Macrohabitat Selection by Vancouver Island Cougar

MACROHABITAT SELECTION BY VANCOUVER ISLAND COUGAR

(Puma concolor Vancouverensis)

Karen ML. Goh

B.Sc, The University of British Columbia, 1994

A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF

MASTER OF SCIENCE

THE FACULTY OF GRADUATE STUDIES

(Faculty of Agricultural Sciences)

We accept this thesis as corifonriing to the required standard

THE UNIVERSITY OF BRITISH COLUMBIA

April 2000

© Karen M.L. Goh 2000 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission.

The University of British Columbia Vancouver, Canada

Date Apri I 7jflf loon ABSTRACT

I examined macrohabitat selection by cougars (Puma concolor vancouverensis) on Vancouver Island between May 1997 and May 1999. During this period, 1285 locations were gathered from 9 female adult cougars using VHF radio telemetry collars. I assessed selection in the context of the ecology of their major prey Columbian black-tailed deer (Odocoileus hemionus columbianus), and hunting cover. I used a logistic regression model, and found seasonal and annual differences in habitat selection. Cougars tended to use mid elevations (ca. 390 m), and were closer to edges (ca. 150 m) than expected randomly. Year round, cougars showed selection for proximity to Old Growth, and Young forests. During winter, cougars avoided Clearcuts. They also tended to stay in Second Growth and Mature forests, while remaining in proximity to Old Growth forests. During spring, cougars avoided areas of low percentage cover of understory. During summer, cougars focused on areas of high quantity deer forage, and avoided being in proximity to areas of low quantity deer forage. Cougar selected hunting cover that was advantageous to them, year round and seasonally. No selection occurred for slope, aspect, and distance to water. Management of cougars should focus on maintaining a large prey base and features of good cougar hunting cover through smaller cutblock operations, and to reduction of human disturbance through fewer cutblocks. Allocation of forested space around 1800 km2 devoid of human activity may ameliorate conditions that cause cougar declines. TABLE OF CONTENTS

ABSTRACT II

TABLE OF CONTENTS Ill

LIST OF TABLES IV

LIST OF FIGURES V

ACKNOWLEDGEMENTS VI

INTRODUCTION 1

STUDY AREA 5

METHODS 6 CAPTURE AND COLLARING 6 LOCATIONS 6 GIS ANALYSIS 7 Forest Cover 9 TRIM 12 ANALYSIS 12 HOME RANGE ANALYSIS 12 STATISTICS 13 MODEL CONCEPTS 15 FINDING THE BEST MODEL 17

RESULTS 19

ASSUMPTIONS 19 RESOURCE SELECTION FUNCTIONS 22 SPRING MODEL (MARCH TO MAY) 22 SUMMER MODEL (JUNE TO NOVEMBER) 25 WINTER MODEL (DECEMBER TO FEBRUARY) 25 YEAR-ROUND MODEL 26

DISCUSSION 27

MATURE PHASE I AND UNDERSTORY REINITIATION 27 FOREST AGE CLASS, EDGE TYPE AND EDGE CONTRAST 28 Spring (March to May) 28 Summer (June to November) 29 Winter (December to February) 30 Year-Round 31 DISTANCE TO EDGE .". : 32 DISTANCE TO WATER 32 ELEVATION 33 SLOPE AND ASPECT 34

REFERENCES 36

APPENDIX I: COMBINATIONS AND FREQUENCY OF EDGE CONTRASTS 41

APPENDIX II: PROBABILITY FUNCTIONS 43

APPENDIX III: FREQUENCY OF HABITAT USE AND AVAILABILITY 44

APPENDIX IV: HOME RANGE SIZES OF VANCOUVER ISLAND COUGAR 45

APPENDIX V: ASSUMPTIONS 46

iii LIST OF TABLES

Table 1. Forest age classes based on successional stages, beneficial to cougars' major prey source, deer. Benefit to deer and cougar are based on consultation with deer experts. N/A = not available 10

Table 2. Habitat variables used in my logistic regression analysis and the categories within each non-continuous habitat variable. There are seven habitat variables, and a total of 28 categories including continuous variables 15

Table 3. The deviance values (-21ogelikelihood) of the Resource Selection Functions for nine adult female cougar

on Vancouver Island, estimated seasonally and year round between 1997-1999. The Dm that has the greatest

difference from D0 provides the best fit. 22

Table 4. Selection processes for each of the four models based only on significant habitat variables 24 LIST OF FIGURES

Figure 1. Study area on Vancouver Island, Canada The latitudes and longitudes bounding the area are: NE corner, 50°27.718'N, 179°58.374'W; SE corner, 50°15.905'N, 168°4.047'W; NW corner, 50°21.869'N, 162°28.148'W; and SW corner, 50°15.789*N, 156°21.735'W 5

Figure 2. Comparison of an unmerged habitat types on the left, to merged habitat types on the right, based on forest age classes 9

Figure 3. Examples of the variances calculated when determining sample size. The sample size required for random points was taken at this point where the variance peaked off for each continuous habitat variable. A sample size of 1000 approximates the asymptote level, and I used n=2000 to be prudent 14

Figure 4. Comparison of different models. Null models were always compared to ISM (Individual Selection Model), Pooled Filter models, and to check for 1) degree of selection, and to ensure that 2) individual cougar selection exists. Significant variables from Filter models were used to produce the Reduced model. Further significant variables were used for the Further Reduced model. The largest difference between the model and its null provided the best model fit 18

Figure 5. Schedule for gathering locations for radio collared cougars on Vancouver Island 20

Figure 6. Extent of 100% MCP home range overlap among 9 adult female cougars on northeastern Vancouver Island B.C., Canada : ,....21

Figure 7. Used versus available distance to edge (m). Selection occurred for shorter distances than expected randomly in each model 23

Figure 8. Used versus available elevation. Selection occurred for lower elevations than would be expected randomly in each model 23

Figure 9. Frequency histogram of cougar locations (top) and random points (bottom). Cougar locations are concentrated at c. 200-500 m with an average of 350 m, whereas random locations are concentrated at c. 350- 700 m with an average of 600 m. Cougars showed significant selection (P = 0.0002) for lower elevation areas with an overall mean elevation of 393.0 ± 5.9 m (95% CI) versus 599.0 ± 4.7 m (95% CI) for random points.35 ACKNOWLEDGEMENTS

Forest Renewal British Columbia, the B.C. Ministry of Environment, Lands, and Parks, and Mountain

Equipment Cooperative, MacMillan Bloedel Co. Ltd., and Western Forest Products provided financial and in-kind support for this research. I am grateful to the cougar houndsmen who generously offered their time for what was sometimes a difficult job. Doug Janz, Tony Hamilton, Helen Schwantje, and the

Conservation Officers, also provided generous support and help throughout. The staff of MacMillan

Bloedel in Sayward, Campbell River, and Nanaimo Divisions were very kind in their efforts to assist us with administration, acquiring GIS data, and general support. I also thank the numerous people that assisted in the project: Michael DeLaronde, Sophie Foster, Hugh Robinson, Avril Shepherd, Corinna

Wainwright, and Kimberly Wickert. I wish to give special thanks to my supervisor David Shackleton, who provided such genuine support, as well as Alton Harestad (who was also a temporary supervisor in

1997 whilst Dr. Shackleton was on sabbatical), Doug Janz, and Charles Krebs for being part of the review process throughout this study. Thanks to Roger Ramcharita, Shawn Taylor, and Heiko Witmer for bending over backwards to help with thesis edits. And a very special thanks to my friends and family and

Aaron Gladders who made this an adventure.

vi INTRODUCTION

Cougars (Puma concolor), the most widespread felid present in the Americas, range from Southern

Alaska and the Yukon to southern Chile. They occur in more climates, biomes, and latitudes than any other wild mammal in the Western Hemisphere (Anderson 1983). Being a higher trophic level species, they are an important species in their ecosystems, and as such have a substantial influence within their environments (Hunter and Price 1992; Power 1992).

Many published studies have contributed to our understanding of cougars including prey types

(Hornocker 1970; Maserand Rohweder 1983; Toweill and Maser 1985; Wehausen 1996), movements

(Beier etal. 1995; Hemker et al. 1984), demographics (Iriarte et al. 1990; Lindzey et al. 1988, 1994;

Spreadbury et al. 1996), social organization and behaviour (Hornocker 1969; Laing and Lindzey 1993;

Seidensticker et al. 1973), and effects of hunting by humans (Logan et al. 1986; Ross et al. 1996; Ross and Jalkotzy 1992; Torres et al. 1996). Little research, however, has been published on the Vancouver

Island subspecies (P. c. vancouverensis), and specifically on its selection of habitat in temperate rainforests.

Forest harvesting and human activity in the forests of North America has created concerns for the viability of wildlife populations. Some wildlife species are at risk of decline due to landscape alterations by humans. The spotted owl (Strix occidentalis) (Andersen and Mahato 1995; Lamberson et al. 1994;

Lehmkuhl and Raphael 1993), marbled murrelet (Branchyramphus marmoratus) (Rodway et al. 1993), grizzly bear (Ursus arctos) (Mattson et al. 1996; Wielgus et al. 1994), and northern goshawk (Accipiter gentiles) (Squires and Ruggiero 1996) are examples of species that face population declines due to forestry-related habitat loss. High elevation logged areas act as a mortality sink for the Vancouver Island

Marmot (Marmota vancouverensis) when they disperse to such areas instead of their natural habitat, subalpine meadows (Bryant 1996; Bryant and Janz 1996).

1 Cougars on Vancouver Island are of interest for several reasons. This subspecies of cougar is known for its greater incidences of attack on humans, relative to mainland subspecies (Beier 1991). It inhabits dense forests and exists in areas of forest harvesting and rapidly expanding human habitation, and is becoming a management and public concern. Furthermore, conflicts between the forest industry and environmentalists have directed attention to whether charismatic species, such as cougar, require pristine habitat.

