groups were detected on 33 transect lines (total of 777.8 km).The model is a half normal key function with no adjustment terms, and did not correct for g(0) < 1.
95
0.1 2 0.09 2007, CV = 0.17 2002, 0.08 CV = 0.25 2009, 0.07 CV = 0.26
0.06 Scenario, CV = 0.20 0.05 SRB HelicopterFlight in Effort hrs/ 10km 0.04 0.48Distance (2010) 0.48Distance (Scenario) 2010, CV = 0.26 0.03 0.2 0.3 0.4 0.5 0.6 2 Estimated Moose Density/km
Figure 3 4. Required helicopter flight effort in hours per 10km 2 versus the estimated moose densities per km 2 from aerial surveys using a simple random block (SRB) survey design and distance sampling within wildlife management unit 353 in Alberta, Canada. Values next to symbols indicate the year the survey was conducted and the coefficient of variation (CV) of the moose density estimate. Distance sampling density estimates were corrected for visibility bias, while SRB surveys are shown uncorrected. Thus, density estimates and CVs of SRB are likely underestimated. The distance sampling “Scenario” is an estimate of the survey time required to achieve an anticipated CV of ± 20% at a 90% CI of the distance sampling survey conducted in winter 2009/2010. Effort does not include fixed wing survey flight time required for pre stratification flights in SRB designs.
96
APPENDIX A. ADDITIONAL INFORMATION CHAPTER 2
Table A 1. Summary of all moose captured between winters 2007/2008 and 2009/2010. Table includes the collar type (Global Positioning System (GPS) or Very High Frequency (VHF) collar), the capture date and the end date or fate of the collar if applicable, the caribou range in or near which the moose has been captured (LSM = Little Smoky, RPC = Redrock Prairie Creek, NAR = Narraway, ALP = A La Peche, JNP = Jasper National Park), sex, number of GPS and VHF locations, and fix rates from individual (moose ID) moose in west central Alberta and east central British Columbia. GPS collars displayed in bold were used for moose resource selection modeling and moose and caribou resource use comparison.
Moose Collar Start End Date/ Caribou #GPS Fix #VHF ID Type Date Fate Range Sex Locations Rate Locations M01 GPS 3/11/2008 2/11/2009 LSM F° 1724 0.76 1 M02 GPS 3/12/2008 Failure RPC M 1 M03 GPS 3/11/2008 1/31/2009 LSM M 1785 0.88 4 M03 VHF 1/31/2009 Deployed LSM M M04 GPS 3/11/2008 Failure LSM F° 1 M05 GPS 3/12/2008 1/21/2009 NAR M 1832 0.94 3 M05 VHF 1/21/2009 Deployed NAR M M06 GPS 3/12/2008 Failure NAR F° 1 M07 GPS 3/12/2008 Failure RPC F° 1 M08 GPS 3/12/2008 Failure RPC F° 1 M09 GPS 3/13/2008 Failure ALP M 1 M10 GPS 3/13/2008 1/22/2009 ALP F° 1534 0.97 7 M10 VHF 1/22/2009 Deployed ALP F° M11 GPS 3/13/2008 Failure ALP M 1 M12 GPS 12/3/2008 2/20/2010 ALP F° 1427 0.77 6 M13 GPS 12/3/2008 2/20/2010 ALP F 2603 0.98 7 M14 GPS 12/3/2008 2/19/2010 ALP F° 1705 0.64 6 M15 GPS 12/3/2008 Failure ALP F 1 M16 GPS 12/4/2008 3/1/2010 RPC F 2640 0.98 4 M17 VHF 1/31/2009 Deployed LSM M 7 M18 VHF 1/31/2009 Deployed LSM M 4 M19 VHF 2/1/2009 Mortality* RPC F° 2 M20 VHF 2/1/2009 Deployed ALP M 6 M21 VHF 2/1/2009 Deployed ALP F° 8 M22 VHF 2/1/2009 Deployed LSM F° 4
97
Table A 1. continued
M23 VHF 2/1/2009 Deployed LSM F° 4 M24 GPS 3/7/2009 2/21/2010 NAR F° 2936 0.96 3 M25 GPS 3/7/2009 2/20/2010 RPC M 4003 0.99 3 M26 GPS 3/8/2009 2/19/2010 LSM F° 2052 0.98 6 M27 GPS 3/7/2009 2/19/2010 RPC M 4096 0.97 4 M28 GPS 3/7/2009 2/18/2010 NAR M 4105 0.96 M29 GPS 3/8/2009 2/20/2010 LSM M 4019 0.7 2 M30 GPS 3/7/2009 2/18/2010 NAR F° 4151 0.98 3 M31 GPS 3/8/2009 Mortality‡ JNP F° 1427 1.00 M32 GPS 3/8/2009 3/10/2010 JNP M 4083 0.97 1 M33 GPS 3/7/2009 2/19/2010 RPC M 4032 0.98 1 N/A / 3/12/2008 Mortality* RPC F N/A / 3/13/2008 Mortality* RPC M Notes: * Capture related mortality; ‡ Unknown natural mortality; ° Female moose tested for pregnancy based on blood serum progesterone levels. All tested female moose were pregnant. A small portion of female moose was not tested, because blood samples were not available.
