South Purcell Mountain Grizzly Bear Linkage Project

Michael Proctor1 Chris Servheen2 Wayne Kasworm3

This report is preliminary and results should not be considered final until the project is complete

Report submitted to:

TEMBEC Industries Inc. Cranbrook BC Contract # 05 FIA RIP 103

BC Ministry of Environment, Ecosystems Branch Victoria BC

National Fish and Wildlife Foundation Washington DC

Wilburforce Science Grants Bozeman

Total project cost $115, 106 TEMBEC 14% - $16,000 BC MoE 18% - %20,500 NFWF 26% - $29785 Wilburforce 32% - $37080

March 2006

1 Birchdale Ecological PO box 920 • Kaslo, BC • V0G 1M0 • Canada • [email protected]

2 US Fish and Wildlife ServiceCollege of Forestry

and Conservation,309 University Hall, University of Montana, Missoula, MT 59812, USA

3 US Fish and Wildlife Service 475 Fish Hatchery Road, Libby, MT 59923, USA

ACKNOWLEDGEMENTS

We would like to thank the many funders who have provided for and continue to support this project. The component reported here was a portion of our larger long-term effort. We thank Kari Stuart-Smith and TEMEBC for their vision to fund work aimed at grizzly bear and ecosystem conservation. We appreciate Matt Austin from the BC Ministry of Environment who has been instrumental in securing continuing Provincial support. We also thank the US-based National Fish and Wildlife Foundation and the US Fish and Wildlife Service for supporting ecological efforts in Canada as well as Wilburforce Science Grants and the Y2Y community, and the Liz Claiborne & Art Ortenberg Foundation. An Alberta Ingenuity Fellowship has supported M Proctor thought much of this effort. John Bergenske has been instrumental in field work efforts as has Gillian Sanders.

2 TABLE OF CONTENTS Acknowledgements…………………………………………………………………………………2 Summary……………………………………………………………………………………………4 Project rationale and description………………………………………………………………....5 Project objectives…………………………………………………………………………………...6 Project activities and methods..……………………………………………………………………6 Field and lab work………...……………………………………………………………………6 Resource Selection Function analysis…..……………………………………………………..7 Results…………………………………………………………………………………………..…...8 Discussion……………………...…………………………………………………………………...10 Deliverables and measures of success summary.………………………………………………...13 Budget summary…………………………………..……………………………………………… 13

Literature Cited…………………………………….………………………………………...…....14

TABLES AND FIGURES

Table 1. Uni-variate logistic regression significant variables...... 9

Figure 1 North American current and historic grizzly bear distribution…………….…...…...16

Figure 2 Study are map………..………………………………………………………………...... 17

Figure 3 Grizzly bear probability of occurrence map…….……………………………………..18

Figure 4 Close up of “linkage zone” predicted along Hwy 3…...…….………...………………..19

3 Summary Grizzly bears in southeast ’s south Purcell Mountain Yahk Grizzly Bear Population Unit are a threatened, fragmented, and declining population. M. Proctor and the USFWS have an international cooperative working relationship for research and management to “recover” this population. Here we report on a two year effort to build a predictive linkage model from the results of DNA surveys and ecological modeling. We genetically sampled wild grizzly bears along and adjacent to Highway 3 between Creston and Cranbrook BC in 2004 and 2005 and combined the resulting presence/absence data with similar results from previous years’ efforts in the region. In total we sampled 170 sites and caught 65 grizzly bears at 54 sites totaling 124 capture events. We used multiple logistic regression and GIS to develop a spatially explicit “resource selection function” model to predict bear occurrence across our 5300 km2 study area. We extrapolated the predictive model to the edges of the south Purcell area encompassing 9500 km2. Our goal was to use this model to help predict linkage areas across the Highway 3 transportation and human settlement corridor, to inform management efforts to enhance successful functional inter-population linkage between areas north and south of Highway 3. Our probability of occurrence model identified one potential linkage zone northeast of Yahk along Highway 3 between Ryan Provincial Park and an area east of Irishman Creek. This model is preliminary and will be integrated with similar models developed at two broader scales to better reflect factors that may influence grizzly bear distribution. We will also validate and improve this model with an independent data set of radio telemetry GPS locations, when that data set is complete.