My research examines cougar-forestry interactions. Specifically, I examine the effects of forestry on the pattern, extent and process of habitat selection by Vancouver Island cougars. To address forest harvesting effects, I observed these processes at the macrohabitat scale, based on biologically important variables that I assume the animal can perceive (Morris 1987; Nyberg and Janz 1990). Macrohabitats are a compilation of patches that have characteristic physical and chemical variables (Morris 1987). The entire suite of patches comprises the possible macrohabitats that exist for a species. This is not to say that the macro-scale is the only or best scale to examine cougar habitat use. All species are likely to show a range of coarse to fine-grained resource selection (Morris 1987) based on differences in spatial, temporal and organizational scales (Eng 1998). The key is to be able to understand the processes that are resulting in the observed habitat use pattern. The scale should not be purely human-based on ecological scale

(Maurer 1985), but should be relevant to the research focus (Eng 1998). The general management

objective of this research is understanding the response of cougars to changes made by forest harvesting.

The macrohabitat scale is appropriate to the scale of forest harvesting.

Four major ecological factors will have a strong effect on how cougar use habitat: prey, hunting cover,

intraspecific competition and interspecific competition (Hornocker 1969, 1970; Laing 1988; Maser and

Rohweder 1983; Seidensticker et al. 1973; Sunquist and Sunquist 1989; Toweill and Maser 1985;

Wehausen 1996).

2 The most pervasive and strongest processes are arguably the first two factors: cougars' prey and their means to obtain them. The activities of the prey are known to strongly influence the predator's behaviour

(Karanth and Sunquist 1995; Sunquist and Sunquist 1989; Rabinowitz and Nottingham 1986). Predators should select the most "profitable" prey measured by energy gain over the need to stalk, capture and handle the prey (Stephens and Krebs 1987). For example, in Utah, Julander and Jeffery (1964) found a high correlation between habitat use by deer (Odocoileus spp.) and cougar habitat use, suggesting a strong relationship between predation and prey habits. Similar relationships between prey and predator habits have also been reported for jaguars (Panthera oncd) (Crawshaw 1991; Rabinowitz and Nottingham 1986;

Schaller and Crawshaw 1980), leopards (Felis bengalensis) (Rabinowitz 1990), and tigers (Panthera tigris) (McDougal 1977). Thus I expect that, in part, cougar will select habitat based on the availability and ecology of their prey.

The distribution of profitable hunting habitat as determined by the interspersion of the vegetation, slope and topography is also expected to affect cougar (Kruuk 1982). There is a high correlation between minimum distance covered in the final ambush and kill success (Sunquist and Sunquist 1989). Stalking cover, needed for the approach, will have a strong effect on ambush distance, and studies have demonstrated a higher kill success rate when cover is available. For example, lions (Panthera led) have a

higher kill success rate in dense riverine thickets than areas of little or no cover (Schaller 1972). Tigers

(Panthera tigris) were able to kill more sambar (Cervus unicolor) in the burned over areas that offered

cover, than in any other time of the year (Sunquist 1981). Cougars (Puma concolor) used mixed swamps

and hammock forests to maximize hunting cover (Belden et al. 1988). Thus, although I expect that the

habitat a cougar will use is determined largely by prey abundance and ecology, its use of habitat is also

influenced by hunting cover.

3 The objective of my study is to determine if cougars select any of the following habitat variables: slope, aspect, elevation, distance to edge, edge type, edge contrasts, distance to water, and forest-age classes.

These variables, for different reasons, are important to both cougar and their primary prey, Columbian black-tailed deer (Odocoileus hemionus columbianus) (Nyberg and Janz 1990; Kremsater and Bunnell

1992). My research examines the strength of the following selection processes and the implications of any selection. I made the following predictions:

1) I expected cougars to exhibit selection for both older and younger forest age classes thick in understory cover because of the high food and cover values that could attract deer, and thus cougars.

2) I also expected cougars to exhibit selection for edge types with high contrasts (e.g., Clearcuts and Old Growth) because they are areas in which ungulates are abundant (Chang et al. 1995; Hanley 1983; Kremsater and Bunnell 1992; Nyberg and Janz 1990; Voller 1998), and where cougar- hunting cover is available (Branch 1995; Elliot et al. 1976; Logan and Irwin 1985; Ockenfels 1994).

3) Related to edge types, I expected that in looking at the cougar locations, there would be positive relationships between the forest type that the location is in and the closest different forest type (herein called edge contrasts). I expected that edge contrasts would be dominated by one forest type that would provide food resources for deer and another forest type that would provide cover for deer. These contrasts would also provide concealment cover for cougars to enhance ambushing of prey.

4) I expected cougar to be closer to edges (distances to edge) than would be expected at random, due to anticipated increased edge habitat use.

5) I expected that cougars do not need to be near water sources. Although deer have water requirements (Nyberg and Janz 1990), distance to a water source is thought to be less important in the wet coastal region where water is readily available. As a result of this, I expected that there should be neutral selection of distance to water by cougars as well.

6) I expected that cougars would show selection for mid and low elevations to alleviate the adverse effects of snow (Telfer and Kelsall 1979) and lack of prey at high elevations.

7) I expected that cougars show selection for high slope because these have been shown together with vegetation and movement, to increase cougar's advantage to attack prey (Beier et al. 1995).

8) I expected use of areas with warmer aspects in the winter because that is where the prey of cougars would alleviate thermal constraints (Kremsater and Bunnell 1992; Nyberg and Janz 1990).

4 STUDY AREA

The study was conducted from May 1997 to May 1999 in a 2500 km2 study area on northeastern

Vancouver Island, British Columbia, Canada (Figure 1). Ranging in elevation from sea level to 1798 m, the two biogeoclimatic ecosystem classification (BEC) zones in the area are Coastal Western Hemlock

(CWH) and Mountain Hemlock (MH) zones. Major tree species in decreasing abundance within the

CWH zone includes western hemlock (Tsuga heterophylla), western redcedar (Thujaplicata), amabilis fir

(Abies amabilis), lodgepole pine (Pinus contorta), and red alder (Alnus rubra). In the MH zone, major tree species in decreasing abundance include mountain hemlock (Tsuga mertensiana), amabilis fir, and yellow-cedar (Chamaecyparis nootkatensis). The CWH is one of the most productive BEC zones in

Canada and is on average the wettest biogeoclimatic zone in British Columbia. It experiences cool summers and mild winters with mean annual temperature between 0 and 10°C for 4-6 months of the year and mean annual precipitation of 1700-5000 mm of which 20-70% is snow (Meidinger and Pojar, 1991).

5 METHODS

Capture and Collaring

Walker and Blue Tick hound dogs were used by local houndsmen to find and pursue cougars until they were treed (Hornocker 1969). A mixture of Ketamine and Medetomidine was administered through a

Capchur dart gun for immobilization. The animals were fitted with Very High Frequency (VHF) radio collars (Telonics, Mesa, Arizona, USA), and tagged in both ears with uniquely numbered disks. A reversal agent, Atipamezole, was used to counter the immobilization.

Locations

Diurnal locations were gathered in 5 ways: at capture, visual, prey kill sites, ground radio-telemetry, and aerial radio-telemetry. Capture locations were obtained at the first capture of the cougar. Visuals were obtained when an animal was seen, or when an observer did not sight the cougar, but was able to hear the cougar (< 50 m) during a walk-in. Kill locations were determined from confirmed sites at which prey remains were found. Ground locations were gathered systematically for each cougar using radio- telemetry (ground and aerial). Each ground location was derived from a minimum of three radio telemetry bearings (i.e., mobile towers) approximately 4 times per week. A LoTek (Newmarket, Ontario,

Canada) radio receiver that quantified signal strength between false (i.e., bounce) and valid signals was used to determine the strongest signal at each tower. A four-element Yagi antenna was used with the

©LoTek receiver. Tower locations were identified with a survey-quality 12 parallel channel GPS receiver (Ashtech Reliance RT, Santa Clara, California, USA) capable of real-time differential correction

(2-5 m error) from a Canadian Coast Guard beacon.

Each tower was identified by its Universal Transverse Mercator (UTM) coordinates and entered into the

LOAS program (Sallee 1999) that uses the algorithms of the Lenth Andrews-M or Maximum Likelihood estimates (Lenth 1981a, b). This estimate provides the probable location of the individual cougar based

6 on observer error, distance to the animal, and distribution of towers around the location. Heezen and

Tester (1967) first suggested the importance of estimating error when locating animals through radio telemetry. Observer error is of primary concern because if telemetry error is large, then sample sizes required for relocation increase significantly (Nams 1989). Furthermore, as the scale of interest becomes finer, the error effect becomes larger. The level of confidence in telemetry locations is inversely proportional to observer error. To correct for this, each observer underwent error testing on test collars to quantify the observer and system bias (White and Garrott 1990). The highest standard deviation value determined for all observers was 1.5 degrees and that value was used in the triangulation calculation.

Any ground locations with an error ellipse > 10 ha were deleted from the database, because of a lack of accuracy relative to my interest in macroscale selection (10 ha would consistently overlap numerous forest polygons).

Cougars were also aerially relocated using radio-telemetry via Cessna airplane about twice a month. A two-element Yagi antennae was attached to each wing strut on each side of the plane, and co-axial cables from the antennae were attached to a ©LoTek receiver and switch box within the cabin. Relocation errors were determined on each flight by placing test collars in fixed locations unknown to the observer.

Relocation of test collars resulted in an average error of 2.4 ± 0.61 ha (95% CI) deviation from the true

location of the collar.

GIS analysis

After all locations were collected, I used a Geographic Information System (GIS) to derive estimates of

slope, aspect, elevation, forest age, distance to edge and to water, edge type and the edge contrast for each

location. All GIS Analyses were conducted using software programs of Arcview, Arcview extensions

Spatial Analyst, and 3D analyst (Environmental Systems Research Institute, Redlands, California, USA), as well as extensions XTOOLS (DeLaune 1999) and Animal Movement (Hooge and Eichenlaub 1997).

7 In addition, two scripts were created, one to derive distance to edge and the second to derive elevation, slope and aspect (Moy 1999a, b). Two GIS data sources were used: forest cover maps of forest ages, waterways, roads, and general topographic feature (MacMillan Bloedel Co. Ltd. and Western Forest

Products Ltd.), and Terrain Resource Inventory Maps (TRIM) of elevation, slope, and aspect (provided by B.C. Ministry of Environment, Lands, and Parks - MoELP).

Slope, aspect and elevation were averaged over a 20-m pixel size. Elevation was analyzed as a continuous variable, but is discussed in terms of three categories: low, medium and high elevations.

