Gbp 2003 03 annrpt 2002

FOOTHILLS MODEL FOREST GRIZZLY BEAR RESEARCH PROGRAM

2002 ANNUAL REPORT (year 4 of a 5 year study)

Prepared and edited by Gordon B. Stenhouse, Robin Munro and Karen Graham March 2003

Citation: Stenhouse, G., Munro R., and K. Graham. 2003. Foothills Model Forest Grizzly Bear Research Program 2002 Annual Report. 162 pp. This is an interim report not to be cited without the express written consent of the senior author.

Disclaimer

This report presents preliminary findings from the first four years of a five-year study on grizzly bears in the Yellowhead Ecosystem. It must be stressed that these data are preliminary in nature and represent data collected during the first three field seasons. All findings must be interpreted with caution. Opinions presented are those of the authors and collaborating scientists and are subject to revision based on the ongoing findings over the course of this study.

EXECUTIVE SUMMARY The FMF Grizzly Bear Research Project focuses on management issues and questions by assessing grizzly bear populations, bear response to human activities, and habitat conditions to provide land managers with tools to integrate grizzly bear “needs” into the land management decision making framework. The study area is approximately 9900 km2 and covers a portion of both mountainous and foothills habitats. A strong gradient in land-use activities and human disturbance exists across the study area. Currently, oil and gas exploration and development, forestry, mining, hunting, settlement, tourism, and recreation dominate the human land use practices and activities. In 2002, a total of 28 grizzly bears were handled, of which 21 (7 M/14 F) were fitted with GPS radio collars. We recollared 16 bears (4 M/12 F) that were originally collared during the first three years of this program. In 2002, there were six known mortalities within our study area. ANIMAL HEALTH: PROGRESS REPORT Our major objective was to assess the effects of prolonged stress on the health of grizzly bears in the study area, by seeking biochemical indicators to quantify the level of chronic stress in a bear and applying multivariate analyses on existing data to seek significant associations among measures of the health and environment of grizzly bears. Effort was expanded to collaborate with, and gain support from, researchers working on similar types of issues in polar bears. The application of multivariate analyses on existing health and environmental data of grizzly bears was initiated in October 2002. A first and necessary step to developing health profiles has been to link all health and environmental data collected from 1999 to 2002 into a single data file that is amenable to statistical analyses. This formidable task was completed in January 2003, and statistical analyses are presently in progress. We also compared select health data between the Foothills Model Forest (FMFGBP) and the Eastern Slopes Grizzly Bear Projects (ESGBP) in effort to determine why female grizzly bears of the ESGBP have an especially long interval between litters (4.4 years) and extremely low reproductive rates (i.e., 0.21 = the average number of viable offspring each adult female is expected to produce each year) (Herrero 2001). Results from a comparison of body condition and reproductive hormone concentrations between Eastern Slopes and FMF bears did not remove the possibility that reduced reproductive output (long interval between litters and low reproductive rate) in Eastern Slopes grizzly bears is a result of low energy uptake (especially in males) causing diminished reproductive function.

i

The other aspect of animal health research initiated this past year was to evaluate the efficacy and safety of the drug combination medetomidine-Telazol (MZT), and the antagonist drug atipamezole, for reversible anesthesia in grizzly bears. These drugs have proven to be excellent for reversible anesthesia of free-ranging polar bears (Cattet et al. 1997), black bears (Caulkett and Cattet 1997), and Scandinavian brown bears (Arnemo et al. 2002). However, preliminary data from five bears captured for the FMFGBP during 2002 do not appear to support the findings of these earlier studies. Although limited results from 2002 suggest MZT and atipamezole do not provide effective reversible anesthesia in grizzly bears, it may be premature to make final conclusions. THE DIET OF GRIZZLY BEARS IN WEST-CENTRAL ALBERTA Between 2001 and 2002, scats from 11 different grizzly bears (3M 8F) were collected during field investigations of GPS radiolocation sites. Based on shifts in the diet of grizzly bears after 2 years of data collection, three seasons were identified for the active part of the year. Roots and ungulates dominated the pre-green up season, or early spring and ends at the end of June. Green vegetation then dominates diet in the late spring until the end of July. Fruits and hedysarum became more voluminous in scats at the beginning of August until week 4 of September. Considerable variation in scat content was noted among years. GRIZZLY BEAR HABITAT USE OF CLEARCUTS IN WEST-CENTRAL ALBERTA: INFLUENCE OF SITE, SILVICULTURE AND LANDSCAPE STRUCTURE

We examined habitat selection for clearcuts in the west-central Alberta foothills by grizzly bears for 3 seasons: hypophagia, early hyperphagia, and late hyperphagia. We compared grizzly bear global position system (GPS) radiotelemetry data with random locations throughout a defined study area, compared the distribution of radiotelemetry locations between daytime and nighttime periods, and developed specific habitat selection models for clearcuts by examining only those locations within clearcuts. Overall when comparing use of clearcuts with matrix habitats, we found that grizzly bears used clearcuts with respect to their availability for hypophagia and late hyperphagia, while selection was evident during early hyperphagia. Selection for clearcuts, however, appeared to be occurring more during nighttime periods than daytime periods suggesting that security issues may be involved. Specific models of habitat selection in cutblocks revealed that silviculture, micro-site terrain, and landscape metrics were important predictors of grizzly bear use for hypophagia and late hyperphagia, while the micro-site terrain model was selected for early hypophagia. Grizzly bears tended to avoid clearcut interiors, heavily using forest edges, while also using clearcuts with high perimeter to edge ratios suggesting security or perhaps ecotone relationships are operating. The impacts of various silvicultural/site preparation treatments were varied and are discussed in more detail.

ii

Finally, age of clearcut was an important predictor of bear use, with intermediate aged clearcuts used most during the first two seasons, and recent and old clearcuts used more during the final season. We discuss the implications of these findings and attempt to relate patterns of use with food resources. DISTRIBUTION OF GRIZZLY BEAR FOODS IN CLEARCUTS AND REFERENCE FOREST STANDS OF WEST-CENTRAL ALBERTA During the summer of 2001 and 2002, we assessed the occurrence and productivity of grizzly bear foods within 355 clearcuts and 183 reference upland conifer stands of west-central Alberta. Ants, Equisetum, Hedysarum, Taraxacum officinale, and Vaccinium myrtilloides occurred with statistically greater frequency in clearcuts compared with upland forests. In contrast, Vaccinium caespitosum, V. membranaceum, and V. vitis-idaea were more likely to occur in upland forests. Finally, Arctostaphylos uva-ursi, Shepherdia canadensis, and ungulate pellets did not differ in occurrence between clearcuts and reference forests. We discuss specific relationships to scarification, micro-site terrain, and canopy in more detail, while providing parsimonious models predicting their occurrence. Model validation suggested that predictions are consistent with independent data. The occurrence of fruit production for berry producing species, given the presence of the species, generally optimized at intermediate levels of canopy cover with little evidence for other relationships. In total, fruit production for 5 species within an average clearcut hectare (ha) was estimated at 22.88 kg. We suggest that food resources in clearcuts are substantial and likely the mechanism driving habitat selection for these areas. GRIZZLY BEAR FOOTHILLS HABITAT FRAGMENTATION BY SEISMIC CUTLINES: PRELIMINARY REPORT ABOUT ‘PARSIMONY OF LANDSCAPE METRICS’ AND ‘CUTLINE EFFECTS ON HABITAT STRUCTURE AND FOOTHILLS GRIZZLY BEAR LANSCAPE USE’ In this preliminary report, we list and recommend landscape-level and class-level metrics, which quantify most of the spatial pattern within the foothills portion of the grizzly bear research project, for generalized landcover types. This list is the result of a detailed parsimony study based on statistical analyses such as principal component and clustering techniques. We also investigated the effects of seismic cutlines on the landscape structure, finding that they cause direct linear increases in edge density and number of patches and power function decreases in mean patch size. These changes appear to be much stronger in landscapes with low natural, or original fragmentation levels.

Inspecting the effects of seismic cutlines and spatial patterns on grizzly bear population landscape use, we found cutlines to be not an important predictor of landscape use.

iii

However, the spatial patterns associated with adding seismic cutlines to the landscape, such as an increase in the variation of inter-patch distances, appear to be important in determining declining population-level landscape use, therefore suggesting seismic cutlines to be important indirect predictors of landscape use. GRAPH THEORY DEVELOPMENT, COST SURFACE ASSESSMENT AND INDIVIDUAL LEVEL CONNECTIVITY ANALYSIS Graph theory is a heuristic methodology, which allows researchers to quantify landscape connectivity at multiple temporal and spatial scales. By representing habitat mosaics as a mathematical ‘graph’ (Keitt et al. 1997), the spatial configuration of patches, level of patch importance, connections, and movement (dispersal) matrices can be analyzed. At the habitat scale, the analysis revealed variable levels of connectivity for individual bears. However, overall results demonstrated that in areas with increased human presence and disturbance, basic habitat connectivity and levels of graph structure decrease. Connectivity results were largely determined by the daily movement rate employed to define functional linkages between habitat patches. Further comparisons could be conducted using the mean daily movement rate for all bears across 1999 and 2000 for additional graph generation. Overall, the research can provide land managers with a validated modeling approach/tool for continued analysis of impacts associated with human development on wildlife populations, including an opportunity to incorporate grizzly bear needs into critical planning. A COMPARISON OF MAPPING PRODUCTS FOR PREDICTING GRIZZLY BEAR HABITAT QUALITY IN WEST-CENTRAL ALBERTA Currently, there is strong divergence in the reliance and use of different mapping products for predicting grizzly bear (Ursus arctos L.) habitats in North America. To date, research studies and government mapping programmes have relied on numerous land use/cover products and methodologies. These include, ecoclassification maps derived from terrain, climate, and vegetation, aerial photographic interpretation, various satellite remote-sensing classifications, and vegetation or productivity surrogates, most notably the greenness model, a tasseled cap transformation of remote sensing bands. Here, we examined the effectiveness of existing, similar mapping products for grizzly bear habitat predictions within the Foothills Model Forest (FMF) of west-central Alberta. We used 4,529 global position system (GPS) radiotelemetry locations retrieved from 16 grizzly bears during the berry season (August 1 to denning) of 2001 and 8,848 random available locations to derive population-level habitat models. In total, we compared the effectiveness of 9 mapping products for distinguishing grizzly bear locations (radiotelemetry) from random locations and hence habitat quality.

iv

Six products were derived from a Landsat 7 TM image acquired from September 2001 (classifications, vegetation surrogates, and raw bands), while 1 was based on terrain from a digital elevation model (DEM), 1 a random or neutral map, and finally a hybrid map composed of land cover polygons from Alberta Vegetation Inventory (AVI) maps (photo-interpretation) and the above Landsat classification. We found that incorporating some form of species composition into mapping products created better grizzly bear habitat models from both a model selection and prediction (validation) perspective. Stratifying the closed conifer stands proved to be particularly important. Use of vegetation indices such as greenness or the raw Landsat bands themselves did not appear to be useful and therefore should be used with caution when using these products alone. We suggest that a classified landscape with compositionally based forest types would provide greater inference and ultimately management of grizzly bears in west-central Alberta. FUTURE HABITAT MODELLING FOR GRIZZLY BEARS IN THE FOOTHILLS OF THE ROCKY MOUNTAINS Predicting the future presents a human dream. The advent of computers, long-term ecological studies and many data freely available from the WWW and from elsewhere make such approaches more achievable. Wildlife habitats and landscapes are constantly changing. Despite natural changes, man-made changes occur as well. A landscape has many functions; it can present for instance habitat for wildlife species such as Grizzly Bears but also provides income to human populations. In recent times, the demands on landscapes have increased manifold times. Based on historical landscape change information and using Woodstock/Stanley software, sound future landscape scenarios ranging from 0 to 100 years are derived. The quality of such landscapes are traced with landscape indicators, e.g. Grizzly Bear habitat index and selected Landscape Metrics. MAPPING AND QUANTIFICATION OF CHANGE IN LANDSCAPE STRUCTURE IN GRIZZLY BEAR HABITAT: PROGRESS REPORT The objectives of this research are to: develop a method to reconstruct landscape structure in the past century, determine the rate of change for the landscape, and create map products that represent a time series of landscape change in the study area, for example 1930, 1950, 1970, … today, and finally explore linkages between alternative landscapes, landscape structure, and management needs. The data compiled will also serve as baseline data for future modelling scenarios of landscape change in addition to being inputs in RSF modelling activities. The projected completion date for this research is April 2003.

v

2002 REMOTE SENSING ACTIVITIES: PROGRESS REPORT This report summarizes the remote sensing activities conducted in support of the Foothills Model Forest Grizzly Bear Project during the calendar year 2002. Activities include: (i) radiometric correction of Landsat imagery, (ii) development of greenness maps, (iii) creation of 1999-2001 change detection maps, (iv) creation of 2001 IDT map, and (v) 2002 image acquisition. Progress in each category is briefly described. APPLICATION OF SCAT DETECTION DOGS TO GRIZZLY AND BLACK BEAR MONITORING IN THE YELLOWHEAD ECOSYSTEM, ALBERTA, CANADA Domestic dogs (Canis familiaris) were used to systematically locate grizzly (Ursus arctos horribilus) and black bear (U. americanus) scat in a 1,000 km2 area of the Yellowhead region, Alberta, Canada. DNA extracted from scat is then used to determine the species, sex and individual identities of the animal that left the sample at each global positioning system (GPS)-recorded location. These data are layered onto a Geographic Information System (GIS) that also includes geo-referenced habitat measures to describe the association between animal abundance and distribution with landscape and environmental conditions. Considerable time was spent this year devising and testing methods to improve amplification success of DNA. These efforts proved enormously effective, enabling us to identify 27 unique grizzly bears over a 1000 km2 Results suggest that the scat canine detection methodology is an efficient, and promising means of systematically collecting wildlife scat over large remote areas for use in addressing a variety of critical wildlife management and research questions.

vi

ACKNOWLEDGEMENTS A program of this scope and magnitude would not be possible without the dedication, hard work and support of a large number of people. The program steering committee of the Foothills Model Forest provided valuable support and assistance to allow the research to proceed in order to address management needs and we thank: Mark Storie, Don Podlubny and Bob Udell for this. The financial support of our many program partners allowed us to focus our attention on the delivery of the program goals within this multi-disciplinary program. A special thank you goes to the 2002 capture crew of: Martin Urquhart, Bernie Goski, John Lee, Dave Hobson, Marc Cattet, Nigel Caulkett, David Barrett, Dwight Borden, Steve Borejko, Neil Brad, Wes Bradford, Jeanette Brooks, Tony Brooks, John Day, Jurgen Deagle, John Elias, Curtis English, Mike Ewald, Rod Fyten, Matt Garnett, Andrew Gustavson, Mark Hoskin, Byron Jensen, Randy Kadatz, Rob Kohut, Garth Lemke, Kim McAdam, Adam Mirus, Kelly Moran, Dennis Palkun, Stuart Polege, Todd Ponich, Blane Rawles, Brad Romaniuk, Ken Schmidt, Daren Schwieger, Geoff Skinner, Charles Shewen, Greg Slatter, Scott Smith, Jim Songhurst, Dennis Urban, Andy Van Imschoot, Kelly Wilson, Terry Winkler, and Tyler Young. Without the dedicated hard work and perseverance of these individuals we wouldn’t have met with another very successful capture program. Exemplary flying skills were provided by John Saunders of Peregrine Helicopters of Hinton and fixed wing pilot Mike Dupuis of Wildlife Observation Air Services. A special word of thanks to Julie Duval for her expertise, and enthusiasm in all areas relating to GIS. A thank you to the vegetation plot crew members Erin Bainbridge, Boon, Karen Brown, Marie-Eve Caron, Freek Cluitsman, Ali Iglesis, Sara Jaward, Terry Larsen, Erin Moore, Mike Price, Stephanie Woelk and Jennifer Wasylyk. Their hard work and keen attitudes ensured a successful summer field season. Thanks also to the dog crew members for a very entertaining summer. Radio room staff at Jasper National Park Dispatch assisted our aerial work by providing excellent communications between all our field crews. Lab work on all DNA hair samples was completed by Dr. David Paetkau at Wildlife Genetics International and Dr. Curtis Strobeck at the University of Alberta. Diet analysis was carried out by Richard Riddell of Wildland Ecological Consulting Ltd. Research support in the field, and with a variety of remote sensing needs, was provided by the program team members of the University of Calgary Geography Department under the direction of Steve Franklin.

vii

The staff at the Hinton Training Centre provided a great deal of assistance in many ways this year and also provided food and lodging for the field crews. Communication efforts for this program were directed by Anna Kaufman, Fiona Ragan, and Patsy Vik. A word of praise goes out to this group for keeping up with media needs and the special communication requirements associated with our program. Sheri Fraser, Fran Hanington, and Denise Lebel once again did an excellent job in managing the administrative details of this program.

viii

TABLE OF CONTENTS

EXECUTIVE SUMMARY ...............................................................................................I

ACKNOWLEDGEMENTS ......................................................................................... VII

1.0 2002 GRIZZLY BEAR CAPTURE AND GPS MOVEMENT DATA ................. 1 1.1 INTRODUCTION ......................................................................................................... 1 1.2 METHODS ................................................................................................................. 2

1.2.1 Study Area......................................................................................................... 2 1.2.2 Bear Capture and Handling ............................................................................. 2

1.3 RESULTS ................................................................................................................... 2

2.0 ANIMAL HEALTH: PROGRESS REPORT ......................................................... 7

2.1 INTRODUCTION ......................................................................................................... 7 2.2 STRESS, HEALTH, AND ENVIRONMENT ..................................................................... 8 2.3 COMPARISON OF SELECT HEALTH DATA BETWEEN THE EASTERN SLOPES (ESGBP) AND THE FOOTHILLS MODEL FOREST GRIZZLY BEAR PROJECTS (FMFGBP) ................ 10 2.4 REVERSIBLE ANESTHESIA OF GRIZZLY BEARS USING MEDETOMIDINE-ZOLAZEPAM-TILETAMINE (MZT) AND ATIPAMEZOLE......................................................................... 14 2.5 REFERENCES ........................................................................................................... 16

3.0 A SUMMARY OF THE 2002 FIELD RESEARCH PROGRAMME: FOODS, BERRIES, AND MICRO-SITE HABITAT SELECTION......................................... 17

3.1 SUMMARY............................................................................................................... 17 3.2 RANDOM PLOTS ...................................................................................................... 17 3.3 USE PLOTS .............................................................................................................. 18 3.4 BERRY PLOTS ......................................................................................................... 18

4.0. THE DIET OF GRIZZLY BEARS IN WEST-CENTRAL ALBERTA........ 19

4.1 INTRODUCTION ....................................................................................................... 19 4.2 METHODS ............................................................................................................... 19

4.2.1 Study Area....................................................................................................... 19 4.2.2 Field Sampling ................................................................................................ 20 4.2.3 Lab Analysis.................................................................................................... 20

4.3 RESULTS ................................................................................................................. 20 4.3.1 Differences among years................................................................................. 22

4.4 REFERENCES ........................................................................................................... 23

5.0. GRIZZLY BEAR HABITAT USE OF CLEARCUTS IN WEST-CENTRAL ALBERTA: INFLUENCE OF SITE, SILVICULTURE AND LANDSCAPE STRUCTURE.................................................................................................................. 24


5.2.1 Study Area....................................................................................................... 25 5.2.2 Grizzly bear locations, available locations, seasons, and time of day ........... 28

ix

5.2.3 Explanatory map variables ............................................................................. 29 5.2.4 Modeling building strategies and statistical methods .................................... 31

5.3 RESULTS ................................................................................................................. 33 5.3.1 Clearcuts versus matrix habitat selection....................................................... 33 5.3.2 Habitat selection within clearcuts .................................................................. 34

5.4 DISCUSSION ............................................................................................................ 40 5.5 MANAGEMENT IMPLICATIONS ................................................................................ 40 5.6 REFERENCES ........................................................................................................... 41

6.0 DISTRIBUTION OF GRIZZLY BEAR FOODS IN CLEARCUTS AND REFERENCE FOREST STANDS OF WEST-CENTRAL ALBERTA.................... 45


6.2.1 Study area ....................................................................................................... 46 6.2.2 Field Sampling ................................................................................................ 49 6.2.3 Explanatory variables..................................................................................... 50 6.2.4 Modeling building strategies and statistical methods .................................... 50

6.3 RESULTS ................................................................................................................. 53 6.31 Grizzly bear food occurrence in clearcuts versus upland forests.................... 53 6.3.2 Distribution of grizzly bear foods in clearcuts ............................................... 54 6.3.3 Occurrence of berries and average productivity for clearcuts....................... 56

6.4 DISCUSSION ............................................................................................................ 61 6.5. MANAGEMENT IMPLICATIONS ............................................................................... 64 6.6 REFERENCES ........................................................................................................... 64

7.0 GRIZZLY BEAR FOOTHILLS HABITAT FRAGMENTATION BY SEISMIC CUTLINES: PRELIMINARY REPORT ABOUT ‘PARSIMONY OF LANDSCAPE METRICS’ AND ‘CUTLINE EFFECTS ON HABITAT STRUCTURE AND FOOTHILLS GRIZZLY BEAR LANSCAPE USE’................................................... 68

7.1 INTRODUCTION ....................................................................................................... 68 7.2 PARSIMONY OF LANDSCAPE METRICS IN THE FOOTHILLS ..................................... 69

7.2.1 Methods........................................................................................................... 69 7.2.2 Results and Recommendations: ...................................................................... 71

7.3 EFFECTS OF SEISMIC CUTLINES ON FOOTHILLS GRIZZLY BEAR HABITAT ............. 75 7.3.1 Methods............................................................................................................ 75 7.3.2 Results ............................................................................................................. 76

7.4 EFFECTS OF SEISMIC CUTLINES AND SPATIAL PATTERNS ON FOOTHILLS GRIZZLY BEAR LANDSCAPE USE .................................................................................................. 79 7.5 REFERENCES ........................................................................................................... 82

APPENDIX 7.6. LIST OF THE 49 CLASS-LEVEL (C) AND 54 LANDSCAPE-LEVEL (L) LANDSCAPE STRUCTURE METRICS CALCULATED FOR THE ANALYSIS (SEE MCGARIGAL ET AL. 2002 FOR A COMPLETE DESCRIPTION OF EACH METRIC). PLAND IS THE COVARIABLE IN THE PARTIAL PRINCIPAL COMPONENTS ANALYSES AT THE CLASS LEVEL (AFTER CUSHMANN ET AL IN REVIEW)........................................................................................ 84

x

APPENDIX 7.7. GENERALIZED RECLASSIFICATION SCHEME FOR PARSIMONY ANALYSIS.86

APPENDIX 7.8. LANDCOVER TYPES FROM THE IDTA MAP, THE RECLASSIFIED GBH MAP (AS USED IN PART 2 OF THIS REPORT), AND THE FINAL RECLASSIFICATION OF LANDCOVER TYPES (AS USED IN PART 3 OF THIS REPORT). STARS INDICATE THE UNCORRELATED COVER TYPES USED IN THE LANDSCAPE USE ANALYSIS................................................... 87

8.0 GRAPH THEORY DE VELOPMENT, COST SURFACE ASSESSMENT AND INDIVIDUAL LEVEL CONNECTIVITY ANALYSIS ............................................. 88

8.1 INTRODUCTION ....................................................................................................... 88 8.1.1 Addressing Connectivity ................................................................................. 88 8.1.2 The Graph Theory Framework ....................................................................... 88

8.2 METHODS ............................................................................................................... 89 8.2.1 Cost Surface Comparisons.............................................................................. 90 8.2.2 Graph Generation........................................................................................... 91 8.2.3 Detailed Graph Calculations.......................................................................... 92

8.3 RESULTS ................................................................................................................. 92 8.3.1 Cost Surface (LCP) Validation ........................................................................ 92 8.3.2 Graph Results.................................................................................................. 94

8.4 DISCUSSION ............................................................................................................ 97 8.5 REFERENCES ........................................................................................................... 98

9.0 A COMPARISON OF MAPPING PRODUCTS FOR PREDICTING GRIZZLY BEAR HABITAT QUALITY IN WEST-CENTRAL ALBERTA ........................... 100

9.1 INTRODUCTION AND OBJECTIVES.......................................................................... 100 9.2 METHODS ............................................................................................................. 101 9.3 RESULTS ............................................................................................................... 109 9.4 SUGGESTIONS, LIMITATIONS, AND CONCLUSIONS ............................................... 115 9.5 REFERENCES ......................................................................................................... 124

10.0 FUTURE HABITAT MODELLING FOR GRIZZLY BEARS IN THE FOOTHILLS OF THE ROCKY MOUNTAINS ....................................................... 127

10.1 INTRODUCTION ................................................................................................... 127 10.2 METHODS ........................................................................................................... 127

10.2.1 General approach and infrastructure......................................................... 127 10.2.2 Data sets...................................................................................................... 128 10.2.3 The natural succession and forestry model ................................................ 128 10.2.4 Fire modelling............................................................................................. 128 10.2.5 Storm event modelling................................................................................. 128 10.2.6 Insect modelling .......................................................................................... 129 10.2.7 Road modelling ........................................................................................... 129 10.2.8 Human settlement modelling....................................................................... 129 10.2.9 Scenarios tested .......................................................................................... 129 10.2.10 Landscape Indicators for Grizzly Bear habitats....................................... 129 10.2.11 Oil and Gas Models ................................................................................. 130 10.2.12 Model evaluation ...................................................................................... 130

10.3 RESULTS ............................................................................................................. 130

xi

10.4 DISCUSSION ........................................................................................................ 131 10.5 REFERENCES ....................................................................................................... 132

11.0 MAPPING AND QUANTIFICATION OF CHANGE IN LANDSCAPE STRUCTURE IN GRIZZLY BEAR HABITAT: PROGRESS REPORT.............. 134

11.1 INTRODUCTION ................................................................................................... 134 11.2 PURPOSE AND BACKGROUND.............................................................................. 134 11.3 DATA AND METHODS.......................................................................................... 135

11.3.1 Spatial Data ................................................................................................ 135 11.3.2 Geographic Information Systems................................................................ 136 11.3.3 Ground Data ............................................................................................... 137

11.4 SUMMARY........................................................................................................... 138 11.5 REFERENCES ....................................................................................................... 139

12.0 2002 REMOTE SENSING ACTIVITIES: PROGRESS REPORT ................ 140 12.1 INTRODUCTION ................................................................................................... 140 12.2 RADIOMETRIC CORRECTION ............................................................................... 140 12.3 GREENNESS MAPS .............................................................................................. 142 12.4 1999-2001 CHANGE DETECTION AND 2001 IDT MAP........................................ 143 12.5 2002 IMAGERY ................................................................................................... 145 12.6 REFERENCES ....................................................................................................... 146

13.0 APPLICATION OF SCAT DETECTION DOGS TO GRIZZLY AND BLACK BEAR MONITORING IN THE YELLOWHEAD ECOSYSTEM, ALBERTA, CANADA ....................................................................................................................... 147

13.1 INTRODUCTION ................................................................................................... 147 13.2 METHODS ........................................................................................................... 147

13.2.1 Back-Tracking Radio-collared Grizzly Bears............................................. 148 13.2.2 Scat collection and processing.................................................................... 148 13.2.3 DNA Extraction and Amplification............................................................. 148

13.3 RESULTS ............................................................................................................. 151 13.3.1 Scat sample age and amplification success ................................................ 151 13.3.2 Scat Sampling Results ................................................................................. 151 13.3.3 Bear distributions based on hair collection and radio-telemetry methods: 152 13.3.4 Between-Dog Team Differences ................................................................. 153

13.4 DISCUSSION ........................................................................................................ 153 13.5 LITERATURE CITED............................................................................................. 155

APPENDIX 1: PUBLICATION/TECHNICAL PAPER LIST................................ 157

APPENDIX 2: FOOTHILLS MODEL FOREST GRIZZLY BEAR RESEARCH PARTNERS................................................................................................................... 162

xii

LIST OF TABLES TABLE 1.1. TOTAL NUMBER OF HANDLED AND COLLARED GRIZZLY BEARS, 1999 - 2002... 3 TABLE 1.2. TOTAL NUMBER OF GPS LOCATION DATA FROM COLLARED BEARS, 1999 -

2002............................................................................................................................. 4 TABLE 1.3. REPRODUCTIVE STATUS OF FEMALES IN WEST-CENTRAL ALBERTA, 1999 -

2002............................................................................................................................. 4 TABLE 1.4. SUMMARY OF GRIZZLY BEAR MORTALITIES IN THE WEST-CENTRAL ALBERTA,

1999 – 2002................................................................................................................. 5 TABLE 1.5. SUMMARY OF MCP HOME RANGES FOR GRIZZLY BEARS IN WEST-CENTRAL

ALBERTA, 1999-2002. ................................................................................................. 6 TABLE 2.1. COMPARISON OF BODY CONDITION INDEX (BCI) VALUES BETWEEN THE

EASTERN SLOPES AND FOOTHILLS MODEL FOREST GRIZZLY BEAR PROJECTS FOR GRIZZLY BEARS CAPTURED DURING EITHER MAY OR JUNE. ....................................... 11

TABLE 2.2. COMPARISON OF REPRODUCTIVE HORMONE CONCENTRATIONS BETWEEN THE EASTERN SLOPES AND FOOTHILLS MODEL FOREST GRIZZLY BEAR PROJECTS FOR FEMALE GRIZZLY BEARS CAPTURED BY LEG-HOLD SNARE DURING EITHER MAY OR JUNE. ......................................................................................................................... 12

TABLE 2.3. COMPARISON OF REPRODUCTIVE HORMONE CONCENTRATIONS BETWEEN THE EASTERN SLOPES AND FOOTHILLS MODEL FOREST GRIZZLY BEAR PROJECTS FOR MALE GRIZZLY BEARS CAPTURED BY LEG-HOLD SNARE DURING EITHER MAY OR JUNE.................................................................................................................................... 12

TABLE 2.4 ANESTHESIA INDUCTION FEATURES OF MEDETOMIDINE (M) AND TELAZOL (ZT) AND REVERSAL FEATURES OF ATIPAMEZOLE IN FIVE GRIZZLY BEARS CAPTURED DURING 2002. ............................................................................................................ 15

TABLE 5.1. AREA (KM2) AND PERCENT COMPOSITION OF LANDCOVER CLASSES WITHIN THE 2677-KM2 SECONDARY STUDY AREA NEAR HINTON, ALBERTA. LANDCOVER CLASSES WERE DETERMINED FROM A REMOTE SENSING CLASSIFICATION (FRANKLIN ET AL., 2001) AND FORESTRY GIS DATA ON CUT-BLOCKS. .................................................... 28

TABLE 5.2. EXPLANATORY MAP VARIABLES USED FOR ASSESSING GRIZZLY BEAR HABITAT SELECTION OF CLEAR-CUTS IN THE UPPER FOOTHILLS OF WEST-CENTRAL ALBERTA, CANADA. ................................................................................................................... 30

TABLE 5.3. SILVICULTURE AND SITE PREPARATION TREATMENTS ASSESSED FOR GRIZZLY BEAR HABITAT SELECTION. ........................................................................................ 31

TABLE 5.4. A PRIORI SEASONAL CANDIDATE MODELS FOR GRIZZLY BEARS DESCRIBING HABITAT SELECTION FOR CLEAR-CUTS IN THE UPPER FOOTHILLS OF WEST-CENTRAL ALBERTA, CANADA. MODEL NUMBER, NAME, AND STRUCTURE ARE PROVIDED....... 33

TABLE 5.5. SEASONAL ESTIMATES OF HABITAT SELECTION FOR CLEAR-CUTS (1) BY GRIZZLY BEARS COMPARED TO MATRIX HABITATS (0; REFERENCE CATEGORY) IN THE UPPER FOOTHILLS OF WEST-CENTRAL ALBERTA, CANADA. ....................................... 34

TABLE 5.6. AIC SELECTED MODELS FOR HYPOPHAGIA, EARLY HYPERPHAGIA, AND LATE HYPERPHAGIA PERIODS. NUMBER OF PARAMETERS (KI), MODEL -2 LOGLIKELIHOOD, AIC, CHANGE IN AIC (∆I) FROM LOWEST MODEL, AND AKAIKE WEIGHTS (WI) OF MODEL SUPPORT ARE REPORTED. ............................................................................... 36

TABLE 6.1. LIST OF GRIZZLY BEAR FOOD ITEMS EXAMINED IN CLEARCUTS AND UPLAND FOREST STANDS OF WEST-CENTRAL ALBERTA. .......................................................... 51

xiii

TABLE 6.2. ENVIRONMENTAL VARIABLES USED TO PREDICT THE OCCURRENCE OF GRIZZLY BEAR FOODS WITHIN WEST-CENTRAL ALBERTA CLEARCUTS. VARIABLE CODE USED FOR CANDIDATE MODELS, VARIABLE DESCRIPTION, UNITS (WITH RANGE FOR NON QUADRATIC PARAMETERS), AND DATA SOURCES ARE PRESENTED. ............................ 52

TABLE 6.3. A PRIORI CANDIDATE MODELS USED FOR ASSESSING DISTRIBUTION OF GRIZZLY BEAR FOODS WITHIN CLEARCUTS OF WEST-CENTRAL ALBERTA. MODEL NUMBER, PARAMETER STRUCTURE (VARIABLES), AND TOTAL NUMBER OF PARAMETERS (INCLUDING CONSTANT) USED FOR CALCULATING AKAIKE WEIGHTS (WI) FOR MODEL SELECTION. ................................................................................................................ 53

TABLE 6.4. FREQUENCY OF OCCURRENCE FOR 13 GRIZZLY BEAR FOODS WITHIN CLEARCUT (N = 355) AND REFERENCE FOREST (N = 183) PLOTS. ODDS RATIO (± S.E.) OF FINDING GRIZZLY BEAR FOODS (H-HERBACEOUS LAYER; S-SHRUB LAYER) WITHIN CLEARCUTS OF WEST-CENTRAL ALBERTA WHEN COMPARED TO REFERENCE UPLAND FOREST STANDS ARE REPORTED FROM LOGISTIC REGRESSION MODELS. MODEL LIKELIHOOD RATIO (LR) χ2 TEST AND ASSOCIATED SIGNIFICANCE (P) LEVELS ARE REPORTED...... 54

TABLE 6.5. AKAIKE (SMALL SAMPLE SIZE) MODEL SELECTION WEIGHTS (WI) FOR GRIZZLY BEAR FOODS IN CLEARCUTS OF WEST-CENTRAL ALBERTA. SHEPHERDIA CANADENSIS IS REPORTED FOR BOTH THE SHRUB-LAYER (S) AND HERBACEOUS-LAYER (H). MODEL WITH THE GREATEST SUPPORT PER INDIVIDUAL SPECIES IS INDICATED IN BOLD FONT.55

TABLE 6.6. AICC SELECTED MODELS WITH CORRESPONDING METRICS OF MODEL SIGNIFICANCE, FIT, AND CLASSIFICATION ACCURACY. ALL MODEL LIKELIHOOD RATIO (LR) χ2 TESTS WERE SIGNIFICANT AT P<0.05. PERCENT DEVIANCE (DEV.) EXPLAINED REPRESENTS THE REDUCTION IN THE LOG LIKELIHOOD FROM THE NULL MODEL. PROBABILITIES FOR HOSMER AND LEMESHOW (1980) GOODNESS-OF-FIT χ2 STATISTIC (P Ĉ) ARE REPORTED FOR MODEL AND DATA FIT, WHILE RECEIVER OPERATING CHARACTERISTIC (ROC) CURVES ARE USED TO ASSESS MODEL CLASSIFICATION ACCURACY. BOTH IN-SAMPLE (TRAINING DATA) AND OUT-OF-SAMPLE (TESTING DATA) DATA WERE USED FOR FIT AND CLASSIFICATION ACCURACY, WHILE INDEPENDENT DATA FROM A CONCURRENT STUDY WERE USED TO ASSESS THE PERCENT CORRECTLY CLASSIFIED (PCC). ................................................................................. 57

TABLE 6.7. ESTIMATED COEFFICIENTS (βI) FOR AIC SELECTED MODELS DESCRIBING THE PROBABILITY OF OCCURRENCE FOR INDIVIDUAL GRIZZLY BEAR FOODS WITHIN CLEARCUTS OF WEST-CENTRAL ALBERTA.................................................................. 58

TABLE 6.8. PREDICTED LOCATIONS OF OPTIMAL GRIZZLY BEAR FOOD OCCURRENCE FOR CLEARCUTS OF WEST-CENTRAL ALBERTA. TO ESTIMATE OPTIMAL LOCATIONS FOR EACH VARIABLE, ALL OTHER FACTORS (VARIABLES) WERE HELD AT THEIR MEAN LEVEL. BOLD FONT INDICATES SIGNIFICANT TREND.................................................. 59

TABLE 6.9. PERCENT FREQUENCY OF FRUIT (AFTER JULY 14), GIVEN THE PRESENCE OF THE SPECIES, FOR CLEARCUTS IN WEST-CENTRAL ALBERTA. ESTIMATED COEFFICIENTS (βI) FOR AIC SELECTED MODELS DESCRIBING THE PROBABILITY OF FRUIT PRESENCE (GIVEN ITS PRESENCE) AND FINALLY THE ESTIMATED OPTIMAL POSITION FOR FRUIT OCCURRENCE. ............................................................................................................ 60

TABLE 6.10. AVERAGE FRUIT PRODUCTION (NUMBER OF BERRIES) PER HECTARE (HA.) WITHIN CLEARCUTS OF WEST-CENTRAL ALBERTA FOR CITES WHERE THE SPECIES WERE PRESENT OR FOR ALL SITES. TOTAL FRUIT PRODUCTION AND ESTIMATED FRESH WEIGHT (KILOGRAMS) PROVIDED FOR ESTIMATED PRODUCTION ON AN AVERAGE

xiv

HECTARE OF CLEARCUT (ALL SITES REGARDLESS OF WHETHER THE BEAR FOOD ITEM WAS PRESENT)............................................................................................................ 62

TABLE 7.1. PRINCIPAL COMPONENT CLUSTERS (METRIC GRADIENTS), IDENTIFIED THROUGH PARTIAL PRINCIPAL COMPONENT AND CLUSTERING ANALYSIS ON 49 CLASS-LEVEL CONFIGURATION METRICS ACROSS 8 GENERALIZED IDT LANDCOVER TYPES, WITH THEIR RESPECTIVE UNIVERSALITY, CONSISTENCY AND STRENGTH. MEANING OF EACH CLUSTER IS PROVIDED BY THE METRIC GRADIENT NAME, LISTING ALSO THE LARGEST POSITIVE AND NEGATIVE LOADINGS. CONSISTENCY WAS ASSESSED AGAINST THE AVERAGE OF ALL WITHIN CLUSTER CORRELATIONS ACROSS ALL CLASSES BEING 0.74. UNIVERSAL AND CONSISTENT METRIC GRADIENTS ARE HIGHLIGHTED FORMING THE MINIMUM SET OF CLASS-LEVEL METRIC GRADIENTS TO BE USED WITHIN THIS STUDY AREA. UNIQUE METRIC GRADIENTS ARE ITALICIZED, AND LOW CONSISTENCY METRIC GRADIENTS APPEAR IN NORMAL FONT. .......................................................... 73

TABLE 7.2. SIX SIGNIFICANT PRINCIPAL COMPONENTS, OR METRIC GRADIENTS OF 54 LANDSCAPE-LEVEL CONFIGURATION METRICS FROM THE GENERALIZED IDT LAND COVER, AS RETAINED FROM PARTIAL PRINCIPAL COMPONENT ANALYSIS. COMPONENTS ARE ORDERED BY THEIR EXPLANATORY POWER (PERCENT EXPLAINED VARIANCE). MEANING OF EACH COMPONENT IS PROVIDED BY THE METRIC GRADIENT NAME, LISTING ALSO THE LARGEST POSITIVE AND NEGATIVE LOADINGS.................... 75

TABLE 7.3. EXPLANATORY MODEL PARAMETERS, SHOWING RESIDUAL DEVIANCE, DEGREES OF FREEDOM, SIGNIFICANCE AND COEFFICIENTS OF MODEL VARIABLES, FOR EARLY SUMMER GRIZZLY BEAR LANDSCAPE USE AT THE POPULATION LEVEL........................ 81

TABLE 8.1: LEAST-COST PATH VALIDATION RESULTS SHOWING MEAN DISTANCES (M) FROM GPS DATA TO MODELED PATHS BROKEN DOWN BY COST SURFACE MODEL, SHOWING SINGLE FACTOR ANALYSIS OF VARIANCE (ANOVA) BASED ON LOG TRANSFORMATION OF MEANS..................................................................................... 94

TABLE 8.2. IDENTIFICATION, GPS SAMPLE SIZES BY YEAR, 95% KERNEL HOME RANGE SIZES, AND MEAN DAILY MOVEMENT RATES USED IN GRAPH THEORY ANALYSIS........ 95

TABLE 8.3. GRAPH ANALYSIS RESULTS SHOWING NUMBER OF NODES (RSF HABITAT PATCHES) USED, NUMBER OF EDGES CREATED USING DAILY MOVEMENT RATE BY FEMALE BY YEAR, AND CONNECTIVITY MEASURES. ................................................... 95

TABLE 9.1. IDENTIFICATION (NAME), SEX (M-MALE; F-FEMALE), GPS RADIOTELEMETRY SAMPLE SIZE (USE), AND RANDOM AVAILABLE LOCATIONS SAMPLED WITHIN 100% MINIMUM CONVEX POLYGON (MCP) HOME RANGES FOR INDIVIDUAL GRIZZLY BEARS (URSUS ARCTOS L.) IN WEST-CENTRAL ALBERTA AND USED FOR MODELING HABITAT SELECTION IN THIS REPORT. DUE TO LIMITED MAPPING PRODUCTS EXTENT, ADDITIONAL GRIZZLY BEAR RADIOTELEMETRY LOCATIONS WERE WITHHELD.......... 105

TABLE 9.2. INTEGRATED DECISION TREE (IDT) APPROACH LAND USE/COVER CLASSIFICATION FOR THE FOOTHILLS MODEL FOREST (FMF) GRIZZLY BEAR PROJECT. RECLASSIFICATIONS OF THE ORIGINAL 23 IDT MAP CLASSES WERE PROVIDED IN A HIERARCHAL MANNER TO SIMPLIFY MAPS INTO ECOLOGICAL CATEGORIES. NOTE THAT THE WELL SITE AND URBAN AREAS WERE MASKED FROM ANALYSES........................ 106

TABLE 9.3. CORRELATION (PEARSON) MATRIX OF TASSELED CAP TRANSFORMATIONS OF LANDSAT TM BANDS FOR RANDOM LOCATIONS (N = 8,848) WITHIN GRIZZLY BEAR MINIMUM CONVEX POLYGON (MCP) HOME RANGES................................................ 106

xv

TABLE 9.4. CORRELATION (PEARSON) MATRIX OF LANDSAT TM BANDS 1 TO 6 FOR RANDOM LOCATIONS (N = 8,848) WITHIN GRIZZLY BEAR MINIMUM CONVEX POLYGON (MCP) HOME RANGES.............................................................................................. 110

TABLE 9.5. ASSESSMENT OF MODEL FIT AND MODEL SELECTION BETWEEN CANDIDATE MAPPING PRODUCTS USED FOR PREDICTING GRIZZLY BEAR HABITAT WITHIN WEST-CENTRAL ALBERTA. COMPLEXITY OF MODEL (PARAMETERS) IS PROVIDED BY K. SCHWARTZ BAYESIAN INFORMATION CRITERIA (SBIC) IS REPORTED AND USED TO SELECT THE MOST PARSIMONIOUS MODEL. CHANGE BETWEEN INDIVIDUAL CANDIDATE MODELS AND LOWEST SBIC MODEL IS REPRESENTED BY ∆ SBIC, WHILE THE WEIGHT OF THE MODEL WITHIN THE SET OF MODELS AND DATA TESTED IS INDICATED BY WI (RANK OF WI IS FURTHER PROVIDED). ALTHOUGH WE HAVE PRESENCE-AVAILABILITY DATA, WE REPORT % DEVIANCE EXPLAINED (% DEV.), MCKELVEY AND ZAVOINA'S R2, AND RECEIVER OPERATING CHARACTERISTIC (ROC) FOR ALL MODELS (SEE BOYCE ET AL. 2002 FOR LIMITATIONS OF THESE METRICS). . 111

TABLE 9.6. RESULTS OF THE SOMER’S D TESTS BETWEEN HABITAT BIN RANK (LOW TO HIGH) AND THE PROPORTION OF TESTING LOCATIONS (N = 679) FALLING WITHIN THOSE BINS AND ORDERED BY SBIC RANK. GOOD MODELS WOULD HAVE A SIGNIFICANT LARGE POSITIVE RELATIONSHIP. JACKKNIFE STANDARD ERRORS, THE PROBABILITY THAT D IS DIFFERENT FROM 0, AND 95% CONFIDENCE INTERVALS ARE PROVIDED, ALONG WITH THE RANK IN THE STRENGTH OF RELATIONSHIP (NOTE SIMILAR RANKS/D VALUES (POOR DISCRIMINATION) FOR SOME MODELS). ............................................ 113