Project rationale and description The goal of this project is to use research and management to begin recovery efforts of the small fragmented grizzly bear population in the south Purcell Mountains south of BC Highway 3. This trans- border population is threatened in British Columbia (Hamilton et al. 2004) and the US (USFWS 1993) and at the southern edge of a contracting distribution (Mattson and Merrill 2002) (Fig 1). The federal designation for the grizzly bear in Canada is special concern (COSEWIC – Committee on the Status of Endangered Wildlife in Canada). Recent research suggests that this is a small (< 50 bears including the US portion) declining population (-3.7%/year; Wakkinen and Kasworm 2004) experiencing limited female connectivity with adjacent populations resulting in an elevated conservation risk (Proctor et al. 2005). This project is a part of a larger coordinated effort for recovery of the south Purcell and south Selkirk populations that is being integrated with the US Fish and Wildlife Service and the BC MoE.

4 Initially we are focusing on reducing human-caused mortality, enhancing and re-establishing inter- population connectivity, improving habitat security, and public education (Proctor et al. 2004). Specifically, this portion of our project is working on identifying the best options for “linkage zones” across Highway 3 in the Purcell Mountains. Because there appears to be some male interchange occurring across Highway 3 in the Purcell Mountains, it is feasible to identify and further develop linkage zones in appropriate areas that would facilitate the secure movement of animals of both sexes through this human transportation corridor. As human development in the area proceeds, the opportunity to establish linkage zones across the Highway 3 transportation and human settlement corridor is diminishing every year. We are using two techniques to identify linkage zones, GPS radio telemetry to identify where bears (likely males) are actually crossing Highway 3, and systematic DNA surveys to estimate core and linkage habitat through Resource Selection Function (RSF) habitat modeling (Manley et al 2002; Apps et al 2004). The reason for using both methods is that it in a system where bears are sparse and movement across Highway 3 is limited, each method will reveal an important part of the picture, and hopefully together they will reveal the best options for linkage habitat. In 2004 and 2005 we did DNA surveys of wild grizzly bears between Creston and Cranbrook BC and combined these results with those of a less intense but broader-based DNA survey in 2001 to underpin our RSF analysis. Here we report that modeling exercise and resulting predictive core and linkage map from presence/absence data for the Highway 3 area. When the radio telemetry portion of the study is complete, we will use our GPS radio telemetry to verify the genetic-based fragmentation results, identify areas where bears might cross Highway 3, and generate data to test, inform, and improve our DNA-based predictive linkage model. We have completed year 2 of 4 in the radio telemetry effort. We strive to integrate industry, government and local landowners and translate our results into practical linkage zone management strategies for the realization of inter-population connectivity. TEMEBC Industries, through a FIA contract, contributed $16,000 toward a portion of the 2005 DNA survey effort and this document fulfills reporting requirements for that contract (05-FIA-RIP-103).

Project activities – methods Field and lab work In June and July 2004 and 2005 we carried out hair-grab DNA surveys of grizzly bears between Creston and Cranbrook BC along and on both sides of BC Highway 3 (Fig. 2). We established 101 DNA

5 barbed-wire scent-baited sampling sites, one in each 25 km2 cell, and collected hair samples (our source of DNA) in four consecutive 2-week periods (Fig 2). These data were combined with results from similar broad-based surveys in 1998 and 2001 in the areas immediately to the north of the 2004-2005 surveys. In total, we sampled 170 sites within our study area. We accessed sites by vehicle and hiking that were located in the best available grizzly bear habitat a minimum of 200m from any road or human trail. This technique is well established throughout North America and follows protocols detailed in Woods et al. (1999). All field methods were conducted according to B.C. government RIC standards. All hair samples were genetically analyzed at Wildlife Genetics International in Nelson BC. WGI is the world’s leading lab for this type of analysis, consisting of microsatellite-based DNA fingerprints, used for reliable individual identification of bears as well as population level inquiries. Grizzly bear samples were distinguished from black bear hair by the presence of a positive allele at a species diagnostic locus (G10J, Paetkau et al. 1998; Paetkau 2003). All grizzly bear samples were genotyped to 6 loci for individual identification. One sample from each individual was used to extend genotypes to 15 loci to be used in population level investigations (Proctor et al. 2003). Genotyping errors were minimized through established protocols detailed in Paetkau (2003).