These elevation categories were created based on observations of snow cover and qualitative characterization of road use. Low elevations (0-300 m) receive little snow and are characterized by the highest area of roads in the study area. Mid elevations (300-600 m) receive snow cover on occasion and have a moderate area of roads in the study area, and high elevations (> 600 m) have snow cover throughout much of the winter, or adverse weather conditions, and are characterized by steep slopes as well as a low area of roads.

Forest classes are divided into 8 forest age classes that will be defined in the upcoming "Forest Cover"

section. Each cougar location can be analyzed in three major ways using this forest age class information.

First, is the actual forest class that the cougar location falls into. Second, is the forest class that the

location is closest to. This is referred to as the Edge Type. It follows that the closest forest class will

always be different from the forest class that the location is in. There are only 8 possible Edge Types to

classify any one location. Furthermore, there can be 100 locations that have an "Old Growth edge type"

but there will not be any information on the actual forest class that these 100 locations fall in. Thirdly,

there is the Edge Contrast that expands on Edge Type by making a relationship between one forest class

and another. Edge contrasts are pairs of each of the 8 forest classes to another of the remaining 7 classes.

That is, there are 8 Classes and 56 possible relationships: (7+6+5+4+3+2+1=28)(2) (Appendix I). For

example, edge contrast of Young to Old Growth refers to a location in Young forest class with Old

8 Growth as its closest different forest class. Edge contrasts brought up in the discussion were ones that had average medium contrast, occurring at least 4 times more or 0.25 times less than expected randomly.

Forest Cover

The forest cover maps for this area are particularly accurate, being a test area for experimental forest harvesting. The area is mapped at a 1:5000 scale (as opposed to the typical 1:20,000 scale) and extensive geo-referencing occurred throughout the area to further improve accuracy. Forest cover maps presented a mosaic of many forest age polygons, separated by a minimum of one year of age. This level of precision was too fine for my purposes, so I grouped ages into 8 forest age classes. This grouping resulted in merging of adjacent polygons (Figure 2). The rationale for reducing the number of forest age classes was to create structurally different successional forest classes (Table 1) of potential significance to cougar's

major prey — deer (Oliver 1992; D. Janz pers. comm. B.C. Ministry of Environment, Lands and Parks,

Nanaimo 1999; A. MacKinnon pers. comm. B.C. Ministry of Forests, Victoria 1999; D. Meidinger/?ens. comm. B.C. Ministry of Forests, Victoria 1999; F. Nuszdorfer pers. comm. B.C. Ministry of Forests-

Nanaimo 1999; B. Nyberg pers. comm. B.C. Ministry of Forests, Victoria 1999).

Figure 2. Comparison of an unmerged habitat types on the left, to merged habitat types on the right, based on forest age classes.

9 8 « S « *- is cn CN vq CN vq a .9 CN cn r-; VO CN oo m v IS i?l u v? a fij ^ to o O O O O in in CO + &2 I VO as CN o i m < vo CN

T3 cu c -a O c o O i— a> o '3 o 00 00 oo CO +J a

3 'C ca 1. w S 03 cu a C D cu cj two o "-" s CU c C CU C£ « > o s o -S o W .. a. CU CU cu a *» B ^1 E c .S * g || u B cu s <-> '3 < C u U CD CU 00 4> >o > CU > S P* CO o a E .S E i S M B 5 3 E if-g CO % CO *1 o <*3 to - « SP oi oi WD ex's eJla WD r, e e <*- •r cu w .S 3 ue iu ca 00T3 'C Q. a o. £ S g < .S o. S. a c/3 OJ op

T3 o O -a ,o -a -a cu o o <2 cu o T3 T3 <2 O 6 cS <£ T3 Q ~a o JS 3 o O o 00 o 1> •a cS o cS •a *3 CU cu •a cu cu CJ > > •a CU L ° •a o o -a Z -a CJ o e CJ » a w C & 3 3 B C 3 cu u cu 2 -a cu 2 SJD T3 T3 CO •4-» o 1 c CU J3 cu E i - O cu o 00 00 i > | g O 6 O i > o B CU CO 8 CU WD 13 w>"3 M cc E C £ CO 8 M "ed •C E B C o. a s Q> 3 /3 /3 /3 -S co -S C/3 C/3 C C C *l co T» co C/3 C/3 C/3 « B 00 c c c c o o ^ "2 o 00 ca ca fl to ^ O j_. ce JS OH i- — ca 'TT' P o 1 •- +-> • SCO QJ -2 CD O y 1 3 o w T3 S .s J T3 3 S a la o 4-* CU C/3 /3 +^ 2 C C/3

10 Stand Initiation I class (0-5 years) represents the forest after a forest harvest. Stand Initiation II class (6-

15 years) is comprised of primarily herbs and shrubs that provide abundant summer forage for deer. At both these stages, the low understory offer deer low to medium security cover from predators and hunters

(Nyberg and Janz 1990), but good food value as well (with the exception in winter). These stages offer cougar some concealment cover at the edges for hunting but cougar remain highly visible to prey in the interior of the forest age classes (Table 1). Because clearcutting was the silvicultural system used to harvest trees, canopy cover is practically non-existent at Stand Initiation I and II, and travel (as measured through understory presence - i.e., increased understory = increased difficulty for travel) through these newly disturbed stands is often difficult for both cougars and deer. The Young class (16-30 years) is dominated by young, small-diameter trees with branches. This class provides medium to high security cover for deer (depending on season) because of the low height of the tree crowns as well as the high density of low branches and still has high food value for deer. For cougars, the Young class provides good concealment cover for stalking prey. During the Stem Exclusion class (31-60 years), the canopy closes almost entirely, and understory growth is minimal, providing favourable thermal cover but little food for deer. The lack of forage reduces deer use, and perhaps cougar use too because of the low concealment cover for stalking prey. During the Understory Reinitiation class (61-90 years) the canopy re-opens and understory growth begins again. At this class, the stands are still quite homogeneous and are generally dominated by single species, similar-aged forest stands. The canopy offers high thermal and security cover, and medium food for deer. In Mature Phase I (91-170 years) and II classes (171-250 years), stand differentiation becomes more marked, with understory growth substantially greater in

Mature Phase II. Stand differentiation begins with some understory growth (Mature Phase 1), while the establishment of canopy differentiation is accompanied by greater understory growth (Mature Phase II).

Both these classes provide medium security, high thermal cover, and medium food for deer. For cougar, these two classes provide good thermal cover as well as some concealment cover for stalking prey, and good travel areas for cougars. Old Growth (250+ years) is the last stage and contains tree species' renewal, gap replacement and substantial amounts of downed wood (Lofroth 1998). Old Growth has a

11 patchy appearance and a multi-layered canopy. It is useful to deer because it provides high thermal and security cover, and high food opportunities within the same patch, which is particularly important in years with high snow accumulation. Old Growth is also useful to cougars because of its good concealment cover against prey, as well as good thermal and travel habitat. The forest classes (Table 1) are readily distinguished from each other allowing analysis and discussions of edge type and edge contrast to be probably more meaningful.

TRIM

TRIM maps are accurate government generated maps, which among other things, detail 3-dimensional information very well. The standards for TRIM maps are high, such as having a maximum 10 m pixel size for satellite imagery, and a 10-micron scan resolution size for 1:10 000 and 1:20 000 maps. TRIM maps are also heavily georeferenced against known control points, further increasing the accuracy of the maps. TRIM maps were used to query topography in my thesis. I converted a Digital Elevation Model

(from the TRIM maps) into a Triangular Irregular Network (TIN) to derive aspect, slope, and elevation for all point queries. For the TRIM work, the coverages had to be converted into GIS grids to create the

TIN. Point queries on grids were based on a 20-m pixel size. I queried slope, aspect and elevation from the TRIM maps.

ANALYSIS

Home Range Analysis

Home ranges were analyzed for each cougar using a 100% Minimum Convex Polygon (MCP). I used the

100% MCP because I later generated random points within each cougar's home range. Because the MCP

forms a border around all points, I was maximizing the possible area within which random points could

fall thus assuming that each cougar had its entire MCP area available to it. This is opposed to other home

range estimates such as Adaptive Kernel that shapes itself more tightly to the distribution of points. 1

used Animal Movement (Hooge and Eichenlaub 1997) in Arcview to calculate home ranges.

12 Statistics

I used exploratory descriptive statistics and multivariate logistic regression resource selection functions

(RSF) to examine differences between cougar locations and random habitat points. Locations were analyzed separately for each cougar, and again with all animals pooled. Statistically, pooling all the animal locations together would be pseudoreplicating (Hurlbert 1984) because my population of inference is the individual cougar. Thus, pooling was done only to compare against the strength of individual selection. Random habitat points were generated using a normally distributed random generator Arcview script (Cederholm 1999) within each animal's 100% MCP home range calculated using the program

Animal Movement (Hooge and Eichenlaub 1997). The number of random points needed to represent each MCP home range was determined by sampling a range of sample sizes (n = 50, 100, 500, 1000,

1500, 2000, 5000). Variances of each continuous independent variable were graphed until the variances began to asymptote. As the asymptote is approached an increase in sampling effort does not result in a substantial change in variance. The sample size that was used indicated the establishment of an asymptote for all independent variables (Figure 3). The resulting number of random points was then queried within the GIS in the same manner as the cougar locations. Points by cougars (used) and random points (available) as well as their corresponding habitat variables were analyzed by several logistic models to determine the best fit of the data. These models followed a Design III sampling protocol A (Manly et al. 1993) in which both the used and available data are sampled, and each animal's use is known. I modelled the data for each of three seasons, as well as year-round. Seasons were based on differences in deer habitat use and requirements (Nyberg and Janz 1990): winter (December to

February), spring (March to May) and summer (June to November). Although there are differences from season to season, the habitat requirements across an entire year are still extremely important to consider

(Nyberg and Janz 1990). Thus an annual model of all data was used to assess habitat selection collectively.

The habitat variables used for all models were elevation, slope, aspect, distance to edge and distance to

13 Distance to Water Distance to Edge

250000

200000

150000 —US— 100000

50000

0 1000 1500 2000 5000 1000 1500 2000 5000

Sample Size Sample Size

Elevation Slope

1000 1500 Sample Size

water, forest-age class, and edge type (Table 2). Edge contrast could not be included as a habitat variable for the logistic regression because there were too many categories to consider within it (see Appendix I).