TABLE 9.7. REGRESSION BETWEEN EXPECTED AND OBSERVED PROPORTION OF VALIDATION OBSERVATIONS WITHIN HABITAT BINS ORDERED BY SBIC RANKS. SLOPE, STANDARD ERROR (S.E.), AND R2 ARE PROVIDED, ALONG WITH F TESTS (1, 8 DF) FOR THE SIGNIFICANCE BETWEEN SLOPE ESTIMATE AND NULL HYPOTHESIS OF EQUALING ZERO (HO = 0.0) AND ONE (HO = 1.0). ............................................................................... 113

TABLE 9.8. ESTIMATED COEFFICIENTS (ΒI) OF HABITAT SELECTION FOR LAND USE/COVER TYPES FOR GRIZZLY BEARS IN WEST-CENTRAL ALBERTA USING THE HYBRID AVI/IDT MAPPING PRODUCT. CLASSES NOT OVERLAPPING ZERO (SELECTION OR AVOIDANCE) ARE INDICATED IN BOLD FONT. ROBUST STANDARD ERRORS (S.E.) ARE REPORTED AS ESTIMATED ACROSS INDIVIDUALS, NOT AUTOCORRELATED LOCATIONS. ................. 117

TABLE 9.9. ESTIMATED COEFFICIENTS (ΒI) OF HABITAT SELECTION FOR LAND USE/COVER TYPES FOR GRIZZLY BEARS IN WEST-CENTRAL ALBERTA USING THE ENHANCED INTEGRATED DECISION TREE (IDT) REMOTE SENSING MAPPING PRODUCT. CLASSES NOT OVERLAPPING ZERO (SELECTION OR AVOIDANCE) ARE INDICATED IN BOLD FONT. ROBUST STANDARD ERRORS (S.E.) ARE REPORTED AS ESTIMATED ACROSS INDIVIDUALS, NOT AUTOCORRELATED LOCATIONS. ................................................. 118

TABLE 9.10. ESTIMATED COEFFICIENTS (ΒI) OF HABITAT SELECTION FOR GRIZZLY BEARS IN WEST-CENTRAL ALBERTA USING THE TERRAIN MAPPING PRODUCT (ELEVATION AND COMPOUND TOPOGRAPHIC INDEX, CTI). ELEVATION IS IN METRES. VARIABLES NOT OVERLAPPING ZERO (SELECTION OR AVOIDANCE) ARE INDICATED IN BOLD FONT. ROBUST STANDARD ERRORS (S.E.) ARE REPORTED AS ESTIMATED ACROSS INDIVIDUALS, NOT AUTOCORRELATED LOCATIONS. ................................................. 118

TABLE 9.11. ESTIMATED COEFFICIENTS (ΒI) OF HABITAT SELECTION FOR LAND USE/COVER TYPES FOR GRIZZLY BEARS IN WEST-CENTRAL ALBERTA USING THE IDT-2001 PRODUCT. VARIABLES NOT OVERLAPPING ZERO (SELECTION OR AVOIDANCE) ARE

xvi

INDICATED IN BOLD FONT. ROBUST STANDARD ERRORS (S.E.) ARE REPORTED AS ESTIMATED ACROSS INDIVIDUALS, NOT AUTOCORRELATED LOCATIONS. ................. 119

TABLE 9.12. ESTIMATED COEFFICIENTS (ΒI) OF HABITAT SELECTION FOR LAND USE/COVER TYPES FOR GRIZZLY BEARS IN WEST-CENTRAL ALBERTA USING THE IDT-2001 SIMPLIFIED A PRODUCT. VARIABLES NOT OVERLAPPING ZERO (SELECTION OR AVOIDANCE) ARE INDICATED IN BOLD FONT. ROBUST STANDARD ERRORS (S.E.) ARE REPORTED AS ESTIMATED ACROSS INDIVIDUALS, NOT AUTOCORRELATED LOCATIONS.................................................................................................................................. 120

TABLE 9.13. ESTIMATED COEFFICIENTS (ΒI) OF HABITAT SELECTION FOR LAND USE/COVER TYPES FOR GRIZZLY BEARS IN WEST-CENTRAL ALBERTA USING THE IDT-2001 SIMPLIFIED B PRODUCT. VARIABLES NOT OVERLAPPING ZERO (SELECTION OR AVOIDANCE) ARE INDICATED IN BOLD FONT. ROBUST STANDARD ERRORS (S.E.) ARE REPORTED AS ESTIMATED ACROSS INDIVIDUALS, NOT AUTOCORRELATED LOCATIONS.................................................................................................................................. 120

TABLE 9.14. ESTIMATED COEFFICIENTS (ΒI) OF HABITAT SELECTION FOR GRIZZLY BEARS IN WEST-CENTRAL ALBERTA USING THE RAW TM LANDSAT BANDS (2 PC AXES) PRODUCT. VARIABLES NOT OVERLAPPING ZERO (SELECTION OR AVOIDANCE) ARE INDICATED IN BOLD FONT. ROBUST STANDARD ERRORS (S.E.) ARE REPORTED AS ESTIMATED ACROSS INDIVIDUALS, NOT AUTOCORRELATED LOCATIONS. ................. 121

TABLE 9.15. ESTIMATED COEFFICIENTS (ΒI) OF HABITAT SELECTION FOR GRIZZLY BEARS IN WEST-CENTRAL ALBERTA USING THE TASSELED CAP TRANSFORMATION (2 PC AXES) PRODUCT. VARIABLES NOT OVERLAPPING ZERO (SELECTION OR AVOIDANCE) ARE INDICATED IN BOLD FONT. ROBUST STANDARD ERRORS (S.E.) ARE REPORTED AS ESTIMATED ACROSS INDIVIDUALS, NOT AUTOCORRELATED LOCATIONS. ................. 121

TABLE 9.16. ESTIMATED COEFFICIENTS (ΒI) OF HABITAT SELECTION FOR LAND USE/COVER TYPES FOR GRIZZLY BEARS IN WEST-CENTRAL ALBERTA USING THE RANDOM MAP PRODUCT. VARIABLES NOT OVERLAPPING ZERO (SELECTION OR AVOIDANCE) ARE INDICATED IN BOLD FONT. ROBUST STANDARD ERRORS (S.E.) ARE REPORTED AS ESTIMATED ACROSS INDIVIDUALS, NOT AUTOCORRELATED LOCATIONS. ................. 122

TABLE 11.1. REMOTELY SENSED DATA USED IN ANALYSIS ............................................. 136 TABLE 12.1. FMF GRIZZLY BEAR PROJECT IMAGE ARCHIVE. ....................................... 143

xvii

LIST OF FIGURES FIGURE 1.1. GRIZZLY BEAR RESEARCH PROGRAM STUDY AREA WITH NATURAL

SUBREGIONS DELINEATED. ........................................................................................... 3 FIGURE 4.1. ANNUAL TRENDS IN THE VOLUME OF RESIDUE OF VARIOUS FOOD ITEMS FOUND

IN GRIZZLY BEAR SCATS COLLECTED IN WEST-CENTRAL ALBERTA BETWEEN 2001 AND 2002........................................................................................................................... 22

FIGURE 5.1. STUDY AREA MAP DEPICTING ELEVATION, LOCAL TOWNS, OVERALL FOOTHILLS MODEL FOREST STUDY REGION (MAP EXTENT), AND SECONDARY FORESTRY STUDY AREA FOR EXAMINING HABITAT..................................................... 26

FIGURE 5.2. MAPPED CLEAR-CUTS BY 5-YEAR AGE CLASS WITHIN THE SECONDARY FORESTRY STUDY AREA IN THE UPPER FOOTHILLS OF WEST-CENTRAL ALBERTA, CANADA USED FOR MODELING GRIZZLY BEAR HABITAT SELECTION. ......................... 27

FIGURE 5.3. ESTIMATED SILVICULTURAL COEFFICIENTS FOR HYPOPHAGIA AND LATE HYPERPHAGIA ............................................................................................................ 38

FIGURE 5.4. SPATIAL PREDICTIONS OF CUT-BLOCK HABITAT QUALITY FOR AIC SELECTED CANDIDATE MODELS BY SEASON (A. HYPOPHAGIA; B. EARLY HYPERPHAGIA; C. LATE HYPERPHAGIA)........................................................................................................... 38

FIGURE 6.1. STUDY AREA MAP DEPICTING ELEVATION, LOCAL TOWNS, OVERALL FOOTHILLS MODEL FOREST STUDY REGION (MAP EXTENT), AND SECONDARY FORESTRY STUDY AREA FOR EXAMINING HABITAT SELECTION RELATED TO CLEARCUT HARVESTING IN WEST-CENTRAL ALBERTA, CANADA. LOCATION OF PRINCIPLE STUDY AREA WITHIN ALBERTA IS DEPICTED IN THE UPPER LEFT PORTION OF THE FIGURE. .... 47

FIGURE 6.2. MAPPED CLEARCUTS BY 5-YEAR AGE CLASS WITHIN THE SECONDARY FORESTRY STUDY AREA IN THE UPPER FOOTHILLS OF WEST-CENTRAL ALBERTA, CANADA USED FOR MODELING GRIZZLY BEAR HABITAT SELECTION. ......................... 48

FIGURE 6.3. PREDICTED PROBABILITY OF OCCURRENCE FOR SHEPHERDIA CANADENSIS (SOLID LINE) AND FOR FRUITS (DASHED LINE) GIVEN THAT THE PLANT WAS PRESENT IN CLEARCUTS OF WEST-CENTRAL ALBERTA.................................................................. 62

FIGURE 7.1. STRATIFICATION OF THE GENERALIZED IDT FOOTHILLS OF THE EXTENDED GRIZZLY BEAR RESEARCH PROJECT BOUNDARY FOR PURPOSE OF THE PARSIMONY ANALYSIS. .................................................................................................................. 70

FIGURE 7.2. STRATIFICATION OF FOOTHILLS PORTION OF ORIGINAL GRIZZLY BEAR RESEARCH PROJECT BOUNDARY INTO 2300HA LARGE HEXAGONS TO ASSESS SEISMIC CUTLINE EFFECTS ON LANDSCAPE METRICS................................................................ 77

FIGURE 7.3. EFFECTS OF ADDING SEISMIC CUTLINES AT DIFFERENT DENSITIES ON THREE LANDSCAPE-LEVEL METRICS ACROSS 104 2300 HA LARGE HEXAGONS OF THE FOOTHILLS GBH MAP. ............................................................................................... 77

FIGURE 7.4. RATE OF CHANGE IN EDGE DENSITY IN RESPONSE TO ADDING SEISMIC CUTLINES AS A FUNCTION OF ORIGINAL LANDSCAPE FRAGMENTATION ...................... 78

FIGURE 7.5. RATE OF CHANGE IN NUMBER OF PATCHES IN RESPONSE TO ADDING SEISMIC CUTLINES AS A FUNCTION OF ORIGINAL LANDSCAPE FRAGMENTATION. ..................... 78

FIGURE 7.6. FLOWCHART OF METHODS USED DURING ANALYSIS OF SEISMIC CUTLINE AND LANDSCAPE STRUCTURE EFFECTS ON GRIZZLY BEAR POPULATION LEVEL LANDSCAPE USE............................................................................................................................. 80

xviii

FIGURE 7.7. EFFECTS OF ADDING SEISMIC CUTLINES ON VARIATION IN INTER-PATCH DISTANCE METRIC. ..................................................................................................... 81

FIGURE 8.1. (A) CLASSIC GRAPH STRUCTURE WITH EDGE EAC=NANC CONNECTS NODES NA AND NC; (B) PATH L (REPRESENTED BY SOLID BLACK LINE) WITHIN GRAPH G IS A SEQUENCE OF NODES CONNECTING NA TO NE. .............................................................. 89

FIGURE 8.2. EXAMPLE OF COST SURFACES USED IN COMPARISON AND LEAST-COST PATH VALIDATION............................................................................................................... 90

FIGURE 8.3. (A) NODES (CENTROIDS) DELINEATED FROM RSF HABITAT PATCHES DEFINED BY 95% KERNEL HOME RANGE; (B) LEAST-COST PATH EDGES CREATED BASED ON COST SURFACE MODELING TO REPRESENT FUNCTIONAL CONNECTIONS BETWEEN PATCHES.91

FIGURE 8.4. LEAST-COST PATH OUTPUT OF FOUR COST SURFACE MODELS DEMONSTRATING VALIDATION TECHNIQUE, DISTANCES OF WITHHELD GPS DATA COMPARED TO PATHS MODELED WERE EXTRACTED USING GIS TECHNIQUES. .............................................. 93

FIGURE 8.5. GRAPHS GENERATED FOR FEMALES G004 (A) AND G020 (B) SHOWING VARIATION IN 2000 CONNECTIVITY LEVELS BETWEEN MOUNTAIN AND FOOTHILLS BEARS RESPECTIVELY, CONNECTION DISTANCES ARE DEFINED BY DAILY MOVEMENT RATES. ....................................................................................................................... 96

FIGURE 9.1. STUDY AREA MAP FOR THE 2001 BOUNDARY (~10,000-KM2) OF THE FOOTHILLS MODEL FOREST (FMF) GRIZZLY BEAR PROJECT AND ASSOCIATED GRIZZLY BEAR (URSUS ARCTOS L.) LOCATION DATA RETRIEVED FROM GPS RADIOTELEMETRY COLLARS DURING THE BERRY SEASON (AUGUST 1 TO DENNING) OF 2001. THE RELIEF MAP SHOWS THE AREAS OF MOUNTAINS AND ASSOCIATED VALLEY-BOWL ORIENTATION OF THE REGION WHERE BEAR LOCATIONS ARE MOST TYPICALLY FOUND. TO THE NORTHEAST, THE STUDY AREA GRADES INTO UPPER AND LOWER FOOTHILL ECOSUBREGIONS WHERE HABITATS PROGRESSIVELY BECOME MORE BOREAL IN NATURE.................................................................................................................... 107

FIGURE 9.2. RELATIONSHIP BETWEEN HABITAT QUALITY RANK (BIN NUMBER) AND PROPORTION OF WITHHELD TESTING LOCATIONS FALLING WITHIN THAT BIN. .......... 112

FIGURE 9.3. COMPARISON OF EXPECTED AND OBSERVED PROPORTION OF VALIDATION LOCATIONS. NOTE THE DEFINED ‘GOLD STANDARD’ REPRESENTING A SLOPE OF 1.0 AND HENCE A SYSTEMATICALLY INCREASING QUALITY OF HABITAT (RELATIVE PROBABILITY OF OCCURRENCE). .............................................................................. 116

FIGURE 10.1. FUTURE LANDSCAPE SCENARIOS (DRAFT) : AGE CLASES 2000 AND 2100 (NATURAL TREE DEATH). DARK GREEN 0 YEARS, YELLOW 170 YEARS, RED >320 YEARS ...................................................................................................................... 131

FIGURE 10.2. FUTURE LANDSCAPE SCENARIOS (DRAFT) : PREDICTED IDT CLASSES 2000 , 2050 AND 2100. PINK VALUES INDICATE FOREST COVER (BASED ON NATURAL DEATH AND RE-GROW). ............................................................................................ 131

FIGURE 11.2. CHANGE THROUGH TIME .......................................................................... 138

xix

1.0 2002 GRIZZLY BEAR CAPTURE AND GPS MOVEMENT DATA

1.1 Introduction

The challenge facing land managers is to learn how to ensure the long-term survival of grizzly bears while addressing human and societal demands on the same land base. If we are to sustain both human use activities and grizzly bears, intensive management based on sound biological information is required. In 1999, the Foothills Model Forest (FMF) initiated a co-operative, international, multidisciplinary, six-year grizzly bear research program in the Yellowhead Ecosystem of west-central Alberta to address these concerns. The primary goal of the research is to assess grizzly bear populations, bear response to human activities, and habitat conditions in order to better understand the requirements of this species and integrate those requirements into the land management decision-making framework. This program is directly linked to the 2000 management framework document entitled “Grizzly Bear Conservation in the Alberta Yellowhead Ecosystem – A Strategic Framework”. The research addresses important management questions for which data are required. An important outcome of this program will be the development of tools and techniques that address landscape level conservation issues, which is a critical component to the successful management of grizzly bear populations throughout Alberta and North America.

The Foothills Model Forest Grizzly Bear Research Program was designed to take a holistic approach to questions concerning grizzly bear response to human activities and landscape conditions. In this regard the project has a series of key program elements that are directly linked to one another. Each program element is the responsibility of research collaborators from a variety of disciplines. Not all these research collaborators are grizzly bear biologists but each researcher brings a unique perspective to the research program objective and makes a significant contribution to the overall team in providing a necessary data component. It is also important to understand that each key program element is reliant on final products prepared by other research collaborators. For example, the development and testing of Resource Selection Function (RSF) models is dependant on the production and delivery of quality habitat map products. In our program the team from the University of Calgary is responsible for the production of these maps. The RSF models also require detailed bear habitat use information that is provided by grizzly bear biologists who capture and collar (GPS) bears each of the five field seasons. The GPS location data is compiled and integrated with extensive GIS data layers to enable the creation of RSF models. Other linkages of this nature are also in place between the DNA program element, the animal health component, graph theory modeling, etc. In short all the program elements are linked and dependant on each other to attempt to understand the complexities of grizzly bear response to human activities and changing landscapes.

Over the past 4 years the Foothills Model Forest Grizzly Bear Program has made significant advances in developing new tools and models to assist in broad scale land management practices and decisions concerning long-term conservation of grizzly bears.

1

In 2002, our research continued focusing on gathering data to further help us determine how human activities impact grizzly bears. The progress to date is summarized in the following document. The research presented in this annual report (year 4 of the 5 year program) is preliminary in nature but certainly provides the reader with a clear understanding of the progress and developments that have taken place since the program began in 1999. Other analysis and research papers are still underway within this program and will be distributed to program sponsors when completed. A full listing of technical papers is available on the following web-site: www.fmf.ab.ca. Our sincere thanks is extended to the many scientists, researchers, and biologists who work on this program with us and have collectively increased our understanding of grizzly bear ecology in this region.

1.2 Methods

1.2.1 Study Area

The study area remains the larger 9800 km2 area, which extends from the eastern portion of Jasper National Park (JNP) into the foothills towards the town of Edson. Hwy 16, and the Brazeau River define the north and south boundaries, respectively (Figure1.1).

1.2.2 Bear Capture and Handling

Methods for both the capture and monitoring of bears remain unchanged from the detailed description found in the year 1 progress report (Stenhouse and Munro 1999).

1.3 Results

In 2002, a total of 28 grizzly bears were handled, of which 21 (7 M/14 F) were fitted with GPS radio collars. We recollared 16 bears (4 M/12 F) that were collared during the first three years of this program. In the fall of 2002, 5 grizzlies (2 F/3 M) were captured and relocated from the Hinton municipal landfill station, including 2 research bears. Unfortunately, the results of the aging analysis for the 2002 captures will not be available until mid-April. Since 1999, a total of 61 individual grizzly bears have been handled and 34 individuals collared. A summary of the grizzly bears handled to date can be found in Table 1.1. The number of GPS locations from these collared bears is shown in Table 1.2. The number of data points for 2002 will likely increase when collars are retrieved from dens sites in the spring (2003).

2

http://www.fmf.ab.ca/

Figure 1.1. Grizzly bear research program study area with natural subregions delineated.

Table 1.1. Total number of handled and collared grizzly bears, 1999 - 2002.

Year Handled Collared Recaptures

1999 24; 11M/13F 19; 8M/11F n/a 2000 26; 11M/15F 21; 9M/12F 12; 4M/8F 2001 31; 10M/21F 23; 7M/16F 17; 6M/11F 2002 28; 10M/17F 21; 7M/14F 16; 4M/12F Total 61; 27M/33F 42; 16M/26F

3

Table 1.2. Total number of GPS location data from collared bears, 1999 - 2002.

Year Number of Bears Number of Locations 1999 13 6051 2000 20 9000 2001 21 11627 2002 21 7293 Total 34 unique individuals 33,971 locations

Reproductive status of radio-collared females has been tabulated on an annual basis through visual observations (Table 1.3). Annual cub survivorship can be tracked by comparing the number of cubs observed in a given year to the previous years observations. Reproductive data from collared females will eventually be used to construct an estimate of whether the sample population is increasing or decreasing. Table 1.3. Reproductive status of females in west-central Alberta, 1999 - 2002.

Year # Family Groups COY Yearlings > 1 Year Unknown

1999 2 - - - 4

2000 6 5 2 3 - 2001 12 12 - 5 1

2002 10 11 6 1 - In 2002, there were 6 known mortalities within our study area (Table 1.4). G036, an adult female, and G029 and adult male were both legally harvested in May. G020 and her yearling cub (gunk#10) were both illegally shot in September. G054, an adult male, was hit by a vehicle on a logging road in the Pembina River area in October. There was also one research mortality this year, G011, an adult female.

4

Table 1.4. Summary of grizzly bear mortalities in the west-central Alberta, 1999 – 2002

Bear Id Date Location Sex Age Kill Type

Research G015 16-May-99 Tri-Creek Area, AB M Adult Research Unknown 01 11-Sep-99 Cardinal Divide, AB F Adult Illegal Unknown 11 24-Sep-99 Embarrass River, AB F Adult Illegal Unknown 02 21-Oct-99 McLeod River, AB F COY Vehicle Unknown 05 24-May-00 McLeod River, AB F Adult Illegal Unknown 04 24-May-00 McLeod River, AB M COY Illegal Unknown 12 31-May-00 Swartz Creek, AB F Unk Legal

Research G032 31-May-00 Jacque Lake, AB F Subadult Research Unknown 03 01-Oct-00 Pembina River, AB F Adult Illegal

Research G026 19-Oct-00 Erith Creek, AB F Adult Unknown Research G102 16-May-01 F Subadult Legal Research G046 27-May-01 Brazeau River, AB M Subadult Natural Research G021 31-May-01 Coral Creek, AB M Adult Legal Research G024 25-Oct-01 Folding Mountain, AB M Adult Illegal Research G036 27-May-02 Warden Creek, AB F Adult Legal Research G029 28-May-02 Warden Creek, AB M Adult Legal Research G011 15-Jul-02 Center Creek, AB F Adult Research Research G020 20-Sep-02 Gregg River, AB F Adult Illegal Unknown 10 28-Sep-02 Gregg River, AB F Yearling Illegal

Research G054 12-Oct-02 Embarrass River, AB M Adult Vehicle The MCP home ranges for collared bears have been summarized in Table 1.5. On average males annual MCPs are 2.3x’s large than female home ranges. The average annual male and female home range size is 2224 km2 and 969 km2, respectively.

5

Table 1.5. Summary of MCP home ranges for grizzly bears in west-central Alberta, 1999-2002.

Bear Id Age at 1st Capture N Annual

(km2) Spring (km2)

Fall (km2)

Females G002 17 1792 694 634 599 G003 5 1344 792 746 376 G004 5 1656 471 408 237 G007 3 222 416 382 97 G010 13 1792 659 636 542 G011 6 360 484 484 21 G012 5 1713 1860 1474 1044 G013 4 193 2045 G016 5 1507 591 485 217 G020 4 2150 987 969 750 G023 11 1686 666 484 495 G026 4 246 1447 1438 164 G027 11 1110 2932 2465 1248 G028 6 1905 1300 1072 936 G034 21 287 799 144 679 G035 3 529 290 65 241 G036 3 623 1064 855 793 G037 3 459 742 304 651 G038 15 438 311 266 160 G040 3 862 1000 930 530 G042 18 1364 1271 410 1151 G048 Subadult 4 G100 Subadult 760 494 395 271

Males G001 Adult 878 1629 1629 814 G005 11 481 1296 1280 732 G006 16 630 1491 1491 186 G008 14 1076 1827 1822 708 G014 9 393 2638 2204 234 G017 7 1145 1694 1684 404 G024 6 1505 4314 2575 3817 G029 13 1980 3571 2485 2971 G033 3 2660 4748 3929 3335 G050 Adult 186 903 G054 Adult 125 358

6

2.0 ANIMAL HEALTH: PROGRESS REPORT Marc Cattet, School of Veterinary Medicine, University of Saskatoon, Saskatchewan, Canada. [email protected] Nigel Caulkett, School of Veterinary Medicine, University of Saskatoon, Saskatchewan, Canada. Gordon B. Stenhouse, Foothills Model Forest, Hinton, Alberta, Canada. [email protected]

2.1 Introduction

Since 2001, a major focus of our group has been to determine if grizzly bears show any evidence of “prolonged stress” and consequent health problems that might possibly result from poor habitat quality or rapidly changing habitat condition. In general, extreme or prolonged stress can have a deleterious effect on animal health, resulting in a state called “distress” (Moberg 2000). An animal in distress uses energy to cope with a threat (or stressor) at the expense of other biological functions including reproduction, tissue growth and maintenance, and immune response. Distress can alter biological function (e.g., failed reproduction, stunted growth, decreased immunity, etc.) and, if unchecked, eventually result in death. The persistence of a population is in part determined by the health of its component individuals. If the health of many animals is affected by prolonged stress or distress, the long-term persistence of the population is threatened. Conversely, if the health of most animals is normal, the population is likely to remain stable or grow with time.

In addition to our major focus on prolonged stress and health, we also investigated two other aspects of animal health during 2002. One was a comparison of select health data between the Foothills Model Forest (FMFGBP) and the Eastern Slopes Grizzly Bear Projects (ESGBP) in effort to determine why female grizzly bears of the ESGBP have an especially long interval between litters (approximately 5 years) and extremely low reproductive rates (i.e., <0.20 = the average number of viable offspring each adult female is expected to produce each year) (Herrero 2001). Although comparable reproductive data are not yet available from the FMFGBP, circumstantial evidence to date suggest female grizzly bears in the FMF population reproduce at an earlier age and reproductive rates are generally higher. The other aspect of animal health research initiated this past year was to evaluate the efficacy and safety of the drug combination medetomidine-Telazol (MZT), and the antagonist drug atipamezole, for reversible anesthesia in grizzly bears. MZT has proven to be an excellent drug combination for anesthesia of free-ranging Scandinavian brown bears (Arnemo et al. 2002), and may potentially offer a number of advantages over the drug combination (xylazine-Telazol or XZT) used currently for the FMFGBP.

7

mailto:[email protected]


2.2 Stress, Health, and Environment

Investigators: Marc Cattet, Nigel Caulkett, Janice Bahr, Judith Van Cleef, Matt Vijayan, John Boulanger, Julie Dugas, and Gordon Stenhouse. In 2001, the animal health component of the project was expanded to look at potential health consequences to grizzly bears as a result of alteration or loss of their habitat. In this regard, the key aspects of health chosen to focus upon were stress and reproduction. Evidence from other species, including humans, has shown that reproductive function is reduced, or fails, in animals suffering from prolonged stress or distress. If habitat alteration or loss is perceived as a significant stress by individual grizzly bears, it follows that their reproductive function may be comprised – a finding that could have serious implications for the long-term persistence of a population. Over the past two years, we have taken a two-prong approach to addressing the question, “Is there evidence from the health and environmental data collected on the Foothills Model Forest Grizzly Bear Project to indicate that the health of bears is being affected by chronic stresses associated with their physical environment?” One approach has been to seek biochemical indicators to quantify the level of chronic stress in a bear, independent of the acute (or short-term) stress imposed by capture and handling. The other approach has been to apply multivariate analyses on existing data to seek significant associations among measures of the health and environment of grizzly bears. Specifically, we want to know if bears living in areas of intense human activity show a characteristic health profile that differs from that of bears living in areas of low human activity and, if so, can we predict where a bear may be living on the landscape based on its health profile.

The development of biochemical indicators of chronic stress has been delayed largely by lack of specific funding for this research direction. Nevertheless, collection and storage of biological samples from captured bears continued throughout 2002 and effort was expanded to collaborate with, and gain support from, researchers working on similar types of issues in polar bears. More specifically, polar bear researchers in Canada and Norway have been concerned for a number of years about the potential effects of climatic warming and environmental exposure to anthropogenic contaminants upon the long-term persistence of polar bear populations (Stirling et al. 1999, Skaare et al. 2001, Polischuk et al. 2002). Despite apparent associations between some environmental variables and certain health characteristics in polar bears, there is presently little information available to determine if these associations actually reflect cause and effect. The ability to measure chronic stress would help significantly to strengthen the evidence for health effects in polar bears as a result of environmental quality or change. During 2003, we will collaborate with a group of polar bear researchers (Nick Lunn – Canadian Wildlife Sevice, Martyn Obbard – Ontario Ministry of Natural Resources, and Stephen Atkinson – Nunavut Department of Sustainable Development) to develop assays for cortisol binding globulin and stress-activated proteins that are specific to grizzly and polar bears, and assess the value of these biochemical constituents as indicators of chronic stress.

8

The application of multivariate analyses on existing health and environmental data to health profile grizzly bears was initiated in October 2002. A first and necessary step to developing health profiles has been to link all health and environmental data collected from 1999 to 2002 into a single data file that is amenable to statistical analyses. This formidable task was completed in January 2003, and statistical analyses are presently in progress. Although results are not yet available, a regression-based approach is being used with categorical and continuous variables to develop models where the health features (the independent variables) of individual animals can be used to predict the presence, absence, or quantity of a specific environmental feature (the dependent variable). In addition, this approach is also being used to detect any potential effects of repeated capture and handling on the health of individual bears.

9

2.3 Comparison of select health data between the Eastern Slopes (ESGBP) and the Foothills Model Forest Grizzly Bear Projects (FMFGBP)

Investigators: Marc Cattet, Nigel Caulkett, Mike Gibeau, Steve Herrero, Janice Bahr, Judith Van Cleef, and Gordon Stenhouse. The Eastern Slopes Grizzly Bear Project (ESGBP) began in 1994 in response to a need for scientific understanding of grizzly bears within an area called the Central Rockies Ecosystem (Herrero 2001). This area encompasses about 40,000 km2 and includes Banff, Yoho and Kootenay National Parks as well as Kananaskis Country and surrounding provincial lands in Alberta and British Columbia. Annually, a trapping and monitoring team has captured and fitted VHF radio transmitters onto grizzly bears within the Bow Valley Watershed and surrounding area. About 25 grizzly bears each year have had active transmitters. The sample of bears has been purposely biased to include more females than males because of the females’ critical role in reproduction. Monitoring has been conducted from air and ground with an attempt to locate all bears about once per week. In addition, there has been intensive daily monitoring of selected female bears. The project was designed to generate at least 100 reproductive years of data regarding adult female grizzly bears, a goal that has been reached in recent years (Herrero 2001).

A major research priority of the ESGBP has been to describe and understand grizzly bear population demography, an analysis based on 9 years of survival and reproductive data focusing on adult females. While results of the demographic analysis are preliminary and unpublished (Garshelis, Gibeau and Herrero unpublished data), a particularly striking finding in Eastern Slopes grizzly bears is their especially long interval between litters (4.4 years) relative to that determined in other studies (typically about 3 years for interior grizzly bear populations). In addition, the reproductive rate of Eastern Slopes grizzly bears is lower than that measured in other grizzly bear studies, i.e., 0.21 versus 0.23 to 0.43 in five other studies. Although comparable demographic characteristics cannot yet be determined for grizzly bears of the Foothills Model Forest Grizzly Bear Project (FMFGBP), circumstantial evidence to date suggest grizzly bears in the FMF population have a shorter interval between litters and a higher reproductive rate than Eastern Slopes grizzly bears. In effort to seek potential explanations for low cub production by Eastern Slope grizzly bears, a comparison of select health parameters was made between Eastern Slopes and FMF bears. The parameters considered were body condition as a reflection of nutrition and reproductive hormone levels as a reflection of reproductive function. The working hypothesis has been that reduced reproductive output in Eastern Slopes grizzly bears is a result of low energy uptake causing diminished reproductive function.

Using the definition of body condition as the “combined mass of fat and skeletal muscle in an animal relative to its body size”, we estimated and compared the body condition of grizzly bears captured in both projects by the Body Condition Index or BCI (Cattet et al. 2002).

10

BCI values are calculated as the standardized residuals from the regression of total body mass against a linear measure of size, body length, and range in value from –3.00 to +3.00. Eastern Slopes grizzly bears tended to be in poorer body condition than FMF grizzly bears captured at the same time of year, a difference that was most notable among adult males (Table 2.1).

Table 2.1. Comparison of Body Condition Index (BCI) values between the Eastern Slopes and Foothills Model Forest Grizzly Bear Projects for grizzly bears captured during either May or June.

Body Condition IndexA (mean ± SE; [n])

Sex (Age Class) ESGBP FMFGBP Statistical SignificanceB (p)

Female (all ages) -0.43 ± 0.13 [22] -0.13 ± 0.13 [37] 0.14ns

- subadult (< 5 yrs) -0.59 ± 0.38 [6] -0.25 ± 0.25 [16] 0.48ns

- adult (≥ 5 yrs) -0.37 ± 0.11 [16] -0.04 ± 0.13 [21] 0.08ns

Male (all ages) -0.16 ± 0.23 [21] +1.00 ± 0.22 [23] < 0.001***

- subadult (< 5 yrs) -0.45 ± 0.35 [9] +0.47 ± 0.29 [10] 0.06ns

- adult (≥ 5 yrs) +0.05 ± 0.31 [12] +1.41 ± 0.29 [13] 0.004**

A Mean BCI values were compared between studies using a t-test for two independent samples. B Statistical significance was assigned when the probability of a Type I error was equal to or less than 0.05.

Non-significant = ns, p ≤ 0.001 = **, and p ≤ 0.001 = ***. As an index of reproductive function, blood serum concentrations of various reproductive hormones were compared between grizzly bears captured in the two studies (Tables 2.2 and 2.3). In both sexes, luteinizing hormone (LH) concentrations were significantly lower in Eastern Slopes bears than in FMF bears. In mammals, LH is secreted from cells of the anterior pituitary gland and stimulates development of the ovaries in females and the testes in males. Further, LH stimulates secretion of sex steroids from the gonads – estrogens (including estradiol) from the ovaries and testosterone from the testes. Diminished secretion of LH can result in failure of gonadal function which manifests in females as cessation of reproductive cycles and in males as failure in production of normal numbers of sperms. Although information is lacking on normal serum concentrations of LH in grizzly bears, these results cannot be used to rule out the possibility of diminished reproductive function in Eastern Slopes bears, especially when considered in conjunction with body condition results (Table 2.1).

11

In mammals, the function of the reproductive system is dependent on the availability of energy in the environment. In several species, fasting and caloric restriction have been shown to cause the suppression of LH secretion, a mechanism that probably prevents energy being wasted for reproduction (Caprio et al. 2001, Gong 2002). Table 2.2. Comparison of reproductive hormone concentrations between the Eastern Slopes and Foothills Model Forest Grizzly Bear Projects for female grizzly bears captured by leg-hold snare during either May or June. Serum ConcentrationA (mean ± SE) Statistical

Hormone (Units) ESGBP (n = 14) FMFGBP (n = 29) SignificanceB (p)

Progesterone (ng/ml) 2.54 ± 0.63 2.82 ± 0.35 0.67ns

Estradiol (pg/ml) 10.6 ± 1.2 13.5 ± 1.0 0.11ns

Luteinizing hormone (ng/ml) 0.13 ± 0.05 0.39 ± 0.08 0.006**

Testosterone (ng/ml) 0.28 ± 0.05 0.29 ± 0.04 0.89ns

A Mean hormone concentrations were compared between studies using a t-test for two independent samples. B Statistical significance was assigned when the probability of a Type I error was equal to or less than 0.05.

Non-significant = ns and p ≤ 0.01 = **. Table 2.3. Comparison of reproductive hormone concentrations between the Eastern Slopes and Foothills Model Forest Grizzly Bear Projects for male grizzly bears captured by leg-hold snare during either May or June. Serum ConcentrationA (mean ± SE) Statistical

Hormone (Units) ESGBP (n = 16) FMFGBP (n = 17) SignificanceB (p)

Luteinizing hormone (ng/ml) 0.07 ± 0.04 0.33 ± 0.09 0.01*

Testosterone (ng/ml) 0.85 ± 0.26 0.91 ± 0.22 0.86ns

A Mean hormone concentrations were compared between studies using a t-test for two independent samples. B Statistical significance was assigned when the probability of a Type I error was equal to or less than 0.05.

Non-significant = ns and p ≤ 0.05 = *.

12

Results from a comparison of body condition and reproductive hormone concentrations between Eastern Slopes and FMF bears cannot be used to disprove the hypothesis that reduced reproductive output (long interval between litters and low reproductive rate) in Eastern Slopes grizzly bears is a result of low energy uptake (especially in males) causing diminished reproductive function. Future research directions should include the assessment of body condition in a larger sample of adult bears, especially males, and the assessment of reproductive function in female and male adult bears. Ideally, assessment of reproductive function should involve ultrasonographic examination of the gonads of both female and male bears, and spermatologic examination of semen samples collected from males. This data should be evaluated in relation to circulating concentrations of reproductive hormones measured in blood serum samples taken at the time of examination.

13

2.4 Reversible anesthesia of grizzly bears using medetomidine-zolazepam-tiletamine (MZT) and atipamezole

Investigators: Marc Cattet, Nigel Caulkett, and Gordon Stenhouse. Since the Foothills Model Forest Grizzly Bear Project (FMFGBP) began, there has been a progression in the anesthetic drugs used to capture grizzly bears. In 1999, Telazol (zolazepam and tiletamine) was the primary anesthetic drug used for the project, selected for its wide reputation as the drug-of-choice for the anesthesia of bears. Nevertheless, although generally effective and safe, Telazol does have some disadvantages including: (1) it must be administered in large volumes; (2) it has poor pain-killing (analgesic) effect; (3) its physiological effects cannot be reversed with an antagonist drug; and (4) bears anesthetized with Telazol may have prolonged recoveries lasting many hours, especially if multiple doses are administered.

From 1999 to 2002, xylazine-Telazol (XZT) replaced Telazol as the primary anesthetic drug for the project (Cattet et al. 2003). Its advantages over Telazol include: (1) administered in smaller volumes; (2) good analgesic effect; and (3) its physiological effects can be potentially reversed with an appropriate antagonist drug, e.g., yohimbine. Although XZT has proven to be safe and effective for the anesthesia of grizzly bears, the reversal of its physiological effects with yohimbine has not been consistent. While some bears have recovered from anesthesia within 5 to 10 minutes following administration of yohimbine, other bears have remained down for 30 minutes or more, apparently unaffected by the antagonist drug. Further, analysis of yohimbine dosage (mg/kg) versus recovery time data does not indicate this to be a dose-dependent effect (Pearson correlation – r = 0.21, p = 0.23, n = 66).

In effort to find a more reversible anesthetic drug, preliminary trials were carried out in 2002 with medetomidine-zolazepam-tiletamine (MZT) and the antagonist drug atipamezole. These drugs have proven to be excellent for reversible anesthesia of free-ranging polar bears (Cattet et al. 1997), black bears (Caulkett and Cattet 1997), and Scandinavian brown bears (Arnemo et al. 2002). However, preliminary data from five bears captured for the FMFGBP during 2002 do not appear to support the findings of these earlier studies (Table 2.4). Induction of anesthesia with MZT was good in only two of five animals, and reversal of anesthesia with atipamezole was moderately good in only one animal.

Although limited results from 2002 suggest MZT and atipamezole do not provide effective reversible anesthesia in grizzly bears, it may be premature to make final conclusions. Aside the small number of bears tested, other factors could have contributed to the substantial variation in results. The relative proportions of medetomidine and Telazol were not uniform among all bears and ranged in ratio from 1:27.5 (M:ZT) to 1:20.

14

Similarly, the amount of atipamezole in proportion to the amount of medetomidine administered varied considerably and ranged from 0.9 to 4.5 times the medetomidine dosage. Lastly two bears were free-ranging and anesthetized by remote injection from a helicopter whereas other bears were first restrained by leg-hold snare before drug administration. Given the relatively small number of animals captured each year during the FMFGBP, it is necessary to control as much as possible for variation in these variables (i.e., drug dosages and proportions, method of capture) before drawing final conclusions on the efficacy of these drugs. Therefore, although testing of MZT and atipamezole will continue in 2003, it will be done under the following conditions:

• Medetomidine and Telazol will be administered initially at a ratio of 1:20 (M:ZT) and a dosage of 0.125 mg/kg M + 2.5 mg/kg ZT based on estimated body weight;

• Atipamezole will be administered at approximately five times the medetomidine dosage which is equivalent to 0.625 mg/kg;

• Testing will be completed on a small number of bears (8 to 10) captured and restrained by leg-hold snare only; and

• Behavior of anesthetized bears will be monitored for a minimum of 20 minutes following administration of atipamezole.

Following testing during 2003, results will be re-evaluated and use of these drugs will cease if reversible anesthesia continues to be unreliable. However, if reversible anesthesia can be achieved consistently and safely, plans will be made to further evaluate the efficacy of these drugs in free-ranging bears anesthetized by remote injection from a helicopter.

Table 2.4 Anesthesia induction features of medetomidine (M) and Telazol (ZT) and reversal features of atipamezole in five grizzly bears captured during 2002.

Bear Induction

Dosage M + ZT (mg/kg)

Induction Time (min) & Quality

Reversal Dosage Atipamezole

(mg/kg)

Reversal Time (min) & Quality

G017 0.14 + 3.74 46 minA – poor 0.12 No data available

G008 0.15 + 4.12 33 minA – poor 0.26 > 92 min – poor

G050 0.08 + 2.29 6 minB – good 0.19 11 min – fair

G033 0.13 + 3.26 40 minC – poor 0.59 47 min – poor

G011 0.14 + 2.72 < 7 minB – good 0.39 > 25 min – poor A Required three darts projected from a helicopter to induce anesthesia. B Required one dart projected from the ground to induce anesthesia. C Required two darts projected from the ground to induce anesthesia.

15

2.5 References

Arnemo, J.M., S. Brunberg, P. Ahlqvist, R. Franzén, A. Friebe, P. Segerström, A. Söderberg, and J.E. Swenson. 2001. Reversible immobilization and anesthesia of free-ranging brown bears (Ursus arctos) with medetomidine-tiletamine-zolazepam and atipamezole: a review of 575 captures. In Proceedings of the Annual Meeting of the American Association of Zoo Veterinarians.

Caprio, M., E. Fabbrini, A.M. Isidori, A. Aversa, and A. Fabbri. 2001. Leptin in

reproduction. Trends in Endocrinology and Metabolism 12: 65-72. Cattet, M.R.L., N.A. Caulkett, M.E. Obbard, and G.B. Stenhouse. 2002. A body

condition index for ursids. Canadian Journal of Zoology 80: 1156-1161. Cattet, M.R.L., N.A. Caulkett, and S.C. Polischuk. 1997. Reversible immobilization of

free-ranging polar bears with medetomidine-zolazepam-tiletamine and atipamezole. Journal of Wildlife Diseases 33: 611-617.

Cattet, M.R.L., N.A. Caulkett, and G.B. Stenhouse. 2003. Anesthesia of grizzly bears

using xylazine-zolazepam-tiletamine or zolazepam-tiletamine. Ursus 14(1) (In press)

Caulkett, N.A. and M.R.L. Cattet. 1997. Physiological effects of medetomidine-

zolazepam-tiletamine immobilization in black bears. Journal of Wildlife Diseases 33: 618-622.

Gong, J.C. 2002. Influence of metabolic hormones and nutrition on ovarian follicle

development in cattle: practical implications. Domestic Animal Endocrinology 23: 229-241.

Herrero, S. 2001. A brief summary of the status of the Eastern Slopes Grizzly Bear

Project (ESGBP). Unpublished report prepared March 10, 2001. 9 pp. Moberg, G.P. 2000. Biological response to stress: implications for animal welfare. In

Moberg, G.P. and Mench, J.A. (Eds) The Biology of Animal Stress – Basic Principles and Implications for Animal Welfare. CABI Publishing, Wallingford Oxon, UK. pp. 1-22.

Polischuk, S.C., R.J. Norstrom, and M.A. Ramsay. 2002. Body burdens and tissue

concentrations of organochlorines in polar bears (Ursus maritimus) vary during seasonal fasts. Environmental Pollution 118: 29-39.

Skaare, J.U., A. Bernhoft, O. Wiig, K.R. Norum, E. Haug, D.M. Eide, and A.E.

Derocher. 2001. Relationships between plasma levels of organochlorines, retinol and thyroid hormones from polar bears (Ursus maritimus) at Svalbard. Journal of Toxicology and Environmental Health (Part A) 62: 227-241.

Stirling, I., N.J. Lunn, and J. Iacozza. 1999. Long-term trends in the population ecology

of polar bears in western Hudson Bay in relation to climatic change. Arctic 52: 294-306.

16

3.0 A SUMMARY OF THE 2002 FIELD RESEARCH PROGRAMME: FOODS, BERRIES, AND MICRO-SITE HABITAT SELECTION

Scott E. Nielsen, Department of Biological Sciences, University of Alberta, Edmonton,

AB T6G 2E9, Canada, email: [email protected] Robin H.M. Munro, Foothills Model Forest, Hinton, AB T7V 1X6, Canada, e-mail:

[email protected] Mark S. Boyce, Department of Biological Sciences, University of Alberta, Edmonton,

AB T6G 2E9, Canada, email: [email protected]

3.1 Summary

From May to September 2002 we established 496 use plots and 485 random vegetation plots within the Foothills Model Forest grizzly bear study area for the purpose of: (1) describing distribution patterns of grizzly bear foods; (2) contrasting random, available locations with GPS use locations at the micro-site habitat selection level; and (3) understanding the spatial and temporal distribution of berry production.

