Time to Event Modelling: Wolf Search Efficiency in Northern Ontario

by Scott Moffatt A Thesis Presented to The University of Guelph

In partial fulfilment of requirements for the degree of Master of Science in Integrative Biology

Guelph, Ontario, Canada Scott Moffatt, March, 2012

ABSTRACT

TIME TO EVENT MODELLING: WOLF SEARCH EFFICIENCY IN NORTHERN ONTARIO

Scott Moffatt University of Guelph

Advisors: John Fryxell and Brent Patterson

With the potentially additive anthropogenic effects of deforestation, climate change, mining, oil and gas extraction there may be rapidly changing predator-prey dynamics in Canadas boreal forest. Of particular concern is whether wolf predation is responsible for the retraction in the historical range of woodland caribou in Northern Ontario. Using time to event modelling, I determined how environmental heterogeneity in heavily forested sites of northern Ontario Canada, and animal movement behaviour, affected wolf kill success. I used Cox proportional hazard models to test several alternative hypotheses, including the wolf functional response, wolf pack characteristics, wolf satiation, landscape and other environmental features. In the proportional hazards model, an increased relative killing efficiency corresponds with a decreased time between predation events. In a comparison of top models, the predator road use hypothesis (defined by distance from road and wolf search velocity) had the most evidence. Wolves used linear features such as roads to quickly navigate their territory while targeting moose habitat near forest access roads. As the most efficient search trajectories occurred near roads, moose and woodland caribou were likely to be at significantly greater risk of predation in sites with high road densities.
I

ACKNOWLEDGMENTS
I would like to thank my committee members John Fryxell and Brent Patterson and Tom Nudds for their support and advice. Thank you to the Fryxell lab who gave me lots of feedback and fresh ideas when I needed them. A special thank you to Tal Avgar, who helped guide me in the modeling and analysis of my data and Daniel Kuefler, Laura Beaudoin and Phil Wiebe who have offered much guidance along the way. Thanks to the wolf projects field assistants including Bronwen Hennigar, Lee Kaiser, Andrew Kirby, Luke Vander Vennen and Joey James who admirably put up with all the hard work and last minute planning. I was financially supported by a National Science and Engineering Research Council (NSERC) Alexander Graham Bell Masters Scholarship, an Ontario Graduate Scholarship, and the University of Guelph. Additional project funding and logistical support was provided by Canadian Forest Service, MNR Wildlife Research and Development Division, Forest Ecosystem Science Co-operative Inc., the Center for Northern Forest Ecosystem Research. Most of all thanks to Morgan Anderson who somehow managed to only partly go insane after over two years working on the project with me. Your experience helped me become a better biologist and this thesis would not have been completed without you. On the other hand, my liver hopes to recover in your absence.

II

TABLE OF CONTENTS
INTRODUCTION ......................................................................................................................................1 . LOGGING EFFECTS...........................................................................................................................2 PREDATION........................................................................................................................................3 MODELING PREDATION EFFICIENCY IN HETEROGENOUS HABITAT.................................4 THESIS OVERVIEW AND OBJECTIVES ........................................................................................6 . METHODOLOGY.....................................................................................................................................9 STUDY AREAS...................................................................................................................................9 WOLF COLLARING.........................................................................................................................12 COLLAR PROGRAMMING.............................................................................................................12 IDENTIFICATION OF KILL SITES.................................................................................................12 DATA PREPARATION AND ANALYSIS ......................................................................................13 . CHARACTERISTICS OF THE SEARCH TRAJECTORIES...........................................................15 STATISTICAL ANALYSIS..............................................................................................................16 FRAILTY TERM ...............................................................................................................................17 . MODEL TESTING.............................................................................................................................17 RESULTS..................................................................................................................................................19 KILL SITE INVESTIGATIONS........................................................................................................19 DATA CENSORSHIP........................................................................................................................21 EFFECT OF REVISITS ON TIME TO PREDATION EVENT........................................................22 TIME TO EVENT REGRESSION ANALYSIS................................................................................25 WEIGHT OF EVIDENCE FOR MECHANISMS: TOP MODEL COMPARISON.........................30 HAZARD RATIOS OF TOP THREE MODELS ..............................................................................30 . DISCUSSION............................................................................................................................................33 INSIGNIFICANT ENVIRONMENTAL AND BEHAVIOURAL VARIABLES.............................35

III

REVISITS...........................................................................................................................................37 IMPROVEMENTS TO MODEL DEVELOPMENT.........................................................................38 INTERPRETATIONS BETWEEN CONTIGUOUS FOREST AND DISTURBED ........................38 . GENERAL CONCLUSIONS.....................................................................................................................39 LITERATURE CITED............................................................................................................................40 APPENDIX A AUDEN SITE WITH KILL SITES, ROADS AND WATER FEATURES...................47 APPENDIX B - COCHRANE SITE WITH KILL SITES, ROADS AND WATER FEATURES............48 APPENDIX C PICKLE LAKE SITE WITH KILL SITES, ROADS AND WATER FEATURES .......49 . APPENDIX D PREDATION PROFILE FOR RECKETTE LAKE PACK............................................50 APPENDIX E COX PROPORTIONAL HAZARD MODELS: MECHANISM RESULTS..................51

IV

LIST OF TABLES
Table 1. Competing mechanisms regarding the variation in the time for wolf packs to kill prey in Northern Ontario in winter 2010-2011.........................................................................................................8 Table 2. Kill site summary for three sites in winter 2010 and winter 2011. A column of probable kills is included which were prey sites not fully identified as kill sites but there was strong though inconclusive evidence suggesting they were....................................................................................................................20 Table 3. All known kills and returns in Auden and Cochrane, winter 2011. Summary stats are averages over the entire search trajectory..................................................................................................................24 Table 4. Cox Proportional Hazard Models. AIC and weights of evidence linking various mechanisms to time to kill moose in Northern Ontario.......................................................................................................31 Table 5. Hazard Ratios and their 95% upper and lower C.I. for the best model predicting the time to kill moose by 10 packs in Northern Ontario, Canada, during winter 2010-2011 .............................................32 .

LIST OF FIGURES
Figure 1. Map of the three study sites in Northern Ontario. Auden and Cochrane study sites were heavily impacted by the commercial forestry industry, and Pickle Lake was the undisturbed control site............11 Figure 2. Sample search trajectory: straight line distance between moose kills at five hour location intervals (green symbols). Map includes roads (red and black lines), waterbodies (light blue) and kills sites (ungulate icons). .................................................................................................................................14 . Figure 3. Revisits reduce the probability of kills and so increase the time to the next predation event. Time to Predation Event (hrs): search trajectories with revisits (dashed line) and without revisits (solid line)..23 Figure 4. Cumulative baseline hazard for time to predation event along a wolf search trajectory. Based on a Cox proportional hazard model with data from 54 moose kills during winter 2010-2011 in Northern Ontario, Canada..........................................................................................................................................26 Figure 5. Kaplan Meier Curves: Probability of a predation event given an environmental or behavioural variable. The time to a predation event increases with distance from waypoint locations to road and water bodies. Higher wolf velocity reduces the time to a predation event. There is higher probability of a predation event in mixed forest stands but snow depth and handling time had little effect on the time to a predation event............................................................................................................................................29

VI

INTRODUCTION
Canadas northern boreal forests have undergone considerable change due to commercial logging, mining, oil and gas extraction. This change may be having dramatic effects on populations of gray wolves (Canis lupus), moose (Alces alces), and woodland caribou (Rangifer tarandus caribou). Of particular concern in Northern Ontario is the dramatic decline in woodland caribou populations. It has been estimated that the woodland caribous range in Ontario has retracted by up to 50% since the mid 1800s (Woodland, & Caribou Recovery Team, 2005). As a result, the boreal ecotype of woodland caribou is designated as threatened by the Ontario Ministry of Natural Resources (OMNR), and the Committee on the Status of Endangered Wildlife in Canada (COSEWIC). In contrast, wolves and moose are not considered at risk either nationally or provincially in Ontario. Woodland caribou habitat is generally characterized by mature coniferous forests, peatlands (bogs and fens), and open areas with terrestrial and arboreal lichen (Fortin et al. 2008; McLoughlin, Dunford, & Boutin, 2005). Woodland caribou tend to avoid areas where logging or forest fires have changed the forest structure to an earlier succession stage (Courtois et al., 2008; Courtois et al., 2007; Schaefer & Pruitt, 1991; Schaefer, 2003). In contrast, moose are often associated with the forest edge, near young deciduous, semi-mature, and mature conifer stands (Ontario Ministry of Natural Resources, 1988). As summarized by Thomas and Gray (2002) range decline in woodland caribou may have been caused by a combination of factors, including habitat loss, human disturbance, loss of connectivity between populations or an increase in predation by gray wolves. Wolves can affect the population dynamics of woodland caribou and moose though direct predation, and indirectly through caribou behavioural response to risk of predation (Gorini et al., 2011). With regard to

indirect predation effects, risk-sensitive foraging theory suggests that prey may avoid higher quality patches in response to predator presence (Lima & Dill, 1990). Accordingly, woodland caribou may spatially segregate themselves from moose and wolf habitat as a predator avoidance strategy (Bergerud, 1992; Holt, 1977; Seip, 1992). In particular, woodland caribou often try to escape predators during the spring calving season by using small islands as safe havens (Bergerud, 1974) and forest wetland habitat (Brown et al., 1986). Moose resource selection in contrast is influenced by cover type, including possible selection for dense forest as a predator avoidance strategy (Kittle et al., 2008). North of approximately 49N latitude, estimates of wolf density are 6 to 7.5 wolves in occupied woodland caribou per 1000 km2 range (Ministry of Natural Resources, 2004:8-9). These wolf densities are close to the tolerances described for woodland caribou in the Ontario forest management guidelines (see Ontario Ministry of Natural Resources, 2001b:9), although, higher tolerances have been previously reported (see Fuller et al., 2003:167). LOGGING EFFECTS Commercial logging significantly alters the makeup of forest stands by creating landscapes of early seral stage vegetation (Edenius et al., 2002; Schaefer, 2003; Wittmer, Sinclair, & McLellan, 2005). Additionally, logging reduces the size of contiguous forest patches and increases the density of roads (Bergerud, 1981). These early successional communities typically result in population increases of ungulates such as moose and/or white tailed deer (Odocoileus virginianus) across Canadas northern boreal forest (Bowman et al., 2010; Mech & Peterson, 2003; Rempel, 1997), which in turn may be supporting higher densities of predators (Wittmer et al., 2007). This relationship may result in the phenomena referred to as apparent competition (Holt, 1977; Holt & Lawton, 1994), whereby the decline of woodland caribou is