Because it was treated as a categorical variable, slope was converted into a dummy variable representing each of the five 20% slope intervals up to 90 °. Similarly, the aspect was divided into four dummy variables each of 90 0 around each cardinal point (i.e., Northern aspect was 45 ° to either side of 360 ° or

315 to 45 °). Also, each of the forest age classes (Table 1) was treated as a separate dummy variable (0 or

1 classification for each location), as was the type of edge. In the year-round model, the year that the data

fell into was classified into a dichotomous variable, with 0 representing the first 12 months of the study,

and 1 representing the second 12 months of the study.

14 Table 2. Habitat variables used in my logistic regression analysis and the categories within each non-continuous habitat variable. There are seven habitat variables, and a total of 28 categories including continuous variables.

Independent habitat variables Categories within a habitat variable (28 total)

Stand Initiation I, Stand Initiation II, Young, Stem Forest-Age Classes Exclusion, Understory Reinitiation, Mature Phase I, and Mature Phase II, and Old Growth

Stand Initiation I, Stand Initiation II, Young, Stem Edge Type Exclusion, Understory Reinitiation, Mature Phases I, and Mature Phase II, and Old Growth

Edge Contrast Combination of any two edge types. Edge Contrast is not (not an independent habitat variable considered an independent habitat variable because the in the logistic regression, but defined number of possible edge type combinations exceed useful here in relation to Edge Type) category numbers in the logistic regression.

Slope 20% intervals up to 90 0

90 ° intervals centered on each cardinal direction: N, S, E, Aspect W Elevation Continuous data (m)

Distance to Edge Continuous data (m)

Distance to Water Continuous data (m)

Model Concepts

For a multivariate logistic regression model, there will be a dependent dichotomous variable compared

against several independent variables. In my study the dependent dichotomous variable is the used and

random cougar locations (0 or 1). There are 7 independent habitat variables used out of 8 possible (Table

2). The 8th habitat variable (edge contrasts) is not considered due to the high number of possibilities

(Appendix I). The logistic regression model looks at the independent variables and distinguishes them

between the dependent variable, thus describing selection between used and random cougar locations.

One can select which habitat variables or categories within those variables (Table 2) they would like to

analyze in any given model. Habitat variables selected for the model can be created based on: 1) the ease

of collecting data in the field or 2) the statistical significance of the variables.

15 Each model can be described and compared to each other by Resource Selection Functions (RSF).

Within the RSF, the values that indicate the strength of selection by each independent habitat variable are called beta coefficients. Beta coefficients describe whether there was selection for or against a habitat variable, and how strong that selection was (for example, a positive beta for elevation indicates positive selection for higher elevation than found randomly). Beta coefficients are analogous to a regression equation slope derived for each habitat variable (independent variables) but differ in that they are created from a multivariate relationship between all the variables (Menard 1995). Any positive beta coefficients

indicate selection for a habitat variable, and vice-versa. Neutral coefficients (equal to zero) indicate no

selection.

The deviance values (-21ogelikelihood or-2LL) that characterize the RSF sum the expressions of the beta

coefficients and determine how well a model fits the data. Each RSF from a model is compared to a

"null selection" model in which the beta coefficients are set to zero (i.e., no selection is occurring). This

is done to check for significant differences between a model and its null model (in which selection has

been set to zero). If there is a significant difference between the actual and null models, selection is

considered to be occurring. The model with the greatest difference from the null model provides the best

fit for the data and indicates the strongest selection:

2 R = fit of the model = Gm / D0

Where:

Gm = Difference between the null model and actual model = D0 - Dm

D0 = -2LL of the null model

Dm = -2LL of the selection model

I created several models to find which variables provide the best fit for predicting cougar habitat on

Vancouver Island. In those instances, the values may not provide a good approximation for the fit of the

model, but can be used comparatively against other models (Manly et al. 1993).

16 For each model, an RSF and beta coefficients were calculated for each individual cougar, and summed to produce an overall deviance total (Ruggiero et al. 1998). These values were used for the Individual

Selection Model or ISM (Ruggiero et al. 1998; Manly et al. 1993). In creating an RSF for each individual cougar, I avoided pseudoreplicating (Hurlbert 1984; Ruggiero et al. 1998). To determine if habitat selection occurred among individuals, a model for pooled data (without regard to individual animals) was compared against both the null model and the ISM deviance total. If the pooled data had a better fit than the ISM (i.e., R2 is higher for the pooled data than the ISM), this indicates that there are insignificant selection processes that differ among individual cougars (Ruggiero et al. 1998).

Finding the Best Model

The model testing scheme is illustrated in Figure 4. The Null Selection model (beta coefficients or selection processes equal 0) was compared to another model that included all seven independent habitat variables (Table 2). With the total number of categories equaling twenty-eight variables (number of categories within the seven habitat variables as outlined in Table 2), the likelihood of finding a highly significant fit at an alpha of 0.05 is expected purely by chance (Menard 1995) and was not used. Thus, this model that incorporated all the variables and categories, was called a "Filter" model. Because of the high probability of finding a significant fit due to chance, it was first used to identify potentially significant independent variables and then those variables were used to derive a "Reduced" model. Both the RSF and the beta coefficients from the Reduced model were considered. The significant and almost

significant (P < 0.15) variables from the Reduced model were examined, and the almost significant variables were included to test if they remained insignificant. A Further Reduced model was created from the Reduced model's significant/near significant variables. The better fit between the Reduced and

Further Reduced model was used to describe cougar habitat selection (Figure 4).

17 This process was completed for each of the four used data sets representing spring, summer, winter and

year-round. The random points used to compare against the actual cougar-use points were identical for

each of the three data sets to maintain a consistent comparison for changes between models. Variables

were entered simultaneously rather than stepwise (one at a time), to avoid the influence of computer

algorithms and random variations in the data on the RSF values (Menard 1995). The resulting RSF

equations become predictive, allowing one to measure a location and estimate its value as potential

cougar habitat (Appendix II).

Null Model Null Model Null Model

Reduced Model Further Reduced Model Filter Model - all -ISM variables - ISM ISM

Significant Significant variables Vs. Vs. variables Vs.

Pooled Filter Pooled Reduced Pooled Further Model Model Reduced Model

18 RESULTS

Nine adult female cougars were radio tracked between May 1997 and May 1999, during which time between 32 and 238 locations were obtained per cat (Figure 5). A total of 1285 locations were collected for all cougars, and 2000 random points were created for each cougar. Raw habitat use data of cougars is outlined in Appendix III. Home range overlap among the females was extensive (Figure 6). The 100%

MCP Home range sizes were on average 273.7 ± 86 km2 (95% CI; Appendix IV).

Assumptions

All of the logistic regression assumptions (Appendix V) were met as follows. The functional form of a logistic regression is assumed for Assumption 1. In all selection models, variables have similar standard errors, suggesting a minimization of irrelevant variables. Nonlinearity in the logit form was met. A change in the logit of the dependent variable (y) is associated with a linear change in the logit of the independent variable (x), not the x actual values. For dichotomous variables, a linear relationship will always exist because there are only two variables being compared. For the non-dichotomous variables of elevation, distance to edge or to water, I found that increases in the independent x variable led to a linear increase in the logit (y) meeting the second assumption. Non additivity is assumed (see Appendix V).

The assumption of low collinearity was also met. For independent variables in all models, there is no instance of high collinearity because the highest R2 value is 0.548. The diagnostics of the model are a series of measurements used to identify cases (i.e., cougar or random locations) that abnormally exert their share of influence on the model, or for which the model works poorly (Menard 1995). In the diagnostics of the model (Menard 1995), all Leverage values in the model are either less than the value of

(12+1 )/l 285=0.01 (the lowest value of all the models, thus most prudent value to use), or slightly greater.

All values were significantly < 1.0, satisfying this diagnostic. Studentized residuals all fell within the range of negative and positive 3, and the Df beta values were < 1.0, meeting logistic regression assumptions (Menard 1995).

19 OS o co 00 co oo os CO CN co CN CN os CN

o o

CM O

CO 22

co CJ 0) +-» CO OQ Q 22

o CN CO oT5 TS o o 22 cn

2> cn

oo TS TS TS o o o 22 21 Resource Selection Functions

I analyzed three data sets representing each season and the complete data set representing year-round use to calculate RSF values (Table 3). Raw data for the habitat classes are available in Appendix III. All

RSF equations for the four models are given in Appendix II.

Table 3. The deviance values (-21ogelikelihood) of the Resource Selection Functions for nine adult female cougar on Vancouver Island, estimated seasonally and year round between 1997-1999. The Dm that has the greatest difference from D0 provides the best fit.

RSF for RSF for the best Individual RSF for the 2 Null Selection Model (Reduced or R Pooled Model Model Further Reduced)

Dm (pooled) D„ Dm Spring (Mar.-May) 3264.84 3065.99 2808.15 0.14 Summer (Jun.-Nov.) 5396.52 5075.12 4819.95 0.11 Winter (Dec.-Feb.) 3432.87 3145.41 2901.04 0.15 Year-Round 9442.33 8562.74 7840.05 0.17

SPRING Model (March to May)

Forest Age Class: Mature Phase I (P = 0.0002), and Stem Exclusion (P = 0.0278) forest ages were

selected against (Table 4).

Edge Type and Contrast: Edge types were not used significantly more than what would be expected,

during the spring. The edge contrast of Mature Phase I forests closest to Stand Initiation I forests were

used 10.4 times more frequently than expected (Table 4). Stand Initiation I forests in proximity to Stem

Exclusion forests were used 10 times more than expected. The use of edge contrast of Young forests next

to Stem Exclusion forests was 0.36 times less than expected randomly.

Distance to Edge: Lower edge distances (P = 0.0022) were selected for (Figure 7) and the average

distance was 139.0 ± 18.3 m (95% CI), compared to 286.0 ± 5.2 m (95% CI) in the random sample.

Distance to Water: Distance to water appeared to have no effect on habitat selection in spring.

22 Elevation/Slope/Aspect: Low elevations were also positively selected for (p = 0.0002). The average elevation used by cougar in spring was 399.0 ± 23.8 m (95% CI) compared to 599.0 ± 4.7 m (95% CI) the random sample (Figure 8). Neither slope nor aspect affected selection.

350

300

^ 250 O) S 200 CD O 150 C 2 to 100 Q 50

0 Random Spring Summer Winter Year-Round Categories

Figure 7. Used versus available distance to edge (m). Selection occurred for shorter distances than expected randomly in each model.