3.2 Random Plots

Random plots in 2002 were double stratified within the Integrated Decision Tree (IDT) landcover / landuse map and with greenness, a tasseled cap transformation of a 1999 (September) Landsat image. This allowed for similar allocation of plots within IDT landcover / landuse categories and across different levels of greenness, despite possible rare occurrence. Given that greenness was a continuous metric, greenness was stratified into 3 different levels (low, intermediate, and high) for each IDT category based on a standard deviation classification. Due to either ecological similarity in IDT categories or time and effort reasons, a number of IDT categories were merged. The following is a list of the IDT categories sampled in 2002 (by low, intermediate, and high greenness levels):

• Alpine/sub-alpine • Anthropogenic (roads, rail, transmission lines, pipelines) • Burn (1997 Gregg River fire) • Closed conifer • Deciduous forest (open and closed) • Herbaceous (<1800 metres) • Mixed forest • Open conifer • Regenerating forest (1956 Gregg River fire from GIS maps) • Shrub-bog-wetland (wet treed, wet open, and shrub)

Notice that we excluded the naturally non-vegetated IDT categories, rock, snow, ice, and water. In addition to these sampling strata, we further sampled clearcuts in order to understand food relationships within these management landscapes.

17




We stratified clearcuts by age, creating 10 separate strata (0-4 yrs, 5-9 yrs, to 45-49 yrs). Again, we labeled random sample locations as to whether they occurred within low, intermediate, or high greenness levels. This 2002 random sampling design was in departure from the 2001 protocol. In 2001, we simply generated random coordinates within seasonal MCP (minimum convex polygons) home ranges of grizzly bears that we were following for micro-site habitat selection analyses. Our switch in protocol for 2002 was done so as to allow for more meaningful examinations of grizzly bear food distributions within IDT categories, greenness, and along successional trajectories of clearcuts. As many additional plots as we could afford were established within historic fire stands in or near the 2001 study area. Spatial models of food distribution, similar to Nielsen et al. (2003) will be produced with these data, therefore allowing for a more mechanistic habitat map (food- or energy-based) of the area.

3.3 Use Plots

Future analyses will contrast random plots (weighted based on sampling stratification protocol) described above with micro-site habitat use locations (GPS grizzly bear location). This will allow for better understanding habitat selection at small within patch scales and for scaling relationships to other scales (patch-level) of selection.

3.4 Berry Plots

To further understand changes in fruit production for Acrtostaphylos uva-ursi, Shepherdia canadensis, Rubus ideaus, and Vaccinium spp., we revisited a subset of the 2001 random plots (permanent transects) in 2002. Unfortunately, field crews often found it difficult to re-locate permanent stakes, making these data invalid for repeated measures design monitoring of temporal changes in fruit production. Although not as powerful, we still have the ability to determine general trends in production by comparing all 2001 and 2002 plots. Finally, to better understand the autoecology of one of the primary critical hyperphagic foods, Shepherdia canadensis, we have marked 1,046 shrubs at 7 sites throughout the study area. Here we were interested in assessing the spatial and temporal variation in fruit production and demographic relationships for this shrub. All of these data are being integrated into databases and analyses will be forthcoming following additional measurements in the summer of 2003.

18

4.0. THE DIET OF GRIZZLY BEARS IN WEST-CENTRAL ALBERTA Robin H.M. Munro, Foothills Model Forest, Box 6330, Hinton, Alberta, T7V 1X6, Canada. [email protected] Gordon B. Stenhouse, Foothills Model Forest, Box 6330, Hinton, Alberta, T7V 1X6, Canada. [email protected]

4.1 Introduction

Evidence to date suggests that food is the most important resource for bears (Powell et al. 1997, Rogers 1976,1977). Large amounts of consistent food resources are necessary for daily and seasonal energetic requirements, as bears must acquire adequate fat reserves to survive denning (McLellan and Hovey 1995). Yearly and/or seasonal variations in food productivity have the potential to substantially alter a number of variables important for bears. For instance, in years when fall food productivity is low, black bears travel great distances to locate high-quality food resources (Garshelis and Pelton 1981, Garshelis et al. 1983, Rogers 1977, 1987). Ultimately, however, the temporal variation in food productivity can influence survival, reproductive fitness, and population viability (Bunnell and Tait 1981, Jonkel and Cowan 1971, Kolenosky 1990, Roger 1987).

Food habit studies provide information about the relative importance of different food to an animal’s diet. For grizzly bears this information is important to understand their life history and help investigators assess the seasonal importance of different habitats. Habitat use has been shown to vary with the availability, distribution and abundance of preferred foods (Hatler 1967, Jonkel and Cowan 1971) and changes in the seasonal use of plant and animal foods by grizzly bears is common. Information on the ecology of grizzly bears is important for their effective management, particularly under the increasing encroachment of human development.

4.2 Methods

4.2.1 Study Area

The study area was located along the eastern slopes of the Canadian Rocky Mountains in west-central Alberta’s Yellowhead Ecosystem (Fig. 1.1). The topographic gradient found in this area generates a diverse pattern of habitats and ecosystems, delineated primarily by elevation and distance east from the Front Ranges. Lying to the east of these Front Range ecosystems is a region (75% of total study area) of low-forested rolling hills referred to as the foothills. The foothills are comprised of a mosaic of lodgepole pine stands, black spruce (Picea mariana)/tamarack (Larix laricna) bogs, open marshes, low gradient riparian areas, clearcuts and roads. Forests in mountainous areas contained mixtures of spruce (Picea englemanii x glauca), and subalpine fir (Abies lasiocarpa). Alpine meadows, and rock outcrops were also common in the mountains.

19



There was a great diversity of mammals in the study area that were potential foods of grizzly bears. Moose (Alces alces), big horn sheep (Ovis canadensis), elk (Cervus elaphus), whitetail deer (Odocoileus virginianus), mule deer (O. hemionus) occurred in the study area throughout the year. Beaver (Castor canadensis), red squirrel (Tamiasciurus hudsonicus) were common. In the mountainous areas marmots (Marmota caligata) and ground squirrels (Spermophilus columbianus) were found at higher elevations. Other large carnivores in the study area included cougar (Felis concolor), wolf (Canis lupus) and black bear (Ursus americanus).

A strong gradient in land-use activities and human disturbance exists across the study area. Currently, oil and gas exploration and development, forestry, mining, hunting, settlement, tourism, and recreation dominate the human land use practices and activities.

4.2.2 Field Sampling

Between 2001 and 2002, scats from 11 different grizzly bears (3M 8F) were collected during field investigations of GPS radiolocation sites. Scats were collected at the site only if the scat age matched those expected from the focal animal. If more than one scat was found additional scats were collected only if it was determined that it potentially contained a different food make up. Scat collection varied with time. The number of scats collected was highest in June, July and August, when maximum man-power was available and lowest in early May and late September, when snow hampered fieldwork and available man-power was diminished. All scats were labelled and frozen for later analysis. Because of limited funding, only 409 of these samples were analysed for diet content.

4.2.3 Lab Analysis

Scat samples were thawed, re-hydrated, and washed with water using 2 mm screens. Any loss of small seeds or other items was noted during the washing. The washed scat was then placed in a container with approximately 1 litre of clean water and agitated to disperse food items. Small sub-samples were drawn and placed in a shallow pan to determine the relative percent of each food item. Percent volumes of food items were ocularly estimated. To account for biased estimates of food habits because of differences in the amount of recognizable fecal residue produced by different foods, correction factors were applied to our estimates of % volumes (Hewitt and Robbins 1996).

4.3 Results

Scat contents could be classified into 9 categories (Table 4.1), categories similar to those found in Mattson et al (1991) and McLellan and Hovey (1995). From late April through to June 1 hedysarum roots dominated the bear’s diet. Throughout the month of June animal matter, primarily ungulates dominate. Other food items of notable volume include horsetails (Equisetum spp.), grasses, and forbs including: cow parsnip (Heracleum lanatum), clover (Trifolium spp.) and dandelions (Taraxacum spp.).

20

Table 4.1. Percent volume of major food items in grizzly bear scats collected in west-central Alberta.

N Graminoids Horsetails Sedges Forbs Fruit Roots Insects Animal Matter MiscellaneousLate April 21 8 0 0 9 7 59 0.1 17 0.2Early May

18 3 8 0 0.6 2 86 0.1 0.6 0.1Late May 42 7 7 0 2 4 49 0.8 29 1Early June 44 8 4 0 5 6 17 1 58 0.1Late June 40 36 2 0 18 4 10 2 27 0.1 Early July 47 51 3 0.1 24 0.2 2 7 10 3 Late July 38 19 0.4 5 33 20 8 6 7 1 Early August 89 19 0.7 0 20 29 16 5 9 1Late August 32 4 0.1 0 7 49 19 4 13 4 Early September 35 15 0 0 3 62 4 0.3 9 6Late September 3 0.3 0 0 0.4 76 4 0.4 0 18

21

Insects, especially ants, were found primarily in scats collected between July and the end of August.

Although over-wintered Vaccinium vitis-ideae was found in May scats, fruit did not become prominent until early August when Shepherdia canadensis and V. membranceum became available. The berries dominated until late September. At this time there was a noticeable increase, again in the amount of hedysarum in the diet. Based on shifts in the diet of grizzly bears after only 2 years of data collection, three seasons were identified for the active part of the year (Figure 4.1). Roots and ungulates dominated the pre-green up season, or early spring and ended the last week of June. Green vegetation then dominates diet in the late spring until the end of July. Fruits and hedysarum became more voluminous in scats at the beginning of August until week 4 of September. These seasons may be subject to change pending 2003 field season results.

GraminoidsHorsetails

Sedges

Forbs

FruitRoots

InsectsAnimal MatterMisc

0%

20%

40%

60%

80%

100%

Late

Apr

il

Early

May

Late

May

Early

Jun

e

Late

Jun

e

Early

Jul

y

Late

Jul

y

Early

Aug

ust

Late

Aug

ust

Early

Sept

embe

r

Late

Sept

embe

r

Figure 4.1. Annual trends in the volume of residue of various food items found in

grizzly bear scats collected in west-central Alberta between 2001 and 2002. 4.3.1 Differences among years

Considerable variation in scat content was noted among years, in 2001 ungulates comprised very little of the spring diet of the bears. Although fruits were always the most abundant food item in late summer scats there was variation in the contribution of genera. In late summer 2001, soopolalllie was most common, while in late summer 2002 huckleberries dominated. Despite these differences some commonalities exist between years. In both years, roots were common in the diet of most bears in the early spring while grasses and forbs tended to appear and dominate the diet by late spring.

22

4.4 References

Bunnell, F.L., and Tait, D.E.N. 1981. Population dynamics of bear – implications. In Dynamics of large mammal populations. Edited by C.W. Fowler and T.D. Smith. John Wildy and Sons, New York. Pp. 75-98.

Garshelis, D. L., and M. R. Pelton. 1981. Movements of black bears in the Great Smoky

Mountains National Park. Journal of Wildlife Management 45:912-925. Garshelis, D.L., H. B. Quigley, C. R. Villarrubia, and M. R. Pelton. 1983. Diel

movements of black bears in the southern Appalachians. International Conference on Bear Research and Management 5:11-20.

Hatler, D.F. 1967. Some aspects in the ecology of black bear (Ursus americanus) in

interior Alaska. University of Alaska, Fairbanks. M.S. thesis. Jonkel, C.J., and I. McT. Cowan. 1971. The black bear in the spruce-fir forest. Wildlife

Monograph 27: 1-57. Kolenosky, G.B. 1990. Reproductive biology of black bears in east-central Ontario. In

Bears-Their biology and Management: Proceedings of the 8th International conference on Bear Research and Management, Victoria, B.C., February 1989. Edited by Laura M. Darling and W. Ralph Archibald. International Association for Bear Research and Management, University of Tennessee, Knoxville. Pp. 385-392

Mattson, D.J., Blanchard, B.M., and Knight, R.R. 1991. Food habits of Yellowstone

grizzly bears, 197701987 Canadian Journal of Zoology. 69: 1619 – 1629. McLellan, B. N., and F. W. Hovey. 1995. The diet of grizzly bears in the Flathead River

drainage of southeastern British Columbia. Canadian Journal of Zoology 73:704-712.

Powell, R. A., J. W. Zimmerman, and D. E. Seaman. 1997. Ecology and behavior of

North American black bears: home ranges, habitat and social organization. Chapman & Hall, London, England.

Rogers, L. L. 1976. Effect of mast and berry crop failures on survival, growth, and

reproductive success of black bears. Transactions of the North American Wildlife and Natural Resource Conference 41:432-438.

Rogers, L. L. 1977. Social relationships, movements, and population dynamics of

blackbears in northeastern Minnesota. Dissertation, University of Minnesota, Minneapolis, Minnesota, USA.

Rogers, L. L. 1987. Effects of food supply and kinship on social behavior, movements,

and population growth of black bears in northeastern Minnesota. Wildlife Monographs 97:1-72.

23

5.0. GRIZZLY BEAR HABITAT USE OF CLEARCUTS IN WEST-CENTRAL ALBERTA: INFLUENCE OF SITE, SILVICULTURE AND LANDSCAPE

STRUCTURE Scott E. Nielsen, Department of Biological Sciences, University of Alberta, Edmonton, AB T6G 2E9, Canada, email: [email protected] Mark S. Boyce, Department of Biological Sciences, University of Alberta, Edmonton, AB T6G 2E9, Canada, email: [email protected]

5.1 Introduction

Identifying critical habitats and potential impacts of human encroachment and land use management on rare or threatened species is a principle topic in conservation biology. For the Rocky Mountain ecosystems of the northern United States and southern Canada, habitat fragmentation and human encroachment, especially through industrial resource activities (forestry, oil, gas, and mining), have been identified as threats to the persistence of grizzly bear (Ursus arctos L.) populations (Banci et al., 1994; Clark et al., 1996; McLellan, 1998). Concurrent with industrial development has been the pervasive change in landscape structure from the suppression of fires over the past number of decades (Andison, 1998). This has led to increased woody encroachment of open areas and widespread succession of early seral or open structured stands (Tande, 1979; Andison, 1998; Rhumtella, 1999). Because grizzly bears are largely associated with early seral stage environments where food resources are greatest, succession to mature stages without further disturbance may ultimately lead to local population declines (McLellan and Hovey, 2001).

In the northeast slopes of the Canadian Rockies, herbaceous communities are uncommon, instead dominated by boreal and foothill forests. Given the extensive suppression of fires, commercial timber harvests now provide one of the only major mechanisms of disturbance and hence development of early seral stage communities. Although timber harvests may reduce habitat security for an area through associated increases in access (McLellan and Shackleton, 1988), the development of these early seral stage communities may be attractive to grizzly bears given the potential production of critical foods including fruits, ants, green herbaceous vegetation, and roots and other subterranean foods (Nielsen et al, 2003a).

Although numerous habitat use and selection studies have been completed for grizzly bears in the Rocky Mountains (McLellan and Shackleton, 1988; Mace et al., 1996; 1999; McLellan and Hovey, 2001; Gibeau et al., 2002; Nielsen et al., 2002; 2003a; Wielgus et al., 2002), few if any studies have assessed specific selection patterns for clearcuts. Examinations of habitat relationship associations have generally revealed consistent avoidance of regenerating clearcuts at the population level when compared with other habitats (Zager et al., 1983; McLellan and Hovey, 2001; Nielsen et al., 2002). Most studies, however, have occurred in mountainous landscapes, where open habitats are not limiting.

24



Few if any have examined how selection for clearcuts occurs in boreal-like foothill environment typical of west-central Alberta. Understanding patterns of habitat use and selection for clearcuts in these areas are crucial given the overlap in grizzly bear distribution and forestry activity. Identifying those clearcuts and site history treatments that are conducive for grizzly bear use provides information essential to the maintenance of grizzly bear populations within the region. Future harvesting can address or mitigate potential negative effects.

Here, we explored grizzly bear habitat selection for clearcuts in west-central Alberta. Nielsen et al. (2003a) suggested that such environments would be attractive to grizzly bears given their abundance of grizzly bear foods. We test this conclusion by examining global position system (GPS) radiotelemetry location data. Our objectives were three-fold: 1) determine differences in grizzly bear selection for matrix habitats (all habitats excluding clearcuts) and clearcuts; 2) examine whether there were differences in use of clearcuts by daytime and nighttime periods; and 3) describe patterns of clearcut selection by season for scarification, age, distance to edge of clearcut, perimeter:edge ratio, and micro-site terrain characteristics.

5.2 Methods

5.2.1 Study Area

The study area was a 2677-km2 landscape located in the eastern foothills of the Canadian Rocky Mountains of west-central Alberta (53°15′N, 117°30′W; Fig. 5.1). Within this area, a total of 525-km2 (19.6% of the area) of forest was harvested through commercial timber harvesting (clearcutting) activities that occurred between 1956 and 2001 (Fig. 5.2). Surrounding these regenerating clearcuts were closed conifer forests (41.4 %) and numerous minor forest (e.g., open conifer, deciduous, etc.) and non-forest (e.g., herbaceous, shrub, etc.) classes (39%) (Franklin et al., 2001; Table 5.1). Closed conifer stands were dominated by lodgepole pine (Pinus contorta) and to a lesser extent three species of spruce (Picea gluaca, P. mariana, and P. engelmannii) that were largely delineated by elevation and soil moisture. Minor forest areas of trembling aspen (Populus tremuloides) or balsam poplar (P. balsamifera), often mixed with other conifers and shrubs, were scattered throughout the region, but most notably in lower elevations or in riparian zones.

25

Figure 5.1. Study area map depicting elevation, local towns, overall Foothills Model

Forest study region (map extent), and secondary forestry study area for examining habitat.

26

Figure 5.2. Mapped clear-cuts by 5-year age class within the secondary forestry study

area in the upper foothills of west-central Alberta, Canada used for modeling grizzly bear habitat selection.

27

Table 5.1. Area (km2) and percent composition of landcover classes within the 2677km2 secondary study area near Hinton, Alberta. Landcover classes were determined from a remote sensing classification (Franklin et al., 2001) and forestry GIS data on cut-blocks.

Landcover class Area (km2) Percent closed conifer 1109.2 41.4 clear-cut 525.2 19.6 mixed forest 401.2 15.0 wetland-open bog 184.0 6.9 closed deciduous 117.9 4.4 wetland-treed bog 94.9 3.5 road/rail/pipeline/well site 76.9 2.9 non-vegetated 34.4 1.3 open conifer 31.5 1.2 shrub 31.2 1.2 other anthropogenic 26.6 1.0 herbaceous 17.6 0.7 water 15.4 0.6 burn 7.8 0.3 open deciduous 3.0 0.1 alpine/subalpine 0.2 0.01

Natural sub-region classification for the area was best described as upper foothills (Achuff, 1994), with elevations varying from 953 and 1975 metres. Summer and winter temperatures averaged 11.5 and –6.0° C respectively, with a normal annual precipitation of 538 mm (Beckingham et al., 1996). Prior to 1950, period stand replacing fires were the primary disturbance regime, averaging 20% of the landscape burned per 20-year period and a 100-year fire cycle (Andison, 1998). In contrast, since the 1950s there has been a substantial reduction in fires to the region associated with the initiation of industrial forestry and fire suppression (Tande, 1979; Andison, 1998; Rhemtulla, 1999). Although some stands in inaccessible areas are now in advanced stages of development due to fire suppression, large areas have or continue to be harvested providing the only major mechanism for renewal of early seral stage communities important for grizzly bear foods. 5.2.2 Grizzly bear locations, available locations, seasons, and time of day

From 1999 to 2002, we captured and collared 21 adult (>5 years old) and sub-adult (3-5 years old) grizzly bears from areas within or just surrounding the secondary study area using aerial darting and leg snaring techniques. Bears were fitted with either a Televilt GPS (global-positioning-system)-Simplex radiocollar or an ATS (Advanced Telemetry Systems) GPS radiocollar. Radio-collars were programmed for acquiring locations at different intervals but were bounded to occur between every 1-hr and 4-hrs.

28

Following retrieval of GPS collars and/or remote uploading of collars, grizzly bear locations were imported into a geographical information system (GIS) and used to delineate 100% minimum convex home range polygons (MCP; Samuel and Fuller, 1994). MCP home ranges were used for assessing available locations for each individual using a random point generator in ArcView 3.2. Sampling intensities for available locations were standardized to 5 locations/km2.

To account for variation in habitat use through time (Schooley, 1994; Nielsen et al., 2003b), we stratified grizzly bear location data into three seasons based on the phenology and use of available food resources for the area (Hamer and Herrero, 1987; Hamer et al., 1991; Munro et al., 2001; Nielsen et al., 2003b). The first season, hypophagia, was defined as those locations occurring between den emergence in April and 14 June. During this season, bears readily fed on roots of Hedysarum spp. and in some instances on carrion. The second season, early hyperphagia, was defined as the period between 15 June and 7 August. During this season, bears normally fed on ants, cow-parsnip (Heracleum lanatum), graminoids, sedges, and horsetail (Equisetum arvense). The third season, late hyperphagia, was defined as the period of 8 August to denning. During this season, bears sought out berries from Canada buffalo-berry (Shepherdia canadensis) and blueberries (Vaccinium spp.) followed by late season digging for Hedysarum spp.

Given that grizzly bears may avoid non-secure areas during daylight periods (Gibeau et al., 2002), we further assessed whether use of clearcuts occurred more than expected during nighttime hours. Daylight hours were defined as the period occurring between 7AM and 7PM, with night time hours corresponding to the period between 7PM to 7AM. Since GPS sampling intervals and collar performance may vary non-randomly over these periods, we first assessed whether acquisition rates for these periods were different for the global dataset.

5.2.3 Explanatory map variables

Clearcut Age, Size, and Silviculture.—Age of clearcut, in years, for each radiotelemetry and available location was determined from a GIS forestry polygon database provided by Weldwood of Canada Ltd. (Hinton, Alberta). Specific age of clearcut for each use or available location was determined for each radiotelemetry location based on the year of harvest and year of radiotelemetry date. Size of clearcut, in km2, was used as a predictor variable to assess whether small cut-blocks were more attractive to grizzly bears from a security or ecotone basis. Silvicultural and/or site preparation data were joined with GIS harvest polygons and stratified into nine separate treatments and a control (no treatment) (Table 5.2; 5.3).

Landscape Metrics.—We assessed the influence of two landscape metrics for clearcut polygons on grizzly bear habitat selection. These metrics were distance to clearcut edge (km) and area (km2) to perimeter (km) ratio. A 10-meter grid was created for the distance to edge (matrix habitat) variable by using the straight-line distance function in the Spatial Analyst extension of ArcGIS 8.1. The area to perimeter ratio, on the other hand was calculated for each polygon based on the estimated clearcut size and perimeter from a GIS.

29

Environmental Site Variables.—To assess how terrain and local site conditions influenced the pattern of habitat selection for grizzly bears in clearcuts, we used a 26.7 m digital elevation model (DEM) to describe the terrain and local micro-site conditions. From the DEM we estimated elevation (km) for each use or available location. We further derived two terrain related variables from the DEM. First, we estimated an index of wetness commonly referred to as the compound topographic index (CTI), previously found to correlate with several soil attributes including horizon depth, silt percentage, organic matter, and phosphorous (Moore et al. 1993, Geissler et al. 1995). Table 5.2. Explanatory map variables used for assessing grizzly bear habitat selection of clear-cuts in the upper foothills of west-central Alberta, Canada. Variable Code Description Type Range age Age of clear-cut (years) linear 0 to 46 age2 Age of clear-cut (years) squared linear 0 to 216 area Area (km^2) of clear-cut linear 0.003 to 2.683 area:perim Area (km^2) to perimeter (km) ratio linear 0.009 to 2.885 cti Compound topographic (wetness) index linear 7.34 to 24.45 cti2 Compound topographic (wetness) index squared linear 53.9 to 597.8 distedge Distance to edge of clear-cut (km) linear 0 to 0.8465 elev Elevation of location (m) linear 974 to 1712 elev2 Elevation of location (m) squared linear 948676 to 2930944scarYN Scarified clear-cut catigorical Yes or No scartype Scarification method catigorical 10 Categories solar Direct solar radiation (WH/m^2) on Day 172 linear 2391 to 4380

30

alarocqu

spelled with an e not an i

Table 5.3. Silviculture and site preparation treatments assessed for grizzly bear habitat selection.

Scarification type Description BLAD Blade (modified) BRAC Bracke DONM Donaren mound DRAG Drag (light or heavy) DRSF Drag- shark fin barrels DSTR Disc trencher EXCA Excavator mound and sc OTHR Other method (hand, unknown) PLOW Plough (Crossley, C&H, C&S ripper) NONE No silvicultural site preparation treatment

A CTI grid was calculated using the spatial analyst extension in ArcView 3.2 and a CTI script from P. Rho (2002). Second, we used the DEM to derive total potential direct incoming solar radiation (WH/m2) for summer solstice (day 172) using the Solar Analyst 1.0 extension in ArcView 3.2. 5.2.4 Modeling building strategies and statistical methods

Clearcut versus matrix habitat selection.—We used logistic regression with seasonal GPS radiotelemetry and available (random) locations to contrast differences in habitat selection for clearcut (1) and matrix habitats (0). Analyses were evaluated at the third-order scale (Johnson, 1980) following a ‘design III’ approach, where the individual identity of the animal is maintained for use and available samples (Thomas and Taylor, 1990). For each season, we calculated resource selection functions (RSF) for used and available resources using the following model structure (Manly et al., 1993):

( )11exp)( xxw β= [Eqn. 1],

where w(x) is the resource selection function (relative probability of occurrence) and β1 the selection coefficients for the clearcut variable x1. We accounted for autocorrelation between observations by assuming the unit of replication to be the individual and estimating robust variances using a Huber-White sandwich estimator among animals (White, 1980; Nielsen et al., 2002). We further corrected for habitat and terrain induced GPS collar bias (Obbard et al., 1998; Dussault et al., 1999; Johnson et al., 2002) by using probability sample weights for grizzly bear locations (not available locations sampled without bias). Probability sample weights were based on local models predicting fix acquisition as a function of terrain and habitat characteristics (Frair et al., in review). We reported results for the categorical variable clearcut as an odds ratio (reference being matrix habitats) based on the exponentiated form of β1.

31

Reported odds ratios were therefore interpreted as the likelihood that grizzly bears were using clearcuts compared with matrix habitats for a particular season. Finally, we tested whether GPS radiotelemetry data associated with clearcuts occurred more in the night or daytime periods by using 2×2 contingency χ2 test. Habitat selection for clearcuts.— For analyses of specific clearcut selection, we selected all locations occurring within clearcut polygons and divided our observations into 2 groups following a k-fold partition design (Fielding and Bell, 1997). The first group, the model-training group, represented a random 85% sub-sample of use and available locations for model development, while the remaining sub-sample (15%), the model-testing group, were used for assessing model performance by within-sample independent validation. The level of stratification among model training and testing groups were based on Huberty’s rule of thumb (Huberty, 1994). Using model-training data and explanatory map variables (Table 5.2) for each season we developed RSF models using estimated coefficients from logistic regression models and following structure (Manly et al., 1993):

( kk xxxxw )βββ +++= L2211exp)( [Eqn. 2]

where w(x) is the resource selection function and βi the selection coefficients for explanatory map variable xi. Linear explanatory map variables (Table 5.2) were assessed for collinearity problems through Pearson correlation (r) tests and variance inflator function (VIF) diagnostics. All variables with correlations (r) >|0.6|, individual VIF scores >10, or the mean of all VIF scores considerably larger than 1 (Chatterjee et al., 2000) were assumed to be collinear and not included within the same model structure. Area of clearcut and area to perimeter ratio were the only correlated (r = 0.68) variables, and thus were not considered together for inclusion in the same model. No further evidence of collinearity was evident using VIF tests. Using explanatory map variables, we generated 6 a priori candidate models (Table 5.4). We evaluated model selection for these 6 candidate models using Akaike’s information criteria (AIC; Burnham and Anderson, 1998, Anderson et al., 2000). Akaike weights (wi) were used to determine the approximate ‘best’ model given the data and candidate models tested. Methods described for controlling for autocorrelation and GPS radiotelemetry bias in the previous section (clearcut versus matrix habitat selection) were similarly used here. Using testing data, we assessed the predictive performance of models by comparing predictions with frequency of within-sample independent testing data (grizzly bear locations/use only) in specified habitat quality bins (Boyce et al., 2002). A total of 10 histogram equalized habitat bins were generated based on the distribution of available habitats in the study area from the AIC selected model. Models that perform well would be characterized by having successively more model-testing GPS radiotelemetry locations in higher value habitat bins, while poor habitat bins would contain few locations. We used a Spearman rank correlation (rs) to assess this relationship among number of observed grizzly bear model-testing locations per bin and bin rank (Boyce et al., 2002). Predictive models would be positive and significant.

32

5.3 Results

5.3.1 Clearcuts versus matrix habitat selection

Following retrieval and/or uploading of collars, 10,127 locations from 21 grizzly bears were recovered from the identified forestry study area. Of these, 2,405 or 23.7% of locations were from within identified clearcut polygons. The selection of clearcuts compared with matrix habitats by grizzly bears varied by season (Table 5.5). During hypophagia, grizzly bears selected clearcut habitats close to that which would be expected based on a comparison with matrix habitats and the availability of the two habitats. The estimated odds ratio for clearcut selection was 1.14 (95% C.I. = 0.88 to 1.46) times that of matrix habitats. In contrast, the early hyperphagia period had much higher rates of selection for clearcuts. During this season, clearcuts were selected over matrix habitats at an odds ratio of 1.56 (C.I. = 1.31, 1.85). Finally, during late hyperphagia, grizzly bears selected clearcut habitats close to that which would be expected, although slightly less than matrix habitats at an odds ratio of 0.85 (C.I. = 0.59, 1.23). Patterns of use for clearcuts and matrix habitats by grizzly bears were different for daytime (7AM to 7PM) and nighttime (7PM to 7AM) periods. Clearcuts were used more than expected during night (χ2 = 5.69, 1 df, P = 0.017). We found no evidence of bias in daytime versus nighttime acquisitions in the global dataset (χ2 = 1.25, 1 df, P = 0.264), thus suggesting a biological effect. Table 5.4. A priori seasonal candidate models for grizzly bears describing habitat selection for clear-cuts in the upper foothills of west-central Alberta, Canada. Model number, name, and structure are provided. Model No. Model Name Model Structure 1 Scarification model age + age2 + scarYN + area 2 Silviculture model age + age2 + scartype + area 3 Site model cti + cti2 + elev + elev2 + solar 4 Landscape model distedge + area:perim 5 Mixture model age + age2 + cti + cti2 + solar + scartype + distedge + area:perim

33

Table 5.5. Seasonal estimates of habitat selection for clear-cuts (1) by grizzly bears compared to matrix habitats (0; reference category) in the upper foothills of west-central Alberta, Canada. Robust 95% C.I. Odds 95% C.I. Season Coeff. S.E. lower upper Ratio lower upper hypophagia 0.128 0.128 -0.124 0.379 1.137 0.883 1.461 early hyperphagia 0.443 0.088 0.270 0.616 1.557 1.310 1.852 late hyperphagia -0.162 0.189 -0.531 0.208 0.850 0.588 1.231

5.3.2 Habitat selection within clearcuts

Hypophagia.—A total of 734 GPS radiotelemetry locations from 14 grizzly bears were acquired from within clearcuts during the hypophagia period. Of the 5 a priori models assessed for the season, the mixture model (#5) showed the greatest AIC support (Table 5.6). During this period, grizzly bears selected intermediate aged clearcuts that were more natural in shape (negative area to perimeter ratio), with use locations being closer to edges than random locations (Table 5.7). There did not appear to be any relationship among grizzly bear locations and the compound topographic index (CTI) of wetness, although the terrain variable of potential direct incoming solar radiation appeared important. Grizzly bears selected for areas of low solar radiation during this season. Lastly, silvicultural treatments were found to be important in selection of clearcuts. Grizzly bears appeared to use areas that were scarified with bracke, dragging, shark-fin barrel drags, disc-trencher, excavator, and plow treatments, although only shark-fin barrel dragging appeared to have a strong consistent effect (Fig. 5.3; Table 5.7). In contrast, bears tended to avoid blade and Donaren mound treated areas, although these treatments were also highly variable (Table 5.7). Estimation of the ‘other’ category was not possible due to perfect avoidance of these clearcuts during hypophagia, which may be more a function of the rarity of these treatments than actual strong avoidance. Predictive accuracy of the AIC selected hypophagia mixture model using withheld model-testing data was good with a significant positive Spearman rank correlation (rs = 0.915,P<0.001). Early Hyperphagia.— A total of 1005 GPS radiotelemetry locations from 15 grizzly bears were acquired from within clearcuts during the hypophagia period. Of the 5 a priori models assessed, the site model (#3) showed the greatest AIC support (Table 5.6). During this period, areas with high levels of direct incoming solar radiation were associated with locations of grizzly bears, while elevation and CTI were weakly related to animal locations. Predictive accuracy of the AIC selected early hyperphagia site model using model-testing data was good with a significant positive Spearman rank correlation (rs = 0.964, P<0.001). Model predictions for clearcuts, represented as habitat quality bins/ranks, are provided for this season and AIC selected model in figure 5.4a.

34

Late Hyperphagia.— A total of 642 GPS radiotelemetry locations from 9 grizzly bears were acquired from within clearcuts during the hypophagia period. Of the 5 a priori models assessed, the mixture model (#5) showed the greatest AIC support (Table 5.6). During this period, coefficients for age of clearcut, area to perimeter ratio, direct potential incoming solar radiation, and the compound topographic index of wetness were overlapping (95% C.I.) 0 suggesting somewhat weak relationships between explanatory variables and habitat selection exhibited by grizzly bears. However, distance to edge of clearcut was strongly negative suggesting security or ecotone relationships, while silviculture again appeared important clearcut selection. Grizzly bears appeared to select for areas that were scarified with bracke and shark-fin barrel dragging, although confidence intervals were large and overlapping zero (Figure 5.3; Table 5.7). In comparison, bears tended to avoid Donaren mound, dragging, disc-trencher, excavator, and plow treatments, but again variations around estimates were large. Like the hypophagia period, estimates of the ‘other’ category were not possible due to perfect avoidance. Predictive accuracy of the AIC selected late hyperphagia mixture model using model-testing data was good with a significant positive Spearman rank correlation (rs = 0.770, P=0.009).

35

Table 5.6. AIC selected models for hypophagia, early hyperphagia, and late hyperphagia periods. Number of parameters (Ki), model -2 loglikelihood, AIC, change in AIC (∆i) from lowest model, and Akaike weights (wi) of model support are reported.

Season Model K -2LL AIC ∆i wi

Hypophagia Scarification model 5 5947.4 5957.4 103.1 <0.001 Silviculture model 13 5914.9 5940.9 86.6 <0.001 Site model 5 6013.8 6023.8 169.5 <0.001 Landscape model 3 6023.1 6029.1 174.8 <0.001 Mixture model 17 5820.3 5854.3 0.0 1.0 Early hyperphagia Scarification model 5 7634.7 7644.7 180.2 <0.001 Silviculture model 13 7575.7 7601.7 137.2 <0.001 Site model 5 7454.5 7464.5 0.0 1.0 Landscape model 3 7641.2 7647.2 182.7 <0.001 Mixture model 17 7522.0 7556.0 91.5 <0.001 Late hyperphagia Scarification model 5 4914.1 4924.1 163.5 <0.001 Silviculture model 13 4830.2 4856.2 95.6 <0.001 Site model 5 4938.0 4948.0 187.4 <0.001 Landscape model 3 4906.3 4912.3 151.7 <0.001 Mixture model 17 4726.6 4760.6 0.0 1.0

36

Table 5.7. Estimated seasonal AIC selected model coefficients. Due to perfect avoidance relating to low sample sizes for the scarification treatment (scartype) ‘OTHR’, this category was not estimated.

Hypophagia Early hyperphagia Late hyperphagia Variable 95% CI 95% CI 95% CI Code Coef. lower upper lower upper Coef. upperage 0.045 0.151 -0.081 -0.2 §age2 -0.260 -0.030 0.210 -0.020cti -0.172 0.269 -0.108 -0.313 0.157 -0.173 0.488cti2 0.007 -0.011 0.008 -0.001 0.016 -0.0003 -0.0126 0.0120

0.025 -0.007 0.056 elev2

Coef. lower

0.098 0.039-0.150 0.430

-0.614 0.0970.025

elev § -0.079 -0.193 0.034 §solar -1.164 -1.794 -0.535 0.889 0.306 1.472 -0.170 -1.720 1.381scartype:

BLAD -0.268 -0.841 0.306 -0.030 -1.727 1.667BRAC 0.593 -0.224 1.411 0.407 -1.272 2.086

DONM -1.745 -4.173 0.683 -0.711 -2.770 1.347DRAG 0.387 -0.595 1.369 -0.358 -2.483 1.767DRSF 0.783 0.067 1.498 0.205 -1.627 2.036DSTR 0.166 -0.768 1.099 -0.273 -1.544 0.998EXCA 0.089 -1.285 1.463 -0.658 -0.301 1.695OTHR - - - - - - PLOW 0.470 -0.010 0.950 -0.343 -1.994 1.308

distedge -2.253 -4.288 -0.217 -3.518 -4.778 -2.259area:perim -4.823 -8.318 -1.329 -5.816 -13.279 1.648

§Coefficients for elev2 and solar are reported at 1000 times their value, while age2 is 100 times its value.

37

Figure 5.3. Estimated silvicultural coefficients for hypophagia and late hyperphagia. Site Preparation Treatment

BLAD BRAC DONM DRAG DRSF DSTR EXCA PLOW

Estim

ated

Coe

ffici

ent o

f Sel

ectio

n

-1.5

-1.0

-0.5

0.0

0.5

1.0

1.5

hypophagia late hyperphagia

a). hypophagia:

Figure 5.4. Spatial predictions of cut-block habitat quality for AIC selected candidate models by season (a. hypophagia; b. early hyperphagia; c. late hyperphagia).

38

b. early hyperphagia:

c. late hyperphagia:

Figure 5.4. (continued) Spatial predictions of cut-block habitat quality for AIC selected candidate models by season (a. hypophagia; b. early hyperphagia; c. late hyperphagia).

39

5.4 Discussion

We found seasonal selection for clearcuts over that of surrounding (matrix) habitats in west-central Alberta. The greatest use occurred during the early hyperphagic and hypophagic periods, with slightly less selection than matrix habitats during the late hyperphagic period. Not only were bears using clearcuts differentially at a seasonal temporal scales, but also at daily temporal scales. Bears were more likely to use clearcuts at nighttime than during daylight periods, suggesting that security was an important issue when bears were selecting clearcuts. Habitat selection models proved predictive for each season, relating to different variables for each period. Distance to edge and edge to perimeter ratio were important for both the hypophagia and late hyperphagic periods further suggesting that security or ecotone issues were important factors determining habitat use. Silviculture treatments for these two periods varied from negative to positive impacts when compared with non-treated clearcuts. Intermediate aged clearcuts were most frequently selected for in hypophagic and late hyperphagic periods. Myrmecophagy and frugivory, together with digging for roots and tubers, is largely considered the most important foraging activities for grizzly bears (Mattson, 1997; Welch et al., 1997; Elgmork and Unander, 1999; Swenson et al., 1999) and, excluding Hedysarum, these species all tended to maximize their occurrence at intermediate aged clearcuts (Nielsen et al., 2003a). Micro-site terrain features appeared to be more important than landscape metrics or silviculture during the early hyperphagic period when bears were most likely to use ants and herbaceous species. Sites with high solar radiation were used during this season suggesting that ants were being used.

5.5 Management Implications

Grizzly bear use of clearcuts was evident for the foothills near Hinton, Alberta. Use, however, occurred differentially depending on micro-site terrain, landscape metrics, and silvicultural history. Approaches to enhance and management grizzly bear habitat within harvest blocks of the foothills therefore appears to be feasible. Harvesting patterns that create higher edge to perimeter ratios and silvicultural practices that enhance grizzly bear foods (Nielsen et al., 2003a) may provide higher quality habitats for grizzly bears. Despite the potential for high quality grizzly bear habitat, methods for reducing sink phenomena must be addressed as these areas contain a high degree of access making mortality risk substantial.

40

5.6 References

Achuff, P.L., 1994. Natural Regions, sub-regions and natural history themes of Alberta; A Classification for Protected Areas Management. Alberta Environmental Protection.

Anderson, D.R., Burnham, K.P., Thompson, W.L., 2000. Null hypothesis testing:

problems, prevalence, and an alternative. Journal of Wildlife Management 64, 912-923.

Andison, D.W., 1998. Temporal patterns of age-class distributions on foothills landscapes in Alberta. Ecography 21, 543-550. Banci, V., Demarchi, D.A., Archibald, W.R., 1994. Evaluation of the population status of

grizzly bears in Canada. International Conference on Bear Research and Management 9, 129-142.

Beckingham, J.D., Corns, I.G.W., Archibald, J.H., 1996. Field guide to ecosites of west-

central Alberta. Natural Resources of Canada, Canadian Forest Service, Northwest Region, Northern Forest Centre, Edmonton, Alberta, Special Report 9.

Beers, T.W., Dress, P.E., Wensel, L.C., 1966. Aspect transformation in site productivity

research. Journal of Forestry 64, 691-692. Boyce, M.S., Vernier, P.R., Nielsen, S.E., Schmiegelow, F.K.A., 2002. Evaluating

resource selection functions. Ecological Modelling 157, 281-300. Burnham, K.P., Anderson, D.R., 1998. Model selection and inference: A practical

information-theoretic approach. Springer-Verlag, New York, New York, USA. Chatterjee, S., Hadi, A.S., Price, B., 2000. Regression analysis by example. Third edition.

John Wiley & Sons, New York, New York, USA. Clark, T.W., Paquet, P.C., Curlee, A.P., 1996. Large carnivore conservation in the Rocky

Mountains of the United States and Canada. Conservation Biology 10, 936-939. Delibes, M., Gaona, P., Ferreras, P., 2001. Effects of an attractive sink leading into

maladaptive habitat selection. American Naturalist 158, 277-285. Dussault, C., Courtois, R., Ouellet, J.P., Huot, J., 1999. Evaluation of GPS telemetry

collar performance for habitat studies in the boreal forest. Wildlife Society Bulletin 27, 965-972.

Elgmork, K., Unander, S., 1999. Brown bear use of ant mounts in Scandinavia. Ursus 10,

269-274.

41

Fielding, A.H., Bell, J.F., 1997. A review of methods for the assessment of prediction

errors in conservation presence/absence models. Environmental Conservation 24, 38-49.

Frair, J.L., Nielsen, S.E., Merrill, E.H., Lele, S., Boyce, M.S., Munro, R.H.M.,

Stenhouse, G.B., 2003. Removing habitat-induced, GPS-collar bias from inferences of habitat selection. Journal of Applied Ecology submitted.

Franklin, S.E., Stenhouse, G.B., Hansen, M.J., Popplewell, C.C., Dechka, J.A., Peddle,

D.R., 2001. An integrated decision tree approach (IDTA) to mapping landcover using satellite remote sensing in support of grizzly bear habitat analysis in the Alberta Yellowhead Ecosystem. Canadian Journal of Remote Sensing 27, 579-592.

Geissler, P.E., Moore, I.D., McKenszie, N.J., Ryan, P.J., 1995. Soil-landscape modeling

and spatial prediction of soil attributes. International Journal of GIS 9, 421-432. Gibeau, M.L., Clevenger, A.P., Herrero, S., Wierzchowski, J., 2002. Grizzly bear

response to human development and activities in the Bow River Watershed, Alberta, Canada. Biological Conservation 103, 227-236.

Hamer, D., Herrero, S., 1987. Grizzly bear food and habitat in the front ranges of Banff

National Park, Alberta. International Conference on Bear Research and Management 7:199-213.

Hamer, D., Herrero, S., Brady, K., 1991. Food and habitat used by grizzly bears, Ursus

arctos, along the continental divide in Waterton Lakes National Park, Alberta. Canadian Field Naturalist 105:325-329.

Huberty, C.J., 1994. Applied Discriminant Analysis. Wiley Interscience, New York. Johnson, C.J., Heard, D.C., Parker, K.L., 2002. Expectations and realities of GPS animal

location collars: results of three years in the field. Wildlife Biology 8, 153-159. Kansas, J.L., Riddell, R.N., 1995. Grizzly bear habitat model for the four contiguous

mountain parks. Second edition. Parks Canada. Mace, R.D., Waller, J.S., Manley, T.L., Lyon, L.J., Zuring, H., 1996. Relationships

among grizzly bears, roads and habitat in the Swan Mountains, Montana. J. Appl. Ecol. 33, 1395-1404.

Mace, R.D., Waller, J.S., Manley, T.L., Ake, K., Wittinger, W.T., 1999. Landscape

evaluation of grizzly bear habitat in western Montana. Conserv. Biol. 13, 367-377. Manel, S., Williams, H.C., Ormerod, S.J., 2001. Evaluating presence-absence models in

ecology: The need to account for prevalence. J. Appl. Ecol. 38, 921-931.

42

Manly, B.F.J., McDonald, L.L., Thomas, D.L., 1993. Resource selection by animals:

Statistical design and analysis for field studies. Chapman & Hall, London. Martin, P., 1983. Factors influencing globe huckleberry fruit production in northwestern

Montana. Int. Conf. Bear Res. and Manage. 5, 159-165. Mattson, D.J., 1997. Selection of microsites by grizzly bears to excavate biscuitroots.

Journal of Mammalogy 78, 228-238. McLellan, B.N., Shackleton, D.M., 1988. Grizzly bears and resource-extraction

industries: Effects of roads on behaviour, habitat use and demography. J. Appl. Ecol. 25, 451-460.