based on an indirect interaction between caribou and a changing landscape composition. This changing landscape supports much greater densities of other prey (moose) and shared predators (Bergerud & Elliot, 1986; Seip, 1992; Wittmer et al., 2005). Rettie et al (1997) suggested caribou whose ranges overlap with moose may face predation mortality that threatens them with extirpation in the absence of positive net immigration. Accordingly, researchers have developed a large body of theory on caribou decline centered on wolf predation. PREDATION In addition to the potential impacts of apparent competition, where landscape change supports higher densities of ungulates and wolves, the heterogeneity of habitats in commercially logged areas complicates our understanding of wolf hunting behaviour and habitat use. Roads, railroads and other linear features in particular change wolf behaviour and predation efficiency. Wolves use linear features such as roads as a means of rapidly moving across rugged terrain (James & Stuart-Smith, 2000; Kunkel & Pletscher, 2000; Whittington, St. Clair, & Mercer, 2005). Bergerud (1981) theorized that as logging reduces the size of contiguous forest patches and increases the density of roads, the travel and searching efficiency of large carnivores is improved. The increase in search efficiency along roads may result in higher predation rates on moose and woodland caribou. New road development may also extend the search effort of wolves further into areas of fens/bog (widely regarded as prime caribou habitat) that they might otherwise avoid (James et al., 2004). This would then reduce the amount of de facto refugia from predation in a given landscape. The heterogeneous resource distribution in the commercially logged landscape of northern Ontario complicates the applicability of conventional behavioural ecology to landscapelevel ecological issues because it could alter optimal foraging patterns of predators, and in prey

create a trade off between predation risk and foraging gain (Lima & Dill, 1990). With the aim of managing populations both locally and over broad geographic spatial scales, researchers are taking a fresh look at classical functional response models, to simplify and understand the complex predator-prey interaction brought on by landscape change. MODELING PREDATION EFFICIENCY IN HETEROGENOUS HABITAT The predator functional response is defined as the relationship between consumption rate (i.e., number of prey eaten per predator per unit time) and prey density (Solomon, 1949). Traditional functional response modelling was based on statistical curve fitting such as with the Hollings (1959) disk equation. Holling (1959) separated the components of predation into the attack (encounter or search) rate and handling time. Handling time in mammalian carnivores is related to prey size (Fryxell et al., 2007), the number of predators in a pack (Hayes et al., 2000; Fryxell et al., 2007), digestive constraints or satiation, and loss to scavengers or other predators (MacNulty et al., 2009; Merrill et al., 2010; Zimmermann et al., 2007). The attack rate, assumed to be constant for a given prey density, is the number of prey items encountered and killed in a given search interval. The reciprocal of this is the time to kill (search time/prey item killed). It is unrealistic, however, to assume a constant search rate. The area searched per unit of time can change with landscape heterogeneity caused by patchiness in logged forest stands (Kauffman et al., 2007), animal behaviour (Bergman et al., 2006), prey spacing (Bianchi et al., 2009; Maier et al., 2005) and with anthropogenic disturbance (Kuzyk, 2002) among other factors. Similarly, prey densities will vary dramatically in heterogenous landscapes. Prey habitat selection is a function of forage variability, dependent on the availability of high quality and/or quantity of browse species (Gorini et al., 2011) compounded with predator avoidance strategies.

To understand how these many variables affect predation success, decomposing wolf movement trajectories into search and handling time allows for a more focused analysis of killing efficiency in a heterogeneous environment. Studying kill locations alone is only representative of the final predation outcome and does not take into account the different stages of a predation event from search to encounter to kill, as well as any failed attack events. Therefore, kill locations are less reliable by themselves for measuring ecological constraints on hunting success (Gorini et al., 2011) especially compared to more mechanistic models that incorporate the effect of various landscape variables over a full search path. Time to kill regression models have been recently proposed as a more effective means of analyzing how habitat heterogeneity affects killing efficiency (McPhee, Webb, & Merrill, 2012). Time to kill models, also referred to as survival or failure-time analyses; relate the time that passes before some event occurs to one or more covariates. A variety of predictor variables such as landscape factors and predator behaviour can be used to estimate search and killing efficiency in a more mechanistic manner than traditional functional response models. Separating handling time from search time allows for an assessment of how a pack moves through its environment between kill sites. The main difficulties in the past have been in separating hunting paths from kill sites and examining all ungulate kills made by a given pack for a defined period of time. Because of this, snow tracking and/or infrequent radio telemetry locations from aerial flights are inadequate techniques for estimating predation rates (Hebblewhite et al., 2009). Snow backtracking (following the trail of a wolf pack on the ground to determine where the animals crossed paths with ungulate prey) has the potential to provide information on searching, encounters, failed attempts and kill rates (e.g. Kunkel et al., 2000). However, snow backtracking cant be achieved when the terrain prevents the use of snowmobile or other equipment to quickly follow a path.

Thus, terrain that allows for backtracking may bias estimates towards easily accessible trajectories along roads or waterways. Hunting characteristics of predators can be much more accurately measured with the improvement in satellite radio collar technology (Merrill et al., 2010). Location data at frequent intervals allows for accurate offsite backtracking when verified with field data from kill sites, denning/rendezvous locations, bedding sites and any available ground backtracking (Hebblewhite & Pletscher, 2002; Zimmerman et al., 2007). Using time to event regression modelling allows for a more mechanistic approach to analyzing hunting paths (Merrill et al., 2010). THESIS OVERVIEW AND OBJECTIVES It is necessary to have a clearer understanding of how landscape changes affect wolfmoose-caribou interactions to plan for the recovery of a self sustaining population of boreal ecotype woodland caribou. The objective of my research was to determine how habitat and landscape heterogeneity, in commercially forested sites of northern Ontario Canada, affected wolf-moose-caribou interactions, using time-to-event modelling. I hypothesize that the landscape changes that have occurred from decades of commercial forestry in Northern Ontario have altered wolf foraging behaviour through increased search efficiency by increasing prey detection, habitat permeability or moose abundance. I predicted that wolves would have a reduced time to predation events in areas of higher prey moose density and early seral stage forests (mixed and deciduous forest stands vs. old growth lowland conifer), particularly in high visibility cuts and near linear corridors. I also predicted wolves would have lower kill efficiency in habitat traditionally occupied by caribou (large contiguous patches of old growth, lowland black spruce particular to bog/fen habitat).

Model testing was based on seven mechanisms with the potential to alter the time to killing prey (Table 1). The mechanisms were based on a number of traditionally studied factors such as pack characteristics (Hayes et al., 2000; Schmidt & Mech, 1997; Thurber & Peterson, 1993), prey density (Vucetich, Peterson, & Schaefer, 2002), search velocity (Holling, 1959), satiation (Arditi, Abillon, & Vieira Da Silva, 1978; Jeschke, 2007; Jost et al., 2005), and a variety of habitat and landscape features (Hopcraft, Sinclair, & Packer, 2005; Kauffman et al., 2007; Kittle et al., 2008). Finally, I present a number of management implications and further research opportunities for wolves, moose and caribou in Northern Ontario, Canada.

Table 1. Competing mechanisms regarding the variation in the time for wolf packs to kill prey in Northern Ontario in winter 2010-2011. Mechanism Functional Response Predator Road Use Prey Detection Prediction Increased prey density and/or wolf velocity decreases time to kill. Wolves disproportionately use roadways to more efficiently search their territory for prey. Wolves more easily detect and kill prey near water or road edges, regenerating forest stand (<20yrs old) and with specific terrain variables Variables Prey Density (moose/25km2) Velocity between point interval (km/hr) Distance to Road (m) Velocity (km/hr) Distance to Waterbody (m) Distance to Road (m) Regenerating Stand Elevation Eskers Slope Mixed Forest Associated with Waterbody Mixed Conifer Lowland Mixed Conifer Upland Deciduous Stand Return Handling Time (hrs) Pack ID Pack Size (4-9) Location (Study Site) Snow Depth (cm) Daily Precipitation (cm) Mean Daily Temperature(oC) Term Type

PD Continuous Velocity Continuous DistRd Continuous Velocity Continuous DistWat DistRd RS Elev Esker Slope MIX WAT MCL MCU DEC Return HT ID PS Site SD TDP MDT Continuous Continuous Discrete Continuous Factorial Continuous Factorial Factorial Factorial Factorial Factorial Factorial Continuous Discrete Discrete Factorial Continuous Continuous Continuous

Cover type

Wolves more efficiently kill prey in forest stands with greater degree of deciduous tree species.

