Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Managing for landscape patterns in the sub-boreal forests of British Columbia Andison, David W. 1996

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

Item Metadata


831-ubc_1996-147223.pdf [ 10.6MB ]
JSON: 831-1.0075275.json
JSON-LD: 831-1.0075275-ld.json
RDF/XML (Pretty): 831-1.0075275-rdf.xml
RDF/JSON: 831-1.0075275-rdf.json
Turtle: 831-1.0075275-turtle.txt
N-Triples: 831-1.0075275-rdf-ntriples.txt
Original Record: 831-1.0075275-source.json
Full Text

Full Text

MANAGING FOR LANDSCAPE PATTERNS IN T H E SUB-BOREAL FORESTS OF BRITISH COLUMBIA By D A V I D W. ANDISON B.Sc. (For.) University of Toronto 1982 A THESIS SUBMITTED IN PARTIAL F U L F I L M E N T OF T H E REQUIREMENTS FOR THE D E G R E E OF DOCTOR OF PHILOSOPHY in T H E 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 the required standard The University of British Columbia May 1996 ® David W. Andison, 1996 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. Forest Resources Management The University of British Columbia 2357 Main Mall Vancouver, BC, Canada V6T 1W5 Date: Abstract Forest managers in North America are attempting to incorporate a growing number of ecological issues across many different temporal and spatial scales. In response, the traditional approach of managing for individual or groups of species and/or functions, is giving way to managing for a more natural time-space array of natural resources. Better known as "landscape management", the strategy relies heavily on understanding the historical, "natural" processes and patterns at the landscape level. The objective of this dissertation is to develop a better understanding of the natural landscape-level dynamics of a fire dominated landscape in northern British Columbia. The age-class distribution was used to show that stand replacing fires had a fire cycle of 80-100 years. Although stands much older than this average persist, there was strong evidence to suggest that very old stands become more susceptible to some form of disturbance. The most striking feature of the age-class distribution was its lack of stability. Although the vast majority of disturbances were very small, and simple in shape, most of the landscape was comprised of very large disturbances. Forest fires were significantly more active in areas with the driest soils. It was also suggested that fire activity may be higher on south and west-facing slopes. Information from the disturbance regime description was incorporated into a spatially explicit landscape model to create multiple landscape scenes of a "natural" disturbance regime. Disturbance size limits and age eligibility restrictions were then imposed in simulation. ii Generally as size and age restrictions increased, interior forest area decreased, edge density increased, age-class distribution favoured younger forests, spatial diversity increased, and temporal diversity decreased. The detailed results demonstrated the importance of using a variety of pattern metrics, in different forms, to grasp the full impacts of each type and degree of restriction. The two methods of investigating sub-boreal landscape dynamics used in this research demonstrated the dangers of using single landscape "snapshots" to represent "natural" conditions. It also raises important questions with respect to the wisdom of management practices which artificially stabilize both spatial and temporal attributes of landscapes. iii Table of Contents Abstract i i Table of Contents , iv List of Tables vii List of Figures ix Acknowledgement xi C H A P T E R 1 - INTRODUCTION 1 1.1 B A C K G R O U N D 1 1.2 GOALS A N D OBJECTIVES 6 1.3 T H E SUB-BOREAL SPRUCE L A N D S C A P E 7 C H A P T E R 2 - SUB-BOREAL L A N D S C A P E E C O L O G Y 10 2.1 L I T E R A T U R E REVIEW 10 2.1.1 DISTURBANCE 11 2.1.2 FOREST FIRES 13 2.1.3 S M A L L - S C A L E DISTURBANCES 20 2.1.4 A R B O R E A L SUCCESSION 21 2.1.5 STAND INITIATION 23 2.2 HYPOTHESES A N D QUESTIONS 27 2.3 D A T A 28 2.4 METHODS A N D RESULTS 31 2.4.1 A G E I N V A R I A N C E 31 A N A L Y S I S 31 DISCUSSION 35 2.4.2 FIRE C Y C L E S 39 A N A L Y S I S 40 DISCUSSION 40 2.4.3 DISTURBANCE RATES 41 A N A L Y S I S 41 DISCUSSION 44 2.4.4 M U L T I P L E FIRE REGIMES 48 A N A L Y S I S 48 DISCUSSION 52 2.4.5 FINE S C A L E TOPOGRAPHIC CONTROLS 53 CREEKS 53 A N A L Y S I S 53 DISCUSSION 54 iv ASPECT 55 A N A L Y S I S 55 DISCUSSION 57 2.4.6 P A T C H SIZES 58 A N A L Y S I S 58 DISCUSSION 64 2.4.7 DISTURBANCE SHAPE 65 A N A L Y S I S 65 DISCUSSION 67 2.4.8 SPECIES / SITE RELATIONSHIPS 67 A N A L Y S I S 68 DISCUSSION 72 2.5 S U M M A R Y ^ 74 C H A P T E R 3 - L A N D S C A P E M O D E L L I N G 80 3.1 L I T E R A T U R E REVIEW 80 3.1.1 S P A T I A L L Y EXPLICIT D A T A 81 3.1.2 DISTURBANCE M O D E L S 84 3.1.3 V E G E T A T I O N D Y N A M I C S M O D E L S 87 3.1.4 L A N D S C A P E M O D E L S 90 3.1.5 M E A S U R I N G PATTERN 93 3.2 A L A N D S C A P E M O D E L FOR THE SBSmkl 95 3.2.1 M O D E L A N D D A T A OVERVIEW 96 3.2.2 DISTURBANCE RATES 99 3.2.3 FIRE SIZES 103 3.2.4 FIRE BEHAVIOUR 105 3.2.5 ESTABLISHMENT 110 3.2.6 SUCCESSION I l l 3.2.7 P A T T E R N ASSESSMENT I l l 3.3 M O D E L V A L I D A T I O N 113 3.3.1 DISTURBANCE SHAPES 114 3.3.2 R E M N A N T ISLAND F O R M A T I O N 116 3.4 S U M M A R Y 118 C H A P T E R 4 - SIMULATION M O D E L L I N G IN T H E SBSmkl 121 4.1 LITERATURE REVIEW . . . 121 4.1.1 L A N D S C A P E PATTERN M E A S U R E M E N T 121 4.1.2 L A N D S C A P E S A N D SIMULATION 125 4.2 PURPOSE 126 4.3 SIMULATION 127 4.3.1 METHODS 127 4.3.2 RESULTS 130 AGE-CLASS DISTRIBUTION 130 INTERIOR FOREST A R E A 134 v E D G E DENSITY 141 P A T C H SIZES 144 A D J A C E N C Y 148 DIVERSITY 151 SPATIAL DIVERSITY 151 T E M P O R A L DIVERSITY 153 4.4 DISCUSSION A N D S U M M A R Y 156 C H A P T E R 5 - DISCUSSION A N D CONCLUSIONS 163 5.1 G E N E R A L DISCUSSION 163 5.2 CONCLUSIONS 170 L I T E R A T U R E CITED 173 APPENDIX A - DEFINING (INVENTORY) A G E - C L A S S 8 189 APPENDIX B - D E T A I L E D SIMULATION RESULTS 193 vi List of Tables Table 2.1. Table 2.2. Table 2.3. Table 2.4. Table 2.5. Table 3.1. Table 3.2. Table 3.3. Table 3.4. Table 3.5 Table 3.6 Table 4.1 Table 4.2. Table 4.3 Table 4.4. Table 4.5 Table 4.6 Table 4.7 Table 4.8 Table 4.9 Table 4.10 Table 4.11 Table 4.12 Table A l . Life table for SBSmkl stands by age-class for the negative exponential model, and Harter and Moore's (1965) and Menon's (1963) Weibull models showing the number of survivors through time based on initial values of 1,000 stems. . 37 Estimated disturbance rates for each 20-year period for the SBSmkl based on rolling back age-classes. Two projections were calculated based on an 80 and a 100 year fire cycle 43 Area and average age by soil-type for the SBSmkl 49 Species percentages by soil-type and slope-class on the SBSmkl 69 Percentage of pure stands by slope-class in the SBSmkl 72 Original and adjusted 20-year disturbance rates for the SBSmkl 100 Fuel-types and rate-of-spread (ROS) values for the landscape model. . . . 108 Rate-of-spread (ROS) slope correction factors for the landscape model fuel-types 108 Edge contrast values for patch definition for the SBSmkl 112 A comparison of actual to simulated percent of fire area in remnant islands by disturbance size-class in the SBSmkl landscape 117 Comparison of actual to simulated areas in island remnants by size-class for disturbances on the SBSmkl landscape 118 Summary of simulated disturbance scenarios for the SBSmkl landscape. . 128 Summary of simulated age-class percentages for the SBSmkl landscape. . 131 Average core area index (percent) for SBSmkl landscape simulations. . . . 136 Average weighted edge density (m/ha) for the SBSmkl landscape simulations 142 Mean patch size (ha) / patch size standard deviation (ha) for SBSmkl landscape simulations 145 Average interspersion / juxtaposition index for young forest (0-40 years) for the SBSmkl landscape simulations 150 Average interspersion / juxtaposition index for pole forest (41-120 years) for the SBSmkl landscape simulations 150 Average interspersion / juxtaposition index for mature forest (120+ years) for the SBSmkl landscape simulations 151 Average Shannon's evenness index for the SBSmkl landscape simulations. 152 Average change over 20 years of percent young forest (0-40 years) for the SBSmkl landscape simulations 153 Average percent change over 20 years of pole forest (41-120 years) for the SBSmkl landscape simulations 154 Average change over 20 years of mature forest (120+ years) for the SBSmkl landscape simulations 154 Number (and percentage in brackets) of sampled plots and stands of mature forest in the SBSmkl in 20 year age-classes 192 Vll Table B l . Simulation results summary of landscape average and standard deviations. 194 Table B2. Simulation results summary of young age-class (0-40 years) averages and standard deviations 195 Table B3. Simulation results summary of pole age-class (41-120 years) averages and standard deviations 196 Table B4. Simulation results summary of old age-class (121 + years) averages and standard deviations 197 viii List of Figures Figure 2.1. SBSmkl time-since-last-fire distribution. Comparison of actual distribution to Harter and Moore's (1965) maximum likelihood model (Weibull A) , Menon's (1963) regression. estimator Weibull model (Weibull B), and the negative exponential model 34 Figure 2.2. Comparison of the actual SBSmkl age-class distribution with two hypothetical models 46 Figure 2.3. SBSmkl age-class distribution by soil drainage class 50 Figure 2.4. SBSmkl age-class distribution by topographic complexity 51 Figure 2.5. Observed areas of old and young forest by aspect on slopes of >10% in the SBSmkl, as a percentage of expected area 57 Figure 2.6. Observed SBSmkl patch size-class distribution for the youngest age-classes. 60 Figure 2.7. Area distributed by patch size-class in the SBSmkl for the youngest age-classes 60 Figure 2.8. Comparison of observed patch sizes (raw data) to estimated disturbance sizes (predicted from model) for age-class 2 on the SBSmkl 62 Figure 2.9. Comparison of actual patch sizes to estimated disturbance sizes and estimated (total) patch sizes for age-class 8 on the SBSmkl 63 Figure 2.10. SBSmkl patch shape by patch size for the youngest age-classes 66 Figure 3.1. Flowchart of SBSmkl Landscape Model 97 Figure 3.2. SBSmkl cumulative 20-year disturbance rate probability based on a 100-year fire return interval, adjusted to allow for reburning 101 Figure 3.3 Disturbance size function for the SBSmkl based on estimated number of disturbance patches from age-class 2 (based on the simulation exercise in Section 2.4.6) 104 Figure 3.4. Cumulative disturbance size function for the SBSmkl using estimated disturbance patches from age-class 2 based on the simulation exercise in Section 2.4.6. 105 Figure 3.5. A comparison of actual to simulated data for disturbance shapes on the SBSmkl landscape 116 Figure 4.1. Frequency distribution of percent young and mature age-classes for the natural (a), 10,000 ha (b) and 60 ha (c) scenarios for the SBSmkl landscape simulations (out of 50 runs) 133 Figure 4.2. Frequency of Total Core Area Index values for the mature seral-stage, as calculated by FRAGSTATS, for the natural simulation scenario, and the 20-10,000 and 10,000 ha size restricted scenarios (a), and the 20-1,000 and 1,000 ha size restricted scenarios (b) based on 50 simulation runs 137 Figure 4.3: Frequency of Total Core Area Index values, as calculated by F R A G S T A T S , for young, pole, and mature age-classes for the unrestricted (natural) simulation scenario for the SBSmkl simulations (based on 50 runs) 140 IX Figure 4.4. Average total core area index, as calculated by FRAGSTATS, for young, pole, and mature seral-stages, for the 10,000 hectares size restriction for both the 20-year and the 40-year eligibility restriction scenarios for the SBSmkl (based on 50 runs) 140 Figure 4.5. Frequency of weighted edge density values for natural, 10,000, and 20-10,000 hectare scenarios (a), and for 1,000 and 20-1,000 hectare scenarios (b) for the SBSmkl (based on 50 simulations) 143 Figure 4.6. Frequency distribution of patch size standard deviation for natural, 10,000, and 1,000 hectare scenarios for the SBSmkl landscape simulations (based on 50 runs) 146 Figure 4.7. Frequency distribution of patch size standard deviations for the natural, 80-year, and 80-year plus buffer simulation scenarios for the SBSmkl (based on 50 runs) 147 Figure 4.8. Average patch size standard deviation for young, pole, and mature seral-stages for natural, 40-year, and 3,000 hectare restriction simulation scenarios. . . 148 x Acknowledgement I would like to begin by acknowledging the assistance, support, and encouragement over the past several years from my committee: Drs. Gary Bradfield, Phil Burton, Brad Hawkes, Hamish Kimmins, Peter Marshall, and David Tait. In particular I would like to thank my supervisor, Dr. Peter Marshall for the time and effort expended on my behalf, and in particular for the well-timed injections of reality. I also appreciate the polite indulgence afforded me by Peter, my committee, and all of my friends and colleagues in the Faculty, with respect to my 'early' ideas. I would also like to thank Craig DeLong and John Parminter from the B .C . Ministry of Forests for their unabated support from day one, both morally and financially, and Kathy Chartrand for her GIS expertise and insight. In addition, I am grateful to the Department of Forest Management at UBC, Forestry Canada, MacMillan Bloedel, and the Natural Sciences and Engineering Research Council of Canada (NSERC) for financial support. Finally, I would like to thank my parents for their own form of patience and support, even though they still don't know what I "do", and Lori for putting up with the whole darned thing. xi CHAPTER 1 - INTRODUCTION 1.1 BACKGROUND Thirty years ago, the value of forested land was measured largely by the amount of timber it held, or could produce. This attitude was borne out of a commonly held view in North America that forests were virtually limitless and infinitely resilient. Forest management reflected this timber-oriented goal by planning primarily for sustainable supplies of renewable fibre. In time, increased public concern over alternative forest values such as recreation, aesthetics, and large game led to the current era of multiple-use principles. The goal of multiple-use management is to allow for alternative uses of the forest within a timber management context. The legacy of timber-based computer models such as F O R P L A N and M U S Y C is indicative of the trend of managing forests based on the productive capacity of the land (Zonneveld 1987, Weintraub and Davis 1992). Today, forests are valued for much more than a renewable source of timber, recreational opportunities, and game species. The roles forests play in regulating water and nutrient flows, fixing atmospheric carbon, influencing climatic patterns, absorbing and transforming atmospheric and water-borne toxins, and providing the opportunity for the continued existence and evolution of flora, fauna, and aquatic species are also recognized (World Resources Institute 1992, Davis 1993). These functions are often referred to as "biological values", and it is believed that they are fundamental to the long-term sustainability of all aspects of forest systems, including the production of timber. Although monetary value for biological values is difficult to determine, 1 there are potentially high costs associated with inappropriate manipulation or removal of forest systems (Davis 1993). Trying to incorporate biological values into the current process of forest management planning has proven extremely difficult. One reason for this difficulty is the lack of knowledge concerning biological systems; however, equally important is the current structure of forest management planning systems. Most of the current planning models focus on production (timber) as the primary value, and allow all other values (biological as well as cultural) to constrain levels of allowable timber extraction qualitatively or quantitatively. There is increasing concern that the continuation of a management system based on timber as the primary value is not likely to effectively manage for these other values (Weintraub and Davis 1992). As a result, public agencies are considering a shift in the primary focus of forest management planning from timber to the maintenance of biological values. The assumption in doing so is that by maintaining a priority on fundamental biological functions, all other values, including a sustainable timber supply, will be maintained as well. These ideals form the foundation of what is now known as "landscape management" (Society of American Foresters 1993). As objects of management, landscapes are defined in two ways. First, they are three-dimensional entities which envelop all biological and physical characteristics including geologic, hydrologic, topographic, pedologic, climatologic, ecologic, physiologic, and zoologic (Troll 1971). In other words, they are ecosystems1. Second, landscapes must be delineated such that the operation, Although an 'ecosystem' is a conceptual entity essentially describing the biota and its environment, the term is often used to refer to any recognizable physical area over which the biota and the environment are relatively homogeneous. This description more accurately describes a biogeocoenose (Kimmins 1987). 2 observation, understanding, and management of one or all of the large-scale2 biological functions, listed eariler, is possible. The Society of American Foresters (1993) refers to landscapes as the "minimum area within which ecological function can be considered renewable", although in reality it is difficult to imagine considering anything less than the biosphere as having that capability. However, based on our current understanding, the most important factors that contribute to the goal of renewability on large scales are the types and behaviour of disturbances, spatial geomorphic heterogeneity, and the colonization, growth, and mortality attributes of the biota (Forman and Godron 1986, Gauthier et al. 1995). Therefore, any area that is large enough in which to understand the interaction of these factors, and observe the patch "mosaic" that results, would be a landscape. In many cases, landscapes are defined as areas in which these factors have a consistent relationship to each other (/. e., internally consistent relationships between disturbance and topography), although in practice this is not always convenient or desirable. The size and boundaries of landscapes are therefore both highly variable and subjective, but in most cases the area is between thousands and hundreds of thousands of hectares. In 1992, the US Forest Service was the first national agency to officially adopt a management policy explicitly defining the primary goal of forest management to be the "sustained health of forest landscapes" as natural systems. Referred to as "ecosystem management", forest management is to be based on the continued function of "natural"3 processes over long periods "Scale", in etiher time or space, is defined by both grain (smallest homogeneous unit of measurement and observation) and extent (the total area of observation). In this text, "large-scale" or "coarse-scale" refer to large grain and broad extents. "Small-scale" or "fine-scale" refer to small grain, and narrow extent. 3 I will consider a "natural" landscape to be one that has experienced a minimum of human influence. 1 will further assume that natural landscapes are qualitatively superior in their ability to maintain the biological values (Noss 1994). This is not to say that humans are not, or never were, a part of landscapes, but rather that historically, the vast majority of known cases of intervention resulted in deterioration of one or more of the ecological functions we now value (Marsh 1864, Leopold 1949, Davis 1993, Noss 1994). 3 of time, within which timber management is to take place (Society of American Foresters 1993). In Canada, policies under the auspices of landscape management have been developed at the national level (Booth et al. 1992), and are inclusive of strategies at the Provincial level in British Columbia (B.C.) (Hamilton 1994) and Ontario (Balsillie 1994). The primary goal of the Canadian version is similarly focused on providing an ecological context for forest management. Despite the conceptual attractiveness of landscape management ideas, it remains to translate these principles into practice (Slocombe 1993). One option is to manage forests on the basis of the needs of each species within a landscape. However, the only chance the species-by-species approach to management has of being successful is when we have a thorough understanding of the interaction between and within species, communities, and their environments. That level of information does not yet exist; in fact, we cannot even name all of the species in a given landscape. Good information is available on some species and communities, but assuming that the well-being of one or several species locally, or the protection of single biogeocoenoses, will lead to the continuation of the full range of biological functions may prove to be incorrect (Boyce and McNab 1992). In fact, local extinctions, periodic high mortality levels, and other perturbations are a part of natural landscape cycles, and the maintenance of the diversity of populations or communities often conflicts with that of landscapes (Bunce and Jongman 1993). Alternatively, landscape functions can be managed through an understanding of the available time-space array of resources provided by a dynamic landscape (Merriam and Wegner 1992). By first understanding, and then managing for, the natural range in variation of the ecological processes most active on a landscape, the risk of losing biological function is minimized, since the rate, intensity, and magnitude of the processes are familiar to the landscape (Noss 1994). 4 This is a passive form of management which presumes there is a direct link between landscape structure and (biological) function (Franklin 1993). Furthermore, it respects that we do not yet have complete information on the existence, importance and resilience of all the functions. In practical terms, this suggests that a tenable strategy for landscape management is to approximate the structures that natural landscapes exhibit (Franklin 1993, Gauthier et al. 1995). Sweden has already developed a broad-based forest management model for the boreal forest based on the knowledge of the natural fire regime (Riilcker et al. 1994); however, there is an increasing demand for a more detailed model of natural landscape dynamics. This strategy has often been referred to as "mimicry" of natural systems. In accepting mimicry as a premise for management, it is also assumed that the structure of landscapes can be identified and quantified through various forms of pattern analysis. A landscape could then be recognized as possessing a well defined range of such structures, representing dynamic "behaviour". Based on our rudimentary understanding of landscape processes, our perception is that "natural" landscapes are highly stochastic, and are capable of a broad range of structural patterns. For instance, the present stand mosaic pattern is but one of many possible natural configurations over time. Similarly, it could be argued that some aspects of the stand mosaic formed by current cutting patterns is natural in the sense that it is statistically possible to observe a similar pattern on a natural landscape, however remotely. The difference is one of degree. It is likely that the probability of the resulting landscape patterns being observed under more natural conditions will decrease dramatically should present harvesting patterns continue. Furthermore, believing that a single landscape is capable of creating such a tremendous range of patterns diminishes the utility of analyzing single pattern "snapshots" in 5 time. A more meaningful question would be whether or not, or in what manner, mosaic patterns resulting from present or proposed management practices through time are similar to those observed on natural landscapes. This is one of the main premises of mimicry, and it has not been rigorously addressed. 1.2 GOALS AND OBJECTIVES The primary focus of this research is to determine how simulated landscape mosaic patterns change under the imposition of alternative disturbance rules. Specifically, I am interested in those aspects of the disturbance regime that forest management would impose on a 'natural' landscape. Computer simulation is employed because comparison of natural to managed landscapes is difficult. Since a given landscape could exhibit a wide range of possible mosaic patterns under natural conditions, observing such behaviour would take centuries, and finding replicate landscapes for experimentation would be impossible. Alternatively, knowledge of landscape processes can be gained through historical reconstruction techniques and other retrospective methods. Computer technology today is such that the ability exists to project this static knowledge temporally in order to see how historical variability manifests itself through time in the form of patterns. The main body of this thesis is organized into three interconnected parts (Chapters 2, 3, and 4), each with its own objective, literature review, methods, results, discussion, and summary. In Chapter 2 the dominant ecological processes at the landscape scale of a sub-boreal spruce (SBS) landscape are identified and described. Some hypothesis testing is presented with respect to the local disturbance regime. Chapter 3 describes the construction and verification of a spatially-6 explicit computer model of this landscape using the results from Chapter 2. In the fourth chapter, the computer model described in Chapter 3 is used to investigate the sensitivity of the landscape mosaic patterns to those disturbance regime parameters that are most likely to change under forest management. The natural disturbance regime parameters of the SBSmkl determined in Chapter 2 and 3 were used as the basis for the model. A general discussion and conclusions are provided in Chapter 5. There are both practical and theoretical contributions from this research. Well defined fire regime information for boreal forest-types is rare to non-existent in B.C. The descriptive information on the SBSmkl fire regime will be of value to the evolution and use of the Provincial Forest Practices Code Biodiversity Guidelines (B.C. Ministry of Forests and B.C. Environment 1995). Furthermore, although numerous single landscape (snapshot) pattern comparisons have been conducted, no known published comprehensive comparison has been done of multiple patterns created by spatial simulations based on empirical disturbance regime information. 1.3 THE SUB-BOREAL SPRUCE LANDSCAPE The British Columbia Ministry of Forests (MoF) is interested in developing landscape management guidelines for several northern interior forest-types (Hamilton 1994). The Sub-Boreal Spruce (SBS) zone (DeLong et al. 1993) has already been targeted for landscape research within the Provincial landscape ecology project, and is an ideal candidate for this research. Natural disturbance cycles are thought to be similar to harvesting rotations, the ecology and silviculture of SBS species and communities is well understood relative to other areas of B.C. , and clear-cut harvesting strategies have the potential to approximate disturbance effects on 7 landscape scales. On the other hand, there is growing concern that the current clear-cut limit of 60-80 hectares is resulting in landscapes that have too many small, regularly spaced disturbances. Such a mosaic pattern, often referred to as "fragmented", is a concern in many North American forest landscapes (Li et al. 1993). Ironically, the fragmentation harvesting strategy was developed in response to concerns about the original harvesting strategy of progressive clearcutting, which was perceived as being a much greater threat ecologically. Alternatives to the fragmented harvesting pattern are being suggested in the new Biodiversity Guidelines for the B.C. Forest Practices Code (B.C. Ministry of Forests and B.C. Environment 1995). I have chosen to study the largest and perhaps simplest variant of the SBS forests. The Mossvale Moist Cool Sub-Boreal Spruce variant (SBSmkl), according to the British Columbia Biogeoclimatic Ecosystem Classification (BEC), covers about 790,000 hectares on an interior plateau north of Prince George (DeLong et al. 1993). The climate of the SBSmkl is typically continental (long winters and moist cool summers), and varies little across its geographic extent. Soils are Grey Luvisols to Podzols, and the topography is gently rolling (DeLong et al. 1993). Principle tree species are lodgepole pine {Pinus contorta Dougl.) and hybrid white spruce (Picea glauca x engelmannii) with minor components of black spruce {Picea mariana (Mill.) B.S.P.), trembling aspen {Populus tremuloides Michx.), black Cottonwood {Populus trichocarpa Torr. & Gray), subalpine fir {Abies lasiocarpa (Hook) Nutt.) and Douglas-fir {Pseudostuga menseisii Mirb. (Franco)). Forest fires, bark beetle {Dendroctonus spp.) and windthrow are the dominant disturbances in the 8 SBSmkl (C. DeLong 4, pers. comm.). Although fires are often associated with the other two events, bark beetle and windthrow also tend to occur on smaller scales than fire. The fire cycle has been estimated to be 100-150 years (B.C. Ministry of Forests and B.C. Environment 1995) and most fires lead to high mortality and even-aged stands (Johnson 1992). It is thought that arboreal succession is not a significant landscape factor in this area because of the short period between fires, and the susceptibility of these species to fire (Johnson 1992, Payette 1993). Existing stand composition is largely a function of the severity, size and uniformity of the initiating fire, the age and composition of the previous stand, and species strategies and preferences (Parminter 1983). In many instances, the composition of the previous stand is thought to be a good indicator of the post-disturbance stand. Regional Ecologist, Prince George Region, B.C. Ministry of Forests. 9 CHAPTER 2 - SUB-BOREAL LANDSCAPE ECOLOGY The purpose of this chapter is twofold; first, to gain a better understanding of disturbance dynamics in boreal-type landscapes, and second, to create the necessary information to parameterize a landscape computer model for the simulation exercise to follow. A review of the relevant literature is followed by a summary of the (eight) main questions and hypotheses that will be addressed, a description of the datasets used, a large results section, and a summary. The results are presented in eight sub-sections, one for each question or hypothesis. Within each of these eight sub-sections, the methods and tests that were used will be described, and the results summarized and discussed. In some instances, modules of the simulation model were employed for more specific hypothesis testing. In each case, the assumptions necessary for the test are outlined, but a full description of the model and its verification is left until Chapter 3. 2.1 LITERATURE REVIEW The mechanisms responsible for the forest mosaic patterns we observe on landscapes are geomorphic processes (pedology, topography, geology), the disturbance regime, and forest stand dynamics (Forman and Godron 1986). It is generally accepted that large-scale disturbance is the dominant process responsible for landscape mosaics in most North American forest-types (Sapsis and Martin 1993, Noss 1994, Gauthier et al. 1995). Within these coarse-scale mosaics are compositional and finer grained structural patterns caused by small-scale disturbances, and succession and stand initiation (two components of forest stand dynamics). Each of these processes affects the sub-boreal forest landscape to varying degrees, and will be reviewed. 10 2.1.1 DISTURBANCE Disturbances are unpredictable, abrupt events causing a change in the structure and/or function of a natural system, often coinciding with the destruction of biomass (Grime 1979, White and Pickett 1985). Common forest disturbances are fire, wind, single or multiple tree mortality (gap formation), ice events, landslides, insect, disease and animal damage, and flooding (White 1979). A l l forest systems have evolved to accommodate and depend on disturbance as a mechanism by which biological functions are maintained (Oliver 1981, Bazzaz 1983, Sapsis and Martin 1993). Disturbances release nutrients tied up in dead and live biomass, increase soil temperatures and the amount of solar radiation reaching the forest floor, increase soil micro-organism activity, and alter seedbed conditions (Heinselman 1980, Rapp 1983, Kimmins 1987). Prolonged absence from disturbance results in declining net primary production (NPP), declining respiration, increased susceptibility to other disturbances, and increased storage of nutrients in dead biomass (Oliver 1981, Reiners 1983, Sapsis and Martin 1993, Sprugel 1985). Disturbance "regimes" are characterised in terms of their type, areal extent, severity (degree of mortality), timing, cycle (time required to disturb an area equivalent in size to the total area in question), and predictability (White 1979, Heinselman 1980, Sousa 1984). The response of a community or ecosystem(s) to disturbance will depend on the interaction of the disturbance characteristics with the life history strategies of dominant species and the physical characteristics of the landscape (White and Pickett 1985). For instance, areal extent affects the ability of species to invade and survive (Reiners 1983), and disturbance intensity determines seedbed conditions and the survival of root stock and stored seed (Oliver 1981). 11 Although disturbance may seem to be a random phenomenon, disturbance regimes and subsequent responses to them are quite landscape specific (di Castri and Hansen 1992). Disturbance regimes that are unfamiliar to forest systems can have significant and unpredictable effects (Runkle 1985, Swaydon 1987, Leemans 1991). The best known such effect is the tremendous build-up of fuel resulting from artificially reducing the disturbance frequency in California forests and grasslands through fire control activities, leading to catastrophic fires (Sapsis and Martin 1993). Reducing disturbance frequencies in northern forest-types also leads to increased fuel build-up (Romme 1982, Ward and Tithecott 1993), as well as increasing the proportion of shade tolerant species, thereby creating more homogeneous landscapes (Methven and Feunekes 1987, Knight 1987). Other notable impacts from altering the natural disturbance regimes include increased erosion, loss of beaches, and declining NPP resulting from flood control; lower productivity, lower species diversity, and lower salt resistance from artificially protecting ocean dunes; and increased susceptibility to pests over time caused by chemical pest control application (White 1979, Odum et al. 1987). Note that in all of the examples given, negative impacts eventually ensued from altering the natural disturbance regime in terms of ecosystem productivity and/or diversity. Harvesting practices resulting in regularly sized and spaced openings across large areas, also represents an unfamiliar disturbance regime and is one of the concerns in the SBSmkl (C. DeLong pers. comm.). Measured responses of different forest-types to fragmentation is quite varied, but it may affect wildlife species dynamics dramatically, even to the point of local extinction (di Castri and Hansen 1992, Hansen et al. 1992, Robinson et al. 1992). Fragmentation can also have potentially negative impacts on water and nutrient cycling, erosion potential, and microorganism, pathogen, and insect activity (Robinson et al. 1992, L i et al. 1993). 12 Lepart and Debussche (1992) consider the inappropriate use of disturbance to be the most serious problem with respect to mankind's intervention in the environment. Landscape management ideally necessitates understanding the natural disturbance regime of a given landscape, and using that knowledge to develop a more familiar (managed) disturbance regime. 2.1.2 FOREST FIRES Forest fires are the most common, natural, large-scale forest disturbance agent worldwide (Suffling 1987). Without fire, many common tree species, from Australian pines to sub-Arctic tundra white spruce, would not persist (Mutch 1970, Bazzaz 1983, Methven and Feunekes 1987). These species are often referred to as fire species because of their evolutionary adaptations to fire. These adaptations include the production of "serotinous" cones (which have a resinous coating that protects seed until such time as the heat from a fire opens them), favourable buildup of flammable litter, and very high juvenile growth rates (Mutch 1970, White 1979). These traits often lead to large areas of pure stands. Despite widespread appreciation of the ecological role of fire in sub-boreal and boreal landscapes, fire is a highly stochastic phenomena, and has proven challenging to study. Forest fire research can be loosely grouped into two classes depending on the scale of study: (1) individual fire behaviour, and (2) (landscape) fire regime. Although it is the fire regime which is most relevant to landscape studies, there are aspects of fire behaviour which are also pertinent. Fire behaviour generally refers to the actions of an individual fire event: what or where it burns, how fast, and at what intensity. Fire behaviour is a function of the type, moisture content, and 13 physical arrangement of the fuel, topography, and weather (Kalabokidis et al. 1991, Turner and Romme 1994). Both the Canadian and American fire danger rating systems use these parameters to predict fire intensity and spread for fire control and prescribed burning purposes (Rothermel 1972, Stocks et al. 1989), although the relative impact of each of these factors varies depending on fire size and intensity. Smaller, less intense crown fires and surface fires are highly influenced by both fuel-type and local topography (Turner and Romme 1994). Quantifying and classifying the effects of both fuel-type and slope has been the focus of considerable fire behaviour research in North America (Anderson 1982, Stocks 1987, Stocks 1989, McAlpine et al. 1991, Van Wagner 1992). The most recent version of the Canadian Forest Fire Danger Rating System (CFFDRS) includes the description of broad fuel types within the Fire Behaviour Prediction (FBP) system (Forestry Canada Fire Danger Group 1992). As an aid to fire managers, the FBP system differentiates sixteen fuel-types according to relative spread rates, intensities, and fuel consumption using the original Canadian fire model (fire weather, topography, date and elevation) (Forestry Canada Fire Danger Group 1992). Under low to moderate fire conditions, CFFDRS fire models (and others) do a reasonable job of predicting fire behaviour. Once fires become large, more intense, and are actively crowning, they are almost exclusively controlled by fire weather conditions (Romme 1982, Zasada et al. 1993, Turner and Romme 1994, Bessie and Johnson 1995). Extended periods of warm dry weather coupled with high winds create conflagrations that are essentially oblivious to topographic or fuel-type changes. Predicting fire behaviour under such conditions becomes more difficult, and fire control is next to impossible. These large fire events are rare, but they are responsible for the vast majority of 14 the area consumed by fire in boreal-type landscapes. The relationship between large and small fire events, and their relative impacts from a landscape perspective are described by the "fire regime". A fire regime is a subset of a disturbance regime, and appropriately refers to long-term landscape level mechanics of fire, such as sizes, frequencies, intensities, and shapes. In other words, a fire regime is the behaviour of fire at landscape scales. The change in scale means that what might be considered random behaviour over a small area and several years, becomes more predictable over large areas and hundreds of years (O'Neill et al. 1986). On landscape scales, fire is far from being a random process. Rather, fire regimes tend to be quite landscape-specific. An example of the phenomenon of scale-dependent perception is the frequency and distribution of the survival of trees within a fire's boundaries. Although boreal wildfires are often referred to as stand replacing (Johnson 1992), there is evidence that severity varies considerably within any given fire (Malanson 1985), meaning that tree survival is common. Within the SBS, Clark (1994) found that both large and small individual trees survived fires in 28% of sampled stands. DeLong and Tanner (1995) and Eberhart and Woodard (1987) found that both the number and sizes of remnant islands of surviving trees increased as the total fire size increased. However, none of these studies could identify permanent landscape features which might explain why a particular tree or island survived. In other words, on a stand scale, survival of individuals or remnant islands may be random, but on a landscape scale, it is possible to describe the occurrence of survivors non-spatially, perhaps as a frequency distribution function. A similar phenomenon occurs with fire sizes. Although it is difficult to predict what size any 15 one particular fire will be, distributions of fire sizes over time often follow a negative exponential function. Taylor et al. (1994) and Ward and Tithecott (1993) found negative exponential fire-size distributions for the Yukon and Ontario, respectively. However, the impact of large fires is such that the associated distribution of area disturbed by each size-class of fire is logarithmic (DeLong and Tanner (1995). In other words, as previously mentioned, large fires are extremely rare, but account for most of the area consumed. A third and final demonstration of a scale-dependent phenomenon of forest fires is fire shape. Although fire shape has not been studied to any great degree, the standard elliptical model of fire has long been recognized as being an oversimplification useful for modelling. Fire perimeters have more recently been described as fractal (McAlpine and Wotton 1993, Clarke et al. 1994, Lorimer et al. 1994). In other words, equivalent amounts of complexity of a fire's edge is revealed with every change in the scale of observation. Many of the numerical parameters and distributions that describe a fire regime are related. The fire return interval, or the average number of years between fire events at a specific location, is the inverse of fire frequency (Johnson 1992). A fire cycle, or the average number of years required to burn an area equivalent to the total landscape in question, is the inverse of the annual percent burned (Johnson 1992). For example, the SBSmkl fire cycle would be the number of years it takes to burn 790,000 hectares. If fires burn equally across all areas of a landscape, and start randomly spatially, then the fire cycle (of the entire landscape) is the same thing as the fire return interval for any given location in the landscape (Johnson and Van Wagner 1985). Furthermore, if fire starts are random, the distribution of fire starts in a single location through time approximates a Poisson distribution (Green 1989). Harrington and Donnelly (1978) 16 successfully used the Poisson model to describe fire starts in several areas of Ontario. The fire return interval for the boreal forest ranges from 20 to 135 years, averaging 65 years (Ward and Tithecott 1993). This translates to an average annual burn rate of about 1.5% of the landscape per year. For the SBSmkl forests, it is thought that the fire return interval is probably longer than in the boreal forest (Parminter 1983), but precise estimates are not available. The rotation age being used in the SBSmkl (the management equivalent of the return interval) is approximately 100 years (J. Pousette5, Pers. Comm.). Unfortunately, very few landscapes accommodate the ideal conditions discussed above. This complicates the definition of a fire regime. Perhaps the most controversial debate with respect to the boreal forest fire regime is whether or not higher levels of fire susceptibility are correlated with increasing stand age (Rowe et al. 1975, Yarie 1981). Van Wagner (1978) first hypothesized that the distribution of fire intervals and the time-since-last-fire distribution will follow a negative exponential model i f fire is age-invariant; that is, i f fire is equally probable in all ages of forest. Typically, age-class distributions associated with disturbances such as fire are considered to be age-invariant in the absence of other information. One of the reasons the negative exponential model is so readily accepted is its simplicity. In the negative exponential model, the average age is the equivalent to both the fire cycle, and the fire return interval (Johnson and Van Wagner 1985). This is consistent with the claim that fires are spatially random, and that the chance of a fire occurring in any one place is described by a Poisson distribution. Under these assumptions, most of the stands in a landscape will be younger than the fire cycle, but because of the random Regional planning specialist, BC Ministry of Forests, Prince George Region. 