98
Table A 2. Proportional availability of categorical landcover classes and means and standard errors (SE) of continuous covariates used to model resource selection by moose at multiple scales in the foothills and mountains of west central Alberta and east central British Columbia, Canada. Availability was defined by a buffer around summer (22 May – 16 Nov) and winter (17 Nov – 21 May) 99% fixed kernel home ranges estimated with global positioning system (GPS) collar data. Buffer size was equal to the seasonal maximum distance of GPS locations for each individual moose. GPS data were collected with 8 moose in the foothills and 9 moose in the mountains between winters of 2007/2008 and 2009/2010.
Foothill Mountain Covariate Summer Winter Summer Winter Closed conifer 59.3 60.9 39.7 44.2 Open conifer 3.6 3.3 8.0 6.7 Mixed forest 5.1 5.4 3.0 2.6 Deciduous forest 3.7 3.7 1.8 1.2 Herbaceous 1.4 1.3 1.0 1.1 Shrub 6.0 6.1 3.5 3.5 % availability Barren 1.5 1.6 0.9 1.2 landcover Muskeg 3.3 3.4 1.5 1.1 classes Water 0.2 0.2 0.6 0.6 Glacier 0.3 0.2 2.0 1.2 Young cutblock 7.4 7.6 1.2 1.2 Old cutblock 1.0 0.9 0.4 0.5 Young burn 0.4 0.5 1.6 0.9 Old burn 0.3 0.4 0.2 0.3 Alpine 6.7 4.5 34.5 33.5 Road density (km/km 2) 0.32 0.33 0.07 0.09 Road density SE 0.005 0.005 0.004 0.004 Line density (km/km 2) 1.62 1.69 0.44 0.50 Line density SE 0.016 0.016 0.009 0.010 Variable mean Elevation (m) 1263 1248 1776 1761 and standard errors (SE) Elevation SE 2.8 2.5 4.2 4.2 Slope (degrees) 7.24 6.74 16.16 15.75 Slope SE 0.084 0.079 0.125 0.123 NDVI 6954.40 NA 5811.16 NA NDVI SE 10.425 NA 23.969 NA
99
Table A 3. Mean and standard deviation of continuous covariates measured at used moose and woodland caribou global positioning system (GPS) locations during summer (22 May – 16 Nov) and winter (17 Nov – 21 May) in the foothills (FH) and mountains (MT) of west central Alberta and east central British Columbia, Canada. Data were collected with 17 moose and 17 caribou GPS collars between winters of 2007/2008 and 2009/2010.
Season / Species Statistic Elevation Road density Line density Distance % Snow NDVI Slope Region (m) (km/km 2) (km/km 2) to Open (Degrees) Summer/FH Moose mean 1249.231 0.25 1.67 248.64 22.28 7158 6.15 SD 176.36 0.36 1.25 296.25 8.50 619 5.85 Summer/FH Caribou mean 1490.92 0.09 1.27 267.46 23.17 6730 8.63 SD 242.82 0.29 1.45 379.26 12.02 1054 8.66 Summer/MT Moose mean 1665.44 0.00 0.49 124.08 26.91 6740 9.85 SD 217.89 0.04 0.62 192.47 10.42 1075 7.59 Summer/MT Caribou mean 1904.94 0.02 0.40 139.78 31.54 5449 12.85 SD 272.47 0.11 0.74 315.03 9.69 2014 9.24 Winter/FH Moose mean 1176.84 0.28 1.86 168.63 93.16 . 4.77 SD 119.89 0.47 1.43 234.15 8.00 . 5.82 Winter/FH Caribou mean 1331.38 0.14 1.65 295.31 92.03 . 5.31 SD 212.31 0.32 1.52 362.80 6.06 . 6.44 Winter/MT Moose mean 1506.12 0.02 0.64 153.27 91.82 . 8.34 SD 243.37 0.11 0.65 209.19 10.31 . 7.65 Winter/MT Caribou mean 1765.63 0.05 0.74 319.97 93.63 . 9.75 SD 308.99 0.17 1.03 408.62 7.77 . 8.18 Notes: NDVI = Normalized Difference Vegetation Index.
100
Figure A 1. Average daily travel speed (km/hr) of 17 global positioning system (GPS) collared moose (data collected between winters 2007/2008 and 2009/2010) and 212 woodland caribou in 7 herds (data collected between winters 1999/2000 and 2008/2009; N. DeCesare, University of Montana, unpublished data) in west central Alberta and east central British Columbia, Canada, to delineate seasons for resource selection function modeling. We identified the start of summer using the timing of calving as 22 May when movements of female moose and caribou decreased. The start of the winter season was identified by the end of caribou and moose fall migration, which we estimated to be approximately 17 November when movement rates decreased.