McLellan, B.N., 1998. Maintaining viability of brown bears along the southern fringe of

their distribution. Ursus 10, 607-611. McLellan, B.N., Hovey, F.W., 2001. Habitats selected by grizzly bears in a multiple use

landscape. Journal of Wildlife Management 65, 92-99. Moore, I.D., Geissler, P.E., Nielsen, G.A., Petersen, G.A., 1993. Terrain attributes:

estimation methods and scale effects. Pages 189-214 in Modeling change in environmental systems, A.J. Jakeman, M.B. Beck, and M. McAleer (eds.), Wiley, London.

Munro, R.H.M., Nielsen, S.E., Stenhouse, G.B., Boyce, M.S., 2001. Microsite habitat

selection by female grizzly bears. Pages 43-49 in Foothills Model Forest Grizzly Bear Research Program 2001 Annual Report, Stenhouse, G., Munro, R. (eds.). Hinton, Alberta.

Nagy, J.A., Hawley, A.W.L., Barrett, M.W., Nolan, J.W., 1989. Population

characteristics of grizzly and black bears in west central Alberta. Alberta Environment Centre, Vegreville, Alberta, Canada. 33 pages.

Nielsen, S.E., Boyce, M.S., Stenhouse, G.B., Munro, R.H.M., 2002. Modeling grizzly

bear habitats in the Yellowhead Ecosystem of Alberta: Taking autocorrelation seriously. Ursus 13, 45-56.

Nielsen, S.E., Munro, R.H.M., Bainbridge, E., Boyce, M.S., Stenhouse, G.B., 2003a.

Distribution of grizzly bear foods within clearcuts and reference forest stands of west-central Alberta. Forest Ecology and Management, In review.

Nielsen, S.E., Boyce, M.S., Stenhouse, G.B., Munro, R.H.M., 2003b. Development and

testing of phonologically driven grizzly bear habitat models. Ecoscience, 1-10.

43

Obbard, M.E., Pond, B.A., Perera, A., 1998. Preliminary evaluation of GPS collars for analysis of habitat use and activity patterns of black bears. Ursus 10:209-217.

Rho, P., 2002. Calculate compound topographic (wetness) index. Department of

Rangeland Ecology and Management, Texas A&M University. Rhemtulla, J.M., 1999. Eighty years of change: The montane vegetation of Jasper

National Park. MSc. Thesis, University of Alberta, Edmonton, Alberta, Canada. Roberts, M.R., Zhu, L., 2002. Early response of the herbaceous layer to harvesting in a

mixed coniferous-deciduous forest in New Brunswick, Canada. Forest Ecology and Management 155, 17-31.

Samuel, M.D., Fuller, M.R., 1994. Wildlife telemetry. Pages 370-418 in T.A. Bookhout

(ed.). Research and management techniques for wildlife and habitats. Fifth edition. The Wildlife Society, Bethesda, Maryland, USA.

Schirokauer, D.W., Boyce, H.M., 1998. Bear-human conflict management in Denali

National Park and Preserve, 1982-94. Ursus 10, 395-403. Schooley, R.L., 1994. Annual variation in habitat selection: patterns concealed by pooled

data. Journal of Wildlife Management 58, 367-374. Swenson, J.E., Jansson, A., Riig, R., Sandegren, F., 1999. Bears and ants: myrmecophagy

by brown bears in central Scandinavia. Can. J. Zool. 77, 551-561. Swets, J.A., 1988. Measuring the accuracy of diagnostic systems. Science 240, 1285-

1293. Thomas, D.L., Taylor, E.J., 1990. Study designs and tests for comparing resource use and

availability. Journal of Wildlife Management 54, 322-330. White, H. 1980. A heteroskedasticity-consistent covariance matrix estimator and a direct

test for heteroskedasticity. Econometric 48:817-838. Tande, G.F., 1979. Fire history and vegetation pattern of coniferous forests in Jasper

National Park, Alberta. Can. J. Bot. 57, 1912-1931. Voss, E.H., 1994. Flora of Alberta: A manual of flowering plants, conifers, ferns and fern

allies found growing without cultivation in the Province of Alberta, Canada. Second Edition revised by Packer, J.G., University of Toronto Press, Toronto, Ontario, Canada. Wielgus, R.B., Vernier, P., Schivatcheva, T., 2002. Grizzly bear use of open, closed, and

restricted forestry roads. Can. J. For. Res. 32, 1597-1606. Zager, P., Jonkel, C., Habeck, J., 1983. Logging and wildlife influence on grizzly bear

habitat in northwestern Montana. Int. Conf. Bear Res. and Manage. 5, 124-132.

44

6.0 DISTRIBUTION OF GRIZZLY BEAR FOODS IN CLEARCUTS AND REFERENCE FOREST STANDS OF WEST-CENTRAL ALBERTA

Scott E. Nielsen, Department of Biological Sciences, University of Alberta, Edmonton, AB T6G 2E9, Canada, email: [email protected] Robin H.M. Munro, Foothills Model Forest, Hinton, AB T7V 1X6, Canada, e-mail: [email protected] Erin L. Bainbridge, Resources and The Environment Program, Faculty of Environmental Design, University of Calgary, Calgary, AB, T2N 1N4, Canada, email: [email protected] Mark S. Boyce, Department of Biological Sciences, University of Alberta, Edmonton, AB T6G 2E9, Canada, email: [email protected]

6.1 Introduction

Identifying critical habitats and potential impacts of human encroachment and land use management on rare or threatened species is a principle topic in conservation biology. For the Rocky Mountain ecosystems of the northern United States and southern Canada, habitat fragmentation and human encroachment, especially through industrial resource activities (forestry, oil, gas, and mining), have been identified as threats to the persistence of grizzly bear (Ursus arctos.) populations (Banci et al., 1994; Clark et al., 1996; McLellan, 1998). Concurrent with industrial development has been the pervasive change in landscape structure from the suppression of fires over the past number of decades (Andison, 1998). This has led to increased woody encroachment of open areas and widespread succession of early seral or open structured stands (Tande, 1979; Andison, 1998; Rhumtella, 1999). Because grizzly bears are largely associated with early seral stage environments where food resources are greatest, succession to mature stages without further disturbance may ultimately lead to local population declines (McLellan and Hovey, 2001).

In the northeast slopes of the Canadian Rockies, herbaceous communities are uncommon, instead dominated by boreal and foothill forests. Without fire in this region, commercial timber harvests provide one of the only major mechanisms of disturbance and hence the development of early seral stage communities. Although timber harvests may reduce habitat security for an area through associated increases in access (McLellan and Shackleton, 1988), the development of these early seral stage communities may be attractive to grizzly bears given the potential production of critical foods including fruits, ants, green herbaceous vegetation, and roots and other subterranean foods.

45




Despite these potential food resources, research has found a general avoidance of regenerating clearcuts in the Rocky Mountains (Zager et al., 1983; McLellan and Hovey, 2001). Nielsen et al. (2003a), however, found selection for clearcuts by grizzly bears in the foothills of west-central Alberta, suggesting that food resources were substantial enough for animals to use such areas. Although numerous habitat use and selection studies have been completed for grizzly bears in the Rocky Mountains (McLellan and Shackleton, 1988; Mace et al., 1996; 1999; McLellan and Hovey, 2001; Gibeau et al., 2002; Nielsen et al., 2002; 2003a; Wielgus et al., 2002), relatively few studies have examined the impact of commercial timber harvesting on habitat food quality (see however, Martin, 1983; Zager et al., 1983; Knight, 1999). Models predicting food resources have recently been shown to outperform general habitat-vegetation classifications, suggesting that further examinations of ‘resource’ gradients are important for predicting and understanding grizzly bear habitat use, particularly at fine temporal scales (Nielsen 2003b).

Here, we explored grizzly bear foods within clearcuts and reference forest stands, attempting to provide further evidence of clearcut habitat selection observed by Nielsen et al. (2003a). Specifically, we investigated how scarification, canopy, age, and micro-site terrain characteristics influenced the occurrence of 13 grizzly bear foods and the production of fruits in clearcuts of west-central Alberta. Our objectives were three-fold: 1) determine differences in grizzly bear food occurrence for reference upland forest stands and clearcuts; 2) develop presence-absence models for grizzly bear foods within clearcuts describing micro-site patterns of occurrence; and 3) describe, for clearcuts, fruit production and differences in occurrence for fruit producing and non-fruit producing sites.

6.2 Methods

6.2.1 Study area

The study area was a 2677-km2 landscape located in the eastern foothills of the Canadian Rocky Mountains of west-central Alberta (53°15′N, 117°30′W; Fig. 6.1). Within this area, a total of 525-km2 (19.6% of the area) of forest was harvested through commercial timber harvesting (clearcutting) activities that occurred between 1956 and 2001 (Fig. 6.2). Surrounding these regenerating clearcuts were closed conifer forests (41.4 %) and numerous minor forest (e.g., open conifer, deciduous, etc.) and non-forest (e.g., herbaceous, shrub, etc.) classes (39%) (Franklin et al., 2001). Closed conifer stands were dominated by lodgepole pine (Pinus contorta Loudon) and to a lesser extent three species of spruce (Picea gluaca, P. mariana, and P. engelmannii) that were largely delineated by elevation and soil moisture. Minor forest areas of trembling aspen (Populus tremuloides) or balsam poplar (P. balsamifera), often mixed with other conifers and shrubs, were scattered throughout the region, but most notably in lower elevations. Natural sub-region classification for the area was best described as upper foothills (Achuff, 1994), with elevations varying from 953 and 1975 metres.

46

Figure 6.1. Study area map depicting elevation, local towns, overall Foothills Model

Forest study region (map extent), and secondary forestry study area for examining habitat selection related to clearcut harvesting in west-central Alberta, Canada. Location of principle study area within Alberta is depicted in the upper left portion of the figure.

47

Figure 6.2. Mapped clearcuts by 5-year age class within the secondary forestry study

area in the upper foothills of west-central Alberta, Canada used for modeling grizzly bear habitat selection.

48

Summer and winter temperatures averaged 11.5 and –6.0° C respectively, with a normal annual precipitation of 538 mm (Beckingham et al., 1996). Prior to 1950, period stand replacing fires were the primary disturbance regime, averaging 20% of the landscape burned per 20-year period and a 100-year fire cycle (Andison, 1998). In contrast, since the 1950s there has been a substantial reduction in fires to the region associated with the initiation of industrial forestry and fire suppression (Tande, 1979; Andison, 1998; Rhemtulla, 1999). Although some stands in inaccessible areas are now in advanced stages of development due to fire suppression, large areas have or continue to be harvested providing the only major mechanism for renewal of early seral stage communities important for grizzly bear foods. 6.2.2 Field Sampling

During the growing seasons (June through August) of 2001-02, we established 355 sample plots within identified clearcuts and 183 sample plots within reference forest stands. All reference forests were upland stands dominated by coniferous tree species having ≥20 % composition of lodgepole pine and not previously disturbed by anthropogenic activities. Based on geographical information system (GIS) fire history maps, reference forest plots ranged in age from 41 to 194 years and averaged 105 (±36.9 SD) years overall. Sampling procedures were the same for both clearcuts and forest stands. Plots were established in a random manner from GIS generated coordinates, although a proportion of plots were based on randomly selected grizzly bears locations identified from global position system (GPS) radiotelemetry data. To ensure approximately equal proportions of plots within different aged clearcuts, we stratified random locations in 2002 to occur within 5-year age classes. We navigated to all field coordinates using hand-held Garmin GPS III plus units, attempting to locate the plot centre to within no more than 10 metres of the coordinates.

At each field plot, we established a 20 metre transect running south-to-north in axis with the 10 m location being the plot centre. Five 0.5-m2 (70.7 mm × 70.7 mm) herbaceous quadrats were established along transects at 5 m intervals. Within these quadrats, we recorded the presence of 11 grizzly bear foods within the herbaceous layer (<1 m height, excluding Shepherdia canadensis at <0.5 m). To estimate the presence of Shepherdia canadensis within the shrub layer (>0.5 m height), we recorded its presence within a belt transect 1 × 20 m (20-m2) in size. At each plot, we further estimated the productivity of Vaccinium spp. and Shepherdia canadensis fruits. Berries were counted within quadrats for Vaccinium spp. and within the belt transect for Shepherdia canadensis. A sub-sample of ripe fruits was weighed for estimation of fresh weight (kg) productivity for each species. All productivity measures refer to per hectare sampling scales and for plots visited on or after July 15 (period at which we reliably felt berries were available for counting [not necessarily ripe]). Finally, we recorded the presence of ants (in mounds and/or woody debris) and ungulate pellets using meander searches within 10 m of either side of the transect (20 × 20 m; 400-m2). All analyses reported here were at the level of the plot. Herbaceous quadrats were therefore combined. Taxonomy of vascular plants follows that of Voss (1994).

49

6.2.3 Explanatory variables

To examine differences in occurrence for grizzly bear foods between clearcut and forest stands, we coded (dummy coding) each plot as to whether it was in a clearcut (1) or forested stand (0). For models specifically examining food occurrence within clearcuts, we further determined the age and silvicultural history of field plots using a GIS forestry database provided by Weldwood of Canada Ltd. (Hinton, Alberta). However, given the small sample of clearcuts visited relative to the availability of different silvicultural preparation treatments, we were forced to dissolve silvicultural history into either scarified (1) or non-scarified (0) treatments using dummy coding. To assess terrain-influenced conditions within clearcuts, we used a 26.7 m digital elevation model (DEM) that described the elevation and local micro-site conditions. From this DEM, we estimated elevation (km) for each clearcut plot. We further derived two terrain related variables from the DEM. First, we calculated an index of wetness commonly referred to as the compound topographic index (CTI). CTI has previously been found to correlate with several soil attributes including horizon depth, silt percentage, organic matter, and phosphorous (Moore et al., 1993, Geissler et al., 1995). We used CTI as a surrogate for soil type and soil moisture, since a soil survey was not available for the area. CTI was calculated from the DEM using the spatial analyst extension in ArcView 3.2 and a CTI script from P. Rho (2002).

Our second DEM derived variable was a site severity index (SSI; Nielsen and Haney, 1998), modified from the Beer’s aspect transformation (Beers et al., 1966). To calculate SSI we derived both an aspect and slope (degrees) grid for the region using the DEM and the spatial analyst extension. Aspect was transformed using a sine wave so as to range from –1 (mesic NE aspects) to +1 (xeric SW aspects). Next, the index was dampened based on slope so that the sine wave steadily was reduced towards 0 as the slope reached 0/flat surface (Nielsen and Haney, 1998). Our final explanatory variable was average canopy cover, estimated for each plot using a spherical densiometer (Lemon, 1956). Spherical densiometer readings were taken above each herbaceous quadrat, in each cardinal direction, and averaged over the entire plot (all five quadrats).

6.2.4 Modeling building strategies and statistical methods

Grizzly bear food occurrence within clearcuts and reference forests.—We used logistic regression to contrast the occurrence of 13 grizzly bear foods (Table 6.1) for clearcuts (1) and upland reference forests (0). Important food resources were based on local reported analyses (Hamer and Herrero, 1987; Nagy et al., 1989; Hamer et al., 1991; Kansas and Riddell, 1995; Munro et al., 2001). Here, we were interested in understanding the difference in occurrence for grizzly bear foods between the two categories and report all logistic regression results as odds ratios with the reference category being forest plots. Reported odds ratios were therefore interpreted as the likelihood that grizzly bear foods were occurring in clearcuts compared with reference forest stands. We used a likelihood ratio χ2 test to determine the significance of individual grizzly bear food models.

50

Distribution of grizzly bear foods in clearcuts.—We examined grizzly bear foods within clearcuts by modeling their occurrence as a function of canopy, age, scarification, elevation, CTI, and SSI. Clearcut plots were divided into 2 groups following a k-fold partition design (Fielding and Bell, 1997). The first group, the model-training group, represented a random 85% sub-sample of plots used for model development, while the remaining sub-sample (15%), the model-testing group, were used for assessing model performance by within-sample independent validation. Table 6.1. List of grizzly bear food items examined in clearcuts and upland forest stands of west-central Alberta. Grizzly bear food item

Type of food/feeding activity Season of use

ants myremocaphagy late spring to early fall Arctostaphylos uva-ursi fruits spring and late summer Equisetum spp. herbaceous summer Hedysarum spp. roots/tuber digging spring and fall Heracleum lanatum herbaceous summer Shepherdia canadensis fruits late summer and fall Taraxacum officinale herbaceous spring and summer Trifolium spp. herbaceous spring and summer ungulates (pellets) scavenging/carnivorous spring to early summer Vaccinium caespitosum fruits late summer and fall Vaccinium membranaceum fruits late summer and fall Vaccinium myrtilloides fruits late summer and fall Vaccinium vitis-idaea fruits late summer and fall We used Huberty’s rule of thumb (Hubery, 1994) to approximate the level of data stratification among model training and testing groups. Using model-training data and explanatory variables (Table 6.2), we developed logistic regression models describing the occurrence of each grizzly bear food item. Linear explanatory variables were assessed for collinearity prior to modeling through Pearson correlation (r) tests and variance inflator function (VIF) diagnostics. All variables with correlations (r) >|0.6|, individual VIF scores >10, or the mean of all VIF scores considerably larger than 1 (Chatterjee et al., 2000) were assumed to be collinear and not included within the same model structure. Age and canopy were found to be correlated (r = 0.66), and thus were not considered together for inclusion in the same model. No further evidence of collinearity was evident using VIF tests. Using these explanatory variables, we generated 6 a priori candidate models (Table 6.3). We evaluated model selection for these 6 models using Akaike’s information criteria (Burnham and Anderson, 1998, Anderson et al., 2000) with a small sample size correction (AICc).

51

Akaike weights (wi) were used to determine the approximate ‘best’ model given the data and candidate models tested for each species/bear food item. We assessed the fit and predictive accuracy of training data and individual models using Hosmer and Lemeshow (1980; 1989) goodness-of-fit χ2 tests (Ĉ) and receiver operator characteristic (ROC) area under the curve estimates respectively. Significant Ĉ values would indicate poor fit between the model and data, while ROC scores were assessed based on their value falling into one of three categories. ROC values ranging from 0.5 to 0.7 were taken to represent ‘low’ model accuracy, while values between 0.7 and 0.9 were considered ‘good’, and finally those above 0.9 were considered to have ‘high’ model accuracy (Swets, 1988; Manel et al., 2001). We used our withheld model-testing data to further assess the fit (Ĉ) and predictive performance (ROC) as model verification. Finally, as further validation, we assessed the predictive capacity of individual AICc selected grizzly bear food models for 136 independent field plots collected for separate purposes within the same study area in 2002. To determine the predictive capacity of models, we estimated the optimal probability cut-off points for AICc selected grizzly bear food models by maximizing both specificity and sensitivity curves simultaneously. Using AICc selected model coefficients, we estimated the probability of occurrence for each grizzly bear food and predicted either a presence (≥cut-off point) or absence (<cut-off point) for each species within each of the 136 independent plots. We estimated the percent correctly classified (PCC) for each species by determining the proportion of total plots correctly classified compared with the total number of observations. We considered models with PCC ≥70% to be considered good. Table 6.2. Environmental variables used to predict the occurrence of grizzly bear foods within west-central Alberta clearcuts. Variable code used for candidate models, variable description, units (with range for non quadratic parameters), and data sources are presented. Variable code Variable description Units and range Data source age age years (0 to 46) GIS forest polygons age2 quadratic of age years GIS forest polygons canopy canopy % canopy (0 to 100) Field measurements canopy2 quadratic of canopy % canopy Field measurements cti compound topographic index (CTI) index (8 to 21) GIS model from DEM cti2 quadratic of CTI index GIS model from DEM elev elevation metres (957 to 1596) DEM elev2 quadratic of elevation metres DEM scar scarification yes or no GIS forest polygons ssi site severity index index (-1 to 1) GIS model from DEM

52

Table 6.3. A priori candidate models used for assessing distribution of grizzly bear foods within clearcuts of west-central Alberta. Model number, parameter structure (variables), and total number of parameters (including constant) used for calculating Akaike weights (wi) for model selection.

Model number Model structure Ki

1 scar + age + age2 4 2 scar + canopy + canopy2 4 3 cti + cti2 + elev + elev2 + ssi 7 4 canopy + canopy2 3 5 scar + age + age2 + cti + cti2 + elev + elev2 + ssi 9 6 scar + canopy + canopy2 + cti + cti2 + elev + elev2 + ssi 9

Occurrence of berries and average productivity for clearcuts.—To determine the occurrence of fruit producing sites, we again used logistic regression and the 6 a priori candidate models. In total, we examined 6 species for fruit production: 4 species of Vaccinium; Arctostaphylos uva-ursi; and Sherperdia canadense (in the shrub layer). We used a conditional modeling strategy where only locations where the species was present were used to examine fruit occurrence. At these sites, we compared plots that lacked fruit production (0) with those where fruit production was present (1). Failure to discriminate the two events was interpreted to mean that berry production was random with respect to the examined variables and candidate models. Due to low sample sizes resulting from the absence of species and/or berry producing sites, we were forced to combine both 2001 and 2002 field seasons despite potential temporal differences. Similarly, Vaccinium caespitosum and V. membranaceum were too uncommon to model individually. Instead, we combined the two species into a Vaccinium caespitosum-membranaceum complex. Finally, we report average productivity of Arctostaphylos uva-ursi, Shepherdia canadensis, and 4 species of Vaccinium within clearcuts.

6.3 Results

6.31 Grizzly bear food occurrence in clearcuts versus upland forests

Ants, Equisetum, Hedysarum, Taraxacum officinale, Trifolium, and Vaccinium myrtilloides were much higher in frequency within clearcuts than upland reference forest stands, proving to be significantly different from one another (Table 6.4). In contrast, Vaccinium cespitosum, V. membranaceum, and V. vitis-idaea were more likely to occur in forest stands. Finally, the occurrence of Arctostaphylos uva-ursi, Heracleum lanatum, Shepherdia canadensis, and ungulate pellets within clearcuts were no different from that of reference forests (Table6.4).

53

6.3.2 Distribution of grizzly bear foods in clearcuts

Based on AICc weights (wi) there was large variation in support for the six a priori candidate models tested (Table 6.5). Only the scarification-canopy model (#2) had little to no support for any one grizzly bear food. Excluding 3 species of Vaccinium that all had high support for the micro-site model (#3), no obvious patterns were evident between candidate models and food groups. Table 6.4. Frequency of occurrence for 13 grizzly bear foods within clearcut (n = 355) and reference forest (n = 183) plots. Odds ratio (± S.E.) of finding grizzly bear foods (H-herbaceous layer; S-shrub layer) within clearcuts of west-central Alberta when compared to reference upland forest stands are reported from logistic regression models. Model likelihood ratio (LR) χ2 test and associated significance (P) levels are reported. Grizzly bear food item

Clearcut frequency

Forest frequency

Odds ratio S.E.

Model LR χ2 P

ants 65.9 26.2 5.44 1.098 78.42 <0.001Arctostaphylos uva-ursi 21.7 19.1 1.17 0.267 0.49 0.485 Equisetum spp. 43.9 24.6 2.40 0.486 20.01 <0.001Hedysarum spp. 10.7 2.7 4.27 2.069 12.29 0.001 Heracleum lanatum 4.2 5.5 0.76 0.320 0.41 0.523 Shepherdia canadensis-H 11.3 7.7 1.53 0.498 1.82 0.177 Shepherdia canadensis-S 17.8 14.2 1.30 0.330 1.12 0.290 Taraxacum officinale 38.9 4.4 13.92 5.253 88.92 <0.001Trifolium spp. 23.4 4.4 6.67 2.554 37.23 <0.001ungulates (pellets) 36.1 39.9 0.85 0.159 0.76 0.385 Vaccinium caespitosum 37.8 49.7 0.61 0.113 7.09 0.008 Vaccinium membranaceum 6.2 22.4 0.23 0.229 28.87 <0.001Vaccinium myrtilloides 14.7 8.7 1.79 0.540 4.02 0.045 Vaccinium vitis-idaea 51.0 81.4 0.24 0.517 50.22 <0.001

54

Table 6.5. Akaike (small sample size) model selection weights (wi) for grizzly bear foods in clearcuts of west-central Alberta. Shepherdia canadensis is reported for both the shrub-layer (S) and herbaceous-layer (H). Model with the greatest support per individual species is indicated in bold font. Grizzly bear food item Model 1 Model 2 Model 3 Model 4 Model 5 Model 6ants 0.013 0.002 <0.001 0.004 0.977 0.003 Arctostaphylos uva-ursi <0.001 <0.001 0.038 <0.001 0.070 0.892 Equisetum spp. <0.001 0.001 0.001 <0.001 0.063 0.935 Hedysarum spp. <0.001 <0.001 <0.001 <0.001 0.003 0.997 Heracleum lanatum 0.012 0.273 0.009 0.690 0.001 0.016 Shepherdia canadensis-H <0.001 <0.001 0.026 <0.001 0.898 0.621 Shepherdia canadensis-S <0.001 <0.001 <0.001 <0.001 0.692 0.308 Taraxacum officinale 0.002 0.188 <0.001 0.515 0.002 0.294 Trifolium spp. 0.006 0.292 0.004 0.623 0.001 0.073 ungulates (pellets) 0.548 0.115 <0.001 0.253 0.072 0.013 Vaccinium caespitosum 0.100 0.059 0.512 0.087 0.135 0.106 Vaccinium membranaceum 0.013 0.020 0.608 0.050 0.094 0.216 Vaccinium myrtilloides <0.001 <0.001 0.681 <0.001 0.166 0.153 Vaccinium vitis-idaea 0.234 0.132 0.034 0.097 0.388 0.114 Using likelihood ratio χ2 tests, we found all AICc selected models to be significant (Table 6.6), although the proportion of deviance explained varied from a low of 2.8% for Trifolium to a high of 31.3% for Hedysarum. There were no significant differences (p Ĉ) between training data and selected models for any individual grizzly bear food using goodness-of-fit tests. Testing data, however, revealed poor fit for 4 species: Equisetum, Hedysarum, Vaccinium membranaceum, and V. vitis-idaea (Table 6.6). Classification accuracy (ROC) for model training data proved poor (0.5─0.7) for 6 of 14 grizzly bear food items and good (0.7─0.9) for the remaining 8 items. Three additional species, however, decreased in accuracy with model testing data suggesting poor prediction for out-of-sample data. All 3 species were from the 4 species that showed poor fit of testing data. Using independent sample plots collected concurrently for other purposes, we found 7 of 12 species to have reasonably good prediction based on percent correctly classified (PCC) values exceeding 70% (Table 6.6). Overall, we found that ants, Arctostaphylos uva-ursi, Shepherdia canadensis, and Vaccinium myrtilloides to have consistently good fit, classification accuracy, and predictive capacity across all data. The other 9 food items proved to be either inconsistent or low in classification accuracy.

55

Estimated coefficients and significant predicted optimal responses showed strong selection for intermediate levels of overstory canopy cover and/or age of clearcut for a number of species, but most notably ants, Arctostaphylos uva-ursi, Hedysarum, Shepherdia canadensis, ungulate pellets, and Vaccinium vitis-idaea (Table 6.7 and 6.8). The scarification variable, although emerging in 8 of 14 AIC selected models, appeared to have strong negative effects on only Hedysarum and Shepherdia canadensis occurrence. Equisetum and Vaccinium vitis-idea, on the other hand, had positive scarification coefficients, albeit somewhat weak in strength. The compound topographic index (CTI) was included in many AICc selected models, but proved to be important for only Arctostaphylos uva-ursi (intermediate levels), Heracleum lanatum (high [wet] values), and Trifolium (high values). Likewise, elevation was included in a number of AICc selected models, but found to be an important predictor for the following 6 foods: Equisetum, Hedysarum, Shepherdia canadensis, Vaccinium caespitosum, V. membranaceum, and V. vitis-idaea. All of these species were predicted with their greatest occurrence at intermediate elevations, with the exception of Vaccinium membranaceum (high elevations) and V. vitis-idaea (low elevations). Finally, the site severity index (SSI) proved to be an important predictor of occurrence for 6 species. Three species, ants, Arctostaphylos uva-ursi, and Vaccinium caespitosum, were more likely to occur at high SSI levels (xeric slopes), while the remaining 3 species, Equisetum, ungulate pellets, and V. membranaceum were more likely at low SSI levels (mesic slopes). 6.3.3 Occurrence of berries and average productivity for clearcuts

For all 5 examined fruit species or species-complexes, we found consistent model selection for the non-linear canopy model (#4) predicting the occurrence of fruits given the presence of the species. Arctostaphylos uva-ursi, Vaccinium caespitosum-membranaceum, and V. vitis-idaea all had significant models overall and predicted optimal occurrences at intermediate canopy levels (Table 6.9). Arctostaphylos uva-ursi fruit production occurred in 45% of sites, Vaccinium caespitosum-membranaceum in 20 % of sites, and finally, Vaccinium vitis-idaea in 36% of sites (Table 6.9). Shepherdia canadensis and Vaccinium myrtilloides were both non-significant overall, but predicted to have optimal fruit occurrence at 0 and 100% canopy levels respectively. In total, fruit production occurred in 68% of sites for Shepherdia canadensis and 46% of sites for Vaccinium myrtilloides (Table 6.9). In some cases, as in Shepherdia canadensis, optimal predicted occurrence of fruit differed from that of the occurrence of the plant (Fig. 6.3). .

56

Table 6.6. AICc selected models with corresponding metrics of model significance, fit, and classification accuracy. All model likelihood ratio (LR) χ2 tests were significant at P<0.05. Percent deviance (Dev.) explained represents the reduction in the log likelihood from the null model. Probabilities for Hosmer and Lemeshow (1980) goodness-of-fit χ2 statistic (p Ĉ) are reported for model and data fit, while receiver operating characteristic (ROC) curves are used to assess model classification accuracy. Both in-sample (training data) and out-of-sample (testing data) data were used for fit and classification accuracy, while independent data from a concurrent study were used to assess the percent correctly classified (PCC). AICc selected model Model % Dev. Training data Testing data Optimal IndependentGrizzly bear food item name and number LR χ2 explained p Ĉ ROC p Ĉ ROC cut-off PCCants Mixture 1 (5) 62.33 16.0 0.338 0.755 0.631 0.742 0.5452 - Arctostaphylos uva-ursi Mixture 2 (6) 78.81 24.1 0.063 0.825 0.342 0.705 0.3120 72.79 Equisetum spp. Mixture 2 (6) 44.58 10.8 0.919 0.719 0.031 0.547 0.4838 87.50 Hedysarum spp. Mixture 2 (6) 62.59 31.3 0.442 0.860 <0.001 0.640 0.0934 91.18 Heracleum lanatum Canopy-only (4) 6.11 5.7 0.545 0.667 0.151 0.378 0.0275 47.06 Shepherdia canadensis-H Mixture 1 (5) 38.78 17.9 0.509 0.798 0.464 0.900 0.1381 95.59 Shepherdia canadensis-S Mixture 1 (5) 80.96 28.9 0.960 0.862 0.470 0.814 0.2115 83.82 Taraxacum officinale Canopy-only (4)

26.18 6.6 0.174 0.660 0.236 0.643 0.3522 64.71Trifolium spp. Canopy-only (4) 9.21 2.8 0.662 0.615 0.793 0.670 0.2569 58.09 ungulates (pellets) Scarification-Age (1) 18.25 4.6 0.958 0.644 0.564 0.604 0.4212 - Vaccinium caespitosum Microsite terrain (3) 12.42 3.1 0.325 0.617 0.596 0.616 0.3840 47.06Vaccinium membranaceum Microsite terrain (3) 14.58 10.7 0.151 0.716 0.027 0.612 0.0527 84.56 Vaccinium myrtilloides Microsite terrain (3) 51.31 19.4 0.306 0.806 0.722 0.750 0.1547 67.65 Vaccinium vitis-idaea Mixture 1 (5) 17.74 4.3 0.424 0.632 0.012 0.487 0.5231 86.03

57

Table 6.7. Estimated coefficients (βi) for AIC selected models describing the probability of occurrence for individual grizzly bear foods within clearcuts of west-central Alberta. Grizzly bear food item Scarify Age §Age2 Canopy §Canopy2 CTI CTI2 Elev §Elev2 SSI Constantants -0.579 0.303 -0.561 0.816 -0.027 -0.021 0.074 3.149 6.890Arctostaphylos uva-ursi -0.836

0.033 -0.582 1.691 -0.064

0.030 -0.156 6.677 -23.496Equisetum spp. 0.768 0.005 -0.249 -0.296 0.025 0.044 -0.173 -1.862 -28.222Hedysarum spp. -2.113 0.138 -2.648 -0.393

0.022 0.082 -0.367 -1.928 -44.268

Heracleum lanatum -0.008

0.304

-3.618Sherperdia canadensis-H -0.693 0.161 -0.259 -0.761 0.035 0.053 -0.252 0.949 -26.783Shepherdia canadensis-S -1.246 0.228 -0.348 -0.469

0.011 0.055 -0.264 2.848 -26.675

Taraxacum officinale -0.031 0.119 0.189Trifolium spp. -0.012

-0.035 -0.816

ungulates (pellets) -0.295 0.143 -0.234 -1.950Vaccinium caespitosum -0.008 0.0002 0.039 -0.154 2.490 -25.035Vaccinium membranaceum 1.819 -0.092 -0.068 0.264 -3.831 30.785Vaccinium myrtilloides 1.070 -0.050 -0.020 0.045 1.357 10.587Vaccinium vitis-idaea 0.647 0.094 -0.243 -0.049 -0.0002 -0.041 0.154 0.235 26.136 §

Note: for ease of reporting coefficients for age2 are 100, for canopy2 1,000, and for elev2 10,000 times their actual size.

58

Table 6.8. Predicted locations of optimal grizzly bear food occurrence for clearcuts of west-central Alberta. To estimate optimal locations for each variable, all other factors (variables) were held at their mean level. Bold font indicates significant trend. Grizzly bear food item Scarification Age Canopy CTI Elevation SSI ants No 26 40 17 1000 1.0 Arctostaphylos uva-ursi No 32 31 13 1000 1.0 Equisetum spp. Yes 0 11 20 1300 -1.0 Hedysarum spp. No 42 30 17 1150 -1.0 Heracleum lanatum Yes 0 100 21 1450 1.0 Shepherdia canadensis-H Yes 36 44 21 1100 -1.0 Shepherdia canadensis-S No 32 43 8 1050 1.0 Taraxacum officinale No 17 0 21 1350 -1.0 Trifolium spp. Yes 26 0 21 1250 -1.0 ungulates (pellets) No 35 51 17 1000 -1.0 Vaccinium caespitosum Yes 18 41 10 1300 1.0 Vaccinium membranaceum Yes 8 100 10 1600 -1.0 Vaccinium myrtilloides Yes 21 87 11 1000 1.0 Vaccinium vitis-idaea Yes 23 59 11 1000 1.0

59

Table 6.9. Percent frequency of fruit (after July 14), given the presence of the species, for clearcuts in west-central Alberta. Estimated coefficients (βi) for AIC selected models describing the probability of fruit presence (given its presence) and finally the estimated optimal position for fruit occurrence. % frequency Canopy Canopy2 Constant OFruit species/group of fruit β S.E. β S.E. β S.E. cArctostaphylos uva-ursi 45.0 0.122 0.063 -0.194 0.101 -1.016 0.700 Shepherdia canadensis-S 67.9 -0.003 0.052 -0.019 0.060 1.317 1.009 Vaccinium caespitosum / membranaceum

20.0 0.068 0.037 -0.101 0.050 -1.778 0.525

Vaccinium myrtilloides 45.5 0.008 0.047 0.020 0.061 -0.491 0.677 Vaccinium vitis-idaea 35.5 0.093 0.030 -0.073 0.032 -2.579 0.609 §

Note: for ease of reporting coefficients for age2 are 100 times their actual value.

60

For clearcut locations where fruiting species were present, fruit production ranged from 22,700 berries/ha for Acrtostaphylos uva-ursi to 200,446 berries/ha for Shepherdia canadensis (Table 6.10). For all 5 species and species-complexes, the average clearcut contained a total of 127,320 berries/ha or an estimated fresh weight production of 22.88 kg/ha (Table 6.10).

6.4 Discussion

We found ants, Equisetum, Hedysarum, Taraxacum officinale, Trifolium, and Vaccinium myrtilloides occurring with higher frequency in clearcuts compared with upland forest stands. Clearcut harvesting was appearing to benefit these species through the disturbance of soils and/or canopy structure. As would be expected and previously reported, both of the exotic species, Taraxacum officinale and Trifolium, responded favorably to clearcutting and mechanical disturbance (Haeussler et al., 1999; Stelfox et al., 2000; Roberts and Zhu, 2002). In contrast, Vaccinium cespitosum, V. membranaceum, and V. vitis-idaea were more likely to occur in forest stands, suggesting that forest harvesting had negative impacts on their incidence of occurrence. We found no evidence suggesting that Arctostaphylos uva-ursi, Heracleum lanatum, Shepherdia canadensis, and ungulate pellets occurred at different frequencies for clearcuts and upland forests. Excluding ungulate pellets, the remaining three species would be expected to have their greatest association with open forests. We would therefore expect clearcuts, many of which were open in nature, to have greater occurrence of these species. The lack of such differentiation suggests some possible reductions in the occurrence of these species if compared to similar open forested reference stands.

61

% Overstory Canopy0 20 40 60 80 100

Prob

abilit

y of

Occ

urre

nce

0.0

0.2

0.4

0.6

0.8

1.0Shepherdia canadensisPresence of Fruit

Figure 6.3. Predicted probability of occurrence for Shepherdia canadensis (solid line)

and for fruits (dashed line) given that the plant was present in clearcuts of west-central Alberta.

Table 6.10. Average fruit production (number of berries) per hectare (ha.) within clearcuts of west-central Alberta for cites where the species were present or for all sites. Total fruit production and estimated fresh weight (kilograms) provided for estimated production on an average hectare of clearcut (all sites regardless of whether the bear food item was present).

Fruit production-

presence sites Fruit production- all sites Grizzly bear food item Mean S.E. Mean S.E. kg Arctostaphylos uva-ursi 22,700 6,201 5,974 6,201 1.11 Shepherdia canadensis-S 200,446 117,380 36,924 22,222 6.65 Vaccinium caespitosum-membranaceum 38,844 15,868 23,658 9,501

5.73 Vaccinium myrtilloides 124,606 45,400 27,053 10,598 3.84 Vaccinium vitis-idaea 47,888 14,115 33,711 10,081 5.56

Total 127,320 22.88

62

Of the 6 a priori candidate models evaluated for the occurrence of grizzly bear foods, only the scarification-canopy model had little to no support for any one grizzly bear food item. The remaining 5 candidate models that included variables describing canopy, scarification, clearcut age, CTI, SSI, and elevation, proved useful descriptors of grizzly bear food occurrence. Goodness-of-fit (Ĉ) and model accuracy (ROC) generally revealed model fit and predictive accuracy, while model validation revealed reasonable accuracy for the majority of grizzly bear foods.

For many species, overstory canopy cover and age of clearcut were strong predictors of food occurrence. The scarification variable, although emerging in most AICc selected models, appeared to have strong negative effects on Hedysarum and Shepherdia canadensis occurrence, while Equisetum and Vaccinium vitis-idea showed positive, albeit somewhat weak, responses. Previous work has shown negative impacts of mechanical scarification for many of the Ericaceae shrubs (Zager et al., 1983; Haeussler et al., 1999; Roberts and Zhu, 2002), suggesting that the root and rhizome structure of many vegetative resprouting species were negatively impacted. Although elevation and compound topographic index (CTI) were included in numerous AICc models, they generally did not prove to be important predictors, in contrast to the site severity index (SSI).

For all 5 fruit species examined, canopy cover was the only variable useful for predicting the occurrence of fruit. Some responses of berry production differed across the canopy gradient from that of plant presence. Shepherdia canadensis was most likely in intermediate canopy levels, with fruit production occurring most often in open sites (Fig. 6.3). Hamer (1996) found similar patterns between canopy and fruit production of Shepherdia canadensis along the East Slopes of the Canadian Rockies near Calgary, Alberta. Fruit production dropped precipitously with increasing canopy cover, especially when canopy cover exceeded 50%. For all 5 species and species-complexes, the average clearcut contained a total of 127,320 berries/ha or an estimated fresh weight production of 22.88 kg/ha. Taken together with herbaceous matter, roots/tubers (Hedysarum), and the substantial biomass of ants available (Elgmork and Unander 1999; Swenson et al., 1999), as well as the potential for ungulate carcasses and/or predation of calves/fawns, the biomass of foods available in clearcuts is substantial. It seems reasonable to assume therefore that the documented use of clearcuts from Nielsen et al. (2003b) was generally due to feeding activities associated with attractive high levels of critical food resources available within an area that is largely food limited. Our results differ from McLellan and Hovey (2001), where general avoidance of clearcuts in the Flathead valley of southeast British Columbia was found, where they suggested that food quality was poor. Differences may have related to the general composition of vegetation and forests between the two areas or the reference area chosen for comparison. Natural regenerating burns, when compared with clearcuts, have typically shown greater productivity for some critical bear foods (Martin, 1983; Zager et al., 1983; Knight, 1999). We suggest that although strong use of clearcuts was documented and the availability of foods was high, selection for these sites may represent attractive sink habitat (Delibes et al., 2001), given the large degree of human access and the risk of mortality associated with these areas (Nielsen et al., unpublished manuscript).

63

6.5. Management Implications

Clearcut harvesting impacts the occurrence and production of grizzly bear foods. Methods of promoting grizzly bear food occurrence through passive management methods, including the technique of scarification/site preparation, may be required. Further active management, such as the planting of fruit producing shrub seedlings, particularly Shepherdia canadensis, may further mitigate negative effects of harvesting. Overall, we found clearcuts in the foothills of west-central Alberta to be major sources of grizzly bear foods. Despite this attraction and possible food related management, further control and/or examination of access related issues within clearcuts is requires. Without control of human access or local education programmes (McLellan, 1998; Schirokauer and Boyd, 1998), the phenomena of ‘attractive sinks’ may lead to local population declines (Delibes et al., 2001).

6.6 References

Achuff, P.L., 1994. Natural Regions, sub-regions and natural history themes of Alberta; A Classification for Protected Areas Management. Alberta Environmental Protection.

Anderson, D.R., Burnham, K.P., Thompson, W.L., 2000. Null hypothesis testing:

problems, prevalence, and an alternative. Journal of Wildlife Management 64, 912-923.

Andison, D.W., 1998. Temporal patterns of age-class distributions on foothills

landscapes in Alberta. Ecography 21, 543-550. Banci, V., Demarchi, D.A., Archibald, W.R., 1994. Evaluation of the population status of

grizzly bears in Canada. International Conference on Bear Research and Management 9, 129-142.

Beckingham, J.D., Corns, I.G.W., Archibald, J.H., 1996. Field guide to ecosites of west-

central Alberta. Natural Resources of Canada, Canadian Forest Service, Northwest Region, Northern Forest Centre, Edmonton, Alberta, Special Report 9.

Beers, T.W., Dress, P.E., Wensel, L.C., 1966. Aspect transformation in site productivity

research. Journal of Forestry 64, 691-692. Burnham, K.P., Anderson, D.R., 1998. Model selection and inference: A practical

information-theoretic approach. Springer-Verlag, New York, New York, USA. Chatterjee, S., Hadi, A.S., Price, B., 2000. Regression analysis by example. Third edition.

John Wiley & Sons, New York, New York, USA. Clark, T.W., Paquet, P.C., Curlee, A.P., 1996. Large carnivore conservation in the Rocky

Mountains of the United States and Canada. Conservation Biology 10, 936-939. Delibes, M., Gaona, P., Ferreras, P., 2001. Effects of an attractive sink leading into

maladaptive habitat selection. American Naturalist 158, 277-285.

64

Elgmork, K., Unander, S., 1999. Brown bear use of ant mounts in Scandinavia. Ursus 10,

269-274. Fielding, A.H., Bell, J.F., 1997. A review of methods for the assessment of prediction

errors in conservation presence/absence models. Environmental Conservation 24, 38-49.

Franklin, S.E., Stenhouse, G.B., Hansen, M.J., Popplewell, C.C., Dechka, J.A., Peddle,

D.R., 2001. An integrated decision tree approach (IDTA) to mapping landcover using satellite remote sensing in support of grizzly bear habitat analysis in the Alberta Yellowhead Ecosystem. Canadian Journal of Remote Sensing 27, 579-592.

Geissler, P.E., Moore, I.D., McKenszie, N.J., Ryan, P.J., 1995. Soil-landscape modeling

and spatial prediction of soil attributes. International Journal of GIS 9, 421-432. Gibeau, M.L., Clevenger, A.P., Herrero, S., Wierzchowski, J., 2002. Grizzly bear

response to human development and activities in the Bow River Watershed, Alberta, Canada. Biological Conservation 103, 227-236.

Hamer, D., Herrero, S., 1987. Grizzly bear food and habitat in the front ranges of Banff

National Park, Alberta. International Conference on Bear Research and Management 7:199-213.

Hamer, D., Herrero, S., Brady, K., 1991. Food and habitat used by grizzly bears, Ursus

arctos, along the continental divide in Waterton Lakes National Park, Alberta. Canadian Field Naturalist 105:325-329.