17 nature of fire spatially, 36.8% of the landscape will have forests older than the fire cycle under ideal conditions (Van Wagner 1978). Other areas of the landscape will burn well before the fire cycle age. If fire is age-selective, the ideal numerical relationships described above no longer hold. The average age will not necessarily be the same as the fire cycle, and the fire cycle will not be the same as the fire return interval (since it will vary with age). The theory of age selection by fire is based on the fact that there is an increase in both vertical and horizontal continuity of fuels, and an increase in the total amount of available fuels with age (Mutch 1970). It has been suggested that i f fire is age-selective, the distribution of return intervals will follow a Weibull model (Van Wagner 1978, Baker 1989a), which is the general case within which the negative exponential model exists. A Weibull distribution of stand ages has been noted for white spruce in Alaskan boreal forests (Yarie 1981), the sub-Arctic boreal forest (Johnson and Rowe 1977), the Upper Mackenzie Valley boreal forest (Rowe et al. 1975) and an eastern sub-boreal forest (Baker 1989a). It should be noted that increasing age selection of fire is only one underlying explanation associated with fitting a Weibull model. Johnson and Rowe (1977) proposed that a Weibull model may also indicate high levels of topographic complexity. Fire activity has been associated with other stand attributes. Bradley et al. (1992) and Johnson (1992) believe that fire susceptibility in pure conifer stands is density dependent, and can be very high during both the juvenile and overmature stages of growth. Cumming and Pelletier (1995) found a relationship between fire frequency and forest stand-type dominance in northern Alberta. Crowning potential has also been associated with drought potential (Turner and Romme 1994). 18 There is also evidence that a single landscape can be divided into sub-regimes according to permanent topographic features. Differential fire sizes have been associated with topographic complexity, and fire susceptibility has been correlated with topographic position (Clark 1990). Suffling et al. (1982) and Barrett and Arno (1991) describe distinct fire regimes for different geophysical areas of very large landscapes (i.e., upland versus lowland, or high elevation versus low). The possibility of multiple fire regimes complicates the application of either the negative exponential or Weibull model. The assumptions necessary to apply either model are that the area in question must be stable, or unchanging with respect to the area of forest and climatic conditions through space and time, and that the total area must be large relative to the largest possible disturbance event (Johnson and Van Wagner 1985). In boreal-type forests, single fires can easily exceed 100,000 hectares, meaning that the area necessary to consider the models must be much larger6. As the size of a landscape increases to accommodate these events, the probability of violating the assumption of unchanging conditions increases. It is difficult to imagine an area much larger than 1,000,000 hectares without at least a moderate change in climate or landform. Despite these shortcomings, the process of testing for negative exponential and Weibull models has proven informative. Among other things, the debate has contributed to the challenge of the "steady state" mosaic phenomenon. Theoretically, fire is a mechanism through which a stable 6 It is interesting to note that Baker (1989a) reworked the data Van Wagner (1978) first used to demonstrate the negative exponential model under a different set of assumptions, and came up with a Weibull model instead. In the discussion, Baker (1989a) points out that there were weaknesses in the arguments for either model, since the landscape in question was not large enough relative to the largest disturbance event. 19 age-class distribution across a landscape of some minimum size persists through time (Naveh 1987). Shugart and West (1981) attempted to quantify this area, hypothesizing that a stable mosaic should occur within areas 50 times the size of the average patch. The theory states that one would observe this stability i f it were possible to observe a sufficiently large landscape without the possibility of multiple regimes. However, not only did Baker (1989b) find no such equilibrium for the southern boreal forests, but he did not even find evidence that it was converging towards one well beyond the predicted threshold. There is general agreement elsewhere that the infrequent occurrence of very large fires in all northern temperate forests denies the existence of steady state landscapes {e.g. Romme 1982, Romme and Knight 1982, Baker 1989b, Turner and Dale 1991, Antonovski et al. 1992, Mladenoff et al. 1993, Payette 1993, Turner and Romme 1994, Cumming et al. 1996). 2.1.3 SMALL-SCALE DISTURBANCES In the absence of large-scale disturbance, small-scale disturbances such as windthrow and bark beetle attacks may also alter stand composition and structure in the SBSmkl. Windthrow in the Prince George Region is the lowest of all areas of B.C., accounting for only a fraction of the area affected by natural disturbance (Mitchell 1995). Windthrow is probably the most difficult disturbance to predict on this landscape, and even more difficult to identify retrospectively since effects are so variable (Everham 1995). Windthrow is thought to be spatially and temporally random and non-selective with respect to species or age on the SBSmkl landscape (D. Wilson 7, pers. comm.), although other studies have shown susceptibility to windthrow changes with both age and species dominance (Foster and Boose 1995). 7 Forest Ecosystem Specialist, BC Ministry of the Environment, Prince George District. 20 Bark beetles (Dendroctonus species) are perhaps the most damaging forest insect in B.C. . Two species of bark beetle are active on the SBSmkl landscape, the mountain pine beetle (Dendroctonus ponderosae Hopkins) and the spruce beetle (Dendroctonus rufipennis Hopkins). Overall susceptibility to mountain pine beetle in the Prince George area is moderate to high according to a susceptibility rating system developed by Shore and Safranyik (1992). Stand-specific susceptibility varies with the amount of pine, age, elevation, latitude, vigour, and stand density (Shore and Safranyik 1992). Attack patterns and preferences mean that mortality patterns from mountain pine beetle tend to be patchy (from a few trees to several hectares) and usually only between 10-30% of a stand is killed (Wood and Van Sickle 1991, Wood and Van Sickle 1993). The spruce beetle is similarly selective in its host and is most often associated with blowdown (Knight and Heikkenen 1980, Wood and Van Sickle 1991). The nature of bark beetle infestations and damage is such that the potential for beetle attacks causing species shifts across a landscape is considerable. On the other hand, it is difficult to imagine mountain pine beetle creating any of the larger patches that exist on the SBSmkl landscape. The association of spruce beetle with windthrow events allows for the possibility of large outbreaks following extensive windthrow damage, although these large patches would be limited to those areas where spruce occurred in almost pure stands. 2.1.4 ARBOREAL SUCCESSION Succession is the sequential change in composition and structure that takes place in biotic communities over time between disturbance events (Kimmins 1987). It has been hypothesized that boreal forests do not experience succession in the sense that single-stem mortality and species 21 replacement rarely occur. The argument is that fires replace stands long before tree replacement begins in most cases (Payette 1993). However, recent research in a neighbouring variant of the SBS concluded that successional pathways were active for those stands that survived beyond some critical age without being disturbed (Clark 1994). Parminter (1983) similarly predicted a gradual shift in boreal species composition of some stand-types in the absence of stand initiating events. These shifts are to some degree predictable given knowledge of shade tolerance, growth rates, and longevity of the species involved. A brief review of these attributes follows. With respect to shade tolerance, lodgepole pine, aspen, and cottonwood require moderate to full sunlight to establish and survive, and hence cannot regenerate under a closed canopy. However, all three experience very high rates of growth, and tend to dominate the stands they occur in from the start (Burns and Honkala 1990). They also tend to have the shortest lifespans. Aspen and cottonwood rarely live beyond 100 years (Bradley et al. 1992), and lodgepole pine is normally limited to 150-170 years (Burns and Honkala 1990), although field sampling in the SBSmkl revealed many specimens of both pine and poplar species in excess of 200 years. Since neither poplar species nor lodgepole pine is tolerant of shade, they are incapable of replacing themselves when fire intervals are greater than their lifespans, and will gradually be replaced with whatever understorey species are present under these circumstances (Payette 1993, Clark 1994). Black spruce, hybrid spruce, and subalpine fir are all shade tolerant, regenerate under closed canopies, and are capable of growing for extended periods of time in the sub-canopy (Pojar et al. 1984). These three species have slow to moderate growth rates, but the extended life span of hybrid spruce and subalpine fir (200-300 years) gives them a long-term advantage. Although black spruce does not live as long (approximately 200 years), it too can regenerate in the absence 22 of disturbance by layering (Burns and Honkala 1990). Douglas-fir is intermediate in terms of both tolerance and rate of growth, and is capable of regenerating underneath itself (Burns and Honkala 1990). In summary, disturbance will not allow time for species composition shifts to occur in most areas of boreal forests, and the initial stand composition will maintain itself until the stand is disturbed again. In those cases where disturbance intervals are prolonged, it is hypothesized that these forests would begin losing aspen and pine between 100 and 200 years of age, and succeeding to hybrid spruce, with a minor, but increasing component of subalpine fir and black spruce (Parminter 1983, Pojar et al. 1984, Clark 1994). 2.1.5 STAND INITIATION Despite the relatively low level of arboreal succession in boreal-type forests, and the low number of tree species involved, these landscapes demonstrate a rich mosaic of species combinations. The stand initiation process is thought to be largely responsible for this richness. The general occurrence of species on ecologically classified sites can be predicted in the SBSmkl. In fact, these relationships form the basis for the identification of site series for the Biogeoclimatic Ecosystem Classification system of B.C. (Pojar et al. 1984). For instance, pure, dense, lodgepole pine occurs on the warm, dry outwash sites, although black spruce can tolerate mixtures with the pine on these sites as well (Pojar et al. 1984). Lodgepole pine and hybrid spruce often overtop a thick understorey of subalpine fir on slightly moister sites, but hybrid spruce is also capable of pure stands on the richest sites (Burns and Honkala 1990). Cooler, 23 mesic sites are often occupied by mixtures of hybrid spruce and subalpine fir. When subalpine fir does make it to the canopy, it is most often on these sites (Burns and Honkala 1990). Draws and lowland areas support low density black spruce, and some poplar species and hybrid spruce (Pojar et al. 1984). Douglas-fir is near the northern limit of its range, and only occurs on warmer, richer, sites such as dry south-facing slopes (DeLong et al. 1993). Despite these generalizations, a single site is capable of supporting different combinations of species mixtures given the chance (Zasada et al. 1993). Soil/site studies attempting to correlate species with specific site conditions in more complex forest-types have rarely been able to account for more than 40% of the variation using site indices (McQuilkin 1976, Verbyla and Fisher 1989, Monserud et al. 1990). Predicting structural characteristics such as stocking, density, or productivity from site features has proven extremely difficult (Harding and Grigal 1986, Eremko 1990). Even in the simple SBSmkl forests, there is reason to believe that sub-boreal species/site relationships are stochastic (Parminter 1983). Even the sampling on which the BEC relationships were derived demonstrates a moderate degree of variation (C. DeLong, pers. comm.). Given that tree-replacing succession is not common, one of the opportunities for the creation of such variability is during stand initiation following disturbance. A closer look at the stand initiation process reveals a range of interacting processes. The success of species reaching, and then surviving, on a given site after disturbance depends on a number of factors. The time of year, type, size, and severity of the disturbance itself, and the species involved are the most important factors (Heinselman 1980, Zasada et al. 1993, Payette 1993). Timing will determine the amount of windborne seed immediately available and the 24 climatic conditions for survival and growth for the first season. The type, size, and severity of the disturbance will determine seedbed conditions, soil nutrient status, and availability of seed (Denslow 1985, Malanson 1985). For instance, large, severe fires will burn off most of the moss and litter layers, and leave very few remnant trees to distribute seed, yet even large fires rarely kill seed in all serotinous cones (Parminter 1983). Other factors determining the success of regeneration are air and soil temperatures during germination, the co-occurrence of seed years with the disturbance, and seed viability (Payette 1993). A l l tree species in the SBSmkl are fire-adapted to some degree, and this is reflected in the various methods of regeneration. Lodgepole pine with serotinous cones is the ultimate fire-adapted species, and is dependent on fires for opening cones and providing suitable seedbed conditions. Large areas of pure, even-aged, lodgepole pine are common in the SBSmkl (Parminter 1983). Black spruce is considered to be a semi-serotinous species, and is also capable of pure stands, although it commonly tolerates mixtures. Lodgepole pine and black spruce also have an advantage in that they produce cones early (10-15 years), often, and their seed stays viable in the cones for many years (up to 50 years for lodgepole pine) (Eremko 1990). On the other hand, serotinous species have great difficulty migrating because their seed cannot travel very far. However, both lodgepole pine and black spruce are capable of producing non-serotinous cones as well (Bradley et al. 1992). Hybrid spruce, Douglas-fir, and subalpine fir do not normally produce cones until 20-50 years of age, and although seed does not stay viable for long after dispersal, all species are relatively prolific and seeds are capable of travelling moderate distances (up to 100 m for spruce, and 200 m for Douglas-fir) (Burns and Honkala 1990, Bradley et al. 1992). Both cottonwood and aspen 25 can produce viable windborne seeds within 10 years of age that can travel great distances (Zasada et al. 1993). The timing of seed production generally varies within a single stand, although crop timing convergence has been noted for some species (Zasada et al. 1993). Alternative regeneration strategies are also evident. Black cottonwood and aspen are both capable of vegetative reproduction from surviving root stock, and black spruce can reproduce by layering (Fowells 1965). For those species that rely on seed dispersal, the dispersal must coincide with the availability of the proper seedbed to enable germination. Lodgepole pine, aspen, cottonwood, and hybrid spruce seeds require mineral soil for germination, and dominate severely disturbed sites when seed is available (Parminter 1983). Seeds of subalpine fir, and Douglas-fir are capable of germinating and surviving on moss layers and are more likely to invade moderate to lightly disturbed areas (Eremko 1990). Black spruce seed is capable of germinating and growing on both mineral soil and moss layers (Burns and Honkala 1990). The range of interacting influences on the post-disturbance stand composition in the sub-boreal forest is too complex to make generalizations. However, there is agreement elsewhere that the single most important factor for predicting post-burn vegetation in boreal-type forests is the stand composition before disturbance (Parminter 1983, Payette 1993). The tendency of species to be found on broad site-types is also useful, and has precedent in the gradient models of Kessell (1976) and Provincial site classification systems (Jones 1986, DeLong et al. 1993). Beyond this, considerably more information is required on the type, size, intensity, and timing of a particular disturbance, as well as species survival, age, and seed periodicity to make a significantly better prediction of post-fire succession. 26 2.2 HYPOTHESES AND QUESTIONS I will only focus on those aspects of the landscape level processes that are most likely to influence coarse-scale mosaic pattern. In other words, I am most interested in where, how big, what shape, and how often fires burn. I am also interested in whether or not fire is influenced by either (permanent) landscape features, or (temporary) vegetation attributes {i.e., age), as well as whether or not vegetation attributes can be predicted from permanent landscape features. Some of these questions can be answered descriptively from available data, while other questions require the testing of hypotheses. QUESTIONS: 1) What is the fire cycle, or the average number of years required to burn the total number of hectares in the SBSmkl landscape? 2) What is the average, and distribution of, the fire disturbance rate? 3) What is the distribution of fire disturbance sizes? 4) What is the distribution of fire disturbance shapes relative to disturbance sizes? H Y P O T H E S E S : 1) The tendency of younger stands to burn is no greater than that of older ones. 2) The SBSmkl contains only a single fire regime; sub-regimes differentiated by topographic or pedologic factors not evident. 3) Fires do not respond to permanent landscape features. 4) Species, or species combinations are not associated with specific soil or site-types. 27 2.3 DATA Two spatially-explicit datasets were used to test the hypotheses and answer the questions posed above. The first was an aggregated, digitized, raster-based mapfile of MoF forest inventory mapsheets for the entire area of the SBSmkl. This mapfile, covering approximately 790,000 ha, was compiled in its original polygon format by Timberline Forest Consultants8 contracted by the Prince George Regional MoF Research unit using the ARC/INFO geographic information system (G1S) software package. Timberline created an ASCII file listing an identification number, patch type (forest, water, opening etc.), area (m2), perimeter (m), and age-class for each polygon. I obtained this file with permission of the Regional MoF research personnel. With respect to the data representing a "natural" landscape, two problems emerged: the prevalence of timber harvesting, and the effect of fire control activities. The existence of cutblocks throughout the area was addressed (by Timberline) by re-classifying all forest openings to their probable current 20-year age-class using an algorithm based on the assumption that the age-class of the harvested patch is that of the most common adjacent age-class. More recent cutblocks could be checked against historical inventories. In most cases, the adjacent age-class was age-class 8 (140-250 years). The effect of fire control activities was addressed by rolling back stand ages. According to local MoF sources, intensive fire control activities in the SBSmkl have been active for the past 35-40 years (C. DeLong, pers. comm.), during which time very little of the forest burnt. The ASCII mapfile revealed that only 1% of the SBSmkl was 40 years of age or less (7,491 hectares). By Prince George, BC. 28 ignoring this 1%, the entire age-class structure was rolled back 40 years to represent the natural age structure of the SBSmkl landscape in 1954. It is not important that the age-class structure used to estimate the disturbance regime metrics was not that of 1994. The final step necessary to prepare this dataset for analysis was to estimate the age-class breakdown of mature and over-mature stands. The MoF forest inventory used 20-year age-classes up to 140 years. The next age-class (age-class 8) covered a 110 year range from 141 to 250 years, and age-class 9 included everything older than 250 years. The data from these two age-classes were not suitable for defining the disturbance regime, and were redefined into 20-year intervals. This required field sampling. The details of this study are given in Appendix A. Soils information was added to approximately 180,000 hectares of the SBSmkl using 1:50,000 maps available from B.C. Ministry of the Environment (MoE) (Dawson 1989). The information on these maps included the landform type (till, lacustrine, organic, fluvial, or a mixture), topographic complexity (flat to very complex), and drainage (rapid, well, moderately well, and imperfect). These data were overlain on the relevant portion of a hard copy of the 790,000 hectare Timberline map. Dot counts were used to calculate the area of each age-class in each of eight soil-type classes. The second dataset was a detailed spatial file compiled for two adjacent forest inventory mapsheets (28,730 hectares) situated within the 790,000 hectare SBSmkl. This contained three overlays: forest inventory, soils, and a digital terrain model (DTM). The forest inventory data originated from two digital GIS files of 1:20,000 MoF inventory mapsheets (93J025 and 93J035). The two mapfiles were purchased through MoF Inventory Branch, and imported into the GIS 29 software package P A M A P (PAMAP 1989). Inventory mapsheets from 1972 were then used to manually digitize the original stand boundaries in areas where recent cutblocks occurred. Where older cutblocks occurred, data from field sampling (of stumps) or historical timber cruises were used to assign an age-class. Where neither of these data were available, it was assumed that the original age of a cutblock was age-class 8. The two forest inventory files were then merged and converted to raster format using 50 m, or 1/4 hectare pixels. A considerable amount of data were available in the inventory mapsheet files. The information carried through to the forest file was limited to the coordinates, percent of each species, age-class, the presence of creeks, and an identification code i f the pixel was not forest cover. The soils overlay for the area of the two inventory mapsheets was created from the same MoE 1:50,000 soils maps mentioned earlier (Dawson 1989). The boundaries of individual soil polygons were manually digitized as a layer on the forest inventory mapsheets to allow the use of water bodies and creeks as landmarks for digitizing. The files were then merged and rasterized at 50 m resolution similar to the forest cover data. The third and final overlay was a 10 m resolution digital terrain model obtained from Lakeland Mills Ltd (Prince George, B.C.), who required it for operational purposes. The 10 m contours were converted to slope and aspect through a P A M A P utility and then added to the other two overlays. The D T M coverage was incomplete (about 85% of the two mapsheets). For those pixels without D T M information, slope and aspect were flagged and set to zero. 30 The three completed overlays were then exported to ASCII files and merged. After verification and editing, the data were converted to binary format. The final spatial database contained 260 x 422 = 114,920 quarter-hectare pixels. 2.4 METHODS AND RESULTS This section includes the methods and results for each of the four questions and four hypotheses concerning the landscape processes of the SBSmkl. The assumption I made for this analysis was that the landscape pattern, according to the age-class mosaic, was exclusively a result of forest fires. Stand replacement events in the SBSmkl include fire, windthrow, and bark beetle infestations, although fire is normally associated with the later two (Bradley et al. 1992), and both windthrow and bark beetle tend to operate at smaller scales. 2.4.1 AGE INVARIANCE HYPOTHESIS 1: The tendency of young stands to burn is no greater than that of older ones. Actually, this hypothesis is part of a larger question concerning the appropriateness of various models in describing fire dynamics on boreal-type landscapes. ANALYSIS Under the assumption that fire is age-invariant (i.e., non-selective with respect to stand age), the SBSmkl time-since-last-fire distribution should approximate a negative exponential function and the average age will equal the fire cycle (Van Wagner 1978). This means that under ideal conditions, approximately 36.8% of the stands will be older than the average age, given that the opportunity for fire occurring on any one area fits a Poisson distribution. If, on the other hand, 31 fire is age-selective, the SBSmkl age-class distribution should resemble a Weibull function, and the fire cycle must be estimated using iterative techniques (Johnson and Van Wagner 1985). Note that the Weibull model reduces to a negative exponential model when the shape parameter is one. Both hypotheses could be tested by comparing the theoretical time-since-last-fire distributions against the actual data using a Kolmogorov-Smirnov (K-S) one sample test (Steel and Torrie 1980). If fire is age-invariant, the negative exponential model for the SBSmkl takes the following form: t exp b where (y) = probability of surviving to age (t) in years, b = the fire cycle (81 years) and t = time (years). In the negative exponential scenario, the average age is the fire cycle, and the inverse of the average age is the average annual disturbance rate. The average age of the SBSmkl yielded an annual disturbance rate of 1.23% per year. In the Weibull model, fire is assumed to be age-selective, becoming increasingly probable in older stands (Baker 1989a). The scale (b) and shape (c) parameters for the Weibull model were estimated using two different techniques, the first a Maximum Likelihood Estimator (MLE) by Harter and Moore (1965) and the second, a regression estimator by Menon (1963) (from Johnson and Van Wagner 1985). The Weibull model for the SBSmkl was: exp 00 b where (y) is the probability of surviving to age (t), scale parameter b = 104 (MLE) and 93 (regression), shape parameter c = 2.9 (MLE) and 1.7 (regression), and t = time (yrs. before fire). 32 Comparing the three models against the actual data using the K-S one sample test was inconclusive. Maximum difference (D) values between observed and hypothesized values for the proportions in each age-class for the negative exponential, the M L E Weibull, and the regression Weibull models were 0.245, 0.236, and 0.211 respectively. Maximum difference values required to reject the null hypothesis were 0.405 for an alpha of 0.05, and 0.369 for an alpha value of 0.1. Since none of the maximum D values exceeded these limits, the null hypothesis could not be rejected for any of the comparisons. In other words, the raw data were not significantly different from any of the three curves tested. However, keep in mind that the K-S test is conservative in that the probability of rejecting the null hypothesis (that the distributions are not significantly different from each other) when it is actually true, is less than the alpha value (Steel and Torrie 1980). The negative exponential and the two Weibull models are plotted against the actual distributions in Figure 2.1 as the probability or frequency (y) of a stand surviving (x) years of age before being disturbed. Visually, it can be argued that the shape of Menon's (1963) Weibull cumulative curve (Weibull B) was more suggestive of the actual cumulative distribution, and the negative exponential model the least appropriate. This is due to the fact that there were few or no very old stands. However, this lack of older stands may be a temporal anomaly. The "temporal artifact" argument is common in landscape studies, and is difficult to refute. However, in this case it is possible to create a test for the SBSmkl using a computer model capable of simulating disturbances according to the assumptions governing the negative exponential model, and then comparing the age-class distributions created with the existing distribution. The hypothesis in this case is that it is statistically possible to observe the current 33 age-class distribution, given the assumptions of the negative exponential model and the historical range of fire frequencies. More specifically, is the lack of over-mature forest we observe a temporal anomaly, or in fact a highly improbable configuration of this landscape historically? 1 0 0 8 0 > DC ZD CO u_ O >-O a LLI DC 6 0 4 0 2 0 0 \ \ \ \ \ \ W \ -, \ \ \ \ \ i < i i »«« 0 2 0 4 0 6 0 8 0 1 0 0 1 2 0 1 4 0 1 6 0 1 8 0 2 0 0 2 2 0 2 4 0 2 6 0 2 8 0 3 0 0 A G E (yrs) I A C T U A L N E G . E X P O N E N T I A L - - - - - W E I B U L L A — — W E I B U L L B Figure 2.1. SBSmkl time-since-last-fire distribution. Comparison of actual distribution to Harter and Moore's (1965) maximum likelihood model (Weibull A) , Menon's (1963) regression estimator Weibull model (Weibull B), and the negative exponential model. A full description of the model is discussed in Chapter 3 and will not be given here. Suffice it to say that the model includes functions for describing the range of both fire frequencies and sizes derived from historical data. However, the model as described in Chapter 3 was altered in one way for this test: the total area of the landscape on which disturbances "burned" was expanded to represent the entire 790,000 hectares from the original 28,730 hectares. Otherwise, the 34 historical disturbance frequencies were estimated according to negative exponential assumptions of equal probability of burning and homogeneously Poisson distributed fire starts. The model was run for 130 20-year periods. The first 10 periods, or 200 years, were removed from the data in an effort to eliminate any bias associated with any one particular starting point in terms of landscape pattern. The average percentages for each age-class over the remaining 120 periods created a perfect negative exponential curve virtually identical to the negative model curve shown in Figure 2.1 (as it should). Of the 120 landscapes created in simulation based on historical data and negative exponential assumptions, none had zero area in stands greater than 200 years. In fact, nothing less than 1% of the landscape was ever left in stands greater than 200 years, and only seven times was there less than 2% of the landscape in stands of this age. The 95% confidence intervals confirm that the simulation model has virtually no chance of ever creating a landscape without stands greater than 200 years. In other words, this test did not support the hypothesis that the current age-class distribution, with its lack of older stands, is a temporal artifact. DISCUSSION The reasons for suspecting an increasing tendency of burning with age have been discussed at length by others (e.g. Mutch 1970, Heinselman 1980). There are also precedents that consider both the negative exponential and Weibull models, and found shape parameters similar to those found for the SBSmkl (e.g. Johnson and Rowe 1977, Yarie 1981, Baker 1989a). These precedents aside, since the Weibull model reduces to the negative exponential model under age invariance, assuming the Weibull model parameters by default seems a reasonable test in and of 35 itself since any shape parameter greater than 1 indicates age selection. In this case the shape parameter (c) jumped from 1.0 (for the negative exponential model) to 1.7 or 2.9 depending on the method of estimation. In light of the corroborative evidence from the simulation test results, it seems reasonable that age selection may be the cause. On the other hand, there is evidence against accepting an age-dependent burning model in the SBSmkl. During the process of field sampling, evidence of fire scars on trees less than 100 years of age was common, indicating that younger stands readily burnt. In fact, almost 20% of the field plots showed some evidence of multiple fires within 100 years (although it was not determined whether the fires were surface or crown). Furthermore, other than for Jack pine (Pinus banksiana Lamb.), the Canadian Forest Fire Behaviour Prediction System does not predict an increase in rate of spread (ROS) for any of the boreal fuel-types with increasing age (Forestry Canada Fire Danger Group 1992). In contrast, a life table for the SBSmkl forests indicated an increase in disturbance rates with age for both Weibull models derived (Table 2.1). The differences in the numbers of survivors was particularly high between the negative exponential model and Harter and Moore's (1965) Weibull (Table 2.1). In other words, the Weibull models predict dramatically increased susceptibility with age, contrary to empirical fire behaviour research. Finally, the simulation test results could be due to the adoption of other assumptions such as stable climate, non-serially correlated frequencies, or the method of estimating disturbance frequencies. I also could have simply missed finding the oldest stands on the landscape during field sampling. It is difficult to fully accept the age selection theory without reservations under the circumstances. 36 Table 2.1. Life table for SBSmkl stands by age-class for the negative exponential model, and Harter and Moore's (1965) and Menon's (1963) Weibull models showing the number of survivors through time based on initial values of 1,000 stems. AGE-CLASS (yrs). RATE OF DISTURBANCE (% per year) # SURVIVORS NEG. EXP. H & M W. MENON W. NEG. EXP. H & M W. MENON W. 0-20 1.23 0.03 0.38 1,000 1,000 1,000 21-40 1.23 0.26 0.83 771 990 920 41-60 1.23 0.69 1.18 602 933 775 61-80 1.23 1.31 1.50 470 804 608 81-100 1.23 2.12 1.79 369 609 448 101-120 1.23 3.10 2.06 287 391 311 121-140 1.23 4.26 2.31 224 204 204 141-160 1.23 5.59 2.55 175 84 126 161-180 1.23 7.09 2.79 137 26 76 181-200 1.23 8.76 3.01 107 6 43 201-220 1.23 10.6 3.23 83 1 23 221-240 1.23 12.6 3.44 65 0 12 241-260 1.23 14.76 3.65 51 0 6 261-280 1.23 17.08 3.85 40 0 3 281-300 1.23 19.56 4.05 31 0 1 It could be argued that neither the negative exponential nor the Weibull model satisfactorily described the SBSmkl age-class structure. In practical terms, neither model seemed to sufficiently explain what was observed; namely that young stands are capable of burning, and that very few very old stands were found. One possible explanation for this is that fire susceptibility is bimodal, where susceptibility peaks just after crown closure, and then again in over-mature stands (Johnson 1992). Another theory is that the SBSmkl landscape may have more than one disturbance regime. Mixed fire regimes have been noted for several coniferous forests (Suffling 37 et al. 1982, Barrett and Arno 1991, Turner and Romme 1994). Section 2.4.4 will explore this possibility further. Finally, although I have assumed that fire was the only disturbance agent creating the observed age-class mosaic, it may be that windthrow and/or bark beetle outbreaks are responsible for the lack of older forests and the Weibull tendencies. Susceptibility to mountain pine beetle increases dramatically with age and tree size (Shore and Safranyik 1992). One could also argue against either model on theoretical grounds since they both depict stable populations, which the SBSmkl is clearly not. As discussed in Section 2.1, one of the assumptions in testing the negative exponential or Weibull model is that the size of the population must be large in comparison to the largest disturbance event. Heinselman (1973), Turner and Romme (1994), and Johnson (1992) agree that the majority of the area burnt on any given landscape occurs because of extreme 50 to 100 year climatic events. When such events occur, fire behaviour becomes a function of fire weather conditions rather than fuel-type, stand age, or topography (Bessie and Johnson 1995). The bimodal nature of the majority of boreal age-class distributions may be evidence in support of this phenomenon (Romme and Knight 1982, Baker 1989b, Turner and Dale 1991, Antonovski et al. 1992, Mladenoff et al. 1993, Payette 1993, Turner and Romme 1994). The SBSmkl was similarly bimodal. The two largest age-classes (burnt between 1915 - 1935, and 1835 - 1855) accounted for almost 42% of the landscape. The amount of area originally covered by these older burns was more than likely greater than 50% of the landscape, indicating that the SBSmkl landscape is not large enough to experience "stability". However, this is as stable as any landscape in B.C. is likely to get with respect to age-class distribution. At 790,000 hectares, the SBSmkl is one of the largest variants of the B E C system. Expanding the total area to try to reduce the effects of single large fires was not an option. The SBSmkl is bordered on all sides by different forest-types and different disturbance 38 regimes. As a result, beyond the 790,000 hectares of the SBSmkl in any direction, the age-class distribution would most likely become less stable, rather than more. Bimodal age-class distributions may indicate that fire on these landscapes may have two behaviourial patterns. The majority of fires respond to fuels, topography, and stand age creating small to medium sized disturbances. During very short, infrequent periods, fire behaviour is almost solely controlled by fire weather creating the very largest disturbances (although portions of these larger fires will respond to fuels, topography, and stand age as well). A model which considers only one of these phenomena, as the negative exponential and Weibull both do, will not capture this type of behaviour. Furthermore, since most of the forest pattern resulted from large fires, there is little physical evidence to suggest that fuels, topography, and stand age are related to fire activity. The evidence suggests that a limited degree of age selection is operating on the SBSmkl landscape. More specifically, very old stands become highly susceptible to disturbance. Whether or not the main mechanism of disturbance is fire, bark beetle, or simply stand break-up is unknown. Of the three models tested, the regression estimator Weibull method of Menon (1963) was marginally better at representing the fire dynamics in the SBSmkl, but the entire genre of Weibull-type models share the problems of being stable, and representing only one phenomenon at one scale of activity. None of them are entirely appropriate'for describing SBSmkl landscape disturbance dynamics. 