UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Landscape spatial patterns and forest fragmentation in managed forests in southeast British Columbia… D’Eon, Robert George 2002

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2003-792110.pdf [ 5.86MB ]
Metadata
JSON: 831-1.0075054.json
JSON-LD: 831-1.0075054-ld.json
RDF/XML (Pretty): 831-1.0075054-rdf.xml
RDF/JSON: 831-1.0075054-rdf.json
Turtle: 831-1.0075054-turtle.txt
N-Triples: 831-1.0075054-rdf-ntriples.txt
Original Record: 831-1.0075054-source.json
Full Text
831-1.0075054-fulltext.txt
Citation
831-1.0075054.ris

Full Text

L A N D S C A P E SPATIAL PATTERNS A N D FOREST F R A G M E N T A T I O N IN M A N A G E D FORESTS PN SOUTHEAST BRITISH COLUMBIA: PERCEPTIONS, MEASUREMENTS, A N D S C A L E by ROBERT GEORGE D ' E O N H.B.Sc.F., Lakehead University, 1987 M.Sc.F., The University of New Brunswick, 1992 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSPOPHY in THE F A C U L T Y OF G R A D U A T E STUDIES Faculty of Forestry We accept this thesis as conforming to thej)E€5tiuired standard THE UNIVERSITY OF BRITISH COLUMBIA December 2002 © Robert G. D'Eon, 2002 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of f p £ g 3 / <S Ol S The University of British Columbia Vancouver, Canada Date DE-6 (2/88) A B S T R A C T Forest spatial patterns are a central topic in contemporary landscape ecology, largely because of concerns about forest fragmentation. Forest fragmentation is thought to be a major threat to biodiversity because remnant forest patches, left from human disturbances such as logging, would support fewer species and be more prone to local extinctions because they are small and isolated from each other, as predicted from an extension of island biogeography theory. These and other theoretical predictions stemming from the forest fragmentation paradigm remain virtually unchallenged by empirical data. I investigated landscape spatial patterns in managed forests of the Slocan Valley in southeast British Columbia, and focused my investigations on theoretical predictions concerning forest fragmentation and the distinction between habitat amount and spatial configuration effects. I first investigated human perception of fragmentation to assess the usefulness of current methods in quantifying landscape spatial pattern and to investigate definitions and confusion about fragmentation. I then used traditional landscape indices to test predictions about fragmentation trends in the Slocan Valley by focusing on the effect of forest harvesting on old growth forest fragmentation. I then created a unique method of assessing landscape connectivity, the inverse of fragmentation, using a scale-dependent, organism-centered technique based on an organism's ability to move between habitat patches. Finally, I tested mule deer scale-dependant selection of forest edges, patch size, and logging roads relative to amount of forest, since these landscape elements are implicated in the fragmentation issue and are either untested or unresolved for mule deer. I found people associate fragmentation with high patch density, which was highly correlated with amount of harvesting, illustrating the confusion between habitat amount and spatial configuration. Landscape indices were of very limited use in deriving absolute values of fragmentation, and are likely best used to compare landscapes and pattern trends. I found little evidence of an old growth forest fragmentation trend in the Slocan Valley. Most predictions concerning a fragmentation trend were falsified. Using an organism-centered method to assess connectivity among old growth patches, I found the landscape to be accessible to all old growth associates at maximum dispersal distances, with the exception of the northern flying squirrel (Glaucomys sabrinus). At median dispersal distances however, only larger more vagile carnivorous birds could access all old growth patches in the landscape. Of particular concern are flying squirrels which had access to only 10 % of the landscape at median dispersal distances. Mule deer displayed selection of landscape elements at the landscape scale only. The best predictors of mule deer winter use were mature forest patch size and amount of mature forest. Because of high correlation between these two variables, distinction between them was difficult and illustrates this persistent problem in empirical work. Empirical field studies are direly needed to test the existing fragmentation theoretical framework. Future work must distinguish between habitat loss effects and independent fragmentation effects. ii T A B L E OF CONTENTS Page Abstract 1 1 Table of contents 1 1 1 List of tables v List of figures v l Acknowledgements v u Preface v i i i Dedication 1 X CHAPTER 1: Forest landscape patterns and fragmentation: a review and introduction 1 CHAPTER 2: Human perceptions of landscape patterns and fragmentation: implications to landscape management 7 Methods 7 Results 11 Discussion 14 CHAPTER 3: Influence of forest harvesting on landscape spatial patterns and forest fragmentation trends in southeast British Columbia 17 Study area 19 Methods 22 Experimental design 22 Landscape index calculations 22 Data analysis 23 Results 27 Discussion 38 CHAPTER 4: Landscape connectivity as a function of scale and organism vagility in a real forested landscape 44 Methods 47 Landscape delineation 47 Distance to edge calculations 48 Connectivity measures and trends 50 Species associations 50 Statistical analyses 55 Results 55 Discussion 62 iii Page CHAPTER 5: Scale-dependant use of six landscape features in a managed forest: the case of mule deer in southeast British Columbia 68 Study area 70 Methods 73 Radiotelemetry 73 Data management and analysis 73 Results 78 Radiotelemetry 78 Resource use 78 Relative selection of landscape features 79 Power analysis 79 Discussion 90 CHAPTER 6: Conclusions and directions 93 Literature cited 98 iv LIST OF TABLES Page Table 2.1. Responses from 30 workshop participants in a study of human perceptions of landscape patterns 13 Table 2.2. Landscape indices for five landscapes used in a study of human perceptions of landscape patterns 14 Table 3.1. Landscape indices used in pattern analyses for 44 landscapes in the Slocan Valley Basin 27 Table 3.2. Mean old growth patch indices of landscapes with no forest harvesting and high amounts of harvesting 31 Table 3.3. Linear regressions between old growth patch indices and amount of harvesting 32 Table 3.4. Mean patch indices among old growth, harvest, and wildfire Patches 33 Table 3.5. Mean coefficient of variation among old growth, harvest, and wildfire patches 34 Table 3.6. Component factor loadings for principle components analyses of old growth, harvest, and wildfire patch indices 35 Table 3.7. Mantel's test of spatial correlation for landscape pattern indices 36 Table 4.1. Estimated dispersal abilities and associated landscape availability for selected old growth associates in the Slocan Valley Basin 54 Table 5.1. Pearson correlations for six landscape parameters describing used and available winter home ranges of seven mule deer in Lemon Creek, British Columbia 88 Table 5.2. Results of six bivariate logistic regressions for 6 landscape features describing used and available home ranges of seven mule deer in Lemon Creek, British Columbia 89 V LIST OF FIGURES Page Figure 2.1. Five landscapes used in a human perception study of landscape patterns 11 Figure 3.1. Landscapes and patches used in a landscape pattern study within the Slocan River Valley 22 Figure 3.2. Examples of regression analyses between amount of old growth and old growth patch indices, and amount of harvesting and old growth patch indices 37 Figure 3.3. Regression analyses between logging road density and proportion of landscapes harvested and three spatial principle components 38 Figure 4.1. Harvest, old growth, and recent wildfire patches within the Slocan Valley ... 52 Figure 4.2. A n example old growth habitat cluster illustrating old growth habitat patches and the cluster boundary 53 Figure 4.3. Patch size frequency distributions for harvest, old growth, and wildfire patches in the Slocan Valley Basin 57 Figure 4.4. Critical distance versus mean distance to edge for harvest, old growth, and wildfire patches in the Slocan Valley 58 Figure 4.5. Critical distance versus DTE_L/CD curves for nine landscapes in the Slocan Valley 59 Figure 4.6. Amount of old growth forest and old growth harvest rate versus 8 for old growth forest patches among nine Landscape Units in the Slocan Valley 60 Figure 4.7. Proportion of the landscape accessible as a function of an organism's ability to move in the Slocan Valley Basin 61 Figure 5.1. Lemon Creek drainage study area in southeast British Columbia 72 Figure 5.2. Mean weekly elevations of four mule deer tracked with GPS radiotelemetry in Lemon Creek drainage 77 Figure 5.3. Mean seasonal home range size of seven mule deer in Lemon Creek 77 Figure 5.4. Landscape-scale use of mature forest and early serai vegetation for seven mule deer in Lemon Creek drainage 81 Figure 5.5. Landscape-scale use of early serai edge and roads for seven mule deer in Lemon Creek drainage 82 Figure 5.6. Landscape-scale use of mature forest and early serai patches for seven mule deer in Lemon Creek drainage 83 Figure 5.7. Home range scale use of early serai vegetation and mature forest for seven mule deer in Lemon Creek drainage 84 Figure 5.8. Home range use of early serai edges and logging roads for seven mule deer in Lemon Creek drainage 85 Figure 5.9. Home range use of mature forest and early serai patches for seven mule deer in Lemon Creek drainage 86 Figure 5.10. Statistical power to detect differences between means of available and used landscape features by mule deer in Lemon Creek drainage 87 vi A C K N O W L E D G E M E N T S I am indebted to Slocan Forest Products Ltd., Slocan Division, for their continual and unfettered support of my research. In particular I thank Alex Ferguson for his continual support from the outset. I am also grateful to Kokanee Forests Consulting for permitting me the freedom to pursue this research within their infrastructure. I am sincerely grateful to my committee members Marie-Josee Fortin, Susan Glenn, and John Nelson and for their valuable feedback, inspiration, and continual supply of ladders, lifelines, and bridges. I am also grateful to my final examiners Gary Bradfield, Hamish Kimmins, and Rob Rempel, for taking the time to impart their wisdom on me. I thank the U B C Faculty of Forestry and Centre for Applied Conservation Research for logistical support. I received personal financial assistance from the University of British Columbia, the National Engineering Research Council of Canada, and a Canfor Corporation fellowship in forest wildlife management. Mule deer radiotelemetry work was funded by Forest Renewal British Columbia funding to Slocan Forest Products. GIS support was provided to me by Dan Mack (Chapters 2 and 3), Ian Parfitt (Chapter 4), and Vince van Tongeren (Chapter 5). Several reviewers provided invaluable input that indisputably improved previous drafts: Hamish Kimmins (Chapter 2), Stephen Sheppard (Chapter 2), Walter Reid (Chapter 3), Dan Rosenberg (Chapter 3), Glenn Sutherland (Chapter 4), and Steve Wilson (Chapter 5). I owe thanks to many people for their input along the way and especially Lenore Fahrig for her discussions on confounding area effects, and Rob Serrouya for his insight into radiotelemetry research. vii PREFACE This dissertation represents the combined work from several related, but distinct, research investigations. A l l of the work stems from on-going research into landscape patterns and organism movement in a forested landscape in the Slocan Valley of southeastern British Columbia, managed primarily by Slocan Forest Products Ltd. for timber and related non-timber resources. Chapters 2 through 5 are presented in publication style format with independent introductions, results, conclusions, and discussions. As a result, each of these chapters are self-contained and may be read independently, but build upon each other and therefore may be read consecutively for added meaning. Chapters 1 and 6 provide an overall introduction and conclusion to the entire dissertation, however specific conclusions pertaining to results are directly discussed within the corresponding chapter. Chapter 1 is presented as a general review of the state of landscape ecology research, particularly pertaining to forest fragmentation in North America. A manuscript derived largely from Chapter 1 was published in The Forestry Chronicle (D'Eon 2002). Chapter 2 represents the results of a study into human perceptions of forest fragmentation that I performed at an ecosystem management conference in October 1998 (D'Eon et al. 2000). This work was also published in The Forestry Chronicle (D'Eon and Glenn 2000). Chapter 3 ' is a study of landscape patterns in managed forests and the effects of forest harvesting on old growth fragmentation patterns. A manuscript from this chapter was submitted for publication in Forest Ecology and Management and is currently in review as of December 2002. Chapter 4 is an investigation into landscape connectivity, the inverse of forest fragmentation, from an organism-centered perspective using a novel quantified approach. This work was published in Conservation Ecology (D'Eon et al. 2002). Chapter 5 represents the results of two years of radiotelemetry data collected on free-ranging mule deer (Odecoileus hemionus), and focuses on selection and avoidance of landscape features created by a predominantly mature forest matrix with interspersed clear-cut harvest patches. This work builds upon on-going mule deer radiotelemetry work in this landscape (D'Eon 2001, D'Eon et al. 2002, D'Eon et al 2003). Results incorporating all final data will therefore be submitted for publication at the conclusion of the study. Finally, Chapter 6 is presented as a general conclusion to the dissertation where I suggest directions for future work. viii To my mother, for encouraging me to pursue my dreams To my father, may he forever be at peace ix CHAPTER 1 FOREST L A N D S C A P E PATTERNS A N D FRAGMENTATION: A REVIEW A N D INTRODUCTION Landscape spatial patterns in managed forests are a central topic in contemporary landscape ecology. The underlying assumption of landscape ecology is that spatial patterning of landscape elements is functionally linked to ecological processes occurring in landscapes (Turner 1989). Interest in forest landscape patterns is largely fueled by concern over one of the most important landscape ecology and conservation issues of recent times, habitat fragmentation (Villard 2002). In turn, concern over habitat fragmentation is no doubt most acute among those dealing with forests because of the stark and dramatic effects of forest harvesting and the extent to which forests, and especially old forests, pervade social conscience (McGarigal and Cushman 2002, Boutin and Hebert 2002). Forest fragmentation, the process of dividing a forest into smaller and more isolated fragments, is commonly thought to originate as a hypothetical suggestion stemming from MacArthur and Wilson's (1967) pivotal work on the theory of island biogeography (Haila 2002). It was suggested that remnant patches of habitat left from human disturbances such as deforestation were comparable to the oceanic islands used by MacArthur and Wilson (1967) to formulate their theory. In the ensuing continental extension of the theory, remnant habitat patches would support fewer species and be more prone to local extinctions, as predicted by MacArthur and Wilson (1967) for small isolated islands - the ultimate result being a reduction in biodiversity. Predicted negative ecological effects of habitat fragmentation have been widely cited (Harris 1984, Saunders et al. 1991, McGargial and McComb 1995, Forman 1997). These predicted effects primarily involve a reduction in patch size, increase in distance between patches, and increased amounts of edge, which ultimately results in local extinction through loss of species dependant on forest interior, isolation effects such as reduced immigration rates, and edge effects such as increased nest predation (Bunnell et al. 1999). Much of the early support for predicted biogeographical effects (i.e., effects related to spatial configuration of habitat patches) in remnant habitat patches came from empirical studies demonstrating correlations between lower species richness and smaller forest 1 fragment size in deforested areas of the eastern United States (e.g., MacClintock et al. 1977, Whitcomb 1977). However, as raised by Helliwell (1976) then Haila (2002), this ensuing species-area relationship says nothing about the effects of isolation, a central point in MacArthur and Wilson (1967), and simply supports the universal axiom that area is related to species richness - a fact seemingly neglected in the fragmentation literature (Haila 2002). Kuhn (1970) referred to a scientific paradigm as a body of scientific achievements and set of beliefs that provides model problems and solutions to a community of practitioners. The fragmentation paradigm, in this sense, has widely proliferated and been applied to virtually every forest ecosystem including the managed forests of western and boreal North America (Harris 1984), to the extent that forest management agencies in many jurisdictions have instituted regulations devised to mitigate fragmentation impacts (e.g., B C Ministry of Forests and BC Ministry of Environment, Lands and Parks 1995). In a prime example, a central tenet of British Columbia's actions in defense of forest fragmentation were guidelines advocating the importance of forest connectivity via the establishment of forested corridors, a strategy widely accepted and recommended by many (Simberloff et al. 1992). However, as highlighted by Haila (2002), what makes oceanic islands evolutionarily and ecologically unique has little relevance to forest fragments created by human disturbance, and makes the extension of island biogeography theory to continental forests, at best, questionable. In particular, managed forests of western and boreal North America have physical attributes that clearly distinguish them from association with MacArthur and Wilson's (1967) oceanic metaphor. The most important of these is that unlike the more impassable stretches of water separating islands, spaces between forest fragments in managed forests are, to varying degrees, passable by many organisms and may or may not present significant barriers to organism movement through the landscape. This conceivably creates large differences between connectivity among islands and connectivity between forest fragments and thus, large differences in the isolation effect of MacArthur and Wilson (1967). As well, as Haila (2002) points out, the temporal evolutionary forces acting upon species communities on islands cannot be compared to those within temporally and spatially dynamic forest fragments created by human disturbance in managed forests. Interestingly, the pervasion of the forest fragmentation paradigm continues despite a paucity of empirical evidence (MacGarigal and Cushman 2002), no doubt due to a sense of 2 urgency to a problem perceived to be of crisis proportions. Most of what we know about forest fragmentation and its effects is based on speculation and untested theory (e.g., Higgs 1981, Harris 1984, Simberloff and Cox 1987, Noss 1987, Saunders et al. 1991, Kareiva and Wennergren 1995, Fahrig 1997 and 2002, With and King 1999). The somewhat sparse scientific evidence generated thus far on the ecological effects of fragmentation is inconclusive at best (Debinski and Holt 2000) because it appears that fragmentation effects are at least as likely to be positive as negative - possibly due to edge effects which can be beneficial or detrimental depending on the species and context (Kremsater and Bunnell 1999). The wide chasm between theory and evidence in this case is likely attributed to two principal causes: (1) the extreme difficulty of conducting good landscape-scale experiments, and (2) a persistent lack of differentiation between habitat loss and fragmentation effects. Well-designed empirical landscape ecology studies, particularly those addressing fragmentation, are rare (McGarigal and Cushman 2002). In science, the strongest conclusions are obtained from manipulative experiments with proper replication and control (Hurlbert 1984) and follow the scientific method of observation, hypothesis formation, and experimentation (Popper 1959, Piatt 1964, Romesburg 1981). McGarigal and Cushman (2002) described the ideal field experiment in the study of fragmentation as having the following features (edited for brevity): (1) experimental units are structurally similar landscapes of the same size, (2) landscape sizes are functionally relevant to the process or organism under consideration, (3) treatments permit manipulation of area and fragmentation effects so that the two can be isolated experimentally, (4) treatments are adequately replicated and randomly assigned to the experimental units, (5) experimental design includes adequate temporal and spatial controls, (6) treatments are implemented systematically to experimental units to preclude bias from natural temporal variation, and (7) the post-treatment sampling period is long enough to ensure real treatment effects are observed. Ironically, McGarigal and Cushman (2002) were not aware of any such experiment, and conceded that it may be impossible due to real-world complexities. Having recognized these difficulties, Hargrove and Pickering (1992) along with McGarigal and Cushman (2002) advocate mensurative experimentation as the most promising source of empirical data in 3 fragmentation studies by acknowledging design deficiencies and examining other evidence to distinguish cause and effect (see also Oksanen 2001). The second cause, the confusion between habitat loss and fragmentation effects, is widespread in the fragmentation literature (Fahrig 2002, McGarigal and Cushman 2002, Schmiegelow and Monkkonen 2002). Habitat loss concerns the net amount of habitat area lost to a particular disturbance; fragmentation effects, such as isolation and edge effects, concern the spatial configuration of habitat independent of habitat loss. Because the two occur simultaneously in virtually every case, confusion persists. Prior to a shift from thinking of fragmentation as a combination of both, to thinking of fragmentation as only a spatial configuration phenomenon (Haila 2002), much of what was referred to in the literature as fragmentation effects included both habitat loss and fragmentation effects (e.g., Diffendorfer et al. 1995, Holt et al. 1995, Robinson et al. 1995, Schumaker 1996). Andren (1994) and Fahrig (1997) brought the distinction between habitat loss and fragmentation effects to the fore. When the distinction has been made in empirical studies, most fragmentation effects were far outweighed by the effects of habitat loss (McGarigal and McComb 1995, Meyer et al. 1998, Trizcinski et al. 1999, but see Villard et al. 1999). However, several predict a threshold effect whereby fragmentation effects, separate from habitat loss effects, can have increasingly large influences on species persistence as habitat amount declines to small amounts (Fahrig 1997 and 2002, With and King 1999). This notion remains untested, however, and the degree to which fragmentation effects occur, i f at all, is clearly unresolved. Hagar and McCoy (1998) cautioned against the acceptance of untested hypotheses because the indiscriminate application of conservation paradigms may lead to misguided research efforts and poor management guidelines (also see Schmiegelow and Monkkonen 2002). Indeed, consensus is building that a current fixation with spatial configuration of habitat elements, particularly related to the issue of fragmentation, is misguided since habitat loss may be more important (Fahrig 1997, McGarigal and Cushman 2002, Schmiegelow and Monkkonen 2002). Resources spent on addressing issues of spatial configuration could represent wasted effort and a deflection from more pressing concerns. In fact, in a departure from current North American forest practices, Boutin and Hebert (2002) recently advocated, as a working hypothesis, that habitat amount should be the sole focus of forest management 4 efforts until habitat loss is projected to be 70-80%, at which point spatial configuration of habitat should also be considered. Regardless of current management direction, there is clearly a need for empirical fragmentation studies to establish the validity of a rich, yet mostly unchallenged, theoretical framework. In this dissertation I investigated landscape spatial patterns in managed forest landscapes in southeast British Columbia and focused on points related to forest fragmentation. Due to a historical lack of definition and confusion in the fragmentation literature, I first examined anthropocentric perceptions of fragmentation to investigate their role in forest landscape research and management (Chapter 2). I used maps of real landscapes illustrating a variety of harvest patterns to determine how forest managers perceived forest fragmentation, and to examine the usefulness of traditional landscape metrics in the context of qualitative and intuitive human perception. I specifically focused on answering the question: are landscape indices more useful in arriving at conclusions of acceptable accuracy to the user, than intuitive human perception in classifying landscape spatial pattern? I then used landscape metrics to quantitatively test theoretical predictions and widely-held assertions about forest patterns and fragmentation in real managed forests (Chapter 3). I tested predicted relationships among patterns of forest patches because of the underlying assumption that for fragmentation effects to occur, spatial patterns of patches must first be consistent with theoretical predictions. I specifically focused my tests on predicted fragmentation trends concerning old growth forest patches in relation to forest harvesting, because of the conservation importance of this issue, and structured the tests based on three general questions: (1) Is forest harvesting fragmenting the existing forest pattern? (2) Is forest harvesting imposing a new fragmented pattern on the landscape? (3) Do logging road patterns affect fragmentation by influencing harvest patch configuration? I followed the advice of Hargrove and Pickering (1992) and McGarigal and Cushman (2002) by targeting landscapes of similar size and structure as my experimental units, in a controlled and replicated mensurative experiment. I also addressed confounding area affects and distinguished between habitat area and spatial configuration as recommended by Fahrig (1997 and 2002). 