Hamer, D., 1996. Buffaloberry [Shepherdia canadensis (L.) Nutt.] fruit production in

fire-successional bear feeding sites. J. Range Manage. 49, 520-529. Haeussler, S., Bedford, L., Boateng, J.O., MacKinnon, A., 1999. Plant community

responses to mechanical site preparation in northern interior British Columbia. Can. J. For. Res. 29, 1084-1100.

Hosmer, D.W.Jr., Lemeshow, S., 1980. Goodness-of-fit tests for the multiple logistic

regression model. Communications in Statistics A9, 1043-1069. Hosmer, D.W.Jr., Lemeshow, S., 1989. Applied logistic regression. John Wiley & Sons,

New York, New York. Huberty, C.J., 1994. Applied Discriminant Analysis. Wiley Interscience, New York. Kansas, J.L., Riddell, R.N., 1995. Grizzly bear habitat model for the four contiguous

mountain parks. Second edition. Parks Canada. Knight, R.E., 1999. Effects of clearcut logging on buffaloberry (Shepherdia canadensis)

abundance and bear myrmecophagy in the Flathead River drainage, British Columbia. Thesis, University of Alberta. Edmonton, Alberta, Canada.

65

Lemon, P.E., 1956. A spherical densiometer for estimating forest overstory density.

Forest Science 2, 314-320. Mace, R.D., Waller, J.S., Manley, T.L., Lyon, L.J., Zuring, H., 1996. Relationships

among grizzly bears, roads and habitat in the Swan Mountains, Montana. J. Appl. Ecol. 33, 1395-1404.

Mace, R.D., Waller, J.S., Manley, T.L., Ake, K., Wittinger, W.T., 1999. Landscape

evaluation of grizzly bear habitat in western Montana. Conserv. Biol. 13, 367-377. Manel, S., Williams, H.C., Ormerod, S.J., 2001. Evaluating presence-absence models in

ecology: The need to account for prevalence. J. Appl. Ecol. 38, 921-931. Martin, P., 1983. Factors influencing globe huckleberry fruit production in northwestern

Montana. Int. Conf. Bear Res. and Manage. 5, 159-165. McLellan, B.N., Shackleton, D.M., 1988. Grizzly bears and resource-extraction

industries: Effects of roads on behaviour, habitat use and demography. J. Appl. Ecol. 25, 451-460.

McLellan, B.N., 1998. Maintaining viability of brown bears along the southern fringe of

their distribution. Ursus 10, 607-611. McLellan, B.N., Hovey, F.W., 2001. Habitats selected by grizzly bears in a multiple use

landscape. Journal of Wildlife Management 65, 92-99. Moore, I.D., Geissler, P.E., Nielsen, G.A., Petersen, G.A., 1993. Terrain attributes:

estimation methods and scale effects. Pages 189-214 in Modeling change in environmental systems, A.J. Jakeman, M.B. Beck, and M. McAleer (eds.), Wiley, London.

Munro, R.H.M., Nielsen, S.E., Stenhouse, G.B., Boyce, M.S., 2001. Microsite habitat

selection by female grizzly bears. Pages 43-49 in Foothills Model Forest Grizzly Bear Research Program 2001 Annual Report, Stenhouse, G., Munro, R. (eds.). Hinton, Alberta.

Nagy, J.A., Hawley, A.W.L., Barrett, M.W., Nolan, J.W., 1989. Population

characteristics of grizzly and black bears in west central Alberta. Alberta Environment Centre, Vegreville, Alberta, Canada. 33 pages.

Nielsen, S.E., Haney, A., 1998. Gradient responses for understory species in a bracken-

grassland and northern dry forest ecosystem of northeast Wisconsin. Transactions of the Wisconsin Academy of Sciences, Arts and Letters. 86, 149-166.

Nielsen, S.E., Boyce, M.S., Stenhouse, G.B., Munro, R.H.M., 2002. Modeling grizzly

bear habitats in the Yellowhead Ecosystem of Alberta: Taking autocorrelation seriously. Ursus 13, 45-56.

66

Nielsen, S.E., Boyce, M.S., Stenhouse, G.B., 2003a. Grizzly bear habitat selection for clearcuts in west-central Alberta: Influence of site, silviculture, and landscape structure. Forest Ecology and Management, In review.

Nielsen, S.E., Boyce, M.S., Stenhouse, G.B., Munro, R.H.M., 2003b. Development and

testing of phonologically driven grizzly bear habitat models. Ecoscience, 1-10. Rhemtulla, J.M., 1999. Eighty years of change: The montane vegetation of Jasper

National Park. MSc. Thesis, University of Alberta, Edmonton, Alberta, Canada. Roberts, M.R., Zhu, L., 2002. Early response of the herbaceous layer to harvesting in a

mixed coniferous-deciduous forest in New Brunswick, Canada. Forest Ecology and Management 155, 17-31.

Schirokauer, D.W., Boyce, H.M., 1998. Bear-human conflict management in Denali

National Park and Preserve, 1982-94. Ursus 10, 395-403. Stelfox, J.G., Stelfox, J.B., Bessie, W.C., Clark, C.R., 2000. Longterm (1956-1996)

effects of clearcut logging and scarification on forest structure and biota in spruce, mixedwood, and pine communities of west-central Alberta. Report.

Swenson, J.E., Jansson, A., Riig, R., Sandegren, F., 1999. Bears and ants: myrmecophagy

by brown bears in central Scandinavia. Can. J. Zool. 77, 551-561. Swets, J.A., 1988. Measuring the accuracy of diagnostic systems. Science 240, 1285-

1293. Tande, G.F., 1979. Fire history and vegetation pattern of coniferous forests in Jasper

National Park, Alberta. Can. J. Bot. 57, 1912-1931. Voss, E.H., 1994. Flora of Alberta: A manual of flowering plants, conifers, ferns and fern

allies found growing without cultivation in the Province of Alberta, Canada. Second Edition revised by Packer, J.G., University of Toronto Press, Toronto, Ontario, Canada.

Wielgus, R.B., Vernier, P., Schivatcheva, T., 2002. Grizzly bear use of open, closed, and

restricted forestry roads. Can. J. For. Res. 32, 1597-1606. Zager, P., Jonkel, C., Habeck, J., 1983. Logging and wildlife influence on grizzly bear

habitat in northwestern Montana. Int. Conf. Bear Res. and Manage. 5, 124-132.

67

7.0 GRIZZLY BEAR FOOTHILLS HABITAT FRAGMENTATION BY SEISMIC CUTLINES: PRELIMINARY REPORT ABOUT ‘PARSIMONY OF LANDSCAPE

METRICS’ AND ‘CUTLINE EFFECTS ON HABITAT STRUCTURE AND FOOTHILLS GRIZZLY BEAR LANSCAPE USE’

Julia Linke, Department of Geography, University of Calgary, Canada [email protected] Steven E. Franklin, Department of Geography, University of Calgary, Canada.

7.1 Introduction

The Rocky Mountain Foothills of Alberta provide habitat to the grizzly bear (Ursus arctos horribilis) and other wildlife. However, this area includes considerable mining, seismic oil and gas exploration, and forest harvesting activities, which inherently influence the configuration and composition of the natural landscape. For instance, in the process of conventional oil and gas exploration, a dense network of 5-10 m wide seismic cutlines is created (Linke, 2002). These cutlines dissect contiguous landscape components, or landscape patches. Other human activities, such as forest harvesting, shape the landscape structure in direct ways, but also in indirect ways, for instance through the addition of roads (McGarical et al. 2001, Reed et al. 1996). Attempts to manage the foothills grizzly bear population within this dynamic, multi-use landscape requires an increased understanding of their landscape use and the influences that structure may have on such use.

Relationships between spatial patterns and wildlife habitat processes have received much attention in the disciplines of environmental management, conservation biology and landscape ecology over the last two decades (Wiens, 1989, Levin, 1992, Diaz, 1996; Davidson, 1998). In order to quantify spatial patterns into single variables, a large list of landscape metrics and indices have been developed (e.g. McGarical et al. 2002, McGarical and Marks 1995, Frohn, 1998). Several studies have successfully related them to habitat use and habitat selection of several wildlife species (for example Stuart-Smith et al. 1997, Chapin, et al. 1998, Knutsen et al 1999, Potvin et al. 2001).

The purpose of this Masters of Science project is to assess and quantify the effects of seismic lines on the landscape structure of grizzly bear habitat. Secondly, the relationship between landscape structure and grizzly bear landscape use is to be investigated. None of these topics have been studied previously.

It is the above outlined context within which several detailed questions arise for this research: 1) Which of the various existing landscape metrics describe best the spatial pattern of the foothills grizzly bear habitat?, 2) What are the quantitative effects of seismic cutlines on the foothills grizzly bear habitat? and 3) What are the effects of seismic cutlines and spatial patterns on grizzly bear landscape use? The following report on progress on these topics will be divided according to these three questions.

68


7.2 Parsimony Of Landscape Metrics In The Foothills

More than 50 class-level and landscape level landscape metrics exists. Some of these metrics are redundant and the landscape structure may be characterized by a few groups of metrics, as identified by earlier studies (McGarical and McComb 1995, Riiters et al. 1995). On the search for a parsimonious suite of independent landscape metrics across grizzly bear habitat classes, we undertook a parsimony analysis to identify important (according to three criteria being universality, consistency and strength) metric groups adhering to methods described in Cushman et al. (in review) based on principal component and clustering analysis. For ease of later comparison of results, we used the same original list of 49 class-level and 54 landscape-level metrics as described by Cushmann et al (in review) (Appendix 1).

7.2.1 Methods

In order to capture the variable spatial patterns within the foothills, we subdivided this study area into 195 non-overlapping 5.4 x 5.4 km sub-landscapes, the size of which was driven by the extent of the study area and the minimum sample size requirement of providing approximately 3-to-1 sample to metric ratio for the following principle component analysis (McGarical et al 2000) (Figure 7.1). The idt habitat map was reclassed into a more generalized classification scheme (Appendix 2, Figure 7.1) to ensure adequate representation of all classes in most sampled sub-landscapes. The following sections summarize the methods and the preliminary findings from this analysis.

7.2.1.1 Class-level Parsimony Analysis: For each of the eight generalized landcover types, we undertook a partial principal component analysis on the correlation matrix of the remaining 49 metrics using the function PROC FACTOR in SAS (SAS, 2002), partialing out "Pland" to remove the effects of landscape composition on the resulting principal components of configuration metrics (Cushman et al in review). The criteria to assess the number of significant principal components we used was the latent root criterion (McGarical et al 2000), identifying significant components with eigenvalues of more than one in cases when the correlation matrix is used in the eigenanalysis. In order to identify principal components that were similar across classes, we applied polythetic agglomerative hierarchical clustering technique with average linkage using the PROC CLUSTER function in SAS (SAS 2002). Similarity of principal components was assessed by comparing their respective component loadings (Cushman, et al in review). This was done by initially creating a distance matrix based on the computation of a Pearson's correlation matrix of all principal components (n=75) across all classes We transformed this correlation matrix to a distance matrix by subtracting each |r| from 1 to create distance values ranging from 0, indicating perfect correlation, to 1, indicating no correlation between the loadings of individual components. Cluster membership was based on the point of inflection of the scree-plot of fusion distances (McGarical et al 2000). This cut-off occurred at a fusion distance of 0.69 resulting in 16 clusters of components.

69

Cushman et al (in review) presented three measures to assess importance of principal components, or metric gradients, which were universality, consistency and strength. Universality was measured as the percentage of classes a metric gradient existed in, and it indicates how globally present a metric gradient is across classes. Metric gradients are considered universal with equal and more than 75% universality. Consistency was calculated as the average within-cluster Pearson's correlation among the loadings of the constituting components.

generalized idt habitat map 1999Closed ForestOpen ForestHerbaceousShrubBogsNon-forested featuresLinear Features (Roads/Pipelines)Disturbed Sites (Cuts and Burns)

5.4km x 5.4 km sub-landscapes

N

6 0 6 12 Kilometers

Figure 7.1. Stratification of the generalized idt foothills of the extended grizzly bear research project boundary for purpose of the parsimony analysis.

It indicates the degree of stability of metric behaviour within a particular gradient, and a consistent gradient is one that is oriented in the nearly the same direction by the constituting metrics (Cushman et al in review). The threshold for highly consistent metric gradient clusters was determined by the average pair-wise correlation across classes. Strength of metric gradients was measured in two ways: the mean eigenvalue, and the mean explained variance across classes. Gradients explaining on average more than 10% of the variance were deemed as particularly strong (Cushman et al in review). We reported all three of these measures, however, universality and consistency are used as main indicators of metric importance.

70

When interpreting metric gradients, metrics were considered with loading higher than 0.4 or less than -0.4 on a given component, and we placed most emphasis on those with loadings greater than 0.60 or less than -0.6, which is in accordance with Rule 1 for determining significance of principal component loadings (McGarical et al 2000). The metrics describing a particular metric gradient were listed with their respective relationships indicated by positive and negative signs. The parsimonious set of independent metrics across all classes is here defined by universal, strong and consistent metric gradients, and only one metric indicative of each gradient is required to capture the observed landscape pattern, since the others in the group would be highly redundant. This set can form the minimum structure attributes necessary to describe structure patterns within the particular landscape under study (Cushman et al in review). A second group of other consistent gradients, which behave class-specific and which cannot be generalized as indicated by low universality scores, is also considered to capture the full suite of dimensions in the study area. These low universality, or unique, gradients should be considered as important metrics for the specific class it represents. A third group of metrics gradients also exists, and these are low consistency metric gradients. These low consistency metric gradients need to be considered in the context that they have variable behaviour with the other metrics within the same group. At this stage, these low consistency metrics are recognized as spatial attributes independent from the other existing metrics.

7.2.1.2 Landscape-Level Parsimony: In order to capture most of the variability in the landscape pattern of the 54 landscape level metrics with the fewest independent suites of configuration landscape metrics, we conducted a principal component analysis on the correlation matrix of the metrics using the function PROC FACTOR in SAS (SAS, 2002). The number of significant principal components, or metric gradients, was again assessed with the latent root criterion (McGarical et al 2000). The resulting principal components constitute an independent set of metric gradients that explain part of the landscape pattern variability. Within each component or metric gradient, the major metrics describing the component were interpreted the same way as mentioned in the class-level analysis. The parsimonious set of independent metrics is here defined by each of the components together explaining most of the variance of the spatial patterns. Metrics indicative of the same gradient are highly redundant, and only one is necessary to describe the existing landscape pattern.

7.2.2 Results and Recommendations:

7.2.2.1 Class-level Parsimony Analysis: With the partial principal component analysis we retained 8-11 components per generalized landcover type, each explaining between 86-88% of the variance of the 49 class-level metrics. Across all eight landcover types, we found five metric gradients that were universal, highly consistent, and variably strong (Table 7.1). These gradients were "Large Patch Dominance ", "Nearest Neighbour", "Edge Contrast", " Aggregation and Perimeter-Area-Ratio", and "Edge/Patch Density". These five metrics gradients together explain on average nearly 70% of the variation in the spatial pattern.

71

This set is recommended as a minimum set of independent class-level landscape metrics to be used in any spatial pattern research project related to this particular study area. Any one metric within each metric gradient will suffice to characterize the spatial patterns since they are highly redundant. The choice of particular metrics ought to be guided by ease of interpretability and specific research question. The linear features landcover type seems to be mainly responsible for reducing the universality within this nearly universal and consistent group of metrics (missing in gradients 3-5, see Table 7.1), indicating some unique spatial patterns within this cover type. This point is emphasized by two particularly strong but unique metric gradients, which exclusively occur within the linear features cover type, "Splitting and Cohesion", and "Core area complexity" (gradients 9 and 10, Table 7.1). These two metric gradients are part of five consistent but more unique, as indicated by low universality scores, gradients (gradients 6-10, Table 7.1). These gradients are independent metric groups to be added to the previously mentioned minimum set to describe the spatial patterns of this study area; however, only in regards to the specific cover types it occurs. The metric gradient "Patch Size Variability" was a highly universal but low consistency gradient (Gradient 11, Table 7.1). This low consistency metric group included five other metric gradients. This third set of metrics, the low consistency metric gradients (gradients 11-16, Table 7.1), explains the last portion of the total spatial patterns existing in the study area; however, since the individual gradients are not stable in their behaviour, these metrics are only pointed out to be independent from the other groups. Within this group, metric gradients cannot be considered as multivariate gradients and therefore, do not indicate high redundancy metrics. At this stage, we recommend to interpret these low consistency metric gradients only as a list of "other metrics", which potentially represent separate aspects of landscape structure, since they do not covary with the other, consistent metric gradients.

In summary, there are five universal and consistent metric gradients, which

together, explain on average nearly 70% of nearly all class-level spatial patterns (gradients 1-5, Table 7.1,). Beyond this basic parsimonious set of metric groups, five unique, but consistent metric gradients exist (gradient 6-10, Table 7.1). These unique gradients should also be considered in regards to the specific cover type they occur in, when attempting to capture the full suite of spatial variability within this landscape with the fewest amount of metrics. Within these 10 metric gradients, one metric per gradient can be used to summarize each pattern. This suggests that the 49 class-level metrics can be reduced to 10 metrics to describe the spatial patterns of the foothills generalized idt land cover. However, the 18 metrics representing the six low consistency metric gradients can only be used as a list of other potential metrics.

72

Table 7.1. Principal component clusters (metric gradients), identified through partial principal component and clustering analysis on 49 class-level configuration metrics across 8 generalized idt landcover types, with their respective universality, consistency and strength. Meaning of each cluster is provided by the metric gradient name, listing also the largest positive and negative loadings. Consistency was assessed against the average of all within cluster correlations across all classes being 0.74. Universal and consistent metric gradients are highlighted forming the minimum set of class-level metric gradients to be used within this study area. Unique metric gradients are italicized, and low consistency metric gradients appear in normal font.

Metric gradient

Contributing Metrics Cover Type Membershi

p*

Univer-sality (%)

Consistency (mean within-

gradient correlation)

Mean Eigenvalu

e

Mean explained Variance

(%) 1.Large Patch Dominance

LPI+,Area_AW+, Core_AW+,

Dcore_AW+,Prox_Mn

all 100 0.79 11.59 22.7

2.Nearest Neighbour

Enn_Mn+, Enn_Aw+

all 100 0.75 1.41 2.7

3.Edge Contrast

TECI+, Econ_Mn+, Econ_Aw+

1,2,3,4,5,6,8

87.5 0.79 3.14 6.17

4.Aggregation and Perimeter-Area-Ratio

Pladj+, Para_Aw+, Cohesion+, AI+,

Para_Cv+, Clumpy+

2,3,4,5,6,8 75.0 0.76 11.87 23.28

5.Edge/ Patch Density

PD+, ED+, LSI+, CWED+, DCAD+

2,3,4,5,6,8 75 0.81 4.69 9.20

6.Patch Shape Complexity

Shape_Mn+, Frac_Mn+, Frac_Cv+, Shape_Cv+, Frac_Aw+,

Gyrate_Mn+, Shape_Aw+

3,4,5,6,8 62.5 0.78 3.54 6.96

7.Proximity and Neighbour-hood Similarity

Prox_Aw+, Simi_Aw+, Prox_Mn+, Simi_Mn+

6,7,8 37.5 0.90 2.18 4.28

8.Core area index

CAI_Mn+, CAI_Aw+, CAI_Cv+

3,4,6 37.5 0.84 2.57 5.04

9. Split and cohesion

Split-, Cohesion+, Econ_Aw+, TECI+

7 12.5 n/a 8.25 16.18

10. Core area

Core_Mn+, Dcore_Cv+,

7 12.5 n/a 6.24 12.24

73

Metric gradient

Contributing Metrics Cover Type Membershi

p*

Univer-sality (%)

Consistency (mean within-

gradient correlation)

Mean Eigenvalu

e

Mean explained Variance

(%) complexity Dcore_Aw+,

CAI_Aw+, Core_Aw+, Dcore_Mn+

11.Patch Size Variability

Core_Cv+, Area_Cv+, Dcore_Cv+

all 100 0.57 2.41 4.73

12. Shape and correlation length of large patches

Shape_Aw+, Frac_Aw+, Gyrate_Aw

1,5,6,7 50 0.69 4.67 9.16

13. Edge contrast of variation

Econ_Cv+ 1,3,5,7 50 0.67 1.25 2.45

14. Proximity index of variation

Prox_Cv+ 1,2,6,8 50 0.62 1.56 0.03

15. Aggregation and Edge

Ai+, Clumpy+, Para_Aw-, Pladj+, CWED-, ED-, LSI-

1,2,7,8 50 0.60 5.66 11.11

16. Mean Patch Size

Area_Mn+, Gyrate_Mn+, Shape_Mn+

1,7 25 0.71 6.95 13.6

* Landcover type number explained in Appendix 2. 7.2.2.2 Landscape-level Parsimony Analysis: Six principal components were retained which overall explained 88% of the variance in spatial structure attributes of 54 landscape-level metrics (Table 7.2). To best describe the spatial pattern of this habitat class, only one constituting metric of each metric gradient is necessary for subsequent analysis, since they are highly redundant. This analysis suggests that the 54 landscape-level metrics can be reduced to six metrics while retaining most of the information, to be chosen among each metric gradient according to study objectives and metric behaviour.

74

Table 7.2. Six significant principal components, or metric gradients of 54 landscape-level configuration metrics from the generalized idt land cover, as retained from partial principal component analysis. Components are ordered by their explanatory power (percent explained variance). Meaning of each component is provided by the metric gradient name, listing also the largest positive and negative loadings. Metric Gradient/Principal Component

Contributing Metrics Cumulative Explained Variance (%)

Patch size/density/edge density/ nearest neighbour

Area_MN+, Core_Mn+, PD-, ENN_MN+, ED-,

AI+, CWED-, Contagion-

53

Patch Variability/Large Patch Dominance

Area_CV+, Dcore_CV, Core_CV+, LPI+, Division-

Area_AW+, TECI-, ECON_MN-

68

Patch shape FRAC_MN+, PARA_Mn-, Shape_MN+

79

Large Patch Shape/Patch Dispersion

Shape_AW+, FRAC_AW+, IJI-, ENN_CV+

83

Edge contrast/ patch richness

PRD+, ECON_AW+, TECI+, SIMI_CV+

86

Area-weighted proximity Prox_AW 88

7.3 Effects Of Seismic Cutlines On Foothills Grizzly Bear Habitat

Using high resolution Indian Remote Sensing (IRS) satellite imagery, seismic lines were mapped inside the 5000 km2 upper Foothills Model Forest Grizzly Bear Research Program study area. The objective was to identify their impacts on landscape structure. Landscape-level landscape metrics were to be used as the indicators. Modeling how the landscape metrics respond to increasing seismic cutline densities is considered as a crucial component for the development of management guidelines to conserve grizzly bears in a natural landscape in transition. We ask two specific questions: 1) In what way are landscape metrics affected by adding seismic cutlines at various densities (are these of linear, exponential, or other relationships)? and 2) are these relationships different across different hexagons (will landscapes with different original fragmentation level respond differently to the addition of seismic cutlines).

7.3.1 Methods

For this preliminary assessment of seismic cutline impacts on the grizzly bear habitat, we stratified the upper foothills landscape into smaller sub-units (Figure 7.2). We used hexagons for the shape of these sub-units, and the size was determined using the mean clustering distance for the seven female and male collared grizzly bears roaming completely within the foothills portion during the early summer (June1-July31) of 1999 and 2000 of the original grizzly bear research project boundary.

75

We used Ripley's K clustering analysis (Ripley, 1976) on each of the individual GPS location data sets, and then computed the number-of-location-weighted mean of these distances (5.9 km, 2300 ha large hexagons). As the basis for the landscape structure assessment, we used the idta landcover map (Franklin et al, 2001), generalized into 16 grizzly bear habitat types (GBH) (as first implemented by Popplewell, 2002; Appendix 3). We resampled the 30m resolution GBH map to 5m, in order to facilitate merging the 5m resolution seismic cutline map layer. We used Fragstats (McGarical et al. 2002) to compute several configuration landscape-level metrics on the 104 2300 ha large hexagons of the GBH map. In order to assess the impacts of seismic cutlines, we once computed these metrics on the GBH map with, and once without the seismic cutlines. The actual effect of seismic cutlines was quantified by computing the percent change in metric per hexagon when adding seismic cutlines. Pooling this information across all 104 hexagons as a function of their actual seismic cutline densities indicated the relationships (e.g. linear, power function, exponential, etc.) between landscape configuration and seismic cutlines. The actual slope, or rate of change, in these relationships was also investigated in regards to the original fragmentation level of each hexagon., by plotting the rate of change in metrics due to additional seismic cutlines in response to the original value of the specific landscape metric under investigation. Here in this report, we only report on a few common landscape metrics, such as number of patches, edge density and mean patch size.

7.3.2 Results

7.3.2.1 Effects of cutlines on landscape metrics We found linear increases in the two metrics, number of patches (R2=0.70) and edge density (R2=0.69) as a function of adding seismic cutlines to the 104 hexagons (Figure 7.3). Mean patch size was reduced when adding seismic cutlines, but in a power function (R2=0.41), indicating relatively lower impacts onto this metric at large cutline densities. These relationships emphasize the direct relationships that additional seismic cutlines can have on the landscape structure of the grizzly bears. Knowing under which levels of landscape configurations grizzly bears can persist, these relationships will become very useful for judging the maximum tolerable level of additional seismic cutlines within their habitat.

76

GBH MapClosed ConiferClosed DeciduousMixed ForestOpen ConiferOpen DeciduousAlpineHerbaceousShrubWet/RiparianNon-ForestedRoad/RailCut 0-2Cut 3-12BurnNo Data

2300 ha hexagons

N

20 0 20 40 Kilometers

Figure 7.2. Stratification of foothills portion of original grizzly bear research project

boundary into 2300ha large hexagons to assess seismic cutline effects on landscape metrics.

Figure 7.3. Effects of adding seismic cutlines at different densities on three landscape-

level metrics across 104 2300 ha large hexagons of the foothills GBH map.

77

7.3.2.2 Original landscape fragmentation specific effects of cutlines on landscape metrics?

When investigating the relationships between the actual rates of change in metrics in response to additional seismic cutlines across the 104 hexagons, we found a generally decreasing rate of change (in a power function) with increasing original fragmentation level of the landscape (Figure 7.4 and 7.5).

Figure 7.4. Rate of change in edge density in response to adding seismic cutlines as a

function of original landscape fragmentation.

Figure 7.5. Rate of change in number of patches in response to adding seismic cutlines

as a function of original landscape fragmentation.

78

The rate of change in landscapes with low initial edge density (Fig.7.4), number of patches (Fig.7.5), or mean patch size (not shown, but similar to Fig.7.5), is much higher than in landscapes with high values in these metrics. This indicates that seismic cutlines will impact the landscape structure much more severe in areas where natural fragmentation is very low, compared to areas with already high natural fragmentation.

Edge density demonstrates a very clear relationship (R2=0.99) (Fig.7.4), while the number of patch metric (Fig.7.5) and mean patch size metric (not shown) have less tight fits (R2=0.42, and R2=0.41 respectively). Additional investigation showed that number of patch and mean patch size metrics are affected by positioning of the seismic cutlines when dealing with very low cutline densities, therefore weakening these relationships. For example, a single small cutline might be positioned in the middle of a large contiguous patch within a low fragmentation landscape. This cutline will therefore not dissect this patch into two entities, but rather the number of patches remains the same as before, and there is no effect on the overall landscape structure, weakening the previously shown general relationship of decreasing effects on landscapes with increasing original fragmentation. These extremely low cutline density cases could be discarded in the future to strengthen these relationships.

7.4 Effects Of Seismic Cutlines And Spatial Patterns On Foothills Grizzly Bear Landscape Use

We used a landscape ecology approach to relate landscape structure with population-level grizzly bear use of the multi-use foothills landscape during early summer over the years 1999 and 2000. The investigated grizzly bear population consisted of 5 female and 2 male bears, which were GPS-collared in the spring soon after emergence from dens. The area available to this population was stratified into 49 km2 hexagon-shaped landscape units (Fig.7.6). The scale of this stratification was determined by the population movement behavior (as indicated by population-level clustering distances using Ripley's K analysis) (Fig.7.6). Ten uncorrelated compositional (Appendix 3) and four configuration landscape metrics (Fig.7.6) were calculated within each landscape unit, and utilization points were pooled or "binned" within each unit (Fig.7.6). Landscape use was related to landscape metrics using a poisson Generalized Linear Model (GLM) (Fig.7.6). The underlying strategy was to find the best fitting model with the fewest number of predictor variables, following the principle of parsimony (Burnham and Anderson, 1998). Final model was selected using stepwise model selection procedures based on the Akaike's Information Criterion (AIC) (Akaike, 1973) with the S-PLUS command stepAIC in the MASS library (Venables and Ripley, 1997).

79

Figure 7.6. Flowchart of methods used during analysis of seismic cutline and landscape

structure effects on grizzly bear population level landscape use. We found declining landscape use associated with increasing proportions of closed forest (Pcforest), of roads (Proads) and in the variation of inter-patch distances (MNN_CV), as indicated by negative coefficients (Table 7.3). These variables together explained nearly 40% of the deviance. The proportion of seismic cutlines was not an important predictor of population level landscape use at this scale. However, further investigating of the effects of seismic cutlines on landscape structure revealed relatively strong relationships (Fig.7.7). Adding seismic cutlines to GBH sub-landscapes (using same scale and methods as in Part 2) appears to cause an increase in the variation of inter-patch distances (R2=0.34) and a decrease in inter-patch distances as such (R2=0.79). This suggests that despite the absence of seismic cutline proportions as a direct predictor of grizzly bear population-level landscape use, seismic cutlines may indeed play an important role for grizzly bears, since they positively affect a landscape metrics, variation in inter-patch distances, which appears to be an important predictor for landscape use. A more detailed version of this analysis can be read in Linke et al (in review).

80

Table 7.3. Explanatory model parameters, showing residual deviance, degrees of freedom, significance and coefficients of model variables, for early summer grizzly bear landscape use at the population level. Model Parameters Population Level Use Deviance: Null Model Final Model

65.19 39.74

Residual degrees of freedom

68

Model Variables: MPS MSI MNN_CV SIEI Pcforest Poforest Pherb Pshrubs Proads Precentcut Poldcut Pburn Pseismic Pstreams

Coefficient.

--- ---

-0.20, ---

-0.50, ---- ----

-0.22 ---- --- --- --- --- ---

Significancea

--- ---

P = 0.021 ---

P < 0.001

P = 0.042 --- --- --- --- --- ---

a based on analysis of deviance F-test

Figure 7.7. Effects of adding seismic cutlines on variation in inter-patch distance metric.

81

7.5 References

Akaike, H. 1973. Information theory and an extension of the maximum likelihood principle. In Petran, B.N. and Csaki ,F. (Eds) International Symposium on Information

Theory. Second edition, Akademiai Kiadi, Budapest, Hungary. Burnham, K. and D.R. Anderson. 1998. Model selection and inference: a practical

information- theory approach. Springer, New York. Chapin, T.G., Harrison, D.J., and D.D. Katnik. 1998. Influence of landscape pattern on

habitat use by American Marten in and industrial forest. Conservation Biology12 (6):1327-1337.

Cushman, S., McGarical, K. and M.C. Neel. In Review. Parsimony in landscape metrics:

strength, universality, and consistency. Landscape Ecology. Davidson, C. 1998. Issues in measuring landscape fragmentation. Wildlife Society

Bulletin 26 (1): 32-37. Diaz, N.M. 1996. Landscape Metrics. A new tool for forest ecologists. Journal of

Forestry 94 (12): 12-16. Franklin, S. E., G. B. Stenhouse, M. J. Hansen, C. C. Popplewell, J. A. Dechka, and D.

R.Peddle. 2001. An integrated decision tree approach (IDTA) to mapping landcover using satellite remote sensing in support of grizzly bear habitat analysis in the Alberta Yellowhead Ecosystem. Canadian Journal of Remote Sensing 27 (6):579-591.

Frohn, R.C. 1998. Remote Sensing for Landscape Ecology. New Metric Indicators for

Monitoring, Modeling, and assessment of Ecosystems. Lewis Publishers, New York. Knutson, M.G., Sauer, J.R., Olsen, D.A., Mossman, M.J., Hemesath, L.M., and M.J.

Lannoo. 1999. Effects of landscape composition and wetland fragmentation on frog and toad abundance and species richness in Iowa and Wisconsin, U.S.A. Conservation Biology 13 (6): 1437-1446.

Levin, S.A. 1992. The problem of patterns and scale in ecology. Ecology 73(6):1943-

1967. Linke, J. 2002. Grizzly Bear Foothills Habitat Fragmentation by Seismic Cutlines

mapped from Indian Remote Sensing (IRS) Imagery. In Stenhouse, G. and R.M. Munro(eds.). 2002. Foothills Model Forest Grizzly Bear Research program 2001 Annual Report 118pp. Linke, J., S.E. Franklin, G.B. Stenhouse, and F. Huettmann. In Review. Grizzly Bear

Population Landscape Use and Structure in Alberta. Ursus.

82

MathSoft Inc. 1999. S-Plus 2000. Professional Release 2. Seattle. McGarigal, K., S. A. Cushman, M. C. Neel, and E. Ene. 2002. FRAGSTATS: Spatial

Pattern Analysis Program for Categorical Maps. Computer software program produced by the authors at the University of Massachusetts, Amherst. Available at the following web site: www.umass.edu/landeco/research/fragstats/fragstats.html

,W.H. Romme, M.Crist, and E. Roworth. 2001. Cumulative effects of roads and

logging on landscape structure in the San Juan Mountains, Colorado (USA). Landscape Ecology 16: 327-349.

,Cushman, S.A. and S. Stafford. 2000. Multivariate Statistics for Wildlife and

Ecology Research. Springer-Verlag, New York. , and B.J. Marks. 1995. FRAGSTATS. Spatial pattern analysis program for

quantifying landscape structure. Gen. Tech. Rep. PNW-GTR-351. Portland, OR: U.S. Department of Agriculture, Forest Service, Pacific Northwest Research Station, 122p.

, and W.C. McComb. 1995. Relationship between landscape structure and

breeding birds in the Oregon Coast Range. Ecological Monographs 65: 235-260. Popplewell, C. 2001. Habitat structure and fragmentation of grizzly bear management units and home ranges in the Alberta Yellowhead Ecosystem. M.Sc. Thesis, unpublished.

University of Calgary, Calgary, Alberta. Potvin F, Lowell K, Fortin M.-J., Belanger L. 2001. How to test habitat selection at the

home range scale: A resampling random windows technique. Ecoscience 8 (3):399- 06. Reed, R.A., Johnson-Barnard, J., and W.L. Baker. 1996. Contribution of roads to forest

fragmentation in the Rocky Mountains. Conservation Biology 10 (4):1098-1106. Ripley, B.D. 1976. The second-order analysis of stationarity processes. Journal of

Applied Problems 13:255-266. Riitters, K.H., O'Neill, R.V., Hunsaker, C.T., Wickham, J.D., Yankee, D.H., Timmins,

S.P., Jones, K.B., and B.L. Jackson. 1995. A factor analysis of landscape pattern and structure metrics. Landscape Ecology 10: 23-40.

SAS. 2002. SAS System. SAS Institute, Cary, North Carolina, USA. Stuart-Smith, A.K., Bradshaw, C.J.A., Boutin, S., Hebert, D.M., and A.B. Rippin. 1997.

Woodland Caribou relative to landscape patterns in Northwest Alberta. Journal of Wildlife Management 61 (3):623-633.

Venables, W.N. and B.D. Ripley. 1994. Modern Applied Statistics with S-Plus. 2nd

Edition. Statistics and Computing, Springer Verlag, New York. Wiens, J.A. 1998. Spatial Scaling in Ecology. Functional Ecology 3: 385-397.

83

84

Appendix 7.6. List of the 49 class-level (C) and 54 landscape-level (L) landscape structure metrics calculated for the analysis (see McGarigal et al. 2002 for a complete description of each metric). PLAND is the covariable in the partial principal components analyses at the class level (after Cushmann et al in review).

Metric Number Level Acronym Name

0 C, L PLAND Proportion of landscape

1 C, L PD Patch density

2 C, L LPI Largest patch index

3 C, L ED Edge density

4 C, L LSI Landscape shape index

5 C, L AREA_MN Mean patch size

6 C, L AREA_AM Area-weighted mean patch size

7 C, L AREA_CV Patch size coefficient of variation

8 C, L GYRATE_MN Mean radius of gyration

9 C, L GYRATE_AM Correlation length

10 C, L GYRATE_CV Radius of gyration coefficient of variation

11 C, L SHAPE_MN Mean shape index

12 C, L SHAPE_AM Area-weighted mean shape index

13 C, L SHAPE_CV Shape index coefficient of variation

14 C, L FRAC_MN Mean fractal dimension

15 C, L FRAC_AM Area-weighted mean fractal dimension

16 C, L FRAC_CV Fractal dimension coefficient of variation

17 C, L PARA_MN Mean perimeter-area ratio

18 C, L PARA_AM Area-weighted mean perimeter-area ratio

19 C, L PARA_CV Perimeter-area ratio coefficient of variation

20 C, L DCAD Disjunct core area density

21 C, L CORE_MN Mean core area

22 C, L CORE_AM Area-weighted mean core area

23 C, L CORE_CV Core area coefficient of variation

85

24 C, L DCORE_MN Mean disjunct core area

25 C, L DCORE_AM Area-weighted mean disjunct core area

26 C, L DCORE_CV Disjunct core area coefficient of variation

27 C, L CAI_MN Mean core area index

28 C, L CAI_AM Area-weighted mean core area index

29 C, L CAI_CV Core area coefficient of variation

30 C, L PROX_MN Mean proximity index

31 C, L PROX_AM Area-weighted mean proximity index

32 C, L PROX_CV Proximity index coefficient of variation

33 C, L SIMI_MN Mean similarity index

34 C, L SIMI_AM Area-weighted mean similarity index

35 C, L SIMI_CV Similarity coefficient of variation

36 C, L ENN_MN Mean nearest neighbor distance

37 C, L ENN_AM Area-weighted mean nearest neighbor distance

38 C, L ENN_CV Nearest neighbor distance coefficient of variation

39 C, L CWED Contrast weighted edge density

40 C, L TECI Total edge contrast index

41 C, L ECON_MN Mean edge contrast

42 C, L ECON_AM Area-weighted mean edge contrast

43 C, L ECON_CV Edge contrast coefficient of variation

44 C, L CLUMPY Clumpiness index

45 C, L PLADJ Proportion of like adjacencies

46 C, L IJI Interspersion/juxtaposition index

47 C, L COHESION Patch cohesion

48 C, L SPLIT Splitting index

49 C, L AI Aggregation index

50 L MESH Mesh size

51 L DIVISION Division index

52 L PRD Patch richness density

53 L SIDI Simpson’s patch diversity

54 L SIEI Simpson’s patch evenness

Appendix 7.7. Generalized reclassification scheme for parsimony analysis.

Original IDTA Landcover Types

Generalized IDT Landcover types

1. Closed Conifer 2. Closed Deciduous 3. Mixed Forest

1. Closed Forest

4. Open Conifer 5. Open Deciduous

2. Open Forest

6. Alpine

not applicable

7. Herbaceous <1800m

3. Grass

8. Shrub < 1800 m

4. Shrubs

9. Wet Open 10. Wet Treed

5. Bogs

11. Rock 12. Snow 13. Shadow 14. Water 15. Urban 16. Wellsites

6. Non-forested Features

17. Pipeline 18. Roads/Rail

7. Linear Features

19. Recent Cut 20. Cut 3-12years 21. Cut >12 years 22. Cut unknown 23. Recent Burn

8. Disturbed Sites

86

Appendix 7.8. Landcover types from the IDTA map, the reclassified GBH map (as used in part 2 of this report), and the final reclassification of landcover types (as used in part 3 of this report). Stars indicate the uncorrelated cover types used in the landscape use analysis.

IDTA Landcover Types GBH Landcover Types Final Reclassification of Landcover Types for Landscape Composition Calculation

Closed Conifer Closed Conifer Closed Forest* Closed Deciduous Closed Deciduous Mixed Forest Mixed Forest

Open Conifer Open Conifer Open Forest* Open Deciduous Open Deciduous

Alpine Alpine Alpine

Herbaceous <1800 m Herbaceous <1800 m Herbaceous*

Shrub < 1800 m Shrub < 1800 m

Shrub and Wetlands*

Wet Open Wetland Wet Treed

Rock Non-Forested Features Non-Forested Features Snow Shadow Water Urban

Pipeline Wellsites and Pipeline Wellsite

Roads/Rail Roads/Rail Roads/Rail*

Recent Cut Recent Cut Recent Cut*

Cut 3-12 yrs Older Cut Older Cut* Cut >12 years Cut unknown age

Recent Burn Recent Burn Recent Burn* Seismic Lines Seismic Lines Seismic Lines

87

8.0 GRAPH THEORY DE VELOPMENT, COST SURFACE ASSESSMENT AND INDIVIDUAL LEVEL CONNECTIVITY ANALYSIS Barbara L. Schwab, Department of Geography, University of Calgary, Canada [email protected] Clarence G. Woudsma, Department of Geography, University of Calgary, Canada [email protected]

8.1 Introduction

8.1.1 Addressing Connectivity

The analysis of habitat fragmentation has become a common method of understanding the environmental impacts of anthropogenic activities. For example, forestry cut-blocks and roads fragment an otherwise contiguous natural environment thereby impacting species, movement patterns and their related habitat. According to Rosenberg et al. (1997) and Beier and Noss (1998), habitat loss, fragmentation, and decreased levels of connectivity are among the most pervasive threats to population viability. As such, connectivity is recognized as a threshold phenomenon (With et al. 1997) in which habitat fragmentation may progress with little effect on a population until critical pathways of connectivity are disrupted; then, a slight change can have dramatic consequences on population persistence. Therefore, the identification of changes in landscape connectivity (structurally and functionally) and the associated response of spatial movement patterns and habitat interactions resulting from landscape change is imperative for the long-term conservation of wildlife populations, specifically grizzly bears. In Alberta, industry resource extraction and exploration continues at an unaltered pace, resulting in associated landscape change, increased habitat fragmentation and decreased levels of connectivity. As such, the main efforts of this research focus on examining the response of functional and structural connectivity to human disturbance and associated landscape change using grizzly bears as a focal species. New approaches to measuring and understanding connectivity based on Graph Theory have been introduced, and while promising, have only recently been tested (Schwab et al. in review).

8.1.2 The Graph Theory Framework

Graph theory is a heuristic methodology which allows researchers to quantify landscape connectivity at multiple temporal and spatial scales. By representing habitat mosaics as a mathematical ‘graph’ (Keitt et al. 1997), the spatial configuration of patches, level of patch importance, connections, and movement (dispersal) matrices can be analyzed. The graph theory framework utilizes the basic elements of nodes or vertices (which for landscape studies typically represent centroids or core areas of habitat patches), edges (connections between patches using least-cost path modeling) and paths (multiple connections between numerous centroids).

88



na

nd

nc

nb

ne

nf

na

nd

nc

nb

eac

a b

Figure 8.1. (a) Classic graph structure with edge eac=nanc connects nodes na and nc; (b) path L (represented by solid black line) within graph G is a sequence of nodes connecting na to ne. As per Harary’s (1969) formal definition of graph theory, a graph G is a set of nodes or vertices (n) and edges (e) such that each edge eac=nanc connects nodes na and nc (Figure 8.1a). In graph G, path L (represented by the solid line in Figure 8.1b) is a specific sequence of vertices or nodes connecting na to ng. The distance of path na to ng is measured by the length of the unique set of edges implicitly defined by the path (Bunn 2000). Using graph theory in a biological context allows for the detailed analysis of how species interact within the landscape. More specifically, it allows one to describe, understand, and potentially predict grizzly bear movement patterns in relation to the surrounding habitat. Additionally, the graph theory approach provides an opportunity to test landscape sensitivity and explore various landscape thresholds as they pertain to grizzly bears. Major data inputs required for the modeling effort include Resource Selection Function (RSF) defined habitat patches (basis for nodes), GIS grid-based landscapes (basis for least-cost path creation), and GPS bear movement data for model validation. Algorithms are further employed to generate connectivity measures.

8.2 Methods

Addressing the question of landscape change due to human disturbance and resulting changes to connectivity requires an integrative modeling approach. The authors adopted and modified the graph theory approach for specific application to grizzly bears. The programs are run using a combination modules linked within ArcInfo to create graph edges and calculate connectivity measures (Bunn et al. 2000). ESRI’s ArcInfo and ArcView GIS software packages provide a working environment that is capable of performing both vector and raster/grid analysis. Research questions will be addressed using RS techniques, GIS applications, and GPS grizzly bear data. Schwab et al. (in review) have previously developed, applied and validated graph theoretic models within GIS to explore habitat connectivity in relation to female grizzly bear movement.

89

Additional efforts are required for further application and validation. Research and analysis to date is summarized in the following three sections: 1) cost surface comparisons, 2) graph generation and 3) detailed graph calculations.

8.2.1 Cost Surface Comparisons

The methodological approach began with the creation and comparison of four primary cost surfaces based on 1) linkage zone models (LZM), 2) subjective weighting (SCS), 3) resource selection function coefficients (RSF), and 4) a homogeneous landscape surface. Each individual cost surface represents an independent base layer or friction surface where the variation in friction is related to the variation in land cover, terrain, human influence as well as other factors (Figure 8.2). For example, suitable grizzly bear habitat and terrain has lower friction values allowing for movement, while human influence elements are represented with higher friction values thus restricting movement. Cost surfaces were developed at the annual level for initial performance comparison.