700

600

500 ? IT 400- - o i«lllllllfl —IE— ™ 300 a>

m 200

100

0 Random Spring Summer Winter Year-Round Category

Figure 8. Used versus available elevation. Selection occurred for lower elevations than would be expected randomly in each model. CO oj c: _> OJ CO o co o XJ cn co OJ a> "co CD i OJ CO OJ JD OJ o •c OJ o o o k o o CO O- JD o JD CC CO OJ JD CD o CC OJ CO OJ OJ OJ OJ co 5 co a CO CO co co .!§ s ro -CZ 2 XJ o o a> o CZ o >- CO o O o l_ a. XJ XJ OJ PJ JD XJ O o 03 oj JD co co OJ CO

CO OJ cz CO o CD o ca XJ co CO co OJ > OJ ca i_ oj OJ JD OJ o t: CO CL o OJ OJ o J= o CD CO o JD JD JD CO oj CO 1= O JD o OJ ro OJ OJ OJ CO OJ .o CO jo c> co CO "co co co o o CO _c CD CZ XJ o o CD Q_ OJ ;o XJ XJ CZ O -3> CJ co OJ JD co CO OJ co

XJ a S3CZ co OJ o o >^ CO co "a o ro co XJ "to co co CZ OJ > ro ro CO JD co oj OJ cu IOJ iC O o OJ o m I 1 0 ^ cS OJ H w OJ o E CC co oj =3 I JD CO co -a c= OJ _o JD E OJ =J CO c: o CO I OJ OJ c= 03 jo CO J= '-5 c "a o o co "co Q_ L O ™ ro XJ O i D co o •z. O o CD PJ £^ ro cr OJ XJ o "o CL CZ o OJ co JD E co J2 OJ _3> co OJ £= CO co XJ => OJ O CO >-

OJ o E o CO "O "ro ro CO is CO co co co S3 a> > CO ro oj i oj CD JD CO CD cz CZ cz E XJ E OJ o o XJ •*- V.O O">J CO c~> OJ "o "o o CZ -rt o o cz sz CD OJ o OJ JD CO c JD =s OJ >- CO o CO oj CD CD OJ a. CO co co ro co £ ro o CO 8 •B CO o - I-ti CO "K o O o Q_ OJ c: XJ ~Z. OJ T1 i_ PrCO o _Z 13 co Z3 ro — X> o CO E co cz "o JD OJ ro £ I co CO o JD OJ co 2 TO >- CD co co

co OJ cu cn CO _ro XI ra CU O & Q. cu cj ro xi cu >. o cu cn I— o cu ro CL .2 < CU o a> > CL u o cu o CO CO

24 SUMMER Model (June to November)

Forest Type: During summer both Mature Phase I (P = 0.0002) and Understory Reinitiation (P = 0.0002) forest classes were selected against (Table 4).

Edge Type and Contrast: Proximity to edge of Stem Exclusion (P = 0.0002) was selected against. In the summer model, locations in Understory Reinitiation forests with closest proximity to Stand Initiation II forests occurred 9.5 times more than expected randomly. The edge contrast of Young to Understory

Reinitiation forests occurred 5.8 times greater than expected, Young forests to Stand Initiation I forests occurred 0.15 times less than expected.

Distance to Edge: Short edge distances (P = 0.0028) was positively selected for. Average edge distance in summer was 159.0 ± 17.2 m (95% CI) compared to 286.0 ± 5.2 m (95% CI) in the random sample

(Figure 7).

Distance to Water: Distance to water was not a significant variable during the summer.

Elevation/Slope/Aspect: The average elevation was 409.0 ± 18.3 m (95% CI) compared to 599.0 ± 4.7 m

(95% CI) found randomly (Figure 8). Slope class 1 was negatively selected for (P = 0.0188).

WINTER Model (December to February)

Forest Type: Mature Phase I (P = 0.0002), Understory Reinitiation (P = 0.0002), and Stand Initiation I

(P = 0.0054) forest age classes were selected against by cougars during winter (Table 4).

Edge Type and Contrast: Old Growth edge type was positively selected for (P = 0.0026). There were

no other significant edge contrasts used more or less than expected during the winter.

Distance to Edge: The average distance to edge 157.0 ± 30.6 m (95% CI) was significantly shorter than the random sample 286.0 ± 5.2 m (95% CI) (Figure 7).

Distance to Water: There was no selection for distance to water.

25 Elevation/Slope/Aspect: Low elevations in the winter were selected for (P = 0.0002). The average elevation used by cougars in winter was 361.0 ± 20.2 m (95% CI) compared to 599.0 ± 4.7 m (95% CI) in the random sample (Figure 8). Slope and aspect were not selected for or against during the winter.

YEAR-ROUND Model

Forest Type: There was also selection against use of Mature Phase I (P = 0.0002), Stand Initiation I forest class (P = 0.031), and Understory Reinitiation forest age classes (P = 0.0466) (Table 4).

Edge Type and Contrast: There was selection for Young edges (16-30 years, P = 0.036) and Old

Growth edges (250 + years). There were no meaningful edge contrasts for the year-round model.

Distance to Edge: Selection (P = 0.014) for shorter edge distances occurred more than expected (Figure

7). On average, cougars used locations that were 153.0 ± 6.3 m (95% CI) away from an edge, whereas the average distance to edge for the random points was 286.0 ± 5.2 m (95% CI).

Distance to Water: There was no selection for distances to water than what was found randomly.

Elevation/Slope/Aspect: Cougars also showed significant selection (P = 0.0002) for lower elevation

areas (Figure 8) with an overall mean elevation of 393.0 ± 5.9m (95% CI) versus 599.0 ± 4.7 m (95% CI)

for random points. There was no selection for slope or aspect. Selection for slope and aspect was not

significant.

26 DISCUSSION Mature Phase I and Understory Reinitiation

All models found selection against Understory Reinitiation and Mature Phase I forests, with the exception of the Spring model that found selection only against Mature Phase I forests. This result could be either an artifact of the data or a true reflection of selection against these forest classes. The delineations of the forest age classes into habitat categories were based on biologically meaningful definitions that were refined by habitat and deer experts (see Methods). However, the proportions of each forest age class in my study area were not equal (Table 1). The lowest two percentages were Understory Reinitiation, and

Mature Phase I forests (with the exception of alpine and rock). In my analysis, 1285 actual points were compared to 18000 random points. The likelihood of the 18000 points identifying the small percentages of Understory Reinitiation and Mature Phase I forests is greater than the ability of the 1285 used points to characterize this rareness. The fact that these two stands represent such a small portion of the forest types may indicate that cougars do not have to cross through them incidentally in their travels either - exacerbating the avoidance of these forest types. Thus, selection against these classes may occur simply by chance because the proportions of each class are so low. The option of combining these two classes

into the other classes would have defeated the original purpose of classifying habitat based on

biologically meaningful differences.

On the other hand, cougars may select against these two classes because of their forest characteristics despite the fact that these forest categories occupy such a small percentage of the area. I speculate that

Understory Reinitiation and Mature Phase I do not have established understories as the older stands or younger stands (see Methods). These features may make the forests unattractive for black-tailed deer

because it does not offer good cover or forage. Cougars may respond to lack of prey and cover by

avoiding Understory Reinitiation and Mature Phase I forest classes.

27 Forest Age Class, Edge Type and Edge Contrast

There were seasonal and annual differences for selection of particular forest classes, as well as specific edge types and edge contrasts. I expected that cougars may use edge types or edge contrasts because the different habitat types available on either side of the edges allow some wildlife to access different resources such as food and cover in close proximity (Clark and Gilbert 1982). In the case of cougars, edge contrasts probably act as better areas for stalking cover for these predators. Belden et al. (1988), for example, found cougars (P. concolor) in Florida utilizing mixed swamps and hammock forest edges, apparently because of better stalking cover. Laing (1988) in Utah also found that cougars would use edge between open and forested areas, possibly to observe and ambush prey. However, use of edges is only

useful to cougar if prey is found at these edges. This appears to be the case in Pacific Coastal forests where combined resources at edges appear to be selected by black-tailed deer (O. h. columbianus)

(Kremsater and Bunnell 1992; Nyberg and Janz 1990; Voller 1998). Below, I will discuss forest age

class, edge type and edge contrast selection for each of 3 seasons, and annually.

Spring (March to May)

During spring, cougar used Stem Exclusion forests less than expected at random. Although this forest

age class has high thermal cover, it contains only minor amounts of forbs, herbs, and shrubs and is likely

less attractive to prey, particularly because during spring, deer are concentrating on optimizing forage

intake. Spring herb and browse species for deer have a lot of new highly digestible growth. This is the

prime season for deer to gain weight lost during the winter (Nyberg and Janz 1990). It is also the fawning

season, and energy is allocated to giving birth and lactating. Cougars are also using edge contrasts of

Young forests to Stem Exclusion forests 0.36 times less than expected randomly. Young forests offer

both high security cover, and food, but there is no large benefit to being close to Stem Exclusion forests.

This is in contrast to the 10 times increase over random points in cougars selecting the Clearcut forest

class (Stand Initiation I forests) adjacent to Stem Exclusion forest age class. While hunting deer, cougar

1 28 use of this edge contrast would be a good strategy because deer would be seeking fresh new forage in

Stand Intitiation I that offer very little cover, but could remain in proximity to Stem Exclusion forests for high security value. The appearance of selection for an edge contrast of Stem Exclusion forest age class, while there is negative selection for Stem Exclusion forest age classes, indicates that 1) the edge contrast involving Stem Exclusion, although rarer is being selected for when compared to the other possible edge contrasts, and 2) the nature of the logistic regression in which edge contrasts were not included is multivariate. It does not analyze for one habitat variable (i.e., forest age class) alone. It will assess selection based on the interactions of all the habitat variables being looked at.