Resource Selection Function Analysis During the winter of 2005 we prepared Geographic Information Systems (GIS) layers for an RSF analysis of the south Purcell region. We generated raster layers with a 100m by 100m cell size for 24 landscape, habitat, and human variables, and correlated those variables with presence/absence data associated with the DNA survey results in a multiple logistic regression (Apps et al. 2004). Variable layers were derived from geographic data from BC Provincial databases. We obtained Baseline Thematic Mapping (BTM) data for landcover classes (avalanche, alpine, burns, barren, old forest, young forest, and recently logged). Terrain Resource Information Management (TRIM, 1:20,000 scale) data provided digital elevation model (DEM) data, road, and stream layers. The DEM was used to derive terrain feature layers, slope, curvature, average solar radiation, and terrain ruggedness. Vegetation Resource Inventory (VRI 1:20,000 scale) data was used to derive layers for forest cover attributes, such as dominant tree species (cedar-hemlock, lodgepole pine, whitebark pine, Douglas fir, and deciduous species) as well as stand age and canopy cover. A regional human settlement layer was also obtained from Provincial databases. A park variable was derived from a protected areas layer, and a main

6 highway layer was digitized off topographic maps. A riparian variable was derived by combining streams, wetlands, and areas of low slope. Because animals select habitat in response to factors that operate at different scales (Manly et al. 2002), our RSF analysis is to be done at 3 scales. The first scale we have modeled is the smallest, meant to simulate the scale that a bear may make daily decisions about habitat selection: a radius of 2.4 km, the average daily movement of a grizzly bear in the nearby Flathead Mts, (unpublished data, B. McLellan). To build this scale into our modeling, we used a moving window over each variable and characterized each cell in the raster layers by calculating a mean value for an area with a radius of 2.4 km. Our sampling unit is 1 DNA site every 25 km2, and we are therefore not measuring fine scale habitat selection, but rather we are exploring habitat selection at several landscape scales. In the near future we will be modeling the medium and large scale that emulate a spring seasonal home range (6.7 km radius) and the average annual home range of grizzly bears in the area (11.8 km radius, the broadest scale at which a grizzly bear may respond to landscape conditions. We then extracted values from all variables for each cell where a DNA site was located and used multiple logistic regression to associate presence and absence of grizzly bears with our landscape, habitat, and human variables. We first used Pearson’s correlation coefficient to eliminate variables that were correlated > 0.7. When confronted with correlated variables, we used the one that was more significantly associated with our presence/absences data in uni-variate analysis. We then used uni- variate logistic regression to rank the correlations of all variables’ fitness for use in a multiple-variate regression. In our multiple logistic regression we only used uncorrelated variables that showed uni- variate significance (P > 0.1). We used model building principles to guide stepwise addition (add one variable at a time and compare to previous best model) and elimination of non-significant variables (start with full model and eliminate least significant variables one at a time; Hosmer and Lemshaw 1989; McGarigal et al. 2000). We then took the multiple logistic regression model that best represented the attribute values for the presence/absence data and applied them to all cells in GIS. This process yielded a map that predicts bear occurrence relative to the variables and their attributes where presence was detected. As an assessment of the fit of our data to the model we did a “confusion matrix” (McGarigal et al. 2000). To carry this out we applied a probability of occurrence predictive threshold where the “cutpoint” is approximately the proportion of sites that captured grizzly bears (0.3136). Cells > 0.3136 were assigned a 1 and cells < 0.3036 were assigned a 0. For all cells that contained a DNA site, these values (0 predicts

7 we should not capture a grizzly bear and 1 predicts we should) were tested against what our model predicted in the confusion matrix. Our goals for this predictive map were the following. We were looking for areas of higher quality habitat that may suggest “core” habitat, that we hypothesized were adjacent to the human settled valley containing Highway 3. Our search for “linkage” habitat would be higher quality areas that extend toward the highway corridor and were mirrored on both sides of the highway, fingers of linkage habitat connected to core areas. Ultimately, we will test the predictions of this RSF map with independent radio telemetry data.