101
APPENDIX B. ADDITIONAL INFORMATION CHAPTER 3
APPENDIX B 1. CALCULATION OF THE MOOSE POPULATION ESTIMATE AND CONFIDENCE INTERVAL FOLLOWING GASAWAY ET AL . (1986)
The variance of each stratum population estimate ˢ ˠ was estimated by: 5 $ # 674 4 , ˢ ˠ ˓ 0 5 Әŵ . ә< (1) 234 4 :4 $ 2 $ 2 where ˓ is the area (km ) of stratum i, ˲3 is the average size (km ) of all SUs surveyed th th in the i stratum, > is the number of SUs surveyed in the i stratum and ˚is the total th $ number of SUs in the i stratum. The stratum sample variance ( $@ ) was estimated by:
5 5 5 $ AC BC D$E4 AC 2CBCFE4 AC 2C $@ , (2) CD# th 2 where ˳H is the number of observed moose in the j SU and ˲H is the number of km in the j th SU.
The observable population estimate ( ˠ ) and the sampling variance ˢ ˠ" for the total survey area was estimated by the sum of the population estimates for each stratum
(ˠ) and the sum of all sampling variances of the observable population estimates for each stratum ( ˢ ˠ 90% Confidence Intervals (CIs) of the total population estimate were calculated at: , (3) ˕J ˠ / ˮ NOPˢˠ where ˮ is the Student’s t statistic, α is 0.1 and ˰R are the degrees if freedom for the observable population estimate (Gasaway et al. 1986).
102
APPENDIX B 2. MULTIPLE COVARIATE DISTANCE SAMPLING
Due to sample size limitations we used conventional distance sampling with a sightability correction factor for decreased sightability on the transect line without further consideration of other covariates influencing the slope of the detection function (Buckland et al. 2004). Modeling detection probability with covariates in a multiple covariate distance sampling (MCDS) framework is recommended when a large component of the variance of the abundance estimate is due to the estimation of the detection function, and this variance can be mostly explained by variables other than distance (Buckland et al. 2004). Our sightability model indicated that ‘distance’ may be less important for predicting moose detection probability than group size, terrain and canopy closure. Thus, to decrease heterogeneity in detection probability we initially considered including predictor variables that might influence moose detection probability in our study area and attempted to model detection probability with different combinations of the covariates terrain, group size and canopy closure.
From a model selection viewpoint, AIC c values of half normal key function models with no adjustment terms (number of adjustment terms was selected based on minimum AIC c as well) with terrain (AIC c = 818.64) or group size (AIC c = 820.36) were indeed better than the AIC c of our top selected model from conventional distance sampling, a half normal model with no adjustment terms (AIC c = 821.41). The half normal models with canopy closure as a covariate had a slightly higher AIC c value (e.g. half normal with no adjustment terms and ‘canopy’ had an AIC c of 822.14). It should be noted that hazard rate models with cosine or simple polynomial adjustment terms often did not converge properly, likely due to the scale parameter inherent to the key function that half normal key functions lack (Marques et al. 2007). Thus, they were either not available for comparison or had much worse AIC c values than the aforementioned models. The top model from MCDS, a half normal key function with no adjustment terms and “terrain” as a covariate, estimated a slightly lower detection probability of 0.55 (versus 0.59 from the top CDS model) and a higher moose density estimate of 0.328 (versus 0.299 of our top model from CDS) along with a lower CV of the density estimate (0.164 for the top MCDS model versus 0.172 for the top CDS model). The model
103 decreased the proportion the detection probability adds towards the variance of the density estimate (component percentage) from 38.0% to 30.9%, thus explaining 7.1% more of the variability of the detection probability. Based on these results, it initially seemed more appropriate to use MCDS, especially since including covariates into detection probability is recommended in diverse habitats as in our study area (Buckland et al. 2001). However, we decided to model the detection function for moose for this study using only conventional Distance Sampling because the estimated CV of 0.448 obtained from bootstrap sampling was much larger than the analytical CV. Within a single model, bootstrap variance and analytical variance should be similar (Buckland et al. 2004). The amount of uncertainty present in the detection probability is influenced by the detection function fitted to the data, thus the key function and potentially adjustment and covariate terms (Buckland et al. 2004). The fact that CVs from bootstrapping and the analytical method differed so much for the top model from MCDS, while those CV estimates are approximately the same for the top model from CDS (which is the same model, just without terrain as a covariate), indicates a very high variability of the covariate terrain. However, pooling data over years or similar survey areas will improve detection functions and subsequently lead to better time specific density estimates, given that detection probability of moose will not vary substantially over years and strata (Buckland et al. 2001). Thus, each distance sampling survey conducted within similar survey areas will lead to an increase in the data set to model detection function and as sample size increases variability of covariates will likely decrease and variance estimates from MCDS will become more robust (Buckland et al. 2004). Collecting data on covariates that have been shown to influence moose sightability during our study (i.e., terrain, group size and canopy closure) during future distance sampling surveys and analysis of distance sampling data in the MCDS engine in program DISTANCE is strongly recommended to improve precision of moose density estimates.
104