5 In a departure from more traditional methods, I then investigated fragmentation impacts in real landscapes by focusing on landscape connectivity, the inverse of fragmentation (Chapter 4). In these investigations I devised a novel method of quantifying landscape connectivity specific to an organism and its ability to travel between patches. In this way, I considered fragmentation effects at the scale of the interaction of an organism and the landscape, rather than more traditional anthropocentric methods of addressing fragmentation. I assessed landscape connectivity across multiple scales based on a range of critical distances representing movement capabilities of selected species. My objectives were to: (1) derive a multi-scale measure of landscape connectivity related to organism vagility, (2) detect critical connectivity thresholds in real landscapes, (3) test a hypothesis that commercial forest harvesting reduces connectivity among old growth patches, and (4) predict consequences to select species related to landscape connectivity in these landscapes. Finally, I used mule deer radiotelemetry data collected within one of my study landscapes to investigate the response of an organism to spatial landscape features often associated with fragmentation, relative to habitat amount, and at different scales (Chapter 5). Mule deer are often associated with interspersion of forest and early serai patches, and the ensuing edge created by this interspersion, and should therefore predictably be influenced by fragmentation effects in this landscape. I therefore tested seasonal selection of logging roads, forest edges, and forest patch size by radiocollared mule deer in my study area. I chose these landscape elements because of their association with predicted mule deer habitat and their obvious association within the forest fragmentation paradigm. Because total habitat amount may be more important than spatial configuration of landscape elements (Fahrig 1997) I concurrently tested seasonal selection of amounts of mature forest and early serai vegetation and investigated their relative influence. Further, because resource selection is predicted to occur at a hierarchy of scales (Johnson 1980, Senft et al. 1987) and is sensitive to the scale of habitat availability (McCLean et al. 1998), I focused my tests on comparisons between use and availability of these landscape features at the landscape (2nd-order selection from Johnson [1980]) and seasonal home range scales (3rd-order selection from Johnson [1980]) to test a hypothesis that selection of these features is scale-dependant. 6 While somewhat diverse and distinct, my specific intent in all of these investigations was to use real data on real landscapes to test theoretical predictions stemming from the fragmentation paradigm. 7 CHAPTER 2 H U M A N PERCEPTIONS OF L A N D S C A P E PATTERNS A N D FRAGMENTATION: IMPLICATIONS TO L A N D S C A P E M A N A G E M E N T A landscape can be defined as a mosaic where the mix of local ecosystems or land uses is repeated in similar form over a kilometers-wide area (Forman 1997). With the advent of tools such as geographic information systems and satellite imagery, viewing landscapes over broad scales has become commonplace. Typically, people believe a landscape should look a certain way without questioning the necessity of that appearance (Nassauer 1988). Humans tend to construct and manage landscapes by making decisions based on what they see and know (Nassauer 1995a,b). Hundreds of quantitative measures of landscape pattern have been proposed (Gustafson 1998). However, accepted quantitative definitions of central concepts in landscape ecology such as forest fragmentation and landscape connectivity remain elusive (Hulshoff 1995, Schumaker 1996, Davidson 1998, Hargis et al. 1998). In this study I investigated human perception of landscape spatial pattern in comparison to more quantitative expressions. I specifically focussed on the question: are landscape indices more useful in arriving at conclusions of acceptable accuracy to the user, than intuitive human perception in classifying landscape spatial pattern? METHODS Thirty resource professionals were enrolled in a 3-hour landscape metrics workshop at an ecosystem management conference in Nelson, British Columbia, 26-28 October, 1998 (D'Eon et al. 2000). Workshop participants were given a 30-minute background lecture focusing on basic principles of landscape ecology. In particular, the issue of forest fragmentation was defined and discussed in the context of how fragmentation may affect landscape pattern. As summarized by Forman (1997; 407), predicted changes include increases in patch number, total boundary length, habitat loss, and habitat isolation, and decreases in average patch size, total interior habitat, and connectivity. A similar exercise was delivered in February 1999 to 38 third-year undergraduate students enrolled in a Bachelor of Science program at the University of British Columbia, Department of Forest 8 Sciences. Students were grouped into three different laboratory sessions of 10, 15, and 13 students per session and therefore treated as separate groups for analytical purposes. Participants were shown five maps (Figure 2.1) of different landscapes ranging from 3,546 to 9,956 ha. Maps were plotted at 1:50,000 scale and mounted on a wall side by side. Each map was plotted on a white background and contained a landscape boundary (determined using ecological boundaries such as heights of land) drawn as a thick black line, creeks drawn as thin blue lines, existing clear-cut harvest patches less than 40 years old shaded in brown, and logging roads indicated as thin dashed black lines (see Figure 2.1). To avoid confusion between fragmentation and habitat loss effects (Fahrig 1997), landscapes were chosen on the basis of similar proportions of total patch area (range = 15 to 24 %), but differing spatial arrangements of patches (Table 2.1). Landscape maps were based on forest inventory data and generally encompassed a single drainage system in steep, mountainous terrain in the Slocan Valley of southeast British Columbia (49°42'N, 117°42'W). The data source for maps was British Columbia 1:20,000 provincial forest cover map series in digital format. Maps were plotted using ARClTSfFO™ software. Under the assumption that existing harvest patches represent future forest spatial patterns, I asked the following question of each participant for each landscape: "Is the illustrated patch pattern in this landscape fragmented?" At this point, a yes or no response by each participant was based on visual inspection of maps and personal perception and opinion. To simulate an individual manager working in relative isolation, discussion of opinions was not permitted until after responses were collected. Following the background lecture, the remaining time was dedicated to a practical exercise where participants, working in teams of five or six people, manually (i.e., without the use of a computer) calculated the following landscape and patch indices for one assigned landscape: proportion of landscape forested (total landscape area and amount of forest land provided), number of patches, patch density (number per 100 ha of forested land), total patch area, mean patch size, total perimeter length, mean patch perimeter, total edge area (assuming 50-m edge effect on each side of patch perimeter), mean patch to edge area ratio, total core area (area of patch inside of 50-m edge effect), mean core area, mean shape index, mean fractal dimension, mean nearest neighbour (minimum edge to edge distance between patches), patch dispersion, and logging road density (kms per 100 ha of forested land). Mean 9 shape index (Patton 1975) and fractal dimension (Ripple et al. 1991) are variations of an area to perimeter ratio where 1.0 represents a perfect shape (perfect circle or straight line) and larger numbers represent increasing departure from perfect shapes and increased shape complexity. Patch dispersion (Clark and Evans 1954) is a measure of non-randomness of patch arrangement and departs froml .0 which represents a random pattern, to < 1.0 indicating a trend towards patch aggregation, or > 1.0 indicating a trends towards regular or uniform spacing. Area was calculated with dot grids, linear distances were measured with scales and analogue clinometers, and arithmetic calculations were performed using hand calculators. Following this exercise, numerical results of each index for each landscape were posted beside the appropriate landscape map. Values posted were those calculated within a geographic information system prior to the workshops to ensure correct values were considered. Participants were then asked to review and discuss the results. In light of this newly provided quantitative data, and information and skills gained in the practical exercise, participants were again asked the same previous question of each of the same landscapes: "Is the illustrated patch pattern in this landscape fragmented?" The degree of fragmentation within a landscape was ranked as a function of the number of individual yes responses. In this way, the landscape with the highest number of yes responses was ranked as most fragmented. Conversely, the landscape with the lowest number of yes responses was ranked as least fragmented. Association between indices and combined group rankings was determined with Pearson correlation and differences between responses before and after index calculations were tested for statistical significance using the Wilcoxon signed rank test (Zar 1984, SPSS 1996). 10 Figure 2.1. Five landscapes used by workshop participants to visually and quantitatively assess landscape patterns at a professional landscape metrics workshop held in Nelson, British Columbia (D'Eon et al. 2000) and in undergraduate student laboratories (Univ. British Columbia, Forest Sciences Dept., Feb. 1999). Maps were in colour and plotted at 1:50,000 scale. 11 RESULTS From combined responses, Landscape 4 was ranked as most fragmented, followed by Landscapes 1, 5, 3, and 2 in descending order of degree of fragmentation (Table 2.1). There was no significant difference between combined responses and rankings of fragmentation before and after calculating landscape indices (Wilcoxon signed rank test: z = -1.744, P = 0.081; Table 2.1). Pearson correlation analyses revealed highly significant associations between number of patches (r = 0.944), patch density (r = 0.937), and patch shape (r = -0.907), and fragmentation rankings (all P < 0.05; Table 2.2). Landscape 2 had the lowest number of patches, lowest patch density, and highest shape index, and was therefore consistently perceived to be not fragmented (Table 2.2). Conversely, landscapes 4 and 1 had the highest number of patches, highest patch densities, and lowest shape indices, and were consistently perceived to be fragmented. The remaining three landscapes had index values ranging between these extremes, and resulted in mixed perceptions of fragmentation among and within groups (Table 2.1). In four cases (out of 150 individual response pairs), within the professional workshop, a participant changed their response between the first and second time of questioning (Table 2.1). A l l 30 workshop participants worked in a forest resource professional capacity within Canada, with 26 located within British Columbia. Two participants worked in Alberta, 1 in Ontario, and the remaining person in the Yukon Territory. By working sector, 11 worked within the industrial sector (e.g., forest licensee), nine government, eight forestry consulting, and the remaining two worked in the academic sector. Within the student laboratories, 32 cases (out of 190 individual responses pairs) occurred where a student changed their initial response; a significantly higher proportion than that within the professional workshop (%2 = 17.79, P < 0.001). A l l participants (professional and student combined) had some previous exposure to basic principles of landscape ecology, but no one considered themselves to have advanced knowledge of the subject. 12 Table 2.1. Responses from 30 professional workshop participants (D'Eon et al. 2000) and 38 undergraduate students (Univ. British Columbia, Forest Sciences Dept., Feb. 1999) before and after a practical exercise on calculating a suite of landscape indices for 5 landscapes. The question asked: is the patch pattern illustrated in this landscape fragmented? Results indicate no significant difference between responses before and after index calculations (Wilcoxon signed rank test: z = -1.744, P = 0.081). Number of yes responses Group t Landscape8 Before After 1 27 26 Professional 2 0 0 (n=30) 3 5 2 4 30 30 5 14 14 1 10 0 Student-1 2 0 0 (n=10) 3 1 1 4 10 10 5 8 0 1 15 14 Student-2 2 0 0 (n=15) 3 7 12 4 14 12 5 9 10 1 13 13 Student-3 2 0 0 (n=13) 3 6 5 4 13 13 5 12 8 1 65 53 Combined 2 0 0 (n=68) 3 19 20 4 67 65 5 43 32 fStudent groups 1 to 3 signify three different laboratory sections. 'Landscapes 1 to 5 correspond to landscapes 1 to 5 in Figure 2.1. 13 Table 2.2. Landscape indices for landscapes used in a professional landscape metrics workshop (D'Eon et al. 2000) and student laboratories (Univ. British Columbia, Forest Sciences Dept., Feb. 1999) for comparison with intuitive human perceptions of landscape pattern. Landscape Landscape Index1 Correlation with rank8 P 1 (2) 2 (5) 3 (4) 4 (1) 5 (3) Total landscape area (ha) -0.793 0.109 6513 8755 8980 3546 9956 Total forested area (ha) -0.257 0.677 5007 4439 5591 2803 9238 Number of patches 0.944* 0.016 28 4 12 29 25 Patch density (#1100 ha) 0.937* 0.019 0.56 0.09 0.21 1.03 0.27 % of forest in patch type -0.279 0.649 24 22 17 15 23 Total patch area (ha) -0.204 0.743 1203 959 922 423 2096 Mean patch size (ha) -0.704 0.185 43.0 239.8 15.1 14.6 48.9 Total patch perimeter (m) 0.405 0.499 97507 49300 13708 48356 26257 Mean patch perimeter (m) -0.693 0.194 3482 12325 1714 1667 3282 Total edge area (ha) 0.405 0.498 947 474 132 466 251 Mean edge to patch area 0.537 0.351 0.79 0.49 1.09 1.10 0.64 Total patch core area (ha) -0.166 0.790 751 724 63 216 273 Mean patch core area (ha) -0.719 0.171 27.8 181.0 12.6 9.8 39.0 Mean patch shape index -0.907* 0.034 1.52 2.31 2.06 1.50 1.53 Mean patch fractal -0.350 0.564 1.27 1.31 1.45 1.32 1.28 Mean nearest neighbour (m) -0.162 0.794 165 30 624 139 113 Patch dispersion 0.205 0.741 0.22 0.01 0.47 0.25 0.08 Log road density (km/100 0.766 0.131 1.34 0.56 0.40 1.09 0.50 tDensities based on total forested area; edge and core area calculated by assuming a 50-m edge effect on either side of patch perimeter; shape index from Patton (1975); fractal dimension calculation from Ripple et al. (1991); patch dispersion index from Clark and Evans (1954). 8Pearson correlation coefficient between combined group rankings and landscape index values. High coefficients signify high association between an index and the degree of fragmentation as determined by survey participants. Significant probability indicated (*) at oc = 0.05. ^Landscapes 1 to 5 correspond to landscapes 1 to 5 in Figure 2.1. Number in brackets is combined ranking of degree of fragmentation (1 = most fragmented, 5 = least fragmented). 14 DISCUSSION In this study, landscapes associated with extreme index values (i.e., smallest or largest value of the five landscapes for a given index) tended to be consistently perceived as either fragmented or not, before and after calculating indices. This is especially true when considering the number of patches, patch density (function of number of patches), and patch shape. This suggests that these very apparent and easily visualized parameters may tend to overshadow other less-obvious parameters. It also suggests that a landscape dominated by a high number of regularly-shaped (e.g., rectangular) harvest blocks was highly associated with what survey participants believe fragmentation to be. Interestingly, there was no significant correlation between fragmentation ratings and amount of harvesting, and the landscape rated as most fragmented (landscape 4, Figure 2.1) had the lowest amount of harvesting as a proportion of total forest area, contrary to an intuitive prediction that harvest amount may influence perceptions of fragmentation. However, since landscapes were specifically selected for their similar amounts of harvesting, conclusions regarding amount of harvesting are unclear in this case. The finding that conclusions about fragmentation generally remained unchanged from initial intuitive perceptions implies that either landscape indices provided little additional information, so that enough information can be collected intuitively to make a correct decision on fragmentation, or what we perceive and call fragmentation is different from what landscape indices measure. In either case, this implies that the current usefulness of landscape indices in helping to define and quantify central concepts in landscape ecology is limited. This situation is perhaps due to a lack of construct validity in landscape metrics work. Construct validity, a concept commonly used in psychology research, concerns the extent to which a test (or landscape metric in this case) measures what its users claim it measures (Johnstone et al. 1997, Chow 1998, Taub 1998). Use of the hundreds of available landscape metrics is often unaccompanied by an explicit description of what each metric measures relative to ecological process. The ecological relevance of landscape metrics in these cases is unclear and left to interpretation. Do we use a landscape metric simply because it is measurable and the data are available, or because it measures what concerns us, in this case 15 fragmentation? This question poses a challenge to clearly define the ecological relevance of landscape metrics in applied ecology work. I do not suggest that the use of landscape metrics is invalid. There are several good examples of the use of landscape metrics in investigating trends and relationships in landscape pattern (Franklin and Forman 1987, Ripple et al. 1991, McGarigal and McComb 1995, Garrabou et al. 1998 ). Indeed, establishing trends in landscape patterns across time and space may be the best utility of landscape metrics at present. However, as quantitative measures of absolute landscape structure relative to ecological process, much more work is required (Carey et al. 1992, Schumaker 1996, Davidson 1998). From an alternative view, the usefulness of landscape indices may not lie in providing a quantitative scale of things such as fragmentation, but rather in providing a means of adjusting the way we view and define them. They may also help us identify our biases in what we perceive as valuable and desirable in landscapes. Ahl and Allen (1996) in discussions of hierarchy theory state that definitions in science are not true or false, but are only more or less useful. They view definitions as a product of an observer's search given a particular intent and perspective (Ahl and Allen 1996;74). They identify two types of definitional entities: those postulated prior to observation, and those derived or modified after empirical observation. In this way, landscape indices can provide useful information in modifying criteria for future work on fragmentation. In this study for example, perceptions of fragmentation were highly associated with number of patches, patch density, and patch shape; suggesting that these particular indices reflect important human values that may be useful in defining landscape structure. However, more work is required to establish the relative roles of human values and ecological significance in the use of landscape indices. Differences in personal experience may also play a role in how landscape patterns are perceived. In this study students were over seven times more likely (odds ratio = 7.4, ln(odds) = 2 + 1.086; SPSS 1996) to change their perception of landscape fragmentation before and after calculating indices. Only 4/150 professional responses versus 32/190 student responses changed after calculating landscape indices. I suggest this is due to stronger-held beliefs and opinions held by professionals with more management exposure and experience. 16 Finally, human perception and cultural values can structure landscapes through landscape management goals based on perception and intuition (Nassauer 1997). In turn, changes in cultural values can result in changes in landscape structure. A good example of this is a historic shift among wildlife ecologists from viewing forest edges as generally beneficial (Leopold 1933) to viewing them as detrimental to many species (e.g., Yahner and Mahan 1996, Harris 1988). This has no doubt contributed to a recent shift in North American forest practices away from small dispersed openings (Franklin and Forman 1987) to a pattern incorporating larger opening sizes (e.g., Province of British Columbia 1996). This landscape management by perception paradigm is enhanced, if not driven, by a lack of reliable quantitative measures of landscape structure. As a result of our inability to quantify central concepts such as fragmentation and connectivity, intuitive human perception will likely continue to have a large influence on management of landscape structure and pattern. 17 CHATPER 3 INFLUENCE OF FOREST HARVESTING ON L A N D S C A P E SPATIAL PATTERNS A N D FOREST FRAGMENTATION TRENDS IN SOUTHEAST BRITISH COLUMBIA. Habitat fragmentation is considered one of the most serious conservation issues of recent decades (Rosenberg et al. 1997) because it is suggested to be one of the most important factors in a current species extinction crisis leading to a loss in biological diversity (Wilcox and Murphy 1985, Gloombridge 1992). Fragmentation is thought to be a problem largely because the theory of island biogeography predicts declines in species richness within habitat fragments as they become smaller and farther apart (MacArthur and Wilson 1967). Species persistence in small fragments, or patches, is predicted to be lower than in contiguous habitat because of higher vulnerability of small populations within small patches to stochastic demographic or environmental events leading to local extinctions (Pimm et al. 1988). Predicted consequences of fragmentation have been widely cited (eg., Harris 1984, Saunders et al. 1991, McGargial and McComb 1995). Much of what is called forest fragmentation is based on human perception rather than rigorously derived conclusions (D'Eon and Glenn 2000). The term "fragmented" is often used to describe landscapes without a clear explanation of what is meant by this term or a quantitative basis for describing a landscape as fragmented. In this case, I echo the concerns of Hager and McCoy (1998) who stated that the acceptance of untested hypotheses has adverse scientific and social consequences by promoting management activities that are based on unsupported assumptions. Commercial forest harvesting is commonly associated with forest fragmentation (Franklin and Forman 1987). It is obvious that forest harvesting generally removes mature forest habitat. Consequently, the number of individuals of a species dependant on this type of habitat will likely go down with large habitat removals within a specific area. However, it is not obvious that changes in spatial patterns of forest elements will be a cause of additional detrimental ecological effects. In fact, the scientific evidence generated thus far is inconclusive at best (Debinski and Holt 2000). This issue is likely related to confusion between the concept of habitat loss, which involves absolute habitat amount, and that of 18 fragmentation, which deals with spatial arrangement of habitat (Andren 1994, Fahrig 1997). This confusion is no doubt fueled by the inseparability of habitat loss and fragmentation in forest management. Any potential fragmentation effects caused by forest harvesting will be accompanied by an associated habitat loss. When the distinction has been made, however, spatial configuration of habitat elements was insignificant compared to habitat amount in deriving ecological effects (McGargial and McComb 1995, Trzcinski et al 1999). In this study I tested predictions concerning forest fragmentation in a managed forest landscape in southeast British Columbia and compared harvest patch patterns to old growth and recent wildfire patch patterns. I structured my tests based on three general questions: (1) Is forest harvesting fragmenting the existing forest pattern? (2) Is forest harvesting imposing a new fragmented pattern on the landscape? (3) Do logging road patterns affect fragmentation by influencing harvest patch configuration? To investigate effects of forest harvesting on existing forest patterns I focused on old growth forest patches because of their relatively high importance to conservation and forest management. I tested the hypothesis that forest harvesting fragments existing old growth by targeting old growth forests and changing their spatial arrangement. If this was true, I predicted that a directional trend in old growth spatial pattern should be observable relative to varying amounts of forest harvesting in landscapes. Specifically, I tested the following six predictions based on Forman's (1997) predicted consequences of forest fragmentation: as the proportion of harvested forest in a landscape increases (1) old growth patch density will increase, (2) old growth patch size will decrease, (3) the ratio of edge area to patch area of old growth patches will increase, (4) core area of old growth patches will decrease, (5) distance between old growth patches will increase, and (6) dispersion patterns of old growth patches will increase. To investigate future forest patterns imposed upon landscapes by harvesting, I focused on spatial patterns of current harvest patches. I did this under the assumption that existing clearcuts regenerate and represent future forest patterns. Specifically, I compared spatial patterns of existing clearcuts to patterns of existing old growth and recent wildfire patches. Patterns of old growth forest and wildfire were assumed to represent the inherent natural heterogeneity of these landscapes. I tested the hypothesis that forest harvesting creates a future fragmented forest pattern. I predicted that i f this was true, then harvest patch 19 patterns will reflect a more fragmented (from Forman 1997) spatial pattern than patterns of existing old growth and wildfire. Further, I tested the hypothesis that forest harvesting creates a simplified and less heterogeneous spatial pattern of forest elements by creating more uniformly shaped and spaced patches (Krummel et al. 1987, Turner and Rusher 1988). If true, I predicted that clearcut patterns will display less variability than patterns of old growth and recent wildfire patches. Finally, I tested the hypothesis that forest harvesting creates a spatial dispersion pattern of forest elements that is markedly different from a natural dispersion pattern by either uniformly spreading out harvesting or confining it to localized areas. If true, I predicted that measures of spatial correlation of harvest patterns would be significantly different than those of old growth and recent wildfire patches. When treated as an element of fragmentation, logging roads are thought to exacerbate forest fragmentation effects by increasing patch density, decreasing patch size, and increasing amounts of edge (Reed et al. 1996, Tinker et al. 1998). However, these findings are based on the assumption that logging roads are an element of fragmentation that divide forests into distinct patches by creating uncrossable barriers to individuals moving through forests. The validity of this assumption is not clear and may be untrue for larger vertebrates that can move across logging roads or use them for increased mobility. The issue is further confounded by wide variability in logging road construction from small temporary access roads that resemble little more than vegetated trails, to large primary roads constructed for long life and heavy vehicle traffic. For these reasons I investigated the influence of logging road patterns on fragmentation by focusing on their influence on harvest patch pattern, rather than treating them as an element of fragmentation. I tested the hypothesis that road density and spatial pattern influence harvest patch size and spacing, and thus fragmentation, by governing harvest allocations based on road networks (Miller et al. 1996). If this was true, I predicted that spatial patterns of harvest patches would be highly associated with road densities. STUDY A R E A Landscape pattern data were derived from a managed forest landscape of 352,253 ha within the Slocan Valley of the Selkirk mountains in southeast British Columbia, Canada (49°N, 117°W; Figure 3.1). Terrain within this mountainous area is generally steep and 20 broken with slope gradients often exceeding 80%. Elevation ranges from 525 m along the main Slocan Valley bottom to 2,800-m mountain peaks. Seventy-five percent of the land area within the defined study area is forested. Forests of this area are within the Interior Subalpine and Southern Columbia regions described by Rowe (1972) and are predominantly within three forest biogeoclimatic subzones described by Braumandl and Curran (1992): Interior Cedar Hemlock Dry Warm subzone at low elevations, Interior Cedar Hemlock Moist Warm subzone at mid elevations, and Englemann Spruce Sub-alpine Fir subzone at higher elevations. Alpine parkland predominates above 2,000-m elevations. Logging within the Slocan Valley began in the late 1800s but was primarily confined to localized selective harvesting. Large-scale commercial logging began around 1950. Side drainages of the Slocan Valley have since been managed for forest harvesting and road building to varying degrees. Many areas, however, within the main valley corridor and a large provincial park have been excluded from forest harvesting. The majority of low elevation areas along the main valley bottom is privately-owned land and has been partially deforested for agricultural and urban development purposes. Private land was excluded from the analyses. Routine forest fire suppression in the area began in the late 1930s (unpublished data, J. Parminter, BC Ministry of Forests,Victoria, BC). 