Unlike winter, when cougars avoided Stand Initiation I forest classes, cougars showed neutral selection during spring and summer for Stand Initiation I likely because the class has abundant opportunities for deer, cougar's primary prey, to obtain forage (Harestad et al. 1982). In the winter, cougar and their prey usually face unfavourable conditions for feeding and winter traveling, as well as lower thermal cover in open areas such as Stand Initiation I. However, during the spring, cougars remaining in proximity to cover enhance their ability to hunt in Stand Initiation I can be advantageous because they can kill prey that are taking advantage of the consistently higher quality food available there over forested areas

(Harestad et al. 1982). For example, cougars select for the edge contrast of Mature Phase I forests to

Stand Initiation I forests 10.4 times more than expected randomly. For cougars, this contrast has the

features of hunting concealment cover in Mature Phase I and prey in Stand Initiation I. For deer, Mature

Phase I forests offer medium food and medium security, but Stand Initiation I are attractive because it has

dense new forage. Alternating between forage and cover areas of deer, optimizes the hunting ability of

cougars.

Summer (June to November)

During summer, in the context of expected deer behaviour, cougars may use forest ages that have high

quantity rather than quality of forage that deer use to build their fat reserves for survival through the rut

29 and winter (Nyberg and Janz 1990). Cougars take advantage of prey in areas of abundant forage, as well as the concealment cover close by. For example, being in the edge contrast of Understory Reinitiation forests with proximity to slightly older clearcuts (Stand Initiation II) that have high quantities of forage, occurred 9.5 times more than expected during the summer. Forests of type Stand Initiation II offer high food, while the Understory Reinitiation forests have high thermal cover and high security for deer (D.

Janz pers. comm. B.C. Ministry of Environment, Lands and Parks, Nanaimo 1999). Being in the edge contrast of Young forests with proximity to Understory Reinitiation forests occurred 5.8 times more than expected randomly. There is good security and high food in the Young forests, but the ease of movement and thermal cover for deer and concealment cover for cougar in the Understory Reinitiation class offers advantages for both deer and cougar. This habitat association is also evident for cougars because they occur 0.15 times less in edge contrast of Young forests with forage, and characterized with good security cover, with closest proximity to Stand Initiation I. Both of these forests offer good food value to deer and may not provide an optimization of different habitat adjacent to one another.

There was a significant lack of cougar locations with edge types of Stem Exclusion forests in which

understory plants are almost non-existent. This is possibly because their main prey, deer, is finding high

quantities of forage elsewhere. Stem Exclusion forests provide high thermal cover but have low food

value, and may be less important to deer in the summer when other forest classes offer both forage and

cover.

Winter (December to February)

Cougars chose Old Growth forest edge types and not Young forest edge types during winter, contrary to

my hypothesis (Table 4). This could be because Old Growth forests may be easier for travel during the

winter. Furthermore, the Old Growth forests may provide alternative foods for deer when the ground is

covered with snow. For example, when younger forests (such as Stand Initiation 1) are covered with

snow, several principal forages of deer are buried (Nyberg and Janz 1990; Schwab et al. 1987) and deer

30 feed on arboreal lichens on older trees (Nyberg and Janz 1990) as well as litterfall and tops of tall shrubs

(Harestad et al. 1982). Old Growth edge types may provide access to this alternative food for deer, and thus may provide further enticement for cougars to use this edge type during the winter.

Additionally, snow imposes thermal and traveling constraints in open areas in winter. This is reflected in the selection against the Stand Initiation I (Clearcut) in the winter and year-round. Clearcutting may be the most serious threat to cougar populations in winter because it can lower the prey base (Smallwood

1994) if there are extensive and diminished areas of winter ranges. This is because Clearcuts, especially

larger ones, will offer little food if the area is covered under snow, has no thermal cover, and creates difficult travel for both deer and cougar. Particularly in winter, lack of prey and stalking cover is pronounced in Clearcuts, dissuading both cougar and deer presence. Furthermore, open areas tend to be avoided by cougars because of low concealment cover for hunting (Seidensticker et al. 1973).

Seidensticker et al. (1973) found that cougars would rather travel along the forested boundaries of an

immense open area, than walk through it. Laing (1988) also found that cougars will travel through Young

forests, but they never crossed Clearcuts due to the lack of cover.

Year-Round

Year-round, cougars had a significantly greater number of locations in Old Growth and Young forest edge

types than expected (Table 4). Snow depth increases with decreasing canopy cover (Harestad et al.

1982). Thus, Old Growth forests have the benefit of snow interception and thermal cover to cougar and

their prey, with the benefit of lichen and conifer litterfall as food for deer. Young forests provide medium

security cover for deer and good concealment against prey for cougar. The security cover of Young

forests is conducive to deer that are spending large amounts of time seeking forage available in Young

forests particularly in summer. It also enhances deer ability to hide, while foraging. If cougars are

foraging optimally, one can expect that they should stay close to edges that deer use, such as Young or

Old Growth edge types, while still being associated with cover for stalking and hunting prey (Koehler and

31 Hornocker 1991; Logan and Irwin 1985; Seidensticker et al. 1973). Year-round, cougars tend to avoid the use of Stand Initiation I forest age class (Clearcuts). Although there are benefits to being in Clearcuts

(such as a dense prey in the summer), the negative effects of Clearcuts (i.e., low stalking cover) may outweigh the benefits (as discussed in the Winter section), causing the overall year-round effect to be negative.

Distance to Edge

Cougars making decisions based on prey habits are most interesting when examining the distances to the nearest edge. The similarities to deer studies (Chang et al. 1995; Hanley 1983; Kremsater and Bunnell

1992; Nyberg and Janz 1990) are almost identical for year-round as well as seasonal habitat use. Whereas random points were approximately 275 m from the closest different forest type (edge), cougar locations were approximately 150 m from edges (Figure 7) both year round and seasonally. Kremsater and Bunnell

(1992) found that distinct edge differences resulted in distinct deer utilization, and that deer were concentrated within 100 m of cover, possibly encouraging cougars to stay within a similar edge width - in my study area, 150 m. Deer would use edges more if cover and forage were typically more available

from two adjacent forest types (Kremsater and Bunnell 1992). Several studies have found greater use by

black-tailed deer along Stand Initiation I, Young, or Old Growth edges, than in the interior of any of these

forest age classes (Hanley 1983; Nyberg and Janz 1990; Kremsater and Bunnell 1992; Chang et al. 1995).

Chang et al. (1995) also found selection for deer in proximity to Old Growth and Young edge types, with

the average edge distance of 135 m. Although Chang et al. (1995) studied Sitka deer (O. h. sitkensis),

their results are very similar to the behaviour of cougars in my study area in terms of edge relationships,

and edge distances.

Distance to Water

From early observational studies, cougars can withstand long periods without water, and thus do not need

to be near a major water source (Young 1946). However, the need for water often draws ungulates close

to water edges (Nyberg and Janz 1990). This in turn could cause high association of cougars to water as

32 well, as is found with jaguars (P. oncd) (Crawshaw 1991). This was not the case in my study area.

Distance to water may be unimportant to deer because water is readily available throughout coastal regions of British Columbia. In either case, proximity to water does not seem to be a major factor in

macrohabitat selection by cougars.

Elevation

I hypothesized that cougars would select both mid and low elevations to minimize bioenergetic costs and

to maximize access to dense prey. Cougars should avoid high elevations due to colder conditions and the

higher availability of prey at mid and low elevations. Cougars did not select low elevations, but

concentrated their movements at mid elevations (c. 200-500 m) for each season and year-round compared

to what would be expected based on random points (Figure 9). At mid elevations, cougars may be able to

maximize edge contrasts of old and younger forests in order to hunt. The study site has the general

characteristic of younger forests at lower elevations, and older forests in mid and high elevations due to

forest harvesting. With many of the Old Growth forests in my study area starting at the mid elevations,

maximization of Old-Young edge contrasts would be greatest at this elevation. Like my study, Logan

(1983) found disproportionate cougar use of elevations as deer changed from low, mid and high

elevations to forage.

Although deer will use elevations from sea level to alpine, cougars may not select areas on the basis of

prey density alone and may be affected by other inter or intraspecific factors that are present at low

elevations (Seidensticker et al. 1973). Cougars did not select very high elevations likely due to a poor

prey base (Logan 1983; Nyberg and Janz 1990) in the winter. Unstable or steep slopes during winter, and

deep snow packs are prevalent at high elevations making high elevations bioenergetically demanding for

travel by cougar and deer (Laing 1988; Nyberg and Janz 1990). The lack of available forage, adequate

thermal cover, and hiding cover make high elevations even more unsuitable during the winter. During the

summer however, deer tend to use higher elevations, and thus one would expect that cougars following

33 their prey habits would do so as well. This was not the case. Cougars use similar elevations throughout the year, suggesting that there may be features or mechanisms there that maximize feeding success at mid-elevations. Perhaps the combination edge types that are primarily available at low-mid and mid elevations allow cougars to maximize concealment cover and thus hunting success.

Slope and Aspect

Slope or aspect classes were not selected seasonally and year-round. I expected there to be selection for south facing aspects in winter. This may not have occurred because deer tend to concentrate use on south facing aspects only in very cold winters in response to critical thermal constraints (Nyberg and Janz 1990) and a global El Nino existed for the duration of our study. For cougar, there are no distinct seasons in terms of birthing, reproducing etc., suggesting less thermal constraints than seasonal animals, and perhaps

less of a motive to use warmer facing aspects other than for access to prey. Furthermore, if cougars travel quickly among watersheds, I might have encountered difficulties detecting selection for warmer south- facing aspects, assuming that travel and hunting time are equal.

I also expected cougars to use steep slopes to enhance predation success. This was not found, and I think

it has to do with either: 1) the vegetation type of the forest; or 2) the pixel size of 20 m used in the

analysis of slope. In areas where vegetation was sparse, cougars use bluffs or slopes for concealment

(Ross and Jalkotzy 1992). In my study area, vegetation is much denser and the need for slope may be less

important than vegetative concealment. However, it is also possible that pixel size used in my analysis

does not incorporate slopes at scales that cougars use. A slope with a pixel size of 20 m can mask

topography features (i.e., edges of roads, small bluffs and cliffs) that cougars are known to use elsewhere

(Sunquist and Sunquist 1989). In this case, cougars not selecting slope may be a construct of the GIS

analysis. Further micro-site examination of topographical features used by cougars (such as road sides,

small bluffs) would suggest a more biologically meaningful pixel size.