39 2.4.2 FIRE CYCLES Question 1: What is the fire cycle? ANALYSIS The methods for estimating the fire cycle have been described above in Section 2.4.1. According to the negative exponential model, the fire cycle was 81 years. According to the M L E and regression estimator methods of the Weibull model, the fire cycle was 93 years and 104 years, respectively. DISCUSSION Fire cycles have always been reported as single numbers, or averages. In this case, using a single number to describe the SBSmkl fire cycle would be misleading since it has already been concluded that the age-class distribution is unstable temporally. This means that average age is an unreliable estimate of the fire cycle since it will be equally unstable. For instance, the estimated average age of the current forest age-class structure using the negative exponential model (81 years) is probably lower than the actual average fire cycle because over half of the area of the landscape was older than 80 years. The fact that a limited degree of age selection is suspected means that the estimates of fire cycle from the two Weibull models, both higher than 81 years, should be considered. The estimated fire cycle from the regression estimator Weibull model of 93 years is most likely closer to an "average" fire cycle. Recognizing a range between 80 and 100 years is a more realistic way of describing the SBSmkl fire cycle. 40 2.4.3 DISTURBANCE RATES QUESTION 2: What is the average and distribution of the disturbance rate? ANALYSIS A "disturbance rate" is the percent area burnt per unit time, usually annually. Disturbance rate is by definition the reciprocal of the fire cycle (area burnt/year = 100/fire cycle) (Johnson 1992), which worked out to 1.23% per year using the average age from the negative exponential model (100/81). However, as with the fire cycle, it would be more appropriate to describe this parameter as a distribution. I lacked the necessary detailed historical fire data to estimate actual, natural annual rates. However, it was possible to estimate 20-year disturbance rates from the large dataset rolled back to 1954 (which accounts for fire suppression efforts). The rate of burn for the last 20 year period was estimated directly from the amount of area in the 0-20 age-class (7.4%). However, each successively older age-class increasingly underestimated the burnt area because some of the originally disturbed area was likely reburnt. Although it was not possible to calculate exactly how much area was reburnt, it could be estimated by rolling back the age-classes, and distributing the area in the most recent age-class in proportion to the abundance of the remaining age-classes. When rolling back the age-classes, the area that existed in age-classes that no longer existed (i.e., too old) had to be accounted for. These "phantom" age-classes are so old, they would have been completely overlain by one or more recent burns. Without including these phantom age-classes, the estimates of areas burnt would have been increasingly underestimated as the roll-back 41 proceeded back through time. However, the problem of estimating what might have been in these phantom age-classes is circular; it was necessary to know what the rate of burning was to make a guess at what might have been there, but it is these burning rates that are sought. Since there was no way to be sure of exactly how much area was originally in each of these age-classes, the best that could be done was to use the fire cycle averages. Two scenarios based on 80 and 100 year fire cycles are presented in Table 2.2. The only difference between the two scenarios is the amount of area allowed into hypothesized older age-classes. The shorter fire cycle had less area in the older stands than the longer one, meaning there was a greater proportion of the area that was redistributed into older age-classes for the landscape at each 20 year time-step. Shorter fire cycles imply higher rates of burn. As expected, there was an increase in the difference between the fire cycle estimates with increasing age-class. The second youngest age-class ranged from only 23.6 to 23.9%, but the seventh estimate, for the 121-140 year age-class disturbance rate, ranged from 41.2 to 36.0% (Table 2.2). Obviously, the degree of confidence in the estimates of burning rates declined with increasing age. For this reason, only the first seven estimates were considered. According to the roll-back, between 23.3 and 22.2% of the SBSmkl was consumed by fire every 20 years9. Of more interest is the variation of disturbance rates, which ranged from 7.4% to at least 40% per 20-year period. The standard deviation for the 80 year cycle was 13.2, meaning that the disturbance rate was lower than 10.1% or higher than 36.5% in one out of three 20-year periods according to this estimation method. These figures cannot be divided by 20 to give annual disturbance rates directly. We expect annual rates to be higher than (20-year rate/20) since some areas may burn more than once in a 20-year interval. 42 Table 2.2. Estimated disturbance rates for each 20-year period for the SBSmkl based on rolling back age-classes. Two projections were calculated based on an 80 and a 100 year fire cycle. Age-Class Existing area (%) Projected area after rollback (%) 80 year cycle 100 year cycle 1-20 7.4 7.4 7.4 21-40 22.2 23.9 23.6 41-60 7.2 10.0 9.8 61-80 9.8 14.8 14.4 81-100 13.0 22.9 21.9 101-120 19.4 43.2 40.6 121-140 10.9 41.2 36.0 141-160 4.0 - -161-180 5.6 - -181-200 0.4 - -M E A N (first 7) 23.3 22.2 S.D.(first 7) 13.2 11.7 There is a way of checking these estimates of disturbance frequency for internal consistency using the computer model, similar to the test used in Section 2.4.1. In particular, the relatively large amount of area in forest older than 80 years of age in the current age-class distribution provides a 'target' by which to compare simulation results. The hypothesis is that it is reasonable to "observe" the current age-class distribution (specifically the large amount of area greater than 80 years old) using a simulation model incorporating the rates of disturbance estimated above. As in Section 2.4.1, the model, fully described in Chapter 3, was altered to represent the entire 43 790,000 hectare landscape from the original extent of 28,730 hectares. The model assumes spatially random fire starts, and uses an historically derived size class distribution and the disturbance frequencies from the 80 year roll-back estimates given in Table 2.2. The model was run for 130 20-year periods, and once again the first 10 periods were eliminated from the data to eliminate any bias associated with any one particular starting point. The amount of area in each landscape snapshot in forest older than 80 years and older than 100 years was tallied and compared to the age-class distribution of the original landscape. The results showed that the chances of observing the current age-class distribution using the disturbance frequencies from the 80-year model in Table 2.2 was marginal, but not impossible statistically. Compared to the actual amount of area in forest greater than 80 years (53%), the model averaged only 29.6%. The 95% confidence interval from the 120 periods extend from 7.9 to 51.4%, meaning that there is less than a 5% chance of finding greater than 53% of the landscape older than 80 years of age. The chances of observing more than 40% of the area in stands older than 100 years of age is much the same as for the 80 year plus area. The 95% confidence interval for area in stands 100 years plus is 3.8 to 39.8%. Overall, although it is possible to find the current (1954) amount of older forest on the SBSmkl landscape, the chances are remote. DISCUSSION The simulation exercise indicated that there is evidence to suggest that actual disturbance rates may range much higher than those estimated in Table 2.2. Considering this, it would not be unrealistic to consider historical 20-year disturbance rates in excess of 50 percent. However, it was possible to observe the current amount of older forest in the simulation, and in any case the 44 simulation test can only be considered a rough guide. There could have been other factors which affected the 1954 age-class distribution creating an unusually large amount of older forest. For instance, age or species selection may have played a factor in what or where fires burnt. Climatic patterns over the past 200 years may have influenced the fire regime. It is also possible that frequency rates are accurate as estimated, but serially correlated through time such that very high rates of burning do not occur near to each other temporally. Generally, the variability of the disturbance rates estimated for the SBSmkl is typical of fire dominated landscapes (Turner and Romme 1994). It is difficult to conclude what the ecological role of this variability might be, but based on our knowledge of the importance of the stand initiation process in boreal-type forests, it may provide an opportunity for greater levels of beta-level (between stand) diversity. This may be a question worth more consideration in the future since present management strategies greatly reduce this variability. Forest management activities will also potentially affect the average disturbance rate since harvesting activities (the disturbance) will be limited to older stands. To demonstrate, Figure 2.2 compares the actual age-class distribution of the SBSmkl with that of a hypothetical "regulated" forest with a harvesting rotation age of 100 years10. Despite the obvious differences caused by the variability of the natural disturbance rates, these two scenarios have almost identical rates of disturbance on average. For a regulated forest with a fire cycle of 90 years, exactly 1.1% of the forest is disturbed each year, the same as the natural disturbance regime on average. However, using an identical harvesting cycle of 90 years eliminates older stands because harvesting is Rotation ages have been based on the fire cycle, or fire return interval estimates in some boreal-type forests. 45 limited to mature stands, while fires can and do occur in stands of any age. In this particular landscape snapshot, 53% of all stands were greater than 80 years of age, and about 40% were greater than 100 years. So even though the average natural fire cycle was 80-100 years, there were areas which survived much longer. This is a by-product of the (Poisson) distribution of fire frequency historically observed spatially within fire dominated landscapes (Green 1989). 25 n 10 30 50 70 90 110 130 150 170 190 210 AGE-CLASS MID-POINT • ACTUAL _ ] VARIABLE ROTATION • 100 YR ROTATION Figure 2.2. Comparison of the actual SBSmkl age-class distribution with two hypothetical models. Increasing the rotation age would help by allowing older stands to exist, but it also reduces the rate of disturbance. For instance, a rotation age of 90, as mentioned means an average disturbance rate of about 1.1% per year. A 120 year rotation designed to leave a greater proportion of older stands, would disturb only 0.83%o of the landscape per year. 46 The third management scenario given in Figure 2.2 directed harvesting to take place in a range of stand ages above 120, allowing older stands to exist in declining amounts. These types of management scenarios are being considered by forest management agencies because they allow a range of older age-classes to exist (Andison et al. 1995). However, this strategy decreased the landscape disturbance rate even further. In this particular scenario, the rate of disturbance was only 0.65%, well below the estimated natural rate. The impact of reduced rates of disturbance on landscape scales is unknown. However, based on the known association between disturbance and nutrient cycling (Kimmins 1987), one possible effect of reduced disturbance rates is a decline in nutrient cycling rates on the landscape. It is also possible that biomass accumulation will increase, resulting in higher risk of fire, but this will be offset to an unknown degree by the removal of biomass during harvesting. In summary, no single number describes the disturbance rate on the SBSmkl landscape. On average, between 10% and 35% of the landscape is consumed by fire every 20 years, but there is evidence to suggest that the maximum may be much higher. The importance of the temporal variability in disturbance rates, combined with the relatively non-selective nature of fire spatially, is unknown, but the imposition of management restrictions on this variability will potentially either eliminate the opportunity for over-mature stands to exist, or decrease the "natural" landscape disturbance rate. 47 2.4.4 MULTIPLE FIRE REGIMES HYPOTHESIS 2: The SBSmkl contains only a single fire regime, and sub-regimes differentiated by topographic or pedologic differences are not evident. Over a 790,000 hectare forest landscape, it is conceivable that more than one fire regime is distinguishable. As discussed in Section 2.4.1, this may be one of the reasons for not observing a stable age-class distribution. The factors commonly used to differentiate different regimes are topographic complexity, soil material, aspect, forest-type, or changing climatic factors (Barrett and Arno 1991, Johnson 1992). ANALYSIS The data limited the number of tests possible. Forest cover-type was not available for the entire landscape and thus could not be tested. Soil and topography data were not available for the entire SBSmkl area, but it was possible to use the 1:50,000 soil-type maps for the southern-most 180,000 hectares, or almost 23% of the SBSmkl landscape. The age-class data were first summarized by the original soil-types to see i f any noticeable trends in average ages were evident (Table 2.3). Expectations were that dry soils are more likely to have the more flammable lodgepole pine and optimal fine-fuel drying and burning conditions, and that flatter topography offers less interruption to fire spread. Thus areas described as dry and flat should have low average ages (fire cycle approximations). Using the same logic, poorly drained soils, and complex topography should have the highest average ages. As noted in Table 2.3, there seemed to be at least two 48 notable exceptions to this hypothesis. The first was the exceptionally long fire cycle on well drained drumlin basal till areas (97.4 years). This was particularly perplexing given that this soil-type is most closely related to the moister basal till eskers, which had a fire cycle of almost 17 years less (Table 2.3). Both soil-types covered extensive uninterrupted areas of the SBSmkl plateau, had an otherwise similar description in terms of steepness and topographic complexity on the MoE soil maps, and often occurred adjacent to each other. Without more detailed information (such as forest cover type) it was impossible to consider testing any other factors which may have been influencing fire susceptibility. Table 2.3. Area and average age by soil-type for the SBSmkl. SOIL-TYPE AREA (ha) % AREA AVG. AGE(yrs) rapidly drained steep tills 8,873 4.5 43.0 rapidly drained eskers 2,969 1.5 53.4 imperfectly drained flat organics 10,004 5.0 78.7 moderately well drained basal till eskers 61,587 31.1 80.5 imperfectly drained fluvial floodplains 6,610 3.3 81.5 moderately well drained lacustrines 11,340 5.7 82.8 imperfectly drained rolling lacustrines 7,819 4.0 84.7 well drained drumlin basal tills 88,919 44.9 97.4 The second notable exception to the relationship between the estimated fire cycle and soil drainage and topographic complexity was with the organic soils. Of all of the soil-types, the expectations were that the wettest would experience fires least often, creating relatively long fire cycle estimates. However the imperfectly drained organic areas had a fire cycle of only 78.8 49 years, close to the landscape average, and were exceeded by much drier soil-types (Table 2.3). There was some concern that the eight original soil-types were inadequately represented. To address this, the data were grouped into classes of soil drainage and topographic complexity, from which age-class distributions and average ages were again compared. The first classification created three levels of soil drainage: rapid, well to moderate, and imperfect. The second classification identified two levels of topographic complexity: flat and complex (Figure 2.3). 3 5 ^ 3 0 -25 H . A G E - C L A S S MID-POINT Legend • RAPIDLY DRAINED [TQ WELL-MOD. DRAINED • IMPERFECTLY DRAINED Figure 2.3. SBSmkl age-class distribution by soil drainage class. 50 Well to moderate, and imperfectly drained soils showed very similar distributions (Figure 2.3). However, on rapidly drained soils, the average fire cycle was over 30 years less than found on areas of either the well to moderately, or imperfectly drained soils (49 years versus 83 and 80 years respectively), with a correspondingly negatively skewed age-class distribution. Classifying by topographic complexity provided no indication of different average fire cycles (both were 81 years) (Figure 2.4). 30 n 2 5 -10 30 50 70 90 110 130 150 170 190 A G E - C L A S S MID-POINT • FINE | TILL Figure 2.4. SBSmkl age-class distribution by topographic complexity. 51 DISCUSSION Several explanations are possible for the lack of significantly older stands on organic soils. The simplest is that the almost pure lowland black spruce stand-types are particularly susceptible to fire. Although the FBP prediction system does not yet deal with lowland black spruce, this scenario has been suggested by fire behaviour researchers (B. Hawkes", pers. comm.). Another possibility is that the small pockets of organic soils tend to be adjacent to highly susceptible stands, and fires tend to spread into them. This hypothesis cannot be tested without cover-type information. Finally, it may simply be an artifact of large fire behaviour, which tends to ignore fuel-types. Lacking the ability to do more extensive analysis on disturbance sub-regimes, one should be cautious in drawing conclusions from these results. Keep in mind that strong evidence was found that the disturbance regime of the SBSmkl was far from stable. Not only did this analysis provide no conclusive evidence that the reason for the instability was the existence of sub-regimes, but most of the age-class distributions for the soil-type or topographic complexity groupings were quite similar, meaning they were just as unstable as the overall age-class distribution. In addition, large fires, which tended to dominate the area, are far less differentiating in what, or where, they burn (Turner and Romme 1994). One only has to look at the results in Table 2.3 to be suspicious of generalizations about fire regime differentiation. Based on soil and topography, the only differentiation that may be legitimate with this limited data set is that rapidly drained sites had a shorter fire return interval than the rest of the landscape. Fire Research, Pacific Forestry Centre, Forestry Canada, Victoria, BC 52 2.4.5 FINE SCALE TOPOGRAPHIC CONTROLS HYPOTHESIS 3: Fires do not respond to permanent landscape features. A series of tests were conducted using the smaller spatial datafile to try to differentiate if, and/or where and when, individual fire behaviour is non-random. These questions are obviously related to those from the previous section. However, Section 2.4.4 was concerned with macro-differences in landscape fire behaviour; for instance, are there areas of several hundreds or thousands of hectares which display significantly different disturbance regime characteristics from others? This section is more concerned with inferring the mechanics of individual fire behaviour (/. e. given that fire starts are random, do fires tend to burn either through or around certain fine scale landscape features, or is fire completely non-selective?) C R E E K S HYPOTHESIS 3a: Fires do not respond to creeks. More precisely, I wish to know whether or not fire edges tend to form at creeks. A N A L Y S I S Detailed edge information was available for the small, 28,730 hectare spatial database such that edge location could be associated with the location of other features. A binomial test of observed versus expected occurrence of edges with the topographic feature was employed. Testing was limited to the most common continuous permanent landscape feature, creeks. Lakes were not used for the analysis because it was not possible to say whether the fire actually jumped a given lake, or simply found a path around it. Rivers were also eliminated from the analysis because their inclusion would have introduced a potential bias in that they may be more than one pixel 53 (50 m) wide in some areas. Limiting the analysis to the smallest class of creeks eliminated both of these problems. It also allowed for potentially stronger conclusions, i f a positive correlation between edges and creeks is found, it likely means that fires are affected by larger, wider fire breaks such as lakes and rivers as well. Using the small (28,730 hectare) spatial datafile, the results were as follows: Total number of forest pixels 103,685 Total number of forest edge pixels 8,882 Total number of eligible creek/river pixels 15,755 No. of forest edge pixels within 1 pixel of a river/creek expected randomly 1,218 No. of forest edge pixels within 1 pixel of a river/creek observed 1,753 The test allowed for a creek to be found on either side of the two pixels that form an edge, so the number of possible pixels in which a creek may be found was double the number of creek pixels, minus the number that occur on the map edge, giving the total of 15,755 shown above. Using a binomial test, the resulting z-value worked out to 15.4 (using a population mean of 1,218 and a standard deviation of 35) which meant that the probability of observing 1,753 creek pixels intersecting forest edges randomly was extremely low (virtually zero). DISCUSSION No references were found in the literature related to these findings. However, the idea of creeks and other water bodies influencing edge formation is consistent with what is know about fire behaviour with respect to fuel continuity and pre-heating requirements (Fuller 1991) assuming creeks generally provide a fuel-break, and are associated with fuels with higher than average moisture content. 54 Although the binomial test was conclusive statistically, clearly the vast majority of creeks had little or no influence on edge formation. It is not difficult to find examples of fires which have easily jumped fuel breaks many times larger than small creeks. This is perhaps a indication of the range of intensity with which fires burn spatially and temporally. For instance, even very large crown fires typically drop down to the ground overnight as cooler, more humid, windless conditions prevail. Similarly, fire flanks burn with far less intensity than the head (Johnson 1992). A fire edge will form at different times and places depending on the burning conditions at that particular time and place. The fact that the most subtle landscape feature (small creeks) were found to influence fires means that fire is likely affected by more significant features as well, but they will probably only influence those parts of a fire that are burning with a very low intensity. ASPECT HYPOTHESIS 3b: Fire return interval is sensitive to aspect. Drier, warmer, south and west slopes are expected to burn more often than cooler and wetter north and east facing slopes. ANALYSIS Unfortunately, the detailed historical fire interval data necessary to ideally study this question were not available for the SBSmkl. However, it was possible to use indirect techniques by correlating aspect to age-class for the 28,730 hectare spatial database (only those pixels with D T M information were included). On this particular area of the SBSmkl , the vast majority of the forested area was older than 120 years, so an average age would not have a great deal of meaning. Instead, the expected areas of the youngest age-classes and the oldest age-class were 55 compared with those observed. A greater than expected number of young forest pixels would indicate more frequent fire activity, while lower than expected values would indicate less frequent fire activity. Those areas with slopes greater than 10% within the area covered by the D T M accounted for about 48,000 pixels. A fairly even balance of pixels were found in each of the eight aspects. Of those pixels with slopes greater than 10%), 2,683 were age-class 1 or 2, and 24,079 were age-class 8 or 9. Although the small landscape represented 3.7% of the SBSmkl , the aspect samples were not taken randomly (they were actually a census of the entire 28,730 hectare block). Thus no statistical inference was possible. The observed occurrence of the youngest and oldest age-classes as a percent of expected (which is always 100%) on eight aspects is shown in Figure 2.5. The results show a tendency for south and west facing slopes to contain more of the younger stands than expected, relative to the north and east facing slopes. East aspects had the lowest (observed / expected) percentage of young stands (68%) and south the highest (142%). The distribution of the oldest age-classes on different slopes showed much less variability, ranging between 94% and 105%) of the expected numbers of pixels. The same four aspects, south, southwest, west, and northwest aspects, showed slightly more than expected older stands compared to north and east aspects. It is interesting to note that the relative frequencies of young stands compared to old stands stayed constant for the south and east slopes compared to the north and east slopes (Figure 2.5). 56 160 140 SS120 < LU tr <100 Q hi h-O UJ 80 Q-X UJ Q w 60 cc LU CO § 40 20 NE SE S SW ASPECT AREAS <40 YEARS OLD ~ AREAS > 140 YEARS OLD Figure 2.5. Observed areas of old and young forest by aspect on slopes of >10% in the SBSmkl, as a percentage of expected area. DISCUSSION The lack of a random sample made it impossible to attach statistical significance to these findings. Nonetheless, it was interesting to find that the pattern of the youngest age-classes indicated the expected, namely that warmer, drier aspects may burn more often than cooler, wetter ones. The results from the older age-classes did not support this hypothesis. This may be a reflection of the large age range which age-class 8 covered (from 140-250 years). Had this age-class been broken down into 20-year age-classes, the results may have more significant. In another conifer-dominated forest type, Hawkes (1979) conducted a study of differences 57 between fire return interval and (among other factors) aspect. He found a positive relationship between short return intervals, and warm dry aspects. A more thorough sampling of historical fire return intervals would be necessary to draw more meaningful conclusions in the SBSmkl . 2.4.6 PATCH SIZES QUESTION 3: What is the distribution of disturbance sizes? A N A L Y S I S The distribution of fire sizes in other fire dominated landscapes has been found to be negative exponential when plotted on a log(area) scale (Ward and Tithecott 1993, Taylor et al. 1994). The expectation was that the distribution of patch sizes for the SBSmkl would be similarly shaped. This required an assumption that each self-contained patch of forest, defined only by age-class, is a single disturbance event. This assumption is not entirely true for two reasons: first, disturbance is not necessarily a spatially contiguous event, and second, the 20-year age-class used allow for multiple events locally. However, these problems should not greatly affect the expectation of finding a negative exponential shape of size-classes. Only the youngest aged patches were used since chances increase with age that the original fire is overlain by another disturbance event12. The database contained 1,743 age-class 1 and 2 patches, totalling 194,000 hectares. The raw SBSmkl size data were transformed to base 10 logarithms, and plotted against frequency in logarithmic increments (Figure 2.6). Figure 2.7 Ideally, only age-class 1 data should be used here, since these have not been overlain by other events. However, the fire activity in age-class 1 was very low, and I was concerned about getting a representative distribution of patch sizes for the landscape. Even though it introduced some bias, I decided to include the patch data from age-class 2 as well. Together, age-class 1 and 2 cover 24% of the landscape. 58 shows the corresponding areas associated with different patch sizes. The results were predictable, but nonetheless striking. The vast majority of the area of the SBSmkl landscape was covered by a small number of very large patches. Although almost 80% of all patches found were 60 hectares or less, they accounted for only eight percent of the area. Patches 500 hectares or less accounted for only 28% of the area, but constituted almost 97% of the numbers of patches (Figures 2.6 and 2.7). In contrast, patches greater than 10,000 hectares constitute less than 0.2% of those counted, but account for 31% of the area. The shape of the patch-size distribution in Figure 2.6 is not negative exponential as expected. There are several possible explanations for the discrepancy. The simplest explanation is that the negative exponential relationship does not always hold. Not all of the distributions found by Taylor et a/. (1994) were negative exponential. Alternatively, the SBSmkl patch size data may fall into a negative exponential distribution i f wider size classes were used (since those of other studies were much wider). However, a simple aggregation exercise did not support this theory. Another possible explanation for the discrepancy is that the size data were in error. Although no independent data were available with which to compare the SBSmkl patch-size data, it was possible to test at least smaller patch numbers for internal consistency. However, to do so required jumping ahead to use the fire module of the landscape computer simulation to simulate fires on the 28,730 hectare landscape. These tests required assuming that the model accurately represented not only fire shapes and sizes, but also remnant (unburnt) island numbers and sizes (see Chapter 3). 59 10 6-L U o DC L U C L 24 J 10 llll.ll..... 100 1000 A R E A (ha) 10000 i i i n 1 0 0 0 0 0 Figure 2.6. Observed SBSmkl patch size-class distribution for the youngest age-classes. 2 0 -1 5 -L U O 1 0 -en L U 0_ . . . . . n i l •lllllllllllllll I I I I I I I [ I I ! 10 i i i i i i i i 100 1000 U l ~ r i TtTT" 10000 TTTTT 1 0 0 0 0 0 A R E A (ha) Figure 2.7. Area distributed by patch size-class in the SBSmkl for the youngest age-classes. 60 The logic of employing simulation is as follows. The greatest proportion of large patch-sizes should be found in the youngest age-class. As patches age, they are overlain and broken up by subsequent disturbances. The patch size-class distribution should therefore change across age-classes. Furthermore, as DeLong and Tanner (1995) found, up to 15% of the area within disturbances are (older) remnant islands. Therefore, in every age-class except the youngest, a portion of the patches observed do not represent disturbances, but island remnants left within other disturbances. From the raw data, there is no way of differentiating between the two types of patches, but we do know that the number of remnant island patches should increase with age-class. The model is capable of very simplistic predictions of the numbers of island remnants. To demonstrate, I assumed that the patch-size distribution for age-class 1 accurately reflected the size distribution of disturbance events, and ran a series of simulations. A total of 115 disturbances were simulated between five and 3,000 hectares and the number and size of islands created within each was recorded. From these data, a series of 17 linear regressions were estimated, one for each island patch size-class up to 100 hectares (using a log scale). These regressions were used to estimate the numbers of islands of each size-class that each age-class 1 disturbance created. The total number of islands in each patch size-class estimated by the simulation represented the total number of patches that are island remnants (i.e. not disturbances). By subtracting this number from the total number of patches observed, an estimate of the disturbance size-class can be calculated. Figure 2.8 shows the estimate of the disturbance size-class distribution estimated by the model compared to the original patch-size distribution, for age-class 2. The difference between the bar pairs is the estimated number of patches that are island remnants. 61 As expected, the differences between the bar pairs were most significant in the smallest patch sizes. In fact, the simulation predicted that none of the smallest patches were disturbances at all. 120, 100000 Patch Size (ha) • RAW DATA H PREDICTED FROM MODEL Figure 2.8. Comparison of observed patch sizes (raw data) to estimated disturbance sizes (predicted from model) for age-class 2 on the SBSmkl. The same procedure can be used progressively through all age-classes to estimate the change in proportions of patch-types between disturbances and remnant islands. The proportion of the total number of patches that were remnant islands increased with increasing age. At age-class 8, the shape of the distribution of disturbance sizes using the model looked quite different than the patch-size distribution as a whole (Figure 2.9). The internal consistency test predicted that most small older forest patches were island remnants and not disturbances. These findings cast doubts on the integrity of the small patch-size data. In fact, the simulation exercise predicted a greater number of small patches than exist in the raw 62 data. In other words, not enough small age-class 8 patches were found. For instance, the number of 2.2 - 2.8 hectare patches of age-class 8 from the raw data was 78, but the simulation exercise predicted 169 remnant island patches in the age-class 8 forest matrix. The inventory, potentially, did not pick up at least 91 of these patches. 200 O100 CD Patch Size (ha) • ACTUAL PATCH SIZE H DISTURBANCE SIZE PREDICTED FROM MODEL • PATCH SIZE PREDICTED FROM MODEL Figure 2.9. Comparison of actual patch sizes to estimated disturbance sizes and estimated (total) patch sizes for age-class 8 on the SBSmkl. The trend of missing small patches was true for all island patch numbers estimates, but most dramatically so in age-class 8. This means that patch size distributions should become increasingly negatively skewed with increasing age (assuming that the distribution of disturbance sizes was constant). However, the patch size data from the SBSmkl revealed virtually identical 63 distributions of patch sizes for every age-class. At this point, the ability of the model to predict island remnant occurrence is moot. If anything, the model has erred on the conservative side since it dealt purely with island remnants, and failed to deal with mere shrinkage of patches. According to the simulations, the only way the raw patch size data could be accurate is i f either island remnants never formed in the SBSmkl (which we know not to be true), or the distribution of disturbance sizes slowly shifted over the last 200 years. Since there was no evidence to suggest that the sizes of disturbances has changed over time, the raw data must be underestimating the number of smaller, older patches of forest. DISCUSSION It is not difficult to imagine how small patches of older forest may be missed in the inventory. Remnant islands, or small areas of older forest are far more noticeable, and more likely to be recognized during inventory stand typing when the matrix is young, and contrast is high. As a stand and its islands age, contrast between the two will fade until it may no longer be possible to delineate distinct islands. The possibility of older islands slowly deteriorating, or being disturbed by other mechanisms such as bark beetle is not supported empirically. The simulation model used for this demonstration created island sizes and numbers consistent with those observed on this landscape across a range of stand and island ages (DeLong and Tanner 1995). This indicates that island deterioration either may not happen for many years after disturbance, or almost immediately afterwards. The results from Section 2.4.1 suggest the former. Aside from the bias problem with the patch size data noted above, there is another reason to be suspicious of the accuracy of the small patch data. The accuracy of the patch information in the 64 raw datafile generally decreases as patches get smaller. A good indication of this is the high proportion (14%) of forested patches in the raw database that are less than two hectares. This was itself an inconsistency since forest inventory maps do not delineate forest polygons below two hectares in size. It is suspected that a large number of these tiny patches were not typed (and recorded as vector information) as distinct patches originally, but were attached to larger polygons. During the process of converting paper maps to digital, rasterized format, there were many potential sources of error. The combined result of all of these difficulties can create a large number of very small polygons that are GIS artifacts and not indicative of either disturbances or remnant islands. On the other hand, given a minimum forest patch size limit of two hectares, it is conceivable that a large number of patches close to this limit were not delineated. Overall, the accuracy of the small patch data is questionable. The combined impact of the bias and accuracy problems with the small patch data is twofold: 1) The distribution of historical disturbance sizes cannot be determined precisely from this type of forest inventory data. The simulation exercise suggests that it may be quite different than the raw data shows, but this cannot be confirmed without empirical study. 2) Following this, the landscape may be more complex, with respect to the mosaic of spatial patterns of age-class patches, than the current data suggests. This would be due primarily to the inventory potentially not picking up smaller patches. 2.4.7 DISTURBANCE SHAPE QUESTION 4: What is the distribution of disturbance shapes relative to disturbance sizes? A N A L Y S I S Disturbance shape complexity was expected to increase with disturbance size. To test for this 65 with the SBSmkl data, a "shape" index was calculated based on a ratio of edge length to area, scaled such that the simplest possible shape, a circle, would yield a shape index of 1. An index of two would then be defined as an area (patch) with twice as much perimeter length as a circle of the same area. The following equation defines shape: SHAPE = (McGarigal and Marks 1994), where P = perimeter, and A = area. A shape index was calculated for each patch in age-class 1 and 2 from the large dataset, and then plotted against log (base 10) of size (Figure 2.10). Only data from age-class 1 and 2 were used because they are more likely to depict the original disturbance patch. 2 0 - , 1 5 -w 1 0 -< X • i i i i 0 1 2 3 4 A R E A log(ha) Figure 2.10. SBSmkl patch shape by patch size for the youngest age-classes. 66 It is obvious from the raw data that shape became more complex with increasing disturbance size, although the difference was only noticeable after about 100 hectares. Up to 100 hectares, the shape index ranged between one and three, rose to around five at 1,000 hectares, and climbed even more dramatically thereafter. The highest shape index was almost 17 for the largest patch found (34,000 hectares). DISCUSSION The significance of increasing shape complexity is difficult to assess without a better understanding of exactly how the shape metric was calculated. The tendency is to read Figure 2.10 to mean that fires became more convoluted, or less circular as they become larger. However, when edge was calculated for patches it included interior edges of remnant islands contained within larger patches as well. As DeLong and Tanner (1995) have already shown for the SBSmkl , the number and sizes of remnant islands increase with increasing patch size. Eberhart and Woodard (1987) found similar increases in the same shape index with increasing fire size in northern Alberta, but they also found that a shape index which counted only the fire perimeter increased by a narrower margin. Given this, it was conceivable that a portion of the increase in shape from the SBSmkl data came from an increase in island remnant frequency, and not from an increase in the complexity of the exterior perimeter. 2.4.8 SPECIES / SITE RELATIONSHIPS HYPOTHESIS 4: Species, or species combinations are not associated with specific soil or site-types. 67 An alternative statement of this hypothesis is that SBSmkl tree species composition cannot be predicted by site characteristics. A N A L Y S I S The 28,730 hectare raster-based dataset described in Section 2.3 was used for this analysis. Although the sample represents 3.7% of the landscape, the pixels were not chosen randomly, thus violating one of the assumptions of binomial testing. In any case, the scale of both the soil mapping, and the inventory mapsheets was coarse relative to the pixel size. Only very general trends were expected, and statistical inferences were not possible. The percentages of each species averaged over all pixels based on five major soil-types (till, fluvial, organic, lacustrine, and lacustrine / till complexes) were calculated. The average percentages of each species was also calculated for five slope-classes within each soil-type making 25 site-types in total (Table 2.4). The standard deviation of the percentage of each species was included as well to give an impression of the relative spread of the data in each class. Only the four main species were included in the summary since subalpine fir and Douglas-fir were found in low percentages in only a fraction of the stands. The dominance of pine on the landscape was immediately apparent, averaging 70% over all pixels. Hybrid spruce was the second most common species at 24% overall, with black spruce and poplar species accounting for only 3% each. The lack of the poplar species was surprising in that it is contrary to what is found in other areas throughout the southern boreal and montane transition forests of northern Canada (Rowe 1972, Clark 1994). 