21 METHODS Experimental design The study area was delineated into 44 distinct landscapes (mean area = 8,006 ha; range = 2,735 - 15,479 ha; SE = 442 ha; Figure 3.1). Landscape boundaries were based on drainage patterns and were generally drawn along heights of land that distinguished two adjacent watersheds. Following terminology provided by Hurlbert (1984), landscapes were then considered the experimental unit in a mensurative experiment with treatments based on past forest harvest levels ranging from 0 to 34.5% of the forest in a harvested state. Landscapes with no past harvesting were considered replicate control landscapes. I recognize an inherent violation of an assumption of homogeneity among landscapes within similar treatments, particularly those considered controls, and thus an inherent pseudoreplication problem (Hurlbert 1984). However, I concur with Hargrove and Pickering (1992) that a classical experimental approach is virtually impossible at very broad scales and that careful pseudoreplication leading to induced and qualified conclusions is necessary. To this end I made every attempt to define landscapes based on similar ecology and close proximity. A l l landscapes are within one large drainage basin and many shared common boundaries. Consequently, all landscapes share a common biogeoclimatic zone structure, underlying geology, and climate (Braumandl and Curran 1992). Landscape index calculations A patch can be considered any area that contains at least one attribute that differentiates it from its surroundings. In this study I chose patch types that can be reliably distinguished from aerial photographs and were relevant to the natural disturbance regime in managed forests of southeast British Columbia. On this basis I identified three patch types within the study area: harvest patches (clear cuts harvested within the past 40 years), old growth patches (forest stands > 140 years for Interior Cedar Hemlock dry warm stands and > 250 years for all others; old growth definitions consistent with BC Ministry of Forests and BC Ministry of Environment, Lands and Parks [1995]), and wildfire patches (wildfire within past 40 years). For each landscape and each patch type the following indices were calculated 23 (Table 3.1): total proportion of the forested land within a patch type, patch density (number of patches per 100 ha of forest), core density (amount of interior patch area per 100 ha of forest assuming a 50-m edge effect extending inwards from a patch perimeter), edge density (amount of edge patch area per 100 ha of forest assuming a 50-m edge effect either side of a patch perimeter), average patch perimeter length, average patch area, average patch core area, average patch edge area, median patch area, average nearest neighbor distance (minimum perimeter to perimeter distance between a patch and the closest other patch within a landscape), patch dispersion index, edge ratio (total edge area divided by total patch area), core ratio (total core area divided by total patch area), shape index, fractal dimension, and logging road density. An assumed edge effect of 50 m from a patch perimeter has been used in previous research of this nature (McGargial and McComb 1995) and has been suggested to be a good approximation of edge effects in western North American forests (Kremsater and Bunnell 1999). Patch dispersion (Clark and Evans 1954) is a measure of non-randomness of patch arrangement and departs froml.O which represents a random pattern, to < 1.0 indicating a trend towards patch aggregation, or > 1.0 indicating a trends towards regular or uniform spacing. Mean shape index (Paton 1975) and fractal dimension (Ripple et al. 1991) are variations of an area to perimeter ratio where 1.0 represents a perfect shape (perfect circle or straight line) and larger numbers represent increasing departure from perfect shapes and increased shape complexity. Coefficient of variation (CV) was calculated on applicable data and used as a relative measure of variability from the mean within a landscape. Al l source data were derived from 1998 British Columbia provincial forest cover map information in digital format. Patch indices were calculated within an ARCLNFO geographic information system platform. Data analysis All data analyses were performed using SYSTAT 8.0 (SPSS 1998) statistical software. I considered tests significant at a = 0.05. Non-normal data distributions were assessed using skewness and kurtosis indicators and transformed using logarithm, squareroot, and arcsine transformations to produce more normal distributions. Skewness or kurtosis were considered extreme if + 2 times their standard error did not include zero (SPSS 1998). 24 To test fragmentation predictions for differences in landscape indices between two levels of harvesting, six landscapes with no harvest patches were compared to 12 landscapes with the highest levels of harvesting in this study (13.1 to 34.5% of forest harvested) using student's /-test. To investigate the assumption of homogeneity among control replicates and my ability to detect differences between control and treatment landscapes in these tests, I compared variability in the proportion of old growth in landscapes among control landscapes relative to treatment landscapes. I also investigated the sensitivity of my results to control sample size («). I did this by systematic removals of control landscapes and thus repeated all Mests (Table 3.2) with control n ranging from 6 to 2. I removed control landscapes in three ways: (1) beginning with the most variable landscape (i.e., farthest from the mean) and proceeding to the least variable, (2) beginning with the least variable landscape, and (3) a random removal. Linear regression techniques were employed to test fragmentation predictions concerning relationships between amounts of forest harvesting and fragmentation effects on old growth forest patterns. Relationships between the total amount of area within a patch type and landscape configuration can be highly correlated (Fahrig 1997). To evaluate harvesting and old growth patch configuration relationships, independent of old growth area, I used partial regression analyses to remove area effects (Legendre 1993, McGarigal and McComb 1995, Trzcinski et al. 1999). In this procedure, the amount of old growth within each landscape was regressed against each of the landscape indices measuring old growth patch configuration. Analysis of residuals was then used to remove confounding area effects for indices showing a significant relationship (regression F-test ratio < 0.05) with old growth area. In these cases, residuals were regressed against harvest level to test my predictions. Original index values were regressed against harvest level where no area effects were detected. Relationships that should logically pass through the origin were evaluated for use of a zero-intercept model. In all cases however, full regression models were used (rather than zero-intercept models) due to significant differences in y-intercepts from zero (Kozak and Kozak 1995). Testing for differences in patch patterns and variability among patch types was performed using one-way analysis of variance. Mean values of old growth patch indices were calculated from six control landscapes with zero harvest levels. In this way old growth 25 parameters were derived free of confounding harvest effects on old growth patches. In cases when significant differences among patch type means were detected Bonferroni post-hoc tests were employed to distinguish differences between patch types (Miller 1985). Principal components analyses (PCA) were performed to reduce the number of variables within each patch type for some tests. In PCAs I distinguished between indices reflecting information on habitat amount (e.g., proportion of forest in patch type) and those reflecting information on spatial configuration (e.g., nearest neighbor). Highly correlated variables (r > 0.9) were first excluded to avoid multicollinearity among variables (Tabachnick and Fidell 1996; 84). PCA factors were retained and used in further analyses when eigenvalues were > 1.0. Component loadings matrices were rotated and sorted using a varimax orthogonal rotation (Tabachnick and Fidell 1996; 647). Spatial correlation between landscapes and pattern indices were assessed using the Mantel test (Mantel 1967, Legendre and Fortin 1989, Glenn et al. 1992). The Mantel test is a generalized regression technique that compares two symmetrical difference matrices to investigate associations between individual distances or dissimilarity based on one characteristic with those calculated from a second characteristic. In this study, differences between index values within landscapes were compared to differences in physical distance between landscapes measured as the straight-line distance between landscape centroids. This test was used to determine if patterns among landscapes were associated with landscape proximity (i.e., landscapes close together have more similar patterns than those far apart) and if differences in spatial correlation were associated with specific patch types. Linear regression analyses were used to test predictions concerning road density and landscape spatial pattern. 26 Table 3.1. Landscape indices used in pattern analyses for 44 landscapes within the Slocan Valley Basin in southeast British Columbia. In each case, variables were screened and transformed if closer to normal distributions could be obtained based on skewness and kurtosis indicators. Index name Description PROP Proportion of the forest in a landscape within a patch type P A T D E N Patch density expressed as number of patches per 100 ha of forest within a landscape C O R E D E N Core density expressed as the amount of core area per 100 ha of forest within a landscape, assuming a 50-m edge effect from the patch perimeter E D G E D E N Edge density expressed as the amount of edge area per 100 ha of forest within a landscape, assuming a 50-m edge effect on either side of a patch perimeter A V G P E R M Average patch perimeter length in meters within a landscape A V G A R E A Average patch area in hectares within a landscape A V G C O R E Average patch core area in hectares within a landscape, assuming a 50-m edge effect from a patch perimeter A V G E D G E Average patch edge area in hectares within a landscape, assuming a 50-m edge effect on either side of a patch perimeter M E D A R E A Median patch area in hectares within a landscape N N Average distance between patches and the nearest patch within a landscape based on the closest straight-line perimeter to perimeter distance. DISPER Dispersion pattern index indicating spatial arrangement of patches (Clark and Evans 1954), ranging from 0 to 2.1491. Values close to 0 indicate aggregated spatial pattern, values close to 1 indicate a random distribution, and values approaching 2.1491 indicate even spacing among patches, or maximum dispersion. E D G E R A T Ratio calculated as the amount of edge area divided by the amount of patch area for a given patch type in a landscape CORERAT Ratio calculated as the amount of core area divided by the amount of patch area for a given patch type in a landscape SHAPE An area to perimeter ratio from Paton (1975). A value of 1.0 indicates a perfect circle. The index increases from 1.0 with shape complexity and departure from a perfect circle. F R A C T A L Variation of an area to perimeter ratio using an estimator of the fractal dimension of a line where the fractal dimension = 21ogP/logA; where P = patch perimeter, A = patch area (Krummel et al. 1987, Ripple et al. 1991). A value of 1.0 indicates a straight line and increases with shape complexity to a theoretical maximum of 2.0 where a line becomes plane filling. 27 RESULTS The number of old growth patches within landscapes ranged from 0 to 40; from 0 to 67 for harvest patches; and from 0 to 13 for wildfire patches. Old growth patches were present in 41 of 44 landscapes and accounted for 0.1 to 23.3 % (x = 6.2, SD = 6.0) of the forested area in landscapes with old growth patches. Harvest patches were present in 38 of 44 landscapes and accounted for 1.3 to 34.5 % (x = 9.2, SD = 7.9) of the forested area in landscapes with harvest patches. Wildfire patches were present in 30 of 44 landscapes and accounted for 0.1 to 62.9 % (x = 4.9, SD = 12.3) of the forested area in landscapes with wildfire patches. No differences were observed between mean old growth patch index values for six control landscapes with no harvesting and 12 landscapes containing the highest levels of harvesting in this study (-2.092 < t < 1.16, all P > 0.054; Table 3.2). Variability in the proportion of old growth among control landscapes was lower (x = 4.5%, range = 2.1 to 6.2%, SE = 0.726, C V = 39.8%) than that for treatment landscapes (x = 6.5%, range = 0 to 23.3%, SE = 1.028, C V = 97.6%). A l l /-tests were non-significant in all control landscape removal experiments with the exception of Nearest Neighbour which was significantly lower among control landscapes when control n was 4 in removal experiment (1) and when control n was 2 in removal experiment (3). Significant relationships were observed in 10 of 12 linear regressions between the amount of old growth in landscapes and individual old growth patch indices (PATDEN, COREDEN, EDGEDEN, A V G P E R M , A V G A R E A , A V G C O R E , A V G E D G E , N N , EDGERAT, CORERAT; all R2 > 0.244, all P < 0.001). To eliminate confounding area effects, residuals from the 10 significant linear regressions were then used to test fragmentation predictions against amounts of harvesting, while the two indices without confounding area affects (MEDAREA and DISPER) were used without alteration (Figure 3.2). A l l subsequent linear regressions between old growth patch indices (or their residuals) and the amount of harvesting in landscapes (PROP_H) were insignificant (all R2 < 0.094, all P> 0.054; Table 3.3). When I compared average measures of patch structure among patch types (old growth, harvest, and wildfire) I found a lower mean fractal dimension for harvest patches 28 than old growth patches (Table 3.4). When compared to wildfire patches, harvest patches consistently had higher values for four indices associated with habitat amount: PROP, PATDEN, COREDEN, EDGEDEN (Table 3.4). As well, harvest patches had lower SHAPE and F R A C T A L values compared to wildfire patches (Table 3.4). Harvest patches also had lower average nearest neighbor (NN) distances and dispersion index values than wildfire patches (Table 3.4). Other index values between harvest and wildfire patches were not statistically different. A l l mean values between wildfire and old growth patches were similar with two exceptions. P A T D E N was significantly lower for wildfire patches compared to old patches, and N N was significantly higher for wildfire patches compared to old patches (Table 3.4). Three of six coefficient of variation indices had significant differences among mean values by patch type (Table 3.5). Coefficient of variation indices of harvest patches were not significantly different than those of old growth patches in every case (Table 3.5). Post-hoc tests revealed higher coefficients of variation of harvest patches than wildfire patches for A V G A R E A , A V G C O R E , and A V G E D G E (all P < 0.012). No differences between old growth and wildfire patch coefficient of variation indices were significant. Eleven indices ( A V G P E R M , A V G A R E A , A V G C O R E , A V G E D G E , M E D A R E A , N N , DISPER, EDGERAT, CORERAT, SHAPE, FRACTAL) related to spatial pattern (as opposed to habitat amount) were considered for principal component analyses (PCA) to derive new variables representing spatial patterns for each patch type. To avoid multi-collinearity among variables, five old growth, four harvest, and five wildfire indices were eliminated prior to PCAs due to high correlations with other indices within each patch type (r > 0.9). In this way, six indices were used in a PCA of old growth indices; seven indices in a P C A of harvest patch indices; six indices in a similar analysis of wildfire patch indices (Table 3.6). These PCAs resulted in three new old growth, three new harvest, and two new wildfire principal component indices (SPATAIL 1,2,3_0, SPATIAL1,2_H, SPATIAL1,2,3_F) from an original suite of 33 indices (11 original indices per patch type). Indices related to habitat amount (PROP, PATDEN, COREDEN, A N D EDGEDEN) were not included in principal component analyses to distinguish between habitat amount and spatial configuration of habitat elements. Rather, these indices were sufficiently represented 29 by PROP due to high correlation between PROP and PATDEN, COREDEN, and E D G E D E N within all patch types (all r > 0 . 8 1 5 ) . Mantel's tests using PROP and SPATIAL principal component indices for all patch types resulted in significant spatial correlations among landscapes in all cases but one (all t> 2 . 5 6 3 , all P < 0 . 0 1 0 ; Table 3 .7 ) . In the one exception, no association between the amount of harvesting in landscapes (PROP_H) and distances between landscapes occurred (f = - 0 . 0 0 3 , P = 0 . 9 9 7 ) . Linear regression analyses between amount of harvesting in landscapes (PROP_H) and logging road densities (ROADS) was highly significant (R2 = 0 . 7 9 5 , P < 0 . 0 0 1 ; Figure 3 .3 ) . In contrast, logging road density versus spatial principal component indices for harvesting (SPATIAL 1,2,3_H) were not significantly associated (all R2 < 0 . 1 0 0 , all P > 0 . 0 5 6 ; Figure 3 . 3 ) . 30 Table 3.2. Mean old growth patch indices of landscapes with no forest harvesting and high amounts of harvesting for 44 landscapes in the Slocan Valley drainage of southeast British Columbia.1, Old growth index (see Table 3.1) Harvest Amount X SD t P PROP (SQ) zero 2.07 0.45 -0.564 0.581 high 2.38 1.28 P A T D E N (SQ) zero 0.35 0.04 -1.055 0.307 high 0.46 0.25 C O R E D E N (SQ) zero 1.35 0.30 -0.612 0.549 high 1.57 0.85 E D G E D E N (SQ) zero 2.32 0.54 -0.598 0.558 high 2.67 1.37 A V G P E R M (LG) zero 7.93 0.23 0.073 0.943 high 7.91 0.54 A V G A R E A (SQ) zero 4.53 0.79 -0.0.57 0.956 high 5.64 1.50 A V G C O R E (SQ) zero 2.99 0.75 0.004 0.997 high 2.99 1.35 A V G E D G E (SQ) zero 5.02 0.56 -0.142 0.889 high 5.09 1.18 M E D A R E A zero 10.29 5.59 0.907 0.378 high 8.11 4.26 N N (LG) zero 5.56 1.08 -0.840 0.414 high 6.04 1.16 DISPER (LG) zero -1.70 1.05 -1.720 0.106 high 0.94 0.76 E D G E R A T (LG) zero 0.22 0.14 -0.531 0.603 high 0.30 0.35 CORERAT (AR) zero 0.44 0.08 0.403 0.693 high 0.42 0.16 SHAPE zero 1.84 0.13 -0.826 0.422 high 1.91 0.20 F R A C T A L zero 1.321 0.01 -2.092 0.054 high 1.338 0.02 fZero harvest n = 6; high harvest landscapes defined as top quartile of all landscapes and ranges from 13.1 to 34.5 % of forest harvested; n = 12. Only landscapes containing old growth patches used for mean and ratio index calculations. Data transformations indicated in brackets (SQ = square root; L G = logarithm; A R = arcsine). 31 Table 3.3. Linear regressions between old growth patch indices and the amount of harvesting (proportion of forest in clearcuts < 20 years old) drainage of southeast British Columbia. in 44 landscapes in the Slocan Valley Old Growth Patch Index1 (see Table 3.1) n R2 P P A T D E N R E S (SQ) 44 0.038 0.206 C O R E D E N R E S (SQ) 44 0.000 0.924 E D G E D E N R E S (SQ) 44 0.001 0.856 A V G P E R M _ R E S (LG) 41 0.003 0.747 A V G A R E A R E S (SQ) 41 0.001 0.829 A V G C O R E R E S (SQ) 41 0.004 0.706 A V G E D G E R E S (SQ) 41 <0.01 0.990 M E D A R E A 41 0.008 0.584 NN_RES (LG) 40 0.094 0.054 DISPER (LG) 40 0.073 0.092 EDGERAT_RES (LG) 41 0.047 0.174 C O R E R A T R E S (AR) 41 0.022 0.359 fOnly landscapes containing old growth patches used for average and ratio calculations, and only those containing > 2 old growth patches used for nearest neighbor and dispersion index calculations; sample size therefore varies. "RES" indicates a new variable using residual values from a regression with the original index and the amount of old growth in a landscape to account for confounding area effects. Data transformations indicated in brackets (SQ = square root; L G = logarithm; A R = arcsine). 32 Table 3.4. Mean patch indices among old growth1 (O), harvest (H), and wildfire (F) patches within 44 landscapes in the Slocan Valley Basin of southeast British Columbia.8 Index Patch type mean Post-hoc (see Table 3.1) 0 H F n F P test P PROP (SQ) 2.073 2.625 1.335 94 7.012 0.001* H - O 1.000 H - F 0.001* F - O 0.892 P A T D E N (SQ) 0.354 0.383 0.150 94 20.521 < 0.001* H - O 1.000 H - F < 0.001* F - O 0.025* C O R E D E N (LG) 0.618 0.965 -0.593 94 10.593 < 0.001* H - O 1.000 H - F < 0.001* F - O 0.258 E D G E D E N (SQ) 2.318 2.398 1.169 94 10.082 < 0.001* H - O 1.000 H - F < 0.001* F - O 0.143 AVGPERM(SQ) 52.873 58.551 62.832 77 0.777 0.463 - -AVGAREA(SQ) 4.527 6.179 6.376 77 0.965 0.386 - -AVGCORE(LG) 2.137 2.978 2.881 77 1.521 0.225 - -AVGEDGE(SQ) 5.016 5.691 5.992 77 0.694 0.503 - -M E D A R E A ( L G ) 2.226 2.746 2.617 77 0.603 0.550 - -SHAPE (SQ) 1.355 1.275 1.376 77 5.739 0.005* H - O 0.470 H - F 0.004* F - O 1.000 F R A C T A L (LG) 0.278 0.250 0.273 77 8.007 0.001* H - O 0.046* H - F 0.001* F - O 1.000 E D G E R A T (LG) 0.221 -0.115 0.166 77 3.251 0.044* H - O 0.391 H - F 0.063 F - O 1.000 CORERATIO (AR) 0.444 0.599 0.494 77 3.703 0.029* H - O 0.185 H - F 0.062 F - O 1.000 N N (LG) 5.558 5.554 7.114 67 14.231 < 0.001* H - O 1.000 H - F < 0.001* F - O 0.012* DISPER (LG) -1.703 -1.538 -0.633 67 6.665 0.002* H - O 1.000 H - -F 0.003* F - O 0.068 TOW growth statistics calculated from landscapes with zero harvest level (n = 6). §Only landscapes containing patches within patch type used in analyses. Post-hoc tests performed only in cases of significant F-test result (all degrees of freedom = 2). Significant Bonferroni P indicated (*) at a = 0.05. Data transformations indicated in brackets (SQ = square root; L G = logarithm; A R = arcsine). 33 Table 3.5. Mean coefficient of variation among old growth^ (O), harvest (H), and wildfire (F) patches for 44 landscapes in the Slocan Valley drainage of southeast British Columbia. 8 Coefficient of P a t c h T ™ e variation index (see Table 3.1) 0 H F n F P Post-hoc test Post-hoc P C V A V G A R E A 118.57 130.30 83.25 67 5.320 0.007* H - 0 H - F O - F 1.000 0.006* 0.499 C V _ A V G C O R E 150.67 152.87 103.01 67 5.438 0.007* H - 0 H - F O - F 1.000 0.006* 0.245 C V A V G E D G E 87.47 95.05 60.39 67 4.506 0.015* H - 0 H - F O - F 1.000 0.012* 0.555 C V _ S H A P E 27.05 28.13 20.99 67 2.220 0.117 - -C V F R A C T A L 2.67 2.46 1.97 67 2.710 0.074 - -C V _ N N 110.67 95.62 103.18 58 0.434 0.650 — — f01d growth statistics calculated from landscapes with zero harvest level (n = 6). §A11 degrees of freedom = 3. Only landscapes containing patches within patch type used in analyses. Post-hoc tests performed only in cases of significant F-test result. Significant post-hoc Bonferroni P indicated (*) at a = 0.05. 