34 300

Elevation (m)

3000

CO 2000 c g "-«—* CO o o _l E o T3 c ocor 1000

o tz CU

CT CU

Elevation (m)

Figure 9. Frequency histogram of cougar locations (top) and random points (bottom). Cougar locations are concentrated at c. 200-500 m with an average of 350 m, whereas random locations are concentrated at c. 350-700 with an average of 600 m. Cougars showed significant selection (P = 0.0002) for lower elevation areas with an overall mean elevation of 393.0 ± 5.9 m (95% CI) versus 599.0 ± 4.7 m (95% CI) for random points. REFERENCES

Anderson, A. E. 1983. A critical review of literature on puma (Felis concolor). Special Report No. 54. Colorado Division of Wildlife, Denver. 91 pp.

Andersen, M. C. and D. Mahato. 1995. Demographic models and reserve designs for the California spotted owl. Ecological Applications 5:639-647.

Beier, P. 1991. Cougar attacks on humans in the United States and Canada. Wildlife Society Bulletin 19:403-412.

Beier, P. 1995. Dispersal of juvenile cougars in fragmented habitat. Journal of Wildlife Management 59:228-237.

Beier, P., D. Choate, and R. H. Barrett. 1995. Movement patterns of mountain lions during different behaviors. Journal of Mammalogy 76:1056-1070.

Belden, R. C, W. B. Frankenbergen, R. T. McBride, and S. T. Schwikert. 1988. Panther habitat use in southern Florida. Journal of Wildlife Management 52:660-663.

Branch, L. C. 1995. Observations of predation by pumas and Geoffroy's cats on the plains vizcacha in semi-arid scrub of central Argentina. Mammalia 59:152-156.

Bryant, A. A. 1996. Reproduction and persistence of Vancouver Island marmots (Marmota vancouverensis) in natural and logged habitats. Canadian Journal of Zoology 74:678-687.

Bryant, A. A., and D. W. Janz. 1996. Distribution and abundance of Vancouver Island marmots (Marmota vancouverensis). Canadian Journal of Zoology 74:667-677.

Cederholm, M. 1999. Random points Arcview Script.

Chang, K. T., D. L. Verbyla, and J. J. Yeo. 1995. Spatial analysis of habitat selection by Sitka Black- tailed Deer in southeast Alaska, USA. Environmental Management 19:579-589.

Clark, T. P., and F. F. Gilbert. 1982. Ecotones as a measure of deer habitat quality in central Ontario. Journal of Applied Ecology 19:751-758.

Crawshaw Jr., P. G. 1991. Jaguar spacing, activity, and habitat use in a seasonally flooded environment in Brazil. Journal of Zoology (London) 223:357-370.

DeLaune, M. 1999. XTOOLS Arcview extension. Oregon Department of Forestry, USA.

Elliot, J. P., I. McTaggart Cowan, and C. S. Holling. 1976. Prey capture by the African lion. Canadian Journal of Zoology 55:1811 -1828.

Eng, M. 1998. Spatial patterns in forested landscapes: Implications for biology and forestry. Pp. 42-75. In J. Voller and S. Harrison (eds.) Conservation Biology Principles for Forested Landscapes. B.C. Ministry of Forests. Canada.

36 Hanley, T. A. 1983. Black-tailed deer, elk, and forest edge in a western Cascades watershed. Journal of Wildlife Management 47:237-242.

Harestad, A. S., J. A. Rochelle, and F. L. Bunnell. 1982. Pp. 343-352. In the Forty-Seventh North American Wildlife and Natural Resources Conference.

Heezen, K. L. and J. R. Tester. 1967. Evaluation of radio-tracking by triangulation with special reference to deer movements. Journal oj Wildlife Management 31:124-141.

Hemker, T. P., F. G. Lindzey, and B. B. Ackerman. 1984. Population characteristics and movement patterns of cougars in southern Utah. Journal of Wildlife Management 48:1275-1284.

Hooge, P. N. and Eichenlaub, B. 1997. Animal Movement Analysis - Arcview extension.Alaska, Biological Science Center, USA.

Hornocker, M. G. 1969. Winter territoriality in mountain lions. Journal of Wildlife Management 33:457- 464.

Hornocker, M. G. 1970. An analysis of mountain lion predation upon mule deer and elk in the Idaho primitive area. Wildlife Monographs 21:1 -39.

Hosmer, D.W. and S. Lemeshow. 1989. Pp. 307. Applied Logistic Regression. New York. NY.

Hunter, M. D. and P. W. Price. 1992. Playing chutes and ladders: heterogeneity and the relative roles of bottom-up and top-down forces in natural communities. Ecology 73:724-732.

Hurlbert, S. H. 1984. Pseudoreplication and the design of ecological field experiments. Ecological Monographs 54:187-211.

Iriarte, J. A., W. L. Franklin, W. E. Johnson, and K. H. Redford. 1990. Biogeographic variation of food habits and body size of the America puma. Oecologia 85:185-190.

Julander, O. and D. E. Jeffery. 1964. Deer, elk, and cattle range relations on summer range in Utah. Transactions of the North American Wildlife Conference 29:404-413.

Karanth, K. U., and M. E. Sunquist. 1995. Prey selection by tiger, leopard and dhole in tropical forests. Journal of Animal Ecology 64:439-450.

Koehler, G. M. and M. G. Hornocker. 1991. Seasonal resource use among mountain lions, bobcats, and coyotes. Journal of Mammalogy 72:391-396.

Kremsater, L. L. and F. L. Bunnell. 1992. Testing responses to forest edges: the example of black-tailed deer. Canadian Journal of Zoology 70:1715-1725.

Kruuk, H. 1982. Interactions between Felidae and their prey species: a review. Pp. 353-374. In S.D. Miller and D.D. Everett (eds.) Proceedings of the Second International Symposium: Cats of the World: Biology, conservation, and management. USA.

Laing, S. P. 1988. Cougar habitat selection and spatial use patterns in southern Utah. Pp. 68. In M.Sc. thesis, University of Wyoming.

37 Laing, S. P. and F. G. Lindzey. 1993. Patterns of replacement of resident cougars in southern Utah. Journal of Mammalogy 74:1056-1058.

Lamberson, R., B. R. Noon, C. Voss, and K. S. Mckelvey. 1994. Reserve design for territorial species: The effects of patch size and spacing on the viability of the northern sported owl. Conservation Biology 8:185-195.

Lehmkuhl, J. F. and M. G. Raphael. 1993. Habitat pattern around northern spotted owl locations on the Olympic peninsula Washington. Journal of Wildlife Management 57:302-315.

Lenth, R. V. 1981a. Robust measures of location for directional data. Technometrics 23:77-81.

Lenth, R. V. 1981 b. On finding the source of a signal. Technometrics 23:149-154.

Lindstedt, S. L., B. J. Miller, and S. W. Buskirk. 1986. Home range, time, and body size in mammals. Ecology. 67:413-418.

Lindzey, F. G., B. B. Ackerman, D. Barnhurst, and T. P. Hemker. 1988. Survival rates of mountain lions in Southern Utah. Journal of Wildlife Management 52:664-667.

Lindzey, F. G., W. D. Van Sickle, B. B. Ackerman, D. Barnhurst, T. P. Hemker, and S. P. Laing. 1994. Cougar population dynamics in Southern Utah. Journal of Wildlife Management 58:619-624.

Lofroth, E. 1998. The Dead Wood Cycle. Pp. 185-214. In J. Voller and S. Harrison (eds.) Conservation Biology Principles for Forested Landscapes. B.C. Ministry of Forests. Canada.

Logan, K. 1983. Mountain lion population and habitat characteristics in the Big Horn Mountains of Wyoming. Pp.101. In M.Sc. thesis, University of Wyoming.

Logan, K. and L. Irwin. 1985. Mountain lion habitats in the Big Horn Mountains, Wyoming. Wildlife Society Bulletin 13:257-262.

Logan, K. A., L. L. Irwin, and R. Skinner. 1986. Characteristics of a hunted mountain lion population in Wyoming. Journal of Wildlife Management 50:648-654.

Manly, B. F. J., L. L. McDonald, and D. L. Thomas. 1993. Pp. 177. Resource selection bv animals: statistical design and analysis for field studies. Chapman & Hall. Britain.

Maser, C. and R. S. Rohweder. 1983. Winter food habits of cougars from northeastern Oregon. Great Basin Naturalist 43:425-428.

Mattson, D. J., S. Herrero, R. G. Wright, and C. M. Pease. 1996. Science and management of Rocky Mountain grizzly bears. Conservation Biology 10:1013-1025.

Maurer, B. A. 1985. Avian community dynamics in desert grasslands: observational scale and hierarchial structure. Ecological Monographs 55:295-312.

McDougal, C. 1977. Pp. 180. The face of the tiger. Rivington, London.

Meidinger, D. and J. Pojar. 1991. Pp. 330. Ecosystems of British Columbia. B.C. Ministry of Forests.

38 Menard, S. 1995. Pp. 98. Applied logistic regression analysis. Thousand Oaks, CA: Sage Publications. Series: Quantitative Applications in the Social Sciences, No. 106

Morris, D. W. 1987. Ecological scale and habitat use. Ecology 68:362-369.

Moy, A. 1999a. Distance to Edge Arcview Script. University of British Columbia, Canada.

Moy, A. 1999b. Slope Aspect Elevation Derivation Arcview Script. University of British Columbia, Canada.

Nams, V. O. 1989. Effects of radiotelemetry error on sample size and bias when testing for habitat selection. Canadian Journal of Zoology 67:1631 -1636.

Nyberg, J. B. and D. W. Janz. (eds.) 1990. Pp. 310. Deer and elk habitats in coastal forests of southern British Columbia. B.C. Ministry of Forests and B.C. Ministry of Environment. Crown Publications. Victoria, Canada.

Ockenfels, R. A. 1994. Mountain lion predation on pronghorn in central Arizona. The Southwestern Naturalist 39:305-307.

Oliver, C. D. 1992. A landscape approach: Achieving and maintaining biodiversity and economic productivity. Journal of Forestry 90:20-25.

Power, M. E. 1992. Top-down and bottom-up forces in food webs: do plants have primacy? Ecology 73:733-746.

Rabinowitz, A. R. 1990. Notes on the behaviour and movements of leopard cats, Felis bengalensis, in a dry tropical forest mosaic in Thailand. Biotropica 22:397-403.

Rabinowitz, A. R. and Nottingham, B. G., Jr. 1986. Ecology and behaviour of the jaguar (Panthera onca) in Belize, Central America. Journal of Zoology (London) 210:149-159.