68' Table 2.4. Species percentages by soil-type and slope-class on the SBSmkl . SOIL-TYPE SLOPE LODGEPOLE. PINE Hybrid SPRUCE B L A C K SPRUCE POPLAR SPP. (no. pixels) CLASS M E A N S.D. M E A N S.D. M E A N S.D. M E A N S.D. ORGANIC 1 27 37 10 21 62 48 1 3 (2,122) 2 35 40 . 20 31 45 50 1 2 3 44 37 34 33 21 41 1 3 4 53 40 32 36 15 36 - -5 51 31 43 30 6 24 - -ALL 34 39 18 29 47 50 1 2 FLUVIAL l 35 38 48 37 2 15 14 28 (5,516) 2 48 36 37 32 5' 21 10 21 3 50 33 33 27 6 24 9 17 4 58 34 29 26 4 20 7 14 5 55 35 33 27 0 4 11 17 A L L 49 36 36 31 4 20 10 20 LACUSTRINE i 72 23 20 16 5 18 4 7 (23,108) 2 68 27 23 19 3 16 5 9 3 56 19 33 22 2 12 8 14 4 52 20 34 22 1 7 12 18 5 50 6 30 20 I 11 18 26 ALL 65 29 25 20 3 15 6 13 TILL 1 72 27 23 22 2 14 2 9 (53,899) 2 74 24 23 20 1 10 1 5 3 • 73 26 24 21 1 8 1 5 4 73 27 24 21 1 7 1 4 5 72 13 27 25 0 1 1 3 A L L 73 24 24 21 1 9 1 5 TILL x 1 84 22 14 19 2 12 - -LACUSTRINE 2 86 18 12 16 1 8 - -(6,796) 3 85 9 15 18 0 4 0 2 4 80 10 19 22 - - 1 6 5 76 5 18 23 - - 6 12 A L L 85 20 14 18 1 8 0 3 GRAND TOTALS 70 28 24 22 3 15 3 9 69 The soil-types alone did a reasonable job of differentiating species combinations. The small landscape area was dominated by tills (59% by area) which were representative of average forest conditions (i.e., the averages of the species compositions on tills were close to the overall averages). Many of the areas of till were pure stands (mostly of lodgepole pine). Only the till / lacustrine complexes had a greater amount of lodgepole pine, averaging 85%. Most of these till / lacustrine areas supported pure lodgepole pine stands, the remainder of the area being a pine/hybrid spruce mixture. The low standard deviations relative to the same species on other soil-types was an indication of the large proportion of pure stands on till / lacustrine complexes. Negligible amounts of black spruce and poplar species were present on both till and till / lacustrine soils. The remaining soil-types tended to support species mixtures. Lacustrine soils were dominated by pine (65%) and hybrid spruce (25%), but showed an average of 6% poplar species, and 3% black spruce. Fluvial soils had less pine (49%), more spruce (36%), and the highest percentage of poplar species of all of the soil-types (10%). Organic soils were the only areas not dominated by pine. Black spruce was the leading species on organic soils (47%) followed by lodgepole pine (33%) and hybrid spruce (18%), with little or no poplar found. The stand composition on organic soils and, to a lesser degree, fluvial soils was highly variable, as their high standard deviations indicate. Using slope-class to further refine soil groups was informative in only three of the five soil-types. For the tills, and till / lacustrine complexes, slope-class had little impact on the average species compositions. This was anticipated to some degree since the standard deviations of species percentages for these two soil-types were relatively low. The only notable exception to this was 70 a slight increase in the amount of poplar species on the steepest slopes of the till / lacustrine complexes at the expense of pine. As there was no parallel trend towards increasing poplar species with increasing slope on the tills, the possible cause of this is unknown. The impact of slope differentiation on the other three soil-types varied. It had a moderate, but noticeable impact on lacustrine soils. The average pine percentage steadily dropped from 72% on flat sites, to 50% on steep sites. Both hybrid spruce and poplar species generally increased over the same span. For fluvial soils, quite the opposite took place. Pine increased from 35% on flat sites, to 55% on steep sites, while hybrid spruce dropped from 48% on flat sites (where it dominated) to 33% on steep sites. Poplar species and black spruce did not change noticeably over the different slope-classes for either lacustrine or fluvial soils. A wide range of species combinations occurred on fluvial soils, as indicated by the high standard deviations for all species on fluvial soil-types. Species percentages on organic soils showed the most dramatic change across different slope-classes. Black spruce dropped from a high of 62% on flat sites, to only 6% on steep sites. In fact, black spruce only dominated on slopes of less than 10% on organic soils. Both pine and hybrid spruce percentages increased from flat to steep slopes, but no single species tended to dominate on these steeper sites. However, on organic soil-types, the average species composition was quite misleading. Where black spruce occurred on organic soils, it was almost always pure. When black spruce was absent or negligible, mixtures of pine and hybrid spruce were found. This resulted in very high standard deviations of species percentages for all slope-classes on organic soils. 71 Several other species X site groupings were tested, including the percentage of a particular species (especially pine) on specific aspects, proximity to creeks and other water bodies, and specific soil-types. The only other notable relationship found between species percentages and permanent landscape features was the association between stand purity and slope-class. Flat sites tended heavily towards pure stands, and steep sites tended to be mixtures (Table 2.5). Fifteen percent of all flat areas had pure stands of one of the three major species, while only two percent of the steepest sites had pure stands. This applied equally to all soil-types and species. Table 2.5. Percentage of pure stands by slope-class in the SBSmkl . Slope-Class (%) Area Pure (%) 0-4 15 ' 5-9 10 10-19 7 20-39 6 40+ 2 DISCUSSION Without the benefit of statistical testing, it was not possible to say which differences in species compositions were meaningful, and which were not. It is also questionable whether or not representing these differences as means and standard deviations was the most appropriate method of-study. However, there is precedent for the use of soil factors in site-species studies (e.g. McQuilkin 1976, Verbyla and Fisher 1989, Monserud et al. 1990, Andison 1993) and the use of both soil and slope in ecological classifications (DeLong et al. 1993). The data did not allow more detailed site-species relationships to be pursued. 72 Despite the restrictions on the investigation, soil-types did a reasonable job of differentiating species groupings. Furthermore, they were consistent with existing knowledge of species preferences and requirements; lodgepole pine is found most often on dry outwash (till) soils (Pojar et al. 1984), pine and hybrid spruce mixtures or hybrid spruce dominated stands on moister, richer sites such as fluvials (Burns and Honkala 1990), and lowland, organic areas tend to black spruce (Pojar et al. 1984). The only unexpected result was the negligible percentages of the two poplar species (trembling aspen and black cottonwood) overall. Since both species are more than capable of occurring on any of these site-types, and the area is well within their ranges (Fowells 1965), the reason for their absence may be related to aspects of the historical disturbance regime. The addition of the slope-classes was moderately useful for refining species compositions on fluvial and lacustrine soils, and was very successful on organic soils. This is particularly notable because one might wonder why any of these soil-types have slopes beyond 5% in the first place. By definition, all three should be flat. That they were not is most likely related to the coarse level of resolution used for the soil maps which allowed some or all of the transition zones surrounding each of these areas to be included within these soil-types. Many of the edges of fluvial and organic soils in particular, are quite steep. There was no way of testing this hypothesis without proper ground truthing, but the results indicate that this is a reasonable explanation. For instance, the expectation was that pure black spruce would dominate on organic sites, but in fact this was only true of the flattest sites. As slope increased, the percent of black spruce declined from 62% on (true organic) flat sites to just 6% on the steepest sites, where drainage conditions would be dramatically different. 73 The prevalence of pure stands on flat areas is more difficult to explain, and would require a more rigorous study to properly address. For instance, the occurrence of pure black spruce on flat organic sites is a well known relationship, and is considered to be largely a function of site. However, pure black spruce sites did not account for the majority of the pure species pixels overall. The rationale for pure pine on dry, flat outwash areas, and pure spruce on heavier lacustrine soils is not as obvious. Whether the degree of purity on these soil-types was related more to site homogeneity, or variation in fire coverage and severity, is unknown. An important observation stemming from this investigation of vegetation dynamics on the SBSmkl landscape was the broad range of species combinations on all site-types. The standard deviations of all 25 soil X slope site-types averaged 25%, and ranged as high as 45-50% in some cases. This was far from being a comprehensive study, yet this sort of variability is typical of more detailed site-species research, some using similar site indicators (e.g. McQuilkin 1976, Verbyla and Fisher 1989, Monserud et al. 1990, Andison 1993). The point is that site information will provide only a general guide to where and when individual species or species combinations will be found even in relatively simple forests. In forest-types where succession is only a minor agent of change, this leaves response of tree species to the stand initiation process and small-scale disturbance events such as windthrow and bark beetle attack as the most probable mechanism causing this variation. 2.5 SUMMARY Several aspects of landscape level phenomena of the SBSmkl were tested and/or quantified in 74 this chapter. Together, these elements are thought to be responsible for the majority of the landscape level dynamics that occurred historically, and created the coarse-scale mosaic pattern that can be observed. Most of these elements are associated with the disturbance regime, which in turn is dominated by forest fire activity. Although vegetation dynamics are active on the SBSmkl landscape, their impact on the mosaic pattern is likely secondary to the fire regime. A reconstructed age-class distribution of the entire SBSmkl landscape indicated a bi-modal shape suggesting a non-equilibrium age-class mosaic caused by the occurrence of very large, infrequent disturbance events. In fact, based on the patch size data, the vast majority of the forest on the SBSmkl landscape in 1954 resulted from very few large fires. For instance, fires greater than 500 hectares accounted for only 3% of all fires, but consumed 72% of the area. These larger fires also had more complex shapes, but this was could be due solely to the increasing frequency and size of remnant islands left unburnt within a fire's perimeter. None of the "steady state" Weibull-class models adequately described the SBSmkl age-class dynamics. The bi-modal age-class distribution of the SBSmkl is typical of boreal-type landscapes, suggesting that: 1) it may not be possible to ever observe a more meta-stable landscape anywhere, and 2) a more suitable model for describing SBSmkl landscape dynamics would be one that considers fire as a two-layer disturbance phenomena. For the most part fires burn many small to medium sized areas, but these are interspersed with very infrequent extreme fire conditions that consume vast areas of forest in a short period of time. These extreme fire events may occur only once every 50-100 years. 75 Accordingly, the landscape was best described by the variation about both sizes and frequencies of disturbance events. For instance, the average length of time between fire events in any given stand was between 80 and 100 years on the SBSmkl, but varied tremendously spatially and temporally. Historically, it was estimated that between 7% to 43% of the entire landscape burnt in a single 20 year period, although a simulation exercise suggested that the range of burning rates per 20-year period may be much greater. This variation allows for the survival of some stands well beyond 100 years. In the reconstructed landscape used in this study, over 50% of the forest was older than 80 years of age (although simulations showed that the presence of this large amount of older forest on the landscape was likely rare historically). Other stands were repeatedly disturbed within a relatively short period of time. Since management disturbance regimes tend to average out disturbance rates and concentrate on older timber, these management strategies may severely reduce the amount of mature and over-, mature timber, and/or reduce the landscape disturbance rate. The consequences of either of these changes are unknown. However less area of mature forest may affect the manner in which plant or animal species use the landscape, and reduced landscape disturbance rates may reduce nutrient cycling rates, resulting in declining timber productivity overall. Although it was not possible to test for the impacts of a wide range of topographic features and forest-types on fire activity, it was found that stands on rapidly drained soils experienced fires more often than stands on other soil-types. This was the only such differentiation that could be made, suggesting that fire is relatively non-selective to permanent landscape features at coarse 76 scales. At finer scales, the current age-class distribution and the noticeable lack of old stands suggests that stands are more likely to be disturbed as they age, but it could not be confirmed that fire is the mechanism by which they are disturbed. More specifically, stands greater than 200 years of age (on the 1954 landscape) were noticeably absent; however, simulations showed that this was inconsistent with an age-invariant model of burning. On the other hand, age sampling provided ample physical evidence that fires were capable of burning through even very young stands. This suggests a threshold of increasing disturbance susceptibility as opposed to the gradual increase over time suggested by Weibull models. Circumstantial evidence was found to suggest that the warmest aspects (south and west), experience more fire activity than cooler aspects (north and east). Finally, although a higher than expected number of fire edges were associated with creeks, fires were generally not strongly influenced by such fuel-breaks. The consistently weak evidence in support of either differential fire regimes or deterministic small-scale fire behaviour suggests that large fires, which dominate the landscape, are largely unaffected by fuel or even major topographic features. Smaller, less intense fires may be highly affected by such features, but because they account for such a small portion of the landscape, only weak trends are observed. This theory is consistent with the implications discussed above concerning the bi-modal shape of the age-class distribution. 77 It was not possible to confirm that disturbance sizes on the SBSmkl exhibit a negative exponential distribution as expected. The patch size data resulting from forest inventory mapsheets was both biased and inaccurate with respect to representing the number of small patches. The bias was revealed by simulations indicating that the data underestimated the number of patches between 2 and 100 hectares in the older age-classes. The degree to which this occurred was unknown, but it is suspected that the overall patch size distribution may be negative exponential, and that the disturbance size distribution may be a negatively skewed normal distribution. The bias also means that the landscape mosaic is actually more complex than inventory maps indicate, meaning that the contribution of very small patches to overall pattern may be underestimated. However, more intensive, rigorous, field sampling would be required to test this. Sub-boreal tree species preferences for coarsely classified site-types according to soil and slope-classes were evident, but clearly all species are capable of occupying a wide range of sites. It is unlikely that arboreal succession is responsible for a large degree of species composition shifts in the SBSmkl. This leaves the stand initiation process, and small-scale disturbances such as bark beetle attacks and windthrow as likely mechanisms for creating the variability in species composition by site-type. The relative impact of each could not be assessed with these data. Although these results on their own contribute to the understanding of the dynamics of boreal-type landscapes, each of the elements discussed in this chapter were also taken into consideration when constructing the landscape simulation model for the SBSmkl. Chapter 3 will quantify, 78 where possible, each of these elements, develop and describe a modelling framework within which they fit, and provide some verification of model output as the next step in this research. 79 CHAPTER 3 - LANDSCAPE MODELLING This chapter summarizes the second of the three parts of this PhD research. In it, the literature relevant to landscape models is reviewed, a suitable method of modelling the SBSmkl landscape is proposed, and then the model parameters arising from the disturbance regime investigation in Chapter 2 are fitted and tested. The simulation model will be used to investigate the temporal and spatial nature of sub-boreal landscape dynamics in Chapter 4. 3.1 LITERATURE REVIEW Any landscape observed today is but one possible arrangement of ecosystems out of countless other possible arrangements — a snapshot in time. In both landscape ecology research and landscape management, it is not the individual snapshots that are of importance, but the range and variability of those snapshots (Baker 1989c). To better understand the dynamics of landscapes, computer models are being increasingly used because they are capable of projecting variation through time, and they allow a simple level of experimenting that is not possible empirically. There is a wide range of model types which could be considered landscape models, from non-spatial Markov models, to raster-based, spatially explicit, diffusion models. This review will cover only spatially explicit computer models of landscapes under three main headings: disturbance models, vegetation dynamics models, and mixed or 'true' landscape models that combine both. The necessary data sets and the methods used to measure spatial patterns resulting from such models will also be discussed. 80 3.1.1 SPATIALLY EXPLICIT DATA Spatially explicit computer models require spatially explicit data sets. Choice of the type of spatial unit, and the scale of resolution at which to measure and model, are important decisions. The two main types of spatially explicit data are "vector-based" and "raster-based". Maps provide polygons, which can easily be converted to vector data as either point or vector coordinates in a Geographic Information System (GIS) (Coulson et al. 1991). Vector data require less computer storage space, are easier to summarize, and tend to be used primarily for mapping and presentation. Raster-based, or pixel data, are normally available from satellite imagery, but can also be converted from vector data within a GIS system. Pixels are far more flexible with which to work since spatial references are much simpler (Coulson et al. 1991). Pixels can be used individually (individual-based) or aggregated by attribute (patch-based) (Flamm and Turner 1994). Pixels (as individuals or patches) are the unit of choice for most models (Baker 1989c, Turner et al. 1989, Antonovski et al. 1992, Mladenoff et al. 1993, Flamm and Turner 1994). Each spatial unit (pixel or polygon) can be assigned one or more discrete (state) or continuous variables, and is assumed to be internally homogeneous (Baker 1989c). Where and how this classification is done will determine the pattern of pixels observed, so it is important that the classification used is ecologically meaningful. Typically, the information contained in each spatial unit is broad-based nominal data, such as age-class, developmental or successional stage, or other compositional or structural land use information. The size of pixel, or the "grain" size, determines what level of detail is to be modelled (Baker 1989c), as well as how much data there are to be collected, stored, and manipulated. The 81 problem of what level of resolution to use is usually associated with pixel data. Small pixels capture more detail, but require larger datafiles, and potentially more complex manipulation i f the model involves interaction among neighbours. Large pixels are easier to store, reference, and manipulate, but there is a risk of losing potentially valuable information (Allen and Starr 1982). Ideally, the choice of pixel size is made by weighing all of these factors against the total area under consideration (the "extent"), and the process(es) to be modelled (Nemani et al. 1992). Usually, as extent increases, grain size increases as well. The best grain size to use may not always be obvious, but spatial data have limits beyond which further manipulation is pointless. For instance, the resolution of the original data collection or mapping is a fact that cannot be improved upon during a rasterization process (Evanisko 1991). Using 10 m pixels for data that were originally mapped at 1:50,000 creates a lot of data, but probably no more potential information than a file created with 100 m pixels. Moving from 100 m to 10 m pixels is an increase in detail that is not important, but creates a practical problem in that it results in a hundredfold increase in the file size. Even when highly detailed data are available and reliable, it may not always be wise to use these data, depending on what it is that is being modelled. In a simulation exercise, Costanza and Maxwell (1994) found that using a coarser level of resolution actually improved overall model effectiveness, because unnecessary fine-grained detail was averaged out. However, beyond some point, the loss of detail may contain information relevant to landscape dynamics, and this could alter the pattern observed. Turner et al. (1989) studied the effects of changing both resolution and extent on seven different landscapes, and found complex relationships between pixel size, degree of aggregation, and the resulting landscape pattern metrics. Baker (1993) and Benson and 82 MacKenzie (1995) also found that landscape metrics responded differently depending on the size of the pixel used for the analysis. It is not surprising that the scale of observation affects the patterns observed in natural systems. Processes such as seed dispersal, competitive ability, forest fires, and climate operate on vastly different time and space scales. Each is difficult enough to understand or predict on its own, but together their behaviour is so complex it is often thought to be random. The result of this interaction is more accurately defined as "heterogeneity" (Merriam 1987), and is unavoidable in natural systems observed at any scale, since natural processes are continuous rather than discrete phenomena (Allen and Starr 1982). What is observed when moving from one level (of resolution and/or extent) to another is a change in the relative influence of each of these phenomenon, and this is reflected in a change in the pattern observed (Allen and Starr 1982). In other words, it is not possible to ever observe a "clean" set of phenomena in natural systems, which explain all or even most of the variability observed. This has serious ramifications for landscape studies: where and when should landscapes be observed? Obviously it is desirable to observe at the scale relevant to the phenomena of greatest interest (Turner and Gardner 1991), but what should be done when these scales differ, and/or overlap with scales of other mitigating phenomena? "Hierarchy theory" may help to find optimum scales of observation. This theory states that ecological processes are not perfectly continuous scalarly, and thus tend to concentrate at certain scales in an hierarchical manner (Allen et al. 1984, Kolasa and Rollo 1991, O'Neill et al. 1986). If a level (of resolution and/or extent) could be found at which the patterns were dominantly a result of the processes of most interest, that would be the "best" scale(s) of observation. Although some testing has been done towards 83 finding these special scales (Costanza and Maxwell 1994, O'Neill et al. 1991), there is currently no means of associating any particular scale with one or more processes (Allen and Starr 1982, O'Neill 1989). In other words, it is extremely difficult to associate meaning with the patterns. Simulation modelling is helping to sort some of this out, but in the end, logic and the experience of others currently serve as the best guides for scale choices in models and measurements. Unfortunately, the choice of data format and/or scale is seldom an option. Landscape studies very often have scales of observation imposed on them due to data availability, convenience, habit, or technology (Evanisko 1991, Levin 1992). This is particularly true of models that access data directly through GIS packages, many of which have rigid interfaces with respect to their data formats and/or scales (Nemani et al. 1992). In many instances, users resort to writing their own software or using public domain interface software such as GRASS (Geographic Resources Analysis System) (Coulson et al. 1991, Flamm and Turner 1994) in order to overcome some of these problems. In the end, map patterns will always emerge from spatial data, but whether or not they reflect ecological conditions, personal biases, a scale-dependent artifact, economic convenience, or some combination of these, is open to debate (Evanisko 1991). The selection of what data are measured, and at what scale(s), is an important choice in landscape modelling. 3.1.2 DISTURBANCE MODELS It is a testament to the perception of disturbance as the dominant landscape process that many of the best landscape models today deal solely with disturbance simulation. The most sophisticated 84 of these are fire behaviour models that are extensions of the North American fire behaviour prediction systems, and represent the final product of over two decades of fire behaviour research. Most of these models focus on individual fire behaviour. They use fuels, topography and fire weather information to make predictions of rates of spread (ROS) and fire intensity in time-steps of hours or days (Fried and Gilles 1988, Green et al. 1990, Stocks et al 1990, Kalabokidis et al. 1991, Vasconcelos and Guertin 1992, Knight and Coleman 1993, Finney 1994). Many of these models are based on the original fire spread formulations of Rothermel (1972), but make some simplifying assumptions from the original model (usually with respect to fire weather information). These models were developed with fire control in mind. They are mathematically complex (particularly the vector-based models), require large amounts of fire weather, fuel, and topographic data, and do not deal with spatial pattern per se. Nor have they yet been linked with dynamic vegetation models. On the other hand, there are potentially tremendous advantages to pursuing further development of fire behaviour models as landscape disturbance models. For instance, it would be possible to track the spatial variation of both fire intensity and crowning. Combined with vegetation information, this could lead to predictions of survival of individuals or groups of trees (remnant islands). As important as arboreal survival may be to landscape dynamics (seed source, structural and compositional diversity), I found no references to models of any type that are capable of predicting this phenomenon after large-scale disturbance. As an alternative to the traditional fire behaviour models, the disturbance spread process has been described in terms of how generic disturbances spread across a heterogeneous landscape composed of internally homogeneous cells or pixels (Green 1989, Turner et al, 1989, Turner and 85 Dale 1991, Hargrove et al. 1993). These are often referred to as "diffusion" or "percolation" models, after the percolation modelling of epidemic spread. The disturbance spreads from cell to cell based on threshold levels of susceptibility. The susceptibility is estimated for each cell individually based on proximity, and attributes such as age, composition, or structural features (Turner and Dale 1991). These models are relatively new, and many have only two levels of susceptibility ("yes" and "no"). One of the most advanced of these models uses simple wind and fuel functions to drive fires across landscapes, and even includes the potential to start spot fires (Hargrove et al. 1993). Most of these models do not yet deal with disturbance shape, or consider variable levels of intensity and/or severity. A slightly more sophisticated version of a percolation model is offered by Clarke et al. (1994). They propose a series of fuel, terrain, and fire weather overlays that contribute to a weighting factor for each pixel that relates to its relative flammability or susceptibility to fire. Once these weights are assigned, an individual fire trail, or a "firelet", moves from an ignition source pixel to one of its eight (unburned) neighbours stochastically. The trail continues until it goes "out" after a pre-determined number of pixels have been consumed. As each trail travels, it can spawn a new fire source from which another trail starts to burn. Fires spread continues in this way, eventually building up dozens of firelet trails, and even leaving unburned remnant islands. Although setting up parameters for the model is subjective, trials against historical fire data have proven moderately successful (Clarke et at. 1994). This method differs from the percolation approach in that each cell possesses the necessary information to "decide" for itself if, or in what direction, to allow the fire to proceed driven by a controlling algorithm or function. These decisions are made independent of what is happening 86 anywhere else the fire might be burning, in sharp contrast to more traditional fire behaviour models described earlier. This is more accurately described as a "cellular automaton" model. Clarke et al. (1994) argue that this technique properly represents the fractal nature of fire boundary shapes. 3.1.3 VEGETATION DYNAMICS MODELS Vegetation dynamics models provide establishment, growth and mortality routines to describe changes in vegetation patches over time. Although there are a large number of stand level vegetation models, I will only discuss those that have been used, or have the potential for use, in spatially explicit landscape models. The simplest type of vegetation model assumes that the previous stand composition, or some earlier serai stage of it, will return (Cattelino et al. 1979). Despite the seemingly rudimentary approach, such a model can be spatially explicit, and may be appropriate i f one is willing to accept deterministic assumptions (Kessell 1979). This would make the most sense in very simple forests such as the boreal. A "transition matrix" approach to stand dynamics tracks changes in the distribution of elements through time. The elements can be defined as a finite number of "states", as with Markov transition probability models, or through differential equations. In the case of Markov transitions, the states are usually some arbitrary classification of stand composition, while for differential models, output can be more complex, depending on the number of elements that are modelled. Transition models have proven useful tools in ecology because of, rather than in spite of, their 87 simplicity. Horn (1975) and White et al. (1985) derived gap replacement models which use the measured frequency of seedlings in gaps to generate probabilities of a species filling each gap. A similar approach has been used to describe successional probabilities (Hobbs 1983, Scanlon and Archer 1991), requiring either time series data, or chronosequence assumptions. Davis and Dozier (1990) summarized initial composition following disturbance, and structural changes over time in transition probability tables against relevant land positions. Transition matrix models are not always ideal. They have been criticized for being limited to having transitions occurring over identical fixed time intervals, being insensitive to historical transitions, always resulting in steady-state configurations, and not addressing causal mechanisms (Turner 1987, Baker 1989c, Scanlon and Archer 1991, Turner and Dale 1991, Antonovski et al. 1992). Recent work has found solutions to most of these problems, but often at a high cost. For instance, the problem of lack of historical reference can be resolved by including the previous state in the transition matrix. However, this increases the number of possible transitions by a power of two (Baker 1989c). The most serious problem with transition probability models is their difficulty in incorporating spatial information (Turner 1987, Baker 1989c, Turner and Dale 1991). Incorporating knowledge of a neighbour's state in the transition for a given pixel has experienced only limited success (Turner 1987), largely because it causes a tremendous increase in the number of transitions to track (Sklar and Costanza 1991). Considering most landscape models that use transitions have in the neighbourhood of 2-12 states, it is unlikely that the several hundred thousand states necessary for even a 5-neighbour model is viable (Rykiel et al. 1993). 88 "Gap models" such as JABOWA, FORET, ZELIG, and FIRESUM use a series of differential equations to describe establishment, diameter growth, and mortality of individual trees on a small homogeneous site (Botkin et al. 1972, Shugart and Noble 1981, Keane et al. 1989). Although the location per se of each tree is not modeled, plot sizes are small (1/10-1/12 hectare). Disturbances are either single-stem or stand (plot) destroying, and growth response is based on climate, internal shading, and spatial environmental variation. A more sophisticated version (ZELIG) uses larger plot sizes and can track locations of individual trees (Smith and Urban 1988). From the stem diameters, plot summaries of leaf area index, biomass, and basal area are derived (Baker 1989c). Since gap models are stochastic, they are normally run 50-100 times and both the mean and variance of the relevant metrics recorded (Weinstein and Shugart 1983). The disadvantages of gap models are that they do not yet operate on scales beyond several hectares, and require considerable calibration data for a single site (Turner and Dale 1991). However, their potential for use in landscape models is perhaps the greatest of any other model type because they are non-equilibrium (i.e., do not necessarily lead to steady state configurations), are capable of responding to environmental gradients and disturbances, and allow lower level dynamics (of individual tree characteristics) to influence higher level (landscape) patterns. In addition, gap models are highly mobile (i.e. the JABOWA framework has been adapted to many areas of the world) (Botkin 1993), and are therefore theoretically capable of defining landscape vegetation dynamics as a spatially distributed matrix of individual models (Shugart and Noble 1981) or model outputs. Another approach to landscape vegetation modelling is the vital attributes approach. First derived by Noble and Slatyer (1980), the characteristic attributes of species such as propagule strategy, 89 growth behaviour, regeneration strategy, and life stages respond to disturbance to initiate and grow each stand (Cattelino et al. 1979). For example, should a fire of moderate severity burn a mature stand of pure lodgepole pine adjacent to a black spruce/aspen lowland, lodgepole pine will return from serotinous cones, but some black spruce will invade from seed near the fires edge, while aspen will likely invade the entire area (because of differential dispersal distances). Pine and aspen will outgrow black spruce, but i f the stand lives beyond 80-100 years, aspen will begin to drop out of the stand since it is intolerant and has the shortest life span. If the stand survives beyond 150-170 years, lodgepole pine will begin to drop out due to old age, and will be replaced by black spruce, which layers, and is more tolerant. The next fire will not return lodgepole pine unless it is within the time period assumed for viability of seed stored in serotinous cones on the site. The main advantage to the vital attributes approach is the direct use of simple, available species information at the appropriate scale. Furthermore, vital attributes models do not converge, or assume historical dominance (i.e., they are not polyclimatic). Although designed as a non-spatial approach, it is considered to have great potential to become spatially explicit (Cattelino et al. 1979). Incorporating edaphic positions or environmental gradients, seedbed conditions, and disturbance effects within the vital attribute framework is difficult, but not impossible. The biggest concern is that the approach is deterministic, as it keys on average behaviour. For boreal-type forests, this may be less of a concern than in more complex hardwood or mixedwood forests. 3.1.4 LANDSCAPE MODELS Models that combine several landscape processes, such as disturbance with vegetation dynamics, 90 into a single model are potentially very complex, although there are excellent examples of very (intuitively) simple versions. One of the first spatially explicit landscape computer models was the gradient model of Kessell (1976). Gradient models are based on the gradient analysis theory of Gleason (1926), which associates the observed range of species behaviour with specific edaphic conditions such as elevation, aspect, topographic position, and soil (Kessell 1976). These relationships are derived by associating the presence and abundance of individual species against actual environmental gradients such as elevation, aspect, topographic position, and soil development (Kessell 1976). Originally used as a fire management model, fire behaviour for a particular location is predicted using Rothermel's (1972) fire behaviour model, fire weather, topography and estimated vegetation gradient information. The vegetation information is assigned deterministically from ordinations of site to species, and predicts fuel-type and fuel-load (Kessell 1979). The main advantage of a gradient model is its ability to assign vegetation based on edaphic and topographic features. This allows pixel-specific topographic and vegetation information to be used in the model, without the requirement of collecting vegetation information for each pixel. However, it assigns only the average vegetation, deterministically reassigns vegetation after disturbance, necessitates extensive sampling and ordination to create a complete file of site-species associations, and requires highly detailed fire weather data to run properly (Kessell 1979). The model is more useful for fire spread and intensity purposes, than for predicting fire pattern and shapes (Turner and Dale 1991). Many of the landscape models are based on one or more of the techniques already mentioned. 91 For instance, pixel-based transition probability models have been expanded to include the spread of disturbance and/or topographic information at landscape scales (Gardner and O'Neill 1991, Turner and Dale 1991). Transition models can also be simplified to track state-changes of single pixels or patches. Such formulations have proven particularly useful in simulating land-use changes in urban/rural interface areas (Flamm and Turner 1994, Muller and Middleton 1994). It is also theoretically possible to extend a gap model such as J A B O W A to landscape scales by essentially making each pixel its own model (set of parameters) which then interacts with other models in parallel through space and time. Keane et al. (1989) took a step in this direction by creating a gap model which incorporates different fire characteristics according to fire weather and one of seven fuel types (FIRESUM). ZELIG, a descendent of the JABOWA/FORET approach, allows adjacent, but otherwise independent grid cells, to interact and influence each other, although the assumption of site homogeneity still holds (Smith and Urban 1988). The realistic expansion of gap models to landscapes in their current form would require calibration of unique model parameters according to site and/or edaphic position. When this information becomes available, this will be a useful modelling approach, but for now, the data and computational power requirements exceed our abilities (Antonovski et al. 1992). There is greater potential in using simplified versions of gap-type models, and generating unique vegetation, topographic, and soil parameters for each pixel similar to the gradient model mentioned above. The only model found in the literature that has taken such an approach is the LANDIS model from Wisconsin (Mladenoff et al. 1993). LANDIS is the combined result of several decades worth of local succession, site/species, and fire behaviour research, and is perhaps the most advanced, complete, landscape model in existence today. 