Results In 2005, we collected 984 bear hair samples, 66 of which were from grizzly bears. We identified 17 different grizzly bears, 4 males and 13 females. Four grizzlies were caught at 4 sites south of Hwy 3 and 13 bears at 10 sites north of the highway. One male was captured south of Hwy 3 that was captured at 2 different locations north of the highway in our 2001 survey. We will target this bear for radio collaring in the coming season. In 2004 we collected 550 bear hair samples which included 55 grizzly samples from 11 individual bears. Four were captured south of Hwy 3 and 9 north. These data were combined with 29 DNA captures from a 2001 south Purcell survey and 17 bears DNA captured in 1998 and 1999. One additional male was captured north and south of Highway 3 during the 2001 survey. Capture locations are summarized in Figure 3. In total we captured 65 individual grizzly bears with multiple captures for many individuals totaling 124 capture events. These captures occurred at 54 of our 170 sampling stations. Our analysis area encompassed 5300 km2 and we extrapolated our model to the edges of the south Purcell ecosystem (9500 km2 Fig. 3). Many combinations of variables were correlated to one another and we chose the variable for multiple regression that had the most significant association with the presence and absence of grizzly bears in our landscape. For instance, alpine and avalanche habitat were highly correlated and alpine was chosen to be an ingredient in the multiple regression model building process. This is important when it comes to interpreting which variables shape grizzly bear distribution. Correlated variables that do not appear in a final multiple regression may be “represented” by the variable they correlate with in the model and still influence bear distribution. Table 1 summarizes the variables and their direction of influence that were significantly associated with presence and absence data in a uni-variate logistic regression and were not correlated to

8 one another. Other variables that were significant in the uni-variate analysis but were highly correlated with the chosen variables, were not use in model building and they are listed in the lower section of Table 1. They were omitted because their correlates in the upper section of Table 1 did a getter job of explaining the data. The most parsimonious model to emerge from model selection contained roads (-), elevation and terrain ruggedness (+), and deciduous (-). We also preformed the model selection procedure and omitted roads as a variable. The reason for this is that road avoidance is one of the primary selective forces on bear distribution (in the study and many others (Waller and Servheen 2005; Mace et al. 1996; Kasworm and Manley 1990; McLellan and Shackleton 1988; Mattson et al. 1987) but we are trying to predict where linkage zones might be located through a roaded environment across the Highway 3 valley. Furthermore, access management may be one of the ultimate tools for linkage management. Figure 3 is a spatially explicit grizzly bear probability of occurrence model of BC Provincial Yahk and the South Purcell Grizzly Bear Population Units (GBPU). We did a “confusion matrix” as an assessment of fit to our data (McGarigal et al. 2000) which scored 0.63.

Table 1. Uni-variate logistic regression significant variables and association with grizzly bear presence around Highway 3 are in Purcell Mountains.

Elevation <0.001 + Terrain ruggedness <0.001 + Alpine <0.001 + Roads 0.001 - Deciduous tree sps. 0.002 - Young forest 0.027 - Highway 3 0.029 - Essf 0.048 + Whitebark pine 0.073 +

1Essf is Englemann Spruce and Sub-alpine fir dominant tree species Correlations: all variables below were significant in uni-variate regression. They were not used because their correlates above explained the data better. Terrain ruggedness was correlated to solar radiation and slope Alpine was correlated with Avalanche, Young forest was correlated with old forest Highway 3 was correlated with human settlement

9 There are several patterns worth noting in the Highway 3 area. First is the presence of what might be considered “core” habitat predictions adjacent to Highway 3. Second, is the presence of a potential linkage zone across Highway 3 just to the northeast of Yahk BC that extends from Ryan Provincial Park east past Irishman Creek (red oval, Fig. 3 & 4 close up).