Satiation

Wolves that make larger kills, and are more satiated, will take longer to make their next kill. The differences in pack, location of study site, gender and/or size of collared individual affects time to kill Weather conditions and/or the accumulation of snow influences time to kill

Pack Characteristics

Weather

METHODOLOGY
STUDY AREAS The study was conducted in three 22,500-km2 sites across Northern Ontario, along the current northern limit of the forestry industries area of undertaking (AOU) (Figure 1). The majority land cover of all three study sites was lowland conifer (>50%), with varying degrees of upland conifer, mixed conifer/deciduous stands and wetland complexes and to a much lesser degree upland deciduous dominant stands. The Auden and Cochrane study areas were disturbed sites below the forestry AOU and the Pickle Lake study area was a control (no commercial forestry) north of the AOU. All three study sites were comprised of typical boreal forest species, in particular black spruce (Picea mariana) and jack pine (Pinus banksiana) with stands of trembling aspen (Populus tremuloides), white birch (Betula papyrifera.) and eastern white cedar (Thuja occidentalis). Less common, though present, were tamarack (Larix laricina), white spruce (Picea glauca) and balsam fir (Abies balsamea). The Auden study site, north and east of Lake Nipigon in the district of Greenstone (Figure 1), is almost entirely within the forestry area of undertaking. After several decades of logging, the site is crisscrossed with cut roads and with variously aged forest stands. Recreational use/outfitting is common in some areas of the study site in particular the southern areas where anglers are common. Only the far north east and west corners of the site and one small section in the center (north of Onamon Lake) have not been commercially logged in the past. The Cochrane study site has ongoing commercial forestry and road development from mining/hydro industries in the southern half of the study site, but very little human disturbance in

the northern half. Being closer to the Hudsons Bay Lowlands, it contains a greater extent of fens and bogs than the other study sites particularly in the northern half. Recreational disturbances from fishing and hunting are relatively high in the southern half but rare in the north with fly-in hunting and fishing impacts only at the odd remote outfitting camp. The site has a higher degree of black spruce cover than the other sites, particularly in the northern areas where there is considerably more fen and bog coverage. All packs were collared in the southern half of the study site, with features roughly analogous to the Auden study site though prey densities in Cochrane tended to be lower than in Auden, due to the lower productivity forests. The Pickle Lake site was intended as a control site because of its very low anthropogenic disturbance. There has never been a commercial forestry industry in the area and because of the small human population, little road development. A single highway leads to the town site of Pickle Lake in the south-east corner of the site, and two winter roads access the native communities of Slate Falls and Cat Lake in the south west and north west of the site respectively. Wolf densities during 2011 were 5.2 +/- 1.1 wolves/1000km2 in Pickle Lake, 6.4 +/-0.3 wolves/1000km2 in Auden and 7.8/1000km2 in Cochrane (unpublished data, University of Guelph, 2011), based on dividing summed pack numbers by territory size. Pack size ranged between 2-9 individuals per pack during the winter of 2011. In the winter, gray wolves in eastern Canada primarily prey on ungulates, almost exclusively moose (Alces alces) but occasionally on woodland caribou (Rangifer tarandus), beaver (Castor Canadensis) and even black bear (Ursus americanus). No white tailed deer were present at any of the three study sites, but deer scat was discovered just south of the sites, suggesting a northward expansion by deer.

10

Figure 1. Map of the three study sites in Northern Ontario. Auden and Cochrane study sites are heavily impacted by the commercial forestry industry, and Pickle Lake is the undisturbed control site.

11

WOLF COLLARING In February 2010, wolves from 18 packs (nine in each of Auden and Pickle Lake sites) were captured by helicopter net gunning and were equipped with Lotek 7000MA or 7000SAW satellite radio collars (Lotek Engineering, Newmarket Ontario). In the summer of 2010, wolves were trapped and collared in both of these sites using soft-catch leg-hold traps (#7 Easy-grip traps; Livestock Protection, Alpine Texas). Wolves were trapped along remote logging roads, near known denning sites and/or along rivers and game trails at remote fly-in sites. In the fall of 2010, the Cochrane study site was added to the project design. Five packs in total were collared in Cochrane (three with satellite radio collars the other with simple VHF collars) using the ground trapping techniques described above. Two additional females that turned out to be solitary transients were also fit with VHF radiocollars. COLLAR PROGRAMMING Satellite collars were programmed to take GPS locations at either 2.5 hour or 5 hour intervals. The longer interval was used in an attempt to extend battery life. One day a month the collars were programmed to take GPS locations every fifteen minutes. GPS locations were sent via the ARGOS satellite system to Ontarios Ministry of Natural Resources (OMNR), where the locations were compiled. With each transmission from the ARGOS satellite, the locations were analyzed to identify wolf cluster sites (potential locations of a kill) using the methods of Webb et al. (2008), as described below. IDENTIFICATION OF KILL SITES Cluster sites, multiple GPS locations where wolves were congregating over a set period of time, were identified every two to three days with each ARGOS satellite upload. An individual cluster site was identified when two GPS locations were within a 150 meter radius.

12

Cluster locations were transferred onto an ftp site for the field crews to access when based out of Auden and Cochrane. When locations were deemed accessible, they were investigated on the ground by snowmobile/snowshoeing/hiking within a week of the predation event. Investigation times were generally limited by upload time, but given the added expense of a more intense schedule, the three day upload schedule was sufficient for identifying carcasses resulting from predation vs scavenging. Wolf predation was identified by signs of disarticulated carcasses, evidence of a chase sequence and/or signs of a struggle (Hayes et al., 2000). Wolves were assumed to be scavenging if the moose or caribou carcass was laying on its sternum, often frozen in its sleep (Ballard et al., 1991), human sign indicated the animal had been shot or struck by a vehicle. Scavenged prey made up sixteen percent of all visited sites with ungulate remains in the winter 2011 field season. DATA PREPARATION AND ANALYSIS Search trajectories began at the first GPS point after the collared wolf left a previous kill site. The trajectory ended at the first GPS point marking the next kill site (Figure 2). All locations between kills were included in the search trajectory including bedding/resting sites not identified by the cluster algorithm. These sites would include small non-ungulate prey items, particularly where beaver may have been present. Handling time at a kill site or at a revisited kill included the time that a wolf spent at the kill cluster (within 200m) and any time away as long as it was an excursion lasting less than 24 hours. Wolves never killed other ungulate prey during these short excursions, so handling time and search time was mutually exclusive. Generally, excursions were to nearby bedding sites outside the 200m buffer from the kill.

13

Figure 2. Search trajectory - straight line distance between moose kills at five hour location intervals (green symbols). Map includes roads (red and black lines), waterbodies (light blue) and kills sites (ungulate icons).

14

CHARACTERISTICS OF THE SEARCH TRAJECTORIES GPS locations of the ten collared packs were analysed at five hour intervals, ending either in no kill (0) or a kill (1). Prey, habitat and landscape variables were determined at each GPS location at the five hour interval. The location data was used to characterize the changes in habitat, landscape and prey density along the wolves search trajectory. In total, sixteen variables were assessed in relation to changing conditions along the search trajectories. Moose abundance data was obtained for the Auden and Cochrane areas from the Ontario Ministry of Natural Resources, and krigged density estimate layers were developed at the University of Guelph using ArcGIS 9.3 (ESRI, 2010; Anderson, 2012). Caribou density estimates were not available at the time of this research, except for the Pickle Lake site. Weather conditions (average daily temperature, average daily precipitation and snow depth) were obtained from Environment Canadas automatic weather observation stations at the town sites of Geraldton (at the south end of the Auden study site) and Cochrane (applied to the Cochrane study site) (Environment Canada, 2011). Fixed weather station data was applied to all packs within each study site. Canopy cover was assessed at the forest stand level using the Ontario Ministry of Natural Resources` Forest Resource Inventory (FRI; OMNR, 2011) . Cover included mixed conifer-lowland, mixed conifer-upland, deciduous, mixed conifer/deciduous, and wetland/waterbody. Age and average slope was also obtained at the stand level from the FRI. Digital elevation models derived from 1:50,000 National Topographic Survey map sheets (available from GeoBase, www.geobase.ca) were used to extract elevation values. ArcGIS layers for water bodies (permanent rivers, creeks, wetlands and lakes) and roads were also obtained from the Ontario Ministry of Natural Resources (OMNR, 2011). Proximity of wolf locations to water and roads was calculated using ArcGIS 9.3 based on a 50x50-m grid cell resolution
15

(Kauffman et al., 2007). Velocity along these routes, and others, was determined by the straight line distance between two consecutive GPS locations over the five hour period measured in km/hr. Pack sizes were determined over the course of the winter during backtracking and kill site investigations. Each pack was given an individual pack ID. The many revisits to previous kills presented an issue with ascertaining which factors affected kill success within a given pack. It was not possible to separate the intent of wolves to look for fresh kills versus simply returning to an old kill. To remove this effect, search trajectories with revisits were shortened to only include locations after the revisit until the next fresh kill. STATISTICAL ANALYSIS All covariates were tested for collinearity using correlation analysis prior to model entry. No pairs of variables analyzed here had Pearson product moment coefficients > 0.50. The effect of covariates on the time for a wolf pack to make a moose kill was assessed using a Cox proportional hazards model (Hosmer & Lemeshow, 1999; Therneau & Grambsch, 2000) conducted using R2.12.0 (Ross Ihaka and Robert Gentleman, 1996). The Cox proportional hazard model is a widely used regression model for survival data: (t;Zi)=0(t)ri(t), where Zi (t) is the set of explanatory variables for individual i at time t. The relative risk for subject i is: ri(t)=eZi(t) Where is a vector of parameters from the linear predictor and 0(t) is an unspecified baseline hazard function (Crawley, 2007). An increase in the relative risk of a wolf pack making a kill coincides with a decrease in the time to the predation event (Cleves et al 2008).
16

FRAILTY TERM Although some covariates describing pack characteristics were available for analysis, a frailty term was nonetheless used because many relevant characteristics were not accounted for sufficiently to avoid potential pseudoreplication of data. The frailty term (using a pack identification number) is equivalent to a mixed effects term. Pseudoreplication would occur from multiple search trajectories for different packs, each pack with unobserved heterogeneity (Hosmer, Lemeshow, & May, 2008; Therneau & Grambsch, 2000). Unobserved heterogeneity would include characteristics associated with individual wolves, such as sex, age, size, social status etc. While some attributes were sometimes known for an individual wolf, it is the unknown characteristics of the larger pack that cannot be accounted for in the statistical analysis, resulting in the need for the frailty term to explain unobserved covariates. MODEL TESTING Prior to analysis, search trajectories of revisits to previous kills were excluded from the search trajectories. A categorical variable (0, 1) was nonetheless used to determine if there was a continued effect on predation risk after wolves leave a kill or revisit a site. Two stages of model testing were conducted: the first stage determined the top models exclusive to each of the seven mechanisms (see Table 1). Akaike`s Information Criterion (AIC) was used to rank models and Akaike weights, wi, were used to assess the relative likelihood of each model in the exclusive set under consideration (Burnham & Anderson, 2002). In a second stage, the models emerging with the greatest weight of evidence for each mechanism (i.e. largest Akaike weights) were then incorporated into a final competition among the set of top models for each mechanism. Additional models were run with the reduced dataset (top model variables from the intramechanism tests) to incorporate variables between defined mechanisms. Model results are

17

reported as the combined effect of all variables. The importance of individual variables within a model were then assessed based on the coefficient being significantly different from zero at =0.05. Graphs of the variables that most affected wolf search efficiency were plotted for each mechanism. Kaplan-Meier survival curves (Hosmer et al., 2008; Ross Ihaka and Robert Gentleman, 1996) plot the change a particular variable has on the time to a kill in comparison to the baseline hazard of the search trajectories.