92 Some of the most promising landscape models are "probabilistic" models. Turner and Dale (1991) refer to probabilistic models as including frequency, spatial distribution, return interval, rotation, periodicity, area, and intensity. In one of the best examples of a working probabilistic model, Antonovski et al. (1992) created a pixel-based model that recognizes eight successional stages in each of four ecotypes. Fire danger, or the probability of a cell burning, is represented simply as temperature X humidity. The burning thresholds for each ecotype were found empirically, and the historical number of days per year that experienced the necessary threshold conditions is then distributed throughout each year. Fire is randomly initiated and passed from cell to cell based on a binary fire-maturity indicator and a stochastic weather element. Seed invasion depends on a cell's relative distance to one of the four ecotypes that survived the fire (Antonovski et al. 1992). Unfortunately, due to lack of information, the final model lumped all stands into a single ecotype. However, it still represents a coherent method that is well within our existing technological grasp, and it could be tailored to the amount and detail of existing information. 3.1.5 MEASURING PATTERN The output of spatially explicit landscape models requires the use of specific metrics that are capable of describing spatial relationships. A key assumption in studying and/or modelling spatial patterns is that there is a relationship between function and landscape structure. For the most part, this claim is untested empirically (Bunce and Jongman 1993), although there is considerable circumstantial evidence to suggest that landscape structure influences animal movement (O'Neill et al. 1986, Hof and Joyce 1992, Mladenoff et al. 1993), water runoff, erosion, disturbance spread (Turner 1987, Turner and Dale 1991), nutrient cycling and exchange between ecosystems 93 (Clark 1990, Shaver et al. 1991), net primary production, speciation, stability (Godron and Forman 1983), biomass accumulation, (Romme and Knight 1982), and even extinction (Krummel et al. 1987, Kareiva and Wennergren 1995). These landscape structures are collectively expressed in terms of "pattern". Finding a way to describe landscape pattern is challenging. Krummel et al. (1987) point out that there is so much complexity to describe, any single technique will only detect certain kinds of patterns. Also, by definition, many landscape studies are not species or land-use specific, and must somehow try to summarize general patterns. For these reasons, it is usual to adopt several metrics, revealing different structural characteristics. Diversity, dominance, adjacency, amount of edge, and contagion or connectivity of the land-types defined by the classification (patches) are the most common elements found in landscape studies (Jenkins 1979, Turner et al. 1989, Hof and Joyce 1992, Hunsaker et al. 1994). Patch size and shape are also commonly reported, shape being represented by either area:perimeter ratios, fractal dimension of edge length, or shape diversity indexes (Ripple et al. 1991, Baker 1993, Mladenoff et al. 1993). Multi-scalar metrics such as fractal dimensions or variance across scales (semivariance) are becoming more common in landscape studies because of their ability to describe the relationship between patterns at different scales (Krummel et al. 1987, Pastor and Broschart 1990, Milne 1991, Turner et al. 1991). The amount and degree of contrast of forest edges is also commonly reported due to the concern over the effect of edges on wildlife species (Merriam and Wegner 1992, Hunsaker et al. 1994). Finally, Hulshoff (1995) proposes that the change of patch shape, size, or number may be useful to reflect landscape dynamics through time. 94 3.2 A LANDSCAPE MODEL FOR THE SBSmkl As the previous Section demonstrated, there is a wide variety of landscape models, each one responding to different needs. In this case, I needed a tool that was capable of comparing different forest disturbance scenarios to each other. A model was required because it allowed the expansion of quantitative information from a single landscape snapshot to multiple (possible) snapshots. In other words, it is the range of all possible patterns that is of interest under each scenario. The purpose reveals most of the necessary requirements for the model. For instance, this model should not only incorporate available information on SBSmkl landscape level processes, but should also capture the natural variability of these phenomena. According to both the available literature and the empirical evidence from the SBSmkl, the three most important aspects of the SBSmkl with respect to pattern formation are those that relate to the fire regime, or (in order of importance) the disturbance frequency, disturbance sizes, and disturbance shapes (Heinselman 1980, Gauthier et al. 1995). Other aspects of the disturbance regime, such as fine-scale fire behaviour tendencies (such as age invariance and the effect of creeks) and differential fire return intervals, will have less of an impact on landscape pattern. The exception to this is the differential fire regime found in Chapter 2 for those forests associated with the driest soil-type, which will be discussed later. This is consistent with the widely accepted hypothesis that most fire disturbances observed on the landscape today are a result of infrequent very large fires, which tend to be weather driven rather than fuel or topography-driven (Romme 1982, Turner and Romme 1994, Bessie and Johnson 1995). Therefore, priority was given to a modelling method which defines frequencies, sizes, and shapes of disturbances over the specifics of fire behaviour, 95 although every reasonable effort was made to approximate fire behaviour as well. The influence of vegetation dynamics is also an important process in the SBSmkl landscape, but is much more difficult to define. Essentially, the variability is so great there is little point in developing extensive vegetation dynamics modules without a great deal more information and research. In any case, this process can be considered part of the secondary layer of effects (along with fire behaviour) due to the overwhelming dominance of the process of disturbance. Therefore, I incorporated as much available information on the vegetation dynamics as I could, given that the fire regime information would take priority in the model. Additional requirements of the model were that it should be simple to understand, have an approach transferable to other landscapes, be PC-based, incorporate readily available sources of forest inventory and terrain data, and utilize software already developed where possible. No models were found that met all these requirements, so one was constructed. 3.2.1 MODEL AND DATA OVERVIEW The operation of the model is outlined in Figure 3.1. The model was written in I B M compatible Turbo C++ for maximum speed and compatibility, and uses a pixel-based spatial database in a binary format for high-speed access, simplicity, and flexibility. The main components of the model are 1) disturbance, 2) revegetation after disturbance, 3) "succession", and 4) the measurement of pattern. 96 LANDSCAPE MODEL FLOWCHART LARGE LANDSCAPE (790,000 hectares) END C A L C U L A T E 20-YEAR FIRE A R E A RANDOMLY SEED FIRE LOCATION no ASSIGN FIRE SIZE patch size analysis SMALL LANDSCAPE (28,000 hectares) M E A S U R E P A T T E R N (caH f r a g f f ) A G E L A N D S C A P E 20 Y E A R S edge contrast rules patch definitions ' growth and mortality rules ASSIGN POST-FIRE COMPOSITION] BURN FIRE FIND A R E A & LOCATION O F FIRE pre-burn species composition site-species analysis sexual maturity rules cellular automaton diffusion model FBP fuel-type ROS stand composition ROS slope correction from FBP 10m Digital Terrain Model fire edge analysis patch shape analysis and validation island size and frequency validation transfer from large landscape Figure 3.1. Flowchart of SBSmkl Landscape Model 97 The model operates at two scales. The first level represents the entire 790,000 hectare plateau, and operates at a relatively low level of resolution (6.4 hectare cells), since it tracks the area burnt as its only state variable (or cell attribute). The second level of the model represents a high resolution (50 m pixels, or 0.25 hectare cells) 28,730 hectare area physically contained within the larger area. This spatial dataset (described fully in Section 2.3) is the area in which disturbance and vegetation modelling will take place, and from which pattern measurements will be taken. The data were extracted from the GIS system P A M A P to allow more flexible access and storage of the datafile (PAMAP 1987). The justification for using a 50 m pixel size is the extraordinary level of detail of the D T M provided by Lakeland Mills Ltd. of Prince George. The contours for the Lakeland D T M were generated by photogrammetric methods based on 10 m contours. In contrast, the Provincial terrain model is produced from satellite imagery based on 30 m contours, and the Ontario D E M (Digital Elevation Model) recommends a scale of resolution of 100-500 m (Mackey et al. 1994). Moving to a 100 m resolution in this case would probably not affect the forest cover information significantly, but would result in considerable loss of useful terrain (slope and aspect) detail. The 50 m pixel was considered a reasonable size with which to consider fire behaviour (B. Hawkes, pers. com.). Creating a pixel size smaller than 50 m was possible, but unnecessary for my purposes since the model will be unable to take advantage of any more detail. In any case, using smaller pixel sizes would increase the file size substantially. For instance, using a 25 m pixel would result in 460,000 records, which translates to a file size of approximately 14 megabytes (MB) in binary format, and 23 M B in ASCII format, compared to 115,000 records (3.6 M B and 8 M B respectively) for the 50 m pixel file. 98 In the small detailed area, age-class and stand composition are tracked as state variables, and permanent information on slope, aspect, soils, and non-forest cover is maintained. The two-scale system eliminates the problem of edge effect since disturbances spread across all 790,000 hectares, but pattern will only be assessed on the small landscape located physically within the larger area. Also, it will eliminate the need for interpolation of the distribution regime information since it was derived for the 790,000 hectare area. In other words, the disturbance stochasticity, from a spatial point of view as well as a temporal one, will be realistic in the small 28,730 hectare area. 3.2.2 DISTURBANCE RATES The model begins with a fire frequency level (percent disturbed) for a 20 year time-step (Figure 3.1). I have already determined that the landscape is far from being "stable" in that the disturbance rates vary tremendously, even grouped into 20-year classes. It is important to capture this variability in the model, so a function describing the first eight estimates of 20-year disturbance rates from Section 2.4.3 was derived (using the results from the 100 year fire cycle). However, before doing this, the 20-year disturbance rates were adjusted to reflect average annual disturbance rates. The 20-year rate of burning is the total area consumed at the end of 20-years. This area would be identical to the total area actually burnt during the 20 year period only if none of the areas burnt more than once during the 20 years. This is unrealistic; I found physical evidence that frequent reburns can and do occur. Assuming reburns means that the areas burnt after each 20 year cycle will be underestimated if the 20-year disturbance rates estimated in Section 2.4.3 are used to derive a disturbance frequency function. 99 The simplest way to adjust these data was to assume that an area was available for burning immediately following a fire. To account for this, each of the burning rate estimates was adjusted upwards by that same probability. For example, the first rate of 7.4% per 20 years became 7.4 + (7.4 * 0.074) = 8.0% per 20-years. Note that this allowed for only one reburn in a 20-year period. It is possible for three or more fires to occur in a single area over a 20-year period, but not nearly as likely, and their effect on the average annual rate would be minimal. The original, and adjusted 20-year disturbance rates are listed in Table 3.1. Table 3.1. Original and adjusted 20-year disturbance rates for the SBSmkl. ORIGINAL 20-YEAR RATES (%) ADJUSTED 20-YEAR RATES (%) 7.4 8.0 23.6 29.2 9.8 10.7 14.4 16.5 21.9 26.7 . 40.6 57.1 36.0 48.9 19.9 23.8 A V E R A G E 27.6 STANDARD DEVIATION 16.3 The values in the second column of Table 3.1 are estimates of 20-year burning rates (adjusted for reburns) which had to be described as a function for the model. Having only eight points with which to fit a function was problematic, so I compared several methods. I started by assuming that the data were distributed normally, and developed a normal curve from the raw data. Figure 3.2 shows the plot of the actual disturbance rate data, plotted cumulatively 100 against a normal cumulative function derived using the sample mean and variance in the normal equation of the form: (y) = (p-x)2 2a2 where (y) = percent burned/20 yrs, x (mean) = 27.6, a (standard deviation) = 16.3 (based on the 100 year fire cycle estimate of the adjusted data in Table 3.1) and p is the cumulative probability from 0 to 1. 0 10 20 30 40 50 60 PERCENT BURNED / 20 YEARS • ACTUAL DATA NORMAL FUNCTION LINEAR FUNCTION ROOT FUNCTION Figure 3.2. SBSmkl cumulative 20-year disturbance rate probability based on a 100-year fire return interval, adjusted to allow for reburning. 101 Another possible approach was to assume that disturbance rate is a linear function. The least squares solution of a linear model is also shown in Figure 3.2, and takes the form: With and R 2 = 0.93 and an SEE = 0.092. Given the low number of data points, it was difficult to argue against either the normal or linear models. However, another way of judging them was to consider the behaviour of both models at the extremes. Both the normal and linear model share an unfortunate characteristic of intersecting the y-axis rather than the x-axis. This translates into a possibility (approximately 4 - 1 0 %) that zero hectares will be disturbed in any given 20-year period. Based on our current knowledge of boreal forest fire behaviour, this seems highly unlikely; it is more reasonable to expect the disturbance rates to intersect the x-axis. Considering the shape of a distribution that would meet this requirement, and still go through the data points, a root function was fit using the nonlinear regression module of SYSTAT (Wilkinson 1988). The following function was created after 66 iterations: where p = probability from 0 to 1. Actually, any fractional root function would have intersected the x-axis in a similar way; however the function noted above to the power of 5.62, aside from minimizing the non-linear loss function, also allowed the area disturbed to be slightly greater than the largest area estimated on the SBSmkl (Figure 3.2). Based on these arguments, this root function was adopted for stochastically approximating area burnt for each 20-year period in the model. Pet burned per 20 yrs = p- 0.0967 0.0169 5.62 Pet burned per 20yrs 102 3.2.3 FIRE SIZES Once the area to be burnt was determined for each 20-year period, the sizes of the individual disturbances had to be chosen (Figure 3.1). Ideally, a function could have been derived from the distribution of historical disturbance size data. However, these data were not available, and it is not the sort of information that is advisable to borrow from other landscapes. Instead, I relied on the patch size data analysis of the large database done in Section 2.4.6. However, as discussed at length in Section 2.4.6, there are doubts concerning the integrity of the small patch data which complicate estimating a disturbance frequency distribution. I considered the most serious problem to be the bias associated with the lack of smaller patches in older age-classes. The best estimate of disturbance sizes would then be the patch sizes of the youngest age-class since it is the least likely to be "missing" small patches. However, I was uncomfortable with these data for two reasons. First, the sample size was relatively small, and second, I was concerned about the accuracy of the patch size data of small patches overall (considering the potential sources of error in both GIS manipulation and the consistency of the delineation of small patches on the paper maps). I decided that a better estimate of the actual size distribution of disturbances was probably that hypothesized by the simulation exercise, adjusting for island remnant patches in age-class 2 (the "predicted from model" data given in Figure 2.8). In the end, this distribution was similar to that of age-class 1. In order to define a function for disturbance sizes for the simulation model, the adjusted data were grouped into logarithmic area classes, and the cumulative data used to fit a three-term Weibull function using the SYSTAT NONLIN module (Wilkinson 1988). The final model was: 103 _ JQ(O.419+0.852 L251y/-[n(l-p)) where a = log(area) burnt in 20 years, and p = probability from 0 to 1. The frequency, and cumulative form of the function are shown in Figures 3.3 and 3.4 respectively, along with the raw data. D I S T U R B A N C E S I Z E (log(ha)) X R A W D A T A F U N C T I O N Figure 3.3 Disturbance size function for the SBSmkl based on estimated number of disturbance patches from age-class 2 (based on the simulation exercise in Section 2.4.6). These individual disturbance patches were then randomly located as rectangles within the larger landscape. If the disturbance did not cross into the smaller model area, nothing more was done and the process of choosing fire sizes and locations continued. Once the 20-year target was reached, disturbance events within the 20 year time-step were complete. If a fire either crossed into, or ignited within the nested 28,730 hectare area, the fire behaviour model took over (Figure 3.1) 104 QgXXX-2 3 D I S T U R B A N C E S I Z E (log(ha)) R A W D A T A F U N C T I O N Figure 3.4. Cumulative disturbance size function for the SBSmkl using estimated disturbance patches from age-class 2 based on the simulation exercise in Section 2.4.6. 3.2.4 FIRE BEHAVIOUR Although several excellent fire behaviour models were available (e.g. McRea 1990, Green et al. 1990, Knight and Coleman 1993, Finney 1994), most required detailed fuel, topography, and weather information, and involved complex spatial formulations. Furthermore, none of them account for the formation of island remnants within the fire boundaries. These models are considered to be the best at representing fire behaviour. Although individual fire behaviour accuracy per se, was not a high priority for this model, I wanted to account for known attributes of fire behaviour, as well as the fire regime information. In other words, it was important that fire behaviour tendencies, over many years and large areas, were well represented. I chose to base my fire behaviour module on the cellular automaton approach of Clarke et al. 105 (1994) described in Section 3.1.2. This method offered several advantages specific to my needs. First, fire sizes could be controlled by simply stopping the fire after the appropriate number of pixels had been consumed by the fire. Second, since it was stochastic, fires started in the same place always behave differently. Third, the inclusion of fuel and topographic information meant that the tendency of fires to burn under different conditions (uphill vs downhill; conifer vs hardwood) could be considered. Finally, shape could be controlled by manipulating the parameters controlling the length of individual firelet trails, the probability of a firelet spawning a new ignition source, and the probability of a firelet going "out". I included three overlays for the weighting of the pixels: fuel-type, topography, and creeks. The fuel-type weighting assignment was based on the rate-of-spread (ROS) figures given for the Canadian Forest Fire Behaviour Prediction (FBP) System fuel-type designations (Taylor and Pike 1995). An Initial Spread Index (ISI) of 15, and Build Up Index (BUI) of 50 were used in the tables to represent "average" fire weather conditions in the Prince George area (Forestry Canada Fire Danger Group 1992). Age and composition information in each pixel were used to assign broad fuel-types closely matching those from the FBP system. Where gaps occurred, I used interpolation (between two existing forest fuel-types) or the closest possible category in the case of non-forested classes, such as non-productive brush and non-productive black spruce. There is currently no mechanism that considers increasing rates of spread in older stands within the FBP fuel-type ROS's that form the basis of how and where fires will spread across the landscape. In other words, the FBP system assumes age invariance. Although the results of 106 Section 2.4.1 were ambiguous, there is reason to believe that a limited degree of age selection is in operation on the SBSmkl landscape13. To allow for age selection, the model adopted the FBP ROS values as they exist below 140 years, and increased them by 50% thereafter. This underplays the tendency of age selection according to the disturbance rates for the Weibull in Table 2.1 in Section 2.4.1, which is reasonable considering the evidence on which the adoption of the Weibull model was made. It also allowed for the possibility of more frequent reburning (which has been observed). The final fourteen fuel-types and their respective ROS values used in the model are given in Table 3.2. The ROS weights given in Table 3.2 are valid only for flat ground. Slope and aspect information combined to form the second data overlay. From the target pixel, the direction (uphill, downhill, or sideslope) was found, and used in conjunction with the slope to adjust the ROS accordingly. The adjustments were taken directly from the Relative Spread factors used for the FBP system (Alexander et al. 1984). The factors by which the fuel-type ROS's were adjusted are given in Table 3.3. The ROS multipliers were applied directly when the fire is moving uphill, and inversely when moving downhill. Note that for slopes less than 10%, no correction was applied. Although there was some indication that south and west facing slopes may be more susceptible to fire than north and east slopes (Section, the evidence was not considered strong enough to warrant complicating the model further. Furthermore, although there was fairly strong evidence that forest stands on very dry, well drained soils burn more often than stands on other 13 As stated in Chapter 2 of this thesis, the observed age selection may not be due, in whole or part, to fire, but nonetheless exists as a landscape feature that I have chosen to incorporate. 107 Table 3.2. Fuel-types and rate-of-spread (ROS) values for the landscape model. FUEL-TYPE (age-class) ROS (m/min) Water 5 Young mixedwood (<= 120 years) 6 * Non-productive brush 7 Non-productive black spruce 8 Mature hardwood (> 120 years) 9 Young mixedwood (<= 120 years) 12 * Immature pine (41 - 120 years) 14 * Mature mixedwood (> 120 years) 17 Young softwood (<= 120 years) 18 * Young spruce (<= 120 years) 20 * Mature pine (> 120 years) 21 Young pine (<= 40 years) 22 * Mature softwood (> 120 years) 27 Mature spruce (> 120 years) 30 * values derived directly from Tay or and Pike (1995). Table 3.3. Rate-of-spread (ROS) slope correction factors for the landscape model fuel-types. SLOPE RANGE ROS MULTIPLIER 0-9% 1.0 10-19% 1.4 20-39% 2.3 > 40% 6.5 108 soils (Section, none of these soils were found within the small 28,730 hectare area. No adjustment for multiple fire regimes was made. The only other ROS adjustment necessary was for water. The model mechanics were such that for pixels that are classified as water, fires will always stop at this edge unless they are assigned a positive ROS value. Forest fires can jump water bodies quite easily in some cases, but, as analysis showed, even very small creeks can influence edge formation (Section For those pixels that are classified as water larger than a creek, an arbitrary ROS value of five was assigned. Subsequent model testing revealed that fires could cross even the largest water bodies using this value, but they had to be persistent to do so. If a fire could not "jump" water, it could always burn around it i f the fire was large enough. For smaller water bodies such as creeks, a different solution was developed. Section showed that most fires burn across creeks, although creeks can influence fire behaviour. Based on this conclusion, the ROS of a forest pixel with a creek running through it was multiplied by an arbitrary factor of 0.6. Subsequent model testing showed that this had little influence on the formation of edges in most cases. The resulting ROS from these three overlays was used as the weight for each individual cell involved in each fire spread decision. Although this model formulation is unique, the concept of combining fuel and slope information to drive fires through a cellular automaton process is not new, and is used in IGNITE (Green et al 1990), P R E P L A N (McRae 1990), FARSITE (Finney 1994), FIREMAP (Ball and Guertin 1992), and LANDIS (Mladenoff et al. 1993). 109 3.2.5 ESTABLISHMENT Once a fire finished burning, the area burnt must be revegetated in a reasonable fashion (Figure 3.1). This was accomplished by considering what were thought to be the two main processes: the tendency of tree species to occupy certain site-types, and the variable effects of disturbance on those tendencies. In other words, if it is assumed that the only two processes that are responsible for species combinations occurring on a given site are site-species associations and disturbance effects, this could be used to create a stochastic model of regeneration. For instance, the averages and standard deviations of each of the four main SBSmkl tree species on 25 soil X slope site-types from Section 2.4.8 offers a reasonable summary of the both the tendencies and variability of the species on each site-type. These tendencies are also more or less in agreement with what is known of the site preferences for each species (Section 2.1.4). However, there is no way of confirming the relative influence of the mechanism(s) by which species shifts occur. For instance, it has been hypothesized that bark beetle infestations may be one of the main mechanisms causing species shifts away from the normative preferences (J. Carlson1 4, pers. comm.). Considering the wide range of site preferences for the species involved, and the lack of empirical evidence with respect to species shifts, I decided to take a very conservative approach to revegetating after fire. For the most part, the existing species combinations were assumed to maintain themselves with the following exceptions. When an area was reburnt within a 40-year period, based on the late sexual maturity of both white spruce (30-60 years) and engelmann spruce (16-25 years) (Fowells 1965), the amount of hybrid spruce declined sharply. If more than 1 4 Phero Tech, Surrey, British Columbia 110 one fire occurred within a 20 year period, the percentage of hybrid spruce was assumed to decline by 80% of the original percentage for each fire after the first. For fires that occurred between 21-40 years after the original fire, the percentage of hybrid spruce was reduced by 50% of the original value. In almost all cases, this resulted in an increase in the lodgepole pine present. 3.2.6 SUCCESSION Succession occurred only once per 20-year period. For those pixels that do not burn, it was necessary to move them forward in time. Generally, the initial floristics model of forest succession (sensu Egler 1954) was assumed in the model, meaning that once the establishment composition was estimated, it remained constant until disturbed again. The simplest way of doing this was to advance the age-class by one and keep the same species percentages. The exceptions to this were mortality rules specific to pine and poplar species based on their shorter lifespans relative to black and white spruce (Fowells 1965, Parminter 1983). For any pixel greater than 160 years, the percentage of pine was reduced by 35% (of the current percentage of pine), and for pixels greater than 120 years, the percentage of poplar species was reduced by 35%. 3.2.7 PATTERN ASSESSMENT At the end of each 20-year period, an ASCII file compatible with the FRAGSTATS spatial pattern analysis software package (McGarigal and Marks 1994) was created. Ideally, different definitions of patches could have been used, but for the purposes of this model, only the simplest possible forms were considered. This coarse grouping, based on the classification of Payne and Bryant (1994), recognized only five patch types: water, non-forested, young forest (0-40 years), pole-sized forest (41-120 years), and mature to overmature forest (greater than 120 years). I l l The metrics that were used to collectively describe pattern are: - proportions, changes in proportions, and evenness (diversity) of patch types, - interior forest, or core areas, - edge density, - the average and standard deviation of patch size. In recognition of the changing relationship between adjacent patches over time, matrices describing the contrast between patch types were defined and used to create a metric describing the equivalent edge length. This essentially translated the edge lengths and contrasts into the equivalent length of high or full contrast edge. Contrast can be defined between 0 (no difference) and 1.0 (full contrast) in FRAGSTATS (McGarigal and Marks 1994). If no contrast matrix is defined for patch types, a contrast of 1.0 is assumed for all patches. The definitions of patch contrasts used for the SBSmkl are based on those suggested by Payne and Bryant (1994), who simply defined "high", "medium", and "low" contrast edges. These are shown in Table 3.4. I translated the values of 1.0 to represent high edge contrast, 0.6 medium contrast, and 0.2 low contrast. I also chose (within FRAGSTATS) not to count the map border as an edge, allowed diagonals to be used when compiling patches, and used a 100 m buffer to define the interior of a patch area. Table 3.4. Edge contrast values for patch definition for the SBSmkl . WATER NON-FOR. Y O U N G POLE M A T U R E WATER 0 1 1 1 1 NON-FORESTED 0 0.6 1 1 Y O U N G 0 0.6 1 POLE 0 0.2 M A T U R E 0 112 3.3 MODEL VALIDATION The model will not be used as a predictive tool, but rather as a means of assessing the relative importance and sensitivity of selected disturbance regime parameters on landscape pattern. The ability of the model to capture the "natural" landscape processes accurately is therefore secondary to its ability to use the best available data in a logical manner consistent with known facts. For instance, assuming fire is in fact the dominant agent of landscape change on this landscape, it is known that fire sizes and frequencies will vary spatially and temporally. I did my best to represent this variability, and in the end, fire sizes and frequencies were valid insofar as the equations that were used to estimate them were either best-fit regression or non-linear estimates of the "adjusted" patch size data from the landscape. However, regardless of what biases or inaccuracies remain in these estimates, comparisons of model output will be relative, and thus all simulations will contain identical levels of bias or inaccuracy. Therefore, although it would have been interesting to know how inaccurate or biased these estimates were, it was not necessary to have this information to conduct the simulation tests. Neither the revegetation, nor the succession components could be validated, although the ideas are supported by literature (Fowells 1965, Parminter 1983, Bradley et al. 1992, Payette 1993, Clark 1994). In any case, the revegetation and the successional components of the model only influence the fuel-type category of pixels, and thus had little or no influence on the resulting landscape patterns, unless dramatic changes in the proportion of hardwood species occurred. Although it could not be confirmed that individual fire behaviour was accurately represented, it was only necessary that the resultant spatial patterns from repeated fires was accurately depicted. 113 In this case, both fire shape and the number and sizes of unburnt remnant islands produced from the model could be compared to empirical data. 3.3.1 DISTURBANCE SHAPES As shown in Figure 3.5, fire shape becomes increasingly complex as fire size increases. Since the raw data counted the perimeter of interior islands as part of the fire perimeter (J. Rustad15 pers. comm.), at least part of the reason for this increased complexity is the occurrence of island remnants. Since the fire model created island remnants, a comparison between the two was possible. Shape could be manipulated in the model by adjusting probabilities for three parameters: the number of times a fire source starts a fire trail, the length of the trail, and whether or not it spawns one or more additional fire sources. Calibrating the parameters was a manual exercise. The testing took place on the actual landscape file, since it included other landscape features that could affect disturbance shape such as non-forested land, creeks, and water bodies. One hundred and fifteen fires between ten and 3,000 hectares were simulated with the model. Above 3,000 hectares, almost all fires ran off the edge of the test landscape. I did not bother simulating many fires less than 100 hectares since disturbances below this size are relatively constant and simple in shape, and generally do not produce remnant islands (Eberhart and Woodard 1987, DeLong and Tanner 1995). Every island two hectares or greater (the minimum patch size in inventory records) was recorded for each disturbance event. The total perimeter and an area:perimeter ratio, or shape index, was calculated as: 1 5 GIS operator, Timberline Forestry Consultants, Prince George, BC 114 SHAPE- P E R I M -*AREA and recorded. The raw data were edited to the same ten to 3,000 hectare limit. Both sets of data are shown in Figure 3.5. For each data set, the following equation was fit; SHAPE =BQ+BXAREA 4 The equation for the raw data sets is as follows: SHAPE = 1.770 + 0.041 log(AREA)4 where R 2 = 0.670, SEE - 0.585, n = 928, sd of b 0 - 0.022, and sd of b, = 0.001. For the simulated data, the equation is: SHAPE = 1.881+0.041 log(AREA)4 where R 2 = 0.923, SEE = 0.506, n = 113, sd of b 0 = 0.089, and sd of b, = 0.001. Since the slopes of the two equations were identical, there was no need to test for significant differences; the null hypothesis that there is no significant difference between the two slopes cannot be rejected. Furthermore, the intercepts were nearly identical, and the difference between them is very small compared to the standard deviation. Based on these results, I concluded that the current fire simulation parameters approximated fire shape to an acceptable degree. The lines defined by the functions virtually fall on top on one another (Figure 3.5). 115 1 1.5 2 2.5 3 3.5 log [AREA(ha)] X RAW DATA # MODEL OUTPUT FUNCTION Figure 3.5. A comparison of actual to simulated data for disturbance shapes on the SBSmkl landscape. 3.3.2 REMNANT ISLAND FORMATION As previously mentioned, the formation of remnant islands is a by-product of this particular fire spread method. Although very little information was available on the number and size of islands that natural disturbances leave, DeLong and Tanner (1995) used the same SBSmkl dataset to take a detailed look at nine disturbances between 159 and 2,239 hectares. Comparison to these data served as the only means of judging the ability of the fire model to create island remnants. DeLong and Tanner (1995) found island remnants between 1 and 73 hectares in size and determined that the total area of island remnants equalled between 3-15% of the total area 116 consumed by fire. The test data from the model created individual islands between 2 and 97 hectares. The total area of island remnants ranged between zero and 21% of the area consumed by fire. The larger range of both sizes and area occupied by islands may be accounted for by the larger range of test data used (10 - 3,000 ha compared to 159 - 2,239 hectares). Data were also available on the areas of islands, in percent, by disturbance size-class (Table 3.5). The simulated data compared favourably to the actual data. Again, the increased range of the model output may be accounted for by the wider range of the test data. Table 3.5 A comparison of actual to simulated percent of fire area in remnant islands by disturbance size-class in the SBSmkl landscape. DISTURBANCE SIZE (ha) PERCENT OF TOTAL FIRE AREA (%) DeLong and Tanner Model data < 500 hectares 4.0 3.3 500 -1,000 hectares 6.0 6.2 > 1,000 hectares 9.0 10.5 Another comparison was made between the proportion of island areas of different sizes within disturbances (Table 3.6). These data were more difficult to interpret because the model specifically eliminated all islands less than two hectares in size, while DeLong and Tanner (1995) counted islands of all sizes. If the islands less than two hectares were ignored in the data from the SBSmkl (in brackets in Table 3.6), the comparison was more meaningful. Compared to the actual data from DeLong and Tanner (1995), the model overestimates the sizes 117 of islands to a moderate degree in disturbances less than 1,000 hectares. The model creates a large portion of islands between 10 and 50 hectares in particular (38%) that were not found by DeLong and Tanner (1995). In contrast, the proportion of different sized islands in disturbances greater than 1,000 hectares from the model compared favourably with empirical data (Table 3.6). Table 3.6 Comparison of actual to simulated areas in island remnants by size-class for disturbances on the SBSmkl landscape. ISLAND PERCENT AREA IN ISLAND REMNANTS SIZE DISTURBANCES <1,000 ha DISTURBANCES >1,000 ha (ha) DeLong & Tanner Model data DeLong & Tanner Model data < 2 ha 49 - 17 -2-5 ha 32 (65) 33 20 (24) 19 5-10 ha 17(35) 25 11(13) 17 10-20 ha 0 27 17 (20) 19 20-50 ha 0 11 27 (35) 23 >50 ha 0 5 8 (10) 22 Original percentage of area in island remnants (percentage of area ignoring data < 2 hectares) Overall, considering the difficulty in approximating island sizes and numbers in the model, the low sample size used (nine) in the comparison by DeLong and Tanner (1995), and the difference in the test range, the model did an acceptable job of creating realistic numbers and sizes of island remnants within fires. 3.4 SUMMARY This Chapter outlined the development and testing of a landscape computer model for the SBSmkl . A model was required that had the following characteristics: 118 - utilizes readily available data and software where possible, - spatially explicit, - stochastic, - captures the natural range of disturbance sizes, rates, and tendencies, and - creates a realistic range of disturbance shapes, including the formation of remnant islands. No existing models were found that met all of these requirements. Therefore, I constructed a series of computer programs using Turbo C++ which created summary landscape mosaic datafiles in 20-year time-steps compatible with the FRAGSTATS spatial pattern analysis program. I used a two-scale, nested dataset to allow for edge effect and the natural spatial stochasticity associated with a Poisson distribution of fire starts. The larger landscape was driven by disturbance size and frequency functions that were defined for the SBSmkl based on the analysis in Chapter 2. Only on the smaller landscape do the fires actually spread. The cellular automaton approach of Clarke et al. (1994) was used as the basis for fire modelling since it allowed for the manipulation of shape and remnant island formation, as well as allowing for the inclusion of pixel-based stochastic burning rules (i.e., every fire will be different). Each 50 m pixel contained information on vegetation summarized as FBP fuel-types and topography. Consistent with the findings from Chapter 2, older stands were allowed to burn more readily than younger stands, and those pixels with creeks or rivers passing through them were not allowed to burn as readily as other pixels. Comparisons of model output to empirical data on fire shapes and island remnant sizes and numbers were acceptable. It was not possible to verify other aspects of the model, other than 119 through checks of logical consistency with known facts. FRAGSTATS was used to summarize patch diversity, contagion, edge density, interior forest area, the average and standard deviation of patch size, and the change over time of the average patch size according to a five class system of patches (water, non-forested, and three ages of forest). These outputs from the model are used in Chapter 4 to investigate the sensitivity of landscape pattern to certain disturbance parameters that cultural management is most likely to affect. 120 CHAPTER 4 - SIMULATION MODELLING IN THE SBSmkl In this chapter the simulation model described in Chapter 3 is used to investigate the relationship between observed landscape patterns, and various aspects of the disturbance regime for the SBSmkl. 4.1 LITERATURE REVIEW Our experience is minimal in both analysing landscape patterns and in modelling disturbance regime behaviour. Landscape pattern analysis has been largely limited to comparisons of two adjacent landscapes, or one landscape at two points in time. Modelling of landscape-level process has been limited to simple theoretical models. Furthermore, studies of the nature of the relationship between landscape metrics, and the response of species or some other ecological function, are rare. This section will briefly review the relevant literature on these topics. 4.1.1 LANDSCAPE PATTERN MEASUREMENT To many, the primary focus of landscape ecology is the description of spatial "pattern", which is essentially a broad-scaled inventory of patches (Gardner and Turner 1991). The potential for describing landscape pattern is tremendous now that technology allows large quantities of high resolution data over extensive areas to be gathered (Gardner and Turner 1991). However, methodologies for summarizing and analysing these data in meaningful and understandable ways are still being developed and tested. This is of particular concern to landscape studies with the sole objective of simply describing landscape pattern. 