Rodway, M. S., H. M. Regehr, and J-P. L. Savard. 1993. Activity levels of marbled murrelets in different inland habitats in the Queen Charlotte Islands, British Columbia. Canadian Journal of Zoology 71:977-984.

Ross, P. I., and M. G. Jalkotzy. 1992. Characteristics of a hunted population of cougars in southwestern Alberta. Journal of Wildlife Management 56:417-426.

Ross, P. I., M. G. Jalkotzy, and J. R. Gunson. 1996. The quota system of cougar harvest management in Alberta. Wildlife Society Bulletin 24:490-494.

Ruggiero, L. F., D. E. Pearson, and S. E. Henry. 1998. Characteristics of American marten den sites in Wyoming. Journal of Wildlife Management 62:663-673.

Sallee, K. L. 1999. Location Of A Signal. Ecological Software Solutions. USA.

Schaller, G. B. 1972. Pp. 480. The Serengeti Lion: A study of predator-prey relations. Chicago University Press. Chicago Press.

39 Schaller, G. B. and P. G. Crawshaw. 1980. Movement patterns of jaguar. Biotropica 12:161-168.

Schwab, F. E., M. D. Pitt, and S. W. Schwab. 1987. Browse burial related to snow depth and canopy cover in northcentral British Columbia. Journal of Wildlife Management 51:337-342.

Seidensticker, J. C, M. G. Hornocker, W. V. Wiles, and J. P. Messick. 1973. Mountain lion social organization in the Idaho primitive area. Wildlife Monographs 35:4-60.

Smallwood, K. S. 1994. Trends in California mountain lion populations. The Southwestern Naturalist 39:67-72.

Spreadbury, B. R., K. Musil, J. Musil, C. Kaisner, J. Kovak. 1996. Cougar population characteristics in southeastern British Columbia. Journal of Wildlife Management 60:962-969.

Stephens, D. W. and J. R. Krebs (1987). Pp 247. Foraging Theory. Princeton University Press, Princeton.

Squires, J. R., and L. F. Ruggiero. 1996. Nest-site preference of northern goshawks in southcentral Wyoming. Journal of Wildlife Management 60:170-177.

Sunquist, M. E. 1981. The social organization of tigers (Panthera tigris) in Royal Chitawin National Park, Nepal. Smithsonsian Contributions to Zoology 336:1 -98.

Sunquist, M. E. and F. C. Sunquist. 1989. Ecological constrainsts on predation by large felids. Pp. 283- 301 In Carnivore behavior, ecology, and evolution. Gittleman, J.L. (ed.). Cornell University Press, NY.

Telfer, E. S. and J. P. Kelsall. 1979. Studies of morphological parameters affecting ungulate locomotion in snow. Canadian Journal of Zoology 57:2153-2159.

Torres, S. G., T. M. Mansfield, J. E. Foley, T. Lupo, and A. Brinkhaus. 1996. Mountain lion and human activity in California: testing speculations. Wildlife Society Bulletin 24:451-460.

Toweill, D. E., and C. Maser. 1985. Food of cougars in the cascade range of Oregon. Great Basin Naturalist 45:77-80.

Voller, J. 1998. Managing for edge effects. Pp. 215-233. In J. Voller and S. Harrison (eds.) Conservation Biology Principles for Forested Landscapes. B.C. Ministry of Forests. Canada.

Wehausen, J. D. 1996. Effects of mountain lion predation on bighorn sheep in the Sierra Nevada and Granite mountains of California. Wildlife Society Bulletin 24:471-479.

White, G. C. and R. A. Garrott. 1990. Analysis of wildlife radio-tracking data. Academic Press, Inc. U.S.A. 383 pp.

Wielgus, R. B., F. L. Bunnel, W. L. Wakkinen and P. E. Zager. 1994. Population dynamics of Selkirk Mountain grizzly bears. Journal of Wildlife Management 58:266-272.

Young, S. P. 1946. Its history, life habits, economic status and control. Pp. unknown. In The Puma, mysterious American cat, (eds. S.P. Young and E.A. Goldman). Monumental Printing Company. USA.

40 APPENDIX I: Combinations and frequency of edge contrasts

There are 56 combinations of edge contrasts. The first digit in each cell represents a forest age class, and the second digit represents the closest different forest class. The two digits together represent an edge contrast relationship

-- 2-1 3-1 4-1 5-1 6-1 7-1 8-1 1-2 - 3-2 4-2 5-2 6-2 7-2 8-2 1-3 2-3 - 4-3 5-3 6-3 7-3 8-3 1-4 2-4 3-4 . „ 5-4 6-4 7-4 8-4 1-5 2-5 3-5 4-5 - 6-5 7-5 8-5 1-6 2-6 3-6 4-6 5-6 - 7-6 8-6 1-7 2-7 3-7 4-7 5-7 6-7 - 8-7 1-8 2-8 3-8 4-8 5-8 6-8 7-8 ~

41 eons u ** a o u cm 01 •D LU

u a O s cr

a o

,?o OH 5l

42 APPENDIX II: Probability Functions

When the best model has been chosen, a probability equation can be created from the model. This equation is used to test if other locations provide similar cougar habitat results given your model of choice. This probability tells you how likely a site is to be good cougar habitat.

The probability equation is much like the simple regression equation in that it incorporates the intercept

(constant) and the beta coefficients of the significant variable:

Equation = Constant + B,(Variable 1) + B2(Variable 2) + ....+ Bn(Variable n)

e ua,ion e uation Probability = e i /( 1 +e « )

The higher the value of an equation, the higher the probability of cougar occurrence assuming that your model is correct. Recall that a negative coefficient indicates selection against a habitat variable. Thus, if a location had several traits that were classified as negative coefficients in the model (i.e., independent variables that were negatively selected for), the equation will have a low number. That low number will then have a low probability of occurrence. Likewise, if a location has traits that were positively selected for, the equation will have a large sum, and the probability that the spot is good cougar habitat, also increases.

The RSF equations for each season, and all year round are:

Summer = (-4.98)(maturel)+(-3.68)(understory)+(-0.00088)(distance to edge)+(-0.0022)(elevation)+(-0.34)(Slope classl)+(0.18)(Edge-young)+(-3.3)(Edge-stem)+(-1.1 )(Edge-primary)+(-1.2)(primary)+(-0.033)(Slope class5) - 2.16

Spring = (1.2)(young)+(-4.6)(maturel)+(-0.096)(Edge-young)+(-0.0022)(Distance to Edge)+(- 0.0024)(Elevation)+(-2.2)(Stem)+(-l .96)(primary)+(0.687)(old)+(-l. 169)(Edge-primary)+(0.104)(Edge-old) - 3.77

Winter = (-6.0)(mature l)+(0.31)(Edge-young)+(0.29)(Edge-mature2)+(0.73)(Edge-old)+(-

0.0029)(Elevation)+(0.14)(N Aspect)+(-4.160XUnderstoiy)+(-1.91)(Edge-primaiy)+(-3.5274XPrimary)+(-0.94)(W aspect) - 2.83

Yearly Model = (2.4)(year)+(-0.56)(primary)+(-5.4)(Maturel)+(-0.00093)(Distance to Edge)+(- 0.0024)(Elevation)+(0.250)(Slope_3)+(0.148)(Slope_l)+(0.034)(Slope_5)+(0.523)(Edge_old)+(0.24)(Edge_SI I)+(0.264)(Edge-young)+(-l .53)(Understory) - 3.77

43 (c0

O0 (0 « CD CO

9.2 co CM « S

•- (0 a> c .2 S « CO 5? O

CO CO 5 oCO CO O

il CO c 3 CO o O o 3

~~CD 00 CD o Co N o i « CD 3 * 8—~~

~~I co g CO > CS O J3 & .3 es o >• CO CO oo o. CN o CS E o co T3 * o~~

~~41 a> CO CD 9 .s a CD CD co lO > 42 ra es JS n CD CO — 00 O CO ra m CO CD 'ra CN *~ £ a > § <2 ra~~

~~a CL^> M C3 •*-» ra c (0 CM CD ra 0) o (0 CD O CL a> ra T3 < CL) 4-* (0 12 2 a> CD o TJ 3 s 5l u. (0 o~~

~~44 APPENDIX IV: Home Range Sizes of Vancouver Island Cougar~~

~~Average cougar home range size was 273.7 ± 86 km2. Time ranges for each cougar are on Figure 5.~~

~~Individual cougar home range sizes are as follows for each of the 9 female cougars followed:~~

~~CAT ID Yearl Year 2 2 298.548 128.878 3 510.3974 124.3305 5 393.3207 146.8457 7 250.6282 0 8 213.7644 0 12 795.8504 278.3391 17 132.1558 241.3978 21 314.5902 133.4242 23 280.897 135.8293~~

~~45 APPENDIX V: Assumptions~~

The model was tested for logistic regression assumptions (Menard, 1995). The first assumption is that the model is correctly specified in its functional form and all relevant independent and no irrelevant independent variables are included. The model is extremely robust to differences in form (Hosmer and Lemeshow 1989; Menard 1985), and thus this specification is the least concerning and is assumed. The second assumption is nonlinearity in the logit form. In logistic regressions, a change in the logit of the dependent variable is associated with a change in the logit of the independent variable, not its actual values. The third assumption is non-additivity. In logistic regressions, a change in the dependent variable is associated with a change in the independent variable, but is not related to the value of one of the other independent variables. If there is a relationship, then there is additivity. Logistic models assume non-additivity. Unless there are theoretical reasons to suspect additivity (i.e., obvious interaction effects), it is assumed (Menard, 1995). The last major assumption is low collinearity. That is that the correlations between any two independent variables are not highly corrected (R2=0.8 or higher). If they are highly correlated, then the beta coefficients for the independent variables may be biased (Menard, 1995). Beyond the assumptions, the analysis of residuals within the model identify cases for which the model works poorly or cases that exert more weight on the model than assumed (Menard, 1995). By checking the diagnostics of the model, one can minimize the weaknesses in the conceptual model. Leverage values for each entry are between 0 and 1, with 0=no influence, and

l=completely influenced. Values greater than (k+l)/N are considered influential. Huge deviations from that value should be reviewed more carefully. Studentized residuals should fall within a range of either -3, or +3, otherwise they are a bad fit. Finally, the Dfbeta describes changes in the individual coefficients when a case is deleted.