34 CD T3 00 CD O T 3 CN 13 PH oo KS OH 00 ON c o NO NO r o in 00 c o c o (N ON ON d d d d d d vo ON r - oo NO 00 m oo oo oo p oo c o CN d d d d d d i n oo oo NO In CN in r o o CD > c3 - d i| 2 X 00 00 T 3 '.ti —< i-i O r n <+* t : O to C3 O Cj oo 00 O £ .s a OO CH C3 a« § 13 > • & B g 8 •5 .2 OH 00 H^ (U <2 £ 00 .S .2 oo * a B & o o <H-H T 3 cj cs S S g cO 00 CD o o U rn .S H OH a a> a o I o o OH 00 OJ o t3 OH CD oo CD o •a OH o t H O o r o 13 "is OH oo rN 13 OH c3 00 CO 13 OH 00 CN 13 OH OO OH 00 CD -3 a NO in , — i o NO , — < oo CO CO oo T—< CO CO r o T—1 CN oo o CN 00 p d d d d d d d d CN OO 00 m oo 00 oo CN CN i n r - ON r o i n o CN i n CN r o o oo ON CN r-; d d d d d d d d ^4 r o oo CN 00 o -3- CN r o NO i n o r o i n ON O NO r o CN o ON CO o CN CN m d d d d d d d CO r o O ON t-- r o ON CN NO I— i ON o o d d d d d d r o H m ON CN OO ON ON o in o CN >— 1 o <—i r-- ON d d d d d d < CN ON o r o CN NO CO Tt- r o o ON t - O l> CN d d d d d d Q w w PH OO W OH < OO h-1 H U IX, CN o CN r o in m CN CD 13 CD 00 W oo ON CD r o CN CN ON CO d N ? .s ° ^ s CO ° H ' C W > A l ^ too Table 3.7. Mantel's test of spatial correlation for landscape pattern indices among 44 landscapes in the Slocan Valley drainage of southeast British Columbia. The null hypothesis tested in each case is: no association between differences in index values and differences in geographic distances1, among landscapes. Index8 (see Table 3.6) Z t PROP_0 4688884 2.563 0.010* SPATIAL1_ _0 66046 4.129 O.001* SPATIAL2 _0 67182 4.376 O.001* SPATIAL3_ _0 64803 3.825 <0.001* PROP_H 611037 -0.003 0.997 SPATIAL1 H 57909 5.2497 O.001* SPATIAL2_ _H 57409 5.142 <0.001* SPATIAL3_ H 54561 4.380 O.001* PROP_F 678078 3.156 0.002* SPATIAL1_ F 20909 7.176 <0.001* SPATIAL2_ _F 24383 8.928 <0.001* fDistance measured as the straight-line distance between landscape centroids. 8 PROP = the total proportion of forests in old growth (O), harvest (H), and wildfire (F) patches; SPATIAL refers to variables derived from principal component analyses of 11 spatial pattern indices within each patch type. "P-values calculated using /-distribution with infinite degrees of freedom. Significance indicated (*) at a = 0.05. 36 a p -a 43 o id P . be O (Proportion of old growth in landscape)1' = 0.701, P < 0.001 • .—I c/3 CI <D -o O a 43 o 1-1 o 'iz> o4 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 i 1 1 r 0 J I I L (Proportion of landscape harvested)' R2 = 0.038, P = 0.206 £3 O p. T 3 43 o •s p o S-l M "3 00 o " i r t • T ~ ~ — r •• • • J I I I L 0 1 (Proportion of old growth in landscape)' (Proportion of landscape harvested)2 R2 = 0.039, P = 0.224 R1 = 0.073, P = 0.092 Figure 3.2. Examples of regression analyses between amount of old growth and old growth patch indices (see Table 3.1), and amount of harvesting and old growth patch indices for 44 landscapes in the Slocan Valley of southeast British Columbia. A l l patch index values were regressed against the proportion of old growth (square root transformed) in landscapes (e.g., a, c); i f a significant regression (P < 0.05) occurred with the proportion of old growth (a), the residuals of this regression were used in an a subsequent regression with the proportion of harvesting in a landscape (b) to eliminate confounding area affects; i f a non-significant regression occurred with proportion of old growth (c), the original index was used in a subsequent regression with proportion of harvesting in a landscape (d). 37 (Logging road density)/ R2 = 0.795, P = 0.001 (Logging road density)2 R2 = 0.100, P = 0.056 » C u c o P . o o 13 P . 'o S3 • .—I P . "is p. C/3 1 1 • • • • • • • • • • # - ••••••• • -• • • • • — • — • • • • • 1 1 0.0 0.5 1.0 1.5 (Logging road density)/ (Logging road density)' R2 = 0.080, P = 0.090 Rf = 0.011, P = 0.532 Figure 3.3. Regression analyses between logging road density, and proportion of landscapes harvested (square root transformed) and three spatial principle components of harvest patch indices (Table 3.6) in the Slocan Valley of southeast British Columbia. Logging road density (square root transformed) is in kms per 100 ha of forest. 38 DISCUSSION Old growth spatial patterns relative to harvesting levels were not consistent with predicted fragmentation trends described by Forman (1997:407). Landscapes with higher harvest levels did not have old growth patches that were consistently smaller and farther apart, nor did they have more old growth edge and less core area proportional to habitat amount than landscapes with little or no harvest levels. Although landscapes are inherently variable and therefore violate to some degree assumptions of homogeneity, I am confident in my results due to their insensitivity to changes in control sample size. The failure to detect trends consistent with old growth fragmentation in this case presents several possibilities: (1) old growth fragmentation is not occurring, (2) amounts of harvesting and old growth are too low to detect a fragmentation trend, or (3) my measures and tests of fragmentation are inadequate to detect a fragmentation trend. Many have predicted critical thresholds in landscape patterns, after which significant effects on ecological process are realized. The overall level of harvesting for all landscapes combined in this study was 9.4 %. Studies focussing on amount of forest harvesting as an element of fragmentation usually cite levels of harvesting that are much higher than those found here. For example, Franklin and Forman (1987) predict a major loss of obligate interior-forest species in landscapes with 30-50 % of the landscape in a harvested state. O'Neill et al. (1988) predict loss of habitat continuity within landscapes when greater than 40 % of habitat is removed. Finally, Fahrig (1997) predicts that species survival is ensured in landscapes when up to 80 % of breeding habitat is removed, regardless of the extent to which this habitat is fragmented. On this basis it is very plausible that current levels of harvest in my study area are too low to either detect a trend towards old growth fragmentation, or are below a threshold under which fragmentation does not occur. The amount of old growth forests in my study area can also be considered relatively low (6.2 %; all landscapes combined). Fire patterns in my study area in the last century have produced a current serai stage distribution heavily skewed towards mid-successional forests (unpublished data, BC Ministry of Forests, Castlegar, BC). Patches of old growth in this case already tend to be relatively small and widely dispersed in harvested and non-harvested areas. Fragmentation, by definition, only occurs when the number of patches increases by the breaking apart of contiguous patches (Fahrig 1997). If whole patches are removed, then 39 fragmentation under this definition is actually decreased since the number of patches decreases (Ripple et al. 1991). Therefore, if old growth patches are small and can be removed in their entirety by harvesting, landscape indices used to measure fragmentation could reflect a decreasing trend in fragmentation. This phenomenon may explain to some extent results of this study. The failure to detect an old growth fragmentation trend may also be related to our questionable ability to quantify and therefore adequately test for fragmentation effects. Hundreds of quantitative measures of landscape pattern have been proposed (Gustafson 1998). However, accepted quantitative definitions of central concepts in landscape ecology such as forest fragmentation remain elusive (Hulshoff 1995, Schumaker 1996, Davidson 1998, Hargis et al. 1998). D'Eon and Glenn (2000) demonstrated that landscape indices are of limited use in distinguishing between human perceptions of fragmentation or are not sensitive enough to differentiate the critical elements of landscape spatial heterogeneity in this context. Further, the spatial scale at which fragmentation is measured is commonly the scale of human perception, but any potential ecological impacts of fragmentation will occur at different scales for different organisms. So while fragmentation as a conservation issue continues to harbor concern on theoretical grounds, it is unclear i f current methods of quantifying landscape pattern adequately address these concerns (Tischendorf and Fahrig 2000). Comparisons among the three patch types revealed that amounts and spatial patterns of harvest patches differed very little, if at all, from amounts and spatial patterns of old growth patches within control landscapes. A l l mean values of harvest patch indices were similar to mean values of old growth indices with one exception. The fractal dimension, a measure of patch perimeter complexity, was lower in harvest patches suggesting that harvest patches tended to be of more regular shapes than old growth patches. However, important indicators of spatial pattern such as patch size and spacing were similar and thus indicate no significant deviation from a more natural patch pattern represented by old growth patches. Fire patches revealed more discrepancies when compared to harvest patch patterns. Indices associated with habitat amounts indicated more total harvesting within the landscape than area represented by wildfire. I suggest that active fire suppression in recent decades may be a confounding factor leading to a reduction in the amount of recent wildfire in these 40 landscapes than would be present without fire suppression. However, considering indices associated with spatial pattern, wildfire patches reflected more complex shapes than harvest patches, similar to the case with old growth patches. Although, wildfire patches tended to have higher spacing index values suggesting that harvest patches are more aggregated and therefore less fragmented from this point of view. Variability indices revealed similar variability between harvest patches and old growth patches, and surprisingly more variability between harvest patches and wildfire patches. This result rejects the notion that harvesting tends to simplify landscape patterns by providing patches with less inherent variability than more natural patches. The opposite appears true in this study. I suggest that this phenomenon is due to shifts in forest policy in four previous decades. Early logging patterns tended towards large continuous clearcuts often hundreds of hectares large. This pattern was reversed towards the 1980s and has resulted in a more recent pattern tending towards smaller dispersed clearcuts often legally restricted to a maximum size (Franklin and Forman 1987, Province of British Columbia 1996). The combination of these shifting patterns appears to have resulted in an overall harvest pattern that displays high spatial variability in these landscapes. Mantel's tests indicated that spatial components among all patch types were spatially correlated. This suggests there is a high degree of association between landscape pattern and distance between landscapes in this study. Spatial correlation among ecological phenomena is not new (MacArthur and Wilson 1967, Hurlbert 1990, Legendre 1993) and was therefore expected for old growth and wildfire patches. However, demonstrating similar spatial correlation within harvest patterns rejects the notion of evenly dispersed harvest patches that would differ from a more natural pattern. In fact, I suggest harvest patterns in all likelihood follow natural patterns due to forest management practices that target similar aged forests and are restricted to localized areas due to access. A n exception, however, occurred in the amount of harvesting which was not spatially correlated among landscapes contrary to amounts of old growth and wildfire that were, suggesting a departure from more natural patterns with regards to harvest rates. This phenomenon is no doubt the result of landscape planning and zoning which delineates distinct areas for harvest and non-harvest; the result of which are areas of heavy industrial use located directly adjacent to areas prohibiting industrial activity such as parks. 41 Logging road densities had no significant association with harvest spatial patterns, thus falsifying a prediction that harvest patch patterns are a consequence of road patterns, or vice versa. The amount of harvesting however, was highly associated with road densities. The combination of these results implies that while the inherently obvious statement that more logging results in more roads is true, it rejects the notion that harvest patch attributes important to the issue of fragmentation such as patch size and spacing are related to road density. This situation is likely explained by road building constraints experienced in rugged mountainous terrain where topography largely governs where roads are built. Miller et al. (1996) failed to find a relationship between average forest stand size and logging road density in Colorado and concluded that topography exerted the greatest influence on stand size, thus supporting this notion. Further, forest management planning in this study area typically begins by planning road networks that will eventually access all potential long-term harvest areas in a planning unit, regardless of short-term objectives. As a result, road networks tend to be constructed for maximum access to timber given a set of terrain constraints, and are therefore independent of patch configuration. To sum, little of the evidence presented here would support a claim that a trend towards old growth fragmentation as a result of forest harvesting is true in these landscapes. Clear predictions pertaining to old growth patch configuration as a result of forest harvesting were falsified. Further, future forest patterns represented by current harvest patches are similar to current spatial patterns of old growth and wildfire, with the exception of patch shape which appears to be more regular in harvest patches. Regardless, while patch shape may influence interception rates of individual organisms moving through landscapes (Gutzwiller and Anderson 1992, Hamazaki 1996), the contribution of patch shape to the issue of forest fragmentation is not clear. Spatial configuration aside, habitat amount and related indices in this study accounted for more differences between harvest patterns and more natural patterns. On this basis I concur with Fahrig (1997) and Trzcinski et al. (1999) and suggest that total habitat amount is more of a concern than spatial configuration of patches. The lack of detected fragmentation effects in this study is consistent with recent suggestions that forest fragmentation may not be a large problem in western North American managed forests (Bunnell 1999). Much of the early evidence to suggest that forest fragmentation may be a concern originated in Europe and eastern North America (Forman 42 1997). Numerous studies cited lower levels of species richness and abundance in remnant forest patches left over from broad-scale deforestation and land conversion to agriculture and urbanization (eg., MacClintock et al. 1977, Wegner and Merriam 1979, Henderson et al. 1985, van Dorp and Opdam 1987). In these cases, arguments were based on the theory that remnant forest patches dispersed throughout deforested landscapes acted similarly to oceanic islands by reducing movement and immigration rates among islands, as described by MacArthur and Wilson (1967). Species unable to move freely through deforested landscapes had lower immigration and movement rates between patches, consistent with the theory of island biogeography. However, movement is likely not as drastically reduced in landscapes dominated by forests managed for long-term timber harvest and other non-timber resources. In contrast to permanently deforested landscapes (e.g., agricultural and urban), managed forests are not permanently deforested, but rather are managed as forests from early to late serai conditions. Edges tend to be of variable contrast depending on silvicultural treatments, and become increasingly diffuse as harvested forests regenerate. Further, regenerating areas provide varying degrees of habitat suitability and may be used to varying extents for dispersal and other life requirements. The predicted effects of forest fragmentation in managed forest landscapes are therefore questionable (Franklin and Forman 1987; Small and Hunter 1988; Schieck et al. 1995, Bunnell et al. 1999). A limitation of this study was the relatively low proportion of harvesting and old growth in these landscapes. I recommend that additional empirical tests of the fragmentation paradigm similar to those presented here be performed in landscapes with larger amounts of harvesting and old growth forest. Another limitation of the study was the use of wildfire patches to reflect attributes associated with a naturally occurring landscape pattern. Despite debate over its influence, active fire suppression may have served a role in modifying wildfire patterns in recent decades (Baker 1993). A n ideal repetition of this study, therefore, would be performed in landscapes with an unsuppressed fire regime. Finally, a finding that harvest patches have similar spatial configuration to more naturally-occurring patches (at least to the extent that I have measured it) could be explained by the naturally-heterogeneous nature of forests in these landscapes. Abrupt changes in terrain and other biophysical attributes associated with steep mountainous areas produce a landscape mosaic that is inherently heterogeneous and therefore may reflect an already naturally-fragmented spatial 43 pattern that does not differ from harvest patterns. The most relevant question then becomes "is harvesting creating a more fragmented pattern than the inherent heterogeneity of the landscape?" Repeating a study of this nature in other less heterogeneous landscapes such as boreal forests may produce differing results. 44 C H A P T E R 4 L A N D S C A P E CONNECTIVITY AS A FUNCTION OF S C A L E A N D ORGANISM VAGILITY IN A R E A L FORESTED L A N D S C A P E Habitat fragmentation is one of the most commonly cited threats to species extinction and an ensuing loss of biological diversity, making it perhaps the most important contemporary conservation issue (Wiens 1996). Lord and Norton (1990) referred to fragmentation as simply the disruption of continuity. The inverse of landscape fragmentation, landscape connectivity, is considered a vital element of landscape structure (Taylor et al. 1993) because it is so critical to population survival (Fahrig and Merriam 1985, Fahrig and Paloheimo 1988) and metapopulation dynamics (Levins 1970). Landscape connectivity can be defined as the degree to which the landscape facilitates or impedes movement among resources patches (Taylor et al. 1993). A direct measure of landscape connectivity therefore must incorporate a measure of some aspect of organism movement through the landscape. Fahrig and Paloheimo (1988) and.Henein and Merriam (1990) measured connectivity as the probability of movement between two resource patches in mathematical models of animal movements. Two other common measures of landscape connectivity are dispersal success and search time, where the first is defined as the immigration rate of an organism into resource patches, the later defined as time spent in transit between resource patches. Tishendorf and Fahrig (2000a) point out the weakness in these measures that higher values of connectivity (high immigration, and low search time) ironically result from more fragmented landscapes (i.e., more smaller patches in a landscape results in higher patch interception rates, and thereby immigration, and lower search times indicating, by definition, higher landscape connectivity). They then advocate measuring cell immigration (dividing patches into equally sized cells and measuring the rate of immigration into cells, whereby movement can occur inside of a large patch but between cells) as a way of overcoming this problem. While methods such as these directly measure animal movement parameters, and are possible in simulation and modeling work, they are often impossible or impractical in empirical studies, despite Tishendorf and Fahrig's (2000a) suggestion that even measuring movement rates on 1% of a landscape sufficiently measures landscape connectivity (but see 45 Arnold et al. 1993, and Pither and Taylor 1998). Indeed, in a review of connectivity work, Tishendorf and Fahrig (2000b) found only four studies that directly measured landscape connectivity and all four used modeling approaches. As well, while the application of percolation theory, the study of connectivity in stochastically generated structures (Stauffer 1985), has been useful in the development of a generalized spatially explicit theoretical framework (With 1997), applications to empirical work are currently limited (but see Keitt et al. 1997). As Tishendorf and Farhig (2000b) concede, in most practical management applications, directly measuring connectivity by measuring animal movement and ensuing rates within landscapes is infeasible for logistical reasons. As a result, most work on real landscapes with the objective of quantifying landscape connectivity has focused on landscape indices that characterize the spatial pattern of a landscape as surrogates for direct measurements of connectivity. The indices are then interpreted and conclusions are drawn regarding landscape connectivity. While literally hundreds of landscape indices have been derived and used (Gustafson 1998), generalizing relationships between landscape indices and ecological processes reflecting landscape connectivity is poorly understood (Tishendorf 2001). This is no doubt due to the static nature of landscape indices that typically characterize spatial configuration of patches and landscapes from a non-organismal perspective and under one set of rules (e.g., habitat availability does not change). As a result, most landscape indices reflect spatial patterns under one set of circumstances, and are usually not linked to any specific organism movement through the landscape. Taking a somewhat more robust perspective, landscape connectivity refers to the functional linkage among habitat patches, either because habitat is connected through structural continuity or because dispersal abilities permit organisms to travel among discrete patches and therefore perceive patches as functionally connected (With et al. 1997). In the latter, organism vagility is one of the most important determinants of landscape connectivity and is why many advocate an organismal perspective when addressing landscape connectivity (Wiens 1989, Schumaker 1996, With et al. 1997, Tishcendorf and Fahrig 2000, McGarigal and Cushman 2002). Taking this view then, landscape connectivity must be considered at the scale of the interaction between an organism and the landscape. Thus, a landscape is not inherently fragmented or connected, but can only be assessed in the context 46 of an organisms ability to move among patches and the scale at which the organism interacts with the landscape (Davidson 1998, With 1999). As well, many ecologists predict non-linear patterns in connectivity suggesting thresholds or abrupt changes in landscape connectivity exist and are scale-dependant (With and Crist 1995, Keitt et al. 1997). If true, connectivity-induced consequences to species living in landscapes should be predictable and relative to species vagility, the scale of interactions between an organism and the landscape, and connectivity thresholds. Commercial forest harvesting is commonly presented as a primary cause of forest fragmentation or a disruption in continuity (Franklin and Forman 1987, McGarigal et al. 2001). As forests are harvested, distances between remnant patches may increase and may represent a reduction in connectivity. However, connectivity is reduced in these cases only if an organism's ability to move between suitable habitat patches is reduced, underlying the importance of an organismal perspective. I assessed landscape connectivity across multiple scales based on a range of critical distances representing movement capabilities of selected i species, selected species. I built upon the habitat-cluster approach taken by Keitt et al. (1997) to provide a generalized measure of connectivity for use with forest cover information in vector format at the scale of forest management. I performed analyses using publicly available forest cover data derived from real managed forest landscapes in southeastern British Columbia. My objectives were to: (1) derive a multi-scale measure of landscape connectivity related to organism vagility, (2) detect critical connectivity thresholds in real landscapes, (3) test a hypothesis that commercial forest harvesting reduces connectivity among old growth patches, and (4) predict consequences to select species related to landscape connectivity in these landscapes. METHODS Landscape delineation Digital forest cover data obtained from the British Columbia Ministry of Forests (Castlegar, BC) were derived from a managed forest landscape of 362,350 ha within the Slocan Valley of the Selkirk mountains in southeast British Columbia, Canada (49°N, 117°W; Figure 4.1). Forest cover classification was provided in vector format and derived from interpretation of 1:20,000 black and white aerial photographs. Terrain within this 47 mountainous area is generally steep and broken with slope gradients often exceeding 80%. Elevation ranges from 525 m along the Slocan Valley bottom to 2,800-m mountain peaks. Forested land made up 71.5% of the study area. Forests in the study area are within the Interior Subalpine and Southern Columbia regions described by Rowe (1972) and are predominantly within three forest biogeoclimatic subzones described by Braumandl and Curran (1992): Interior Cedar Hemlock Dry Warm subzone at low elevations, Interior Cedar Hemlock Moist Warm subzone at mid elevations, and Englemann Spruce Sub-alpine Fir subzone at higher elevations. Alpine parkland predominates above 2,000-m elevations. Logging within the Slocan Valley began in the late 1800s but was primarily confined to localized selective harvesting. Large-scale commercial logging began around 1950. Side drainages of the Slocan Valley have since been managed for forest harvesting and road building to varying degrees. Many areas within the main valley corridor and a large provincial park, however, have been excluded from forest harvesting. The majority of low elevation areas along the main valley bottom is privately owned land and has been partially deforested for agricultural and urban development purposes. Private land was excluded from the analyses. Routine forest fire suppression in the area began in the late 1930s (J. Parminter, unpublished data, BC Ministry of Forests,Victoria, BC). To investigate general connectivity trends across the entire study area I considered the study area a single landscape and performed analyses on this basis. As well, I investigated relationships between varying forest harvest levels and old growth connectivity within the study area by comparing connectivity among nine landscape units ([LU]as defined by the government of British Columbia, Ministry of Forests) making up the study area (Figure 4.1). LUs ranged from 18,014 ha to 58,858 ha (x = 40,261; SE = 4,699) and were delineated using watershed boundaries primarily. Distance to edge (DTE_L) calculations I reclassified forest cover data into three patch types within landscapes: harvest patches (clear cuts harvested within the past 40 years), old growth patches (forest stands > 140 years for Interior Cedar Hemlock dry warm stands and > 250 years for all others; old growth definitions consistent with BC Ministry of Forests and BC Ministry of Environment, Lands and Parks [1995]), and wildfire patches (wildfire within past 40 years). I then 48 analyzed habitat connectivity using these three patch types separately in a binary manner where each patch type was considered habitat and all remaining landscape was considered non-habitat. In this way, each patch type was considered separately where patches of habitat were dispersed within a matrix of non-habitat (Figure 4.1). Within an ARCINFO geographic information system platform, I calculated the edge to edge distance between patches of the same patch type. I used the notion of a critical distance representing an organism's ability to travel between habitat patches as a fundamental element of landscape connectivity. Patches within a critical distance were considered connected and formed a habitat cluster. I varied critical distance from 100 m to the minimum critical distance where all patches in the landscape became connected into one cluster. I defined the boundary of each cluster as the 100% mean convex polygon (MCP) boundary surrounding outside patches in the cluster (Figure 4.2). Since a M C P boundary such as this creates points within the polygon beyond the critical distance from the centroid, boundaries were modified to follow a buffer around patches equal to the critical distance in cases where the M C P boundary created areas within a cluster beyond the critical distance to the edge of a patch. In this way, cluster boundaries were created that encircled all patches in the cluster and ensured all locations within the cluster were within the critical distance to a patch. I determined the centroid of each cluster and forced centroids into the cluster in cases where the calculated centroid was outside of the cluster (this can occur with geometric shapes similar to a " C " or "L") . From the centroid I projected 36 lines to the cluster boundary at 10° increments (Figure 4.2). I calculated the mean distance of these lines which represented the mean distance to a cluster edge (DTE_C). At each critical distance I calculated the mean distance to cluster edges within a landscape (DTE_L) as the mean DTE_C for all clusters within the landscape. I considered DTE_L a measure of the average distance an organism could move within a landscape within its habitat type (i.e., for a given patch type), given an ability to travel between habitat patches equal to the critical distance. In doing so, I recognize the assumption of a simplistic binary landscape model which forces the following major assumptions: all patches of habitat and non-habitat are equally suitable, travel routes between patches are linear, the only barrier to movement between patches is distance, and all species have equal gap-crossing abilities. 