121 For the most part, landscape ecology has employed existing spatial pattern analysis methods. Unfortunately, generic pattern metrics tend to be both plentiful and complex. For instance, FRAGSTATS, a spatial pattern analysis program for quantifying landscape structure (McGarigal and Marks 1994) calculates over 50 spatial pattern metrics from a single landscape image. This is problematic for non-specific exploratory or descriptive landscape studies in terms of which metrics best describe landscape pattern. On the other hand, many pattern metrics are highly correlated. For example, Riitters et al. (1995) condensed 55 metrics of landscape pattern and structure down to just six through pairwise tests and factor analysis. The problem of which metrics to use in order to best capture pattern now becomes less of an issue than which metrics are the easiest to understand. The tendency of descriptive studies of landscape pattern has been to use the most intuitive measures such as edge density, patch size range, patch density, adjacency, nearest neighbours, and diversity (Jenkins 1979, Turner et al. 1989, Hof and Joyce 1992, Hunsaker et al. 1994). Most such comparative studies have come to similar conclusions. Generally, as landscapes become more developed and less "natural", forest patches become simpler, more abundant and smaller, interior forest area decreases, edge density increases, distance between similar patch-types increases, and contagion decreases (Ripple et al. 1992, Mladenoff et al. 1993, Luque et al. 1994, Wickham and Norton 1994). Pattern also has a temporal dimension which must be considered together with the spatial aspect. Landscapes are widely recognized as being highly dynamic entities, and thus landscape research is also largely concerned with change (Forman and Godron 1986, Risser 1987, Dunn et al. 1990, Noss 1990, Forman 1995). For instance, Romme and Knight (1982) determined that dramatic 122 changes over large areas were a common occurrence in Yellowstone Park, and hypothesize that patterns of change over 100 km 2 areas occurred on a cyclical basis. It is the change in spatial pattern which is the focus of landscape studies concerned with ecological impacts. Changes in landscape spatial pattern are thought to affect wildlife, water and nutrient flows (Romme and Knight 1982), and even disturbance processes. The effects of landscape patterns on disturbance processes were discussed in Chapter 2. However, most of the empirical research, and virtually all of the experimental work concentrates on the impacts of landscape patterns on wildlife and plant species. The range of research extends quite wide. For example, Holt et al. (1995) artificially fragmented a large abandoned agricultural site to study long-term effects of fragmentation on plants. After only six years, they found that larger patches were more species rich, and that clonal and transient species were more likely to disappear in the smaller patches (Holt et al. 1995). Van Apeldoorn (1994) found that the red squirrel (Sciurus vulgaris L.) in Europe responded to the size and distance between patches of suitable habitat, as well as to corridors. On a different scale, Wallace et al. (1995) found that elk (Alces alces) and bison (Bison bison) feeding patterns in Yellowstone Park were more related to broad-scale choices of forage quantity than to finer-scaled factors such as forage quality, as had been previously thought. With respect to boreal landscape studies, there is some concern that the disappearance of deciduous pockets in Sweden's boreal forest due to management activities is having a negative impact on birds. Enoksson et al. (1995) studied the impacts of isolation of deciduous habitat on six bird species. They found that increasing isolation of the deciduous patches had a negative 123 impact on densities for two species, a positive impact for one species, and no impact on the remaining three species. Similarly, Hansson (1994) found mixed responses among several species of large and small vertebrates and several bird species to clear-cuts and edges in central Sweden's boreal forest. Several species responded positively to edges and openings, a few responded negatively, but most species remained unaffected. He hypothesized that the relative (evolutionary) youth of the boreal forest has not yet allowed the evolution of specialists, which have narrow habitat tolerance ranges. Rather, the boreal forest may be dominated by highly adaptable generalists (Hansson 1994). He found it difficult to label species as either "edge" or "interior" in contrast to non-boreal studies. For instance, Mclntyre (1995) found a very strong response of edge and interior bird species to patch sizes in forest patches in the Georgia Piedmont. These sorts of landscape studies are becoming more common (than simple pattern description) as the need for linkages between patterns and responses increases. However, the danger in relying on species response data is that short-term responses may not indicate longer-term thresholds. For instance, it has been suggested that unoccupied areas of otherwise suitable habitat within a landscape act as habitat alternatives (Noss 1994). If this is true, measured population levels of some species may not appreciably change as habitat is reduced until either a critical threshold is reached (where the alternate habitat have all been eliminated), or some other disturbance takes place (eliminating a large part of the currently used habitat). Although there is some evidence to suggest that exceeding natural levels of pattern change leads to the loss of biological values, our understanding of exactly where such "natural boundaries" are, and how sensitive species are to changes within those boundaries, is limited (Dunn et al. 1990, Noss 1990b). The point is that reliance on species response data alone may lead to the elimination of 124 the resilience of a species with respect to disturbance response. Ultimately, this may be the most important role of landscape ecosystems. 4.1.2 LANDSCAPES AND SIMULATION Describing landscape patterns and measuring species responses to pattern changes are only two aspects of the study of landscapes. Ultimately, landscape dynamics can only be fully understood by focusing on the processes that are responsible for the observed spatial patterns. The shift from studying pattern to process is an important one, but our ability to observe these processes over extended periods is limited. In light of these difficulties, simulation modelling has become prevalent. Computer models are used to better understand how manipulating a small number of landscape processes manifests itself over time and space - as pattern. Landform development, hydrological cycles, species migrations, speciation and gene flow, predation, and disturbance are collectively responsible for the dynamic range of patterns observed (Noss 1994), but to date most landscape simulation models deal only with disturbance. The first landscape dynamics models were very simple, involving only one or two parameters. The elegant simplicity of such models provides strength and clarity. For instance, so-called "neutral" percolation models have been valuable for understanding the spread of entities (disturbances, organisms, nutrients etc.) through space and time. These models track the spread of an entity across a landscape occupied by cells that are either available or unavailable for the entity to move through. By using different frequencies and spatial arrangements of the available cells, threshold levels of available cell frequencies were found, below which the entity could not move freely across the landscape (O'Neill 1988, Gardner and O'Neill 1991). These models will no doubt 125 continue to be important in understanding the spread of disturbances, and movement of organisms across landscapes. Some simulation models deal solely with the effect of changing disturbance regime parameters on the resulting pattern. L i et al. (1993) used simulation to compare patterns of five different harvesting strategies on indices of fragmentation (edge, shape, relative patchiness, and interior fragmentation index) concentrating on cutblock spacing (dispersed to clustered) and size and shape. Predictably, they found that the maximum dispersion scenario created higher levels of fragmentation as cutting frequency increased. They also found that larger cuts and the presence of uncut reserves resulted in less fragmentation, roads always increased fragmentation, and stream corridors could both increase and decrease fragmentation (Li et al. 1993). Other simulation efforts went further to include predicted species responses. Gustafson and Crow (1994) simulated several management scenarios on a landscape to consider the effect of parasitism by brown-headed cowbirds (Molothrus ater) on bird nests near clear-cut edges. They found that smaller cuts created increased edge, but noted a non-linear response of vulnerability of nests to parasitism. Although proposals for similar simulation methods to that developed in this thesis were found, (e.g. Gauthier et al. 1995) no results were available. 126 4.2 PURPOSE As discussed in the previous section, the first step to understanding landscape patterns is to understand and describe the range of natural processes at the landscape level for that particular landscape. This has been done for the SBSmkl, as described in Chapter 3. The purpose of the simulation exercise is to take the next step and explore the relative influence of different components of the disturbance regime on various aspects of pattern. The lack of related research means that this analysis will largely be exploratory. The disturbance parameters that were modified for the simulations were those most likely to be influenced by forest management. These were: 1) sizes of disturbances; 2) the temporal variation in frequency of disturbance; and 3) the eligibility of stands (since fires burn young stands, but we generally do not harvest them). The simulation exercise that follows will test the response of landscape pattern to constraining these three aspects of the disturbance regime in the model. 4.3 SIMULATION 4.3.1 METHODS The landscape model described in Chapter 3 was used to create 60 landscape scenes in 20-year intervals under 36 different disturbance scenarios (Table 4.1). There were essentially two types of restrictions imposed: disturbance size restrictions, and eligibility restrictions. The size restrictions simply control the size of allowable disturbance. Both upper and lower size restrictions were imposed; for instance, "0-10,000" or "10,000" means no lower limit was imposed, but individual disturbances larger than 10,000 hectares were not allowed. "20-3,000" 127 means a lower limit of 20 hectares was imposed, and no individual disturbances larger than 3,000 hectares were allowed (Table 4.1). "NONE" means that the original, "natural" disturbance size distribution function derived in Chapter 3 was used, and "60 only" means that a constant disturbance size of 60 hectares was set. Eligibility restrictions controlled both the rate of disturbance and availability of pixels to a spreading disturbance. I have lumped them together for reference purposes. Table 4.1 Summary of simulated disturbance scenarios for the SBSmkl landscape. SIZE RESTRICT, (hectares) ELIGIBILITY RESTRICTIONS NONE A V E R A G E RATE OF DIST. (21%) ADJUSTED A V E R A G E RATE OF DISTURBANCE (18%) > 20 YRS ONLY > 40 YRS ONLY > 80 YRS ONLY > 80 YRS + BUFFER NONE X X X X X X 0 - 10,000 X X X X X X 0 - 3,000 X X X X X X 0 - 1,000 X X X X X X 60 only X X X X X X 20 - 10,000 X X - - - -20 - 3,000 X X - - - -20 - 1,000 X X - - - -The eligibility restrictions were meant to be incremental. No eligibility restrictions ("NONE") means that all fuel-types were available for disturbance, and the full range of 20-year disturbance rates described in Chapter 3 was imposed. The first eligibility restriction limited the 20-year disturbance rate to the average rate (of approximately 21%), but otherwise allowed disturbances 128 to move freely across the landscape. The remaining four eligibility restrictions confined the spread of disturbances to increasingly older age-classes. The disturbance rate for these scenarios therefore had to be adjusted downwards to account for the fact that disturbances were no longer allowed to overlap one another within the first 20-year period. The adjusted disturbance rate was approximately 18%. The ">20 yrs" restriction did not allow multiple disturbances within the first 20 years, which is the equivalent of not allowing "reburns". The ">40 yrs" and ">80 yrs" allowed no pixels less than 40 and 80 years of age (respectively) to be disturbed after the landscape was aged. The last column of Table 4.1 (">80 yrs + buffer") represents two eligibility restrictions: disturbance was not allowed in pixels less than 80 years of age, and adjacent disturbances could not merge. To prevent individual disturbances from joining, I imposed a 350 m buffer around each new disturbance. This buffer was only effective within the first 20-year period. Not all combinations of size X eligibility restrictions were simulated. The sheer numbers of simulations (60 x 36 = 2,160 runs) was such that I decided not to extend the 20 hectare (lower) size restriction to all eligibility restriction scenarios. The first ten observations from each run of 60 were eliminated from each dataset to remove any bias associated with a common starting point. Each of the 50 remaining landscape scenes were summarized to measure pattern. The program used to do this (FRAGSTATS) was given only five patch-types: water, non-forested, young forest (0-40 years), pole forest (41-120 years), and mature forest (120+ years). Given the close relationship between the pattern metrics (discussed in the previous section), and the desire to keep things as simple as possible, I chose to summarize 129 five spatial pattern metrics: 1) weighted edge density (mature to young edges are "full" edges, while young or mature to pole edges are weighted partially - see Chapter 3 for details), 2) patch size average and standard deviation, 3) juxtaposition (the probability of a patch being beside all other patch types), 4) core area index (the area of a patch beyond 100 m of an edge, expressed as a percentage of the total area), 5) Shannon's evenness index (measure of the degree to which all patch types are evenly represented on the landscape). I also calculated and summarized the averages of, and changes in, the percentages of each of three forest age-classes (young, pole, and mature) between each 20 year period. 4.3.2 RESULTS In the interest of clarity, I have chosen to discuss each of the metrics progressively, building on information from each preceding section. Depending on the metric, the results have been summarized at one or two levels: the landscape as a whole, and by age-class. Furthermore, since there is virtually no research with which to corroborate these findings, and the potential for extensive analysis exists, I will only discuss those aspects of the results which were, in my opinion, the most striking and relevant. For a full listing of the simulation results, see Appendix B. AGE-CLASS DISTRIBUTION The simplest way of describing a landscape is by the average percentage of area in each age-class. The age-class averages will not be affected by changes in simulated size restrictions, so the averages were summarized by eligibility restrictions (Table 4.2). The total number of runs averaged (N) is also given in Table 4.2 for reference. 130 Table 4.2. Summary of simulated age-class percentages for the SBSmkl landscape. AGE-CLASS ELIGIBILITY RESTRICTIONS NONE A V E R A G E RATE OF DIST. ADJUSTED A V E R A G E R A T E OF DISTURBANCE (the average rate adjusted to disallow reburns) > 20 YRS ONLY > 40 YRS ONLY > 80 YRS O N L Y > 80 YRS + BUFFER Young (0-40) 37 36 35 39 31 28 Pole (41-120) 39 39 38 40 46 47 Mature(121+) 13 14 16 10 13 14 (N) 400 400 250 250 250 250 For the unrestricted scenarios, young forest accounted for 37% of the landscape, pole forest another 39%, and mature forest 13% (Table 4.2). These represent the long-term averages of the percentage of the landscape in each age-class, given the distribution of disturbance frequencies derived in Chapter 3. Virtually identical proportions were found for the scenarios restricting the disturbance rate to the average. This is to be expected since the average rate of disturbance has not changed, and no age restrictions have been imposed. Logically, the impact of increasing age restrictions should focus disturbance activity away from younger forests at the expense of older. Although the 40-year restriction did this to a small degree, the 20-year restriction did not show this trend (Table 4.2). It is unknown why this happened, although it may be related to the mechanics of disturbance movement used by the model. The shifts in age-class percentages were more significant for the 80-year restriction, but not as predicted. Instead, mature forest was held at 13%, young forest declined to 31%, and pole forest 131 increased to 46%. The reason for this is partly model artifact. The unrestricted scenario could virtually deplete mature forest on occasion, allowing very low percentages of mature age-class forest. When the model was restricted to disturbing only those areas greater than 80 years of age, the disturbances were severely limited in their ability to move freely across the landscape. The model formulation was such that the model searched for eligible areas, but in some cases, the necessary number of hectares was too difficult for the model to find. As it turned out, there seemed to be a minimum threshold of somewhere close to six percent of mature forest which this particular series of simulations could not get below. As a result, average young forest percentages were lower than expected. The 80-year plus buffer restriction was understandably more prone to this problem (Table 4.2). Although size restrictions had no effect on the average age-class percentages, they did have an impact on the range. As Figure 4.1 demonstrates, size restrictions generally narrowed the frequency distribution of percentages for each age-class (based on the 50 model runs). In other words, size restrictions prevented unusually large or small amounts of any one age-class from occurring on the 29,000 hectare measurement area. A similar narrowing of age-class percentage range was observed for the pole age-class. Furthermore, this tendency increased as size restrictions increased (from 10,000 to 60 hectares), although only these two extremes are shown in Figure 4.1(b and c). The narrowing effect on the frequency distribution is more noticeable at the upper end (of the distribution), but it had potentially critical impacts on the lower end too. This was particularly relevant for the mature age-class. For instance, both the 10,000 hectare and the 60 hectare restriction scenarios produced landscape scenes that had maximum mature age-class percentages 132 (a) I I I I I I 10 20 30 40 .10 60 Percent Area 70 80 90 I Young i l l ! Mature (b) I 20 J I I I I I 20 30 40 50 Percent Area fill 70 I Young 1 Mature (c) Hli .1(1 40 I Young Percent Area ] Mature Figure 4.1. Frequency distribution of percent young and mature age-classes for the natural (a), 10,000 ha (b) and 60 ha (c) scenarios for the SBSmkl landscape simulations (out of 50 runs). 133 near that of the unrestricted scenario (Figure 4.1). However, neither of these two scenarios (nor any other ones for that matter) could produce landscape scenes with very low percentages of mature forest. The unrestricted "natural" scenario created landscape scenes with less than 5% mature forest 20 times out of 50. The lowest percentage of mature forest that the 60 hectare scenario could manage was 6%, and this occurred only once. Size restrictions increasingly limited the ability of the 29,000 hectare area to contain high and low amounts of older forest. Increasing eligibility restrictions also created a narrower range of age-class percentages. In fact, once the size restrictions were added to the eligibility restrictions, this tendency became much more prominent. For instance, the corresponding standard deviation of the percentage of young forest for the natural scenario in Figure 4.1(a) was 23%, compared to 16% and 12% for the 10,000 hectare and 60 hectare scenarios, respectively (Figures 4.1(b and c)). The standard deviations for the 10,000 hectare and 60 hectare scenarios with the added restriction of the average rate of disturbance, were 10% and 2% respectively. Stated another way, the 95% confidence limits of young forest percentage for the natural scenario was 0-86%. In comparison, the 95% confidence limits of the young forest percentage for the scenario restricted to 60 hectares and the average disturbance rate was only 29-37%. An identical pattern was noted for pole and mature age-classes. INTERIOR FOREST AREA The average core area index (CAI) is the average interior forest area expressed as a percentage of total forest area. In this case, I defined only that area of a patch beyond 100 m (two pixels widths) of an edge as being interior forest or "core". The 100 m limit was chosen as a convenient arbitrary average, keeping in mind that the effective forest edge distance can vary 134 from several to hundreds of meters depending on the species or process (Murcia 1995). CAI responded to both disturbance size restrictions and eligibility restrictions, although much more so to the size restrictions (Table 4.3). The largest single response of CAI was to the elimination of the opportunity for disturbances over 10,000 hectares. CAI dropped by 10% (from 56% to 46%) when the otherwise unrestricted scenario was limited to 10,000 hectares (Table 4.3). For context, that translates to a decrease from 448,000 hectares to 368,000 hectares of interior forest over the 800,000 hectare landscape. Reducing the maximum disturbance size to 3,000 hectares only decreased the CAI by another 4% to 43%, and restricting the maximum disturbance size to 1,000 hectares only decreased it to 39% (Table 4.3). Surprisingly, the 60 ha restriction also averaged 39% interior forest, implying that no further loss of interior forest resulted between the 1,000 hectare upper limit and the 60 hectare scenarios. By adding a lower size limit restriction of 20 hectares to the upper size limit restrictions, CAI percentages increased seven to eight percent in each case. For example, for the scenarios with no eligibility restrictions, the 1,000 hectare size restriction created only 39% core area, while the 20-1,000 hectare restriction produced 47% core area (Table 4.3). The impact of not allowing disturbances less than 20 hectares in size on landscape CAI was surprising, and warranted further investigation. The frequency distributions of simulated CAI values for the mature age-class for a sample of upper and lower size restrictions were compared to isolate the impact of imposing a 20 hectare lower limit (Figure 4.2). Two pairs of scenarios were compared: the 10,000 hectare and the 20-10,000 hectare scenario in Figure 4.2(a), and the 1,000 hectare and the 20-1,000 hectare scenario 135 in Figure 4.2(b). Table 4.3 Average core area index (percent) for SBSmkl landscape simulations. SIZE RESTRICT, (hectares) ELIGIBILITY RESTRICTIONS NONE A V E R A G E RATE OF DIST. ADJUSTED A V E R A G E RATE OF DISTURBANCE (the average rate adjusted to disallow reburns) >20 YRS ONLY > 40 YRS ONLY > 80 YRS ONLY > 80 YRS + BUFFER NONE 56 55 48 52 49 46 0 - 10,000 46 44 44 49 50 39 0 - 3,000 43 41 41 44 45 36 0 - 1,000 39 40 37 42 39 32 60 only 39 37 36 38 41 34 20 - 10,000 53 52 - - - -20 - 3,000 49 50 - - - -20 - 1,000 47 46 - - - -Beginning with the two 10,000 hectare restricted distributions, there appears to be little or no difference between either the ranges or the averages of mature forest CAI (Figure 4.2(a)). So despite the fact that the 20 hectare lower limit increased landscape CAI averages from 46% to 53%> for these scenarios (Table 4.3), mature forest CAI did not increase. Thus, the gains in CAI between no lower size limit, and the 20 hectare lower size limit for the 10,000 hectare scenarios must be due to increases in young and pole age-classes' interior forest percentages. In fact, the percentage of young forest interior area increased from 51% to 61% with the addition of the 20 hectare lower limit for these two scenarios (see Appendix B). 136 5 10 15 20 25 30 35 40 45 50 55 60 Total Core Area Index • Natural • 20-1,000 • 1,000 Figure 4.2. Frequency of Total Core Area Index values for the mature seral-stage, as calculated by FRAGSTATS, for the natural simulation scenario, and the 20-10,000 and 10,000 ha size restricted scenarios (a), and the 20-1,000 and 1,000 ha size restricted scenarios (b) based on 50 simulation runs. 137 In contrast, the 20 hectare lower limit had a noticeable impact on mature forest CAI for the 1,000 hectare scenarios. In fact, landscape CAI and mature forest CAI values increased by almost the same amount when the 20 hectare lower limit was imposed (Appendix B and Table 4.3). However, in this case both of the 1,000 hectare scenarios had mature forest CAI frequency distributions much narrower than that of the natural scenario (Figure 4.2(b)). For instance, 12 times out of 50 the natural scenario created landscape scenes with more CAI than the greatest amount created by the 20-1,000 hectare scenario. Similarly, 17 times out of 50 the natural scenario created landscape scenes with less CAI than the least amount created by the 20-1,000 hectare scenario. In other words, the 20 hectare lower limit may have had a significant impact on landscape CAI averages, but still could not create the range of mature forest CAI values demonstrated by the natural scenario. A similar pattern was noted for each age-class CAI . Eligibility restrictions had a mixed impact on landscape CAI averages. Disturbance rate averaging had no impact on landscape CAI, while the 80-year plus buffer restriction had the greatest impact (Table 4.3). The decrease in CAI for the 80-year plus buffer restriction is not surprising. For every other restriction, disturbances could occur adjacent to each other. When disturbances were not allowed to merge, percentages of interior areas declined. For example, the average core area dropped by 3-11% between the five scenarios with the 80-year restriction and the five with the buffer plus 80-year restrictions (Table 4.3). The other three age-restricted scenarios (the 20-year, 40-year and 80-year) also resulted in a decrease in CAI values compared to the natural scenario. However, there was not a consistent decline in CAI with increasing eligibility restrictions as one might expect. In fact, the 40-year and 80-year eligibility restriction scenarios created more interior forest area than the 20-year 138 scenarios. The reason for this was shifting distributions of interior forest area by age-class. Interior forest area is distributed unequally between different age-classes even on "natural" landscapes. To demonstrate, Figure 4.3 depicts the range of CAI percentages from the "natural" scenario for the three forest patch-types. Young forest has a high percentage of interior area because the patches are uninterrupted (contiguous). Pole forest is moderately fragmented by younger disturbances, but has the advantage of covering a wide age range (41-120 years), so it still has relatively high CAI values. The mature patches tend to be the smallest and most spatially dispersed due to the influence of many overlying disturbances, and therefore have the least amount of interior forest area (Figure 4.3). When disturbances were increasingly limited by age eligibility (from 20 to 40 years), chances became even greater that mature forest would be disturbed and fragmented, resulting in less mature forest interior area. On the other hand, chances were greater that younger forest would remain uninterrupted or contiguous, resulting in more young forest interior forest areas. So the landscape CAI figures from Table 4.3 increased between the 20-year and 40-year eligibility restriction only because a slightly greater proportion of the landscape was accounted for by age-classes with the greater percentages of interior forest (young and pole). Figure 4.4 demonstrates this phenomenon for one pair of (20-year and 40-year restricted) scenarios. The 80-year restriction did not increase landscape CAI over the 40-year restriction because the decrease of mature forest interior area and the increase of pole forest interior area balanced out. Young forest interior area was not a factor in this case since the upper limit of the young age-class is only 40 years. 139 Figure 4.3. Frequency of Total Core Area Index values, as calculated by FRAGSTATS, for young, pole, and mature age-classes for the unrestricted (natural) simulation scenario for the SBSmkl simulations (based on 50 runs). young pole mature • 20 YEAR RESTRICTION 40 YEAR RESTRICTION Figure 4.4. Average total core area index, as calculated by FRAGSTATS, for young, pole, and mature seral-stages, for the 10,000 hectares size restriction for both the 20-year and the 40-year eligibility restriction scenarios for the SBSmkl (based on 50 runs). 140 Although not shown, the eligibility restrictions had the same effect as the size restrictions on the frequency distribution of the interior forest area of the three age-classes. Generally, as eligibility restrictions increased, the range of interior area percentages decreased. Appendix B provides both averages and standard deviations of CAI values for all three age-classes. EDGE DENSITY Edge density and interior area are highly (inversely) related, so I will not spend time repeating the observations made earlier. Furthermore, only landscape-level edge density will be discussed since it is not appropriate to break weighted edge scores down by age-classes (since the weighting would bias the age-class averages). Keep in mind that actual, full edge density is much higher. Recall from Table 3.4 that the weights given to edges between young and pole, and pole and mature age-classes were less than that of an edge between young and mature age-classes. Average edge density figures corroborate the same trends noted in the previous section. Generally, the greatest responses were to upper limits of disturbance sizes, and in particular to the restriction of disturbances to less than 10,000 hectares (Table 4.4). Average edge density also consistently increased in response to the addition of buffers, and decreased with the addition of a lower disturbance limit of 20 hectares (Table 4.4). To further investigate the impact of limiting disturbance sizes from below, edge density frequency distributions were compared. Figure 4.5 shows the frequency distribution of the same two pairs of scenarios compared in the previous section; the 0-10,000 and 20-10,000 size limits, and the 0-1,000 and 20-1,000 size limits. None of the four scenarios have eligibility restrictions. 141 Table 4.4. Average weighted edge density (m/ha) for the SBSmkl landscape simulations. SIZE RESTRICT, (hectares) ELIGIBILITY RESTRICTIONS NONE A V E R A G E R A T E OF DIST. ADJUSTED A V E R A G E RATE OF DISTURBANCE (the average rate adjusted to disallow reburns) >20 YRS ONLY > 40 YRS O N L Y > 80 YRS O N L Y > 80 YRS + BUFFER NONE 29 30 33 32 31 32 0 - 10,000 34 35 35 33 31 36 0 - 3,000 37 37 37 35 34 38 0 - 1,000 38 37 39 37 38 41 60 only 37 38 39 38 36 40 20 - 10,000 31 31 - - - -20 - 3,000 32 .32 - - - -20 - 1,000 34 34 - - - -In contrast to the response of interior forest area, the range of the frequency distribution of edge density was unaffected by size restrictions (Figure 4.5). Even the more restrictive 1,000 hectare scenarios allowed a range of edge densities very similar to that of the natural scenario (Figure 4.5(b)). Recall that the range of CAI decreased dramatically for the 1,000 hectare scenarios in Figure 4.2. The relevance of this is that edge density and interior forest area do not appear to be linearly correlated, as originally assumed. It also means that creating "natural" averages and distributions of edge densities may simply be a matter of manipulating upper and lower size restrictions. Unfortunately, only the 20 hectare lower size limit was simulated, so this hypothesis cannot be tested. 142 a 3 — Figure 4.5. 15 10 5 (b) 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 Weighted Edge Density (m/ha) • Natural • 20-1,000 • 1,000 Frequency of weighted edge density values for natural, 10,000, and 20-10,000 hectare scenarios (a), and for 1,000 and 20-1,000 hectare scenarios (b) for the SBSmkl (based on 50 simulations). 143 P A T C H SIZES Average patch size is often used in landscape pattern summaries, but can be very misleading. Patch size distributions are rarely normally distributed, and tend to have a large number of very small patches, and a small number of very large patches. Although average patch size is summarized here, I chose to focus the discussion on standard deviation of patch sizes as an indication of the dispersion of patch sizes. In contrast to edge density, the ranges of patch size standard deviations were quite wide across all simulation scenarios (Table 4.5). This is because a single large patch could have a large impact on the standard deviation. Once again, it is notable that the standard deviation fell from 561 to 306 hectares by eliminating disturbances greater than 10,000 hectares (Table 4.5). Further restrictions on maximum allowable patch size reduced the standard deviation more, but even when maximum disturbance sizes were limited to 1,000 hectares, average standard deviation was still 217 hectares. The relatively large impact of the 10,000 hectare size limit is the same pattern noted for both the interior forest and edge density results. Also consistent with previous findings, adding the lower disturbance size limit of 20 hectares increased standard deviation relative to the scenarios with no lower restrictions. However, in this case the effect was only moderate. For example the 3,000 hectare scenario produced a patch size standard deviation of 282 hectares while the 20-3,000 hectare scenario created a standard deviation of 311. The standard deviation actually decreases between the 1,000 hectare and 20-1,000 hectare scenarios (Table 4.5). This seemingly conflicts with previous findings on CAI and 144 edge densities and may be another indication of a more complex underlying relationship between these metrics. Table 4.5 Mean patch size (ha) / patch size standard deviation (ha) for SBSmkl landscape simulations. SIZE RESTRICT, (hectares) ELIGIBILITY RESTRICTIONS NONE A V E R A G E RATE OF DIST. ADJUSTED A V E R A G E R A T E OF DISTURBANCE (the average rate adjusted to disallow reburns) > 20 YRS ONLY > 40 YRS ONLY > 80 YRS ONLY > 80 YRS + BUFFER NONE 39 / 561 36 / 512 25 / 374 31 /443 28 / 361 28 / 311 0 - 10,000 23 / 306 22 / 251 22 / 284 27 / 397 29 / 329 23 / 237 0 - 3,000 21 / 282 20 / 202 20 / 217 24 / 304 23 / 252 20 / 229 0 - 1,000 19 / 217 23 / 153 18 / 144 22 / 223 19 / 217 19 / 208 60 only 22 / 192 22 / 96 19 / 104 21 / 144 22 / 223 20 / 191 20 - 10,000 35 / 414 34 / 377 - - - -20 - 3,000 32/311 29 / 291 - - - -20 - 1,000 22 / 153 22 / 197 - - - -It is interesting to note the trends of average patch sizes among the size restricted scenarios. For instance, for the scenarios with no eligibility restrictions, the average patch size decreased from 23 to only 19 hectares between the 10,000 hectare and the 1,000 hectare scenarios (Table 4.5). However, the 60 hectare scenario managed to create an average patch size of 22 hectares. Obviously the influence of small patches is quite large, and this demonstrates the danger of using average patch size alone as a landscape metric. Another example of the tremendous impact of small patches on the average patch size is the differences between the 10,000 hectare and the 20-10,000 hectare scenarios. The effect of adding 145 a lower restriction of 20 hectares in this case increased the average patch size from 23 to 35 hectares for the scenarios with no eligibility restrictions (Table 4.5). The frequency distributions of the standard deviations were also affected by restrictions. The patch size standard deviation for the natural scenario ranged from just over 250 to over 1,000 hectares. However, neither of the size-limited scenarios shown in Figure 4.6 produced a single landscape scene with a patch size standard deviation over 650 hectares, yet both had patch size standard deviations well below 250 hectares. Furthermore, even though average standard deviation decreased from 306 to 217 hectares between the 10,000 hectare and 1,000 hectare scenarios (Table 4.5) there is very little difference between the ranges (Figure 4.6). 21) 15 c u 3 — U — Figure 4.6. 10 5 100 200 300 400 500 600 700 800 Patch Size Standard Deviation (ha) I Natural 900 1000 10,000 1,000 Frequency distribution of patch size standard deviation for natural, 10,000, and 1,000 hectare scenarios for the SBSmkl landscape simulations (based on 50 runs). 146 Eligibility restrictions affected the patch size standard deviation frequency distributions in much the same way as did size restrictions. Generally, as age restrictions increased, the range of patch size standard deviations declined. As with the size restrictions, this trend was not evident from the averages in Table 4.5. For instance, although the average standard deviation of patch sizes of the 80-year restriction was a respectable 361 hectares compared to 561 for the natural scenario (Table 4.5), Figure 4.7 shows that the unrestricted, natural scenario range of standard deviations was double that of the 80-year scenario. 14 12 10 >> 8 a u — I-. b Figure 4.7. ll 100 200 300 400 500 600 700 Patch Size Standard Deviation (ha) I 80-Year M i l l 900 1000 • • Natural I 80-Year plus Buffer Frequency distribution of patch size standard deviations for the natural, 80-year, and 80-year plus buffer simulation scenarios for the SBSmkl (based on 50 runs). One final point of interest for the patch size standard deviation results was the contribution of each age-class to the overall averages reported in Table 4.5. Understandably, patch size standard deviation was almost always the greatest for young forest, and the least for mature forest. In fact, 147 average standard deviation for the mature age-class never exceeded 100 hectares for all 36 scenarios. As previously mentioned, chances were very poor that large areas of older forest could persist intact. Three examples of standard deviation by age-class are given in Figure 4.8. 2000 Figure 4.8. Average patch size standard deviation for young, pole, and mature seral-stages for natural, 40-year, and 3,000 hectare restriction simulation scenarios. A D J A C E N C Y To measure adjacency I chose to use an "interspersion / juxtaposition index" (IJI). The IJI I used ranges from zero to 100. Index values of 100 refer to maximum interspersion of patches, where all patches are equally adjacent to all other patch types in equal proportions (McGarigal and Marks 1994). This means that every patch in a simple landscape may touch every other patch type, but may still score an IJI significantly less than the maximum of 100. Furthermore, this analysis was not limited to the three forest patch-types, but included two fixed area patch-types: 148 water and non-forested land. Since these two permanent patch-types are well distributed spatially, it makes it more difficult to observe a very low IJI score (the lowest score found for all 36 x 50 = 1,800 runs was 41). Since my interest was mainly concerned with the adjacency patterns of forested patches, I summarized IJI only for the three forest patch-types. Overall, the effect of restrictions to the disturbance regime on IJI appear negligible. Unlike the previous pattern metrics, no trends were associated with increasing disturbance size restrictions, nor were there any consistent impacts on the eligibility-restricted scenarios. Furthermore, in all three age-classes, the IJI of the natural scenario was neither the greatest nor the least of the 36 scenarios (Tables 4.6, 4.7, and 4.8). However, there were some details worth mentioning. Unlike the preceding landscape metrics, limiting disturbance sizes from above had no effect on adjacency. However, the IJI of the six scenarios that use a 20 hectare lower size restriction were noticeably higher than those of most other scenarios, including the natural scenario (Table 4.6). For instance, while the six scenarios that had the 20 hectare lower limit averaged IJI of 84, the corresponding six scenarios with no lower limit averaged only 80 (Table 4.6). In each case, the average IJI increased by one to seven percent. The same effect can be seen in the pole age-class (Table 4.7) and the mature age-class (Table 4.8). The elimination of small patches increased the chances of any given patch bordering on all other patch-types evenly. The only other notable feature of the adjacency analysis is the behaviour of the pole age-class for the 80-year restriction. Three out of the five 80-year restriction scenarios showed very high IJI's. This can be at least partially attributed to the relatively large amount of area in the pole age-class that these scenarios maintained. 149 Table 4.6 Average interspersion / juxtaposition index for young forest (0-40 years) for the SBSmkl landscape simulations. SIZE RESTRICT, (hectares) ELIGIBILITY RESTRICTIONS NONE A V E R A G E RATE OF DIST. ADJUSTED A V E R A G E RATE OF DISTURBANCE (the average rate adjusted to disallow reburns) > 20 YRS ONLY > 40 YRS ONLY > 80 YRS O N L Y > 80 YRS + BUFFER NONE 79 83 78 79 85 85 0 - 10,000 81 82 82 79 85 79 0 - 3,000 78 81 81 79 80 75 0 - 1,000 76 82 81 78 76 71 60 only 81 82 81 80 70 68 20 - 10,000 82 86 - - - -20 - 3,000 85 85 - - - -20 - 1,000 83 85 - - - -Table 4.7 Average interspersion / juxtaposition index for pole forest (41-120 years) for the SBSmkl landscape simulations. SIZE RESTRICT, (hectares) ELIGIBILITY RESTRICTIONS NONE A V E R A G E RATE OF DIST. ADJUSTED A V E R A G E RATE OF DISTURBANCE (the average rate adjusted to disallow reburns) > 20 YRS ONLY > 40 YRS ONLY > 80 YRS O N L Y > 80 YRS + BUFFER NONE 85 88 78 86 91 88 0 - 10,000 84 86 86 86 93 88 0 - 3,000 83 86 86 87 93 87 0 - 1,000 81 86 85 85 81 84 60 only 84 85 84 85 87 84 20 - 10,000 88 90 - - - -20 - 3,000 89 90 - - - -20 - 1,000 88 89 - - - -150 Table 4.8 Average interspersion / juxtaposition index for mature forest (120+ years) for the SBSmkl landscape simulations. SIZE RESTRICT, (hectares) ELIGIBILITY RESTRICTIONS NONE A V E R A G E RATE OF DIST. ADJUSTED A V E R A G E RATE OF DISTURBANCE (the average rate adjusted to disallow reburns) > 20 YRS ONLY . > 40 YRS ONLY > 80 YRS O N L Y > 80 YRS + BUFFER NONE 76 77 76 77 79 78 0 - 10,000 75 78 . 