18

RESULTS
KILL SITE INVESTIGATIONS In the winter of 2010, approximately 92% of the 152 unique clusters sites were visited between February and the first week of April in the Auden study site. Twenty sites were visited in the Pickle Lake study site, which were accessible from the main highway in early spring. Without helicopter support, the vast majority of Pickle Lake cluster sites and twelve sites in the Auden site were not visited in 2010. Twenty six moose kills were located in Auden and four in Pickle Lake. Many of these kills occurred before the collars were deployed and were only identified when the wolves returned to the kill at a later date. Short handling time and sparse remains were also highly suggestive of a revisit for many of these kills. For the majority of winter 2010-11, sixteen packs were equipped with functional collars in three study sites (nine packs in Auden, three packs in Cochrane and four packs in Pickle Lake). Attempts were made to identify kills for all packs using similar methods to the previous season, but with the addition of helicopter support in remote locations. Sixty-nine kill sites were identified in the Auden study area, 24 in the Cochrane study site and 11 in the Pickle Lake study site (Table 2). See Appendix A-C for maps of kill locations in all three study sites in relation to roads and water features.

19

Table 2. Kill site summary for three sites in winter 2010 and winter 2011. A column of probable kills is included. These were prey sites where strong though inconclusive circumstantial evidence suggested they were kill sites. STUDY SITE AUDEN COCHRANE PICKLE LAKE TOTAL WINTER WINTER 2010 2011 26 69 N/A 5 30 24 11 104 WINTER 2011 PROBABLE 4 4 8 TOTAL SCAVENGE KILLS W2011 2 99 28 16 143 4 1 7

20

DATA CENSORSHIP The dataset from winter 2011 in Auden and Cochrane was used exclusively for this analysis because it was not possible to date many of the kills identified in 2010. In Pickle Lake, two of the four packs were located at First Nations open-pit dumpsites and did not make any known moose kills. The packs appeared to have large, defended territories and telemetry data suggested regular patrolling of the territories. But other than the occasional scavenge on human killed moose, the wolves seemed to rely exclusively on the dump sites for their nutritional needs. Unfortunately, or tellingly, the other collared individuals in these packs died early in the winter season from malnutrition. Five Pickle Lake packs were re-collared by helicopter in late March of 2011. In the last two weeks of winter, a total of eleven kills were found, mostly when wolves returned to previous kills from earlier in the winter. Accordingly, for similar reasons to the winter 2010 dataset (unknown date of kill), the Pickle Lake winter 2011 kills were also excluded from this analysis. Winter 2011 kill sites (and their respective search trajectories) from the Auden and Cochrane sites were appropriate for this analysis. There was a cumulative 69 kill sites and 2 scavenges for the 10 wolf packs. However, search trajectories were not made for all 71 sites because of errors in the ARGOS upload that left some large time gaps along some of the search trajectories of unrecovered collars (mostly for packs where the collars were unrecovered when they went offline). In Auden, a pack of 2 individuals was also excluded from the analysis because they rarely killed ungulate prey in some cases spending >600 hours near a single kill site. In addition this pair spent several days bedded down at different beaver lodges. Another pack was excluded in Cochrane because the radio collar was not uploading locations regularly and the collar was not recovered to download a reliable stream of location data. Finally, search
21

trajectories were not included for each of the first kills for 10 packs, since prior to that kill their predation history was not investigated on the ground. A total of 54 search trajectories leading to moose kills were accordingly analyzed from 10 packs in total (8 packs in Auden, 2 in Cochrane). EFFECT OF REVISITS ON TIME TO PREDATION EVENT During the winter 2011 field season in Auden and Cochrane, 43% of all kills were revisited by the same wolf pack later in the season on one of more occasions. On average an adult moose kill was revisited 1.1 times and a young of the year (YOY) moose had 0.44 revisits (Table 3). Handling time for kills averaged 145 hours in the two study sites and 28 hours during revisits. Wolves spent longer searching for prey and had a longer search distance (km) before returning to a previously killed carcass than the average time of search trajectories leading to a fresh kill. The relative risk SE of killing moose prey decreased when wolves returned to previous kills (Hazard Ratio=0.133, 0.01, P<0.001) (Figure 3).

22

Figure 3. Revisits reduce the probability of kills and so increase the time to the next predation event. Time to Predation Event (hrs): search trajectories with revisits (dashed line) and without revisits (solid line).

23

Table 3. Distance and time parameters associated with all known kills and returns in Auden and Cochrane, winter 2011. Summary stats are averages over the entire search trajectory. EVENT TYPE KILL REVISIT TOTAL EVENTS 93 76 AVERAGE TIME TO EVENT (HRS) 73 99 AVERAGE SEARCH DISTANCE (KM) 27.4 38.8 HANDLING TIME (HRS) 145 28

24

TIME TO EVENT REGRESSION ANALYSIS To analyze the time to a predation event, the Cox proportional hazard model is composed of a baseline hazard (risk) that changes over time at baseline levels of covariates, exclusive of covariate and main effect parameters that describe how the hazard varies in response to these variables. This baseline is used to determine difference in search efficiency when particular variables are analyzed in the Cox proportional hazard models. Plotting the cumulative baseline hazard indicated that the risk of killing increased moderately from 0 to 150 hours, and increased dramatically after 150 hours (Figure 4). To assess the variability in search efficiency, the analysis was conducted in two stages. In the first stage, models were run separately for variables within each of the mechanisms outlined in Table 1. In the second stage, the best models within each subset were compared against each other to determine variables with the overall greatest effect on wolf killing efficiency. Kaplan-Meier survival curves were produced to visualize the difference covariates can have on the time to a predation event (Figure 5). MECHANISMS INFLUENCING TIME TO KILL Predator road use - To determine if roadways allowed a more efficient search for prey, predator road use was assessed by distance of location waypoints to road and velocity variables. With regards to wolves road use, the model for velocity, measured as path length for every fivehour interval, and distance to road (meters) held the most weight of evidence (Appendix E Table 1). The relative risk SE of killing moose increased with wolf velocity (HR=1.17 0.05, P=0.05) and decreased with distance from road (HR=0.99 0.0001, P=0.02).

25

Figure 4. Cumulative baseline hazard for time to predation event along a wolf search trajectory. Based on a Cox proportional hazard model with data from 54 moose kills during winter 2010-2011 in Northern Ontario, Canada.

26

Prey dependent functional response - In a comparison of moose density and wolf velocity, the model for velocity alone, measured as path length for every 5 hour interval, held the most weight of evidence (See Appendix E - Table 2). The relative risk SE of killing moose prey (HR=1.09, 0.055, P=0.08) increased with wolf velocity. There was no evidence that risk changed with prey density (HR=0.992 0.091, P=0.93). Weather - The model for snow depth, measured in cm at the Geraldton and Cochrane weather stations, held the most weight of evidence (27%) (Appendix E - Table 3). However, the relative risk SE of killing moose prey did not seem to be affected by snow depth (HR=1.01 0.01, P=0.35). In addition risk was not significantly related to total daily precipitation (HR=0.93 0.08, P=0.35) or average daily temperature (HR=0.98 0.02, P=0.37). Prey detection - Prey detection was assessed with a combination of variables that may make it easier for wolves to see/find prey. Variables including distance from road and water, cutblocks, and terrain (Appendix E Table 4 for model results). Distance to road, measured as meters from any class of road (highway to overgrown forestry cut road), and held the most weight of evidence (22%). It was followed closely by distance to waterbody at 19%. The relative risk SE of killing moose prey decreased with distance from road (HR=0.99 0.001, P=0.03) and increased with distance from a waterbody (HR=1.00 0.002, P=0.240). There was no evidence that topographic variables (individually or otherwise) affected the time to kill: Elevation (HR=1.00 0.04, P=0.690), Slope (HR=0.92 0.08, P=0.300), Esker (HR=1.42 0.39, P=0.360). Likewise, regenerating stands (<20yrs old, generally clear cuts or burns) had little influence on time to kill (HR=1.15 0.56, P=0.8). Cover type - Cover type, determined from stand type using Ontario FRI, included 4 forest types and a water feature for estimating risk of predation in different forest stands (Appendix E

27

Table 5 for model tests and results). The model for mixed forest type, measured as any stand on the Ontario FRI with a roughly equal percentage of coniferous and deciduous trees, held the most weight of evidence (25%). The relative risk SE of killing moose prey increased (HR=1.64 0.31, P=0.17) when wolves searched in mixed forest stands, though not significantly. There was even less evidence that other cover types (individually or otherwise) affected the time to kill Mixed Conifer Upland (HR=0.82 0.36, P=0.58), Mixed Conifer Lowland (HR=0.93 0.31, P=0.81), Deciduous (HR=0.64 0.63, P=0.480), Water body (HR=0.80 0.50, P=0.65). Satiation - To assess the affect of satiation on wolves search efficiency, handling time and revisits were modelled (Appendix E - Table 6 for model tests and results). The model for handling time, measured as the hours spent handling the previous prey, held the most weight of evidence (57%; Table 6). The relative risk of killing moose prey increased with handling time at the previous kill (HR=1.00 0.02, P=0.056). Pack characteristics - Pack characteristics such as pack size, site location and pack ID were modelled (Appendix E Table 7 for models and results). The model for pack size, measured as the number of individuals in the pack, held the most weight of evidence (57%; Table 7). The relative risk SE of killing moose prey did not seem to be affected by pack size though (HR=1.05 0.174, P=0.77). In addition risk was little affected by site (HR=0.65 0.68, P=0.52) or pack ID (HR=0.98 0.08, P=0.10).