49 Connectivity measures and trends I used the slope of the linear regression model between the critical distance and the logarithmic transformation of the corresponding DTE_L as a measure of overall landscape connectivity for a given patch type and landscape. The slope of this line (8) is predictably higher in landscapes with higher patch connectedness since steeper slopes represent more habitat access (DTE_L) for each incremental rise in dispersal ability (critical distance). The opposite is also true in that lower slope values represent landscapes where incremental increases in dispersal ability return lower increases in available habitat, to the extreme case of a flat line (slope = 8 = 0) where increases in dispersal ability do not provide any increase in available habitat. I then used linear and curvilinear regression analyses between 8 and old growth amount and harvest rates among LUs to investigate connectivity trends relative to harvest rates. I calculated L U harvest rates (HR) as the amount of harvested area in a landscape (H) divided by the amount of old growth (OG) and harvested area combined (i.e., HR = H/(H+OG)* 100%). I added existing harvested area to existing old growth area in the denominator of this calculation because I wanted to calculate old growth harvest rate as a function of all old growth forests existing prior to the onset of commercial logging in the study area. In doing so, I assumed that all recent harvesting (i.e., since 1961) was old growth harvesting. I believe this to be justifiable because of British Columbia's long-standing policy and tradition of harvesting the oldest forests first. It is also congruent with observable harvest patterns in the study area that tend to target older forests. Species associations I investigated connectivity primarily among old growth patches and vertebrates associated with old growth forest because of the relative importance of old growth forest to conservation. From an original list of 74 old-growth associates known to occur in my study area (Bunnell 2000) I focused on the northern goshawk (Accipiter gentiles), marten (Martes americana), and cavity-nesting birds and mammals (Table 4.1). I excluded species associated primarily with riparian old growth forest (e.g., amphibians and cavity-nesting waterfowl) and more generalist species (e.g., red squirrel [Tamiasciurus hodsonicus]), and 50 concentrated on species with documented dependencies on terrestrial old growth forest structures, particularly for reproductive habitat. Although distance between patches is a prime determinant of landscape connectivity, patch size could be a factor to many species in that only patches large enough to provide suitable resources to an organism would be considered part of it's available or connected habitat (With 1999). To investigate this phenomenon I considered minimum patch size requirements (MPSR) for individual species. I defined MPSR by building upon Allen (1987) as the minimum amount of contiguous habitat forming a patch required, before a patch can be used or occupied by a species. For most species however, a minimum patch size is either not required, unknown, or not consistently demonstrated with data (see Bunnell et al. 1999 for review). I therefore used 2 ha as a general MPSR to assess landscape connectivity for most species because it is in the range suggested as a suitable resolution for a large suite of species including most forest passerines and forest-dwelling small mammals (With 1999), and because it was the finest resolution in my source data. However, two old-growth associates in my study area, the northern goshawk and marten, have sufficiently documented data on MPSRs that I felt comfortable assigning them. The northern goshawk is a forest-dwelling raptor that nests in the study area and has a high association with old forests for nesting (Graham et al. 1999). I assigned a 12-ha MPSR for the northern goshawk based on Reynolds et al.'s (1992) forest management recommendation for nest site reserves. Similarly, I assigned a 15-ha MPSR for marten based on data and recommendations reported by Snyder and Bissonette (1987) and Chapinetal. (1998). To investigate consequences of species dispersal ability and patch connectivity, I calculated median and probable maximum dispersal distances for individual species using data and methods provided by Sutherland et al. (2000). Using these estimates, I calculated the amount of the landscape accessible to species as a function of dispersal ability by calculating the total amount of cluster area available as a proportion of the landscape area at each critical distance. 51 CN Old growth patch Cluster boundary. Distance to cluster edge Figure 4.2. A n example old growth habitat cluster illustrating old growth habitat patches (shaded) and the cluster boundary (outer line). Illustrated as well are 36 lines radiating from the cluster centroid which were used to calculate a mean distance to cluster edge (DTE_C). In this example, there are 38 old growth patches at a critical distance of 1,000 m forming a 7,009-ha cluster with a D T E C = 4,980 m. 53 Table 4.1. Estimated dispersal abilities and associated landscape availability for selected old growth associates within the Slocan Valley Basin of southeastern British Columbia. Dispersal ability8 Proportion of landscape accessible (%)' Weight Median Probable maximum (km) Median Maximum Species (kg) t (km) dispersers dispersers Northern goshawk 1.1370 17.00 192.10 100 100 Barred owl 0.5060 23.86 269.62 100 100 Boreal owl 0.1670 56.00 632.80 100 100 Great gray owl 1.3909 44.66 504.69 100 100 Hawk owl 0.2516 15.47 174.83 100 100 Saw-whet owl 0.1072 9.12 103.02 100 100 Black-backed woodpecker 0.0666 1.29 14.57 20 100 Three-toed woodpecker 0.0612 1.27 14.35 20 100 Hairy woodpecker 0.0625 1.27 14.41 20 100 Pileated woopecker 0.2660 1.65 18.70 24 100 Boreal chickadee 0.0098 0.91 10.32 15 100 Mountain chickadee 0.0101 0.92 10.38 15 100 Pygmy nuthatch 0.0106 0.93 10.47 15 100 Red-breasted nuthatch 0.0098 0.91 10.32 15 100 White-breasted nuthatch 0.0211 1.05 11.85 17 100 Brown creeper 0.0084 0.89 10.04 15 100 Northern flying squirrel 0.1070 0.43 4.90 10 70 Marten 0.6610 2.39 26.97 24 100 fFrom Banfield 1974, Eckert 1987, and Dunning 1993. Weights are for females or species average for species without sexual dimorphism. ^Derived from available data (northern goshawk and boreal owl) and equations reported in Sutherland et al. (2000). Probable maximum distance is a threshold distance where the probability of a dispersing individual exceeding it is P < 0.001, and is calculated as median dispersal distance multiplied by 11.3 (Sutherland et al. 2000). ^Based on the amount of habitat cluster area available for a given dispersal ability, calculated as a proportion of the landscape size (Figure 4.7). 54 Statistical analyses Linear and curvilinear regression analyses and student's t-tests were performed using SYSTAT 8.0 (SPSS 1998) statistical software. Tests were considered significant at a = 0.05. Non-normal data distributions were assessed using skewness and kurtosis indicators and transformed using logarithmic transformations to produce more normal distributions. Skewness or kurtosis were considered extreme if + 2 times their standard error did not include zero (SPSS 1998). RESULTS A l l three patch types displayed a predictable negative exponential patch size distribution (Figure 4.3; note logarithmic scale). Harvest patch sizes (x = 32.8 ha, SE = 2.76, n = 785, range = 2.0 - 1018.4) were similar (t = 0.672, P = 0.501) to old growth patch sizes (x = 30.0 ha, SE = 2.81, n = 481, range = 2.1 - 787.8), but smaller (r = 3.126, P = 0.002) than wildfire patches (x = 84.3 ha, SE = 16.24, n = 123, range = 2.2 - 1336.6). Total patch areas for harvest, old growth, and wildfire patches were 25,771 ha, 14,448 ha, and 10,372 ha, respectively. At MPSR = 2 ha, harvest patches had the highest index of connectivity (8 = 0.259) considering the entire study area as one landscape, followed by old growth patches (8 = 0.216) and wildfire patches (8 = 0.169; Figure 4.4; slopes significantly different [test of slope homogeneity: t = 8.036, P < 0.001]). At MPSR = 12 ha, connectivity of old growth patches within the entire study area was reduced to 8 = 0.206, and to 8 = 0.204 at MPSR =15 ha. Loglinear regression R for critical distance (CD) versus DTE_L were high (> 0.923) and all regressions were highly significant (P < 0.001) in these cases and therefore considered good fits of the data for comparing slopes (Figure 4.4). I identified two types of critical thresholds in my data. The first is the minimum CD where all patches became connected and created one large habitat cluster (MFN_CD). This threshold, at MPSR = 2 ha, for harvest, old growth, and wildfire patches occurred at CD = 5,400 m; 8,000 m; and 8,500 m; respectively (Figure 4.4). Among individual LUs at MPSR = 2 ha, MIN_CD ranged from 2,900 m to 9,400 m (x = 6,311; SE = 718.5; n = 9). Final cluster sizes for harvest, old growth, and wildfire patches were 208,568 ha; 345,500 ha; and 216,965 ha; respectively. 55 The second critical threshold I identified was a point where CD = D T E L or the ratio between CD and D T E L = 1.0. At MPSR = 2 ha, DTEJL among harvest patches exceeds the CD at all points along the curve (i.e., DTE_L/CD > 1.0; Figure 4.4). Old growth patch D T E L / C D < 1.0 from CD = 900 m to CD = 4,100 m; and > 1.0 at all other CD (Figure 4.4). Wildfire patch D T E L / C D < 1.0 from CD = 800 m to CD = 7,800 m; and > 1.0 at all other CD (Figure 4.4). Among individual LUs only one (Figure 4.5a) had values of DTE_L/CD > 1.0 at all CD. A l l LUs had D T E L / C D values > 1.0 at low values of CD (CD < 400). One L U had D T E L / C D values > 1.0 for all CD except between CD = 1,200 m and CD = 3,000 m (Figure 4.5b). Four landscapes had DTE_L/CD values largely < 1.0 except at maximum CD of 5,600 m and 7,200 m (Figure 4.5c,d,e,f). Three LUs had D T E L / C D values < 1.0 except at CD < 400. Old growth patch connectivity among LUs ranged from 8 = 0.1014 to 8 = 0.5207 (x = 0.2237, SE = 0.0405, n = 9) at MPSR = 2 ha. Corresponding old growth harvest rates ranged from 47.3% to 90.2% (x = 65.1, SE = 4.25, n = 9; Figure 4.6). Among LUs I found significant linear regression between 8 (log transformed) and proportion of old growth in the landscape (R2 = 0.553, P = 0.022; Figure 4.6), and a significant curvilinear regression between 8 (log transformed) and old growth harvest rate (R2 = 0.375, P < 0.001; Figure 4.6). Estimated species dispersal distances ranged from 0.43 km (northern flying squirrel) to 56.00 km (boreal owl [Aegolius funereus]) for median dispersal distances with corresponding probable maximum distances of 4.90 km and 632.80 km (Table 4.1). A l l species with the exception of the northern flying squirrel had the ability to access all old growth patches in the landscape at maximum dispersal distances (Table 4.1, Figure 4.7). Median dispersers, however, were more limited with only the northern goshawk and owls with median dispersal abilities providing access to 100 % of old growth patches in the landscape. Smaller birds and mammals including marten were able to access only 10 % to 24 % of the landscape at median dispersal distances (Figure 4.7). 56 100 Figure 4.3. Patch size frequency distributions for harvest, old growth, and wildfire patches in the Slocan Valley Basin of southeast British Columbia, September 2001. 57 35 -, 30 -25 -o o o 20 -X _ i i 15 -i LU I-Q 10 -5 -0 -Old growth Harvest Wildfire DTE L LU h-Q o 5.5 5 4.5 4 3.5 3 2.5 2 Harvest Old growth Wildfire 3 2 3 4 5 6 7 8 9 Cr i t ical d i s tance (m x 1,000) Figure 4.4. (top) Critical distance (CD) versus mean distance to edge (DTE_L) for harvest, old growth, and wildfire patches within a 325,350-ha landscape in the Slocan Valley of southeast British Columbia. The dashed line represents a 1:1 ratio between DTEJL and CD, where points above the line represent DTE_L/CD > 1.0, below the line DTE_L/CD < 1.0. (bottom) Critical distance versus DTE_L (logarithm transformed) illustrating linear regression models. Regression equations for harvest patches: y = 0.2586x + 2.817, R2 = 0.923, P < 0.001; old growth patches: y = 0.2163x +2.6935, R2 = 0.986, P < 0.001; wildfire patches: y = 0.1693x + 2.5925, R2 = 0.959, P < 0.001. The slopes of regression models were used as an index of landscape connectivity (5) varying between 0 and 1.0, where high values of 8 represent high connectivity among patches. 58 Figure 4.5. Critical distance (CD) versus DTEJL/CD curves for nine Landscape Units in the Slocan Valley of southeastern British Columbia. The dashed line at D T E L / C D = 1 is the threshold where mean distance to cluster edge, an index of habitat availability, is equal to critical distance, a measure of dispersal ability. Labels a to f are landscape units where DTE L/CD > 1.0 at some CD. 59 0 2 4 6 8 10 12 Proportion of old growth in landscape (%) 40 50 60 70 80 90 100 Proportion of old growth forest harvested (%) Figure 4.6. Amount of old growth forest (top) and old growth harvest rate (bottom) versus 8 (log transformed) for old growth forest patches among nine Landscape Units in the Slocan Valley of southeastern British Columbia. 8 is a derived index of landscape connectivity ranging between 0 and 1.0, where high values represent high connectivity among patches. 60 Figure 4.7. Proportion of the landscape accessible as a function of an organism's ability to move (critical distance) in the Slocan Valley Basin of southeast British Columbia. Shown here are curves for minimum patch size requirements (MPSR) of 2, 12 and 15 ha. Proportion of the landscape was calculated as the proportion of the entire landscape within habitat clusters at each critical distance. 61 DISCUSSION Changes in connectivity for all three patch types were scale-dependent with proportionally larger increases in connectivity attained at high critical distances (CD). As theory predicted, critical thresholds, where small changes in pattern produce abrupt responses (Turner and Gardner 1991), were found at relatively high CDs that produced abrupt changes in connectivity until a final threshold was attained where all patches were connected and formed a single cluster (With and Crist 1995, Keitt et al. 1997). My results are consistent with O'Neill et al. (1988) who, in an application of percolation theory (Stauffer 1985, Orbach 1986), predicted a relationship between decreasing habitat amount and an organism's ability to traverse non-habitat that results in the formation of a percolating cluster (a habitat cluster that spans the landscape). In this way, an organism that can cross large distances wil l be able to use resources that are sparsely dispersed. In my case, abrupt changes in connectivity were attributable to distant patches that, once becoming accessible at high CDs, sharply increased habitat availability. Thresholds in this case represented a minimum ability an organism must have to perceive the entire landscape as connected and thereby travel to every patch. Harvest patches in this landscape were more connected than old growth patches which were more connected than wildfire patches, based on relative values of 8. Current harvest patch patterns are particularly important in light of the persistent legacy left by current harvest patterns on future forest patterns (Bunnell et al. 1999, Nelson and Wells 2000). This has important implications to future forest patterns that may result from regenerated harvest blocks. If left to attain old growth conditions, current harvest patterns represent a future old growth forest pattern that is more connected than existing old growth forest patterns. This rejects a notion that harvest patches may represent a more fragmented patch pattern in this case (Krummel et al. 1997). Two major caveats in this case however, are that (1) old growth stands resulting from regenerated cut blocks may not represent similar quality to existing old growth stands, and (2) regenerated cut blocks in a managed forest may be harvested prior to attainting old growth structure. Patch size was not correlated with connectivity since wildfire patches, with the lowest connectivity index, were almost three times larger (mean differences) than harvest and old growth patches. However, I found strong correlation between connectivity and total patch amount. There were more harvest patches than old growth patches making up more total 62 patch area, followed by the number and total amount of wildfire patches. As well, individual landscape units displayed an increasing trend in old growth patch connectivity with increasing old growth amount. I found similarly consistent correlation between habitat amount and other landscape indices in this landscape (Chapter 3). The strong relationship between amount of habitat and connectivity in this case supports the importance of habitat amount in discussions of landscape structure (Fahrig 1997) and the measurement of landscape connectivity (Tishendorf 2001). Old growth harvesting was negatively correlated with old growth connectivity and thus supports a hypothesis that forest harvesting reduced old growth connectivity in this landscape. However, since connectivity is strongly related to habitat amount in this case, the relationship between reduced connectivity with increased harvesting is no doubt related to a reduction in the amount of old growth in the landscape since the onset of commercial forest harvesting. This has important management implications in that, i f true, by simply removing total amount of habitat, regardless of spatial configuration of the removal, connectivity within remaining habitat will be reduced. This supports recent assertions that a current forest management focus on spatial configuration of removals, rather than the amount of total removal, is misguided (Fahrig 1999, Trzcinski et al. 1999). Nonetheless, accepting a view that overall reductions in a habitat type result in lower connectivity among patches of a given habitat type, says little about whether or not remaining patches are connected. As stated, landscapes are not inherently disconnected or not, but must be evaluated from an organismal perspective and at the scale of interaction between the landscape and the organism. It is entirely conceivable that habitat reductions leading to lower connectivity values do not result in a fragmented landscape if an organism does not perceive it so, or vice versa. At the default MPSR of 2 ha, the minimum critical distance where all patches became connected was lowest for harvest patches (5,400 m) followed by old growth (8,000 m) and wildfire patches (8,400 m). Thus, species associated with early serai habitat provided by harvest blocks and an ability to move 5,400 m through this landscape would perceive the entire landscape as connected and be able to travel to all patches. Similarly, species dependant on old growth forest or recent wildfire patches would require an ability to travel 8,000 m and 8,500 m, respectively, before viewing the entire landscape as connected. 63 Among old growth associates all species but the northern flying squirrel (Glaucomys sabrinus) have estimated probable maximum natal dispersal abilities in excess of 8,000 m. This suggests that at least some individuals of these species seeking new reproductive habitat have the ability to disperse to and inhabit all old growth patches in this landscape. While long distance dispersal is rare (Sutherland et al. 2000, Turchin 1988), maximum dispersal ability has large implications to metapopulation dynamics. In a metapopulation model, it is the ability of some individuals to recolonize distant patches that lowers the probability of local extinctions (Levins 1970). I suggest, therefore, that all old growth associates I considered, with the exception of the northern flying squirrel, have a low probability of local extinction that is attributable to a lack of connectivity among old growth patches in this landscape. Short dispersal distances are more frequent and strongly influence age and sex structure and abundance within populations (Sutherland et al. 2000). My estimates indicate that only the larger more vagile carnivorous birds could access all old growth patches at median dispersal distances. Median dispersing woodpeckers seeking old growth structure in this landscape would be limited to approximately 20% of the landscape, chickadees and nuthatches limited to approximately 15% of the landscape, and marten limited to 24%. As well, any impacts of reduced landscape access are further exacerbated in that these values represent the sum area within all habitat clusters combined, which are not continuous but inherently isolated from each other by distances in excess of the critical distance. These findings suggest distribution and abundance of smaller less vagile species may be affected in this landscape as a consequence of reduced connectivity among old growth patches. Of particular concern is the northern flying squirrel, which is typically associated with old forest structure for food and denning requirements (Carey et al. 1997). My results indicate that only 70% of the landscape is accessible to northern flying squirrels even at maximum dispersal distances, and only 10% of the landscape at median dispersal distances, i f flying squirrels are obligate users of old growth structure. On this basis northern flying squirrel populations may be limited in this landscape by a lack of connectivity among old growth patches, which has implications to minimum viable population requirements as individuals are isolated from each other. A caveat, however, to this conclusion is that flying squirrels may be facultative old growth users (Ransome and Sullivan 1997; D. Ransome, 64 personal communication). If so, effects of dispersed old growth habitat would be reduced by flying squirrels using other habitat and thereby increasing their access to the landscape. Dispersal and other movements away from a patch of suitable habitat to another suitable patch is generally considered costly because moving individuals may face increased mortality rates associated with higher predation risk and the physiological costs of moving through unfamiliar or hostile habitat (Sutherland et al. 2000 and refs therein). Therefore, individuals that move beyond the limits of available habitat (i.e., travel farther than available patches) incur costs associated with moving without the benefit of increased suitable habitat access. This notion is scale-dependant and should apply to the scale of the organism and it's interaction with the landscape. For example, an organism with an ability to move between suitable patches of 500 m, will incur unreciprocated costs by moving 500 m from a patch i f the next suitable patch is 1,000 m away. Indeed, Keitt et al. (1997) predicted that selection pressure may favor species with dispersal abilities equal to the scale of distances between habitat patches in the landscape because of the optimal balance between the benefits of increase habitat and movement costs. Among patches in my study area I used the ratio between DTE_L and the associated critical distance (CD), as a measure of the optimal balance between movement costs and increased habitat access. I identified this optimal balance where DTE_L (a measure of accessible habitat) equaled critical distance (CD; an organism's ability to move). For old growth patches within the entire landscape this optimal balance occurred at CD = 900 m and 4,100 m. Between these two points (900 - 4,100 m) represents a range of movement ability that exceeds accessible habitat and therefore higher movement costs without the benefits of increased habitat access. Conversely, either side of this range (< 800 m and > 4,200 m) represents movement abilities that are below accessible habitat and therefore movement risks that are rewarded with increased access to habitat. When considering individual landscape units (LU), I found similar patterns where all landscapes displayed ratios above 1.0 for small CDs (< 