Discussion The results presented in this report are preliminary, as identification of recommended linkage zones is approximately 60% complete. We have only had this data for less than 2 weeks and will continue modeling predictive linkage habitat. We plan to model at 2 additional scales, (seasonal and annual home range), recognizing that bears may be responding to landscape features at these scales. We also plan to use a multi-model approach using Akaike’s Information Theoretic methods and model averaging to build a more comprehensive predictive model (Burnham and Anderson 1998). The goals of this modeling exercise were challenging as we are trying to predict linkage zones in an area where bears are fragmented, sparse, and avoiding a large highway and human settled valley. One goal was to identify “core” grizzly bear habitat adjacent to Highway 3 that would offer the best possibility of facilitating movement of grizzly bears across the highway and through the associated human environment. Linkage zones may be more successful if connected to areas of high quality bear habitat. Our probability of occurrence map meets this objective. We identified several areas where higher quality habitat is adjacent to the human highway corridor, particularly in the western portion of the study area. A second, more ambitious goal was to identify any areas where reasonably good habitat extends into the human corridor and across Highway 3 that might be a potential linkage zone. We found one such area just to the northeast of Yahk between Ryan Provincial Park and Tochty along Highway 3 (Fig. 3 & 4 close up, red oval). Prior to this modeling effort, we hypothesized that this area might be a linkage zone because we DNA captured one individual bear north and south of Highway 3 and near the highway within this linkage zone. Interestingly, when examining the independent radio telemetry locations (not provided in this report) one bear uses habitat in this potential “linkage zone” right up to the highway on the north side. Our probability of occurrence model did not predict a “linkage zone” in a location near the confluence of Kidd Creek and Highway 3 where we know from radio telemetry locations that a male grizzly bear crosses Highway 3. Our model did however predict “core” areas adjacent to this crossing location. It was our hypothesis that due to sparse numbers of bears using this area that both DNA based

10 RSF modeling and radio telemetry would be necessary to uncover bear habitat use and this situation appears to support that supposition. While bears may prefer higher quality habitat, they must move through areas of lower quality particularly in fragmented landscapes. This individual appears to be targeting the adjacent good habitat just south of Hwy 3 that our model predicted. Apps (1997) did an assessment of potential linkage habitat along Highway 3 between Creston and Cranbrook by using 4 variables: linear disturbance, human density, riparian areas and visual cover. While there was no specific bear location or habitat use data in the modeling exercise, it identified 3 potential linkage zones. One was very similar to the linkage area we identified with our RSF model, one was very similar to the preliminary zone we found with radio telemetry, and the third is not supported by our results. Our assessment of fit (confusion matrix score) for the model to the data was only 63% (>70% is considered good). One reason contributing to this slightly lower result may be that we did this survey from vehicle access and hiking (no helicopter use). From examination of the DNA captures and the high quality predicted areas, it is apparent that our sampling method reached the edges of better quality habitat where we DNA-captured grizzly bears. This pattern suggests that our sampling effort may not have sampled a purely representative sample of the landscape. There may be value in using helicopters in less roaded habitats to reach remote areas for better representative sampling. On the other hand, our goal was to model the relatively lower quality disturbed habitat at lower elevations near the Highway 3 corridor. The likelihood of this habitat being used by bears may be greater early in their season (June – July) when we sampled. We therefore centered our grid on this habitat type, while trying to also sample across the variation of our variables. In a preliminarily exploration of our model’s fit to independent data there is good resonance between the DNA-derived predictive model and our independent radio telemetry locations. It is our plan to use the final radio telemetry dataset to validate this DNA-derived model, and ultimately vice versa. This model will be used to begin the process of linkage zone identification that will be complete when modeling has been finalized and radio telemetry data has been integrated into the analysis. This phase of our efforts has already begun with input into TEMBEC’s High Value Conservation Forests (HVCF) planning. Landuse planning is an iterative process and our research program will share results as they evolve and improve with relevant industries, agencies and the public as necessary. Linkage zone identification for bears is a recent phenomena in the scientific literature as reports of fragmentation increase. Graves et al. (2006) used GPS radio locations of bears in and around salmon