28

(Roads)

(Water)

(Velocity)

(Stand Type)

(Snow Depth)

(Handling Time)

Figure 5. Kaplan Meier Curves: Probability of a predation event given an environmental or behavioural variable. The time to a predation event increases with distance from waypoint locations to road and water bodies. Higher wolf velocity reduces the time to a predation event. There is higher probability of a predation event in mixed forest stands but snow depth and handling time had little effect on the time to a predation event.

29

 WEIGHT OF EVIDENCE FOR MECHANISMS: TOP MODEL COMPARISON

The leading candidate variables affecting wolf search efficiency were compared in a second stage of analysis to assess effects of risk of a predation event across the proposed mechanisms (Table 4). In a comparison of the top models related to each mechanism, the Predator Road Use model (distance from road and wolf search velocity) had the greatest weight of evidence (17%). The two next best models were nearly as plausible, with nearly as high Akaike weights (15% each). These were the landscape model and the Predator Road Use model combined with Snow Depth (Table 4). HAZARD RATIOS OF TOP THREE MODELS For the top model, predator road use, the relative risk of killing moose prey increased with wolf velocity (HR=1.17 0.05, P=0.05) and decreased with distance from road (HR=0.99 0.0001, P=0.02; Table 5). The landscape model was a combination of the top landscape specific variables from the first stage of analysis. The relative risk of killing moose prey increased with proximity to roads (HR=0.99 0.0001, P=0.01) but did not significantly increase with snow depth (HR=1.01 0.01, P=0.30) or in mixed forest stands (HR=1.62 0.36, P=0.18). Adding snow depth to predator road use didnt change relative risks of killing moose. Risk still increased with close proximity to roads (HR=0.99 0.0001, P=0.01) but did not significantly increase with snow depth (HR=1.01 0.01, P=0.17) and increased with wolf search velocity (HR=1.13 0.06, P=0.05).

30

Table 4. Cox Proportional Hazard Models. AIC and Weights of evidence linking various mechanisms to time to kill moose in Northern Ontario MODEL Predator road use Predator Road Use, Weather Landscape Cover type Landscape, velocity Functional response Prey detection Weather Satiation Pack characteristics Landscape, prey density Landscape, prey density, velocity Full model Null model Pack dynamics
1 2

VARIABLES DISTRD1, SR2 DISTRD, SR, SD3 MIX4, DISTRD, SD MIX MIX, DISTRD, SD, SR SR DISTRD SD HT5 PS6 MIX, DISTRD, SD, PD7 MIX, DISTRD, SD, PD, SR ALL TOP VARIABLES HT,PS

K 4 5 5 3 6 3 3 3 3 3 6 7 8 3 4

AICC 0 0.22 0.22 0.67 1.36 1.61 1.41 2.15 3.03 3.41 4.2 6.45 9.35 21.91 7.16

WI 0.17 0.15 0.15 0.13 0.09 0.08 0.08 0.06 0.04 0.03 0.02 0.01 0 0 0

DISTRD Distance from waypoint location to road (m) SR Velocity of wolf pack between consecutive waypoints(km/h) 3 SD Snow Depth (cm) 4 MIX mixed deciduous and coniferous forest stand 5 HT Handling Time of prey item (hrs) 6 PS Number of individual wolves per pack 7 PD Prey Density (moose/25km2)

31

Table 5. Hazard Ratios and their 95% upper and lower C.I. for the best model predicting the time to kill moose by 10 packs in Northern Ontario, Canada. Model Predator Road Use Variable Velocity (km/hr) Distance to Road (m) Landscape Mixed Forest Stand Snow Depth (cm) Distance to Road (m) Predator Road Use, Weather Velocity (km/hr) Distance to Road (m) Snow Depth (cm) HR 1.17 0.99 1.62 1.01 0.99 1.13 0.99 1.01 LOWER UPPER P 0.999 0.999 0.802 0.993 0.999 0.999 0.999 0.996 1.26 0.99 3.29 1.02 0.99 1.29 0.999 1.02 0.05 0.02 0.18 0.30 0.01 0.05 0.01 0.17 EFFECT + + + + +

32

DISCUSSION
Roads, high movement velocity, and to a lesser degree mixed forest stand types and snow depth, all predicted the rate at which wolves killed moose. Other pack characteristics, prey density and environmental variables such as terrain ruggedness, other canopy types, water features and weather had much less impact on the time between predation events. These results suggest that wolves favour roads disproportionately to more efficiently search their territory for prey. This contradicts other studies suggesting that wolves are less abundant in areas with high road density and human activity (Jensen, Fuller, & Robinson, 1986; Mech et al., 1988; Thiel, 1985). In these studies, wolf predation sites were not significantly closer to road and trails than were wolf locations during search or random points (Atwood, Gese, & Kunkel, 2009; James et al., 2009). In fact, predation sites were farther away from the roads than expected by randomness (Kunkel & Pletscher, 2000; McLoughlin et al., 2004). However, roads and other linear corridors that receive little human use may be attractive to wolves as easy travel routes (Thurber et al., 1994). Use of linear features such as roads by wolves to quickly and efficiently navigate their territory is well documented (Bergerud, 1988; James, 1999; Mcloughlin et al., 2009; Whittington et al., 2005). The disparity between the results of my research and the other cited work may result either from a difference in analyses, or site-specific differences in Ontarios northern boreal forest that influence wolf space use. In regards to the later, a common link between these other studies analyzing kill sites is that they have been conducted in areas with high human density and road use, particularly southern areas of wolf range or heavily impacted northern sites such as in Albertas oil and gas region. The study sites analyzed for this project on the other hand were

33

much further north in the interior boreal forest and did not have the same anthropogenic footprint as southern populations. While in the past these study sites in northern Ontario were heavily used for forestry, with the decline of Ontarios commercial forestry industry, there is now minimal human traffic on most roads in our study areas. This is particularly true in the winter season when the majority of the pack territories within the study sites were largely inaccessible to human traffic. Without anthropogenic disturbance, the wolves were at liberty to use the roads (both tertiary cut roads and large unmaintained primary roads). Another reason for the differences in results between studies may derive from potential differences between analyses based on time to event modelling versus those based on resource selection functions. With time to event modelling, the search trajectory of a wolfs movement through their environment towards a kill is incorporated into the model. Resource selection functions compare locations to kill sites while ignoring the search process and differential habitat use that may occur during prey search (Fritts & Mech, 1981; Kunkel & Pletscher, 2000). In addition to increased search efficiency near roads, movement behaviour (velocity) also affected the time between predation events. Increased velocity by wolves resulted in a reduced time between predation events. Mixed forest stands, containing both conifer and deciduous trees species, also reduced the time for wolves to make a moose kill. Moose are known to browse on balsam fir, trembling aspen, white birch and other species of shrub such as red osier dogwood (Cornus sericea) and willow species (Salix spp.; Van Beest et al., 2010). Mixed forests generally contain the highest diversity of these species, much more so than the large tracts of lowland black spruce - the predominant stand type in all our study sites. This suggests that wolves preferentially search within stand types that have the highest probability of harbouring moose. The lack of

34

autocorrelation between the moose density and mixed forest type variables here likely results from the coarse scale of the prey density maps. Forestry tends to create a patchwork of successional stages in regenerating forest stands (Weir & Johnson, 1998). An increase in edge habitat via logging can increase species richness (Haeussler et al., 2002) and density of moose in those areas (Rempel, 1997). Studies in Quebecs boreal forest found that over seventy years of commercial logging resulted in the conversion of 37% of the landscape from coniferous to mixed forest and 19% from transitional to deciduous stands (Boucher, Arseneault, & Sirois, 2006). These secondary mixed forest stands are also more likely to be closer to roads, allowing wolves to quickly move quickly across the landscape (Frair et al., 2008; Rempel, 1997). INSIGNIFICANT ENVIRONMENTAL AND BEHAVIOURAL VARIABLES Many environmental factors have been widely cited as impacting predation rates such as prey density, snow depth, pack size, and satiation, but these had mild to no significant effects on the time between predation events. For example, while wolf population density at a regional scale has been linked to the availability of prey in North America (Fuller & Murray, 1998), local prey density was not found to have a significant effect on the time to kill in this study. A companion project to this study also found that wolves in these study sites did not select for areas of high moose density within the core areas of their home ranges (Anderson, 2012). The available data on moose spatial distribution may be at too coarse of a scale, however, to be useful for accurately interpreting density at a scale that would be appropriate for assessing if wolves are truly altering search patterns towards prey dense areas. Researchers have been looking into the effect of snow depth on predation success for years. They have consistently found higher kills rates in years with high snow accumulation (Carbyn, 1983; Nelson & Mech, 1986). Other studies have found a significant effect of the North
35

Atlantic Oscillation (NAO) on predation rates, with high snow rates resulting in higher predation rates (Fuller, 1991; Hebblewhite, 2005; Huggard, 1993). In this study, snow depth had a weak, positive effect on the time to a predation event. As my analysis did not compare multiple years though, I cannot adequately test for multi-year climactic effects. It was surprising, however, not to see a more significant difference in snow depth that could be attributed to the time of year. The study started in November with low snow accumulation and ended in April with a meter of snow pack. But perhaps during the winter of 2010-2011 the wolves did not have as significant an advantage over ungulate prey as snow hardness and/or density were not sufficient to prevent wolves from sinking deeply in the snow (unpublished data, Canadian Forestry Service). Future research will examine snow depth, density and hardness in relation to habitat type across all study sites, but these data were unavailable at the time of my study. The only snow depth data I had available, from Environment Canada weather stations, were not appropriate for making fine scale interpretations within the search trajectories. Wolves use frozen lakes and rivers to move quickly across their territory, rather than extensively searching for prey in one particular subsection (Burkholder, 1959; Jedrzejewski et al., 2001). However, similar to snow depth, risk of predation only mildly increased with distance from water corridors and was not significant. Pack size also did not significantly influence the time between predation events, even though it is commonly cited as increasing overall predation rates (Hayes et al., 2000). While the study by Hayes et al. in the Yukon had greater variability in pack size, other studies suggest that pack size does not result in wolves taking down more and bigger prey items (Thurber & Peterson, 1993). However, further research has found kill rates and type of prey attacked may have less to do with pack sizes and more to do with individual variability in the age and composition of the breeding adults large older males and packs with