400 m), most intermediate CDs below 1.0, and near or above 1.0 at relatively high CDs. This suggests that species with either relatively small or large movement abilities in this landscape receive benefits of increased habitat access without unnecessary movement costs; the opposite is true for species with intermediate movement abilities. This is consistent with my predictions that larger more vagile species can access all patches in this 65 landscape and view it as connected, while smaller species with intermediate movement abilities are more restricted. Presumably, species on the far left of the curve (CD < 400 m) are so restricted by their movement abilities that once in a suitable habitat patch, moving beyond the habitat cluster they are in is unlikely, and are therefore almost always within a suitable habitat cluster and provides an explanation for this observation. Minimum patch size requirements (MPSR) produced predictable results: connectivity was reduced among old growth patches when minimum patch size constraints were imposed. As well, the amount of the landscape available to organisms was reduced in a similarly predictable manner. This phenomenon is, again, no doubt related to the ultimate result of patch size constraints that exclude patches from the analyses: a reduction in total habitat amount. My results in this case are consistent with results from simulations in and predictions from artificial landscapes (Dale et al. 1994, With 1999). While intuitively appealing, however, the application of a minimum patch size which governs whether or not patches are included in investigations of connectivity in real landscapes, is currently limited. Due to either a demonstrated lack of a patch size requirement or a paucity of data, I could justify imposing patch size constraints for only two species on my list of old growth associates. I caution that the distinction must be made between home range size, for which there are many data on many species, and documented cases of species avoiding the use of patches below a certain size, for which few data exist. The later is vital to an understanding of use, and therefore connectivity, among patches in real landscapes. And, since the influence of minimum patch size in investigations of connectivity can be large, as demonstrated in this study, I strongly recommend future empirical work include the gathering of minimum patch size use. M y method of measuring landscape connectivity is consistent with recommendations for organismal-based procedures and, very importantly, is a function of the scale at which an organism interacts with the landscape. While it is not a direct measure of animal movement, it incorporates elements of organism movement at a wide spectrum of scales - a failure of most landscape indices in the study of landscape connectivity. More importantly though, I believe a large advantage of my method is its applicability in real landscapes with real organisms, its intuitive simplicity, and its feasibility in practical situations. It is particularly suited, by design, for use with digital forest cover information in vector format. I suggest 66 this method could be useful for species-specific assessments for any habitat type. As well, this measure of connectivity decreased with decreases in habitat amount and, presumably, increases in an ensuing fragmentation effect - a persistent problem with other methods in which the opposite is true (Tischendrof and Fahrig 2000a). While useful for analytical purposes, a major limitation of my methods is the assumption of a simplistic binary landscape model where movement between patches is linear and gaps between patches are equally unsuitable to every organism. These assumptions are likely false to varying degrees in most cases. Indeed, organisms most likely perceive habitat suitability along a gradient and travel along non-linear routes that facilitate their movement (Taylor et al. 1993, With et al. 1997). However, the importance of distance between patches, the basis of my methods, is indisputable. Rather, I suggest the shortcomings of a binary landscape model can be reduced by including additional modifications based on empirical data specific to a landscape and species of interest. 67 CHAPTER 5 SCALE-DEPENDANT USE OF SIX L A N D S C A P E FEATURES IN A M A N A G E D FOREST: THE CASE OF M U L E DEER IN SOUTHEAST BRITISH C O L U M B I A Mule deer (Odecoileus hemionus hemionus) in the interior mountainous regions of North America are often associated with use of low-elevation mature forests in winter followed by movement to higher-elevations and more open habitat in summer (Garrott et al. 1987, Thomas and Irby 1990, Armeleder et al. 1994, Schackleton 1999, D'Eon 2001). The proximal cause of this strategy is usually attributed to deep snow accumulations at high elevations during winter and ultimately by seasonal changes in the quality and quantity of available forage within the annual home range of the animal (Garrott et al. 1987). While habitat use of mule deer has been reported for several populations, published accounts primarily address preference or avoidance of habitat type classifications and biophysical attributes (e.g., Armeleder et al. 1994, Nicholson et al. 1997, D'Eon 2001). Deer (Odocoileus spp.) response to landscape features implicated as factors in a current forest fragmentation problem (Saunders et al. 1991), such as forest edges, logging roads, and patch size in managed forests has been either mixed or unstudied, and is therefore either unclear or unknown. This is especially true of mule deer which have been less studied than other deer species, despite an obvious relationship between these landscape elements within a fragmentation context and traditional deer habitat management. Habitat edges were historically seen as ecologically beneficial and particularly beneficial to deer (Leopold 1933, Thomas 1979). Indeed, many deer habitat management regimes in North America are premised on Thomas' (1979) cover to forage ratio recommendations that stipulate patch interspersion of mature forest and early serai vegetation. Edges, however, are now commonly associated with forest fragmentation and negative ecological effects (see Kremsater and Bunnell 1999 for review). As a result, traditional views of edge in relation to deer must be re-examined. This is especially true in light of recent work demonstrating mixed responses to edge by deer, prompting some to suggest that the value of edge habitat to deer cannot be assumed without local verification (Kirchhoff and Schoen 1983, Kremsater and Bunnell 1992). 68 Roads have been identified repeatedly as incurring negative ecological effects (e.g., Forman and Alexander 1998, Underhill and Angold 2000). Logging roads, specifically, have been cited as elements of forest fragmentation (Reed et al. 1996, Tinker et al. 1998) because they may create barriers to movement and create edge. Roads have been well-documented contributors to direct ungulate mortality through vehicle collisions and increased hunter access (Romin and Bissonette 1996, Cole et al. 1997). As well, deer in some cases have been shown to avoid areas near well-traveled or open roads (Rost and Bailey 1979, Nyberg and Janz 1990). However, use or avoidance of logging roads by mule deer in managed forests where roads are of variable quality, often remain unplowed in winter, and vehicle traffic and speeds are low, is largely unknown. Finally, patch size, while an important parameter in investigations of the role of landscape structure in species abundance and distribution (see Bunnell et al. 1999 for review) has, to my knowledge, eluded study in the published literature in relation to mule deer habitat use. This has occurred despite the obvious implications of patch size to interspersion of mature forest and early serai vegetation and the ensuing edge created from this interspersion. In this study I tested seasonal selection of logging roads, forest edges, and forest patch size by radiocollared mule deer in a forested mountainous landscape in southeast British Columbia. I chose these landscape elements because of their association with predicted mule deer habitat and their obvious association within the forest fragmentation paradigm. Because total habitat amount may be more important than spatial configuration of landscape elements (Fahrig 1997) I concurrently tested seasonal selection of amounts of mature forest and early serai vegetation and investigated their relative influence. Further, because resource selection is predicted to occur at a hierarchy of scales (Johnson 1980, Senft et al. 1987) and is sensitive to the scale of habitat availability (McCLean et al. 1998), I focused my tests on comparisons between use and availability of these landscape features at the landscape (2nd-order selection from Johnson [1980]) and seasonal home range scales (3 r d-order selection from Johnson [1980]) to test a hypothesis that selection of these features is scale-dependant. In this application, I refer to landscape scale as a deer's selection of a seasonal home range area within a larger landscape context. Conversely, seasonal home range scale refers to selection of attributes within the confines of a chosen seasonal home range. 69 STUDY A R E A Mule deer were captured and monitored within the Lemon Creek drainage, a 21,924-ha mountainous landscape within the Selkirk Mountains of southeastern British Columbia, approximately 23 km northwest of Nelson (49° 42' N , 117° 25' W; Figure 5.1). Elevations within the study area range from 548 m at the mouth of Lemon Creek to 2,405-m mountain peaks. Terrain is generally steep and broken with slope gradients exceeding 100% and slope aspects varying from 1 to 360°. Annual precipitation averages 659 mm (Environment Canada weather station, New Denver, British Columbia). Average daily July high and low temperatures are 25.8 0 C and 11.6 0 C, respectively; average daily February highs and lows are 2.6 0 C and -4.0 0 C, respectively. Snow usually covers 100% of the ground throughout the study area from mid-November until late April and is deepest in mid-February. Mid-winter snow depths commonly exceed 40 cm at the lowest elevations and are several meters at higher elevations. The study area is within the Interior Cedar Hemlock Dry Warm (ICHdw), Interior Cedar Hemlock Moist Warm (ICHmw2), and Englemann Spruce Subalpine Fir (ESSFwcl and ESSFwc4) biogeoclimatic zones (Braumandl and Curran 1992). The ICHdw zone occurs from the lowest elevations in the study area to approximately 1,000 m; above which ICHmw2 extends to approximately 1,450 m; above which ESSFwcl and ESSFwc4 extend to the treeline at approximately 1,950 m. Forests in ICHdw consist mostly of mixed serai stands of Douglas-fir (Pseudotsuga menziesii), white birch (Betula papyriferd), western larch (Larix occidentalis), and western white pine (Pinus monticola); in ICHmw2 forests are mostly serai mixes of western hemlock (Tsuga heterophylld) and western redcedar (Thuja plicata); in ESSFwcl and ESSFwc4 forests are mostly serai mixes of Engelmann spruce (Picea engelmanni) and subalpine fir (Abies lasiocarpd). Common shrubs in ICHdw and ICHmw2 are falsebox (Paxistima myrsinites), Douglas maple (Acer glabrum) and black huckleberry (Vaccinium membranaceum); in ESSFwcl and ESSFwc4 common shrubs are white-flowered rhododendron (Rhododendron albiflorurri) and black huckleberry. Forests make up 70.9% of the land cover within the study area. Approximately 96.5%o of these forests are dominated by coniferous species with forest canopy closure varying from 0% in recent clearcuts and natural openings to 100% in dense stands (British Columbia Ministry of Forests data). Within forests, mature forests (age > 80 years) make up 70 80.1% and early serai vegetation (<20 yrs) including logged stands make up 16.2%. Logged stands make up 80.5% of early serai vegetation. Broad-scale commercial logging in the area began in 1950 resulting in a current landscape characterized by dispersed clearcuts within a mature forest matrix. 71 BRITISH COLUMBIA •*V..ir. N 2,000 0 2,000 gure 5.1. Lemon Creek drainage study area in southeast British Columbia. Lemon Creek flows east to west from high elevation (> 2,000 m) headwater areas to 548 m elevation at its mouth. Thin solid lines are logging roads, dotted lines are streams, and shaded areas are harvest patches. 72 METHODS Radiotelemetry Between 18 and 27 February, 1999, field crews captured three male and three female adult mule deer in clover traps baited with alfalfa and salt (Clover 1954, D'Eon et al. 2003). We fitted deer with global positioning system (GPS) radiocollars obtained from Advanced Telemetry Systems (Isanti, M N , USA) that contained Garmin GPS 25LP receivers (Wildlink 1990) and remote release mechanisms. Radiocollar battery life was limited to approximately 1 yr. We therefore removed and retrieved radiocollars, fitted them with new batteries, and subsequently redeployed them on newly captured and different deer (five males, one female) between 4 and 18 March, 2000. Collars were then retrieved within 1 yr of deployment. At programmed intervals GPS radiocollars communicate with GPS satellites orbiting the earth and store GPS locations in memory (see Rodgers et al. [1996] for a thorough description). Briefly, when a GPS radiocollar is retreived, location data are downloaded into a computer. I programmed GPS radiocollars to obtain GPS locations at 4-hr intervals in the first deployment in February 1999, and 6-hr intervals the next year to extend battery life. However, a location is recorded only in cases when > 2 satellites are successfully contacted and called a "fix". Fix rates vary depending on environmental factors and animal behaviour and are much lower than 100% in field studies (Dussault et al. 1999, Bowman et al. 2000, D'Eon et al. 2002). Reported location accuracies for recorded fixes are + 100 m 95% of the time during the first year of deployment (Wells 1986) and were elevated to + 30.6 m 95% of the time during the second year due to changes in the United States Department of Defense GPS policy for civilian use (D'Eon et al. 2002). Data Management and Analysis I downloaded data from retrieved radiocollars and screened them for obvious anomalies and impossible data that I deleted (D'Eon et al. 2002). I stratified locations of individual deer into seasons by inspecting spatial distribution of locations and elevational movements of individuals (Apps et al. 2001). Mule deer in this drainage display a consistent migratory pattern from low-elevation winter ranges near the mouth of Lemon Creek to high-elevation summer ranges (Figure 5.2), similar to mule deer migration patterns described by Garrott et al. (1987). I identified four seasons for each deer: summer, fall migration, winter, 73 and spring migration. Winter and summer home ranges tended to be relatively localized areas where deer concentrate in one area; whereas spring and fall migration ranges are larger and are primarily movement phases (Figure 5.3). I created 100% minimum convex polygon seasonal home ranges for each deer (White and Garrott 1990). I calculated mean values among seasonal locations for each individual deer rather than pooling all locations among all individuals since I had many locations on few individuals, as suggested by White and Garrott (1990) and Aebischer et al. (1993). I then performed statistical analyses by comparing mean deer use (i.e., n = number of deer) of landscape features to mean values of what was available to each deer (design 3 from Thomas and Taylor [1990]). Doing so avoided pseudoreplication and inflated sample size problems (Hurlbert 1984). To distinguish between resource use at landscape and seasonal home range scales, I compared deer use of landscape features to what was available at these two scales. In these comparisons, I considered the entire drainage (21,924 ha) to represent a landscape scale. To provide a measure of landscape-scale availability to each deer associated with each seasonal home range, I created polygons representing all area within the study area between the maximum and minimum elevations used by each deer within seasonal home ranges. The landscape was stratified by elevation in this way to avoid confounding correlations between elevation and landscape features such as logging road and harvest patch density. Therefore, each seasonal home range polygon (termed "use") for each deer had a corresponding landscape availability polygon (termed "available"). Within each use and available polygon for each deer, I calculated the following parameters within a geographic information system: logging road density, early serai edge density, proportion of the polygon in mature forest, proportion of the polygon in early serai vegetation, mean mature forest patch size, and mean early serai patch size. I calculated the amount of early serai vegetation and mature forest to investigate edge use in relation to the two components creating edge in this case. I calculated logging road density as the total length of these features within a polygon divided by the polygon area to provide a density value in m/ha. Patches were delineated as continuous areas of mature forest and early serai vegetation. I defined early serai edge as the linear interface between early serai vegetation (forest < 20 yrs including clearcuts, natural openings, and shrub communities) and mature 74 forest (> 80 yrs), and similarly calculated edge density as the total length of edge within a polygon divided by the polygon area. I obtained all digital forest cover and road information from British Columbia provincial forest cover and terrain resource inventories. To compare seasonal home range use of landscape features to those available to deer at this scale, I compared deer locations within seasonal home ranges to an equal number of randomly selected locations within each seasonal home range. In this way, random locations within seasonal home ranges were considered a measure of resource availability at the seasonal home range scale. I calculated road and edge use as the proportion of deer locations < 100 m of a road or edge within each deer seasonal home range. These values were compared to values calculated in the same way for random locations. I calculated mean patch size use for each deer as the mean patch size associated with each location (i.e., size of the patch that each location was within), and repeated a similar calculation for corresponding random locations. As well, I calculated the proportion of deer locations in mature forest and early serai vegetation within each deer seasonal home range. These values were similarly compared to the same parameters calculated for random locations. I screened all variables for normality using skewness and kurtosis indicators and transformed severely non-normal distributions using logarithmic transformations (Fowler et al. 1998). I tested statistical differences between mean deer use (n = number of deer) and means of available landscape features with Student's two-sample /-tests when sample variances were similar (i.e., standard deviations were same order of magnitude) and used Welch's approximate t when sample variances were unequal (Zar 1984). I used Pearson correlations with Bonferroni corrected probabilities to investigate correlations among landscape features (SPSS 1998). I used bivariate logistic regression with Bonferroni corrected pobabilities to evaluate the ability of each landscape feature to predict deer use. Models were compared using McFadden's rho2 and log-likelihood (G) tests, with higher values of both statistics indicating better predictors (Tabachnick and Fidell 1996). I used bivariate, rather than multivariate, logistic regression to avoid multi-collinearity among independent variables due to high inter-variable correlations (Tabachnick and Fidell 1996). I performed retrospective power analyses to investigate the influence of sample size on the power of my data to detect increasingly large effect sizes (Steidl et al. 1997, Gerard et al. 1998). I calculated statistical power (1 - p) for proportional changes in the available mean 75 for all Mests between use and available resources using variability estimates within the data, oc = 0.05, and varying sample size from 6 to 15 deer (Zar 1984). I calculated the proportion of all tests with power > 0.8 because this is generally considered adequate power (Thomas 1997). I performed all statistical tests other than power calculations using SYSTAT 8.0 statistical software (SPSS 1998) and considered significance at oc = 0.05. Power analyses were calculated manually using spreadsheet software. 76 2500 2000 •B 1500 1000 500 •••• / S . V . 0 • • • • •• _L February 1999 September 1999 March 2000 October 2000 May 2001 Figure 5.2. Mean weekly elevations of four mule deer tracked with GPS radio-telemetry from February 1999 to February 2000, and three mule deer tracked from March 2000 to May 2001 in Lemon Creek drainage, British Columbia. 5000 Sex • Both • Female • Male Figure 5.3. Mean seasonal home range size of seven radiocollared mule deer in Lemon Creek, British Columbia, 1999-2001. Seasons: s = summer, fm = fall migration, w = winter, sm = spring migration. Error bars are 1 SE of the mean for six deer (three males, three females) in s, fm, and sm; seven deer (four males, three females) in w. Seasonal home ranges were delineated from 100% minimum convex polygon of seasonal locations. 77 RESULTS Radiotelemetry Four of six radiocollars deployed in year 1, and two of six deployed in year 2, functioned properly and provided useful data within all seasons. A third collar in year 2 functioned properly during the winter season but fell off the animal during spring migration and therefore provided only winter data. The remaining collars malfunctioned by either not recording data or recording spurious and unuseable data which is a common phenomenon in contemporary GPS radiotelemetry (D'Eon et al. 2002). Consequently, sample size for data analyses was six deer (three females, three males) in summer and spring and fall migration, and seven deer (three females, four males) in winter. The number of locations recorded for individual deer ranged from 75 to 409 in winter, 9 to 328 in spring migration, 50 to 573 in summer, and 67 to 1058 in fall migration. The number of locations varied among individuals because of differences in timing of initiation of movements, length of time spent in each season, and fix rates (rate of collar successfully obtaining fixes at the set attempt interval). Resource Use Despite a tendency for smaller female home ranges (Figure 5.3), no consistent sex bias was detected in any resource selection tests at either the landscape or stand scale. Seasonal use of landscape features was not significantly different between male and female deer at either scale (all f-test P > 0.05). On this basis I combined data and did not distinguish between sex for subsequent resource use analyses. Landscape scale: Deer use of landscape features at the landscape scale was similar to landscape availability in spring migration, summer, and fall migration for amounts of mature forest and early serai vegetation and use of edges and logging roads (Figures 5.4 and 5.5). In winter however, use of mature forests was greater than available (avail, x = 64.2 %, SD = 0.89; home range x = 90.0, SD = 13.06; Welch's t6 = -5.204, P = 0.002), and use of early serai vegetation was lower than available (avail, x =18.3 %, SD = 1.01; home range x = 6.9 %, SD = 11.09; Welch's t6 = 2.716, P = 0.034). As well, winter use of edges was lower than available (avail, x = 28.1 m/ha, SD = 0.46; home range x = 16.4 m/ha, SD = 10.80; Welch's te - 2.61 A, P = 0.044), and winter use of logging roads was lower than available (avail, x = 78 15.3 m/ha, SD = 2.46; home range x = 8.1 m/ha, SD = 8.16; tI2 = 2.251, P = 0.044). Mean mature forest patch size was significantly greater in home range polygons than available polygons in all seasons but summer (Figure 5.6). Early serai patch size use differed significantly only in fall migration where mean early serai patch size was greater in home range polygons than available polygons (Figure 5.6; avail, x =111 ha, SD = 50.5; home range x = 444 ha, SD = 245.9; Welch's t5.4= -2.612, P = 0.044). Seasonal home range scale: Deer use of landscape features was similar to home range availability, as a function of random locations within seasonal home ranges, in all seasons and among all landscape features (all /-test P > 0.06; Figures 5.7, 5.8, and 5.9). In one case, however, the proportion of deer locations in early serai vegetation during spring migration was considerably higher than available, but not significantly so (avail, x = 17.8 %, SD = 5.39; use x = 36.2 %, SD = 20.50; Welch's t6 = -2.130, P = 0.08). Relative selection of landscape features Since use of landscape features differed only at the landscape scale, I investigated post-hoc relative use of features only at the landscape scale. Correlation among landscape features at the landscape scale was high (Table 5.1). Of 15 individual tests, eight had significant correlations (Bonnferroni-corrected P < 0.05). In particular, amount of mature forest was highly correlated with road density, mature forest patch size, and amount of early serai vegetation (all Bonferroni-corrected P < 0.001) and also closely associated with edge density (r = -0.715, Bonferroni-corrected P = 0.061). Among the 6 landscape features, mature forest patch size and amount of mature forest in home range polygons were the best predictors of deer use at the landscape scale (Table 5.2; McFadden's rho2 - 0.709 and 0.690, respectively). Amount of early serai vegetation, logging road density, early serai patch size, and edge density were not significant predictors (Table 5.2; all G < 6.149, all Bonferroni corrected P> 0.078). Power Analysis 79 No tests at any sample size between n = 6 and n = 15 had power to detect a 1 0 % difference between use and available means at oc = 0 .05 (Figure 5 . 1 0 ) . At an effect size of 5 0 % of the available mean and n = 6, 3 7 . 5 % of all r-tests had power > 0 . 8 ; at n = 15 this rose to 67.5%o of tests having power > 0 . 8 . At an effect size of 1 0 0 % of the available mean and n = 6 , 7 5 % of all Mests had power > 0 .8 ; at n = 15 this rose to 90%> of tests having power > 0 .8 . 80 c o Polygon type • Available • Home range Figure 5.4. Landscape-scale use of mature forest and early serai vegetation for seven mule deer tracked with GPS radiotelemetry from February 1999 to May 2001 in Lemon Creek drainage, British Columbia. Values are means (± 1 SE) of individual deer seasonal home ranges and corresponding available polygons. Available polygons were delineated by using elevational contours within the entire drainage equal to individual deer maximum and minimum elevations in each season. Seasons: w = winter, sm = spring migration, s = summer, fm = fall migration. Sample size was six deer in sm, s, and fm; seven deer in w. Asterisks indicate significant /-test at oc = 0.05. 