11 bearing streams in Alaska to characterize “corridors” used by bears to cross a major highway. Underpinning this effort is an area where bears are known to cross a major highway to reach high quality salmon resources. Waller and Servheen (2006) characterized grizzly bear crossing locations along a highway in Montana using logistic regression and GPS locations. What these two studies have in common that differs from our effort is that both have multiple crossing events by multiple bears, enough to identify generalized characteristics of crossing areas. In other words, fragmentation was not as advanced in these systems as it is in our system. Our challenge is to identify linkage habitat in a system that has been significantly fractured. Genetic data suggest that movement is limited across this highway (Proctor et al. 2005) and this is supported by our radio telemetry results so far. We have 8 radio collared bears (to date) and only one male has crossed the highway. That individual crossed Highway 3 multiple times in one location, suggesting that some landscape characteristics may be important, at least for this bear. We also now have evidence from a DNA capture, that another male crossed Highway 3 at this same location. While identifying linkage zones for grizzly bears in the south Purcell Highway 3 area is challenging, it remains a worthwhile endeavor. Fragmentation is by no means complete (Proctor et al. 2005), grizzly bear recovery is a feasible goal (Pyare et al. 2004; Swenson et al. 1998), and development pressures are increasing, so the need exists and our work is timely. Our research group has experienced considerable willingness on the part of industry, government, and the public to accommodate coexistence with grizzly bears, limited often by research into workable solutions.

12 Deliverables and measures of success summary DNA survey results of successful DNA sampling sites across the landscape. COMPLETE Predictive linkage model for the Hwy 3 area of the south Purcell Mts. COMPLETE Report to funding contributors containing field results and predictive linkage model. COMPLETE Science, industry, government, and public-derived linkage zone management plan. IN PROGRESS

Budget for South Purcell Grizzly Bear Linkage Project

Budget for TEMBEC FIA report S Purcell 2005 GB Linkage Project

Expenses IncomeSource 2001 DNA survey Field 17321.89 37080 Wilburforce Lab 16365.65 Biologist 3393 37080.54 37080

2004 DNA survey Field 18,785 12,000 BC MoE1 Lab 11000 17785 NFWF2 (USFWS)3 Biologist 2000 2000LCAOF4 31,785 31,785

2005 DNA survey Field 21240.52 8500 BC MoE Lab 15,000 16000 TEMBEC Analysis Biologist 10,000 12000 NFWF (USFWS) 10000 UA AI5 Post Doc Salary 46240.52 46500

115106.1 115365

1NFWF is National Fish and Wildlife Foundation (US based) 2USFWS is US Fish and Wildlife Service 3BC MoE is BC Ministry of Environment 4LCAOF is Liz Claiborne & Art Ortenberg Foundation 5UA AI is University of Alberta - Alberta Ingenuity Fellowship

.

13 Literature cited

Apps, C. 1997. Identification of grizzly bear linkage zones in the highway 3 corridor of southeast British Columbia and southwest Alberta. B.C. Ministry of Environment, Lands, and Parks and WWF Canada. Cranbrook, B.C. Canada. 45 pp.

Apps, C. D., B. N. McLellan, J. G. Woods, and M. F. Proctor. 2004. Estimating grizzly bear distribution and abundance relative to habitat and human influence. Journal of Wildlife Management 68:138-152.

Burnham, K.P. and D.A. Anderson. 1998. Model selection and inference: a practical information- theoretic approach. Springer-Verlag. New York. 353 pp.

Graves, T., S. Farley, M.I. Goldstein, C. Servheen, C. Schwartz, and S Arthur. Submitted. Identification of functional corridors with movement characteristics of brown bears on the Kenai Peninsula, Alaska. Journal of Wildlife Management.

Hamilton, A.N., D.C. Heard, and M.A. Austin. 2004. British Columbia grizzly bear (Ursus arctos)pPopulation estimate 2004. British Columbia Ministry of Water, Land, and Air Protection, Biodiversity Branch. Victoria, B.C. 7 pp.