36

both alphas surviving make more successful kills on large prey (Sand et al., 2006). This has been further corroborated by MacNaulty et al. (2011) using the Yellowstone National Park wolf dataset. Here they showed that the relationship between wolf kill success and pack size is nonlinear and that an individuals withholding effort (free riding) may explain why success does not increase across large group sizes. REVISITS Wolves regularly revisited previous kill sites in all three study sites over the two winter field seasons. The broad utilization among packs (every pack revisited kills on multiple occasions) suggests this is a common behavioural characteristic, perhaps due to food limitation. Wolves revisited 43% of previous kills a total of 76 times and on more than half the search trajectories used in this analysis. The regularity of revisited kill sites throughout the winter often months later suggested this behaviour will have important implications on search effort, predation rates and functional response models. Wolves have been found to make more revisits to old kills when there are deep snow conditions (Fuller, 1991; Kolenosky, 1972) such as the year of this study. Its likely that wolves returned to previous kills when they were unsuccessful at making another kill. Excluding handling times at the re-visited kill sites still resulted in an extended time to the next predation event. The sheer number of returns suggests that revisiting previous kills is an important energetic strategy to boost metabolism and maintain energy stores until a new kill can be made. The energetic boost may have been limited though, because the carcasses being returned to tended to be mostly consumed except for leg bones and in large adult moose, skull and spinal column.

37

IMPROVEMENTS TO MODEL DEVELOPMENT Further improving empirically based movement modelling will require either greater field effort or adjustment in radio collar technology. Either would facilitate development of a more mechanistic model of search efficiency by creating more accurate search paths. Excluding bedding sites or incorporating this behaviour into this analysis may affect the significance of some explanatory variables, by introducing or eliminating potential biases. Finding bedding sites can be done effectively with a combination of field effort, and/or altering the GPS location fix interval. Changing the fix interval to take more frequent locations (at the expense of collar battery life) while using accelerometer data and a time threshold could accurately determine resting sites. Using a shorter fix interval uses more battery power but with the low life expectancy of collars on wolves (from high levels of wear and tear) more data will be collected. Once accelerometers have been calibrated to behaviours, less intense field work could still effectively estimate predation rates. Accelerometers could allow researchers to assess how much movement the collars are experiencing as an animal moves through its environment. When the animal is running or eating a prey item, the increase in collar movement results in a higher activity reading. Conversely, when an animal is motionless, such as when its bedded down, the activity readings are low or even zero (Juarez et al., 2010). With a shorter fix schedule on the GPS collars GPS locations recorded hourly or quarter hourly a cluster analysis similar to the one used to find the kill sites could be used with a shorter time threshold to determine bedding sites. Eliminating these bedding sites may improve model outcomes. INTERPRETATIONS BETWEEN CONTIGUOUS FOREST AND DISTURBED To fully understand how wolf search behaviour is altered by logging and road creation, it is necessary to have more information on patterns in the areas north of the forestry AOU.

38

Emphasis should be placed on the Pickle Lake control site to ascertain baseline predation rates on moose and caribou in this relatively undisturbed site. A comparison between sites, particularly the Auden and Pickle Lake site which are geographically close and have similar habitat features, would help to fully assess the impact of the commercial forestry industry on wolf and ungulate interactions in northern Ontario. Telemetry data collected will aid with modelling movements of wolves and the two main species of ungulate prey (moose and caribou). However, with regards to predation, not enough emphasis was placed on ensuring packs were collared at the appropriate time of year for winter field work. Without baseline predation rates in Pickle Lake managers will not be able to compare heavily impacted and un-impacted sites.

GENERAL CONCLUSIONS
To determine how landscape changes have affected wolf-moose interactions, this observational study, including sites with high and lower degree of landscape changes, found that wolves reduced the time between predation rates by being closer to roads and trails, and with increased wolf velocity. To a lesser degree, mixed forest stand types and snow depth also affected wolf killing efficiency. It is clear that in habitat heavily impacted by commercial logging, wolves are using the road network to quickly move through their territories and encounter and kill moose prey. Quantifying the source of variation in wolf search/killing efficiency is an important challenge in linking the impacts of predators to their prey, particularly in heavily impacted sites south of the forestry area of undertaking in Northern Ontario. This higher killing efficiency of wolves near roads and in early seral stage forests will have direct effect on woodland caribou populations. This issue will need to be addressed by wildlife managers when developing woodland caribou conservation strategy in Ontarios north.

39

LITERATURE CITED
Anderson, M. 2012. Wolf responses to spatial variation in moose density in northern ontario. Unpublished masters thesis, University of Guelph, Guelph, Canada. Arditi, R., Abillon, J. M., & Vieira Da Silva, J. 1978. A predator-prey model with satiation and intraspecific competition. Ecological Modelling, 5(3), 173-191. Atwood, T. C., Gese, E. M., & Kunkel, K. E. 2009. Spatial Partitioning of Predation Risk in a Multiple PredatorMultiple Prey System. Journal of Wildlife Management, 73(6), 876-884. Ballard, W. B., Ayres, L. A., Roney, K. E., Spraker, T. H., The, S., Management, W., & Jan, N. 1991. Immobilization of Gray Wolves with a Combination of Tiletamine Hydrochloride and Zolazepam Hydrochloride. Journal of Wildlife Management, 55(1), 71-74. Bergerud, A. T., & Elliot, J. P. 1986. Dynamics of caribou and wolves in northern British Columbia. Canadian Journal of Zoology, 64(7), 1515-1529. Bergerud, A T. (1988). Caribou, wolves and man. Trends in Ecology & Evolution, 3(3), 68-72. Bergerud, A T. 1974. Decline of caribou in North America following settlement. The Journal of Wildlife Management, 38(4), 757770. Bergerud, A T. 1992. Rareness as an antipredator strategy to reduce predation risk for moose and caribou. In D. R. McCullough & R. H. Barrett (Eds.), Wildlife 2001 populations. 10081021. Bergerud, A.T. 1981. The decline of moose in Ontarioa different view. Alces, 17, 3043. Bergman, E. J., Garrott, R. A., Creel, S., Borkowski, J. J., Jaffe, R., & Watson, E. G. R. 2006. Assessment of prey vulnerability through analysis of wolf movements and kill sites. Ecological Applications, 16(1), 273-284. Bianchi, F. J. J. A., Schellhorn, N. A., & Van Der Werf, W. 2009. Foraging behaviour of predators in heterogeneous landscapes: the role of perceptual ability and diet breadth. Oikos, 118(9), 1363-1372. Boucher, Y., Arseneault, D., & Sirois, L. 2006. Logging-induced change (1930 2002) of a preindustrial landscape at the northern range limit of northern hardwoods , eastern Canada. Canadian Journal of Forest Research, 517(2), 505-517 Bowman, J., Ray, J. C., Magoun, A. J., Johnson, D. S., & Dawson, F. N. 2010. Roads, logging, and the large-mammal community of an eastern Canadian boreal forest. Canadian Journal of Zoology, 88(5), 454-467. Brown, W. K., Huot, J., Lamothe, P., Luttich, S., Par, M., Martin, G. S., & Theberge, J. B. 1986. The distribution and movement patterns of four woodland caribou herds in Quebec and Labrador. Rangifer, 6(2), 43-49.

40

Burkholder, B. L. 1959. Movements and behavior of a wolf pack in Alaska. Journal of Wildlife Management, 23(1), 1-11. Burnham, K. P., & Anderson, D. R. 2002. Model selection and multimodel inference: a practical information-theoretic approach. Springer. Carbyn, L. N. 1983. Wolf predation on elk in Riding Mountain National Park, Manitoba. Journal of Wildlife Management, 47(4), 963-976. Courtois, R., Gingras, A., Fortin, D., Sebbane, A., Rochette, B., & Breton, L. 2008. Demographic and behavioural response of woodland caribou to forest harvesting. Canadian Journal of Forest Research, 38(11), 2837-2849. Courtois, R., Ouellet, J.-P., Breton, L., Gingras, A., & Dussault, C. 2007. Effects of forest disturbance on density, space use, and mortality of woodland caribou. Ecoscience, 14(4), 491-498. Crawley, M. J. 2007. The R Book. (J. W. S. Ltd, Ed.) (Vol. 150, p. 953). Edenius, L., Bergman, M., Ericsson, G., & Danell, K. 2002. The role of moose as a disturbance factor in managed boreal forests. Alces, 36(1), 57-67. Fortin, D., Courtois, R., Etcheverry, P., Dussault, C., & Gingras, A. 2008. Winter selection of landscapes by woodland caribou: behavioural response to geographical gradients in habitat attributes. Journal of Applied Ecology, 45(5), 1392-1400. Frair, J. L, Merrill, E. H., Beyer, H. L., & Morales, J. 2008. Thresholds in landscape connectivity and mortality risks in response to growing road networks. Journal of Applied Ecology, 45(5), 1504-1513. Fritts, S. H., & Mech, L. D. 1981. Dynamics, movements and feeding ecology of a newly protected wolf population in northwestern Minnesota. Wildlife Monographs, 80(80). Fuller, T K. 1991. Effect of snow depth on wolf activity and prey selection in north central Minnesota. Canadian Journal of Zoology, 69(2), 283-287. Fuller, T K, & Murray, D. L. 1998. Biological and logistical explanations of variation in wolf population density. Animal Conservation, 1(3), 153-157. Gorini, L., Linnell, J. D. C., May, R., Panzacchi, M., Boitani, L., Odden, M., & Nilsen, E. B. 2011. Habitat heterogeneity and mammalian predator-prey interactions. Mammal Review, 42(2). Gustine, D. D., Resources, N., Studies, E., British, N., George, P., Vn, B. C., Columbia, B., et al. 2006. Calf Survival of Woodland Caribou in a Multi-Predator Ecosystem. Wildlife Monographs, 165, 1-32. Haeussler, S., Bedford, L., Leduc, A., Bergeron, Y., Kranabetter, J. M., & Silvicultural, J. M. 2002. Silvicultural disturbance severity and plant communities of the southern Canadian boreal forest. Silva Fennica, 36, 307-327.