77 74 79 72 0 - 3,000 73 75 76 73 74 68 0 - 1,000 71 77 75 72 71 67 60 only 72 77 75 73 72 72 20 - 10,000 81 80 - - - -20 - 3,000 78 80 - - - -20 - 1,000 76 81 - - - - D I V E R S I T Y Two aspects of diversity were addressed: spatial and temporal. S P A T I A L D I V E R S I T Y Spatial diversity measures are perhaps the most difficult pattern metric to interpret because the scales are dimensionless. Nevertheless, diversity measures can be informative in a relative sense. I chose to use Shannon's evenness index, which measures relative abundance of patch-types. A score close to zero implies one patch-type dominates, and a score of one means all patches are present in equal proportions. Since water and non-forested patch-types are permanent landscape features present in low quantities, an evenness score of one is not possible. Finally, it is important to keep in mind that this measures the evenness of a 29,000 hectare sample area. 151 Average evenness scores ranged between 0.71 and 0.88 (Table 4.9). The natural scenario averaged the least "even" distribution of patch-types, and all types and combinations of disturbance regime restrictions resulted in an increase in evenness. This is logical because the unrestricted scenario would be expected to create the greatest range of patch (percentage) distributions spatially and temporally. Table 4.9 Average Shannon's evenness index for the SBSmkl landscape simulations. SIZE RESTRICT, (hectares) ELIGIBILITY RESTRICTIONS NONE A V E R A G E RATE OF DIST. ADJUSTED A V E R A G E RATE OF DISTURBANCE (the average rate adjusted to disallow reburns) > 20 YRS ONLY > 40 YRS ONLY > 80 YRS ONLY > 80 YRS + BUFFER NONE 0.71 0.75 0.77 0.76 0.83 0.84 0 - 10,000 0.81 0.84 0.84 0.77 0.83 0.84 0 - 3,000 0.79 0.85 0.85 0.82 0.82 0.82 0 - 1,000 0.81 0.87 0.87 0.82 0.81 0.82 60 only 0.84 0.88 0.88 0.85 0.80 0.82 20 - 10,000 0.79 0.82 - - - -20 - 3,000 0.83 0.84 - - - -20 - 1,000 0.87 0.87 - - - -It is interesting to note that this is the only landscape metric discussed so far in which averaging the disturbance rate showed any consistent difference from the original range of disturbance rates. For instance, shifting from the average disturbance rate scenario to the natural scenario for the 60 hectare size limit caused the evenness index to drop from 0.88 to 0.84. Most of the other comparisons between unrestricted and average rate restricted scenarios showed similar differences. 152 T E M P O R A L D I V E R S I T Y Hulshoff (1995) first introduced the idea of measuring landscape "change", although the original interpretation was for changes in patch sizes and shapes over time. To a degree, change has already been discussed in a general sense throughout the previous results, but I wanted a metric that measured a rate of change. Therefore I interpreted temporal diversity simply as the ability of the landscape to change proportions of patch-types over time. In this case I averaged the change in percent of the contribution of each of the three forest patch-types between consecutive 20-year time-steps in Tables 4.10, 4.11 and 4.12. Keep in mind that these change values only measure the change in age-class percentages across the 29,000 hectare sample area. The change values for the entire landscape for all eligibility-restricted scenarios would be much smaller. Table 4.10 Average change over 20 years of percent young forest (0-40 years) for the SBSmkl landscape simulations. ELIGIBILITY RESTRICTIONS SIZE RESTRICT, (hectares) NONE A V E R A G E RATE OF DIST. ADJUSTED A V E R A G E RATE OF DISTURBANCE (the average rate adjusted to disallow reburns) > 20 YRS ONLY > 40 YRS O N L Y > 80 YRS ONLY > 80 YRS + BUFFER NONE 18 15 15 15 9 10 0 - 10,000 12 8 8 12 8 5 0 - 3,000 10 5 5 8 5 6 0 - 1,000 10 3 4 5 10 3 60 only 9 1 2 3 2 1 20 - 10,000 12 11 - - - -20 - 3,000 12 5 - - - -20 - 1,000 11 3 - - - -153 Table 4.11 Average percent change over 20 years of pole forest (41-120 years) for the SBSmkl landscape simulations. SIZE RESTRICT, (hectares) ELIGIBILITY RESTRICTIONS NONE A V E R A G E RATE OF DIST. ADJUSTED A V E R A G E RATE OF DISTURBANCE (the average rate adjusted to disallow reburns) > 20 YRS ONLY > 40 YRS ONLY > 80 YRS O N L Y > 80 YRS + BUFFER NONE 16 14 13 14 9 9 0 - 10,000 11 8 7 11 7 5 0 - 3,000 9 5 5 7 4 6 0 - 1,000 9 3 3 5 9 3 60 only 8 1 2 2 2 1 20 - 10,000 11 10 - - - -20 - 3,000 10 5 - - - -20 - 1,000 10 3 - - - -Table 4.12 Average change over 20 years of mature forest (120+ years) for the SBSmkl landscape simulations. SIZE RESTRICT, (hectares) ELIGIBILITY RESTRICTIONS NONE A V E R A G E RATE OF DIST. ADJUSTED A V E R A G E RATE OF DISTURBANCE (the average rate adjusted to disallow reburns) > 20 YRS ONLY > 40 YRS ONLY > 80 YRS ONLY > 80 YRS + BUFFER NONE 4 3 2 2 4 4 0 - 10,000 4 2 3 . 2 2 2 0 - 3,000 3 2 2 2 2 2 0 - 1,000 3 1 1 1 3 1 60 only 2 1 1 1 1 1 20 - 10,000 3 3 - - - -20 - 3,000 3 2 - - - -20 - 1,000 j 1 - - - -154 Not surprisingly, the greatest amounts of change occurred in the percentages of the young and pole age-classes. For instance, the six scenarios that were not size restricted for the young serai stage averaged 14% change (Table 4.10), the six for the pole age-class averaged 12% change (Table 4.11), but the six for the mature age-class averaged only 3% change (Table 4.12). Generally, size restrictions had a greater impact than eligibility restrictions on reducing the average percent change. However, the combination of size and eligibility restrictions was particularly influential in affecting percent change. In most cases, any scenario that had a combination of both types of restrictions was well under half the average change value of the natural scenario for all age-classes (Tables 4.10, 4.11, and 4.12). The large impact of averaging the disturbance rate on the average percent change is particularly notable. For instance, in the young age-class, the eight scenarios that had no eligibility restrictions averaged 12% change. The same eight size restricted scenarios with the average disturbance rate restriction averaged only 6% change (Table 4.10). The degree to which averaging the disturbance rate affected the percent change also increased with increasing size restrictions (Tables 4.10, 4.11, and 4.12). It is interesting that the spatial diversity analysis in the previous section using Shannon's evenness index found much the same thing. Finally, the particularly low values of change associated with almost all of the 60 hectare scenarios, in all age-classes, is remarkable in that the 60 hectare scenarios had not otherwise stood out in previous findings. Disregarding the scenarios with no eligibility restrictions, none of the 60 hectare size restricted scenarios averaged more than 3% change on average, for any of the age-classes (Tables 4.10, 4.11, and 4.12). 155 4.4 DISCUSSION AND SUMMARY Our limited experience in dealing with landscape patterns warrants both the caution and the seeming repetition with which the simulation results were presented. Repetition was necessary to allow more ways of "seeing" a landscape. Caution was required to prevent reading too much into such results since little or no interpretation concerning ecological relevancy is possible. Nevertheless, the simulation exercise still provided some valuable insights. Most of the landscape pattern metrics used were sensitive to all disturbance regime restrictions chosen. Generally, restrictions decreased interior forest area, patch size standard deviation, and age-class percentage change over time, and increased edge density and spatial diversity, compared to the unrestricted "natural" scenario. Adjacency was the only pattern metric which showed an inconsistent response to the disturbance regime restrictions. These overall results are not particularly surprising since these are the logical consequences of "evening out" the disturbance process spatially and temporally. The results are also much the same as the findings of landscape pattern studies on the effect of increasing cultural activity on actual landscapes (Ripple et al. 1992, Mladenoff et al. 1993, Luque et al. 1994, Wickham and Norton 1994). Again, this is not surprising since the disturbance restrictions were deliberately chosen with cultural changes (i.e. through timber harvesting activities) to the disturbance regime in mind. However, the details of how, and to what degree differences in disturbance regime parameters manifested themselves in the various pattern metrics was less predictable. Size restrictions generally affected the landscape averages more than eligibility restrictions. As 156 the maximum allowable disturbance size decreased, the pattern trends noted above became more prevalent. An unexpected aspect of the size restriction results was the large impact of restricting maximum disturbance size to 10,000 hectares, particularly in contrast to the relatively moderate impact exhibited by further maximum size restrictions to 3,000 and 1,000 hectares. The unrestricted simulation scenarios allowed disturbances of almost 250,000 hectares, but the chances of a disturbance of between 10,000 and 250,000 hectares was extremely remote (about 6 in 100,000). Obviously these very large disturbance events are vital to maintaining "natural" patterns. The dominance of large disturbances was also noted for the 1954 landscape snapshot (Chapter 2). The most notable change between the 10,000 hectare and the 1,000 hectare size-limited scenarios was the reduction in the range of the metrics, as opposed to the averages. This was best demonstrated by the age-class percentages, but occurred for other metrics. This is as much an issue for the lower bounds as with the upper bounds. For instance, the "natural" scenario was more than capable of literally depleting the mature forest on the 29,000 hectare model area. This raises the question of how, or rather at what scale, age-class distributions should be managed. The impact of eliminating very small disturbances on landscape pattern was more complicated. For the six scenarios where it was possible to compare 20 hectare lower limits with no lower limits on disturbance sizes, the differences were quite large for some metrics. The impression was that by simply adding the appropriate lower size limit to match an upper size limit, it may be possible to create landscapes which in some ways "look" like natural ones. However, here it became important to look at more than one metric, and beyond simple landscape averages. At least three such contrasts were noted: 157 1) If one used average patch size alone to judge the suitability of landscape patch sizes, the 20 hectare lower limit added to a 10,000 hectare upper limit would create landscapes not unlike "natural" ones (35 versus 39 hectares respectively). However, the standard deviation of the patch sizes drops from 561 hectares to 414 hectares. This reflected the lack of both upper and lower extremes in the restricted scenario. 2) Edge density figures also indicated that the 20 hectare lower size limit was effective in creating more "natural" landscapes. However, it was important to consider where this extra edge was being created. The juxtaposition index of all six scenarios that use a 20 hectare lower size restriction were noticeably higher than those of the natural scenario. This indicated that there was a greater number of edge types of all combinations for the 20 hectare size-limited scenarios. 3) Finally, average percentages of mature interior area were misleading in that they indicated that a lower size limit of 20 hectares would closely mimic the natural scenario patterns. However, the frequency distributions of the same metric revealed that the scenarios limited by size from both above and below severely restricted the range of possible interior areas of mature forest. A l l of the scenarios which restrict sizes to 60 hectares were remarkable in that the differences between the average metrics of the 0-1,000 scenarios and the 60 hectare scenarios (for all eligibility restrictions) were only marginal. As discussed earlier, the most likely explanation is that the average size of the 0-1,000 hectare scenarios is very close to 60 hectares. However, where differences did occur, they were important. For instance, the standard deviation of patch 158 sizes was lower for the 60 hectare scenarios. This may seem obvious, but the difference was not so much because of the inability to create large patches (since they could, and often did merge), but the inability to create small ones. More importantly perhaps, the 60 hectare scenarios severely restricted the temporal dynamics of the landscape pattern. This was demonstrated by both the temporal change results, and the extremely narrow range of each of the frequency distributions of the metrics. Eligibility restrictions influenced landscape patterns in ways that were different from the size restrictions. Although averaging the disturbance rate had a negligible impact on the amount of interior forest areas, edge density, and average patch size, it had a large impact on temporal diversity. The impact on temporal diversity may seem somewhat tautological, but recall that the metric summaries are only for a 29,000 hectare area. The smaller the sample area relative to the landscape, the greater the temporal diversity since the model distributes disturbances randomly. Therefore, these results only imply that temporal diversity in smaller areas of the landscape (29,000 hectares) is far greater for unrestricted disturbance regime scenarios. However, keep in mind that the model distributed disturbances randomly through space. Had the model clustered disturbances, this would have mitigated this issue to some degree. As the allowable minimum age of disturbance increased, percentages of the oldest age-class declined since it became more likely that older forests would be disturbed. This increased disturbance pressure on the mature age-class resulted in less mature forest interior area since the areas remaining tended to be smaller and more isolated. This also forced the model to "look" for eligible areas over a smaller area. As a result, chances increased that disturbances would run together, forming larger contiguous areas. The combined result led to more "natural" average 159 landscape metrics with the 40-year limit, despite the decline of mature forest area, and mature forest interior area. The 80-year scenario was not always able to find enough area to disturb because of the spatially random nature of the disturbance module. It is unknown what the landscape pattern metrics of the 80-year scenarios would have been i f the model could have taken all possible eligible forest areas when it needed to. However, without a doubt, both the total, and interior area of the mature age-class would have declined even further. The effect of the 80-year plus buffer restriction is also difficult to assess because these scenarios were not able to always disturb the necessary amount of area either. However, by comparing the buffer scenarios with the non-buffer 80-year scenarios, it is evident that there was a noticeable decline in interior area percentages and an increase in edge densities with the buffers. None of the other metrics showed consistent differences between these two sets of scenarios. In the end, the simulation exercise was as much about developing an expanded appreciation for the various "pattern" metrics, as it was about the effect of changing landscape disturbance regime parameters on those landscape patterns. For instance, landscape averages are clearly one dimensional values, and taken alone, would not have fully described the SBSmkl landscape pattern. Despite the fact that the average metric figures were often comparable, the frequency distributions of any given metric through the 50 model runs revealed that the unrestricted simulation, or the "natural" runs, exhibited a much greater range than any other scenario. The extreme pattern scenes created by the unrestricted scenario are important since they are more than just possible scenes, they are virtually as probable as the average (in other words the frequency 160 distribution was relatively flat). When changes to "natural" disturbance regime parameters results in a restriction of the response range, then the dynamic behaviour of the landscape, and hence its pattern, has been altered. This ties in with the issue of temporal diversity. Small to medium-sized areas (similar in size to the sample area in this study) experience dramatic change over short periods of time. If disturbance regime restrictions reduce that variability, the behaviour of the landscape is altered. In this study, it was not uncommon that simulation scenarios virtually eliminated temporal variability. This also raises the issue of the most appropriate scale(s) at which we should be measuring and managing landscapes. The temporal change comparisons suggest a dynamic dimension to landscapes that is perhaps still not fully appreciated. For instance, the adequacy of preservation of constant percentages of representative forest areas as a habitat conservation measure is a questionable management tactic since age-class cycling will occur whether or not areas are protected from human-caused disturbance (Noss 1990, Sinclair et al. 1995). Throughout the analysis, it was not difficult to find that some effect on pattern resulted from every type and degree of restriction. This was expected, but the range and complexity of pattern responses to restrictions was not. This led to many unanswered questions, but this is also to be expected. In fact, the generation of more directed, specific questions is one of the desired consequences of such exploratory studies. This exercise has exposed several aspects of pattern dynamics that will potentially help direct future landscape ecology research efforts, including more serious consideration of the temporal dimension, the relationship between pattern metrics, the relationship between specific aspects of pattern and ecological response, and alternative disturbance regime parameters which may affect pattern (e.g., the spatial distribution of 161 disturbances (random vs regular vs clumped) on very large scales may be an important landscape parameter not addressed in this study). Perhaps of greater importance, the simulations demonstrated the depth of our lack of knowledge concerning landscapes and patterns. This exercise will become a part of that body of research which forms only the beginning of ecological landscape understanding. 162 CHAPTER 5 - DISCUSSION AND CONCLUSIONS Discussion and conclusions relating to the specific findings of this research have already been presented at the ends of the respective chapters and will not be repeated here. Instead, this chapter will summarize these research findings in the context of the science of landscape ecology, and the practice of landscape management. 5.1 GENERAL DISCUSSION The context for this research is the prevailing desire to mimic or approximate natural processes and patterns in forest management practices. This assumes that: 1) We possess the necessary knowledge to define "natural" processes and patterns, and 2) We have both the ability and willingness to "mimic nature". One cannot mimic without knowledge of that which is to be mimicked. While a suitable knowledge-base of forest systems at the stand level may be available, our knowledge at landscape scales tends to be simplistic and anecdotal. If nothing else, this research on the SBSmkl landscape has shown how misleading, and therefore how dangerous, reliance on this knowledge can be. For example, as suspected, there was evidence to suggest that small fuel-breaks (such as creeks) had some influence on fire behaviour. However most fires, or most parts of fires, clearly ignored them. It may also come as a surprise that on larger scales, natural landscapes may be simpler, and less diverse, than managed landscapes, or that disturbances over 10,000 hectares are so vital 163 to maintaining "natural" patterns. Perhaps the greatest surprise was the degree to which temporal pattern was a part of this particular landscape, and the degree to which changes to the disturbance regime affected it. Not only did disturbance regime restrictions significantly narrow the response range of most spatial pattern metrics, but the ability of the landscape to change over time was severely compromised. One of the reasons that landscape-level knowledge tends to be overly simple and/or misleading is that a large part of the existing knowledge-base of landscapes has come from landscape "snapshots". Snapshots are useful for describing spatial pattern relationships, but do not capture dynamics. Using landscape snapshots as the basis on which "natural" landscape patterns are defined and understood is risky. Not only is it possible that that one particular snapshot may be an anomaly (as the 1954 snapshot of the SBSmkl may be), but the tendency is to become overly concerned with spatial issues and ignore the temporal dimension altogether. For instance, what is the relevance of mimicking the average amount of young age-class forest on the SBSmkl landscape when that particular amount of young forest occurred perhaps in one decade out of ten? In a way, the focus on spatial pattern is epitomized by the acceptance of the Weibull models of age-class distributions as appropriate metaphors for landscape dynamics. Perhaps the most questionable landscape theory currently under discussion is that observing age-class stability is simply a matter of finding a large enough area. One of the most striking findings of this research is the lack of stability of the age-class distribution of the SBSmkl landscape. This was most dramatically demonstrated by the range of historical 20-year disturbance rates (from seven to over 40 percent), but was also evident in the range of fire cycle estimates (from 81 to 104 years). In addition, the simulation exercise provided ample evidence that perhaps the most important pattern 164 element of this landscape historically was simply that it was highly dynamic. Until now, stable models have been useful tools for understanding landscape dynamics at the simplest level, but it would be a mistake not to move past them. That this landscape is not stable should no longer be a surprise; enough information already exists to begin formulating a theory based on the interaction of processes occurring at different scales in time and space. For example, consider the following list of factors that influence landscape fire behaviour, and the scales at which they operate. Beginning at the smallest scale, results suggest that SBSmkl stands beyond a certain age become more susceptible to being disturbed. This phenomenon occurs on small spatial scales, and is a relatively slow process temporally. This study also found that permanent features of the landscape (creeks and other fuel breaks) may influence individual fire behaviour. These occur on small to medium spatial scales, and are normally temporally permanent. Larger scale features of the landscape which may influence long-term fire frequency, such as dry soils, are also permanent temporally. Although not discussed in this study, across the entire landscape, general weather patterns (which influence fire behaviour), as well as lightning patterns will vary spatially and temporally over very short periods of time. Finally, over very long periods of time, climate inevitably changes, influencing the entire area. Even with this incomplete list of factors that influence the behaviour of fire over a landscape over many decades, it is not difficult to imagine how they would combine to create a seemingly random pattern of landscape fire behaviour. Furthermore, under this multi-phenomenon model, even i f the SBSmkl were ten times as big, it is unlikely that it would exhibit any more stability than the current age-class distribution. However, Weibull models of age-class distributions will always demonstrate stability because they only consider one phenomenon. Knowing this, there 165 is no point in pursuing stable models to describe landscapes, and continue to be misled into believing they represent landscape behaviour. Landscape understanding can also be misled by focusing on pattern rather than process. Pattern research alone can too easily lead to the conclusion that the pattern may be little more than random, which means "management" has ultimate flexibility. For instance, based on selected landscape metrics, it could be argued that over the short-term (20 years) it is unlikely that forestry activities could possibly create a landscape that is "unnatural" since the range of natural behaviour is so great. It could further be argued that because our efforts to curb natural forms of disturbance are limited, natural disturbance will be affecting landscape pattern anyways. However, to rely on pattern, and ignore process, means our assumptions (about what we think we know) may be creating a very different landscape much faster than we imagine. I have already discussed the dangers of using landscape snapshots as pattern benchmarks, but there are other aspects that may be of concern. For instance, the BC Biodiversity Guidebook characterizes the SBSmkl as having a fire cycle of between 100 and 150 years (BC Ministry of the Environment and BC Environment 1995). If management were to use the average fire cycle (125 years) literally as a harvesting rotation, there is little doubt that the species composition of the landscape would become more uniform spatially. Similarly, although my research strongly suggests that older forested areas were highly mobile temporally and spatially, the Biodiversity Guidebook advocates regularly spaced permanent reserves with a fixed area. What danger will these areas pose to the rest of the landscape in terms of conduits for insects and disease? Finally, despite the fact that this research has shown that the natural rate of disturbance is approximately 1.1% per year, i f that rate of harvesting is imposed on top of other size and eligibility restrictions, 166 the amount of mature forest will dwindle to virtually zero (see Chapter 2). This is clearly not acceptable, and in response we have reduced disturbance rates significantly (to 0.68% per year -see Chapter 2), with unknown impacts on the landscape ecosystem. The preceding discussion demonstrates how this research contributed to the knowledge-base of landscapes-level research. However, is there now enough knowledge of SBSmkl landscapes to practice "mimicry" of natural patterns and processes? Unfortunately, there is no answer to this question. Research will continue, and each piece of new knowledge will bring us closer to that goal, but there is no way of knowing what is unknown, or of what consequence it is ecologically. Ironically, as the amount of landscape-level knowledge grows, it will become more difficult to practice landscape management (as defined in this study). Recall that the second assumption of the strategy of mimicry as a forest management paradigm is that we have both the ability and the will to mimic. The will to mimic is culturally defined, and cultural preferences are not likely to change as fast as new knowledge is gained. For instance, despite the findings in this study, large clearcuts are not likely to be acceptable on any boreal landscapes in Canada today, let alone increasing disturbance rates to include immature timber. Even i f society were somehow fully aware of the (negative) ecological impacts, there are still cultural issues that currently take precedence, such as economics. Absolute mimicry of landscape processes and patterns is therefore apparently out of the question. The mimicry concept is thus limited by both knowledge and culture, but for the moment, more so by knowledge. At this point we have the luxury of managing in the absence of extensive knowledge of landscapes as ecosystems. Ultimately, as information such as that from this 167 research becomes available, we will have to make more informed, but more difficult decisions about the degree to which we will choose to mimic. For instance, it is likely much more acceptable to society to have larger cutblocks than variable rotation lengths, and/or spatially clumped harvesting on large scales. However, the simulations suggested that the latter two strategies would result in more "natural" landscape patterns. The problem is that by just increasing the cutblock size, fragmentation re-emerges at a larger scale when disturbances are placed randomly. The evolution of landscape management will therefore be a function of both landscape ecology research, and cultural choices. Landscape ecology research can anticipate this relationship in several ways: 1) An alternative model of landscape behaviour must be formulated and effectively communicated to replace the stable model currently assumed. Logically, it should consider multi-scalar phenomenon as "hierarchies" (sensu O'Neill et al. 1986). The challenges in studying hierarchical systems are twofold: first, producing reasonable results (prediction), and' second, interpreting these results. Many of the interpretation problems have been discussed already. One simply cannot know everything about why a fire behaved as it did years or decades later. It would seem that the best way of dealing with this is to compliment historical reconstruction research with active experimentation on scales much larger than that normally associated with fire behaviour research. There are obvious problems with conducting such experiments, but our ability to "control" forest fires is such that there will be more than enough study sites created in the future. The challenge will be in being prepared to take advantage of them at the time. The current fire research in Yellowstone Park is a good example of such a strategy. 168 2) Work must continue on studying the relationship between process and pattern in models to explore behaviourial ranges and thresholds, and create hypotheses for empirical research. Modelling is the ideal medium to explore the pattern/process relationship because it allows one to investigate the potential interactions of processes responding to different frequencies. Although the factors influencing landscape pattern are often difficult to quantify, it is reasonable to expect to be able to adequately predict at least ranges of some of these factors. Susceptibility of fuel-types, the influence of permanent landscape features, and even burning conditions and climate change are predictable phenomena within limits. 3) Orient experimental research according to a) hypotheses generated by modelling, and b) management flexibility. Although absolute mimicry may not be possible, we still have the ability to control a large number of disturbance regime parameters. This research has shown that patterns can almost always be isolated and described, but whether or not there is any ecological relevance is still unknown. It will be the experimental work that will allow us to make informed choices about the most appropriate patterns. This evolution of landscape management will take time, but it has already begun in B.C. with the introduction of the Biodiversity Guidebook (B.C. Ministry of Forests and B.C. Environment 1995). These guidelines concentrate on patch sizes and age-class representation, but it is hoped that the list will grow. The next step may be the introduction of the temporal dimension, allowing that mature and over-mature areas of the forest are a part of the shifting mosaic phenomenon, and not semi-permanent reserves. This would eventually lead to the acceptance of the idea that percentages of mature and over-mature areas on a given landscape will fluctuate over extended periods of time. 169 Ideally, this would all culminate with the acceptance of the integration of natural disturbance processes with cultural disturbance events. This is where understanding process, and not just pattern becomes relevant, and considering landscapes as hierarchical is useful. For example, long-range predictions of relative levels of risk associated with different areas of a landscape can be made and included in landscape plans as discontinuous, low to medium frequency disturbance events. Overlain on this would be harvesting strategies as continuous, high frequency events (annual). The average harvesting rate can be held constant (without necessarily adversely affecting woodflow), and the overall disturbance rate (i.e., harvesting plus natural disturbance) would be allowed to wander over time (satisfying the criteria of landscape change), as would the concentration of disturbances spatially. Furthermore, if and when these natural disturbance events occurred, they would simply be considered a part of the long-term ecological strategy, but not a part of the harvesting strategy (i.e., they should not be salvage logged). Once such a system is in place, the relative proportions of allowable natural to cultural disturbances can be determined for different areas depending on fire control activities. 5.2 CONCLUSIONS This research has provided a view of boreal-type landscapes that is much different and much more dynamic than previously thought. By doing so, it has also demonstrated that we still know discouragingly little about landscape dynamics. It would not be an exaggeration to say that more is known about the landscape dynamics of the (relatively simple) SBSmkl than most other areas of equivalent size in Canada. To what degree then, can we claim to be practicing forest management based on "natural" principles? 170 The future will no doubt bring better knowledge about natural processes and patterns, but it will also bring increasing cultural pressure against the implementation of various aspects of landscape manipulation that would result in more "natural" landscapes. Merging these two factors (ecological knowledge and culture) form the basis of so-called landscape or ecosystem management. The ecological knowledge is meant to form the foundation of management choices (culture). However, the (largely untested) concept of landscape management has created expectations, or ideals, that may be difficult to live up to. As Salwasser (1994) concludes, landscape management may not end up being a new paradigm at all, but simply "...a luxury to be pursued only after all other needs are met". The only reason that there is general agreement now that it will "work" is that definitions of landscape management vary widely. There is not even agreement on how it will differ from current forest management practices (Gerlach and Bengston 1994, Irland 1994, Salwasser 1994). Not surprisingly, most of the disagreement hinges on different perceptions of the depth and breadth of ecological knowledge, the foundation of landscape management. It would seem then, the most immediate priority is to reach a common understanding of: a) how much ecological knowledge exists (particularly at the landscape scale), b) the limitations to "mimicking nature" that we are aware of today, and the potential ecological consequences, and c) what the most pressing ecological concerns are, or, what it is that we don't know. The responsibility for most of these fall to the landscape ecologist. It is expected that the third factor, that of understanding what is unknown, will be the most challenging, but the most informative. Judging by this research, it will certainly be the longest list. For their part, the 171 greatest challenge for managers will be accepting and allowing for the unknown. This could take many forms from simply being conservative, to developing far less rigid, "cookbook" approaches to forest management. In the end, it may be how we deal with the unknown that defines landscape management as a new paradigm. We cannot now, nor will we ever, completely "mimic" natural patterns on these forest landscapes. Given this, we must take responsibility for understanding the magnitude of the changes that we will impose, and their impact on the ecological resources we are attempting to sustain. This research has taken that first step towards understanding those differences. 172 LITERATURE CITED Alexander, M.E. , B.D. Lawson, B.J. Stocks, and C.E. Van Wagner. 1984. User guide to the Canadian Forest Fire Behaviour Prediction System: Rate of spread relationships (interim edition). Canadian Forestry Service Fire Danger Group, CFS. 76 p. Allen, T.F.H., R.V. O'Neill , and T.W. Hoekstra. 1984. Interlevel relations in ecological research and management: Some working principles from hierarchy theory. U S D A Rocky Mtn. For. and Range Exp. Stn. General Tech. Rep. RM-110., Fort Collins, Colorado, 11 p. Allen, T.F.H. and T.B. Starr. 1982. Hierarchy: Perspectives for ecological complexity. University of Chicago. 310 p. Anderson, H. 1982. Aids to determining fuel models for estimating fire behaviour. U S D A For. Serv. Gen Tech. Rep. INT-122. Andison, D.W. 1993. Variability in young jack pine plantations of Northern Ontario. Regeneration Intelligence Forestry Consultants report No. 9301, Prepared for the Ontario Ministry of Natural Resources. Vancouver, BC, Canada. 20 p. Andison, D.W., D.F. Clark, P.J. Burton, F.L. Heinzelman. 1995. A Review of the Cassiar TSA, British Columbia. Available from BC Ministry of Forests, Fort Nelson, BC. 7 p. Antonovski, M.Ya. , M.T. Ter-Mikaelian and V . V . Furyaev. 1992. A spatial model of long-term forest fire dynamics and its applications to forests in western Siberia. In: Shugart, H.H. , R. Leemans, and G.B. Bonan (eds) A Systems Analysis of the Global Boreal Forest. University Press, Cambridge, p. 373-403. Baker, W.L. 1989a. Landscape ecology and nature reserve design in the Boundary Waters canoe area, Minnesota. Ecology 70(1): 23-35. Baker, W.L. 1989b. Effect of scale and spatial heterogeneity on fire-interval distributions. Can. J. For. Res. 19 pp. 700-706. Baker, W.L. 1989c. A review of models of landscape change. Landscape Ecology. 2(2): 111-133. Baker, W.L. 1993. Spatially heterogeneous multi-scale response of landscapes to fire suppression. Oikos, 66: 66-71. Ball, G.L. and D.P. Guertin. 1992. Improved fire growth modelling. Int. J. Wildland Fire 2(2): 47-54. Balsillie, D. 1994. Applying the ecosystem approach to resource management in Ontario. In: Jordan, J.K. and P.W.C. Uhlig (eds) Ecosystem management strategies for the Lake Superior Region. A conference and workshop in Deluth, Mn. May 16-19 1994. p. 8-11. 173 Barrett, S.W. and S.F. Arno. 1991. Classifying fire regimes and defining their topographic controls in the Selway-Bitterroot wilderness. In: Andrews, P.L. and D.F. Potts (eds), Proc. of the 11th Conf. on Fire and Forest Meteorology. April 16-19, 1991, Missoula, Montana. Society of American Foresters, Bethseda, M D . pp. 299-307. Bazzaz, F.A. 1983. Characteristics of populations in relation to disturbance in natural and man modified ecosystems. In: Mooney, H.A. and M . Godron (eds), Disturbance and Ecosystems: Components of Response. Ecol. Stud. 44. Springer-Verlag, Germany, p. 259-275. Benson, B.J. and M.D. MacKenzie. 1995. Effects of sensor spatial resolution on landscape structure parameters. Landscape Ecology 10(2): 113-120. Bessie, W.C. and E.A. Johnson. 1995. The relative importance of fuels and weather on fire behaviour in subalpine forests. Ecology 73(3): 747-762. Booth, D.L. , D.W.K. Boulter, D.J. Neave, A . A . Rotherham, D.A. Welsh. 1992. Natural forest landscape management: a strategy for Canada, (draft). 14 p. Botkin, D.B. 1993. Forest Dynamics: A n Ecological Model. Oxford University Press. New York. 309 p. Botkin, D.B., J.F. Janak, and J.R. Wallis. 1972. Some ecological consequences of a computer model of forest growth. J. Ecol. 60: 849-872. Boyce, S.G. and W.H. McNab. 1992. Management of forested landscapes. Paper presented at Annual convention, SAF, Richmond, V A , Oct. 25-28, 1992. 15 p. Bradley, A.F. , W.C. Fischer, N . V . Noste. 1992. Fire ecology of the forest habitat types of Eastern Idaho and Western Wyoming. USDA For. Serv. Gen. Tech. Rep. INT-290. 59 p. British Columbia Ministry of Forests and BC Environment. 1995. Biodiversity guidebook of the forest practices code of British Columbia. Province of BC. 99 p. Bunce, R.G.H. and R.H.G. Jongman. 1993. An introduction to landscape ecology. In: Bunce, R.G.H., L.Ryszowski, and M.G. Paoletti (eds), Landscape Ecology and Argoecosystems. Lewis Publishers, Ann Arbor, M i . 241 p. Burns, R . M . and B.H. Honkala. 1990. Silvics of North America, Vol . 1: Conifers. U S D A For. Serv. Agriculture Handbook 654. 675 p. Cattelino, P.J., Noble, I.R., Slatyer, R.O. and Kessell, S.R. 1979. Predicting the multiple pathways of plant succession. Envir. Manage. 3: 41-50. Clark, D.F. 1994. Post-fire succession in the sub-boreal spruce forests of the Nechako Plateau, Central British Columbia. M.Sc. Thesis. U . Vic, Victoria, BC, Canada. 124 p. 174 Clark, J.S. 1990. Landscape interactions among nitrogen mineralization, species composition, and long-term fire frequency. Biogeochemistry, 11: 1-22. Clarke, K .C . , J.A. Brass, and P.J. Riggan. 1994. A cellular automaton model of wildfire propagation and extinction. Photo. Eng. & Remote Sensing 60(11): 1355-1367. Costanza, R. and T. Maxwell. 1994. Resolution and predictability: An approach to the scaling problem. Landscape Ecology 9(1): 47-57. Coulson, R.N. , C.N. Lovelady, R.O. Flamm, S.L. Spradling, and M.C . Saunders. 1991. Intelligent geographic information systems for natural resource management. In: Turner, M . G . and R.H. Gardner, (eds), Quantitative Methods in Landscape Ecology. Ecol. Stud. 82. Springer-Verlag, New York. p. 154-172. Cumming, S.G., P.J. Burton, and B. Klinkenberg 1996. Canadian boreal mixedwood forests may have no "representative" areas: some implications for reserve design. Ecography (in press). Cumming, S.G. and R.T. Pelletier. 