81 jS 50 1 ^ 4 0 I 30 C D ~° 20 CD ^ U $ 10 co o LU w sm fm w sm s Season Polygon type • Available • Home range Figure 5.5. Landscape-scale use of early serai edge and roads for seven mule deer tracked with GPS radiotelemetry from February 1999 to May 2001 in Lemon Creek drainage, British Columbia. Values are mean densities (± 1 SE) within individual seasonal home ranges and corresponding available polygons. Available polygons were delineated by using elevational contours within the entire drainage equal to individual deer maximum and minimum elevation use in each season. Seasons: w = winter, sm = spring migration, s = summer, fm = fall migration. Sample size was six deer in sm, s, and fm; seven deer in w. Asterisks indicate significant Mest at cc = 0.05. 82 12000 CD 10000 CD CL 00 s a CD i— -t—' CD w sm s 3000 CD £ 2 5 0 0 CU « 2000 §_ 1500 CD oo CD LU sm s Season Polygon type • Available • Home range Figure 5.6. Landscape-scale use of mature forest and early serai patches for seven mule deer tracked with GPS radiotelemetry from February 1999 to May 2001 in Lemon Creek drainage, British Columbia. Values are means (± 1 SE) of individual deer seasonal home ranges and corresponding available polygons. Available polygons were delineated by using elevational contours within the entire drainage equal to individual deer maximum and minimum elevations in each season. Seasons: w = winter, sm = spring migration, s = summer, fm = fall migration. Sample size was six deer in sm, s, and fm; six deer in w. Asterisks indicate significant /-test at oc = 0.05 83 Early Serai Mature forest Location type • Random • Deer Figure 5.7. Home range scale use of early serai vegetation and mature forest within seasonal home ranges of six mule deer tracked by GPS radiotelemetry from February 1999 to May 2001 in Lemon Creek drainage, British Columbia. Use calculated as the proportion of deer locations within early serai and mature patches; available calculated as the proportion of random locations within early serai and mature patches. Values are means (± 1 SE) of individual deer means and corresponding random means. Seasons: w = winter, sm = spring migration, s = summer, fm = fall migration. Sample size was six deer in sm, s, and fm; seven deer in w. 84 Early serai edges Logging roads sm s Season Location type • Random • Deer Figure 5.8. Home range use of early serai edges and logging roads within seasonal home ranges of seven mule deer tracked by GPS radiotelemetry from February 1999 to May 2001 in Lemon Creek drainage, British Columbia. Use calculated as the proportion of deer locations within 100 m of an edge or road. Available calculated as the proportion of random locations within 100 m of an edge or road. Values are means (± 1 SE) of individual deer means and corresponding random means. Seasons: w = winter, sm = spring migration, s = summer, fm = fall migration. Sample size was six deer in sm, s, and fm; seven deer in w. 85 CO _c CD N 'co - C o ro o. to CD i_ ro 12000 10000 8000 h 6000 4000 2000 iff w sm s fm 12000 Location type • Random • Deer sm s Season Figure 5.9. Home range use of mature forest and early serai patches within seasonal home ranges of seven mule deer tracked by GPS radiotelemetry from February 1999 to May 2001 in Lemon Creek drainage, British Columbia. Use calculated as the mean patch size of patches associated with individual deer locations. Available calculated as the mean patch size of patches associated with random locations. Values are means (± 1 SE) of individual deer means and corresponding random means. Seasons: w = winter, sm = spring migration, s = summer, fm = fall migration. Sample size was six deer in sm, s, and fm; seven deer in w. 86 20 40 60 80 Effect size (% of available mean) 100 Figure 5.10. Statistical power (1 - /3) to detect differences between means of available and used landscape features by mule deer in a radiotelemetry study in Lemon Creek, British Columbia. The y-axis is the proportion of 48 /-tests performed for 6 variables within four seasons at two scales. The bottom line is n = 6 with n increasing consecutively to n = 15 on the top line. Estimates of variance for consecutive n are based on observed variances at n = 6 and 7. Statistical significance was set at cc = 0.05. 87 Table 5.1. Pearson correlations for six landscape parameters^ describing used and available (combined) winter home ranges of seven mule deer in a GPS radiotelemetry study in Lemon Creek, British Columbia from February 1999 to May 2001. Bonferroni corrected probabilities in brackets with statistical significance indicated (*) at a = 0.05. Road Edge Mature Early serai % early % mature density density forest patch size serai forest patch size Road density 1.0 Edge density 0.863 1.0 (0.001)* Mature forest -0.688 -0.572 1.0 patch size (0.099) (0.489) Early serai 0.740 0.464 -0.453 1.0 patch size (0.037)* (1.000) (1.000) % early serai 0.931 0.872 -0.796 0.537 1.0 (O.001)* (0.001)* (0.010)* (0.716) % mature -0.877 -0.715 0.892 -0.554 -0.942 forest (0.001)* (0.061) (O.001)* (0.596) (O.001)* fRoad density = logging road density, Edge density = early serai edge density, mature forest = forest > 80 yrs, early serai = forest < 20 yrs and shrub communities. Mature forest patch size, early serai patch size, and % early serai were logarithm transformed to attain more normal distributions. 88 Table 5.2. Results of six bivariate logistic regressions for six landscape features describing used and available home ranges of seven mule deer in winter in a GPS radiotelemetry study in Lemon Creek, British Columbia from February 1999 to May 2001. The binary dependant variable was home range type (use or available) with each landscape feature as the independent variable in each regression. Landscape feature1 McFadden's rho2 Mature forest patch size 0.709 13.766 O.001* % mature forest 0.690 13.384 O.001* % early serai 0.317 6.149 0.078 Logging road density 0.246 4.769 0.174 Early serai patch size 0.142 2.763 0.576 Edge density 0.035 0.671 1.000 fRoad density = logging road density, Edge density = early serai edge density, mature forest = forest > 80 yrs, early serai = forest < 20 yrs and shrub communities. Mature forest patchsize, early serai patchsize, and % early serai were logarithm transformed to attain more normal distributions. 'calculated as 2 x [ L L ( N ) - LL(0)] , where L L ( N ) = the log-likelihood of the full model and LL (0 ) = the log-likelihood of a constants only model. P-value is calculated for x 2 with 1 degree of freedom and using a Bonferroni correction. 89 DISCUSSION Selection of landscape features by mule deer in this study was scale-dependant. Mule deer did not demonstrate preference or avoidance for any landscape features at the seasonal home range scale, but clearly demonstrated preference and avoidance of features at the landscape scale, particularly in winter. These findings support Johnson's (1980) and Senft et al.'s (1987) predictions that resource selection occurs at a hierarchy of scales. This has important implications. Concluding that mule deer, in this case, did not make choices among landscape features, based on a seasonal home range analysis alone, would be misleading. As Johnson (1980) implied, mule deer in this case may indeed have chosen winter home ranges based on landscape-scale attributes and once localized into winter home ranges used landscape features in a random fashion. This illustrates the importance of scale considerations in ungulate resource selection analyses (Apps et al. 2001), which is rarely done (Mysterud and Ostbye 1999). Winter was primarily the time of year where preference or avoidance of landscape features occurred. A large body of literature supports the notion that winter is a critical time for mule deer and that they are most stressed at this time of year (e.g., Bartmann 1984, Unsworth et al. 1999); it is therefore not surprising to observe this seasonal phenomenon. Winter habitat of mule deer is strongly associated with mature forest cover that is associated with lower and more favourable snow depths (Gilbert et al. 1970, Armeleder et al. 1994, D'Eon 2001). Not surprising then was a winter preference for greater amounts of mature forest and avoidance of early serai vegetation observed in my study. Interestingly, mature forest patch size was the best predictor of winter deer use at the landscape scale. However, its close association with amount of mature forest makes concluding that deer were selecting patch size alone, suspect. Differentiating between habitat amount and spatial configuration of habitat elements is problematic since they are so inextricably related (Fahrig 1997, Trzcinski et al. 1999). The spatial configuration of patches of a given habitat type is inherently related to, and often a function of, the amount of the habitat in the landscape (Fahrig 1997). Unequivocal differentiation of effects due to habitat amount from those due to spatial configuration of habitat thereby becomes difficult, i f not impossible. When the distinction has been made, however, spatial configuration of habitat elements was insignificant compared to habitat amount in deriving ecological effects 90 (McGarigal and McComb 1995, Trzcinski et al 1999). In a review (Andren 1994) and meta-analysis (Bender et a l l 998) of patch-size effects among a variety of species and systems, neither study concluded that any patch size effects occurred. On this basis, it is plausible and perhaps more conceivable that deer were attracted to wintering areas containing large amounts of mature forest which inherently contained large patches, and vice versa. Logging road density was not a significant predictor of winter deer use at the landscape scale, despite significantly lower road densities in winter home ranges than available in the landscape. Since vehicle traffic and related human disturbance is almost non-existent during the winter months in this landscape, it would seem unlikely that deer avoidance would be related to direct human disturbance. Rather, it is more plausible, here again, that lower road densities, highly correlated with large amounts of mature forest, are the result of large amounts of forests, and conversely small amounts of early serai (in this landscape, most early serai vegetation is the result of harvesting, which is highly correlated with road density). This illustrates the importance of distinguishing between and accounting for confounding area effects. Deer response to forest edges, despite the historical paradigm of considering edge beneficial to deer (Leopold 1933, Thomas 1979), has been mixed and has provided results that question the validity of this paradigm (see Kremsater and Bunnell 1992 for review). Kremsater and Bunnell (1992) suggested that deer preference for edge is governed by the inherent interspersion of edge in the landscape and explains differences among studies. They found that studies reporting no edge preference by deer were typically in landscapes where edge habitat was finely grained or evenly interspersed in the landscape. Conversely, they found deer tended to prefer edges when edge was not well interspersed. Edge in my study area tended to be well interspersed in the landscape. Indeed, at the home range scale, edge habitat accounted for approximately 30 - 50 % of random seasonal locations (Figure 5.8). Consistent with Kremsater and Bunnell (1992) then, use of edges by deer at this scale was high (Figure 5.8; 29 - 49 % of seasonal locations) but in proportion to edge availability. At the landscape scale deer selected winter home ranges with less edge than available in the landscape. However, as in the case of logging roads, it is the high negative correlation between amount of contiguous mature forests and edge that precludes a conclusion that deer avoided edge when selecting winter home ranges. Rather, higher predictive power associated 91 with amounts of contiguous mature forests again provides a more plausible explanation for this observation, particularly since edge was not a significant predictor of winter deer use (Table 5.2). Because of the relationship between spatial configuration and habitat amount, it is not surprising that I found high correlation and confounding effects between amount of mature forest and other landscape features, consistent with other results of landscape pattern investigations in this area (Chapter 3). In particular, the relationship between amount of mature forest and mature forest patch size is so close (Table 5.1; r = 0.892, P < 0.001) that the two parameters are almost equally powerful predictors of winter deer use at the landscape scale (Table 5.2). I suggest however, that in this population, deer selected winter home ranges on the basis of landscape-scale amounts of mature forest, and thus in doing so chose areas that inherently had large contiguous mature forest patches and lower amounts of other landscape features such as logging roads, edges, and early serai patches. Due to the difficulty in differentiating habitat amount from habitat configuration in mensurative studies (Hurlbert 1984) such as mine, I recommend future work engage in manipulative approaches, i f possible, to dissect this conundrum (McGarigal and Cushman 2002). Finally, contemporary GPS radiotelemetry is expensive and prone to malufunctions which tends to result in low sample sizes among studies. Statistical power was arguably adequate to distinguish trends in this study. This is likely due to remarkably similar behaviour and habitat use patterns among individuals in this case, which resulted in relatively low variability. However, low sample size may have been a concern in this study if individuals had differed more. Additional statistical power accrued by small increases in sample size was demonstrated and has implications for choosing radiotelemetry methods in studies of this nature. Alternative methods such as V H F radiotelemetry with lower per unit costs and higher reliability typically result in higher sample sizes than GPS radiotelemetry. It is therefore extremely important to investigate sample size issues prior to initiating GPS radiotelemetry studies to ensure adequate statistical power will be obtained. 92 CONCLUSIONS A N D DIRECTIONS The work presented within this dissertation on landscape spatial patterns in the Slocan Valley of southeast British Columbia continues today. The information gained and methods refined through my research presented here, provides a solid basis for continued work of this nature. Numerous findings stand out as significant and represent major contributions to the growing literature on spatial patterns and forest fragmentation in managed forests. In Chapter 1,1 highlighted the two most important barriers to contemporary forest fragmentation work: (1) the extreme difficulty of conducting good landscape-scale studies, and (2) the persistent confusion between habitat loss and fragmentation effects. The challenge to future work is clear. In Chapter 2,1 revealed that much of what is regarded as forest fragmentation is based on differing human interpretations, and that landscape management will likely continue to be influenced by human perception. Landscape indices were of limited value in measuring absolute levels of fragmentation, and are likely best used as relative measures for landscape comparisons in time and space. In Chapter 3, clear predictions stemming from the fragmentation literature concerning predicted spatial pattern trends due to fragmentation were falsified in my study area. Assumptions about fragmentation effects on spatial patterns in managed forests, therefore, must be examined carefully and tested. In order for fragmentation effects to occur on organisms living and moving through landscapes, predicted effects on spatial habitat patterns must be occurring and are only relevant to the inherent heterogeneity of the landscape. In Chapter 4,1 created a unique method of measuring landscape connectivity based on an organism-centered perspective of the landscape, in contrast to more traditional anthropocentric methods. In my study area, harvest patches were most connected followed by old growth and wildfire patches. If permitted to acquire old growth structure, existing harvest patch patterns could represent a future old growth pattern that is more connected than existing old growth patches. Changes in landscape connectivity were scale-dependant, and a function of the scale of interaction between an organism's ability to move and the landscape. Most old growth-dependent vertebrates I considered would view this landscape as connected at maximum dispersal abilities, with the exception of the northern flying squirrel which has more limited dispersal abilities. In Chapter 5,1 demonstrated that selection of landscape 93 features by mule deer was scale-dependant and only evident at the landscape scale, thus illustrating the importance of a landscape-scale perspective. As well, the link between habitat amount and habitat configuration was very evident and illustrated the challenge of distinguishing the two in a cause and effect relationship. While a solid basis for future work exists, numerous challenges and difficulties are ahead. Perhaps most importantly, fragmentation, as a scientific paradigm, suffers from a lack of unequivocal definition and quantification. Indeed, people tend to associate forest fragmentation with areas of high harvest block density, regardless of spatial pattern parameters (D'Eon and Glenn 2000). Patch density is highly correlated with harvest amount and illustrates the pervasive confusion between habitat loss and fragmentation effects (Fahrig 1997). As well, this notion is completely independent of inherent natural heterogeneity of the landscape and therefore without reference as to whether or not an imposed pattern is more or less fragmented than the natural pattern, and therefore may or may not pose an ecological threat. However, recent work provides promising clarification. Fahrig (2002), McGarigal and Cushman (2002), Schmiegelow and Mdnkkdnen 2002, and Boutin and Herbert (2002) emphatically contend that fragmentation effects are independent of habitat loss effects and must be distinguished in future fragmentation research. Consensus therefore appears to be building which will no doubt provide more clarity, and in turn, a better basis for future fragmentation work. Landscape indices, because they are relatively easy to calculate and use, have been widely applied in landscape pattern work, and specifically in fragmentation studies. However, their usefulness in defining absolute levels of fragmentation is probably very limited, and it is unlikely that pursuit of such indices will be fruitful. Indeed, despite numerous attempts at a universally accepted landscape metric that unequivocally measures fragmentation (e.g., Ripple et al. 1991, Schumaker 1996, Davidson 1998), none exists. Quite likely the best utility of landscape indices, or more likely a group of uncorrelated indices (e.g., Hulshoff 1995, Riitters et al. 1995, Garrabou et al. 1998), is as benchmarks to measure future or past trends in landscape pattern, or their use to compare landscape patterns among landscapes and patch types. The failure of landscape indices to provide an absolute measure of fragmentation is most likely because they do not measure spatial pattern in relation to the interaction of an organism and the landscape. It is this distinction that makes organism-94 centered techniques more useful since a landscape is not inherently fragmented or not, regardless of spatial pattern parameters, and can only be assessed in the context of an organism and its interaction with the landscape (With et al. 1997, Tishcendorf and Fahrig 2000). Landscape metrics have, however, provided a useful stepping-stone in the evolution of the fragmentation paradigm in illustrating the inseparability of habitat amount with spatial configuration. In this way, the suggestions derived from Ahl and Allen (1996) that the usefulness of indices may lie in their influence on future work by modifying criteria and definitions, is very relevant. Organism-centered techniques are likely the next step in assessing landscape patterns, particularly fragmentation patterns. While distance between landscape elements relevant to the organism under consideration is a critical factor in determining movement ability through a landscape, future work should incorporate more sophisticated elements of movement. In most cases, organism movement is more complicated than a linear distance between two habitat elements (Turchin 1996). Numerous landscape features could conceivably enhance or inhibit organism movement between points. These elements and their effects on movement are organism-dependent and could conceivably be included in landscape pattern assessments. For example, i f an organism under consideration lacks the ability to cross water (i.e., streams, rivers, lakes) this could greatly influence movement paths and the connectivity between habitat elements. In this example and using the techniques I derived, habitat patches separated by water would not be considered connected and would not form a habitat cluster, regardless of their distance apart. Incorporating a simple rule such as this would add empirical reality and likely have a significant effect on assessment outcomes. Measurement of landscape patterns in the assessment of fragmentation trends must be related to the inherent natural heterogeneity of the landscape (Bradshaw and Fortin 2000). One would expect different results from a description of landscape patterns in an inherently heterogeneous mountainous landscape from one in a less heterogeneous flat landscape. Differences, therefore, in imposed patterns can only be assessed relative to underlying inherent patterns. A n imposed pattern, or legacy, representing a significant change from inherent patterns in one landscape, may not be so in another. Most importantly, spatial configuration effects such as fragmentation effects may or may not be expected based on relative departures from inherent patterns. It would be prudent, therefore, to replicate the 95 research I have presented here in a variety of landscape types varying from relatively homogenous (e.g., boreal) to relatively complex (e.g., coastal mountains). It is perhaps the quantification of landscape patterns relative to inherent patterns that could provide the best source of empirical data in the short-term. While fragmentation effects on organisms are extremely difficult to detect (McGarigal and Cushman 2000), spatial patterns are relatively easy to measure and compare. Since any fragmentation effects would be the result of a change in pattern, quantifying these changes provides a relatively easier, albeit short-term, method of determining the expectation of detecting fragmentation effects on organisms. Clearly there is a great need for well-designed empirical fragmentation studies to validate many theoretical predictions (McGarigal and Cushman 2002). As a minimum, future studies must distinguish between habitat loss and fragmentation effects. Habitat amount is clearly important to organism abundance and distribution and may be more important than spatial configuration of habitat elements. In most of my investigations, habitat amount best explained observed phenomena. However, distinguishing between the two effects on organisms in the field remains a challenge, and perhaps the greatest challenge to future work. As well, future studies should strive to investigate the full range of habitat loss, from 0 to 100 % in areas of high intensity forest management, and the full range of fragmentation, from fully aggregated to perfect, even dispersion. As well, the full range of habitat availability, again from 0 to 100 %, should be explored and incorporated into study designs when possible. This is important since some predict threshold effects in landscapes (Fahrig 2002). If thresholds exist, then studies considering only one part of the range could be missing critical levels above or below a threshold where effects change dramatically. Finding these thresholds, i f they exist, is another of the great challenges to future studies of this nature (Boutin and Hebert 2002). Temporal shifts in forest policy provided a plausible explanation for observed patterns in current harvest patch configuration in my research. There is little doubt that forest policy will continue to change and evolve in my study area and elsewhere (Kimmins 1991 and 2002). In the face of such uncertainty, tools such as forest simulation modeling (e.g., Nelson 1999 and 2001) provide avenues for investigating impacts of changing forest policy on landscape spatial patterns through time. Since forest harvesting is a major disturbance 96 agent contributing to landscape patterns in managed forests, investigating and predicting impacts from future forest policy on landscape patterns through the use of simulation modeling presents an important pursuit. Finally, forest fragmentation, and indeed landscape ecology, is arguably on the verge of a paradigm shift. Kuhn (1970) described periods of steady and incremental progression within an accepted set of beliefs, or paradigm, as normal science (Pinter and Pinter 1998). I believe forest fragmentation, as a paradigm, has progressed through a normal science phase from early predictions of negative ecological effects through to tangible management prescriptions designed to mitigate these effects. With Farhig's (1997) work on the distinction between habitat loss and fragmentation effects, I believe we have arrived at the cusp of a shift away from thinking of fragmentation effects as large and detrimental, to viewing them as minor, i f at all real (Boutin and Hebert 2002, Fahrig 2002, Haila 2002, McGarigal and Cushman 2002, Schmiegelow and Monkkonen 2002). The debate is far from over and hinges on our ability to gather better empirical data. However, i f true, the implications of relatively small spatial configuration effects relative to habitat amount effects are profound and challenge the very basis of landscape ecology. 97 LITERATURE CITED Aebischer, N . J., P. A . Robertson, and R. E. Kenward. 1993. Compositional analysis of habitat use from animal radiotracking data. Ecology 74:1313-1325. Ahl , V . , and T. F. Allen. 1996. Hierarchy theory: a vision, vocabulary, and epistemology. Columbia University Press, New York, New York, USA. Allen, A . W. 1987. Habitat suitability index models: barred owl. U.S. Fish and Wildlife Service. Biological Report 82, Washington, DC, USA. Andren, H. 1994. Effects of habitat fragmentation on birds and mammals in landscapes with different proportions of suitable habitat: a review. Oikos 71:355-366. Apps, C. D., B. N . McLellan, T. A . Kinley, and J. P. Flaa. 2001. Scale-dependent habitat selection by mountain caribou, Columbia Mountains, British Columbia. Journal of Wildlife Management 65:65-77. Armeleder, H. M . , and M . J. Waterhouse. 