Hosmer, D. W., and S. Lemeshow. 1989. Applied logistic regression. John Wiley and Sons, New York, New York, USA.

Kasworm, W.F. and T.L. Manley. 1990. Road and trail influences on grizzly bears and black bears in northwest Montana. International Conference on Bear Research and Management 8:79-84.

Mace, R.D., J.S. Waller, T.L. Manley, L.J. Lyon, and H. Zuring. 1996. Relationships among grizzly bears, roads, and habitat use in the Swan Mountains, Montana. Journal of Applied ecology. 33:1395-1404.

Manly, B. F. J., L. L. McDonald, and D. L. Thomas, T. L. McDonald, and W. P. Erickson. 2002. Resource selection by animals: statistical design and analysis for field studies, 2nd edition. Kluwer Academic Publishers, Norwell, Massachusetts, USA.

Mattson, D.J., R.R. Knight, and B. Blanchard. 1987. The effects of development and primary roads on grizzly bear habitat use in Yellowstone National Park, Wyoming. International Conference on Bear Research and Management. 7:259-274.

Mattson, D.J. and T. Merrill. 2002. Extirpations of grizzly bears in the contiguous . Conservation Biology. 16:1123-1136.

McLellan, B.N. and D.M. Shackleton. 1988. Grizzly bears and resource extraction industries: effects of roads on behavior, habitat use, and demography. Journal of Applied Ecology. 35:451-460.

McGarigal, K. S. Cushman, and S. Stafford. 2000. Multivariate Statisticas for Wildlife and Ecology Research. Springer-Verlag. New York. 283pp.

14 Pyare, S., S. Cain, D. Moody, C. Schwartz, and J. Berger. 2004. Carnivore re-colonization: reality, possibility and a non-equilibrium century for grizzly bears in the Southern Yellowstone Ecosystem. Animal Conserv. 7:1-7.

Proctor, M.F., C. Servheen, S. Miller, W. Kasworm, and W. Wakkinen. 2004 A comparative analysis of management options for grizzly bear conservation in the U.S.-Canada trans-border area. Ursus 15:145-160..

Swenson, J.E., F. Sandergren, and A. Söderberg. 1998. Geographic expansion of an increasing brown bear population: evidence for presaturation dispersal. Journal of Animal Ecology. 67:819-826.

U.S.F.W.S. 1993. Grizzly bear recovery plan. U.S. Fish and Wildlife Service. Missoula Montana. 181 pp.

Wakkinen, W.L., and W.F. Kasworm. 2004. Demographic and population trends of grizzly bears in the Cabinet–Yaak and Selkirk ecosystems of British Columbia, , Montana, and Washington. Ursus 15 Workshop Supplement: 65–75.

Waller, J.S., and C. Servheen. 2005. Effects of transportation infrastructure on grizzly bears in northwestern Montana. Journal of Wildlife Management69:985-1000.

Woods, J.G., D. Paetkau, D. Lewis, B.N. McLellan, M. Proctor, and C. Strobeck. 1999. Genetic tagging of free-ranging black and brown bears. Wildlife Society Bulletin. 27:616-627.

15 Figure 1a.Current and historic distribution of grizzly bears in North America. b. Close up of regional grizzly bear distribution, Blue estimates current grizzly bear distribution, green represents protected areas (BNP is Banff National Park, GNP is Glacier National Park, and PWC is Purcell Wilderness Conservancy), brown lines are major highways, and our study area is within the red oval. a. b.

16 Figure 2. Study area map with 2004 (yellow) and 2005 (white) DNA grids. Yellow dots are sampling sites and those that exist outside of grid were sampled in 2001 and 2002.

17 Figure 3. Probability of occurrence of grizzly bears in the south Purcell Mountains of southeast BC. Map includes DNA sampling locations and occurrence results (DNA captures). The red oval locates a potential linkage zone across Highway 3.

18 Figure 4. Close up of Purcell Mountain Highway 3 area grizzly bear probability of occurrence model. The red oval locates a potential linkage zone across Highway 3.

19