1995. Fire incidence, forest cover and riparian buffers in Northeastern Alberta. Forest Sciences Department, U B C (unpublished report). Davis, W. 1993. The global implications of biodiversity. M . A . Fenger et al. (eds.), Our Living Legacy. Proc. of a Symp. on Biological Diversity. Victoria, BC. pp. 23-46. Davis F.W. and J. Dozier. 1990. Information analysis of a spatial database for ecological land classification. Photogrammetric Engineering and Remote Sensing. 56(5): 605-613. Dawson, A . B . 1989. Soils of the Prince George - Mcleod Lake area. BC Ministry of the Environment, Victoria. 219 p. DeLong, S.C. and D. Tanner. 1995. Managing the pattern of forest harvest: Lessons from wildfire. Biodiversity and Conservation, (in press). DeLong, C , D. Tanner, and M.J. Jull. 1993. A field guide for site identification and interpretation for the southwest portion of the Prince George forest region. BC Ministry of Forests, Victoria, BC. 290 p. Denslow, J.S. 1985. Disturbance mediated coexistence of species. In: Pickets, S.T.A. and P.S. White (eds), The Ecology of Natural Disturbance and Patch Dynamics. Academic Press, p. 307-323. di Castri, F. and A.J . Hansen. 1992. The environment and development crisis as determinants of landscape dynamics. In: Hansen, A.J . and F. Di Castri (eds), Landscape Boundaries: Consequences for Biological Diversity and Ecological Flows. Ecol. Stud. 92. Springer-Verlag, Germany, p. 3-18. 175 Dunn, C P . , D . M . Sharpe, G.R. Guntenspergen, F. Stearns, and Z. Yang. 1990. In: Turner, M.G. and R.H. Gardner, (eds), Quantitative Methods in Landscape Ecology. Ecol. Stud. 82. Springer-Verlag, New York. p. 173-198. Eberhart, K . E . and P .M. Woodard. 1987. Distribution of residual vegetation associated with large fires in Alberta. Ca. J. For. Res. 17: 1207-1212. Egler, F.E. 1954. Vegetation science concepts. I. Initial floristic composition - a factor in old-field vegetation development. Vegetatio 4: 412-417. Enoksson, B., P. Angelstram, and K. Larsson. 1995. Deciduous forest and resident birds: the problem of fragmentation within a coniferous forest landscape. Landscape Ecology. 10(5): 267-276. Eremko, R.D. 1990. A review of natural lodgepole pine regeneration in the Chilcotin. F R D A Report No. 126. BC Ministry of Forests and Forestry Canada. 20 p. Evanisko, F. 1991. Dimensions of scale in landscape analysis. In: Systems Analysis in Forest Resources, USDA For. Serv. S. For. Exp. Stn. Gen. Tech. Rep. SE-74, Charleston, S.C.. p. 81-88. Everham III, E . M . 1995. A comparison of methods for quantifying catastrophic wind damage to forests. In: Wind and Trees, Countts, M.P. and J. Grace (eds.). Cambridge University Press. Great Britain, pp. 340-355. Finney, M . A . 1994. FARSITE Fire Area Simulator, Users Guide and Technical Documentation. Systems for Environmental Management, Missoula, MT. 54 p. Flamm, R.O. and M . G . Turner. 1994. Alternative model formulations for a stochastic simulation of landscape change. Landscape Ecology 9(1): 37-46. Forestry Canada Fire Danger Group. 1992. Development and structure of the Canadian Forest Fire Behaviour Prediction System. Information Report ST-X-3, Forestry Canada, Ottawa. 63 P-Forman, R.T.T. 1995. Some general principles of landscape and regional ecology. Landscape Ecology. 10(3):133-142. Forman, R.T.T. and M . Godron. 1986. Landscape Ecology. J. Wiley and Sons. N Y . Fowells, H.A. 1965. Silvics of forest trees of the United States. Agric. Handbook No. 271. USDA, For. Serv. 762 p. Foster, D.R. and E.R. Boose. 1995. Hurricane disturbance regimes in temperate and tropical forest ecosystems. In: Wind and Trees, Countts, M.P. and J. Grace (eds.). Cambridge University Press. Great Britain, pp. 305-339. 176 Franklin, J.F. 1993. Preserving biodiversity: species, ecosystems, or landscapes? Ecol. Appl. 3: 202-205. Fried, J.S. and J.K. Gilles. 1988. Stochastic representation of fire occurrence in a wildland fire protection planning model for California. For. Sci. 34(4): 948-959. Fuller, M . 1991. Forest fires: An Introduction to Wildland Fire Behaviour, Management, Firefighting, and Prevention. John Wiley and Sons Inc. 238 p. Gardner, R.H. and R.V. O'Neill. 1991. Pattern, process and predictability: the use of neutral models for landscape analysis. In: Turner, M.G. and R.H. Gardner, (eds), Quantitative Methods in Landscape Ecology. Ecol. Stud. 82. Springer-Verlag, New York. p. 289-307. Gardner, R.H. and M . G . Turner. 1991. Future directions in quantitative landscape ecology. In: Turner, M . G . and R.H. Gardner, (eds), Quantitative Methods in Landscape Ecology. Ecol. Stud. 82. Springer-Verlag, New York, p 519-525. Gauthier, S., A . Leduc, and Y . Bergeron. 1995. Forestry dynamics modelling under a natural fire cycle: A tool to define natural mosaic diversity for forest management. Can. For. Serv., Petawawa, Canada, (internal report). Gerlach, L.P. and D.N. Bengston. 1994. If ecosystem management is the solution, what's the Problem? J. of Forestry 92(8): 18-21. Gleason, H.A. 1926. The individualistic concept of the plant association. Amer. Midi . Nat. 21: 92-110. Godron, M . and R.T.T. Forman. 1983. Landscape modification and changing ecological characteristics. In: Mooney, H.A. and M . Godron (eds), Disturbance and Ecosystems: Components of Response. Ecol. Stud. 44. Springer-Verlag, Germany, p. 12-28. Green, D.G. 1989. Simulated effects of fire, dispersal and spatial pattern on competition within forest mosaics. Vegetatio. 82:139-153. Green, D.G, A . Tridgell, and A . M . Gil l . 1990. Interactive simulation of bushfires in heterogeneous fuels. Mathematical and Computer Modelling 13(12): 57-66. Grime, J.P. 1979. Plant Strategies and Vegetation Processes. Wiley, New York. Gustafson, E.J. and T.R. Crow. 1994. Modelling the effects of forest harvesting on landscape structure and the spatial distribution of cowbird brood parasitism. Landscape Ecology. 9(4): 237-248. Hamilton E.H. 1994. Biodiversity research and extension strategy. BC Min. of For. 20 p. (draft) 177 Hansen, A . , D.L. Urban, and B. Marks. 1992. Avian community dynamics: The interplay of landscape trajectories and species life histories. In: Hansen, A.J . and F. Di Castri (eds), Landscape Boundaries: Consequences for Biological Diversity and Ecological Flows. Ecol. Stud. 92. Springer-Verlag, Germany, p. 170-195. Hansson, L . 1994. Vertebrate distributions relative to clear-cut edges in a boreal forest landscape. Landscape Ecology. 9(2): 105-116. Harding, R.B. and D.F. Grigal. 1986. Site quality influences on biomass estimates for white spruce plantations. For. Sci. 32(2):443-446. Hargrove, W.W., R. H . Gardner, M.G. Turner, W.H. Romme, and D.G. Despain. 1993. A grid-based model for simulating fire patterns in heterogeneous landscapes. Presentation at the 8th Annual US Landscape Ecology Symposium. Oak Ridge Tenn. March 24-27, 1993. Harrington, J.B. and R.E. Donnelly. 1978. Fire probabilities in Ontario's boreal forest. In: 5th National Conference on Fire and Forest Meteorology. Atlantic City, N.J. American Meteorologic Society, pp 1-4. Harter, H.L. and A . H . Moore. 1965. Maximum-Liklihood estimation of the parameters of gamma and Weibull populations from complete and from censored samples. Technometrics 7(4): 639-643. Hawkes, B.C. 1979. Fire history and fuel appraisal studies of Kananaskis Provincial Park, Alberta M.Sc. thesis. U . Alberta, Edmonton. 173 p. Heinselman, M . L . 1973. Fire in the virgin forests of the Boundary Waters Canoe area, Minnesota. Quaternary Research 3, pp. 329-382. Heinselman, M . L . 1980. Fire and succession in the conifer forests of northern North America. In: West, D.C., H.H. Shugart, and D.B. Botkin (eds.), Forest Succession: Concepts and Applications. Springer-Verlag, New York. pp.374-405. Hobbs, R.J. 1983. Markov models in the study of post-fire succession in heathland communities. Vegetatio 56: 17-30. Hof, J.G. and L . A . Joyce. 1992. Spatial optimization for wildlife and timber in managed forest ecosystems. Forest Science 23(3): 489-508. Holt, R.D., G.R. Robinson, and M.S. Gaines. 1995. Vegetation dynamics in an experimentally fragmented landscape. Ecology 76(5): 1610-1624. Horn, H.S. 1975. Markovian properties of forest succession. In: Cody, M . L . and J .M. Diamond (eds), Ecology and Evolution of Communities. Cambridge, Mass., Belknap Tress, p. 196-211. Hulshoff, R . M . 1995. Landscape indices describing a Dutch landscape. Landscape ecology 10(2): 101-111. 178 Hunsaker, C.T., R.V. O'Neill , B.I. Hackson, S.P. Timmins, D.A. Levine, and D.J. Norton. 1994. Sampling to characterize landscape pattern. Landscape Ecology 9(3): 207-226. Irland, L .C. 1994. Getting from here to there: Implementing ecosystem management on the ground. J. of Forestry 92(8): 12-17. Jenkins, S.H. 1979. Seasonal and year-to-year differences in food selection by beavers. Oceologia 44:112-116. Johnson, E.A. 1992. Fire and vegetation dynamics: Studies from the North American Boreal Forest. Cambridge U . Press, Great Britain. 129 p. Johnson, E.A. and J.S. Rowe. 1977. Fire and vegetation change in the western Subarctic. A L U R 75-76-61. Ottawa, Canada. Johnson, E.A. and C.E. Van Wagner. 1985. The theory and use of two fire history models. Can. J. For. Res. 15: 214-220. Jones, R.K. 1986. Soil survey and its use in forest management in Ontario. In: G . M . Wickware and W.C. Stevens (co-chair). Site Classification in Relation to Forest Management. Gov't of Can., Can. For Serv., Sault Ste Marie, Ont. COJFRC Symp. Proc. O-P-14. pp. 29-37. Kalabokidis, K .D. , C M . Hay, and Y . A . Hussin. 1991. Spatially resolved fire growth simulation. In: 11th Conference on Fire and Forest Meteorology, April 16-19, 1991, Missoula, Montana, pp. 188-195. Kareiva, P. and U . Wennergren. 1995. Connecting landscape patterns to ecosystem and population processes. Nature 373(28): 299-302. Keane, R.E., S.F. Arno, and J.K. Brown. 1989. FIRESUM - A n ecological process model for fire succession in western conifer forests. USDA For. Serv. Gen. Tech. Rep. INT-266. 20 p. Kessell, S.R. 1976. Gradient modelling: A new approach to fire modelling and wilderness resource management. Environmental Management 1(1): 39-48. Kessell, S.R. 1979. Gradient modelling: Resource and fire management. Springer-Verlag, New York. 432 p. Kimmins, J.P. 1987. Forest Ecology. MacMillan Publishing Co., New York. 531 p. Knight I. and J. Coleman. 1993. A fire perimeter expansion algorithm based on huygens' wavelet propagation. Int. J. Wildland Fire 3(2): 73-84. Knight, B.F. and H.J. Heikkenen. 1980. Principles of Forest Entomology. 5th ed. McGraw-Hil l , N Y . 461 p. Knight, D.H. 1987. Parasites, lightning, and the vegetation mosaic in wilderness landscapes. In: 179 Goigel-Turner, M . (ed), Landscape Heterogeneity and Disturbance. Ecol. Stud. 64. Springer-Verlag, Germany, p. 59-83. Kolasa, J. and C D . Rollo. 1991. Introduction. The heterogeneity of heterogeneity: A glossary. In: Kolasa, J. and S.T.A. Pickett (eds), Ecological Heterogeneity. Ecological Studies 86. Springer-Verlag. p. 1-23. 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. Leemans, R. 1991. Canopy gaps and establishment patterns of spruce (Picea abies (L.) Karst) in two old-growth coniferous forests in Central Sweden. Vegetatio. 93:157-165. Leopold, A . 1949. A Sand County Almanac and Sketches Here and There. Oxford University Press. N Y . Lepart, J. and M . Debussche. 1992. Human impact on landscape patterning: Mediterranean examples. In: Hansen, A.J . and F. Di Castri (eds), Landscape Boundaries: Consequences for Biological Diversity and Ecological Flows. Ecol. Stud. 92. Springer-Verlag, Germany, p. 76-106. Levin, S.A. 1992. The problem of pattern and scale in ecology. Ecology 63(6): 1946-1967. L i , H . , J.F. Franklin, F.J. Swanson, and T.A. Spies. 1993. Developing alternative forest cutting patterns: A simulation approach. Landscape Ecology 8(1): 63-75. Lorimer, N.D. , R.G. Haught, and R.A. Leary. 1994. The fractal forest: Fractal geometry and applications in forest science. USDA For. Serv. N C For. Exp. Stn. Gen. Tech. Rep. NC-170. St. Paul, Minn. 43 p. Luque, S.S., R.G. Lathrop and J.A. Bognar. 1994. Temporal and spatial changes in an area of the New Jersey Pine Barrens landscape. Landscape Ecology. 9(4): 287-300. Mackey, B.G. , D. W. McKenney, C A . Widdifield, R.A. Sims, K. Lawrence, and N . Szcyrek. 1994. A new digital elevation model of Ontario. Northern Ontario Development Agreement Technical Report TR-6. Dept. of Nat. Res. Can., C.F.S. Sault Ste Marie, Ont. 26 p. Malanson, G.P. 1985. Simulation of competition between alternative shrub life history strategies through recurrent fires. Ecological Modelling, 27: 271-283. Marsh, G.P. 1864. Man and Nature. Harvard University Press, Cambridge, Mass. 472 p. McAlpine, R.S., B.D. Lawson, and E. Taylor. 1991. Fire spread across a slope. In: Andrews, P.L. and D.F. Potts (eds), Proc. of the 11th Conf. on Fire and Forest Meteorology. April 16-19, 1991, Missoula, Montana. Society of American Foresters, Bethseda, M D . pp. 218-225. 180 McAlpine, R.S. and B . M . Wotton. 1993. The use of fractal dimension to improve wildland fire perimeter predictions. Can. J. For. Res. 23: 1073-1077. McGarigal, K. and B.J. Marks. 1994. FRAGSTATS - Spatial pattern analysis program for quantifying landscape structure. V . 2.0. Oregon State U . , Corvallis OR. Mclntyre, N.E. 1995. Effects of forest patch size on avian diversity. Landscape Ecology. 10(2): 85-100. McQuilkin, R.A. 1976. The necessity of testing of soil-site equations. Soil Sci. Soc. Am. J. 40:783-785. McRae, R.H.D. 1990. Use of digital terrain data for calculating fire rates of spread with the P R E P L A N computer system. Mathematical and Computer Modelling 13(12): 37-48. Menon, M . V . 1963. Estimation of the shape and scale parameters of the Weibull distribution. Technometrics. 5: 175-182. Merriam, G. 1987. Landscape ecology: The ecology of heterogeneous systems. In: Moss, M.R. (ed), Landscape Ecology and Management. Proc. of the First Symp. of the Can. Soc. for Land. Ecol. and Manage. U . of Guelph. p. 43-50. Merriam, G. and J. Wegner. 1992. Local extinctions, habitat fragmentation, and ecotones. In: Hansen, A.J . and F. Di Castri (eds), Landscape Boundaries: Consequences for Biological Diversity and Ecological Flows. Ecol. Stud. 92. Springer-Verlag, Germany, p. 150-169. Methven, I. and V . Feunekes. 1987. Fire games for park managers: Exploring the effect of fire on landscape vegetation patterns. In: Moss, J.M. (ed), Landscape Ecology and Management. Proceedings of the first symposium of the Canadian Society for Landscape Ecology and Management. U . of Guelph, Ontario, p. 101-109. Milne, B.T. 1991. Lessons from applying fractal models to landscape patterns. In: Turner, M . G . and R.H. Gardner, (eds), Quantitative Methods in Landscape Ecology. Ecol. Stud. 82. Springer-Verlag, New York. p. 199-235. Mitchell, S.J. 1995. A synopsis of windthrow in British Columbia: occurrence, implications, assessment and management. In: Wind and Trees, Countts, M.P. and J. Grace (eds.). Cambridge University Press. Great Britain, pp. 448-304. Mladenoff, D.J., G.E. Host, J. Boeder, and T.R. Crow. 1993. LANDIS: A spatial model of forest landscape disturbance, succession, and management. In: Second International Conference on Integrating Modelling and GIS. in draft. Mladenoff, D.J., M . A . White, and J. Pastor. 1993. Comparing spatial pattern in unaltered old-growth and disturbed forest landscapes. Ecological Applications 3(2): 294-306. 181 Monserud, R.A., U . Moody, and D.W. Breuer. 1990, A soil-site study for inland Douglas-fir. Can. J. For. Res. 20:686-695. Muller, M.R. and J. Middleton. 1994. A Markov model of land-use change dynamics in the Niagara Region, Ontario, Canada. Landscape Ecology 9(2): 151-157. Murcia, C. 1995. Edge effects in fragmented forests: implications for conservation. TREE 10(2): 58 - 62. Mutch, R.W. 1970. Wildland fires and ecosystems - A hypothesis? Ecology. 51(6): 1046-1051. Naveh, Z. 1987. Biocybernetic perspectives of landscape ecology and management. In: Moss, J .M. (ed), Landscape Ecology and Management. Proceedings of the first symposium of the Canadian Society for Landscape Ecology and Management. U . of Guelph, Ontario, p. 23-34. Nemani, R., S. Running, L . Band, and D. Peterson. 1992. Regional hydro-ecological simulation system: An illustration of the integration of ecosystem models in a GIS. In: Goodchild, M . , B. Parks, and L . Skynext (eds), Integrating GIS and Environmental Modelling, Oxford, London. 13 P-Noble, R. and R.O. Slatyer. 1980. The use of vital attributes to predict successional changes in plant communities subject to recurrent disturbances. Vegetatio. 43:5-21. Noss, R.F. 1990. Can we maintain biological and ecological integrity? Conservation Biology. 4(3): 241-243. Noss, R.F. 1990b. Indicators for monitoring biodiversity: A hierarchical approach. Conservation Biology 4(4):355-364. Noss, R.F. 1994. At what scale should we manage biodiversity? Oregon State University, Corvallis, Or. 25 p. (draft). Odum, W.E., T.J. Smith III, and R. Dolan. 1987. Suppression of natural disturbance: Long term ecological change on the outer banks of North Carolina. In: Goigel-Turner, M . (ed), Landscape Heterogeneity and Disturbance. Ecol. Stud. 64. Springer-Verlag, Germany, p. 123-135. Oliver, C D . 1981. Forest development in North America following major disturbance. For. Ecol. Manage. 3:153-168. O'Neill , R.V. 1989. Perspectives in hierarchy and scale. In: May, R . M . and S.A. Levin (eds), Perspectives in ecological theory. Princeton University Press, p. 140-156. O'Neill , R.V., D.L. de Angelis, J.B. Waide, and T.F.H. Allen. 1986. A hierarchical concept of ecosystems. Monographs in Population Biology 23, Princeton University Press. 253 p. 182 O'Neill , R.V., B.T. Milne, M.G. Turner, and R.H. Gardner. 1988. Resource utilization scales and landscape pattern. Landscape Ecology 2(1): 63-69. O'Neill , R.V., S.J. Turner, V.I . Cullinan, D.P. Coffin, T. Cook, W. Conley, J. Brunt, J .M. Thomas, M.R. Conley, and J. Gosz. 1991. Multiple landscape scales: An intersite comparison. Landscape Ecology 5(3): 137-144. P A M A P . 1989. P A M A P Geographic Information System, Version 2.2, Victoria, BC. Parminter, J. 1983. Fire-ecological relationships for the Biogeoclimatic zones of the Cassiar timber supply area: summary report. In: BC Ministry of Forests, Northern Fire Ecology Project: Cassiar Timber Supply Area. Internal report. BC Ministry of Forests, Victoria, BC . 64 p. Pastor, J. and M . Broschart. 1990. The spatial pattern of a northern conifer-hardwood landscape. Landscape Ecology. 4(1): 55-68. Payette, S. 1993. Fire as a controlling process in North American boreal forest. In: West, D.C., H.H. Shugart, and D.B. Botkin (eds.), Forest Succession: Concepts and Applications. Springer-Verlag, New York. pp. 144-169. Payne, N.F. and F.C. Bryant. 1994. Techniques for wildlife habitat management of uplands. McGraw-Hill, N Y , 840 p. Pojar, J. R. Trowbridge, and D. Coates. 1984. Ecosystem classification and interpretation of the Sub-Boreal Spruce zone, Prince Rupert Forest Region, British Columbia. Land Management report no. 17. BC Ministry of Forests. 319 p. Rapp, M . 1983. Problems of disturbance on the nutrient cycling in ecosystems. In: Mooney, H.A. and M . Godron (eds), Disturbance and Ecosystems: Components of Response. Ecol. Stud. 44. Springer-Verlag, Germany, p. 117-128. Reiners, W.A. 1983. Disturbance and basic properties of ecosystem energetics. In: Mooney, H.A. and M . Godron (eds), Disturbance and Ecosystems: Components of Response. Ecol. Stud. 44. Springer-Verlag, Germany, p. 83-98. 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(1): 23-40. Ripple, W.J., G.A. Bradshaw, and T.A. Spies. 1991. Measuring forest landscape patterns in the Cascade Range or Oregon, USA. Biological Conservation. 57: 73-88. Risser, P.G. 1987. Landscape ecology: State of the art. In: Goigel-Turner, M . (ed), Landscape Heterogeneity and Disturbance. Ecol. Stud. 64. Springer-Verlag, Germany, p. 3-14. Robinson, G.R., R.D. Holt, M.S. Gaines, S.P. Hamburg, M . L . Johnson, H.S. Fitch, and E.A. Martinko. 1992. Diverse and contrasting effects of habitat fragmentation. Science 257: 524-526. 183 Romme, W.H. 1982. Fire and landscape diversity in subalpine forests of Yellowstone National Park. Ecological Monographs. 52(2): 199-221. Romme, W.H. and D.Ft. Knight. 1982. Landscape diversity: the concept applied to Yellowstone Park. Bioscience 32(8): 664- 670. Rothermel, R.C. 1972. A mathematical model for predicting fire spread rate and intensity in wildland fuels. USDA For. Serv. Res. Paper INT-115. 40 p. Rowe, J.S. 1972. Forest Regions of Canada. Dept. of the Envir. CFS Publication No. 1300. 172 p. Rowe, J.S., D. Spittlehouse, E. Johnson, and M . Jasieniuk. 1975. Fire studies in the upper Mackenzie Valley and adjacent Precambrian uplands. A L U R 74-75-61, Ottawa, Canada. Rulcker, C , P. Angelstam, and P. Rosenberg. 1994. Ecological forestry planning: A proposed planning model based on the natural landscape. SkogForsk, Redogorelse nr. 8, 4 p. Runkle, J.R. 1985. Disturbance regimes in temperate forests. In: Pickets, S.T.A. and P.S. White (eds), The Ecology of Natural Disturbance and Patch Dynamics. Academic Press, p. 17-33. Rykiel, E.J., W.C. Forsythe, W.M. Childress, and B. L i . 1993. Enormous increases in numbers of state transitions with neighbourhood size. Presentation at the 8th Annual US Landscape Ecology Symposium, Oak Ridge, Term. March 24-27, 1993. Salwasser, H. 1994. Ecosystem management: Can it sustain diversity and productivity? J. of Forestry 92(8): 6-10. Sapsis, D.B., and Martin. 1993. Fire, the landscape, and diversity. A theoretical framework for managing wildlands. Presented at the 12th Conference on Fire and Forest Meteorology, Oct. 26-28, 1993, Jekyll Is. GA. p. 270-278. Scanlon, J.C. and S. Archer. 1991. Simulated dynamics of succession in a North American subtropical Prosopic savanna. J. Veg. Sci. 2:625-634. Shaver, H.R., K.J . Nadelhoffer, and A.E . Giblin. 1991. Biogeochemical diversity and element transport in a heterogeneous landscape, the North Slope of Alaska. In: Turner, M . G . and R.H. Gardner, (eds), Quantitative Methods in Landscape Ecology. Ecol. Stud. 82. Springer-Verlag, New York. Shore, T.L. and L. Safranyik. 1992. Susceptibility and risk rating systems for the mountain pine beetle in lodgepole pine stands. Pacific Forestry Centre Information Report BC-X-336. Forestry Canada, Victoria, BC. 12 p. Shugart, H.H, and I.R. Noble. 1981. A computer model of succession and fire response of the high-altitude Eucalyptus forest of the Brindabella range, Australian Capital Territory. Aust. J. of Ecol. 6: 149-164. 184 Shugart, H.H. and D.C. West. 1981. Long-term dynamics of forest ecosystems. Amer. Sci. 69: 647-652. Sinclair, A.R.E. , D.S. Hix, O.J. Schmitz, G.G.E. Scudder, D.H. Turpin, and N.C . Larter. 1995. Biodiversity and the need for habitat renewal. Ecological Applications 5(3): 579-587. Sklar, F.H. and R. Costanza. 1991. The development of dynamic spatial models for landscape ecology: a review and prognosis. In: Turner, M.G. and R.H. Gardner, (eds), Quantitative Methods in Landscape Ecology. Ecol. Stud. 82. Springer-Verlag, New York. p. 239-288. Slocombe, S. 1993. Implementing ecosystem-based management. Bioscience 43(9): 612-622. Smith, T .M. and Urban, D.L. 1988. Scale and resolution of forest structural pattern. Vegetatio. 74:143-150. Society of American Foresters. 1993. Task force report on sustaining long-term forest health and productivity. Bethseda, Maryland: SAF. 83 p. Sousa, W.P. 1984. The role of disturbance in natural communities. Ann. Rev. Ecol. Sys. 15:353-391. Sprugel, D.G. 1985. Natural disturbance and ecosystem energetics. In: Pickets, S.T.A. and P.S. White (eds), The Ecology of Natural Disturbance and Patch Dynamics. Academic Press, p. 335-352. Steel, R.G.D. and J.H. Torrie. 1980. Principles and Procedures of Statistics: A Biometrical Approach. 2nd ed. McGraw-Hill. Stocks, B.J. 1987. Fire behaviour in immature jack pine. Can. J. For. Res. 17: 80-86. Stocks, B.J. 1989. Fire behaviour in mature jack pine. Can. J. For. Res. 19: 783-790. Stocks, B.J., B.D. Lawson, M.E. Alexander, C.E. Van Wagner, R.S. McAlpine, T.J. Lynam, and D.E. Dube. 1989. The Canadian forest fire danger rating system: An overview. The Forestry Chronicle. 65: 450-457. Stocks, B.J., D.J. McRae, T.J. Lynham, G.R. Hartley. 1990. A photo-series for assessing fuels in natural forest stands in northern Ontario. COFRDA Report 3304. 162 p. Suffling, R. 1987. Catastrophic disturbance and landscape diversity: The implication of fire control and climate change in subarctic forests. In: Moss, J .M. (ed), Landscape Ecology and Management. Proceedings of the first symposium of the Canadian Society for Landscape Ecology and Management. U . of Guelph, Ontario, p. 111-120. Suffling, R., B. Smith, and J. Dal Molin. 1982. Estimating past forest age distributions and disturbance rates in North-western Ontario: A demographic approach. J. Envir. Mgmt. 14: 45-56. 185 Swaydon, R.W. 1987. Population responses to environmental disturbance. In: Disturbance in grasslands: causes, effects and processes. Geobotany 10: 15-31. Taylor, S.W., K. Kepke, N . Parfitt, and C.C. Ross. 1994. Wild fire frequency in Yukon ecoregions. Canadian Forest Service, Pacific Forestry Centre, Victoria, BC. 22 p. (draft). Taylor, S.W. and R.G. Pike. 1995. Field guide to the Canadian Forest Fire Prediction (FBP) System. Pacific Forestry Centre, CFS, Victoria, BC. (draft). Troll, G. 1971. Landscape ecology, geo-ecology and biocenology - a terminology study. Geoforum, 8: 43-46. Turner, M.G. 1987. Spatial simulation of landscape changes in Georgia: A comparison of 3 transition models. Landscape Ecology 1(1): 29-36. Turner, M.G. , and V . H . Dale. 1991. Modelling landscape disturbance. In: Turner, M . G . and R.H. Gardner (eds), Quantitative methods in landscape ecology. Ecol. Studies 82, Springer-Verlag. p. 322-351. Turner, M.G. and R.H. Gardner. 1991. Quantitative methods in landscape ecology: An introduction. In: Turner, M . G . and R.H. Gardner (eds), Quantitative methods in landscape ecology. Ecol. Studies 82, Springer-Verlag. p. 3-14. Turner, M.G. , R.V. O'Neill , R.H. Gardner, and B.T. Milne. 1989. Effects of changing spatial scale on the analysis of landscape pattern. Landscape Ecology. 3(3/4): 155-162. Turner, M.G. , and W.H. Romme. 1994. Landscape dynamics in crown fire ecosystems. Landscape Ecology 9(1): 59-77. Turner, S.J., R.V. O'Neill , W. Conley, M.R. Conley, and H.C. Humphries. 1991. Pattern and scale: statistics for landscape ecology In: Turner, M . G . and R.H. Gardner (eds), Quantitative methods in landscape ecology. Ecol. Studies 82, Springer-Verlag. p. 17-34. van Apeldoorn, R.C., C. Celada and W. Nieuwenhuizen. 1994. Distribution and dynamics of the red squirrel (Sciurus vulgaris L.) in a landscape with fragmented habitat. Landscape Ecology 9(3): 227-235. Van Wagner, C.E. 1978. Age-class distribution and the forest fire cycle. Can. J. For. Res. 8: 220-227. Van Wagner, C.E. 1992. Prediction of crown fire behaviour in two stands of jack pine. Can. J. For. Res. 23: 442-449. Vasconcelos, M.J. and D.P. Guertin. 1992. FIREMAP. Simulation of fire growth with a geographic information system. Int. J. Wildland Fire 2(2): 87-96. 186 Verbyla, D.L. and R.F. Fisher. 1989. An alternative approach to conventional soil-site regression modelling. Can. J. For. Res. 19:179-184. Wallace, L . L . , M . G . Turner, W.H. Romme, R.V. O'Neill , and Y . Wu. 1995. Scale of heterogeneity of forage production and winter foraging by elk and bison. Landscape Ecology 10(2): 75-84. Ward, P.C. and A . G . Tithecott. 1993. The impact of fire management on the boreal landscape of Ontario. OMNR, Aviation, Flood and Fire Management Branch Pub. No. 305. 12 p. Weinstein, D.A. and H.H. Shugart. 1983. Ecological modelling of landscape dynamics. In: Mooney H.A. and M . Godron (eds), Disturbance and Ecosystems: Components of Response. Ecol. Studies 44. Springer-Verlag, Germany, p. 31-43. Weintraub, A . and L. Davis. 1992. Hierarchical planning in forest resource management: Defining the dimensions of the subject area. In: Workshop on Hierarchical Approaches to Forest Management in Public and Private Organizations. May 25-29, Toronto, Ont. p. 1-26. (In draft). White, P.S. 1979. Pattern, process, and natural disturbance in vegetation. The Botanical Review. 45(3): 229-299. White, P.S., M.D. MacKenzie, and R.T. Busing. 1985. A critique on overstorey/understorey comparisons based on transition probability analysis of an old growth spruce-fir stand in the Appalachians. Vegetatio 64: 37-45. White, P.S. and S.T.A. Pickett. 1985. Natural disturbance and patch dynamics: an introduction. In: Pickets, S.T.A. and P.S. White (eds), The Ecology of Natural Disturbance and Patch Dynamics. Academic Press, p. 3-13. Wickham, J.D. and D.J. Norton. 1994. Mapping and analyzing landscape patterns. Landscape Ecology. 9(1): 7-24. Wilkinson, L . 1988. SYSTAT: The system for statistics. Evanston, II, SYSTAT Inc. Wood, C.S. and G.A. Van Sickle. 1991. Forest insect and disease conditions, B C and Yukon -1990. Pacific Forestry Centre Information Report BC-X-326. Forestry Canada, Victoria, BC . 33 p. Wood, C.S. and G.A. Van Sickle. 1993. Forest insect and disease conditions, B C and Yukon -1993. Pacific Forestry Centre Information Report BC-X-346. Forestry Canada, Victoria, BC. 33 p. World Resources Institute. 1992. Global biodiversity strategy. World Conservation Union, U .N . Environmental Programme, Food and Agriculture Organization and U.N. Education, Scientific and Cultural Organization. Yarie, J. 1981. Forest fire cycles and life tables: a case study from interior Alaska. Can. J. For. Res. 11: 554-562. 187 Zasada, J.C., T.L. Sharik, and M . Nygren. 1993. The reproductive process in boreal forest trees. In: Shugart, H.H. , R. Leemans, and G.B. Bonan (eds) A Systems Analysis of the Global Boreal Forest. University Press, Cambridge, p. 85-125. Zonneveld, I.S. 1987. Landscape ecology and its application. In: Moss, M.R. (ed), Landscape Ecology and Management. Proc. of the First Symp. of the Can. Soc. for Land. Ecol. and Manage. U . of Guelph. p. 3-16. 188 APPENDIX A - DEFINING (INVENTORY) AGE-CLASS 8 How do age-class 8 stands (141-250 years) break down into 20-year intervals in the SBSmkl? An accurate estimate of how the ages of older stands is distributed is necessary to define an age-class distribution for the landscape as a whole. The results described in this Appendix are used in Chapter 2 for determining fire regime parameters. DATA Two sources of data were available: 1) Cruise data from a local timber company. Operational cruise data were obtained from Lakeland Mills Ltd for all cutblocks in mapsheets 93J025 and 93J035. However, not all stands that were cruised had age information. A total of 48 samples in 45 stands were available. At least two trees per cutblock were sampled, but the age data for some cutblocks were not able to be differentiated since data were lumped. 2) Sample increment core and stump ring counting. Several age-class 8 stands were selected from maps for sampling from 1:20,000 inventory mapsheets. Sampling was not random; rather it was based on accessibility. A l l stands that were the required age, and within 500 m of the nearest road access were considered for sampling. Some large stands were sampled more than once. In addition, stumps were aged in the most recent cutblocks to either cross-reference cores, or to represent the adjacent age-class 8 stand. At least two dominant or co-dominant trees per sample location were aged. 189 The 59 sites sampled in 38 stands represented most of the accessible age-class 8 stands on the two central mapsheets (93J025 and 93J035). However, this sample represents only 10.5% of the age-class 8 stands according to the inventory records. An additional 24 sites in 24 stands were sampled from other SBSmkl mapsheets. Although all of the sampled sites were legitimately within the boundaries of the SBSmkl, 16 of these stands were close to the Englemann Spruce Subalpine Fir (ESSF) transition on the eastern edge. After consultation with the Regional Ecologist, the decision was made not to include the data from these sites. A number of cored sites within the SBSmkl also had to be removed from the analysis because of data collection problems (mostly sample size). In the end, 67 sites in 46 stands were accepted. The age data from the two sources overlapped to some degree. The full data set used to estimate the age-class breakdown included 105 sites in 83 stands. With both types of data, the age-class of the oldest pine was taken as the age-class of the sample location. When it was obvious that the oldest pine was either an outlier "survivor" or improperly sampled (counted), the second oldest pine was used to age the site. The oldest, or second oldest pine was used because it is most likely to indicate the timing of the last stand replacing disturbance event, minus the period of seeding in. Spruce was used to age stands only i f suitable pine was not found, or very old multi-aged stands were suspected. For those sites that clearly had two age-classes present, the most recent disturbance was used since I am ultimately interested in the frequency of forest fire disturbance events. 190 RESULTS According to the British Columbia Ministry of Forests forest inventory, age-class 8 stands include all stands aged 141 to 250 years. In order to break this down into 20 year classes, I extended the classification as follows: AGE RANGE CLASS 141-160 8 161-180 9 181-200 10 201-220 11 221-240 12 241-260 13 261-280 14 281-300 15 The data were compiled in two ways: by sample and by stand. From Table A l it can be seen that the results were almost identical for the two methods. In both cases, almost half of the age-class 8 stands were in the 141-160 year category, another one third in the 161-180 range, another 10% or so each in the next two oldest classes, and only one percent in the 221 -240 year category. No stands were found older than 240 years old within the plateau proper. The age-class 8 stands sampled in the ESSF transition zone which were not included in the estimate were often over this limit, although considerably more hybrid spruce and Douglas-fir were found in these areas. Lodgepole pine over 240 years was extremely rare anywhere within the SBSmkl . The STANDS percentages were used for the calculations in Chapter 2. 191 Table A l . Number (and percentage in brackets) of sampled plots and stands of mature forest in the SBSmkl in 20 year age-classes. Age-Class 8 9 10 11 12 TOTAL Age Range (yrs) 141-160 161-180 181-200 201-220 221-240 PLOTS results 49 (47) 34 (32) 9(9) 12(11) 1 (1) 105 (100) STANDS results 40 (48) 22 (27) 8(10) 12(14) 1 (D 83 (100) It should also be noted that these results were compiled using the youngest age cohort in duel-age stands since the ages are to represent the time since the last stand initiating event. Twenty-one of the plots (20% of the sample) showed more than one age cohort. It is not known whether this is a result of sampling close to historical fire edges, or a reflection of true multi-aged stands since the field sampling was not designed to address this question. 192 APPENDIX B - DETAILED SIMULATION RESULTS 193 Table B l . Simulation results summary of landscape average and standard deviations. SCENARIO PS (ha) PSSD (ha) ED (m/ha) TCAI (%) SIEI ave s.d. ave s.d. ave s.d. ave s.d. ave s.d. N A T U R A L 39 7 561 198 29 2 56 4 0.71 0.13 10,000 ha 23 3 306 108 34 3 46 3 0.81 0.07 10,000-20 35 3 414 125 31 2 53 3 0.79 0.07 3,000 21 2 282 108 37 3 43 3 0.79 0.06 3,000-20 32 3 311 94 32 2 49 2 0.83 0.04 1,000 19 2 217 99 38 3 39 3 0.81 0.06 1,000-20 28 2 258 87 34 3 47 2 0.82 0.04 60 22 2 192 100 37 4 39 2 0.84 0.04 A V E R A G E 36 4 512 122 30 2 55 3 0.75 0.08 10,000 22 1 251 79 35 1 44 3 0.84 0.04 10,000-20 34 3 376 99 31 2 52 2 0.82 0.06 3,000 20 1 201 47 37 1 41 2 0.85 0.02 3,000-20 32 1 291 61 32 1 50 1 0.84 0.02 1,000 23 1 153 30 37 1 40 1 0.87 0.01 1,000-20 29 1 197 43 34 1 46 1 0.87 0.01 60 22 1 96 15 38 1 37 1 0.88 0.01 >20 YEARS 25 4 374 141 33 3 48 4 0.77 0.11 10,000 22 2 284 84 35 • 2 44 2 0.84 0.05 3,000 20 1 217 52 37 2 41 2 0.85 0.03 1,000 18 1 144 32 39 1 37 1 0.87 0.02 60 19 1 104 18 39 1 36 1 0.88 0.01 >40 YEARS 31 3 443 115 32 2 52 2 0.76 0.08 10,000 27 2 397 77 33 1 49 2 0.77 0.04 3,000 24 2 304 86 35 1 44 2 0.82 0.04 1,000 22 1 223 40 37 1 42 1 0.82 0.02 60 21 1 144 36 38 1 38 1 0.85 0.02 >80 YEARS 28 2 361 81 31 2 49 2 0.83 0.05 10,000 29 1 329 86 31 2 50 1 0.82 0.04 3,000 23 2 252 57 34 1 45 2 0.82 0.03 1,000 19 2 217 99 38 3 39 3 0.81 0.06 60 22 1 223 52 36 1 41 1 0.80 0.01 > 80 +BUFFER 28 2 310 114 32 3 46 3 0.84 0.06 10,000 23 1 237 71 36 2 39 1 0.84 0.03 3,000 20 1 229 91 38 2 36 2 0.82 0.04 1,000 19 1 207 59 41 1 32 1 0.82 0.02 60 20 1 191 48 40 1 34 1 0.82 0.01 PS - average patch size TCAI - total core area index PSSD - patch size standard deviation SIEI - Simpson's evenness index ED - weighted edge density 194 Table B2. Simulation results summary of young age-class (0-40 years) averages and standard deviations. SCENARIO PCT. C H . PS (ha) PSSD (ha) TCAI (%) IJI ave. ave. ave. s.d. ave. s.d. ave. s.d. ave. s.d. Natural 40 18 283 283 1,475 1,411 61 11 79 n 10,000 37 12 124 82 730 644 51 7 81 9 10,000-20 38 12 222 108 1,033 766 61 5 82 8 1,000 37 10 83 54 400 426 45 6 76 9 1,000-20 37 11 142 54 516 378 55 4 83 5 3,000 38 10 100 50 584 460 50 6 78 8 3,000-20 36 12 162 50 657 383 58 4 85 5 60 34 9 107 51 330 395 46 4 81 6 Average 43 15 216 123 1,294 805 60 6 79 7 10,000 37 8 113 56 624 462 51 5 82 6 10,000-20 37 11 216 115 989 727 60 5 86 6 1,000 34 3 71 16 268 96 50 3 82 3 1,000-20 33 3 129 28 389 170 54 3 85 3 3,000 36 5 86 30 418 209 48 5 81 4 3,000-20 37 5 168 57 680 394 59 4 85 5 60 33 1 89 8 159 26 45 1 82 2 >20 years 42 15 235 189 1,401 1,140 53 9 78 10 10,000 33 8 111 71 635 562 49 6 82 6 1,000 34 4 69 15 274 128 43 3 81 3 3,000 34 5 90 37 466 310 47 5 81 5 60 34 2 93 12 202 45 42 1 81 2 >40 years 43 15 216 123 1,294 805 60 6 79 7 10,000 41 12 148 82 945 637 55 6 79 7 1,000 40 5 95 25 500 217 49 3 78 3 3,000 37 8 104 47 620 417 51 5 79 6 60 36 3 110 18 312 117 45 1 80 3 > 80 years 25 9 101 56 489 401 53 7 85 5 10,000 28 8 135 62 529 358 56 4 85 4 1,000 37 10 83 54 400 426 45 6 76 9 3,000 30. 5 109 49 414 251 51 5 80 7 60 33 2 83 12 224 81 43 2 70 2 > 80+buffer 25 10 67 38 327 283 49 9 85 6 10,000 27 5 44 10 166 75 42 4 79 4 1,000 29 3 32 6 94 35 32 3 71 3 3,000 29 6 38 12 142 78 37 4 75 4 60 30 1 49 5 90 20 34 1 68 3 PCT - percent land C H - average percent change per 20 years PS - patch size PSSD - patch size standard deviation TCAI - total core area index IJI - interspersion/juxtaposition index 195 Table B3. Simulation results summary of pole age-class (41-120 years) averages and standard deviations. SCENARIO PCT. CH. PS (ha) PSSD (ha) TCAI (%) IJI ave. ave. ave. s.d. ave. s.d. ave. s.d. ave. s.d. Natural 39 16 87 96 727 725 56 12 85 10 10,000 38 11 36 25 367 325 47 9 84 7 10,000-20 40 11 67 46 553 428 56 8 88 7 1,000 39 9 30 18 285 265 40 9 81 7 1,000-20 39 10 49 28 379 294 50 7 88 5 3,000 39 9 33 21 345 272 45 7 83 5 3,000-20 40 10 59 36 471 354 54 6 89 4 60 40 8 41 27 299 291 42 8 84 5 Average 39 14 51 50 465 453 54 9 86 9 10,000 38 8 30 12 296 166 47 5 86 4 10,000-20 40 10 59 31 503 317 56 7 90 5 1,000 39 3 35 6 230 73 44 3 86 2 1,000-20 39 3 46 11 296 117 51 3 89 2 3,000 40 5 30 11 275 154 45 5 86 2 3,000-20 40 5 51 15 407 173 55 4 90 3 60 38 1 31 3 151 39 41 1 85 1 >20 years 34 13 34 37 347 411 46 12 78 10 10,000 39 7 35 18 389 272 47 7 86 4 1,000 38 3 25 8 196 87 41 3 85 1 3,000 39 5 30 12 290 146 45 5 86 3 60 38 2 25 4 155 50 39 2 84 1 >40 years 39 14 51 50 465 453 54 9 86 9 10,000 40 11 45 31 469 375 51 7 86 6 1,000 40 5 31 8 275 115 46 3 85 3 3,000 42 7 41 19 425 284 49 5 87 5 60 40 2 31 8 209 107 43 2 85 2 > 80 years 46 9 100 48 835 439 57 5 91 3 10,000 47 7 95 47 752 430 58 4 93 2 1,000 39 9 30 18 285 265 40 9 81 7 3,000 48 4 73 24 612 265 55 3 93 1 60 49 2 56 8 469 155 49 2 87 1 > 80+buffer 44 9 86 48 703 473 53 6 88 4 10,000 47 5 67 18 555 238 47 3 88 2 1,000 49 3 58 12 504 197 40 2 84 2 3,000 48 6 66 28 569 350 44 4 87 3 60 49 1 62 8 464 134 42 1 85 1 PCT - percent land . PSSD - patch size standard deviation C H . - average percent change per 20 years TCAI - total core area index PS : patch size IJI - interspersion/juxtaposition index 196 Table B4. Simulation results summary of old age-class (121+ years) averages and standard deviations. SCENARIO PCT. CH. PS (ha) PSSD (ha) TCAI (%) IJI ave. ave. ave. s.d. ave. s.d. ave. s.d. ave. s.d. Natural 10 4 10 9 63 84 30 16 76 n 10,000 15 4 8 4 59 52 33 10' 75 8 10,000-20 12 3 11 6 62 48 35 12 81 8 1,000 13 3 6 3 32 30 23 8 71 7 1,000-20 13 3 9 3 37 19 32 7 76 7 3,000 12 3 6 3 29 24 24 8 73 6 3,000-20 13 3 11 4 51 35 35 6 78 6 60 15 2 8 2 28 13 27 6 72 6 Average 7 3 6 3 35 28 28 10 77 8 10,000 14 2 7 2 47 24 32 7 78 5 10,000-20 12 3 11 4 62 36 38 7 80 6 1,000 16 1 9 1 41 16 28 3 77 2 1,000-20 18 1 12 2 60 14 38 3 81 3 3,000 14 2 7 1 38 14 29 3 75 3 3,000-20 12 2 10 2 45 18 36 5 80 4 60 18 1 8 1 29 6 29 2 77 2 >20 years 14 2 8 8 69 94 27 17 76 11 10,000 17 3 8 3 66 41 36 7 77 6 1,000 17 1 7 1 47 24 30 4 75 3 3,000 16 2 7 3 50 38 31 6 76 5 60 17 1 7 1 26 5 29 3 75 2 >40 years 7 2 6 3 35 28 28 10 77 8 10,000 8 2 5 2 25 14 24 6 74 7 1,000 9 1 5 1 20 7 22 4 72 4 3,000 11 2 6 2 29 13 26 7 73 4 60 13 1 6 1 21 6 25 4 73 3 > 80 years 18 4 10 4 85 66 39 9 79 6 10,000 14 3 8 3 46 26 36 7 79 5 1,000 13 3 6 3 32 30 23 8 71 7 3,000 11 2 6 8 30 56 25 8 74 7 60 7 1 3 0 12 3 19 3 72 2 . > 80+buffer 20 4 12 5 83 44 38 7 78 7 10,000 15 2 7 2 38 17 27 5 72 4 1,000 11 1 4 1 17 6 16 3 67 3 3,000 12 2 5 1 19 7 19 4 68 5 60 10 1 4 0 13 3 16 2 73 2 PCT - percent land PSSD - patch size standard deviation CH - average percent change per 20 years TCAI - total core area index PS - patch size IJI - interspersion/juxtaposition index 197 


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



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"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items