Linkage Zone Model Subjective Cost Surface Model

RSF Cost Surface Model Homogeneous Surface Model

Figure 8.2. Example of cost surfaces used in comparison and least-cost path validation. To test the utility of cost surfaces for modeling movement, the four different permeability surfaces were evaluated using Least-Cost Path (LCP) modeling (Walker and Craighead 1997, Purves and Doering 1999). Least cost paths between defined GPS start and end locations were generated with validation evaluated using comparisons between the least-cost path generated and withheld female 2001 GPS telemetry data.

A path segment was defined as a sequence of consecutive GPS points characterized with a start and end location. All remaining ‘middle’ points were withheld for validation. 90

See Schwab et al. (in review) for specific details. Each cost surface was used to generate a total of 211 least-cost path segments between identified start and end points. Euclidean distance (m) from the withheld GPS location to the generated path was used to validate how well the path represented actual movement.

Females with cubs-of-the-year (COYS) were excluded from the analysis due to differing movement patterns. The purpose of the validation is to determine which cost surface ‘best’ represents female grizzly perceived potential movement in the study area. Although the LCP approach does not represent actual movement on the landscape, it does provide an appropriate estimation for general movement and functional distances between habitat patches versus straight-line or linear connections as per Amstrup et al. (2000). The cost surface model that performed ‘best’ in model validation was used to further define ‘edges’ or linkages between identified habitat patches within the graph theory modeling procedure.

8.2.2 Graph Generation

The basis of graph construction is the creation of nodes derived from the selection of habitat patches, where each centroid or node represents the center or core area of a habitat patch (Figure 8.3a).

a b

Figure 8.3. (a) Nodes (centroids) delineated from RSF habitat patches defined by 95% kernel home range; (b) least-cost path edges created based on cost surface modeling to represent functional connections between patches.

To delineate habitat patches, a resource selection function (RSF) model was generated using 7,054 female grizzly bear GPS telemetry locations from 1999 and 2000, and 46,981 random locations. The resulting RSF surface depicted the relative probability of occurrence for female grizzly bears at any given grid location in the study area. To further identify habitat nodes, we used all grid locations with a relative probability of occurrence > +1.5 SD from the mean probability of occurrence.

91

Thus, nodes represented areas of high female grizzly bear habitat potential. Small isolated habitat patches were unlikely to be used by bears and therefore all nodes less than 2.5 hectares were eliminated from the procedure. Centroids or nodes were then extracted to represent the center of each identified habitat patch and used in further graph generation. The second element in graph construction is the creation of least-cost path edges that represent connections between patches or nodes (Figure 8.3b). Edges are expressed as a distance matrix D, whose elements dij are the functional distances between patches i and j. Edge creation is carried out using C programming created to duplicate the costdistance and costpath functions within ArcInfo which are the basic functions of LCP modeling. The LCP approach approximates the actual distance covered by each female grizzly as it moves from one patch to the next in a heterogeneous landscape based on the alternative cost surfaces previously described.

8.2.3 Detailed Graph Calculations

Once graphs are generated, statistical and spatial indices are further calculated based on defined daily movement or dispersal rates. These algorithms are currently under review and require further testing before being employed to generate final connectivity values. To date, simple connectivity indices such as gamma and alpha have been employed (Schwab et al. in review). Yet, more detailed analysis based on the probability of dispersal and patch importance is important to understanding connectivity relative to grizzly bears. The results of this research can be expected spring 2003. Further description of specific connectivity measures and graph theory applications are discussed in detail within Bunn et al. (2000) and Urban and Keitt (2001).

8.3 Results

8.3.1 Cost Surface (LCP) Validation

Cost surface validation was conducted by calculating the distance each of 372 withheld GPS data locations were from their respective path using GIS techniques (Figure 8.4). This procedure involved intersection withheld GPS data with raster distance surfaces based on modeled paths. Figure 8.4 depicts contrasting examples of the modeled paths. It is important to recognize that there is considerable room for discrepancy between the actual GPS points and the generated path given that what happens during the 4 hour period is unknown. The model assumes that the bear is likely moving from the start to the end point but as the example on the right depicts, the route followed may be convoluted and depends on many unknowns such as bear activity and state. Results in Table 8.1 indicate a range in mean distance from GPS points to generated paths of 160 to 700 meters. High distances suggest that the LCP procedure is more likely to generate corridors for movement rather than specific movement paths. However, given the large size of individual home ranges and study area, in addition to data accuracy and scale, the results appear acceptable.

92

Figure 8.4. Least-cost path output of four cost surface models demonstrating validation technique, distances of withheld GPS data compared to paths modeled were extracted using GIS techniques.

Statistical comparison of mean distances for each cost surface was accomplished with a single factor analysis of variance (ANOVA) using LOG transformation of the individual records. Results indicated that overall, the resource selection function (RSF) cost surface performed better than the other three based on the mean distance between validation point and generated path (Table 8.1). Seven of the 10 bears used in LCP modeling had consistently lower mean distances with the RSF cost surface. However, statistical significance (P = 0.031) was only demonstrated with female G028. Refer to Schwab et al. (in review) for more detailed explanation. Overall, it was decided to use the RSF cost surface for further graph generation based on its consistent lowest mean distance.

93

Table 8.1: Least-cost path validation results showing mean distances (m) from GPS data to modeled paths broken down by cost surface model, showing single factor analysis of variance (ANOVA) based on LOG transformation of means.

Mean Distance (m) of Points to Path

Bear ID Number of Paths

Total GPS

Points LZM SCS RSF Homogeneous Signif. G002 8 14 348.45 288.49 274.58 335.61 P = 0.641G003 54 108 300.57 332.45 295.28 315.47 P = 0.409G016 36 56 234.68 212.58 182.43 274.01 P = 0.373G027 1 1 379.01 442.3 458.31 467.68 --- G028 20 32 238.75 213.1 160.33 340.07 P = 0.031G036 20 39 321.68 247.73 228.28 292.68 P = 0.460G038 16 18 292.35 266.05 241.33 319.69 P = 0.374G040 9 8 833 668.29 494.18 760.45 P = 0.400G042 4 4 446.33 449.03 224.95 208.44 P = 0.391G100 43 92 480.33 478.88 503.46 472.64 P = 0.527All Bears 211 372 387.51 359.89 306.31 378.67 ---

8.3.2 Graph Results

Using the RSF cost surface (based on performance of validation comparison) as a surrogate landscape for animal movement, individual connectivity for 3 female bears was modeled over a 2 - year period to explore changes in simple connectivity levels. Daily movement distances were calculated for each female bear by summing the annual linear distances (km) between consecutive GPS points and dividing by total days (Table 8.2). This distance dictates the mean distance traveled by each bear on a daily basis. Therefore, only paths less than or equal to this distance are retained for edge or least-cost path connections between habitat patches. The resulting graph is further analyzed to quantify levels of habitat connectivity within each kernel home range.

94

Table 8.2. Identification, GPS sample sizes by year, 95% kernel home range sizes, and mean daily movement rates used in graph theory analysis.

1999 2000 1999 / 2000

Bear ID GPS

Locations

Home Range (km2)

Movement Rate

(km/day) GPS

Locations

Home Range (km2)

Movement Rate

(km/day)

Mean Movement Rate

(km/day) G004 794 150.722 4.267 321 177.894 3.912 4.0895 G016 696 79.881 4.195 296 42.469 1.725 2.96 G020 684 255.911 5.554 448 523.928 2.425 3.9895

All Bears 2174 486.514 4.672 1065 744.291 2.687 3.368

Individual habitat connectivity was modeled for G004, G016 and G020 for 1999 and 2000. Levels of connectivity were assessed using gamma (γ) and beta (β) indices (Table 8.3). The gamma index calculates the number of edges to the maximum possible number of edges in each graph with values ranging from 0 - 1. The beta index is simply the number of edges connected to the graph nodes or habitat patches and thus, the ratio ranges from 0 and greater.

Table 8.3. Graph analysis results showing number of nodes (RSF habitat patches) used, number of edges created using daily movement rate by female by year, and connectivity measures.

Connectivity

Measures Bear ID Year

Number of Nodes

Number of Edges Gamma (γ)

Beta (β)

Exposure to Human Use

G004 1999 66 261 0.122 3.955 Medium 2000 72 274 0.107 3.806 Medium

35 206 0.346 5.886 2000 12 6 0.09 0.5 Low G020 1999 125 1146 0.148 9.168 High 2000 232 625 0.023 2.694 High

G016 1999 Low

Results demonstrated varying connectivity levels between bear due to variation in home range size, the resulting number of habitat patches and the daily movement rate. Both gamma and beta indices indicated higher connectivity levels for all bears in 1999 when compared to 2000 (Table 8.3). Overall, female G016 demonstrated highest levels of habitat connectivity across all bears for 1999. In contrast, female G020 demonstrated lowest levels of habitat connectivity across all bears for both 1999 and 2000.

95

a b

Figure 8.5. Graphs generated for females G004 (a) and G020 (b) showing variation in

2000 connectivity levels between mountain and foothills bears respectively, connection distances are defined by daily movement rates.

Connectivity levels were found to remain consistent for G004 across both years. Results were largely dependant on daily movement rates employed to define graph connections (Table 8.2).

Levels of habitat connectivity were also found to correspond with exposure to human use (Table 8.3). Figure 8.5 depicts differing levels of graph connectivity for mountain and foothills bears respectively. Highest levels of connectivity were found in home ranges where human disturbance and landscape fragmentation was limited. Decreased levels of habitat connectivity, visually demonstrated by the appearance of subgraphs (Figure 8.5b), were found in home ranges were human disturbance and resulting landscape fragmentation had increased. Overall, we found the graph theory approach an effective method for examining and quantifying habitat connectivity at the level of the individual female home range. Further research of the graph theory model presented here, may provide new methods for examining connectivity levels and movement patterns associated with grizzly bear populations at the landscape level.

96

8.4 Discussion

Model developments for landscape studies increasingly are combining the functional capabilities of RS, GIS and GPS (Beier 1993, Shumaker 1996, Anderson and Danielson 1997, Walker and Craighead 1997, Purves and Doering 1999, Bunn et al. 2000, Roberts et al. 2000). Combining the graph theoretic modeling approach with the spatial analysis abilities of geographic information systems (GIS) and acquired global positioning system (GPS) grizzly bear movement data allows for the delineation of spatial movement patterns in relation to remote sensing (RS) classified landscape habitats. Understanding grizzly bear movement and habitat use through GIS-based methods has the potential to assist resource managers with land use decisions related to the conservation of grizzly bears (Dugas and Stenhouse, 1999). The utility of this however, is predicated on a solid foundation of understanding landscape ecology and grizzly bear biology.

Grizzly bear GPS telemetry data in this case provided an opportunity to validate the creation of graph edges based on least-cost path modeling, as well as delineate habitat patches through RSF modeling. Results revealed that while LCP modeling does not represent actual movement on the landscape, it did provide an effective method for retrieving functional movement distances. As such, the results help to explain how grizzly bears perceive the landscape based on movement, habitat type and connectivity patterns.

Currently, connectivity has only been addressed at the habitat level scale. The

analysis revealed variable levels of connectivity for individual bears. However, overall results demonstrated that in areas with increased human presence and disturbance, basic habitat connectivity and levels of graph structure decrease. Connectivity results were largely determined by the daily movement rate employed to define functional linkages between habitat patches. Further comparisons could be conducted using the mean daily movement rate for all bears across 1999 and 2000 for additional graph generation.

Overall, the research can provide land managers with a validated modeling approach / tool for continued analysis of impacts associated with human development on wildlife populations, including an opportunity to incorporate grizzly bear needs into critical planning. Additionally, the proposed Species at Risk Act (SARA) highlights the need to identify and protect critical habitat required by threatened or endangered species. As such, results of this project will provide land managers with transferable methods and products that define both critical habitat and necessary linkages or corridors for animal movement and maintained landscape connectivity. Finally, results will aid in understanding the impact of human development and associated landscape change on structural and functional connectivity, spatially and temporally.

97

8.5 References

Amstrup, S.C., Durner, G.M., Stirling, I., Lunn, N.J. and F. Messier. 2000. Movements and distribution of polar bears in the Beaufort Sea. Canadian Journal of Zoology 78: 948-966.

Anderson, G.S. and B.J. Danielson. 1997. The effects of landscape composition and

physiognomy on metapopulation size: the role of corridors. Landscape Ecology 12: 261-271.

Beier, P. 1993. Determining Minimum Habitat Areas and Habitat Corridors for Cougars.

Conservation Biology 7(1): 94-108. Beier, P. and R.F. Noss. 1998. Do Habitat Corridors Provide Connectivity? Conservation

Biology 12(6): 1241-1252. Bunn, A.G., Urban D.L. and T.H. Keitt. 2000. Landscape connectivity: A conservation

application of graph theory. Journal of Environmental Management 59: 265-278. Dugas, J. and G.B. Stenhouse. 1999. Grizzly Bear Management: Validating Existing

Cumulative Effects Models. Thirteenth Annual Conference on Geographic Information Systems. Vancouver. pp. 157-160.

Harary, F. 1969. Graph Theory. Addison-Wesley, Massachusetts.

Keitt, T., Urban, D.L. and B.T. Milne. 1997. Detecting Critical Scales in Fragmented Landscapes. Conservation Ecology online: http://www.consecol.org/vol1/iss1/art4.

Purves, H. and C. Doering. 1999. Wolves and People: Assessing cumulative impacts of

human disturbance on wolves in Jasper National Park. 1999 ESRI Users Conference; published online at http://www.esri.com/library/userconf/proc99/proceed/papers/pap317/p317.htm

Roberts, S.A., Hall, G.B. and P.H. Calamai. 2000. Analysing forest fragmentation using

spatial autocorrelation, graphs and GIS. International Journal of Geographical Information Science 14(2): 185-204.

Rosenberg, D.K., Noon, B.R. and E.C. Meslow. 1997. Biological Corridors: Form,

Function and Efficacy. BioScience 47(10): 677-687.

Shumaker, N.H. 1996. Using Landscape Indices to Predict Habitat Connectivity. Ecology 77(4): 1210-1225.

Schwab, B.L., Woudsma, C.G., Stenhouse, G.B., Franklin, S.E. and S.E. Nielson. ****.

Connections That Matter: A Graph Theoretic Analysis of Grizzly Bear Movement in the Yellowhead Ecosystem, Alberta, Canada, Ursus, in review.

Urban, D. L. and T. H. Keitt. 2001. Landscape connectedness: a graph theoretic

perspective. Ecology, 82:1205-1218. 98

http://www.consecol.org/vol1/iss1/art4

http://www.esri.com/library/userconf/proc99/proceed/papers/pap317/p317.htm

Walker, R. and L. Craighead. 1997. Analyzing Wildlife Movement Corridors in Montana

Using GIS. 1997 ESRI Users Conference; published online at http://www.grizzlybear.org/col/lcpcor.htm

With, K.A., Gardner, R.H. and M.G. Turner. 1997. Landscape connectivity and

population distributions in heterogeneous landscapes. OIKOS 78(1): 151-169.

99

http://www.grizzlybear.org/col/lcpcor.htm

9.0 A COMPARISON OF MAPPING PRODUCTS FOR PREDICTING GRIZZLY BEAR HABITAT QUALITY IN WEST-CENTRAL ALBERTA Scott E. Nielsen, Department of Biological Sciences, University of Alberta, Edmonton, AB T6G 2E9, Canada, email: [email protected]

9.1 Introduction and Objectives

Identification of grizzly bear (Ursus arctos L.) habitat is a critical component to the protection, management, and sustainability of existing populations, as well as the potential restoration of future populations (Boyce & Waller, 2003). Currently, however, there is strong divergence in the reliance and use of different mapping products for predicting grizzly bear habitats in North America. To date, research studies and government mapping programmes have relied on numerous land use/cover products and methodologies. These include, ecoclassification maps derived from terrain, climate, and vegetation, aerial photographic interpretation, various satellite remote-sensing classifications, and vegetation or productivity surrogates. Generally, most contemporary grizzly bear habitat studies have focused on satellite remote sensing products (Craighead et al., 1985). The popularity of remote sensing has likely been due to consistent temporal availability, large aerial footprint, and the relatively well-defined and automated methodologies available. Use of these products have largely been either land use/cover classifications, such as in Franklin et al. (2001) four our study area, or a greenness model, which is a vegetation index derived from a tasseled cap transformation of satellite bands (Crist & Cicone, 1984; Manley et al., 1992). Although greenness has been related to leaf area index and net primary productivity (White et al., 1997; Waring & Running, 1998), there is as of yet any field-defined relationships with grizzly bear occurrence and habitat. In contrast to satellite remote-sensing products, there currently exists a wealth of geographical information system (GIS) databases relating to land cover mapping from aerial photographic interpretations. In Alberta, these areas primarily reside within Forest Management Agreements (FMA), where such mapping is required. Although such products can differ in quality across FMA’s and usually map open plant community’s areas as non-forested, they provide substantial information on species composition within forest stands. Other GIS data sources potentially useful for predicting grizzly bear habitats include representations of terrain, referred to as digital elevation models (DEM). With a DEM one can not only estimate elevation, but also a number of other terrain derived variables including slope, aspect, drainage, and numerous indices of terrain complexity.

The objectives of this report are to compare the effectiveness of various GIS and remote sensing-based products for use in predicting grizzly bear habitats in west-central Alberta. The current (2003) expansion of the Foothills Model Forest grizzly bear study provides a critical juncture and opportunity for assessing future directions and/or changes in mapping. The expansion not only requires additional data sources and/or imagery, but methods for dealing with large-scale mapping and social-political-industrial boundary issues where quality and quantity of spatial GIS information can change dramatically along non-ecological boundaries. Here, we specifically contrast the effectiveness of 9 existing mapping products and address where future improvements can be directed to provide managers and scientists with the ‘best’ possible habitat maps and models for grizzly bears. 100

9.2 Methods

9.2.1 Mapping products: The candidate models.— In total, we compared 9 mapping products (referred to as candidate models for statistical comparisons) for their effectiveness in predicting grizzly bear occurrence. Six were derived from a Landsat 7 TM imagery acquired from a September 2001 scene (classifications, vegetation surrogates, and raw bands), 1 based on terrain features from a digital elevation model (DEM), 1 random or neutral map, and finally a hybrid map composed of land cover polygons from Alberta Vegetation Inventory (AVI) maps (photo-interpretation) and select classes and spatial locations (Parks Canada) from a Landsat classification. Below we describe each of these products in more detail:

1. IDT-2001. Our first map product was an integrated decision tree (IDT) approach satellite remote sensing classification from a 2001 Landsat 7 TM September scene that identified 23 individual land use/cover categories (Franklin et al., 2001). We simplified these 23 classes into 17 categories (Table 9.2). Urban and well site classes were both masked out as they were extremely rare and were not sampled by radiotelemetry or random available locations. In addition, closed and open deciduous forests were lumped into a deciduous forest class as open deciduous was also too rare for modeling. Finally, clear-cut age classes 0-to-2-years, 3-to-12-years, >12-years, and cut age unknown were aggregated into a single class called clear-cut. These age classes were based on GIS information and did not represent information derived from classifications of remote sensing data.

2. IDT-2001 Simplified A. This map product was simply a reclassification of the

IDT-2001 map with aggregations of specific classes totally 12 categories (Table 9.2). Rock, snow, ice, and water were aggregated into a non-vegetated class, while road/rail and pipeline were lumped into an anthropogenic class, and finally shrub and wet open classes were aggregated into a shrub-bog-wetland category.

3. IDT-2001 Simplified B. This map product was a further simplification of the

original IDT map into 8 more-or-less general land use/cover categories (Table 9.2). From the IDT-2001 Simplified-A map, forest classes were aggregated into either open or closed classes, while alpine/subalpine and herbaceous areas were lumped into a single category called herbaceous. And finally, wet treed was added to the shrub-bog-wetland complex.

4. Enhanced IDT map. The current study area is largely composed of conifer

forests, especially outside of Jasper National Park. Overall, the IDT map depicts this well (e.g. 39% of extent is in the closed conifer class), but unfortunately does not differentiate the different types of conifer stands present. Such differences in habitats may potentially be important for detecting selection of critical resources and likely to be necessary for matching forest-based management. To address this problem, we stratified the closed conifer stands into 3 spectrally unique categories, therefore creating a map with 18 land use/cover classes. We must qualify this classification, however, with the fact that these classes are currently only spectral classes and may not necessarily represent meaningful ecological categories.

101

5. Hybrid IDT/AVI. As point out for the above mapping product, we currently lack well-defined compositionally based land cover classifications within the current IDT product. Although the Enhanced IDT map attempts to stratify conifer stands, there already exists a wealth of land cover information on Crown FMA lands where Alberta Vegetation Inventory (AVI) has largely been completed. To assess the utility of these products for grizzly bear predictions, we have merged AVI and IDT products to produce a seamless land use/cover map. Since we lacked AVI data within Jasper National Park, the IDT classification was necessary for this region. However, in the foothills, we used AVI for forested landscapes and IDT for open, shrub and wetland classes. Furthermore, given the fact that grizzly bears are typically associated with riparian areas, especially during the spring, we developed a GIS to predict riparian areas based on slope, hydrographic features, and open habitat classifications. This riparian model was then draped onto the IDT/AVI map. In total the Hybrid IDT/AVI map had 24 land use/cover categories and represented our most complex model. Further modifications or reclassifications of this map are possible, but complicated by the fact that we are using two separate philosophies in mapping; one based on composition (AVI), and the other on a more general vegetation physiognomy (IDT).

6. Tasseled cap transformations. Instead of directly classifying landscapes into

land use/cover categories from a remote sensing image, a number of researchers have used vegetation surrogates or indices for predictions of animal occurrences, including grizzly bears. The two most commonly used indices are that of the Normalized Difference Vegetation Index (NDVI), commonly used with the AVHRR (1-km resolution) satellite data, and greenness one of three tasseled cap transformations of satellite bands, most typically from the (30-m resolution) Landsat satellite. Here, we attempt to use all 3 tasseled cap transformations, greenness, wetness, and brightness from the September 2001 image used for IDT classifications. Although the tasseled cap transformations are theoretically orthogonal, we found these transformations to be correlated based on a Pearson correlation matrix (Table 9.3). Therefore, in order to use information from the three bands, we used principal components (PC) analysis to summarize the data into non-correlated axes that can then be used as independent model predictors. We used the first two axes of the PC in model development as they explained 95.5% of the variance in greenness, wetness, and brightness.

7. Raw Landsat 7 TM bands. Similar to the argument of the previous tasseled cap

mapping product, we were interested in examining how continuous data from raw Landsat 7 TM remote sensing bands (1 to 5 and 7) might work in predicting grizzly bear occurrence without the need of a land use/cover classification. Many of these TM bands were related, however, and examinations of correlations revealed strong colinearities that would prevent their use in RSF models. It was therefore necessary to use a principal components (PC) approach as in the tasseled cap methods above. Again we used the first two axes of the PC for model development as they explained 96.1% of the bands variance.

102

8. Terrain. Instead of relying on satellite remote sensing methods of either land use/cover classes or vegetation surrogates that can be difficult and/or costly to develop, we explore the effectiveness of terrain-based models using only a digital elevation model (DEM) for predicting grizzly bears. Here, we used two DEM derived variables from a 25-m grid of the study area: 1) the elevation in metres; and 2) a compound topographic index (CTI), commonly referred to as wetness. CTI has previously been found to be correlate with several soil attributes including horizon depth, silt percentage, organic matter, and phosphorous (Moore et al. 1993; Geissler et al. 1995). High CTI values are considered to represent wet or mesic soils with high productivity. We hypothesized that both variables would be non-linear in manner and thus fit quadratics.

9. Random or neutral landscape. Finally, as a reference or null model, we generated a neutral landscape with 23 hypothetical land use/cover (habitat) categories and randomly assigned one of these 23 categories to a habitat patch. The landscape patch configuration was generated in a systematic manner from a ‘fishnet’ grid of 2 ha sized square shaped pixels (patch) that started arbitrarily in the corner of the 2001 study boundary. Each random habitat category had the same likelihood of being included within each fishnet-generated patch (probability ~ 0.0435). Therefore, unlike that of the ‘real’ landscape where land use/cover categories are distributed from rare to common, each random habitat category had equal proportions/composition within the 2001 study boundary.

9.2.2 Grizzly bear and available location data.—Grizzly bears (Ursus arctos L.) were captured in the FMF study area near Hinton, Alberta using aerial darting and leg snaring techniques. Sub-adult and adult animals, stratified by defined Bear Management Units (BMUs), were fitted with either a Televilt Simplex (Lindesburg, Sweden) or ATS (Advanced Telemetry Solutions, Isanti, Minnesota) global position system (GPS) radiotelemetry collar. Collars were programmed to acquire a fix at no longer than 4-hour intervals and data were retrieved from monthly or bi-monthly aerial uploads or via collar downloading once recovered. Spatial locations of grizzly bears were integrated into a geographical information system (GIS) and used for generating minimum convex polygon (MCP) home ranges and for analyses of habitat selection based on mapping productions. Although the FMF project has collected over 33,800 radiotelemetry GPS locations to date (1999 to 2002), we examined here only a subset of these locations to control for seasonal and yearly effects (Nielsen et al. 2002; 2003). This subset represented 4,529 locations from 16 individuals gathered during the berry season (August 1 to denning) of 2001 (Table 9.1, Figure 9.1). We chose 2001 and the berry season as it corresponded to the period of imagery (September 2001 Landsat 7 TM scene) used for remote sensing-based mapping products. To contrast use locations for habitat selection modeling, we required a random or available sample of resources. To derive this sample, we used the estimated annual 100% minimum convex polygon (MCP) home range for each animal and generated a random sample of locations within each MCP using a GIS. These available locations were sampled with an intensity of 1 random location per 1-km2. With this random available sampling design we assumed that any location within the MCP home range was available to that individual grizzly bear at some point during the berry season.

103

Given that our analyses and sampling follows the individual animal, our methodology follows a design III habitat selection approach using presence-availability data (Thomas & Taylor, 1990; Manly et al., 1993; 2002). With these data and designs we generated resource selection functions (RSFs) using mapping products as predictors of grizzly bear occurrence (relative). 9.2.3 Grizzly bear habitat model building strategy.—We evaluated patch or third-order (Johnson, 1980) resource selection for grizzly bears based on presence (GPS radiocollar locations) and available (random) locations. We followed an information theoretic approach to model design and selection (Burnham & Anderson, 1998; Anderson et al., 2000), where nine candidate models were compared for their usefulness in predicting the relative occurrence of grizzly bears using Schwartz’s Bayesian information criteria (SBIC). An information-theoretic approach selects for the most parsimonious model, with the SBIC differing from the more widely used Akaike information criteria (AIC) in having larger penalties for circumstances where larger sample sizes are used (Schwartz, 1978). Using Huberty’s rule of thumb (Huberty, 1994; Fielding & Bell, 1997), we stratified our use locations into a model training and testing data (validation) dataset following an 85% (training) to 15% (testing) data partition. Models were developed using only training data, while testing data were used to assess the predictive capacity of the model. Random available locations were all considered training data (no data partitioning), as we were only interested in testing (validating) the prediction of use locations. Using training data, we developed resource selection function (RSF) models for each candidate model using the following structure,

w(x) = exp(β1 x1 + β2 x2 + … + βk xk),

where w(x) is the resource selection function and βi are selection coefficients based on environmental covariates xi (Manley et al., 1993; 2002). Logistic regression was used to estimate the coefficients for each explanatory variable. For categorical variables, we used deviance coding to estimate the effect of each category from the average effect among all categories. This is in contrast to models using indicator coding (effect of each category to a reference category), which is more common, but requires that the interpretation and inference be based on the reference group. For these models, any estimated coefficients >0 would be interpreted as selection, while <0 avoidance. Coefficients at or near 0 (coefficients overlapping 0) were interpreted as use of those habitats or resources with respect to their availability (neither selection, nor avoidance).

104

Table 9.1. Identification (name), sex (M-male; F-female), GPS radiotelemetry sample size (use), and random available locations sampled within 100% minimum convex polygon (MCP) home ranges for individual grizzly bears (Ursus arctos L.) in west-central Alberta and used for modeling habitat selection in this report. Due to limited mapping products extent, additional grizzly bear radiotelemetry locations were withheld.

Name Sex Use Random Total G02 F 56 71 127G03 F 288 188 476G04 F 274 50 324G12 F 376 852 1228G16 F 289 205 494G20 F 160 184 344G23 F 235 230 465G24 M 432 1550 1982G28 F

M 348 609 957

G29 559 1805 2364G33 M 449 1630 2079G36 F 208 735 943G38 F 259 154 413G40 F 30 89 119G42 F 124 240 364G100 F 442 256 698Total 4529 8848 13377

105

Table 9.2. Integrated decision tree (IDT) approach land use/cover classification for the Foothills Model Forest (FMF) grizzly bear project. Reclassifications of the original 23 IDT map classes were provided in a hierarchal manner to simplify maps into ecological categories. Note that the well site and urban areas were masked from analyses. IDT# Original IDT IDT-2001 IDT-2001 Simplified A IDT-2001 Simplified B1 closed conifer closed conifer closed conifer closed forest 2 closed deciduous deciduous forest deciduous forest

alpine/subalpine

shrub<1800m wet open

non-vegetated

14 water

closed forest 3 mixed forest mixed forest mixed forest closed forest 4 open conifer open conifer open conifer open forest 5 open deciduous deciduous forest deciduous forest open forest 6 alpine/subalpine alpine/subalpine herbaceous 7 herbaceous<1800 herbaceous<1800 herbaceous<1800 herbaceous 8 shrub<1800m shrub-bog-wetland shrub-bog-wetland 9 wet open shrub-bog-wetland shrub-bog-wetland 10 wet treed wet treed wet treed shrub-bog-wetland 11 rock rock non-vegetated 12 snow snow non-vegetated non-vegetated 13 shadow shadow non-vegetated non-vegetated

water non-vegetated non-vegetated 15 road / rail line road / rail line anthropogenic anthropogenic 16 pipeline pipeline anthropogenic anthropogenic 17 well site masked out… masked out… masked out… 18 urban masked out… masked out… masked out… 19 cut 0-2 years cut cut cut 20 cut 3-12 years cut cut cut 21 cut > 12 years cut cut cut 22 cut unknown age cut cut cut 23 recent burn recent burn recent burn recent burn Total 23 17 12 8 Table 9.3. Correlation (Pearson) matrix of tasseled cap transformations of Landsat TM bands for random locations (n = 8,848) within grizzly bear minimum convex polygon (MCP) home ranges.

greenness wetness brightness

greenness 1.000 wetness 0.650 1.000 brightness -0.537 -0.851 1.000

106

Figure 9.1. Study area map for the 2001 boundary (~10,000-km2) of the Foothills Model

Forest (FMF) grizzly bear project and associated grizzly bear (Ursus arctos L.) location data retrieved from GPS radiotelemetry collars during the berry season (August 1 to denning) of 2001. The relief map shows the areas of mountains and associated valley-bowl orientation of the region where bear locations are most typically found. To the northeast, the study area grades into upper and lower foothill ecosubregions where habitats progressively become more boreal in nature.

107

To overcome non-random errors typical of GPS radiotelemetry data (Obbard et al., 1998; Dussault et al., 1999; Rettie & McLoughlin, 1999), we used sample probability weights to adjust biased fix rates in GPS radiocollar use locations (Nielsen & Boyce, 2002; Frair et al., unpublished manuscript). Sample weights represented the inverse of the probability that the observation was included due to sampling error based on empirical models of GPS fix acquisition from test sites that examined the influence of terrain and habitat within west-central Alberta. Since spatio-temporal autocorrelation between GPS locations can bias variance estimates (Lennon 1999, 2000; Nielsen et al., 2002), we further used robust cluster estimates of variances around individual animals. In doing so, we assume the individual animal to be the unit of replication, not the telemetry locations (Otis & White 1999; Nielsen et al., 2002). Independence was therefore assumed for individuals, but not necessarily within an individual (i.e., GPS radiotelemetry location). 9.2.4 Model Comparisons and predictive accuracies.—Candidate models were compared using Schwartz’s Bayesian information criteria (SBIC) weights (wi) that provided the relative likelihood of a model being the best, given the data and models tested. We evaluated the predicted accuracy of each model using the independent out-of-sample testing data (15% partitioned validation dataset). Predictive accuracy of models were assessed by comparing the proportion of testing data (679 radiotelemetry locations) within 10 frequency-adjusted (equalized) bins ranked from low to high (Boyce et al., 2002). Using this method, we assume few testing locations within low habitat quality (relative probability of occurrence) bins, while progressively more locations in subsequent higher quality habitat bins. We use a number of metrics to assess this assumed relationship, including a Somer’s D statistic between bin number and frequency adjusted proportion of locations and a regression slope analysis between the expected proportion of locations and observed proportion of locations. The Somer’s D test, with jackknifed standard errors and significance, is very much like a Spearman rank correlation, ranging from –1 (strong negative association) to 1 (strong positive association). A significant positive Somer’s D would indicate a predictive model where high quality habitat bins contained higher densities of withheld (testing data) animal locations, while low quality habitat bins contained few animal locations (Boyce et al., 2002). Alternatively, we fit a least squares regression line between the observed and expected proportion of testing locations within predicted habitat bins. A slope of 1.0 with little to no variance around that estimate (standard error) or an R2 approaching 1.0 was assumed to be the ‘gold standard’ or perfect model. We tested that the slope was different from 0.0 (random map with equal proportion of testing data in each bin, e.g. 10% of bear locations in each of 10 bins) and 1.0 (our defined gold standard where each bin progressively increases in quality at a standard rate) using a F-test. Although confusion metric approaches rely on crisp set identity and are therefore considered conservative to invalid for presence-availability designs (Boyce et al., 2002), we still assess the classification accuracy of models using a receiver operating characteristic (ROC) curve from within-sample training data. ROC values ranging between 0.5–0.7 were taken to represent low model accuracy, while values between 0.7–0.9 were considered to have good model accuracy, and finally those above 0.9 indicated high model accuracy (Swets, 1988; Manel et al., 2001). Given limitations in ROC analyses, we were further forced to assume non-weighted models (e.g., no GPS bias) and therefore should be further viewed with caution. Finally, we report the proportion of the deviance explained from a null model for each mapping product and McKelvey and Zavoina's R2.

108

9.3 Results

9.3.1 Candidate model comparisons.—Contrasting the nine candidate models using SBIC, we found substantial evidence (wi ≅ 1.0 or 100 %) for concluding that the hybrid model was the best model given the data and sets of models tested (Table 9.5). It therefore appears, from a model selection standpoint, that the inclusion of species composition and a riparian model within the land use/cover map were useful steps for understanding grizzly bear habitat selection. Following the hybrid model, SBIC selected the enhanced IDT map (2nd), again suggesting that some type of species composition stratification is important, a terrain model (3rd), and the IDT-2001 map (4th) (Table 9.5). Our terrain model was based on both elevation and the compound topographic index (~soil wetness). It is interesting that a DEM-driven model alone had greater support than the IDT-2001 map, our principal FMF mapping product for use in project. Inclusion of a topographic ruggedness index from the DEM may further improve terrain-driven models. The least SBIC supported models were the Random map (9th), as might be expected, the tasseled cap transformation (8th), and the raw Landsat bands (7th) (Table 9.5). The low support for the tasseled cap transformation is also interesting given its wide and increasing popularity in carnivore and grizzly bear modeling within the North American Rocky Mountains. As with the SBIC ranked candidate models, percent deviance explained, ROC, and R2 (McKelvey and Zavoina's) values were ranked in a similar manner (Table 9.5). All of these metrics, however, pointed to the large amount of residual variation left unexplained. Much of this can be explained by the population-level design. Substantial differences in selection strategies are known to exist for this population (Nielsen et al., 2002) making single models of selection at the population level quite general and noisy. 9.3.2 Model validation: Predictive capacity.—Although we found substaintial SBIC support for the hybrid model, the predictive capacity of the model was not as high as some other candidate models (Table 9.6). Somer’s D tests between habitat bin rank (from low to high) and the proportion of testing locations falling within those bins revealed that the IDT-simplified A product was the most predictive followed by the IDT-2001 and tasseled cap products (Table 9.6). The hybrid AVI/IDT map was 4th overall, suggesting that although it explained the most variation within the data and maintained parsimony, its predictive capacity was possibly unbalanced in nature. In figure 9.2, it can be seen that the hybrid model was under predicting observations in bin 8 and to a lesser degree bin 6, while maintaining low numbers of observations in the low quality habitat bins numbered 1 to 5. Despite Somer’s D differences, all maps, excluding the random map, were significant at the P<0.001-level suggesting strong predictive capacity.

109

Table 9.4. Correlation (Pearson) matrix of Landsat TM bands 1 to 6 for random locations (n = 8,848) within grizzly bear minimum convex polygon (MCP) home ranges. Band-1 Band-2 Band-3 Band-4 Band-5 Band-7Band-1 1.000 Band-2 0.976 1.000 Band-3 0.964 0.988 1.000 Band-4 0.472 0.591 0.572 1.000 Band-5 0.820 0.883 0.912 0.689 1.000 Band-7 0.892 0.924 0.948 0.544 0.964 1.000

110

Table 9.5. Assessment of model fit and model selection between candidate mapping products used for predicting grizzly bear habitat within west-central Alberta. Complexity of model (parameters) is provided by K. Schwartz Bayesian information criteria (SBIC) is reported and used to select the most parsimonious model. Change between individual candidate models and lowest SBIC model is represented by ∆ SBIC, while the weight of the model within the set of models and data tested is indicated by wi (rank of wi is further provided). Although we have presence-availability data, we report % deviance explained (% Dev.), McKelvey and Zavoina's R2, and receiver operating characteristic (ROC) for all models (see Boyce et al. 2002 for limitations of these metrics). Model K % Dev. M&Z R2 ROC SBIC ∆ SBIC wi RankHybrid AVI / IDT 24 7.14 0.124 0.665 15448.2 0.0 ~1.0 1 Enhanced IDT 18 5.65 0.107 0.652 15694.8 246.6 2.9E-54 2 Terrain 4 5.04 0.121 0.635 15700.4 252.2 1.7E-55 3 IDT 2001 16 5.13 0.099 0.641 15780.2

6

0.058 16047.3Random

331.9 8.3E-73 4 IDT simplified A 11 4.57 0.084 0.634 15843.8 395.6 1.3E-86 5 IDT simplified B 8 4.05 0.076 0.614 15901.4 453.2 4E-99 TM Bands 2 2.89 0.060 0.599 16037 588.7 1E-128 7 Tasseled cap 2 2.82 0.596 599.0 8E-131 8

22 0.37 0.007 0.538 16565.6 1117.4 2E-243 9

111

Bin rank (habitat quality) 1 2 3 4 5 6 7 8 9 10

Prop

ortio

n of

test

ing

loca

tions

with

in b

in i

0.00

0.05

0.10

0.15

0.20

0.25

0.30

enhIDT IDT01 IDTsimpA IDTsimpB Hybrid Random Tasslecap Bands Terrain

High quality habitat Low quality habitat

Expected region for low quality habitats

Expected region for high quality habitats

Expected for random map (equal proportion)

Figure 9.2. Relationship between habitat quality rank (bin number) and proportion of withheld

testing locations falling within that bin.

112

Table 9.6. Results of the Somer’s D tests between habitat bin rank (low to high) and the proportion of testing locations (n = 679) falling within those bins and ordered by SBIC rank. Good models would have a significant large positive relationship. Jackknife standard errors, the probability that D is different from 0, and 95% confidence intervals are provided, along with the rank in the strength of relationship (note similar ranks/D values (poor discrimination) for some models).

Somer’s 95% C.I. Somer’s D

Jackknife Model D S. E. P lower upper Rank Hybrid AVI / IDT 0.711 0.155 <0.001 0.408 1.014 4 Enhanced IDT 0.733 0.194 <0.001 0.352 1.114 3 Terrain 0.733

<0.001

<0.001

0.194 <0.001 0.352 1.114 3 IDT 2001 0.867 0.153 0.567 1.166 2 IDT simplified A 0.956 0.067 <0.001 0.825 1.086 1 IDT simplified B 0.689 0.186 <0.001 0.325 1.053 5 TM Bands 0.733 0.163 0.413 1.053 3 Tasseled cap 0.867 0.153 <0.001 0.567 1.166 2 Random 0.289 0.306 0.344 -0.310 0.888 6

Table 9.7. Regression between expected and observed proportion of validation observations within habitat bins ordered by SBIC ranks. Slope, standard error (S.E.), and R2 are provided, along with F tests (1, 8 df) for the significance between slope estimate and null hypothesis of equaling zero (Ho = 0.0) and one (Ho = 1.0).

Model slope Ho = 0.0 Ho = 1.0 Model Slope S. E. R2 F P F P Hybrid AVI / IDT 1.046 0.606 0.298 12.30 0.008 0.02 0.880 Enhanced IDT 0.871 0.163 0.782 28.70 <0.001 0.63 0.449 Terrain 0.843 0.131 0.838 41.45 <0.001 1.44 0.264 IDT 2001 0.861 0.170 0.761 25.53 <0.001 0.67 0.437 IDT simplified A 0.865 0.169 0.766 26.17 <0.001 0.64 0.448 IDT simplified B 0.227 TM Bands

0.250

0.745 0.196 0.644 14.45 0.005 1.72 0.687 0.133 0.770 26.81 <0.001 5.57 0.046

Tasseled cap 0.677 0.112 0.821 36.56 <0.001 8.30 0.021 Random 0.137 0.084 2.66 0.141 105.05 <0.001

113

In support of SBIC model selection weights, we found that the estimated slopes between observed and expected testing validation proportions (Figure 9.3) were ranked in a similar manner to that of the SBIC models (Table 9.7). Slopes ranged from 0.137 (random map) to 1.046 (hybrid model). Tests of slope, however, revealed that the random map, TM bands, and tasseled cap product had slopes significantly different from 1.0, our gold standard. All products, excluding the random map showed significant differences from 0.0 or random noise (Table 9.7). The enhanced IDT and hybrid AVI/IDT map products appeared to best follow the gold standard. 9.3.3 Habitat selection coefficients.—We present coefficients, robust standard errors (adjusted for the level of the individual), and 95% confidence intervals for variables from individual candidate models. For the SBIC selected hybrid product, avoided classes included, anthropogenic industrial lands, closed Engelmann spruce-subalpine fir (ESSF) forest, open pine forest, closed and open white spruce forest, closed mixed conifer forest, closed wet conifer forest, clear-cuts, and non-vegetated areas (Table 9.8). In contrast, open conifer (IDT), mixed forest (IDT), wet open (IDT), wet treed (IDT), and the no class category were selected for. Riparian did not appear to be an important predictor of grizzly bear habitat during the berry season, but likely may be important for other seasons.

The enhanced IDT-2001 map showed strong avoidance for recent burn, clear-cuts, shadow, and snow, while strong selection for alpine/subalpine, closed conifer 2, closed conifer 3, deciduous forest, herbaceous, open conifer, and shrub areas (Table 9.9). It is interesting that for the hybrid model a number of forest stands were strongly avoided, but no such pattern was observed with the enhanced IDT-2001 product. The terrain product showed selection for intermediate elevations with little overall relationship with the compound topographic index (CTI) (Table 9.10). For the IDT-2001 product, patterns were similar to the enhanced IDT map, excluding the closed conifer selection observed in the enhanced product (Table 9.11). Here, with closed conifer being lumped into one large (39% of landscape) class, we failed to show any interesting selection dynamics as the estimate was near zero. This suggests that it is important to map and understand forested conifer habitats in more detail. Simplifications of the IDT-2001 map product reflected the original product, although numerous classes were combined (Tables 9.12 & 9.13). The two principal component (PC) axes representing the six raw Landsat TM bands showed significance for PC 2, but not for PC 1 (Table.9.14). PC2 was primarily positively related to band 4 and to a lesser degree negatively related to band 1. In contrast, the use of 2 principal components axes of the tasseled cap transformations greenness, wetness, and brightness showed significant relationships with the occurrence of grizzly bears (Table 9.15). Finally, for the random map product we found that all categories, excluding random habitat 18, were non-significant (Table 9.16). We are unaware of the reasons why habitat 18 had significant avoidance, but given the number of categories (23) the likelihood of one category being significant due to chance is high.