1994. Winter habitat use by mule deer in the central interior of British Columbia. Canadian Journal of Zoology 72:1721 -1725. Arnold, G. W., D. E. Steven, J. R. Weeldenburg, and E. A . Smith. 1993. Influences of remnant size, spacing pattern and connectivity on population boundaries and demography in Euros Macropus robustus living in a fragmented landscape. Biological Conservation 64:219-230. Baker, W. 1993. Spatially heterogeneous multi-scale response of landscapes to fire suppression. Oikos 66:66-71. Banfield, A . W. F. 1974. The mammals of Canada. University of Toronto Press, Toronto, Ontario, Canada. Bartmann, R. M . 1984. Estimating mule deer winter mortality in Colorado. Journal of Wildlife Management. 48:262-267. BC Ministry of Forests and BC Ministry of Environment, Lands and Parks. 1995. Forest practices code biodiversity guidebook. British Columbia Ministry of Forests and Ministry of Environment, Lands and Parks, Queen's printer, Victoria, British Columbia, Canada. Bender, D. J., T. A . Contreas, and L. Fahrig. 1998. Habitat loss and population decline: a meta-analysis of the patch size effect. Ecology 79:517-533. Boutin, S., and D. Hebert. 2002. Landscape ecology and forest management: developing an effective partnership. Ecological Applications 12:390-397. Bowman, J. L. , C. O. Kochanny, S. Demarais, and B. D. Leopold. 2000. Evaluation of a GPS collar for white-tailed deer. Wildlife Society Bulletin 28:141-145. Bradshaw, G. A. , and M-J. Fortin. 2000. Landscape heterogeneity effects on scaling and monitoring large areas using remote sensing data. Geographic Information Sciences 6:61-68. Braumandl, T. F., and M . P. Curran. 1992. A field guide for site identification and interpretation for the Nelson Forest Region. British Columbia Ministry of Forests, Nelson, British Columbia, Canada. Bunnell, F. L. 1999. What habitat is an island? Pages 1-31 in J.A. Rochelle, L .A . Lehmann, and J. Wisniewski, editors. Forest fragmentation: wildlife and management implications. Koninklijke Brill , Leiden, Netherlands. Bunnell, F. L., R. J. Wells, J. D. Nelson, and L. L.Kremsater. 1999. Patch size, vertebrates, and effects of harvest policy in southeastern British Columbia. Pages 271-293 in J. A . 98 Rochelle, L .A . Lehmann, and J. Wisniewski, editors. Forest fragmentation: wildlife and management implications. Koninklijke Brill , Leiden, Netherlands. Bunnell, F. L. 2000. Vertebrates and stand structure in the Arrow IFPA. Arrow Innovative Forestry Practices Agreement Technical Report, Slocan, British Columbia, Canada. Carey, A. B., S. P. Horton, and B. L. Biswell. 1992. Northern spotted owls: influence of prey base and landscape character. Ecological Monographs 62:223-250. Carey, A . B., T. M . Wilson, C. C. Maguire, and B. L. Biswell. 1997. Dens of northern flying squirrels in the Pacific Northwest. Journal of Wildlife Management 684-699. Chapin, T. G., D. J. Harrison, and D. D. Katnik. 1998. Influence of landscape pattern on habitat use by American marten in an industrial forest. Conservation Biology 12:1327-1337. Chow, S. L. 1998. In defense of the construct validity approach. Theory and Psychology 8:481-487. Clark, P. J., and F. C. Evans. 1954. Distance to nearest neighbour as a measure of spatial relationships in populations. Ecology 35:445-453. Clover, M . R. 1954. A portable deer trap and catch-net. California Fish and Game 40:367-373. Cole, E. K. , M . D. Pope, and R. G. Anthony. 1997. Effects of road management on movement and survival of Roosevelt elk. Journal of Wildlife Management 61:1115-1126. Dale, V . H. , S. M . Pearson, H. L. Offerman, and R. V . O'Neill. 1994. Relating patterns of land-use change to faunal biodiversity in the Central Amazon. Conservation Biology 8:1027-1036. Davidson, C. 1998. Issues in measuring landscape fragmentation. Wildlife Society Bulletin 26:32-37. Debinski, D. M . , and R. D. Holt. 2000. A survey and overview of habitat fragmentation experiments. Conservation Biology 14: 342-355. D'Eon, R. G. 2001. Using snow-track surveys to determine deer winter distribution and habitat. Wildlife Society Bulletin 29:879-887. D'Eon, R. G. 2002. Forest fragmentation and forest management: a plea for empirical data. Forestry Chronicle 78:686-689. D'Eon, R. G., and S. M . Glenn. 2000. Perceptions of landscape patterns: do the numbers count? Forestry Chronicle 76:475-480. D'Eon, R. G., J. A . Johnson, and E. A . Ferguson, editors. 2000. Ecosystem management of forested landscapes: directions and implementation. U B C Press, Vancouver, British Columbia, Canada. D'Eon, R. G., S. M . Glenn, I. Parfitt, and M-J. Fortin. 2002. Landscape connectivity as a function of scale and organism vagility. Conservation Ecology 6(2): 10. [online] URL: http://www.consecol.org/vol6/iss2/artl0. D'Eon, R. G., R. Serrouya, G. Smith, and C. Kochanny. 2002. GPS radiotelemetry error and bias in mountainous terrain. Wildlife Society Bulletin 30:430-439. D'Eon, R. G., G. Pavan, and P. Lindgren. 2003. A small drop-net versus Clover traps for capturing mule deer. Northwest Science, in press. Diffendorfer, J. E., M . S. Gaines, and R. D. Holt. 1995. Habitat fragmentation and movement of three small mammals (Sigmodon, Microtus, and Peromyscus). Ecology 76:827-839. 99 Dunning, J. B. 1993. CRC handbook of avian body masses. CRC Press, Boca Raton, Florida, USA. Dussault, C , R. Courtois, J-P. Ouellet, and J. Huot. 1999. Evaluation of a GPS telemetry collar performance for habitat studies in the boreal forest. Wildlife Society Bulletin 27:965-972. Eckert, A . W. 1987. The owls of North America. Weathervane, New York, New York, USA. Fahrig, L. 1997. Relative effects of habitat loss and fragmentation on population extinction. Journal of Wildlife Management 61:603-610. Fahrig, L. 1999. Forest loss and fragmentation: which has the greater effect on persistence of forest-dwelling animals? Pages 87 - 95 in: Rochelle, J. A. , L. A . Lehmann, and J. Wisniewski, editors. Forest fragmentation: wildlife and management implications. Koninklijke Bril l N V , Leiden, The Netherlands. Fahrig, L. 2002. Effect of habitat fragmentation on the extinction threshold: a synthesis. Ecological Applications 12:346-353. Fahrig, L., and G. Merriam. 1985. Habitat patch connectivity and population survival. Ecology 66:1762-1768. Fahrig, L. , and J. Paloheimo. 1988. Determinants of local population size in patchy habitats. Theoretical Population Biology 34:194-213. Forman, R. T. 1997. Land mosaics. The ecology of landscapes and regions. Cambridge University Press, Cambridge, United Kingdom. Forman, R.T., and L.E. Alexander. 1998. Roads and their major ecological effects. Annual Review of Ecology and Systematics 29:207-231. Fowler, J., L. Cohen, and P. Jarvis. 1998. Practical statistics for field biology. John Wiley & Sons, Chichester, United Kingdom. Franklin, J. F., and R. T. Forman. 1987. Creating landscape patterns by forest cutting: ecological consequences and principles. Landscape Ecology 1:5-18. Garrabou, J., J. Riera, and M . Zabala. 1998. Landscape pattern indices applied to Mediterranean subtidal rocky benthic communities. Landscape Ecology 13:225-247. Garrott, R. A. , G. C. White, R. M . Bartmann, L. H . Carpenter, and A. W. Alldredge. 1987. Movements of female mule deer in northwest Colorado. Journal of Wildlife Management 51:634-643. Gerard, P. D., D. R. Smith, and G. Weerakkody. 1998. Limits of retrospective power analysis. Journal of Wildlife Management 62:801-807. Gilbert, P. F., O. C. Wallmo, and R. B. Gil l . 1970. Effect of snow depth on mule deer in Middle Park, Colorado. Journal of Wildlife Management 34:15-23. Glenn, S. M . , S. L. Collins, and D. J. Gibson. 1992. Disturbances in tallgrass prairie: local and regional effects on community heterogeneity. Landscape Ecology 7:243-251. Gloombridge, B., editor. 1992. Global biodiversity: state of the earth's living resources. Chapman and Hall, New York, New York, USA. Graham, R. T., R. L. Rodriguez, K. M . Paulin, R. L. Player, A . P. Heap, and R. Williams. 1999. The northern goshawk in Utah: habitat assessment and management recommendations. General Technical Report RMRS-GTR-22. US Department of Agriculture, Forest Service, Rocky Mountain Research Station, Odgen, Utah, USA. Gustafson, E. J. 1998. Quantifying landscape spatial pattern: what is the state of the art? Ecosystems 1:143-156. 100 Gutzwiller, K. J., and S. H. Anderson. 1992. Interception of moving organisms: influences of patch shape, size, and orientation on community structure. Landscape Ecology 6:293-303. Hagar, H. A . , and K . D. McCoy. 1998. The implications of accepting untested hypotheses: a review of the effects of purple loosestrife {Lythrum salicana) in North America. Biodiversity and Conservation 7: 1069-1079. Haila, Y . 2002. A conceptual genealogy of fragmentation research: from island biogeography to landscape ecology. Ecological Applications 12:321-334. Hamazaki, T. 1996. Effects of patch shape on the number of organisms. Landscape Ecology 11:299-306. Hargis, C. D., J. A . Bissonette, and J. L. David. 1998. The behavior of landscape metrics commonly used in the study of habitat fragmentation. Landscape Ecology 13:167-186. Hargrove, W. W., and J. Pickering. 1992. Pseudoreplication: a sine qua non for regional ecology. Landscape Ecology 6:251-258. Harris, L. D. 1984. The fragmented forest. University of Chicago Press, Chicago, Illinois, USA. Harris, L. D. 1988. Edge effects and conservation of biotic diversity. Conservation Biology 2:330-332. Helliwell, D. R. 1976. The effects of size and isolation on the conservation value of woodland sites in Britain. Journal of Biogeography 3: 407-416. Henderson, M . T., G. Merriam, and J. Wegner. 1985. Patchy environments and species survival: chipmunks in an agricultural matrix. Biological Conservation 31:95-105. Henein, K. , and G. Merriam. 1990. The elements of connectivity where corridor quality is variable. Landscape Ecology 4:157-170. Higgs, A . J. 1981. Island biogeography theory and nature reserve design. Journal of biogeography 8:117-124. Holt, R. D., G. R. Robinson, and M . S. Gaines. 1995. Vegetation dynamics in an experimentally fragmented landscape. Ecology 76:1610-1624. Hulshoff, R. M . 1995. Landscape indices describing a Dutch landscape. Landscape Ecology 10:101-111. Hurlbert, S. H . 1984. Pseudoreplication and the design of ecological field experiments. Ecological Monographs 54:187-211. Hurlbert, S. H . 1990. Spatial distribution of the montane unicorn. Oikos 58:257-271. Johnson, D. H. 1980. The comparison of usage and availability measurements for evaluating resource preference. Ecology 61:65-71. Johnstone, B., D. Holland, and J. E. Hewett. 1997. The construct validity of the category test: is it a measure of reasoning or intelligence? Psychological Assessment 9:28-33. Karieva, P., and U . Wennergren. 1995. Connecting landscape patterns to ecosystem and population process. Nature 373:299-302. Keitt, T. H. , D. L. Urban, and B. T. Milne. 1997. Detecting critical scales in fragmented landscapes. Conservation Ecology 1(1): 4. [online] URL: http://www.consecol.org/voll/issl/art4 Kimmins, J. P. 1991. The future of the forested landscapes of Canada. Forestry Chronicle 67:14-18. Kimmins, J. P. 2002. Future shock in forestry. Forestry Chronicle 78:263-271. 101 Kirchoff, M . D., and J. W. Schoen. 1987. Forest cover and snow: implications for deer habitat in south-east Alaska. Journal of Wildlife Management 51:28-33. Kremsater, L., and F. L. Bunnell. 1992. Testing responses to forest edges: the example of black-tailed deer. Canadian Journal of Zoology 70:2426-2435. Kremsater, L., and F. L. Bunnell. 1999. Edge effects: theory, evidence and implications to management of western North American forests. Pages 117-153 in J.A. Rochelle, L .A. Lehmann, and J. Wisniewski, editors. Forest fragmentation: wildlife and management implications. Koninklijke Bril l , Leiden, Netherlands. Kozak, A. , and R. A . Kozak. 1995. Notes on regression through the origin. Forestry Chronicle 71:326-330. Krummel, J. R., R. H. Gardner, G. Sugihara, R.V. O'Neill, and P. R. Coleman. 1987. Landscape patterns in a disturbed environment. Oikos 48: 321-324. Kuhn, T. S. 1970. The structure of scientific revolutions. University of Chicago Press, Chicago, Illinois, USA. Legendre, P. 1993. Spatial autocorrelation: trouble or new paradigm? Ecology 74:1659-1673. Legendre, P., and M-J. Fortin. 1989. Spatial pattern and ecological analysis. Vegetatio 80:107-138. Leopold, A . 1933. Game management. Charles Scribner and Sons, New York, New York, USA. Levins, R. 1970. Extinction. Pages 77-107 in M . Gerstenhaber, editor. Lectures on mathematics in the life sciences. Volume 2. American Mathematics Society, Providence, Rhode Island, USA. Lord, J. M . , and D. A . Norton. 1990. Scale and the spatial concept of fragmentation. Conservation Biology 4:197-202. MacArthur, R. H. , and E. O. Wilson. 1967. The theory of island biogeography. Princeton University Press, Princeton, New Jersey, USA. MacClintock, L. , R. F. Whitcomb, and B. L. Whitcomb. 1977. Evidence for the value of corridors and minimization of isolation in preservation of biotic diversity. American Birds 31:6-13. Mantel, N . 1967. The detection of disease clustering and a generalized regression approach. Cancer Research 27: 209-220. McGarigal, K. , and S. Cushman. 2002. Comparitive evaluation of experimental approaches to the study of habitat fragmentation effects. Ecological Applications 12:335-345. McGarigal, K. , and W. C. McComb. 1995. Relationships between landscape structure and breeding birds in the Oregon coast range. Ecological Monographs 65:235-260. McGarigal, K. , 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. McCLean, S., M . A . Rumble, R. M . King, and W. L. Baker. 1998. Evaluation of resource selection methods with different definitions of availability. Journal of Wildlife Management 62:793-801. Meyer, J. S., L. L. Irwin, and M . S. Boyce. 1998. Influence of habitat abundance and fragmentation on Northern Spotted Owls in Western Oregon. Wildlife Monographs 139:1-51. 102 Miller, J. R., L. A Joyce, R. L. Knight, and R. M . King. 1996. Forest roads and landscape structure in the southern Rocky Mountains. Landscape Ecology 11:115-127. Miller, R. 1985. Multiple comparisons. Pages 679-689 in S. Kotz, and N . L. Johnson, editors. Encyclopedia of statistical sciences, volume 5. John Wiley and Sons, New York, New York, USA. Mysterud, A. , and E. 0stbye. 1999. Cover as a habitat element for temperate ungulates: effects on habitat selection and demography. .Wildlife Society Bulletin 27:385-394. Nassauer, J. I. 1988. The aesthetics of horticulture: neatness as a form of care. HortScience 23:973-977. Nassauer, J. I. 1995a. Culture and changing landscape structure. Landscape Ecology 10:229-237. Nassauer, J. I. 1995b. Messy ecosystems, orderly frames. Landscape Journal 14:161-170. Nassauer, J. I. 1997. Placing nature: culture and landscape ecology. Island Press, Washington, D.C., USA. Nelson, J. 1999. Operations manual - ATLAS/FPS. Faculty of Forestry, University of British Columbia. Nelson, J., and R. Wells. 2000. The effect of patch size on timber supply and landscape structure. Pages 195-207 in R. G. D'Eon, J. Johnson, and E. A. Ferguson, editors. Ecosystem management of forested landscapes: directions and implementation. U B C Press, Vancouver, British Columbia, Canada. Nelson, J. 2001. Assessment of harvest blocks generated from operational polygons and forest-cover polygons in tactical and strategic planning. Canadian Journal of Forest Research 31:682-693. Nicholson, M . C , R. T. Bowyer, and J. G. Kie. 1997. Habitat selection and survival of mule deer: tradeoffs associated with migration. Journal of Mammalogy 78:483-504. Noss, R. F. 1987. Corridors in real landscapes: a reply to Simberloff and Cox. Conservation Biology 1:159-164. Nyberg, J. B., and D. W. Janz, editors. 1990. Deer and elk habitats in coastal forests of southern British Columbia: a handbook for forest and wildlife managers. British Columbia Ministry of Forests, Special Report Series Number 5, Victoria, British Columbia, Canada. Oksanen. L. 2001. Logic of experiments in ecology: is pseudoreplication a pseudo-issue? Oikos 94:27-38. O'Neill, R. V . , B. T. Milne, M . G. Turner, and R. H . Gardner. 1988. Resource utilization scales and landscape pattern. Landscape Ecology 2:63-69. Orbach, R. 1986. Dynamics of fractal networks. Science 231:814-819. Patton, D. R. 1975. A diversity index for quantifying habitat "edge". Wildlife Society Bulletin 3:171-173. Pimm, S. L. , H . L. Jones, and J. Diamond. 1988. On the risk of extinction. American Naturalist 132:757-785. Pinter, G. G., and V . Pinter. 1998. From epistemology to rational science policy: Popper versus Kuhn. Perspectives in Biology and Medicine 41:291-298. Pither, J., and Taylor, P. D. 1998. An experimental assessment of landscape connectivity. Oikos 83:166-174. Piatt, J .R. 1964. Strong inference. Science 146:347-352. 103 Popper, K. R. 1959. The logic of scientific discovery. Basic Books, New York, New York, USA. Province of British Columbia. 1996. Forest practices code of British Columbia act. Revised statutes of British Columbia, chapter 159. Queen's Printer, Victoria, British Columbia. Ransome, D. B., and T. P. Sullivan. 1997. Food limitation and habitat preference of Glaucomys sabrinus and Tamiasciurus hudsonicus. Journal of Mammalogy 79:538-549. Reed, R., J. Johnson-Barnard, and W. Baker. 1996. Contribution of roads to forest fragmentation in the Rocky Mountains. Conservation Biology 10:1098-1106. Reynolds, R. T., R. T. Graham, R. M . Hildegard, R. L., Bassett, P. L. Kennedy, D. A . Boyce, Jr., G. Goodwin, R. Smith, and E. L. Fisher. 1992. Management recommendations for the northern goshawk in the southwestern United States. General Technical Report R M -217. US Department of Agriculture, Forest Service, Rocky Mountain Forest and Range Experiment Station, Ft. Collins, Colorado, USA. Ripple, W., G. Bradshaw, and T. Spies. 1991. Measuring forest landscape patterns in the Cascade range of Oregon, U.S.A. Conservation Biology 57:73-88. Riitters, K . H. , R. V . O'Neill, C. T. Hunsaker, J. D. Wickham, D. H. Yankee, S. P. Timmins, K. B. Jones, and B. L. Jackson. 1995. A factor analysis of landscape pattern and structure metrics. Landscape Ecology 10:23-39. Robinson, S. K. , F. R. Thompson, T. M . Donovan, D. R. Whitehead, and J. Faaborg. 1995. Regional forest fragmentation and the nesting success of migratory birds. Science 267:1987-1990. Rodgers, A . R., R. S. Rempel, and K. F. Abraham. 1996. A GPS-based telemetry system. Wildlife Society Bulletin 24:559-566. Romesburg, H . C. 1981. Wildlife science: gaining reliable knowledge. Journal of wildlife management 45:293-313. Romin, L. A. , and J. A . Bissonette. 1996. Deer-vehicle collisions: status of state monitoring activities and mitigation efforts. Wildlife Society Bulletin 24:276-283. Rosenberg, D. K. , B. R. Noon, and E. C. Meslow. 1997. Biological corridors: form, function, and efficacy. Bioscience 47:677-687. Rost, G. R., and J. A . Bailey. 1979. Distribution of mule deer and elk in relation to roads. Journal of Wildlife Mangement 43:634-641. Rowe, J. S. 1972. Forest regions of Canada. Canadian Forest Service, Publication Number 1300, Ottawa, Ontario, Canada. Saunders, D., R. Hobbs, and C. Margules. 1991. Biological consequences of ecosystem fragmentation: a review. Conservation Biology 5:18-32. Schieck, J., K. Lertzman, B. Nyberg, and R. Page. 1995. Effects of patch size on birds in old-growth montane forests. Conservation Biology 5:1072-1084. Schmiegelow, F. K. , and M . Monkkonen. 2002. Habitat loss and fragmentation in dynamic landscapes: avian perspectives from the boreal forest. Ecological Applications 12:375-389. Schumaker, N . 1996. Using landscape indices to predict habitat connectivity. Ecology 77:1210-1225. Senft, R. L. , M . B. Coughenour, D. W. Bailey, L. R. Rittenhouse, O. E. Sala, and D. M . Swift. 1987. Large herbivore foraging and ecological hierarchies. Bioscience 37:789-799. 104 Shackleton, D. 1999. Hoofed mammals of British Columbia. Royal British Columbia Museum Handbook, U B C Press, Vancouver, British Columbia, Canada. Simberloff, D., and J. Cox. 1987. Consequences and costs of conservation corridors. Conservation Biology 1:63-71. Simberloff, D., J. A . Fair, J. Cox, and D. W. Mehlman. 1992. Movement corridors: conservation bargains or poor investments? Conservation Biology 6:493-504. Small, M . F., and M . L. Hunter. 1988. Forest fragmentation and avain nest predation in forested landscapes. Oecologia 76:62-64. Snyder, J. E., and J. A . Bissonette. 1987. Marten use of clear-cuttings and residual forest stands in western Newfoundland. Canadian Journal of Zoology. 65:169-174. SPSS 1996. SYSTAT® 7.0 for windows. SPSS Inc. Chicago Illinois, USA. SPSS 1998. SYSTAT® 8.0 for windows. SPSS Inc., Chicago, Illinois, USA. Stauffer, D. 1985. Introduction to percolation theory. Taylor and Francis, London, United Kingdom. Steidl, R. J., J. P. Hayes, and E. Schauber. 1997. Statistical power analysis in wildlife research. Journal of Wildlife Management 61:270-279. Sutherland, G. D., A . S. Harestad, K. Price, and K. P. Lertzman. 2000. Scaling of natal dispersal distances in terrestrial birds and mammals. Conservation Ecology 4(1): 16. [online] URL: http://www.consecol.org/vol4/issl/artl6 Tabachnick, B. G., and L. S. Fidell. 1996. Using multivariate statistics. HarperCollins College Publishers, New York, New York, USA. Taub, J. M . 1998. Eysenck's descriptive and biological theory of personality: a review of construct validity. International Journal of Neuroscience 94:145-197. Taylor, P. D., L. Fahrig, K . Henein, and G. Merriam. 1993. Connectivity is a vital element of landscape structure. Oikos 68:571-573. Thomas, D. L., and E. J. Taylor. 1990. Study designs and tests for comparing resource use and availability. Journal of Wildlife Management 54:322-330. Thomas, J. W., editor. 1979. Wildlife habitats in managed forests: the Blue Mountains of Oregon and Washington. United States Department of Agriculture, Forest Service Handbook Number 553. Thomas, L. 1997. Retrospective power analysis. Conservation Biology 11:276-280. Thomas, T. R., and L. R. Irby. 1990. Habitat use and movement patterns during migration by mule deer in southeastern Idaho. Northwest Science 64:19-27. Tinker, D. B., C. A . C. Resor, G. P. Beauvais, K. F. Kipfmueller, C. I. Fernandes, and W. L . Baker. 1998. Watershed analysis of forest fragmentation by clearcuts and roads in a Wyoming forest. Landscape Ecology 13:149-165. Tishendorf, L. 2001. Can landscape indices predict ecological processes consistently? Landscape Ecology 16:235-254. Tischendorf, L., and L. Fahrig. 2000a. How should we measure landscape connectivity? Landscape Ecology 15:633-641. Tishendorf, L., and L. Fahrig. 2000b. On the usage and measurement of landscape connectivity. Oikos 90:7-19. Trzcinski, M . K. , L. Fahrig, and G. Merriam. 1999. Independent effects of forest cover and fragmentation on the distribution of forest breeding birds. Ecological Applications 9:586-593. 105 Turchin, P. 1988. Quantitative analysis of movement: measuring and modeling population redistribution in animals and plants. Sinauer Associates, Sunderland, Massachusetts, USA. Turchin, P. 1996. Fractal analyses of animal movement: a critique. Ecology 77:2086-2090. Turner, M . G. 1989. Landscape ecology: the effect of pattern on process. Annual Review of Ecology and Systematics 20:171-97. Turner, M . G., and R. H. Gardner. 1991. Quantitative methods in landscape ecology: an introduction. Pages 3-14 in: M . G. Turner and R. H. Gardner, editors. Quantitative methods in landscape ecology. Springer-Verlag, New York, New York, USA. Turner, M . G., and C. L. Rusher. 1988. Changes in the spatial patterns of land use in Georgia. Landscape Ecology 1:241 -251. Underhill, J.E., and P.G. Angold. 2000. Effects of roads on wildlife in an intensively modified landscape. Environmental Review 8:21-39. Unsworth, J. W., D. F. Pac, G. C. White, and R. M . Bartmann. 1999. Mule deer survival in Colorado, Idaho, and Montana. Journal of Wildlife Management 63:315—326. van Dorp, D., and P. F. M . Opdam. 1987. Effects of patch size, isolation, and regional abundance on forest bird communities. Landscape Ecology 1:59-73. Villard, M - A . 2002. Habitat fragmentation: major conservation issue or intellectual attractor? Ecological Applications 12:319-320. Villard, M - A . , M . K. Trzcinski, and G. Merriam. 1999. Fragmentation effects on forest birds: relative influence of woodland cover and configuration on landscape occupancy. Conservation Biology 13:774-783. Wegner, J. F., and G. Merriam. 1979. Movements by birds and small mammals between a wood and adjoining farmland habitats. Journal of Applied Ecology 16:349-358. Wells, D. E. editor. 1986. Guide to GPS positioning. Canadian GPS Association, Fredericton, New Brunswick, Canada. Whitcomb, R. F. 1977. Island biogeography and "habitat islands" of eastern forests. I. Introduction. American Birds 31:17-23. White, G. C , and R. A. Garrott. 1990. Analysis of wildlife radio-tracking data. Academic Press, New York, New York, USA. Wiens, J. A . 1989. Spatial scaling in ecology. Functional Ecology 3:385-397. Wiens, J. 1996. Wildlife in patchy environments: metapopulations, mosaics, and management. Pages 53 to 84 in: D. R. McCullough, editor. Metapopulations and wildlife conservation. Island Press, Washington, DC, USA. Wilcox, B. A. , and D. D. Murphy. 1985. Conservation strategy: the effects of fragmentation on extinction. American Naturalist 125:879-887. Wildlink. 1990. Wildlink data acquisition system. Reference manual. Advanced Telemetry Systems, Isanti, Minnesota, USA. With, K . 1997. The application of neutral landscape models in conservation biology. Conservation Biology 5:1069-1080. With, K . 1999. Is landscape connectivity necessary and sufficient for wildlife management? Pages 97-115 in: Rochelle, J. A. , L. A . Lehmann, and J. Wisniewski, editors. Forest fragmentation: wildlife and management implications. Koninklijke Bril l N V , Leiden, The Netherlands. With, K . A . , and T. O. Crist. 1995. Critical thresholds in species' responses to landscape structure. Ecology 76:2446-2459. 106 With, K. , R. H . Gardner, and M . G. Turner. 1997. Landscape connectivity and population distributions in heterogeneous environments. Oikos 78:151 -169. With, K . A . , and A . W. King. 1999. Extinction thresholds for species in fractal landscapes. Conservation Biology 13:314-326. Yahner, R., and C. Mahan. 1996. Depredation of artificial ground nests in a managed, forested landscape. Conservation Biology 10:285-288. Zar, J. H. 1984. Biostatistical analysis. Prentice-Hall Inc., Englewood Cliffs, New Jersey, USA. 107 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.831.1-0075054/manifest

Comment

Related Items