41

Hayes, R. D., Baer, A. M., Wotschikowsky, U., & Harestad, A. S. 2000. Kill rate by wolves on moose in the Yukon. Canadian Journal of Zoology, 78(1), 49-59. Hebblewhite, M. 2005. Predation by wolves interacts with the North Pacific Oscillation (NPO) on a western North American elk population. Journal of Animal Ecology, 74(2), 226-233. Hebblewhite, M. Pletscher, D. H. 2002. Effects of elk group size on predation by wolves. Canadian Journal of Zoology, 80(5), 800-809. Hebblewhite, M., Paquet, P. C., Pletscher, D. H., Lessard, R. B., & Callaghan, C. J. 2009. Development and application of a ratio estimator to estimate wolf kill rates and variance in a multiple-prey system. Wildlife Society Bulletin, 31(4), 933-946. Holling, C. S. 1959. The components of predation as revealed by a study of small mammal predation of the European pine sawfly. Canadian Entomologist, 91(5), 293-320. Holt, R. D. 1977. Predation, apparent competition, and the structure of prey communities. Theoretical Population Biology, 12(2), 197-129. Holt, R. D., & Lawton, J. H. (1994). The ecological consequences of shared natural enemies. Annual Review of Ecology and Systematics. 25, 495-520. Hopcraft, J. G. C., Sinclair, A. R. E., & Packer, C. 2005. Planning for success: Serengeti lions seek prey accessibility rather than abundance. Journal of Animal Ecology, 74(3), 559-566. Hosmer, D. W., Lemeshow, S. 1999. Applied survival analysis : regression modeling of time to event data. New York, New York, USA: A Wiley-Interscience Publication, John Wiley & Sons Inc. Huggard, D. J. 1993. Effect of snow depth on predation and scavenging by gray wolves. Journal of Wildlife Management, 57(2), 382-388. James, A. R. C., Boutin, S., Hebert, D. M., Rippin, A. B., & White, J. 2004. Spatial separation of caribou from moose and its relation to predation by wolves. Journal of Wildlife Management, 68(4), 799809. James, A. R. C. & Stuart-Smith, A. K. 2000. Distribution of caribou and wolves in relation to linear corridors. The Journal of Wildlife Management, 64(1), 154-159. James, A. R. C. 1999. Effects of industrial development on the predator-prey relationship between wolves and caribou in northeastern Alberta. Doctor. University of Alberta. James, ARC, Stuart-Smith, A. K. 2009. Distribution of caribou and wolves in relation to linear corridors. The Journal of Wildlife Management, 64(1), 154-159. Jedrzejewski, W., Schmidt, K., Theuerkauf, J., Jedrzejewska, B., & Okarma, H. 2001. Daily movements and territory use by radio-collared wolves (Canis lupus) in Bialowieza Primeval Forest in Poland. Canadian Journal of Zoology, 79(11), 1993-2004. Jensen, W. F., Fuller, T. K., & Robinson, W. L. 1986. Wolf, Canis lupus, distribution on the Ontario-Michigan border near Soult Ste. Marie. Canadian Field Naturalist, 100, 363366.
42

Jeschke, J. M. 2007. When carnivores are full and lazy. Oecologia, 152(2), 357-364. Jost, C., Devulder, G., Vucetich, J. A., Peterson, R. O., & Arditi, R. 2005. The wolves of Isle Royale display scale-invariant satiation and ratio-dependent predation on moose. Journal of Animal Ecology, 74(5), 809-816. Juarez, C. P., Rotundo, M. A., Berg, W., & Fernndez-Duque, E. 2010. Costs and benefits of radio-collaring on the behavior, demography, and conservation of owl monkeys (Aotus azarai) in Formosa, Argentina. International Journal of Primatology, 5(1), 69-82. Kauffman, M. J., Varley, N., Smith, D. W., Stahler, D. R., MacNulty, D. R., & Boyce, M. S. (2007). Landscape heterogeneity shapes predation in a newly restored predator-prey system. Ecology Letters, 10(8), 690-700. Kittle, A. M., Fryxell, J. M., Desy, G. E., & Hamr, J. 2008. The scale-dependent impact of wolf predation risk on resource selection by three sympatric ungulates. Oecologia, 157(1), 16375. Kolenosky, G. B. 1972. Wolf predation on wintering deer in east-central Ontario. The Journal of Wildlife Management, 36(2), 357-369. Kunkel, K. E., & Pletscher, D. H. 2000. Habitat factors affecting vulnerability of moose to predation by wolves in southeastern British Columbia. Canadian Journal of Zoology, 78(1), 150-157. Kuzyk, G. 2002. Wolf distribution and movements on caribou ranges in west-central Alberta. Unpublished masters thesis, University of Alberta, Edmonton, Alberta. Lima, S. L., & Dill, L. M. 1990. Behavioral decisions made under the risk of predation: a review and prospectus. Canadian Journal of Zoology, 68(4), 619-640. MacNulty, D R, Smith, D. W., Mech, L. D., Vucetich, J. A., & Packer, C. 2011. Nonlinear effects of group size on the success of wolves hunting elk. Behavioral Ecology, 23(September), 75-82. MacNulty, D R, Smith, D. W., Mech, L. D., & Eberly, L. E. 2009. Body size and predatory performance in wolves: is bigger better? Journal of Animal Ecology, 78(3), 532-539. Maier, J. A. K., Hoef, J. M. V., Mcguire, A. D., Bowyer, R. T., Saperstein, L., & Maier, H. A. 2005. Distribution and density of moose in relation to landscape characteristics : effects of scale 1. Canadian Journal of Forest Research, 2243(9), 2233-2243. McLoughlin, P. D., Dunford, J. S., & Boutin, S. 2005. Relating predation mortality to broadscale habitat selection. Journal of Animal Ecology, 74(4), 701-707. McLoughlin, P. D., Walton, L. R., Cluff, H. D., Paquet, P. C., & Ramsay, M. A. 2004. Hierarchical habitat selection by tundra wolves. Journal of Mammalogy, 85(3), 576-580. McLoughlin, P. D., Dzus, E., Wynes, B., Boutin, S., & Press, A. 2009. Declines in Populations of Woodland Caribou. Journal of Wildlife Management, 67(4), 755-761.

43

Mcphee, H. M., Webb, N. F., & Merrill, E. H. 2012. Hierarchical predation : wolf (Canis lupus ) selection along hunt paths and at kill sites. Canadian Journal of Zoology, 90, 555-563. McPhee, H. M., Webb, N. F., & Merrill, E. H. 2012. Time-to-kill: measuring attack rates in a heterogenous landscape with multiple prey types. Oikos, 121(5), 711-720. Mech, L D, Fritts, S. H., Radde, G. L., & Paul, W. J. 1988. Wolf distribution and road density in Minnesota. Wildlife Society Bulletin, 16(1), 85-87. Mech, L D, & Peterson, R. O. 2003. Wolf-prey relations. wolves behavior ecology and conservation. The University of Chicago Press (pp. 131-160). Merrill, E., Sand, H. ., Zimmermann, B., McPhee, H., Webb, N., Hebblewhite, M., Wabakken, P., et al. 2010. Building a mechanistic understanding of predation with GPS-based movement data. Philosophical Transactions of the Royal Society B Biological Sciences, 365, 2279-2288. Nelson, M. E., & Mech, L. D. 1986. Relationship between snow depth and gray wolf predation on white- tailed deer. Journal of Wildlife Management, 50(3), 471-474. Ontario Ministry of Natural Resources. 1988. Timber Management Guide for the Provision of Moose Habitat. Palacios, V., & Mech, L. D. 2010. Problems with studying wolf predation on small prey in summer via global positioning system collars. European Journal of Wildlife Research, 57(1), 149-156. Rempel, R. 1997. Timber-management and natural-disturbance effects on moose habitat evaluation. Journal of Wildlife Management, 61(2), 517-524. Wildlife Society. Rettie, W. J., Sheard, J. W., & Messier, F. 1997. Identification and description of forested vegetation communities available to woodland caribou: relating wildlife habitat to forest cover data. Forest Ecology and Management, 93(3), 245-260. Ihaka R. & R. Gentleman. 1996. R: A language for data analysis and graphics. Journal of Computational and Graphical Statistics, 5(3), 299-314. Sand, H., Wikenros, C., Wabakken, P., & Liberg, O. 2006. Effects of hunting group size, snow depth and age on the success of wolves hunting moose. Animal Behaviour, 72(4), 781-789. Schaefer, J. A., & Pruitt, W. O. 1991. Fire and woodland caribou in southeastern Manitoba. Wildlife Monographs, 116(116), 339. Schaefer, J. A. 2003. Long-term range recession and the persistence of caribou in the taiga. Conservation Biology, 17(5), 1435-1439. Schmidt, P. A., & Mech, L. D. 1997. Wolf pack size and food acquisition. The American Naturalist, 150(4), 513-517.