114

9.4 Suggestions, Limitations, and Conclusions

We suggest that species related land cover classifications, particularly for the conifer category, be further developed either through satellite remote sensing-based methods or through the incorporation of AVI data. We found that such products in the FMF study area were strongly selected for in Schwartz’s Bayesian information criteria (SBIC) weights and for the most part by prediction-based assessments. If satellite remote sensing classifications prove too difficult to distinguish forest stand types, AVI may provide reasonable classifications. However, areas that lack AVI data, such as Jasper National Park, may require aerial photographic interpretation. In either case, the improvement of land use/cover classifications towards a tree species-based mapping programme, regardless of the method, would ultimately gain added acceptance by current land use practitioners (especially forestry partners where composition is crucial) and conservation initiatives requiring more ‘realistic landscapes’. Further benefits would likely include more detailed landscape and patch metrics analyses, population viability analyses, and basic inferences regarding habitat selection ultimately influencing management. For example, grizzly bear habitat use of closed conifer stands within the IDT-2001 product (Table 9.11) was largely uninformative having a selection coefficient near zero, meaning that animals neither avoided nor selected for these areas. However, when closed conifer was further separated into species and spectral classes (Table 9.8 & 9.9), there were strong distinctions between those classes that were avoided and those that were selected. In contrast to generating additional forest classes, the inclusion of a riparian category did not prove useful during the berry season (Table 9.8). It is likely, however, that a riparian category would prove to be valuable for the pre-berry period when bears use those areas for movement and feeding (primarily Hedysarum spp.) activities. Previous work for the area has found substantial variation in selection between individuals and among seasons (Nielsen et al., 2002), even at temporal scales as small as one month (Nielsen et al., 2003). Selection behaviours are also likely to vary among years (Schooley, 1994), as critical resources used by grizzly bears are temporally dynamic at annual periods (e.g., berry productivity). Changes to these critical food resources are likely to cause switching patterns in habitat selection. These results presented herein therefore are limited in their utility and application of estimated coefficients. It is possible that under different food seasons and/or years, or even for different individuals, product effectiveness could vary. We feel, however, that given the importance of the hyperphagic (~berry) period, this season would be the most critical for habitat mapping and testing of product effectiveness. Furthermore, land use/cover information was most accurate for the year 2001 with the clearest contiguous satellite scene from that year. Although these limitations may be seen initially as major restrictions, it provides stimulus for further exploration of potential differences and hence a better understanding of populations and resource dynamics (a subject of the senior authors dissertation).

115

Expected proportion of validation locations0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20

Obs

erve

d pr

opor

tion

of v

alid

atio

n lo

catio

ns

0.00

0.05

0.10

0.15

0.20

0.25

0.30Gold StandardenhIDT IDT01 IDTsimpA IDTsimpB Hybrid Random Tasslecap Bands Terrain

Low quality habitats High quality habitats

Expected region for low quality habitats

Expected region for high quality habitats

Expected by chance (random map)

Figure 9.3. Comparison of expected and observed proportion of validation locations. Note the

defined ‘gold standard’ representing a slope of 1.0 and hence a systematically increasing quality of habitat (relative probability of occurrence).

116

Table 9.8. Estimated coefficients (βi) of habitat selection for land use/cover types for grizzly bears in west-central Alberta using the hybrid AVI/IDT mapping product. Classes not overlapping zero (selection or avoidance) are indicated in bold font. Robust standard errors (S.E.) are reported as estimated across individuals, not autocorrelated locations. Robust 95% C.I. Land use/cover type βi S. E. lower upper alpine/subalpine (IDT) 0.607 0.476 -0.326 1.540 anthropogenic industrial -0.690 0.174 -1.031 -0.349 closed Engelmann spruce-subalpine fir forest -0.867 0.384 -1.620 -0.114 open Engelmann spruce-subalpine fir forest -0.308 0.246 -0.791 0.175 closed conifer (IDT) 0.403 0.285 -0.156 0.962 open conifer (IDT) 1.334 0.470 0.413 2.256 closed pine forest -0.362 0.275 -0.901 0.177 open pine forest -0.354 0.150 -0.648 -0.060 closed white spruce forest -1.137 0.485 -2.088 -0.185 open white spruce forest -0.578 0.259 -1.086 -0.071 mixed forest (IDT) 1.997 0.356 1.298 2.696 closed mixed conifer forest -0.931 0.228 -1.378 -0.484 open mixed conifer forest -0.337 0.296 -0.917 0.243 closed mixed forest 0.087 1.019 -1.911 2.084 open mixed forest 0.176 0.216 -0.248 0.599 deciduous forest 0.457 0.263 -0.058 0.971

wet treed (IDT) 0.514

0.052

-0.081

closed wet conifer forest -1.000 0.308 -1.603 -0.397 open wet conifer forest -0.429 0.511 -1.431 0.573 wet open (IDT) 1.024 0.377 0.286 1.763

1.972 0.964 2.979 clearcut -0.999 0.089 -1.172 -0.825 herbaceous 0.175 -0.292 0.396 non-vegetated -0.742 0.287 -1.305 -0.179 riparian 0.221 -0.513 0.351 no class 0.706 0.248 0.221 1.191

117

Table 9.9. Estimated coefficients (βi) of habitat selection for land use/cover types for grizzly bears in west-central Alberta using the enhanced Integrated Decision Tree (IDT) remote sensing mapping product. Classes not overlapping zero (selection or avoidance) are indicated in bold font. Robust standard errors (S.E.) are reported as estimated across individuals, not autocorrelated locations.

Robust 95% C.I. Land use/cover type βi S. E. lower upper alpine/subalpine 1.027 0.392 0.260 1.795 recent burn -2.532 0.959 -4.411 -0.653 closed conifer 1 -0.147 0.208 -0.555 0.260 closed conifer 2 0.253 0.125 0.008 0.497 closed conifer 3 1.031 0.247 0.546 1.516 clearcut -0.681 0.131 -0.937 -0.425 deciduous forest 0.799 0.182 0.443 1.156 herbaceous 0.938 0.192 0.562 1.313 mixed forest 0.460 0.317 -0.162 1.082 open conifer 1.115 0.282 0.562 1.668 pipeline -0.299 0.314 -0.914 0.316 road & rail -0.300 0.182 -0.657 0.058 rock -0.300 0.252 -0.795

-2.341

-0.118

0.195 shadow -1.337 0.423 -2.167 -0.508 shrub 0.878 0.335 0.221 1.535 snow -1.396 0.482 -0.452 water 0.242 0.526 -0.788 1.272 wet-open -0.162 0.290 -0.730 0.406 wet-treed 0.413 0.271 0.943

Robust

Table 9.10. Estimated coefficients (βi) of habitat selection for grizzly bears in west-central Alberta using the terrain mapping product (elevation and compound topographic index, CTI). Elevation is in metres. Variables not overlapping zero (selection or avoidance) are indicated in bold font. Robust standard errors (S.E.) are reported as estimated across individuals, not autocorrelated locations.

95% C.I. Variable βi S. E. lower upper Elevation 0.013 0.003 0.007 0.019 Elevation2 -3.82E-06 8.66E-07 -5.51E-06 -2.12E-06 CTI -0.021 0.089 -0.194 0.153 CTI2 0.005 0.003 -0.001 0.011

118

Table 9.11. Estimated coefficients (βi) of habitat selection for land use/cover types for grizzly bears in west-central Alberta using the IDT-2001 product. Variables not overlapping zero (selection or avoidance) are indicated in bold font. Robust standard errors (S.E.) are reported as estimated across individuals, not autocorrelated locations.

Robust 95% C.I.

Land use/cover type β lower i S. E. upper 1.048 0.389 1.810

recent burn -2.527 0.954 -4.396 -0.657 closed conifer

1.235

mixed forest

0.082 0.143 -0.198 0.362 clearcut -0.609 0.138 -0.880 -0.339 deciduous forest 0.848 0.198 0.461 herbaceous 0.999 0.195 0.616 1.382

0.438 0.315 -0.179 1.054 open conifer 1.149 0.270 0.620 1.677 pipeline 0.063 0.407 -0.734 0.861 road & rail -0.163 0.178 -0.511 0.185 rock -0.224 0.254 -0.722 0.273 shadow -1.300 0.417 -2.118 -0.483 shrub 0.954 0.363 0.243 1.665 snow -1.359 0.481 -2.303 -0.416 water 0.275 0.524 -0.753 1.303 wet-open -0.103 0.300 -0.690 0.484 wet-treed 0.430 0.285 -0.129 0.989

alpine/subalpine 0.285

119

Table 9.12. Estimated coefficients (βi) of habitat selection for land use/cover types for grizzly bears in west-central Alberta using the IDT-2001 simplified A product. Variables not overlapping zero (selection or avoidance) are indicated in bold font. Robust standard errors (S.E.) are reported as estimated across individuals, not autocorrelated locations.

Robust 95% C.I. Land use/cover type βi S. E. lower upper alpine/subalpine 0.924 0.400 0.139 1.708 anthropogenic -0.269 0.169 -0.601 0.063 recent burn -2.651 0.931 -4.476 -0.826 closed conifer -0.042 0.147 -0.330 0.246 clearcut -0.733 0.128 -0.984 -0.482 deciduous forest 0.724 0.195 0.342 1.106 herbaceous 0.875 0.192 0.498 1.251 mixed forest 0.314 0.318 -0.309 0.936 non-vegetated -0.538 0.226 -0.981 -0.096 open conifer 1.024 0.286 0.464 1.584 shrub-bog 0.068 0.653 0.298 -0.516 wet-treed 0.306 0.278 -0.239 0.851

Table 9.13. Estimated coefficients (βi) of habitat selection for land use/cover types for grizzly bears in west-central Alberta using the IDT-2001 simplified B product. Variables not overlapping zero (selection or avoidance) are indicated in bold font. Robust standard errors (S.E.) are reported as estimated across individuals, not autocorrelated locations.

Robust 95% C.I. Land use/cover type βi S. E. lower upper

-0.145 0.208 -0.553 0.262 -2.527 0.880 -0.803

closed forest 0.218 0.224 -0.221 clearcut

1.048 0.365 0.333 1.762 non-vegetated -0.415 0.199 -0.805 -0.024 open forest 1.142 0.264 0.624 1.660 shrub-bog-wetland 0.290 0.327 -0.351 0.932

anthropogenic recent burn -4.251

0.658 -0.610 0.177 -0.956 -0.263

alpine/herbaceous

120

Table 9.14. Estimated coefficients (βi) of habitat selection for grizzly bears in west-central Alberta using the raw TM Landsat bands (2 PC axes) product. Variables not overlapping zero (selection or avoidance) are indicated in bold font. Robust standard errors (S.E.) are reported as estimated across individuals, not autocorrelated locations.

Robust 95% C.I. Landsat bands PC axis βi S. E. lower upper PC of bands 1 -0.073 0.045 -0.161 0.015 PC of bands 2 0.496 0.095 0.310 0.681

Table 9.15. Estimated coefficients (βi) of habitat selection for grizzly bears in west-central Alberta using the tasseled cap transformation (2 PC axes) product. Variables not overlapping zero (selection or avoidance) are indicated in bold font. Robust standard errors (S.E.) are reported as estimated across individuals, not autocorrelated locations.

Robust 95% C.I.

Tasseled cap PC axis βi S. E. lower upper 0.1641 0.0691 0.029 0.299

0.250 0.656 PC of tasseled cap 1 PC of tasseled cap 2 0.4525 0.1036

121

Table 9.16. Estimated coefficients (βi) of habitat selection for land use/cover types for grizzly bears in west-central Alberta using the random map product. Variables not overlapping zero (selection or avoidance) are indicated in bold font. Robust standard errors (S.E.) are reported as estimated across individuals, not autocorrelated locations.

Robust 95% C.I. Land use/cover type βi S. E. lower upper Random habitat 1 0.038 0.212 -0.378 0.453 Random habitat 2 0.180 0.225 -0.260 0.620 Random habitat 3 -0.120 0.091 -0.298 0.058 Random habitat 4 0.189 0.145 -0.095 0.473

-0.392 0.141

-0.263

Random habitat 13

0.156

Random habitat 18 0.095

-0.120

-0.138 Random habitat 23 0.055

Random habitat 5 -0.009 0.195 0.373 Random habitat 6 -0.213 -0.488 0.063 Random habitat 7 0.027 0.119 -0.207 0.260 Random habitat 8 -0.036 0.100 -0.232 0.160 Random habitat 9 0.024 0.105 -0.182 0.229 Random habitat 10 -0.018 0.125 0.226 Random habitat 11 0.071 0.163 -0.249 0.391 Random habitat 12 0.119 0.113 -0.103 0.341

-0.024 0.111 -0.241 0.194 Random habitat 14 -0.047 0.123 -0.289 0.194 Random habitat 15 0.112 0.085 -0.055 0.278 Random habitat 16 -0.071 -0.376 0.235 Random habitat 17 0.057 0.180 -0.296 0.410

-0.454 0.142 -0.731 -0.176 Random habitat 19 -0.135 0.117 -0.365 Random habitat 20 0.146 0.136 0.413 Random habitat 21 0.247 0.425 -0.586 1.080 Random habitat 22 0.133 -0.399 0.123

0.182 -0.302 0.412

122

Finally, it should be pointed out that current remote sensing and forest GIS-based land use/cover maps (such as those used in this report) may not reflect the scale and/or resource that grizzly bears perceive when making selection decisions (Nielsen et al., 2003). Remote sensing and forest GIS layers relate more to the perception and scales of resources that humans find important for management, not necessarily that of grizzly bears. Added uncertainty (accuracy) in the GIS and remote sensing data make interpretations even fuzzier. Although the IDT products contained an overall accuracy near 80% (Franklin et al., 2001), some important habitats, such as open conifer, had accuracies closer to that of 50%. The use of indirect resource/habitat gradients provides substantial prediction uncertainty, particularly in space and time, a goal of this project (Austin et al., 1990; Austin & Gaywood 1994; Nielsen et al., 2003). Variables, such as elevation, closed conifer, or greenness, are really only (and hopefully) a surrogate for some resource (e.g. foods) we failed to directly measure and/or map. As an alternative, Nielsen et al. (2003) developed food-based models and found substantial improvements in explanation of grizzly bear occurrence. Future developments in food-based mapping are underway by Nielsen that may provide alternative approaches for habitat mapping. Unfortunately, these methods still will be limited by accuracy of GIS and in some instances remote sensing information and are computationally difficult to derive, especially at the scales of the expanded study area.

In conclusion, despite possible limitations in GIS and remote sensing data, we found that most mapping products were significantly predicting the occurrence of grizzly bears. Identification of tree species within forest stands for satellite remote sensing and AVI-based methods appears to be important and requires further attention. Use of vegetation surrogates, however, performed rather poorly and should be cautioned against as sole predictors for grizzly bear habitats. Previous work (Nielsen, unpublished) has found greenness to work rather well in the mountains where it distinguished rock (low greenness and bear use), from forests (medium greenness and bear use) and herbaceous alpine areas (high greenness and bear use), but poorly in the foothills where such simple distinctions are not possible.

123

9.5 References

Anderson, D.R., K.P. Burnham & W.L. Thompson, 2000. Null hypothesis testing: problems, prevalence, and an alternative. Journal of Wildlife Management, 64:912-923.

Austin, M.P., A.O. Nicholls & C.R. Margules, 1990. Measurement of the realized qualitative

niche: Environmental niches of five Eucalyptus species. Ecological Monographs, 60:161-177. Austin, M.P. & M.J. Gaywood, 1994. Current problems of environmental gradients and species

response curves in relation to continuum theory. Journal of Vegetation Science, 5:473-482. Boyce, M.S., P.R. Vernier, S.E. Nielsen, & F.K.A. Schmiegelow, 2002. Evaluating resource

selection functions. Ecological Modelling, 157:281-300.

Crist, E.P. & R.C. Cicone, 1984. Application of the tasseled cap concept to simulated thematic mapper data. Photogrammetric Engineering and Remote Sensing, 50:343-352.

Boyce, M.S. & J.S. Waller, 2003. Predicting the number of grizzly bears in the proposed

recovery area of the Bitterroot ecosystem. Wildlife Society Bulletin, in press. Burnham, K.P. & D.R. Anderson, 1998. Model selection and inference: A practical information-

theoretic approach. Springer-Verlag, New York.

Chatterjee, S., A.S. Hadi & B. Price, 2000. Regression analysis by example. Third edition. John Wiley & Sons, New York, New York.

Criaghead, J.J., F.L. Craighead & D.J. Craighead, 1985. Using satellites to evaluate ecosystems

as grizzly bear habitat. Pages 101-112, In: Proceedings , Grizzly Bear Habitat Symposium, USDA Forest Service Intermountain Research Station, Ogden, Utah.

Dassault, C., R. Courtois, J.P. Ouellet & J. Huot, 1999. Evaluation of GPS telemetry collar

performance for habitat studies in the boreal forest. Wildlife Society Bulletin, 27:965-972. Fielding, A.H. & J.F. Bell, 1997. A review of methods for the assessment of prediction errors in

conservation presence/absence models. Environmental Conservation, 24:38-49. Frair, J.L, S.E. Nielsen, E.H. Merrill, M.S. Boyce, S. Lele, R. H.M. Munro & G.B. Stenhouse,

2003. Incorporating habitat-biased GPS location error into resource selection models. Unpublished manuscript.

Franklin, S.E., G.B. Stenhouse, M.J. Hansen, C.C. Popplewell, J.A. Dechka & D.R. Peddle, 2001. An integrated decision tree approach (IDTA) to mapping landcover using satellite remote sensing in support of grizzly bear habitat analysis in the Alberta Yellowhead Ecosystem. Canadian Journal of Remote Sensing, 27:579-592.

Geissler, P.E., I.D. Moore, N.J. McKenszie, and P.J. Ryan, 1995. Soil-landscape modeling and

spatial prediction of soil attributes. International Journal of GIS, 9:421-432.

124

. 2000. Red-shifts and red herrings in geographical ecology. Ecography, 23:101-113.

Moore, I.D., P.E. Gessler, G.A. Nielsen, and G.A. Petersen, 1993. Terrain attributes: estimation methods and scale effects. Pages 189-214 in Modeling change in environmental systems, A.J. Jakeman, M.B. Beck, and M. McAleer (eds.), Wiley, London.

Johnson, D.H., 1980. The comparison of usage and availability measurements for evaluating resource preference. Ecology, 61:65-71.

Lennon, J.J., 1999. Resource selection functions: taking space seriously. Trends in Ecology and

Evolution, 14:399-400.

Manel, S., H.C. Williams & S.J. Ormerod, 2001. Evaluating presence-absence models in

ecology: the need to account for prevalence. Journal of Applied Ecology, 38:921-931. Manley, T.L., K. Ake & R.D. Mace, 1992. Mapping grizzly bear habitat using Landsat TM

satellite imagery. Pages 231-240, In: J.D. Greer, editor. Remote sensing and natural resource management. American Society of Photogrammetry and Remote Sensing, Bethesda, Maryland, USA.

Manly, B.F.J., L.L. McDonald & D.L. Thomas, 1993. Resource selection by animals: statistical

design and analysis for field studies. Chapman & Hall, London, United Kingdom. Manly, B.F.J., L.L. McDonald, D.L. Thomas, T.L. McDonald & W. P. Erickson, 2002. Resource

selection by animals: statistical design and analysis for field studies. Second Edition. Chapman & Hall, London, United Kingdom.

Nielsen, S.E. & M.S. Boyce, 2002. Resource selection functions and population viability

analyses. Pages 17–42, in G.B. Stenhouse & R.H.M. Munro eds. Foothills Model Forest Grizzly Bear Research Program: 2001 Annual Report, March 2002, Hinton, Alberta, Canada.

Nielsen, S.E., M.S. Boyce, G.B. Stenhouse & R.H.M. Munro, 2002. Modeling grizzly bear

habitats in the Yellowhead Ecosystem of Alberta: taking autocorrelation seriously. Ursus, 13:533-542.

Nielsen, S.E., M.S. Boyce, G.B. Stenhouse & R.H.M. Munro, 2003. Development and testing of

phenologically driven grizzly bear habitat models. Ecoscience: (in press). Obbard, M.E., B.A. Pond & A. Perera, 1998. Preliminary evaluation of GPS collars for analysis

of habitat use and activity patterns of black bears. Ursus, 10:209-217. Otis, D.L. & G.C. White, 1999. Autocorrelation of location estimates and the analysis of

radiotracking data. Journal of Wildlife Management, 63:1039-1044.

125

Rettie, W.J. & P.D. McLoughlin, 1999. Overcoming radiotelemetry bias in habitat-selection studies. Canadian Journal of Zoology, 77:1175-1184.

Schooley, R.L., 1994. Annual variation in habitat selection: patterns concealed by pooled data.

Journal of Wildlife Management, 58:367-374. Schwartz, G., 1978. Estimating the dimension of a model. Annals of Statistics, 6:461-464. Swets, J.A., 1988. Measuring the accuracy of diagnostic systems. Science, 240:1285-1293. Thomas, D.L., & E.J. Taylor, 1990. Study designs and tests for comparing resource use and

availability. Journal of Wildlife Management, 54:322-330. Waring, R.H. & S.W. Running, 1998. Forest ecosystems: analysis at multiple scales. Second

Edition. Academic Press, San Diego, California, USA. White, J.D., S.W. Running, R. Nemani, R.E. Keane & K.C. Ryan, 1997. Measurement and

remote sensing of LAI in Rocky Mountain montane ecosystems. Canadian Journal of Forest Research, 27:1714-1727.

Zweig, M.H. & G. Campbell, 1993. Receiver-operating characteristic (ROC) plots: a

fundamental tool in clinical medicine. Clinical Chemistry, 39:561-577.

126

10.0 FUTURE HABITAT MODELLING FOR GRIZZLY BEARS IN THE FOOTHILLS OF THE ROCKY MOUNTAINS

Falk Huettmann, Department of Geography, University of Calgary, Canada [email protected] Steven E. Franklin, Department of Geography, University of Calgary, Canada

10.1 Introduction

Landscapes are constantly changing. Despite natural changes such as succession, man-made changes occur as well, e.g. forestry. A landscape has many functions; it presents habitat for wildlife but also income to human populations. In recent times, the demands on landscapes have increased manifold times, whereas methods and approaches to trace these changes are still less developed. Here work in progress on a predictive modelling approach for a multi-use landscape, the Foothills of the Rocky Mountains in Alberta, Canada, is presented in order to trace such changes and offer solutions to safeguard Grizzly Bears in the future.

Modelling has received increased attention in recent years to describe and understand ecological patterns and processes (e.g. White et al. 1999, Keane et al. 1999). Models on a landscape scale deal with natural events such as fire, insects, succession, storm events and others (e.g. Voinov et al. 1999). Additional models were developed that address man-made changes such as forestry, road locations, human settlements and global change (e.g. Gilruth et al. 1995, Agarwal et al. 2000). Although many of these subjects have been expressed as simple or complex models, most of them still deal with aspatial approaches, e.g. summarizing effects for a certain spatial unit but not providing spatially explicit information. Likely, such approaches are insufficient when modelling ecological patterns and processes on a landscape scale. Due to increased development of GIS and the improved understanding of spatial interactions in (landscape) ecology, spatially explicit models are nowadays used (Remsoft 2002).

This project tries to derive in a sound fashion the future state of Grizzly Bear habitat . Grizzly Bears were already labeled Umbrella, Flagship and Indicator species. In the absence of relevant spatial future habitat models for wildlife, it is believed that future habitat modelling for this species might show very well how future modelling could be applied to wildlife in general.

10.2 Methods

10.2.1 General approach and infrastructure

We used the software packages of Woodstock and Stanley (REMSOFT 2002) to model the future landcover for Grizzly Bear habitat. This allowed us to implement realistic forest management scenarios combined with aspatial and spatial approaches towards a reliable future landscape scenario, as it relates to Grizzly Bears.

127


10.2.2 Data sets

The study area lacks a consistent mapping coverage. Some sections are mapped with an AVI, e.g. from landowners such as Weyerhaeuser, Weldwood and Sunpine. Other sections have not received any GIS mapping even. This situation makes the IDT map as input for a consistent and large-scale landcover map ideal (Franklin et al. 2001). However, assigning ages to polygons can present a difficult task (see also Andison 1998). This was achieved in ArcView 3.2 by overlaying polygons over the centre AVI GIS maps with a known age. Other areas were aged by overlaying polygons from the fire history maps provided for free by the Alberta Government via WWW and elsewhere. The remaining polygons were aged by assigning average age from ground-plots collected over the years 1999 to 2001 (Franklin unpublished) and overlaid for individual IDT classes. The resulting dataset presents the best age information available for the study area (Fig. 10.1). The area of the National Park and some ‘White Zones’ were specifically labelled since they present sections that have no relevant modifying management activities. Another stratification was made by elevation classes in order to capture growth and climate zones; a DEM (Digital Elevation Model) was used. Inoperable areas, e.g. with a steep slope, were also mapped. Habitat requirements for Grizzly Bear habitat were derived from a long-term study, based on GPS telemetry with satellites (e.g. compare also Linke et al. in review, Nielsen et al. in press).

10.2.3 The natural succession and forestry model

Natural succession is difficult to describe when forest stand polygons and habitat aggregates such as the IDT polygons are used. Likely, some of the Growth and Yield curves are not very accurate for the study area. However, we tried to implement known succession series as accurately as possible, and as realistic as possible in Woodstock and Stanley, using local and available information, e.g. Burns and Honkala (1990), Hunag et al. (2001), Alberta Environmental Protection Land and Forest Service Resource Data Division (1985). These natural patterns can be overruled by Forest Activities For Forest Harvest Modelling known silvicultural harvest techniques are implemented, e.g. Alberta Ground Rules, Detailed Forest Management Plans and Smith et al. (1997).

10.2.4 Fire modelling

So far, existing and specific fire models, such as Landis (eg. Doyon and Duinker 2000, Andison pers. com.) were not used. However, it was started to implement localized knowledge (Andison 1998) and the results from the Fire Polygon Database for the study area in a Woodstock analysis using the occurrence of fire in a random fashion. These results were implemented into Woodstock and Stanley as constraints in the optimization section.

10.2.5 Storm event modelling

Spatial storm models do not exist for the study area. From local knowledge, the occurrence of storm events was created in a random fashion in Woodstock. These results were implemented into Woodstock and Stanley as constraints in the optimization section.

128

10.2.6 Insect modelling

Spatial insect models do not exist for the study. As with storm events, insect infestations were manifested in a random fashion in Woodstock. These results were then implemented into Woodstock and Stanley as constraints in the optimization section.

10.2.7 Road modelling

Accepted predictive road models are not known, other than the approach implemented by PATCHWORKS (Forest Corp. unpublished). For now, it was decided to develop a future landscape model first, and predict roads afterwards on top of this existing GIS layers. However, most road locations are driven by locations with high forest volume in accessible areas. Some forest stands can well get harvested when accessible by a road and thus are affected by road proximities. The present model cannot address the circularity of these relationships in the prediction of roads and forests. The current implementation uses predicted forest volume and a DEM as a predictor where roads would occur, and thus implements future road locations manually. Currently, a PATCHWORKS implementation is investigated, which is supposed to present the most realistic scenarios known for the topic. This approach would encompass a ‘roll-over’ modelling approach, which is coming forward in next modeling rounds. 10.2.8 Human settlement modelling

The occurrence of human settlements appears to show specific patterns. Here we model them as point patterns, which expand from existing settlements. Additional new villages are predicted where major new road intersections are predicted by the forestry and road model. However, the study area is presently very sparsely settled making spatial predictions of future settlements less complex.

10.2.9 Scenarios tested

This component of the model is not fully implemented, yet. In order to show effects of several landscape management actions, several scenarios are to be simulated. It was decided to use 1) a scenario that 'continues business' as usual. Other scenarios deal with 2) very relaxed forest harvest levels, 3) unsustainable high harvest levels, and 4) with a landscape management regime that takes Grizzly Bear habitat and conservation needs fully into account. Based on the best information as possible, two Global Change scenarios are also to implemented: 5) increasing temperature, and 6) decreasing temperatures.

10.2.10 Landscape Indicators for Grizzly Bear habitats

This work is currently in progress. Besides local and expert knowledge, meaningful indexes for Grizzly Bear habitat are derived from Resource Selection Function (RSF) approaches (Nielsen et al. in press) and from overlaying GPS bear data with the IDT map, e.g. stratified by pre- and post-berry seasons and by Grizzly Bear gender and age. Future work might include landscape metrics of mean patch size, edge density and patch type diversity to quantify and trace the modelled future landscape and its use.

129

10.2.11 Oil and Gas Models

Oil and Gas exploitation plays a major role designing the landscape of the study area. Currently, information on local policy and business plans are requested, as well as geological maps used to predict where major activities will occur. 10.2.12 Model evaluation

Results from predictive future models cannot be ground-truthed or compared, other than from competing models which currently don't exist for the study area. Therefore, a statistical model evaluation in the classical sense (e.g. Pearce et al. 2001) is not possible. It is intended to use historical landscapes, e.g. derived from the 1950’s from Aerial Photo work (K. Montgomery unpublished) and see whether the current landscape can be predicted reliably. Other approaches are also assessed. This component of the project is receiving major attention in order to provide spatial predictions for the future with known certainty.

10.3 Results

The current results are a first start on the complex predictive modeling topic but show what information the results can provide. Currently, no alternative scenarios are run and modeling input data are still to be improved and updated, e .g roads. Findings presented in Fig 10.1 and 10.2 are not based on the mentioned ‘roll-over modelling’ approach (‘dynamic feedback loop’) and thus assume mostly natural death and the simple re-occurrence of species after the life spans is reached.

130

Figure 10.1. Future landscape scenarios (draft) : Age Clases 2000 and 2100 (natural tree death). Dark green 0 years, Yellow 170 years, red >320 years.

Figure 10.2. Future landscape scenarios (draft) : Predicted IDT classes 2000 , 2050 and 2100. Pink values indicate Forest Cover (based on natural death and re-grow).

10.4 Discussion

Predicting the future presents a human dream. The advent of computers, long-term ecological studies and free data over the WWW make such approaches more achievable.

Our modeling approach is spatial but does not take all spatial interactions fully into account. For instance, with the current software version of Woodstock and Stanley one cannot implement true seedling distances, or neighbouring effects from adjacent forest stands. In order to fully address such effects might require a cellular automaton, or even more complex models; more research is needed to evaluate the gain from such approaches.

Each sub-model, e.g. fire and insects, was treated as being independent. Such a model is not making use of all the synergies and interactions that occur in an ecosystem. More research is needed to address how relevant such considerations are for a realistic future model.

Here an approach and a method was presented to model future landscapes relevant to Grizzly Bear habitat. Obviously, this work presents a start for a complex modeling investigation. However, developing such tools is crucial to safeguard wildlife and their habitats long-term. The eventual goal of this study is to link future landscape scenarios with Grizzly Bear populations to derive a Population Viability Analysis PVA, (Nielsen et al. in progress) for sound and long-term conservation and management of wildlife (compare also Gustafson et al. 2001).

131

10.5 References

Agarwal, C., G.L. Green, M. Grove, T. Evans and C. Schweik. 2000. A review and assessment of land-use change models Dynamics of Space, Time and Human Choice. Center for the Study of Institutions, Population and Environmental Change, Indiana University, Bloomington. USA.

Alberta Environmental Protection Land and Forest Service Resource Data Division. 1985.

Alberta Phase 3 Forest Inventory: Volumes and Stem Numbers for Forest Types. Central Alberta: Volume Sampling Regions 4, 5 and 6. Edmonton. 90 pp.

Andison, D.W. 1998. Temporal patterns of age-class distributions on foothills landscapes in

Alberta. Ecography 21: 543-550. Burns, R.M. and B. H. Honkala. 1990. Silvics of North America, Volume 1 (Conifers), Volume2

(Hardwoods). Agriculture Handbook 654. 877 pp. Doyon, F. and P.D. Duiker. 2000. Fire regime simulation of the Whitecourt forest using

LANDIS.BAP Report#4. Prepared for Millar Western Forest Products's Biodiversity Assessment Project.

Franklin, S. E., G. B. Stenhouse, M. J. Hansen, C. C. Popplewell, J. A. Dechka, and D. R.

Peddle. 2001. An integrated decision tree approach (IDTA) to mapping landcover using satellite remote sensing in support of grizzly bear habitat analysis in the Alberta Yellowhead Ecosystem. Canadian Journal of Remote Sensing 27 (6):579-591.

Gilruth, P. T., S. E. Marsh and R. Itami. 1995. A dynamic spatial model of shifting cultivation in

the highlands of Guinea, West Africa. Ecological Modelling 79: 179-197 Gustafson, E.J., N. J. Murphy and T. R. Crow. 2001. Using a GIS model to assess terrestrial

salamander response to alternative forest management plans. Jounral of Environmental Management 63:281-292.

Huang, S. D. Morgan, G. Klappstein, J. Heidt, Y. Yang, and G. Greidans. 2001. A Growth and

Yield Projection System for Natural and Regenerated Stands within an Ecologycally based, Enhanced Forest Management Framework: Yield Tables for Seed-origin Natural and Regenerated Lodgepole Pine Stands. Forest Management Branch, Edmonoton Alberta, Canada. 193 pp.

Keane, R.E., P. Morgan and J. D. White. 1999. Temporal patterns of ecosystem processes on

simulated landscapes in Glacier National Park, Montana, USA. Landscape Ecology 14: 311-329.

Linke, J., S.E. Franklin, G.B. Stenhouse, and F. Huettmann. In Review. Grizzly Bear

Population Landscape Use and Structure in Alberta. Ursus.

132

Nielsen, S.E., M.S. Boyce, G.B. Stenhouse, R.H.M. Munro. 2002. Modeling grizzly bear habitats in the Yellowhead Ecosystem of Alberta: Taking autocorrelation seriously. Ursus, 13:533-542.

Pearce, J., S. Ferrier and D. Scotts. 2001. An evaluation of the predictive performance of

distributional models for flora and fauna in north-east New South Wales. Jounral of Environmental Management 62: 171-184.

REMSOFT 2002. Woodstock 2.0 and Stanley 4.5 Modeling Manuals + Software, Fredericton

NB, Canada. Smith, D.M., B. C. Larson, M.J. Kelty, P. Mark and S. Ashton. 1997. The practice of

silviculture: Applied Forest Ecology. Ninth Edition. J.Wiley and Son. 537 pp. Tang, S.W., J, F. Franklin and D. R. Montgomery. 1997. Forest harvest patterns and landscape

disturbance processes. Landscape Ecology 12(6): 349-363. Voinov, A., R. Constanza, L Wainger, R. Boumans, F. Villa, T. Maxwell and H. Voinov. 1999.

Patuxent landscape model: integrated ecological economic modeling of a watershed. Environmental Modelling & Software 14: 473-491

White, M. A., P. E. Thornton, and S. Running. 1999A continental phenology model for

monitoring vegetation responses to interannual climatic variability. Global Biochemical Cycles Vol 11: 217-234.

133

11.0 MAPPING AND QUANTIFICATION OF CHANGE IN LANDSCAPE STRUCTURE IN GRIZZLY BEAR HABITAT: PROGRESS REPORT

Kirk P. Montgomery, Department of Geography, University of Calgary, Canada [email protected]

11.1 Introduction

This report documents the research that I plan to accomplish within the FMF grizzly bear research project for the years 2003 – 2004 in addition to summarizing previous work. This research is also a critical component towards completing the requirements of the degree of Master of Science in Geography awarded by the University of Calgary.

The goal of this research is to provide the FMF grizzly bear research project valuable insight into landscape change within the study area and to provide the broader scientific community with new methods for the creation of multi-temporal mapping products derived from multiple data sources. What follows is a short discussion of the purpose and background to this research, a description of the data and methodologies used, and an early look at some of the results and deliverables.

11.2 Purpose and Background

Quantifying change in a landscape is critical in understanding the past and in predicting the future in terms of landscape structure and the FMF goal of quantifying grizzly bear habitat. Specifically, the objectives of this research are to:

i. Develop a method to reconstruct landscape structure in the past century.

ii. Determine the rate of change for the landscape. iii. Create map products that represent a time series of landscape change in the study

area, for example 1930, 1950, 1970, … today. iv. Explore linkages between alternative landscapes, landscape structure, and

management needs.

There are three areas of interest in this research; firstly, we know the study area has changed, but we don’t know the amount, rate or causes of this change. Determining the rate and cause of change will provide grizzly bear managers with the trends of past landscape use in the study area. It will allow them to predict with greater accuracy future scenarios of development. Secondly, we will be in a position to roughly estimate landscape use by bears at different times in history through collaboration with other research partners utilising Resource Selection Function (RSF) Modelling. Estimates of bear landscape use have been successful with current-day landscape structure (Popplewell, 2001; Linke, in review).

The third reason to undertake this research is to solve the methodological challenges involved in answering the objectives so that the methodologies can be applied generally in other landscape research.

134


The simulation of historical landscapes based on air photos, remotely sensed imagery, and GIS data layers are a complicated yet worthy task. This research will provide techniques which are valid for the simulation of historical landscapes. The next section of this report describes the sources and types of data required to carry out this research as well as a generalized methodology that will be used to create the final deliverables.

11.3 Data and Methods

There are several different types of data required for this research concerning historical landscapes. This project is ambitious because it tries to replicate the types of data available today, but as if they were acquired within the last fifty years. Data not acquired specifically for use in this research will be supplied from one of these sources:

• Foothills Model Forest • Other participants in the research project • The University of Calgary Library • Resource extraction industries in the study area

Generally, there are two classes of data that will be used in order to accurately model historical landscape structure; spatial data and ground data. Spatial data refers to data such as aerial photographs, remotely sensed imagery, and GIS data layers. Each is described in more detail below.

11.3.1 Spatial Data

11.3.1.1 Airborne Data Airborne data circa 1950 was acquired for the years before remotely sensed imagery was widely available (before 1972). These aerial photographs were orthorectified and mosaiced to form a seamless image for two bear management units (the Gregg and the Cardinal) within the more broad study area. The next phase of the research involves taking these images and interpreting them in conjunction with a GIS to produce thematic coverages that represent the features of interest found on the air photo mosaic. Once this is complete, a relationship will be established between elements of the air photo mosaic and field data. This is explained further under the Ground Data section below. 11.3.1.2 Spaceborne Data Remotely sensed data offer many advantages over conventional data sources. It allows us the ability to image large geographic areas in their entirety and evaluate landscape patterns and the aerial extent of resources in a temporally and spatially comprehensive way (Kepner et al. 2000). It also affords us the ability to obtain measurements in areas or at times that are not practical. Table 11.1 below documents the specific types of remotely sensed data acquired for this research.

135

Table 11.1. Remotely sensed data used in analysis

SENSOR Image Date Ground Resolution (m)

Aerial Photographs 1948-1952 15 CORONA August 29,1963 90 LANDSAT MSS August 04, 1974 80 LANDSAT TM July 31, 1985 30 LANDSAT TM September 02, 1988 30

Remotely sensed imagery make an ideal data source because the FMF grizzly bear project study area is large. However, remotely sensed imagery is not generally available before 1972 when the first Landsat satellite was launched and became operational. Prior to the Landsat series of satellites, the United States government operated the top-secret surveillance satellite ‘Corona,’ from 1960 through 1972 (Ruffner, 2001). Two 1967 Corona images were acquired in order to aid analyses. These images have an approximate ground resolution of about 90 metres making them difficult to use in order to find subtle small scale changes in the landscape but are adequate for use in detecting large scale features such as cut blocks.

After 1972, the LANDSAT series of satellites provides the needed spatial and temporal coverage of the study area in order to produce accurate change detection maps. Figure 11.1 shows the various image types that will be used in this analysis with aerial photography and CORONA data being greyscale and the LANDSAT images being false colour. All of these images have been orthorectified and geocorrected followed by a correction for the effect of the Earth’s atmosphere.

Once this has been completed variables such as greenness (an indicator of plant biomass or the amount of green vegetation present at a location) can be easily produced. Greenness maps are required by the FMF grizzly bear research project because greenness has been demonstrated as being a good predictor of bear habitat use (Mace et al., 1999; Boulanger, 2001).

Additionally, I will be creating habitat classification maps for historical landscapes from this image data set. These maps will follow and resemble the modern day results of the classification methodology developed for the creation of the integrated decision tree (IDT) classification maps (Franklin et al., 2001). Figure 11.2 above shows potential change over time between 1949 and 2000.

11.3.2 Geographic Information Systems

The forestry industry uses geographic information systems (GIS) data layers to plan and account for current and future resource extraction activities. The data provided by them regarding their future cut plans has been attained in order to simulate future landscapes. These GIS coverages include roads and trails, fire history, and cut history as examples.

136

11.3.3 Ground Data

Ground data refers to information collected ‘in the field.’ Two previous field seasons worth of data have been completed and will be a valuable resource in evaluating landscape change over time. This data set includes field plot data from field sampling sites used in the creation of the IDT map products. Field data recorded include a general description of the site, the dominant form and type of vegetation, and tree characteristics such as diameter at breast height (dbh).

In addition to this data from previous field seasons, the summer of 2002 provided additional valuable information needed for research activities. These parameters include leaf area index (LAI) measured through the use of a ceptometer and a LAI-2000, and stand properties such as crown closure, tree height, stand age, and dbh. With this ground data, a relationship will be determined based on collected ground data and forest parameters from the spatial data. For example, the prediction of greenness will be made based on relationships between the forest structure parameters (LAI, dbh, etc.) and aerial photo tone, texture, and ground cover.

Figure 11.1. Simulated IDT.

137

Figure 11.2. Change through time

11.4 Summary

The research to be completed and described here will aid resource managers in their ability to understand landscape changes in the FMF grizzly bear study area over the past 50 years. Deliverables include (1) a methodology for determining past landscapes based on historical aerial photos and which then resemble satellite image products of more recent vintage, (2) a series of map products documenting previous landscapes and showing landscape change up to the present including greenness maps and classification maps, and (3) an estimation of landscape use for alternative landscapes by grizzly bears. The data compiled will also serve as baseline data for future modelling scenarios of landscape change in addition to being inputs in RSF modelling activities. The projected completion date for this research is April 2003.

138

11.5 References

Boulanger, J. 2001. A preliminary comparison of grizzly bear habitat mapping techniques. unpublished report.

Franklin, S.E., Stenhouse, G.B., Hansen, M.J., Popplewell, C.C., Dechka, J.A., and Peddle, D.R.

2001. An Integrated Decision Tree Approach (IDTA) to Mapping Landcover Using Satellite remote Sensing in Support of Grizzly Bear Habitat Analysis in the Alberta Yellowhead Ecosystem. Canadian Journal of Remote Sensing, 27: 579 – 592.

Kepner, W., Watts, C., Edmonds, C., Maingi, J., Marsh, S., and Luna G. 2000. A landscape

approach for detecting and evaluating change in a semi-arid environment. Environment Monitoring and Assessment, 64: 179 – 195.

Linke, J., S.E. Franklin, G.B. Stenhouse, and F. Huettmann. In Review.Grizzly Bear Population

Landscape Use and Structure in Alberta. Submitted to Ursus. Mace, R.D., Waller, J.S., Manley, T.L., Ake, K., and Wittinger, W.T. 1999. Landscape

evaluation of grizzly bear habitat in western Montana. Conservation Biology, 13:367-377. Popplewell, C. 2001. Habitat Structure and Fragmentation of Grizzly Bear Management Units

and Home Ranges in the Alberta Yellowhead Ecosystem. unpublished MSc thesis. Ruffner, K.C. 2001. CORONA and the Intelligence Community.

http://www.cia.gov/csi/studies/96unclass/corona.htm (January 30, 2003).

139

12.0 2002 REMOTE SENSING ACTIVITIES: PROGRESS REPORT Greg McDermid, Department of Geography, University of Calagary, Canada [email protected]

12.1 Introduction

This report briefly summarizes the remote sensing activities conducted in support of the Foothills Model Forest Grizzly Bear Project during the calendar year 2002. Activities are divided into five categories: (i) radiometric correction, (ii) greenness maps, (iii) 1999-2001 change detection, (iv) 2001 IDT map, and (v) 2002 image acquisition. Progress in each category is described briefly in the following sections.

12.2 Radiometric Correction

First-cut greenness map products used by researchers in the grizzly bear project exhibited unexpected patterns caused by atmospheric and illumination conditions. For example, the overall greenness of a June 1999 image was higher than that in the subsequent September scene; the opposite of what we might expect out of an index that is highly correlated to the phenology of green vegetation. Users also noted unexplained temporal variability in ‘pseudo-invariant’ features such as rocks and clear lakes – targets that should display little or no change over time. These observations suggested issues related to the preparation of these products, and caused us to re-evaluate our radiometric processing routines.

Radiometric processing, in the form of atmospheric and/or topographic correction is a common methodological step in the processing of remote sensing imagery. The electromagnetic radiation signals collected by satellites are modified by gases and aerosols in the Earth’s atmosphere in complex and often contradictory ways. In certain circumstances – such as multi-scene project that call for a common radiometric scale – the effects of atmospheric attenuation must be accounted for (Song et al., 1999).

Confounding the radiometric issue is the lack of a common, reliable method for performing atmospheric correction. Much of the literature on the subject is concerned with absolute calibration of individual images, whereby sensor characteristics, atmospheric conditions, and illumination angles are modeled explicitly (e.g. Vermonte and Kaufman, 1995). However, the detailed atmospheric observations required to drive these models are rarely available, leading most users to rely on standard atmosphere parameterization from commercial image processing packages. Unfortunately this is often an unsatisfactory approach; commonly producing poor or unexpected results (Cohen et al., 2001).

A group of alternative solutions – known collectively relative calibration – are procedures whereby individual scenes are normalized with respect to each other, using little or no ancillary data (Hall et al., 1991; Narayana et al., 1995).

140


The techniques vary widely, and can be considered to include anniversary dates (Dobson et al., 1999), histogram matching (Homer et al., 1997), dark object subtraction (Chavez, 1988), and linear transformation (McGovern, et al., 2002). To date no single ‘best’ calibration procedure has emerged, likely due to the wide range of ground and atmospheric conditions encountered in practice.

A series of tests were performed to determine the most effective radiometric processing procedure for our purposes. Three candidate procedures were examined: (i) absolute atmospheric correction using PCI’s ATCORR procedure, (ii) relative normalization using regression of pseudo-invariant features, and (iii) raw digital numbers, with no normalization. Figure 12.1 summarizes the RMS errors observed from a sample of pseudo invariant features (clear lakes, flat asphalt and gravel surfaces, etc.) drawn from seven Landsat scenes in the project archive, relative to values from a September 2001 ‘master’ image. Low RMS errors would indicate low temporal variability and, therefore, successful radiometric correction.

0

20

40

60

80

100

120

Aug29/98

Jun4/99

Sep8/99

Oct19/99

Aug17/00

Sep27/00

Jul3/01

Image Date

RM

SE

RawAtmospheric CorrectionRadiometric Normalization

Figure 12.1. RMS errors illustrate the variability of pseudo-invariant features observed across seven Landsat scenes using three strategies for radiometric correction.

The raw (uncorrected) images generated a mean RMS error of 51.3. Atmospheric correction with ATCORR – the methodology used in generating the original remote sensing products – produced only a slight improvement of 37.0. The radiometric normalization routine produced the smallest RMS errors by far: with a mean of 1.7. As a result, we concluded that the original pre-processing methodology using absolute atmospheric correction produces poor results, and should be abandoned in favour of the new relative normalization procedure. All subsequent products described in this document reflect the adoption of these new procedures.

141

12.3 Greenness Maps

‘Greenness’ is one of the products of the tasseled cap transformation of Landsat TM or ETM+ imagery (Crist and Cicone, 1984). Previous research has shown that this and the other tasseled cap indices – brightness and wetness – can provide a useful means of summarizing the information content of a satellite image into a small number of discrete variables that are highly related to key vegetation attributes such as species, age, and structure (Cohen et al., 1995).

The tasseled cap transformation was performed on each of the images in the grizzly bear archive (Table 12.1). In addition, a ‘post-berry’ greenness mosaic (Figure 12.2) composed of imagery from September 8, 1999 and September 14, 2001 was constructed.

Figure 12.2. Late summer greenness mosaic, composed of imagery from September 8, 1999 and September 14, 2001.

142

Table 12.1. FMF Grizzly Bear Project Image Archive. Sensor Path Row Date Description TM 44 23 8/29/98 Perfect

TM 45 23 4/6/99 Cumulous clouds over east, haze over northwest, belts of

stratus along front range

TM 45 23 9/8/99 Thin stratus in north.

TM 44 23 10/19/99 Thin stratus in north. Snow in mountains. Medicine Lake empty

ETM+ 45 23 8/17/00 Stratus over northeast. Thin cumulous over front range

ETM+ 44 23 9/27/00 Thin stratus over western and middle

ETM+ 45 23 7/3/01 Scattered puffball cumulous

ETM+ 44 23 9/14/01 Perfect

ETM+ 44 23 6/13/02 Cumulous in north and south. Central OK. ETM+ 45 23 8/23/02 Perfect

12.4 1999-2001 Change Detection and 2001 IDT Map

Change detection between satellite orthomosaics from 2001 and 1999 was performed using the methodology developed previously by Franklin and Hansen and reported in detain in a prior FMF Technical Report. Summarized briefly, the technique uses an enhanced wetness difference threshold to derive a raw change image. The change pixels are then subjected to a series of GIS decision rules to classify pixels into various categories of change. Unfortunately, the technique does not reliably account for all the new roads in the study area, due to the 30-meter pixel size of the imagery. As a result, roads changes were digitized manually.

The 1999-2001 change detection revealed 284km of new roads, 296ha of mine expansion, 128 new well sites, and 6012ha of new cutblocks. These change features were incorporated into the existing IDT map to create a new 2001 landcover product. The resulting 2001 IDT landcover map is shown in Figure 12.3.

143

Figure 12.3: 2001 IDT map, incorporating the results of the 1999-2001 change detection analysis.

144

12.5 2002 Imagery

Two Landsat ETM+ scenes were acquired in support of future remote sensing activities in the present study area: one scene from June 13 (path 45, row 23), and a second from August 28 (path 44, row 23). An uncontrolled mosaic of the two images is shown in Figure 12.4. Geometric and radiometric pre-processing of the new imagery is presently underway.

Figure 12.4: Uncorrected mosaic of 2002 imagery. The scene on the left is August 23; the scene on the right is June 13.

145

12.6 References

Chavez Jr., P. S., 1988: An improved dark-object subtraction technique for atmospheric scattering correction of multispectral data; Remote Sensing of Environment, Vol. 24, pp. 459-479.

Cohen, W. B., T. A. Spies, and M. Fiorella, 1995: Estimating the age and structure of forests in a

multi-ownership landscape of western Oregon, U.S.A., International Journal of Remote Sensing, Col. 16, pp. 721-746.

Cohen, W. B., T. K. Maiersperger, T. A. Spies, and D. R. Otter, 2001: Modelling forest cover

attributes as continuous variables in a regional context with Thematic Mapper data; International Journal of Remote Sensing, Vol. 22, No. 12, pp. 2279-2310.

Dobson, J. E., E. A. Bright, R. L. Ferguson, D. W. Field, L. L. Wood, K. D. Haddad, H., J. R.

Jensen, V. V. Klemas , R. J. Orth, and J. P. Thomas, 1999: NOAA Coastal Change Analysis Program (C-CAP): Guidance for Regional Implementation. NOAA Technical Report NMFS 123, U.S. Department of Commerce.

Hall, F. G., D. E. Strebel, J. E. Nickeson, and S. J. Goetz, 1991: Radiometric rectification:

Toward a common radiometric response among multidate, multisensor images; Remote Sensing of Environment, Vol. 35, pp. 11-27.

Homer, C., R. Ramsey, T. C. Edwards, Jr., and A. Falconer, 1997: Landscape cover-type

modelling using a multi-scene Thematic Mapper mosaic; Photogrammetric Engineering and Remote Sensing, Vol. 63, pp. 59-67.

Narayana, A., H. U. Solanki, B. G. Krishna, and A. Narain, 1995: Geometric correction and

radiometric normalization of NOAA AVHRR data for fisheries applications; International Journal of Remote Sensing, Vol. 16, pp. 765-771.

Song, C., C. E. Woodcock, K. C. Seto, M. P. Lenney, and S. A. Macomber, 1999: Classification

and change detection using Landsat TM data: When and how to correct atmospheric effects? Remote Sensing of Environment, Col. 75, pp. 230-244.

Vermonte, E. and Y. J. Kaufman, 1995: Absolute calibration of AVHRR visible and near-

infrared channels using ocean and cloud views; International Journal of Remote Sensing, Vol. 16, pp. 2317-2340.

146

13.0 APPLICATION OF SCAT DETECTION DOGS TO GRIZZLY AND BLACK BEAR MONITORING IN THE YELLOWHEAD ECOSYSTEM, ALBERTA, CANADA Samuel K. Wasser, Center for Conservation Biology, Department of Biology, Box 351800, University of Washington, Seattle, WA 98195, USA, ph: 206-543-1669, FAX: 206-543-03410; email for Wasser: [email protected]. Gordon Stenhouse, Foothills Model Forest, Box 6330, Hinton, Alberta, CANADA T7V 1X6.

13.1 Introduction

We field-tested the scat detection dog method, surveying for grizzly and black bear scat in a 5,400 km2 study area in and around Jasper National Park, Alberta, Canada, during a 6-week period in 1999 (mid-May-July) and in a smaller 1, 000-km2 area during an 8-week period in 2001 and 2002 (July-August). Approximately 43% of the 1999 study area was within the park, whereas the remaining 57% was within a multi-use area to the east, exposed to a wide variety of land use activities (e.g., forestry, mining, oil and gas development and exploration, transportation corridors, trapping, tourism, hunting, commercial outfitting, and public recreational use). One hundred percent of the 2001-2002 study area was in the multiple-use area outside the park. The elevation and degree of natural habitat fragmentation is significantly higher in the south versus north of the park, whereas tourism is most heavily concentrated in the north of the park. The elevation does not change between the north and south, outside the park. However, the north part of the multi-use area is characterized by heavy resource extraction and is thus the most disturbed portion of the entire study area. The southern portion of the multi-use areas is predominantly mature lodgepole pine, and relatively low levels of human use at this time.

13.2 Methods

During 1999, 40, 9 x 9 km grid cells, evenly distributed over the entire 5400 km2 study area, were each searched 3 times at 2-week intervals. Grizzly bear population size estimates, conducted using DNA hair snagging techniques, over the 5400 km2 study area were revealed to be approximately 100 individuals in 1999. (Stenhouse and Munro 1999). To better accommodate the low population estimate, grid cell sizes were reduced to 5 x 5 km and the number of scat search sessions per cell increased to 5 times each during 2001. Hair snag sampling occurred in 1999 only. Scat sampling during 2001 was also restricted to three areas (12 cells per area) in the multi-use area that differed markedly in amount of human disturbances and based on 1999 data, the number of bears in each area. A different dog team was assigned to each cell every time the cell was searched to control for between-dog differences in sample detection rates. All transects were conducted with the dog off-leash, always remaining in-sight of its handler, maximizing the area covered by each dog. Dogs were also trained to not chase wildlife and wore a bear bell to provide wildlife advance warning of our presence. This proved highly effective as no bears were every seen by any field crew member during scat-dog searches. Each dog team walked for ~ 7 hrs, covering a 5-9 km transect.

147


In 1999, transects extended outward from 1 km of the hair snag in the higher quality grizzly (from existing CEA model habitat quality mapping) bear habitat per cell. In 2001-2002, the five transects per cell were spaced 2 weeks apart and also occurred in a different location per transect. As with the positioning of hair snag sites, scat dog transects were always walked in areas where bears were presumed most likely to occur.

13.2.1 Back-Tracking Radio-collared Grizzly Bears

A second form of sampling was also developed, relying on location data from GPS radio collared grizzly bears to enable detection dogs to collect repeated scat samples from known individuals whose home range, and hence disturbances encountered, over the sampling period were also known. Back-tracking data from radio collared grizzly bears were collected in 2001-2002 by up-loading into our hand-held GPS units the previous 2 weeks of 4 hr interval telemetry points from an individual grizzly bear’s GPS collar. A dog team was then positioned at each end of a portion of that bear’s 2 week route. These two teams walked from one waypoint to the next until they met in the middle, collecting all scats encountered along the way. 13.2.2 Scat collection and processing

Scats of all ages were collected. Scat samples were divided into 4 age classes in the field based on moisture, color and odor strength: 0-1 days, ≤2 days, ≤2 weeks, ≤1 month, and > 1 month. Sample exposure and contents were also recorded. At the time of collection, the location of each scat was recorded using a hand held GPS unit (Garmin Pro II, Loma Linda, CA). This enabled us to layer all DNA data derived from the sample on to a Geographic Information System (GIS) that also includes data sets on habitat quality and human disturbance features within the study area (Stenhouse & Munro, 1999). We noted the habitat type, weather, wind speed, whether the sample was detected by the dog, with or without handler assistance, and whether the sample was found on a road, seismic line, hiking or game trail or in vegetation. DNA is unevenly distributed in scat (Wasser et al., 1997). Hence, all individual scat samples were thoroughly mixed with a gloved hand, on location, prior to removing a subsample for subsequent analyses. In 1999, ~ 30 g of scat was loosely wrapped in a coffee filter and placed in a zip-loc bag containing silica as a preservative, at a ratio of 4g silica per 1g of scat (Wasser et al., 1997). Samples were stored in a freezer at the end of each field day and remained there for 0-8 weeks until transported to our laboratory for extraction and analyses. The preservation method was changed in 2001 to improve DNA amplification success. Approximately 15 g of the well-mixed sample was placed in a screw-top vial containing 2.5 mls 90% ethanol per g feces (Wasser et al., 1997; Murphy et al., in press). 13.2.3 DNA Extraction and Amplification

All scat samples were freeze-dried, sifted through a steel mesh colander, and thoroughly mixed. Fecal DNA is extracted based on modifications of Wasser et al (1997). Briefly, 1600 µl of Qiagen ASL Buffer (provided in the QIAmp stool kit) is added to 200 mg of well-mixed, freeze-dried feces. Samples are vortexed (1 min), incubated (1 hr at 70°C) and then centrifuged (13,000 rpm for 3 min).

148

All supernatant is transferred to a new tube containing inhibitex tablets for removal of PCR inhibitors (provided in kit), vortexed briefly, incubated (1 min at rm temp) and centrifuged (13,000 rpm for 3 min). All supernatant is transferred to a clean tube and centrifuged. Six hundred µl of the supernatant is transferred to a fresh tube containing 25 µl of Proteinase K, 600 µl AL buffer (provided in kit) are added, vortexed briefly and incubated (1 hr at 70°C). After adding 600 µl of ethanol, all lysates are mixed by inverting tubes several times and loaded in 2-3 centrifugations onto the spin columns, washed with 500 µl AW1 (provided in kit), centrifuged (20 sec), washed with 500 µl AW2 (provided), centrifuged (1 min) and the DNA eluted in 200 µl of Qiagen AE buffer after a 1 min incubation. Samples are then purified using the Geneclean III®non-spin Kit (Q-BIOgene Inc., Carlsbad , CA) at 55°C.

A portion of the control region of mtDNA is PCR amplified using fluorescent primers, analyzed on an ABI 3100 Genetic Analyzer (Perkins Elmer Applied Biosystems, Foster City, CA), and compared to known black and brown bear control samples, interspersed throughout the same run, to ascertain the taxonomic identity of the unknown fecal samples (Paetkau & Strobeck, 1996; Woods et al., 1999; Clarke, 2001). The samples are set up in 20 µl volumes containing 1.5mM MgCl2, 0.25 µg/µl BSA, 0.2uM dNTPs, 0.13uM 6-FAM labeled LTRPROB13, 0.13uM HSF21 (Wasser et al., 1997), 0.65U Taq DNA Polymerase (Promega, Madison, WI) and 2.0ul of the fecal DNA extract. The reactions are cycled in a 9700 Thermocycler (Perkins Elmer Applied Biosystems) once for denaturation at 94°C for 2 minutes, annealing at 53°C for 30 sec and extension at 72°C for 30 sec, and then repeated 34 times with the denaturation time shortened to 30 sec at 92°C. A final elongation for 3 min at 72°C completes the amplification. PCR reactions are diluted 1:80 and 1 µl of the dilution is added to 10 µl of deionized formamide containing 0.0129% GeneScan TAMRA-400HD ROX (Perkin Elmer-Applied Biosystems). After a 5 min denaturation at 95°C, the solution is snap cooled in stratagene benchtop cooler at –20°C for 5 min, and then subjected to capillary electrophoresis with a 5 sec injection time on an ABI3100. Data are initially analyzed on GeneScan software (Perkin Elmer-Applied Biosystems), after which allele sizes are called by GENOTYPER software (Perkin Elmer-Applied Biosystems).

Upon confirmation of species identity, the grizzly bear scat samples were profiled for individual identities using 6 µsatDNA loci, G01A, G10B, G10C, G01D, G10J, G10M (Paetkau & Strobeck, 1994; Paetkau et al., 1995). The decision to use 6 µsat loci was based on: (a) the Psib < 0.05 criterion requires > 4 loci in grizzly and black bears (Woods et al. 1999); and (b) analyses of multiple grizzly bear populations by D. Paetkau (personal communication) revealed that samples examined at ≥ 6 loci, which do not amplify for at least 5 of these loci, tend to have an excess of homozygotes resulting from allelic drop-out. Thus, the above criteria are a means of decreasing the likelihood of allelic drop-out in the analyzed samples. Heterozygosities for these loci ranges from high (G10B; He=0.81) to moderate (all others, 0.68 < He < 0.76). Samples were also characterized for gender using primers SRY41F, SRY121R (Taberlet et al., 1993), and ZFX/ZFY (Woods et al., 1999) in 1999 and CCB-01 forward and SE47/48 (bovine) amelogenin primers in 2001 (Ennis & Gallagher, 1994). The amelogenin locus has the advantage over the ZFX/ZFY-SRY combination in that the former reaction amplifies both male and female bands in the same PCR reaction, reducing the likelihood of false positive female identifications.

149

Microsatellite primers are modified with 5’ fluorescent tags and DNA amplified using a modified protocol by Paetkau (unpublished): 20 µl volumes contain 25mM MgCl2, 10 mg/µl BSA, 10 mM dNTPs, 20µM each of forward reverse primers (Paetkau & Strobeck, 1994; Paetkau et al., 1995), 5U/µL Taq DNA Polymerase, TaqStart antibody and 4X TaqStart buffer (Promega, Madison, WI) and 2.0ul of the fecal DNA extract. The reactions are cycled in a 9700 Thermocycler (Perkins Elmer Applied Biosystems) once for denaturation at 94°C for 60 sec, followed by 3 cycles at 94°C for 60 sec, annealing at 54°C for 20 sec and extension at 72°C for 5 sec, then repeated 33 cycles at 94°C for 15 sec, annealing at 54°C for 20 sec and extension at 71°C for 1 sec. A final elongation for 30 sec min at 72°C completes the amplification. One µl of the PCR reaction is added to 10 ul of deionized formamide containing 0.0129% GeneScan TAMRA-400HD ROX (Perkin Elmer-Applied Biosystems) and then processed as described above for mtDNA. Known grizzly bear serum samples (n=19) from radio collared individuals are interspersed throughout the runs, along with mock extraction blanks and negative DNA controls, to ensure correct identification of alleles and the detection of any contamination, respectively. All samples are extracted in duplicate with each extract PCR amplified at least twice for nuclear DNA to increase the likelihood of detecting alleles subject to allelic drop-out. The multiple PCR strategy is similar to that recommended for hair by Tablerlet et al (1996). However, adding a second extraction is better suited to the uneven distribution of DNA in feces (Wasser et al., 1997; Wasser, ). Microsatellite loci were also analyzed to assure that they conform to Hardy-Weinberg Equilibrium and hence that there is not an excess of homozygotes at each locus (Taberlet et al., 1999). Similar analyses were conducted to detect potential null alleles (Paetkau & Strobeck, 1995). Once the microsatellite and gender analyses were complete, all grizzly bear data were independently examined for individual uniqueness, as well as the number of allele differences between unique individuals by D. Paetkau (Wildlife Genetics International, Nelson, BC). Any two samples that differed by only 1 or 2 alleles, across all loci, were re-analyzed (i.e, PCR repeated and/or re-examined via GENOTYPER) at the discrepant loci to determine if the differences were due to technical error (i.e., allelic drop-out, failed PCR, or a miscalled allele), and hence whether the samples actually represented the same individual. As a final cross-check, samples collected in the same cell and sampling session were compared. Those samples that still differed at only one locus (i.e., one sample was heterozygous and the other sample homozygous for one of the heterozygous alleles at that locus) were assumed to differ because of unconfirmed allelic drop-out and, therefore, called as the same individual. Any sample that could not be confirmed in these ways, at a minimum of 6 out of 7 loci (including gender) was discarded. These combined methods were devised to produce the most conservative estimate of the number of unique individuals in the population.

150

13.3 Results

13.3.1 Scat sample age and amplification success

Scat detection dogs collected 400 scats in 1999, 480 scats in 2001 and 750 scats in 2002. We estimated the age (i.e., time since defecation) of each scat to determine its impact on DNA amplification success. Scat age proved difficult to accurately ascertain in our first year of study, causing our aging criteria to be constantly revised. For 2001-2002, sample ages were determined separately for data collected during cell transects and backtracking of individual bears. Sixty two percent of transect samples were estimated to be ≤ 2 weeks old and 85% ≤ 1 month old. For the back-tracking surveys, 78% of samples were estimated to be ≤2 weeks old and 94% ≤ 1 month old. Sample age was also significantly correlated with DNA amplification success (r=0.132, p <0.05), dropping off sharply in samples > 2 weeks of age. These DNA amplification impacts were further examined using predictors of sample age: Amplification success declined with scat odor strength (r=0.216, p < 0.0008); moisture (ANOVA, F=3.74, p <0.003); and presence of mold (ANOVA, F=8.64, p <0.004). In 1999, approximately 65% of scats were successfully amplified for mtDNA and scnDNA used to determine species and gender, respectively, and only 40% of scats successfully amplified for microsatellite DNA. We suspected that DNA amplification success from scat was compromised by improper freezing of silica preserved samples that year. Scats were stored in a walk-in freezer, also used to store beaver carcasses used for grizzly bear trapping by capture crews. The freezer was frequently being opened and undoubtedly took a long time to re-cool because of its large storage space. These problems were corrected with a dedicated freezer and ethanol preservation (see methods) in 2001. Amplification success of mtDNA increased to 93%, scnDNA for gender to 81%, and microsatellite DNA to 73% in 2001. Samples from 2002 are still being analyzed. 13.3.2 Scat Sampling Results

The detection dog method collected 2.2 black bear per grizzly bear scat samples in 1999 compared to 0.7 black bear per grizzly bear in 2001. These between year differences appear to result from differences in the areas sampled over these two periods. Sampling occurred inside and outside the park in 1999. Sampling occurred outside the park only in 2001. Black bear scats in 1999 were more commonly detected inside than outside the park, whereas grizzly bear scat collections showed the reverse pattern (F=60.26, p < 0.0001 for species; F=6.85, p < 0.01 for inside versus outside the park, ANOVA). The higher percentage of grizzly bear relative to black bear scat collected in 2001 is thus consistent with our exclusive sampling outside the park during that year.

Individual identities could not be determined in 1999 owing to poor DNA amplification success (see above). Microsatellite DNA and gender analyses were conducted for grizzly bear scat only in 2001, revealing 43 unique individuals over the 1, 000-km2 study area (25 unique individuals detected during transects and an additional 2 unique individuals detected during backtracking surveys). The sex ratio of these 43 individuals was 1.79 females per male.

151

The number of samples collected per grid cell was highly correlated with the number of unique individuals identified in that cell (r=0.84, p < 0.0001), and the ratio of number of individuals per number of samples in each cell was highest for the cells in the central portion of the study area. During transect surveys, an average ratio of 0.71 different individuals was detected per sample collected, based only on samples that amplified well enough to ascertain individual identity. The mean number of different individuals detected per cell, across all 5 sessions, was 2.87. As expected, the above ratio was substantially lower (0.18) during backtracking surveys.

The distribution of ursids also differed significantly from the north to south. In 1999, there was minimal species overlap inside the park. Black bear scat samples collected inside the park were concentrated in the northern portion and to a lesser degree in the central part of the park. Very few black bear samples were found in the more mountainous southern portion of the park. By contrast, no grizzly bears were found in the northern-most part of the park. Grizzly bear scat samples collected inside the park were most heavily concentrated in the more mountainous central and southern portions of the park.

A somewhat different pattern was found in the multi-use area outside the park, consistent with both 1999 and 2001 data. There was exceedingly high species overlap in the northern portion of the multi-use area, which has the highest level of human use within the entire study area. That area is characterized by forestry clearcuts, high density of all-weather roads, high levels of human activity, and two large coal mines located to the north west of the town of Cadomin. Both black and grizzly bears were most abundant in the northern portion of the multi-use area in 1999, with the number of each species progressively declining to the south. Moreover, the highest concentration of bears in the north were found along major roads. The distribution of black bear samples was also similar in 2001. However, the number of grizzly bear samples and the number of unique individuals was highest in the western-Central portion of the study area in 2001. Twenty of the 31 unique grizzly bears identified during transects in 2001 were detected in the central portion of the multi-use study area compared to 10 unique grizzly bears in the north and only one unique individual in the south. Indeed, very few black or grizzly bear scat samples were collected in the large lodgepole pine dominated forest of the central to southern portion of the multi-use area, south of Cadomin, in 1999 or 2001, with the exception of the area close to the primary road on the eastern border of the study area.

13.3.3 Bear distributions based on hair collection and radio-telemetry methods:

Hair samples were concurrently collected over the same study area during 1999 in a separate study conducted by Mowat and Strobeck (see Mowat, 2000; Woods et al., 1999 for details). In 1999 scat detection dog and hair snag methods collected comparable numbers of samples (~400 each), despite 50% more cells being sampled for hair than were sampled for scat by dog teams (i.e., 64 versus 40 grid cells, respectively). The hair sampling detected 41 unique grizzly bears over the 5,400 km2 study area in 1999 (7.6 individuals/1,000 km2). Mark-recapture analyses from these hair DNA samples estimated the number of grizzly bears in this population to be 14.7/1,000 km2. Using the scat detection method we detected 25 individuals (Mark-recapture analyses on our 2001 scat data are still pending).

152

Part of the hair-scat difference likely results from the fact that detection dogs physically cover a larger portion of the cell in search of a sedentary object, while hair snags are sedentary, luring in bears from considerable distances by olfactory means. One outcome of this difference is that multiple hits per hair snag in a single sampling session rarely detect more than one individual, compared to a 71% chance of detecting a different individual with each new scat sample collected in a single session in that same grid cell (see above). This difference becomes even more extreme considering that hair snag lures remain in place for 10 days per session, compared to ≤ 7 hrs that the dog team spent in that same cell during that session. Despite the above differences, it is noteworthy that the respective distributions over the landscape of grizzly and black bear samples collected by these two methods were quite similar to one another in 1999.

The distribution of grizzly bear scat was also compared to concurrent GPS radio-collar data from 19 grizzly bears, distributed throughout the 5,400 km2 study area, providing their Universal Transverse Mercator (UTM) locations every 4 hrs in 1999 and 2001. The radio-collared grizzly bear distributions were quite similar to the distribution of grizzly bear scat detected in both 1999 and 2001. Eight of the unique individuals detected from scat in 2001 were matched to previously radio collared bears. However, only two of those individuals were radio collared in that same year. On the other hand, all scat detected from radio collared bears were found in areas consistent with telemetry results in the years they were collared. This type of data is important when researchers are trying to determine survival rates from radio telemetry data only.

13.3.4 Between-Dog Team Differences

The total number of scat samples detected/hr varied markedly across dogs, ranging from 0.34-1.12 scats per hr (mean =0.75±0.17) in 1999 and 0.45-1.11 scats per hr (mean = 0.76±.12) in 2001. The black to grizzly bear ratio of samples detected also differed between dogs, ranging from 0.63-3.76 (mean=2.27±0.73) in 1999 and 0.35-1.89 (mean=0.99±0.31) in 2001. The sex ratios (females per male) of individuals identified from scat in 2001 ranged across dogs from 0.33-2.69 (mean=1.72±0.44).

13.4 Discussion

The sample collection abilities of scat detection dogs lends considerable strength to noninvasive genetic approaches to wildlife management. Scat detection dogs provide comparatively high scat sample acquisition. Sampling appears to be reliable based on the correspondence between radio-collar telemetry data and scat distribution data of grizzly bears, as well as the close correspondence between number of grizzly bear samples collected and number of unique grizzly bears detected per grid cell. Scat detection dogs can be utilized more opportunistically than other methods because the method does not require any prior set-up. However this method still requires careful planning to maximize the efficiency of the dogs. This also makes scat detection dog methods ideal for population and trend monitoring on some predetermined basis as well as for cross-sectional monitoring of large, new areas.

Microsatellite results suggest that multiple scat samples are more likely to represent multiple individuals compared to hair samples (see also Woods et al. 1999). This was expected since scat were collected over a 5-7 km transect compared to hair collected at a single location in that same grid cell and sampling period.

153

As is the case for hair collections, factors such as age/sex differences in home range size could impact scat encounter rates by dogs and may need to be adjusted by the models used to estimate population size from such data. Differences in sampling efficiencies between dogs is another potential source of bias from the detection dog method. However, those impacts can readily be accounted for by randomizing the cells searched by all dogs on repeat visits.

Since handlers necessarily always work the same dog, it was impossible to determine whether these between-dog team differences were the result of the handler, the dog, or the areas sampled. On the other hand, experience training over 300 narcotics detection dog/handler teams (B. Davenport, unpublished data), and subjective observations in our study, suggest that between-dog team differences due to errors (missed samples) are most often due to handler error (e.g., moving the dog off the sample without realizing that the dog had detected it). Careful dog selection and increased handler training should help reduce such biases. If the ratio of handler assisted versus independent finds becomes too excessive, the dog (and/or handler) should be removed from the sampling team.

Lack of population closure is another consideration in detection dog studies, depending on the models used to estimate population size and the range in age of scat samples encountered in the field (Kendall 1999). Lack of closure should not be a serious problem in the present study since (a) the majority of scat samples encountered were estimated to be < 1 mo. old.; and (b) older samples had a reduced DNA amplification success rate. Kendall et al. (2001) attempted to maximize closure by restricting scat collections to trails cleared one or more days prior to sampling. However, sampling only trails may have its own associated biases (Kendall et al. 2001). While detection dogs work equally well on-or-off trail, we found that far more samples were detected off-trail than on trail during our past 2 years of sampling (S. Wasser, unpublished data). Removing samples from trails prior to sampling also results in loss of information regarding the individuals that were there beforehand. Sample removal could be especially problematic if some animals were reluctant to traverse the sampling areas during the subsequent sampling period because of the immediate presence of human scent. Considerable time was spent this year attempting to improve amplification success of microsatellite DNA from grizzly bear scat. These efforts proved highly successful. Additional comparisons of this nature would be extremely valuable, enabling us to make direct comparisons between concurrently collected data sets, both in terms of final population estimates and sampling biases. These results would be extremely valuable for determining how best to proceed with long-term monitoring of grizzly bears in North America over the coming years.

154

13.5 Literature Cited

Clarke, C., D. Immell, and S. K. Wasser. 2001. Technical considerations for hair genotyping methods. Western Black Bear Workshop, 7, 24-29.

Ennis, S. & Gallagher, T. F. 1994. A PCR-based sex-determination assay in cattle based on the

bovine amelogenin locus. Animal Genetics, 25, 425-427. McLellan, B. N., Hovey, F. W., Mace, R. D., Woods, J. G., Carney, D. W., Gibeau, M. L.,

Wakkinen, W. L. & Kasworm, W. F. 1999. Rates and causes of grizzly bear mortality in the interior mountains of British Columbia, Alberta, Montanta, Washington and Idaho. Journal of Wildlife Management, 63, 911-920.

Mowat, G., and C. Strobeck. 2000. Estimating population size of grizzly bears using hair capture,

DNA profiling, and mark-recapture analysis. Journal of Wildlife Management, 64, 183-193.

Murphy, M., Waits, L. P., Kendall, K. C., Wasser, S. K., Higbee, J. A. & Bogden, R. in press. An

evaluation of long-term preservation methods for brown bear (Ursus arctos) faecal DNA samples. Conservation Genetics.

Paetkau, D., Calvert, W., Stirling, I. & Strobeck, C. 1995. Microsatellite analysis of population

structure in Canadian polar bears. Molecular Ecology, 4, 347-354. Paetkau, D. & Strobeck, C. 1994. Microsatellite analysis of genetic variation in black bear

populations. Molecular Ecology, 3, 489-495. Paetkau, D. & Strobeck, C. 1995. The molecular basis and evolutionary history of a

microsatellite null allele in bears. Molecular Ecology, 4, 519-520. Paetkau, D. & Strobeck, C. 1996. Mitochondrial DNA and the phylogeography of Newfoundland

black bears. Canadian Journal of Zoology, 74, 192-196. Stenhouse, G. B. & Munro, R. M. 1999. Foothills Model Forest Grizzly Bear Research Program,

1999 Annual Report, 139 pp. Hinton, Alberta.: Foothills Model Forest. Taberlet, P., Griffin, S., Goossens, B., Questiau, S., Manceau, V., Escaravage, N., Waits, L. P. &

Bouvet, J. 1996. Reliable genotyping of samples with very low DNA quantities using PCR. Nucleic Acids Res., 24, 3189-3194.

Taberlet, P., Mattock, H., C., D.-P. & Bouvet, J. 1993. Sexing free-ranging brown bears Ursus

arctos using hairs found in the field. Molecular Ecology, 2, 399-403. Taberlet, P., Waits, L. P. & Luikart, G. 1999. Noninvasive genetic sampling: Look before you

leap. Trends in Ecology and Evolution, 14, 323-327.

155

Wasser, S. K. unpublished data. Wasser, S. K., Houston, C. S., Koehler, G. M., Cadd, G. G. & Fain, S. R. 1997. Techniques for

application of faecal DNA methods to field studies of ursids. Molecular Ecology, 6, 1091-1097.

Woods, J. G., Paetkau, D., Lewis, D., McClellan, B. N., Proctor, M. & Strobeck, C. 1999.

Genetic tagging free-ranging black and brown bears. Wildlife Society Bulletin, 27, 616-627.

156

Appendix 1: Publication/Technical Paper List 1999 Dugas, J. and G.B. Stenhouse. 1999. Grizzly Bear Management: Validating Existing

Cumulative Effects Models. Thirteenth annual conference on geographic information systems. Vancouver, B.C. 1999.

GeoAnalytic Inc. 1999. Application of Evidential Reasoning to the Classification of Grizzly

Bear Habitat using Landsat TM and Ancillary Data, Milestone Report 1. December 31, 1999. For: Canada Centre for Remote Sensing, Contract number: 23413-9-D220-01/SQ.

Lee, J.L. and G.B. Stenhouse. 1999. Comparison of Grizzly Bear telemetry location data with a

grizzly bear habitat model. Foothills Model Forest Report. 29pp. Stenhouse, G.B. 1999. The Foothills Model Forest Grizzly Bear Research Program. A research

initiative in support of “A Framework for the Integrated Conservation of Grizzly Bears”. Work plan for 1998-1999. 120pp.

Stenhouse, G.B. and R. Munro. 1999. Foothills Model Forest Grizzly Bear Research Program

2000 Annual Workplan (year 2). 2000

Dechka, J., S. Franklin, D. Peddle, and G. Stenhouse. 2000. Land cover mapping and landscape fragmentation analysis in support of grizzly bear habitat management. Presented at Geographic Information Systems and Remote Sensing for Sustainable Forest Management: Challenge and Innovation in the 21st Century, Workshop, February 23-25, 2000, Edmonton, AB.

Cattet, M.R.L., N.A. Caulkett, and G.B. Stenhouse. 2001. The comparative effects of chemical immobilizing drug and method of capture on the health of free-ranging grizzly bears. Paper presented at the 13th International Conference on Bear Research and Management. May 21-25, 2001, Jackson, WY.

Cattet, M.R.L., N.A. Caulkett, G.B. Stenhouse, and M.E. Obbard. 2001. The development and assessment of a body condition index for polar bears and its application to black bears and grizzly bears. Paper presented at the 13th International Conference on Bear Research and Management. May 21-25, 2001, Jackson, WY.

Franklin, S.E., D.R. Peddle, J.A. Dechka, and G.B. Stenhouse. 2000. Grizzly bear habitat mapping in the Alberta Yellowhead Ecosystem using evidential reasoning with Landsat TM, DEM and GIS data. Paper presented at the Sixth Circumpolar Conference on Remote Sensing of Arctic Environments, Yellowknife, NWT, June 2000.

157

Franklin, S.E., G.B. Stenhouse, M.J. Hansen, C.C. Popplewell, J.A. Dechka, and D.R. Peddle. 2000. An Integrated Decision Tree Approach (IDTA) to mapping landcover using satellite remote sensing in support of grizzly bear habitat analysis in the Alberta Yellowhead Ecosystem. Canadian Journal of Remote Sensing. (submitted).

GeoAnalytic Inc. 2000. Application of Evidential Reasoning to the Classification of Grizzly

Bear Habitat using Landsat TM and Ancillary Data, Milestone Report 2. January 31, 2000 For: Canada Centre for Remote Sensing, Contract number: 23413-9-D220-01/SQ.

Stenhouse, G.B. and R. Munro. 2000. Foothills Model Forest Grizzly Bear Research Program 1999 Annual Report. 110 pp.

Skrenek, J., D. Hodgins and G.B. Stenhouse. 2000. Managing cumulative effects on Grizzly

Bears: An inter agency and multi-stakeholder strategy in the Alberta Yellowhead Ecosystem. Paper presented at “Environmental Cumulative Effects Management Conference" November 1-3, 2000, Calgary, Alberta.

Stenhouse, G.B. and G. Mowat. 2000. Grizzly Bear DNA Hair Inventory Project Results.

Paper presented at the Managing for Bears in Forested Environments Conference, October 17-19, 2000, Revelstoke, B.C.

Wasser, S. and G.B. Stenhouse. 2000. Grizzly Bear Inventory using trained dogs and DNA scat

analysis. Paper presented at the Managing for Bears in Forested Environments Conference, October 17-19, 2000, Revelstoke, B.C.

2001 Boulanger, J., G. Stenhouse and R. Munro. 2001. Causes of heterogeneity bias when DNA

mark-recapture sampling methods are applied to grizzly bear populations. Journal of Applies Ecology. In press.

Cattet, M.R.L., N.A. Caulkett, and G.B. Stenhouse. 2001. The comparative effects of chemical immobilizing drug and method of capture on the health of free-ranging grizzly bears. Paper presented at the 13th International Conference on Bear Research and Management. May 21-25, 2001, Jackson, WY. Manuscript submitted to Ursus for publication.

Cattet, M.R.L., N.A. Caulkett, and G.B. Stenhouse. 2001. The development and assessment of a body condition index for polar bears and its application to brown bears and black bears. Paper presented at the 13th International Conference on Bear Research and Management. May 21-25, 2001, Jackson, WY. Manuscript submitted to Ursus for publication.

Cattet, R.L., K. Christison, N. Caulkett and G.B. Stenhouse. 2001. Effects of method of capture

on chemical immobilization features and physiological values in grizzly bears. Submitted paper, Journal of Wildlife Diseases, December 2001.

158

Frair, J., E. Merrill, M. Boyce, S. Lele, G. Stenhouse, R. Munro and S. Nielsen. 2001. Incorporating habitat-biase3d locational error into habitat use models. Paper presented at the 8th Annual Conference of the Wildlife Society. Reno, Nevada, September 2001.

Franklin, S.E., D.R. Peddle, J.A. Dechka, and G.B. Stenhouse. 2001. Evidential reasoning with

Landsat TM, DEM, and GIS data for landcover classification in support of grizzly bear habitat mapping. International Journal of Remote Sensing. 2001 (in press).

Logan, R.J., G.B. Stenhouse, and R.F. Ferster. 2001. Cheviot Mine: A catalyst for space age

research towards regional conservation of the grizzly bear. Paper presented at the 2001 Annual conference of the Canadian Institute of Mining, Metallurgy and Petroleum. Quebec City, Quebec, April 2001.

Mucha, D.M., G.B. Stenhouse and J. Dugas. 2001. A 3D landscape visualization tool using satellite imagery for grizzly bear management in the Alberta Yellowhead Ecosystem. Paper submitted to the 2001 Alberta Chapter of the Wildlife Society Annual Meeting, March 3-5, Banff, Alberta.

Munro, R.H.M., S.E. Nielsen, G.B. Stenhouse, and M.S. Boyce. 2001. The influence of habitat quality and human activity on grizzly bear home range size. Paper presented at the 13th International Conference on Bear Research and Management. May 21-25, 2001, Jackson, WY. Manuscript submitted to Ursus for publication.

Nielsen, S., M. Boyce, and G.B. Stenhouse. 2001. Can you have too much data? The Problem of spatial autocorrelation in habitat selection studies. Paper presented at the 8th Annual Wildlife Society Conference, Reno Nevada, September 24-30th, 2001.

Nielsen, S.E., M.S. Boyce, G.B. Stenhouse and R. Munro. 2001. Using Resource Selection

Functions in Population Viability Analysis of Yellowhead Grizzly Bears. Paper presented at Alberta Conservation Association “Partners in Conservation Conference”, February 10, 2001, Nisku, Alberta.

Nielsen, S.E., M. Boyce, G.B. Stenhouse, and R. Munro. 2001. Resource selection of grizzly bears in the Yellowhead Ecosystem of Alberta, Canada. Poster presented at the 13th International Conference on Bear Research and Management. May 21-25, 2001, Jackson, WY. Manuscript submitted to Ursus for publication.

Nielsen, S.E., M. Boyce, G. Stenhouse and R. Munro. 2001. Habitat selection by Yellowhead grizzly bears. Paper presented at the 8th Annual Conference of the Wildlife Society. Reno, Nevada, September 2001.

Nielsen, S., M. Boyce, G.B. Stenhouse and R. Munro. 2001. Incorporating food phenology

models for grizzly bear predictions: can we improve upon static habitat maps? Submitted paper, December 2001.

Popplewell, Charlene. 2001. Habitat structure and fragmentation of Grizzly Bear management

units and home ranges in the Alberta Yellowhead ecosystem. University of Calgary, MSc Thesis.

159

Popplewell, C. G.B. Stenhouse, M. Hall-Beyer and S.E. Franklin. 2001. Using remote sensing and GIS to quantify the Landscape structure and habitat fragmentation within grizzly bear management units in the Yellowhead Ecosystem, Alberta. Poster presented at the 13th International Conference on Bear Research and Management. May 21-25, 2001, Jackson, WY. Manuscript submitted to Ursus for publication.

Stenhouse, G.B. 2001. Foothills Model Forest Grizzly Bear Research Program. Invited paper

presented at the 2001 Environmental Research and Technology Development Forum for the Upstream Oil and Gas Industry. Sponsored by Petroleum Technology Alliance Canada (PTAC). January 31, 2001, Calgary, Alberta.

Stenhouse, G.B. 2001. The Foothills Model Forest Grizzly Bear Research Program: Building on

Partnerships. Invited paper presented at the Alberta Conservation Association “Partners in Conservation Conference”, February 10, 2001, Nisku, Alberta.

Cattet, M.R.L., N.A. Caulkett, and G.B. Stenhouse. 2003. Anesthesia of grizzly bears using xylazine-zolazepam-tiletamine or zolazepam-tiletamine. Ursus 14(1) (In press)

Stenhouse, G.B. 2001. Grizzly Bear Conservation in the Northern East Slopes of Alberta: the integration of land management direction and grizzly bear research. Paper presented at the 13th International Conference on Bear Research and Management. May 21-25, 2001, Jackson, WY. Manuscript submitted to Ursus for publication.

Stenhouse, G.B. and R. Munro. 2001. Grizzly Bear Mortality in the Yellowhead Ecosystem –

Anomaly or Trend? Paper presented at the 2001 Annual Conference of the Northwest Section of the Wildlife Society. Banff, Alberta, March 2-4, 2001.

2002 Cattet, M.R.L., N.A. Caulkett, M.E. Obbard, and G.B. Stenhouse. 2002. A body condition

index for ursids. Canadian Journal of Zoology 80: 1156-1161. Caulkett, N.A. and M.R.L. Cattet. 2002. Anesthesia of bears. In Heard, D. (Ed). Zoological

Restraint and Anesthesia. International Veterinary Information Service (www.ivis.org), Ithaca, New York.

Caulkett, N.A., M.R.L. Cattet, and G.B. Stenhouse. 2002. Comparative physiological effects of immobilizing agents in North American bears. Paper presented at the 14th International Conference on Bear Research and Management. July 30th to August 1st, 2002, Steinkjer, Norway.

Nielsen, S.E., M.S. Boyce, G.B.Stenhouse and R.H.M Munro. 2002. Modeling grizzly bear

habitats in the Yellowhead Ecosystem of Alberta: taking autocorrelation seriously. Ursus (in press).

2003

160

Cattet, M.R.L., K. Christison, N.A. Caulkett, and G.B. Stenhouse. 2003. Physiological responses of grizzly bears to different methods of capture. Journal of Wildlife Diseases (Accepted for publication)

Nielsen, S.D., G.B. Stenhouse, R.H.M. Munro. 2003. Development and testing of phenologically

driven grizzly bear habitat models. Ecoscience 10:2480-1490. Popplewell, C. G.B. Stenhouse, M. Hall-Beyer and S.E. Franklin. 2002. Using remote sensing

and GIS to quantify the Landscape structure and habitat fragmentation within grizzly bear management units in the Yellowhead Ecosystem, Alberta. Ursus (in press).

Schwab, B., C. Woudsma, G.B. Stenhouse, S.E. Franklin, and S.E. Nielsen. 2002. Connections

That Matter: Graph Theoretic Analysis of Grizzly Bear Movement in the Yellowhead Ecocsystem, Alberta, Canada. Ursus (in press).

161

Appendix 2: Foothills Model Forest Grizzly Bear Research Partners

(1999- 2002)

BP Canada Energy Company

Cardinal River Coals Ltd.

Hinton Training Centre

Ainsworth Lumber Alberta Conservation Association Alberta Energy Company Alberta Sustainable Resource Development/Alberta Environment Alberta Newsprint Anderson Resources Ltd AVID Canada BC Oil and Gas Commission Environmental Fund Blue Ridge Lumber (1981 Ltd)

Burlington Resources Canada Centre for Remote Sensing Canadian Resources Ltd Canadian Forest Products Canadian Hunter Canadian Wildlife Service

ConocoPhillips Canada Resources Ltd. Foothills Model Forest FRIAA G&A Petroleum Services GeoAnalytic Ltd. Gregg River Resources

Inland Cement Komex International Ltd.

Luscar Sterco (1977) Ltd Millar Western Pulp Ltd Mountain Equipment Coop National Science and Engineering Research Council (NSERC) Northrock Resources Parks Canada (Jasper National Park) Peregrine Helicopters Petro-Canada Precision Drilling Ltd. PTAC (Petroleum Technology Alliance of Canada) Rocky Mountain Elk Foundation Suncor Sundance Forest Industries Sunpine Forest Products Ltd Talisman Energy Telemetry Solutions The Centre for Wildlife Conservation (USA) Trans Canada Pipelines University of Alberta University of Calgary University of Lethbridge University of Sasketchewan University of Washington Veritas Ltd. Weldwood of Canada Ltd Weyerhaeuser Canada Ltd World Wildlife Fund

162

Gbp 2003 03 annrpt 2002

Documents

Transcript of Gbp 2003 03 annrpt 2002