44

Seip. 1992. Factors limiting woodland caribou populations and their interrelationships with. Canadian Journal of Zoology. Solomon, M. E. 1949. The Natural Control of Animal Populations. Society, 18(1), 1-35. Therneau, T. M., & Grambsch, P. M. 2000. Modeling survival data: extending the cox model. Springer. Thiel, R. P. 1985. Relationship between road densities and wolf habitat suitability in Wisconsin. American Midland Naturalist, 113(2), 404-407. Thurber, J. M., & Peterson, R. O. (1993). Effects of population density and pack size on the foraging ecology of gray wolves. Journal of Mammalogy, 74(4), 879-889. Thurber, J. M., Peterson, R. O., Drummer, T. D., & Thomasma, S. A. 1994. Gray wolf response to refuge boundaries and roads in Alaska. Wildlife Society Bulletin, 22(1), 61-68. Van Beest, F. M., Mysterud, A., Loe, L. E., & Milner, J. M. 2010. Forage quantity, quality and depletion as scale-dependent mechanisms driving habitat selection of a large browsing herbivore. Journal of Animal Ecology, 79(4), 910-922. Vucetich, J. A., Peterson, R. O., & Schaefer, C. L. 2002. The effect of prey and predator densities on wolf predation. Ecology, 83(11), 3003-3013. Wasser, S. K., Keim, J. L., Taper, M. L., & Lele, S. R. 2011. The influences of wolf predation, habitat loss, and human activity on caribou and moose in the Alberta oil sands. Frontiers in Ecology and the Environment, 9(10), 546-551. Webb, N. F., Hebblewhite, M., & Merrill, E. H. 2008. Statistical methods for identifying wolf kill sites using global positioning system locations. Journal of Wildlife Management, 72(3), 798-807. Weir, J. M. H., & Johnson, E. A. 1998. Effects of escaped settlement fires and logging on forest composition in the mixedwood boreal forest. Canadian Journal of Forest Research, 28(3), 459-467. Whittington, J., Cassady, C., Clair, S., & Mercer, G. 2004. Path tortuosity and the permeability of roads and trails to wolf movement. Ecology And Society, 9(1). Whittington, J., ST Clair, C. C., & Mercer, G. 2005. Spatial responses of wolves to roads and trails in mountain valleys. Ecological Applications, 15(2), 543-553. Wittmer, H. U., McLellan, B. N., Serrouya, R., & Apps, C. D. 2007. Changes in landscape composition influence the decline of a threatened woodland caribou population. Journal of Animal Ecology, 76(3), 568-579. Wittmer, H. U., Sinclair, A. R. E., & McLellan, B. N. 2005. The role of predation in the decline and extirpation of woodland caribou. Oecologia, 144(2), 257-267.

45

Woodland, O., Recovery, C., & February, T. 2005. Recovery Strategy for Forest-Dwelling Woodland Caribou ( Rangifer tarandus caribou ) in Ontario. Ontario Ministry of Natural Resources. Ontario Parks Species at Risk Section, Peterborough Ontario. Zimmermann, B., Wabakken, P., Sand, H., Pedersen, H. C., & Liberg, O. 2007. Wolf movement patterns: a key to estimation of kill rate? Journal of Wildlife Management, 71(4), 11771182.

46

APPENDIX A AUDEN SITE WITH KILL SITES, ROADS AND WATER FEATURES

47

APPENDIX B - COCHRANE SITE WITH KILL SITES, ROADS AND WATER FEATURES

48

APPENDIX C PICKLE LAKE SITE WITH KILL SITES, ROADS AND WATER FEATURES

49

APPENDIX D PREDATION PROFILE FOR RECKETTE LAKE PACK

Kill ID A04-2011 A17-2011 A29-2011 A36-2011 A52-2011 A52-2011 A52-2011 A53-2011 A53-2011

Time of Predation Event 30/12/2010 20:59 15/01/2011 17:00 28/01/2011 8:30 08/02/2011 10:02 16/02/2011 18:00 01/03/2011 10:00 05/03/2011 11:34 05/03/2011 21:31 24/03/2011 20:33

End of Prey Handling 12/01/2011 9:00 26/01/2011 15:28 31/01/2011 14:31 12/02/2011 6:29 24/02/2011 5:59 02/03/2011 16:00 05/03/2011 13:59 18/03/2011 22:03 26/03/2011 0:00

Handling Time (hrs) 300 214 78 92 180 30 3 193 28

Search Time (hrs) 80 41 187 107.5 100 42 7 142

Type Kill Kill Kill Kill Kill Return Return Kill Return

50

APPENDIX E COX PROPORTIONAL HAZARD MODELS: MECHANISM RESULTS

Table 1. Cox Proportional hazard models relating variables (Mechanism: Predator Road Use) to time to kill moose in Northern Ontario, Canada. MODEL VARIABLES LL K AICc AICc Wi M1 DISTRD -154.07 3 314.18 1.41 0.32 M2 M3 DISTRD, VELOCITY DISTRD*VELOCITY -152.35 -154.32 4 312.77 5 318.64 0.00 5.98 0.65 0.03

51

Table 2. Cox Proportional hazard models relating variables (Mechanism: Variations in Hollings Equation) to time to kill moose in Northern Ontario, Canada.

MODEL VARIABLES M1 PREY DENSITY M2 M3 M4 VELOCITY VELOCITY, PREY DENSITY VELOCITY*PREY DENSITY

LL -155.23 -154.17 -154.32 -154.57

K AICc 3 316.50 3 314.38 4 316.71 5 319.25

AICc 2.12 0.00 2.33 4.87

Wi 0.20 0.57 0.18 0.05

52

Table 3. Cox Proportional hazard models relating weather variables to time to kill moose in Northern Ontario, Canada.

MODEL M1 M2 M3 M4 M5 M6 M7

VARIABLES TDP TDP, MT TDP, SD, MT SNOW DEPTH (SD) SD, TDP SD, MT MT

LL -154.26 -153.82 -153.00 -153.64 -152.96 -153.6 -154.28

K 3 4 5 3 4 4 3

AICc 314.56 315.71 316.11 313.32 313.99 315.27 314.60

AICc 1.24 2.39 2.79 0 0.67 1.95 1.28

Wi 0.14 0.08 0.07 0.27 0.19 0.10 0.14

53

Table 4. Cox Proportional hazard models relating variables (Mechanism: Prey Detection) to time to kill moose in Northern Ontario, Canada. MODEL VARIABLES
M1 M2 M3 M4 M5 M6 M9 M10 M11 M12 M13 M14 M16 M17 M18 M19 M20 M21 M22 DISTRD+ESKER+SLOPE+ELEV DISTRD+RS DISTRD+RS+DISTWAT DISTRD+DISTWAT SLOPE+ELEV SLOPE+RS DISTRD+ESKER DISTRD+ESKER+DISTWAT DISTRD+ESKER+SLOPE+DISTWAT DISTRD+ESKER+SLOPE RS+DISTWAT DISTWAT+DISTRD+ESKER+RS+SLOPE DISTWAT+DISTRD+ESKER+ELEV+RS+SLOPE DISTANCE TO ROAD DISTWAT ESKER SLOPE ELEV RS

LL
-156.37 -154.26 -153.04 -154.04 -156.3 -156.3 -156.46 -154.75 -155.54 -155.24 -154.24 -155.56 -156.22 -154.07 -154.22 -155.48 -155.65 -155.26 -155.26

K AICc
6 4 5 4 4 4 4 5 6 5 4 7 8 3 3 3 3 3 3 324.89 316.59 316.19 316.15 320.67 320.67 320.99 319.61 323.23 320.59 316.55 325.32 328.70 314.18 314.48 317.00 317.34 316.56 316.56

AICc
0.71 2.41 2.01 1.97 6.49 6.49 6.81 5.43 9.05 6.41 2.37 11.14 14.52 0.00 0.30 2.82 3.16 2.38 2.38

Wi
0.00 0.07 0.08 0.08 0.01 0.01 0.01 0.01 0.00 0.01 0.07 0.00 0.00 0.22 0.19 0.05 0.05 0.07 0.07

54

Table 5. Cox Proportional hazard models relating variables (Mechanism: Canopy Stand Type) to time to kill moose in Northern Ontario, Canada.

Model Variables M1 MIX+MCU+MCL+DEC+WAT M2 M3 M4 M5 M6 M7 M8 M9 M10 M11 M12 M13 M14 M15 M16 M17 M18 M19

LL -154.88 -154.59 -153.86 -153.85 -153.76 -155.07 -154.78 -155.63 -155.15 -153.81 -155.5 -154.57 -155.13 -154.84 -153.66 -155.65 -154.92 -153.69 -154.25

K AICc 7 323.96 6 321.33 5 317.83 4 315.77 3 313.56 3 316.18 4 317.63 5 321.37 6 322.45 4 315.69 4 319.07 4 317.21 5 320.37 3 315.72 4 315.39 3 317.34 3 315.88 5 317.49 4 316.57

AICc 10.44 7.81 4.31 2.25 0.04 2.66 4.11 7.85 8.93 2.17 5.55 3.69 6.85 2.20 1.87 3.82 2.36 3.97 3.05

Wi 0.00 0.01 0.03 0.08 0.25 0.07 0.03 0.00 0.00 0.08 0.02 0.04 0.01 0.08 0.10 0.04 0.08 0.03 0.05

MIX+MCU+MCL+DEC MIX+MCU+MCL MIX+MCU MIX MCU MCU+MCL MCU+MCL+DEC MCU+MCL+DEC+WAT MCL+MIX MCL+DEC MCL+WAT MCL+DEC+WAT WAT WAT+MIX DEC MCL MCL+MIX+WAT MIX+DEC

55

Table 6. Cox Proportional hazard models relating variables (Mechanism: Satiation) to time to kill moose in Northern Ontario, Canada.

Model M1 M2 M3 M4

Variables HT RETURN RETURN, HT RETURN*HT

LL -154.88 -155.88 -156.14 -154.07

K 3 3 4 5

AICc 315.80 317.80 320.35 318.25

AICc 0.00 2.00 4.55 2.45

Wi 0.57 0.21 0.06 0.17

56

Table 7. Cox Proportional hazard models relating variables (Mechanism: Pack Characteristics) to time to kill moose in Northern Ontario, Canada.

Model M1 M2 M3 M4 M5 M6 M7

Variables ID PACK SIZE (PS) SITE SITE, PS SITE, PS, ID ID, PS ID, SITE

LL -158.96 -155.07 -155.83 -155.46 -159.17 -159.24 -159.32

K AICc 3 323.96 3 3 4 5 4 4 316.18 317.70 318.99 328.45 326.55 326.71

AICc 7.78 0.00 1.52 2.81 12.27 10.37 10.53

Wi 0.01 0.57 0.27 0.14 0.00 0.00 0.00

57